Ion of Chromatographic Selectivity - 1986

JOURNAL OF CHROMATOGRAPHY LIBRARY - volume 35
optimization of chromatographic selectivity

a guide to method development
This Page Intentionally Left Blank
JOURNAL OF CHROMATOGRAPHY LIBRARY - volume 35
optimization of
chromatographic
selectivity
a guide to method development
Peter J. Schoenmakers
Philips Research Laboratories, P. 0. Box 80.000,5600JA Eindhoven, The Netherlands
ELSEVl ER
Amsterdam - Oxford - New York - Tokyo 1986
ELSEVIER SCIENCE PUBLISHERS B.V.
Sara Burgwhartstraat 25
P.O. Box 21 1 , l OOO AE Amsterdam, The Netherlands
Distributors for the United Statas and C . d a :
ELSEVIER SCIENCE PUBLISHING COMPANY INC.

52, Vanderbilt Avenue
New York. N Y 10017, U.S.A.
ISBN 0-444-42681-7 (Vol. 35)

ISBN 0444-41616-1 (Series)
0 Elsevier Science Publishers B.V.. 1986
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system or
transmitted i n any form or by any means, electronic, mechanical, photocopying. recording or other-
wise, without the prior written permission of the publisher, Elsevier Science Publishers B.V./Science &
Technology Division, P.O. Box 330, 1000 A H Amsterdam, The Netherlands.
Special regulations for reeders in the USA - This publication has been registered with the Copyright
Clearance Center Inc. (CCC), Satem, Massachusetts. Information can b e obtained from the CCC about
conditions under which photocopies of parts of this publication may be made in the USA. All other
copyright questions, including photocopying outside of the USA, should b e referred t o the publisher.
Printed in The Netherlands

CONTENTS
Journal of Chromatography Library (other volumes in the series) IX
PREFACE XI11
ACKNOWLEDGEMENTS XV
CHAPTER I INTRODUCTION 1
1.1 CHROMATOGRAPHY 1
1.2 SEPARATION - T H E COLUMN 2
1.2.1 Retention times and capacity factors 2
1.2.2 Distribution coefficients 4
1.2.3 Selectivity 5
1.2.4 The phase ratio 5
1.3 RESOLUTION 7
1.4 EFFICIENCY 8
1.4.1 The plate number 9
1.5 OPTIMIZATION 9
1.6 PEAK CAPACITY 14
1.7 METHOD DEVELOPMENT 15
1.7.1 An organized approach 15
1.7.2 Method development in the laboratory 18
REFERENCES 19
CHAPTER 2 SELECTION OF METHODS 20

2.1 CLASSIFICATION O F CHROMATOGRAPHIC TECHNIQUES 20
2.2 SELECTION O F CHROMATOGRAPHIC METHODS 21
2.2:1 Expert systems 23
2.3 CHARACTERIZATION A N D CLASSIFICATION METHODS 24
2.3.1 Polarity; Solubility parameters 24
2.3.2 The Rohrschneider characterization scheme 27
2.3.3 The Snyder solvent classification scheme 31
2.3.4 Summary 35
REFERENCES 36
CHAPTER 3 PARAMETERS AFFECTING SELECTIVITY 37

3.1 GAS CHROMATOGRAPHY 37
3.1.1 Gas-liquid chromatography (GLC) 37
3.1.2 Gas-solid chromatography (GSC) 43
3.1.3 The use of retention indices 45
3.2 LIQUID CHROMATOGRAPHY 47
3.2.1 Liquid-liquid chromatography (LLC) 52
3.2.2 Liquid bonded phase chromatography 56
V
3.2.2.1 Reversed phase chromatography (RPLC) 56
3.2.2.2 Polar bonded phases 74
3.2.3 Liquid-solid chromatography (LSC) 76
3.3 SEPARATION OF IONS IN LC 82
3.3.1 Ion-exchange chromatography (IEC) 82
3.3.2 Ion-pair chromatography (IPC) 93
3.4 SUPERCRITICAL FLUID CHROMATOGRAPHY (SFC) 101
3.5 CLASSIFICATION O F PARAMETERS 105
3.5.1 Summary of parameters for selectivity optimization 108
REFERENCES 113
CHAPTER 4 OPTIMIZATION CRITERIA 116

4.1 INTRODUCTION 116
4.1.1 Separation of two peaks 116
4.1.2 Separation in a chromatogram 117
4.2 ELEMENTAL CRITERIA 119
4.2.1 Peak-valley ratios 119
4.2.2 Fractional peak overlap 123
4.2.3 Separation factor 125
4.2.4 Discussion 127
4.3 CHROMATOGRAMS 131
4.3.1 Sum criteria 131
4.3.2 Product criteria 134
4.3.3 Minimum criteria 140
4.3.4 Other criteria 144
4.3.5 Summary 145
4.4 COMPOSITE CRITERIA 146
4.4.1 Number of peaks 146
4.4.2 Analysis time 148
4.4.3 Column independent time factors 151
4.4.4 Time corrected resolution products 153
4.5 RECOMMENDED CRITERIA FOR THE GENERAL CASE 158
4.6 SPECIFIC PROBLEMS 158
4.6.1 Limited number of peaks of interest 158
4.6.2 Programmed analysis 165
4.6.3 Dealing with solvent peaks 167
REFERENCES 169
CHAPTER 5 OPTIMIZATION PROCEDURES 170

5.1.1 Univariate optimization 173
5.1.2 Local vs. global optima 176
5.1.3 Characteristics of optimization procedures 177
5.1.4 Definitions 179
5.2 SIMULTANEOUS METHODS WITHOUT SOLUTE RECOGNITION 179
5.3 THE SIMPLEX METHOD 183
VI
5.4 REDUCTION O F T H E PARAMETER SPACE 188
5.4.1 Full factorial designs 188
5.4.2 Scouting techniques 191
5.5 INTERPRETIVE METHODS 199
5.5.1 Simultaneous interpretive methods 200
5.5.2 Iterative designs 220
5.5.3 Summary 233
5.6 PEAK ASSIGNMENT A N D RECOGNITION 233
5.6.1 Single channel detection 236
5.6.2 Dual-channel detection 239
5.6.3 Multichannel detection 24 1
5.7 SUMMARY . 245
REFERENCES 250
CHAPTER 6 PROGRAMMED ANALYSIS 253

6.1 T H E APPLICATION O F PROGRAMMED ANALYSIS 253
6.2 PARAMETERS AFFECTING SELECTIVITY I N PROGRAMMED
ANALYSIS 257
6.2.1 Temperature programming in G C 258
6.2.2 Gradient elution in LC 260
6.3 OPTIMIZATION O F PROGRAMMED ANALYSIS 266
6.3.1 Optimization of programmed temperature G C 269
6.3.1.1 Sequential methods 269
6.3.1.2 Interpretive methods 273
6.3.1.3 Discussion 275
6.3.1.4 Selectivity optimization 276
6.3.1.5 Summary 276
6.3.2 Optimization of programmed solvent LC 276
6.3.2.1 Simplex optimization 277
6.3.2.2 Systematic optimization of program parameters 279
6.3.2.3 Interpretive methods for selectivity optimization 284
6.3.2.4 Discussion 290
6.3.2.5 Summary 292
REFERENCES 294
CHAPTER 7 SYSTEM OPTIMIZATION 296

7.2 EFFICIENCY OPTIMIZATION 299
7.2.1 Open columns vs. packed columns 299
7.2.2 Gas chromatography (open columns) 300
7.2.3 Liquid chromatography (packed columns) 302
7.2.4 Summary 305
7.3 SENSITIVITY OPT1M IZATION 305
7.4 INSTRUMENT OPTIMIZATION 310
7.4.1 Gas chromatography (open columns) 314
7.4.2 Liquid chromatography (packed columns) 316
VII
7.4.3 Summary 318
REFERENCES 318
LIST OF SYMBOLS AND ABBREVIATIONS 321
AUTHOR INDEX 329
SUBJECT INDEX 333
VIII
JOURNAL OF CHROMATOGRAPHY LIBRARY
A Series of Books Devoted to Chromatographic and Electrophoretic
Techniques and their Applications
Although complementary to the Journal of Chromatography, each volume in the Library
Series is an important and independent contribution in the field of chromatography and
electrophoresis. The Library contains n o material reprinted from the journal itself.
Other volumes in this series
Volume 1 Chromatography of Antibiotics (see also Volume 26)

by G.H. Wagman and M.J. Weinstein
Volume 2 Extraction Chromatography
edited by T. Braun and G. Ghersini
Volume 3 Liquid Column Chromatography. A Survey of Modern Techniques and
Applications
edited by 2.Deyl, K. Macek and J. Janak
Volume 4 Detectors in Gas Chromatography
by J. 8evEik
Volume 5 Instrumental Liquid Chromatography. A Practical Manual on High-Per-
formance Liquid Chromatographic Methods (see also Volume 27)
by N.A. Parris
Volume 6 Isotachophoresis. Theory, Instrumentation and Applications
by F.M. Everaerts, J.L. Beckers and Th.P.E.M. Verheggen
Volume 7 Chemical Derivatization in Liquid Chromatography
by J.F. Lawrence and R.W. Frei
Volume 8 Chromatography of Steroids
by E. Heftmann
Volume 9 HPTLC - High Performance Thin-Layer Chromatography
edited by A. Zlatkis and R.E. Kaiser
Volume 10 Gas Chromatography of Polymers
by V.G. Berezkin, V.R. Alishoyev and I.B. Nemirovskaya
Volume 11 Liquid Chromatography Detectors (see also Volume 33)
by R.P.W. Scott
Volume 12 Affinity Chromatography
by J. Turkova
Volume 13 Instrumentation for High-Performance Liquid Chromatography
edited by J.F.K. Huber
Volume 14 Radiochromatography. The Chromatography and Electrophoresis of
Radiolabelled Compounds
by T.R. Roberts
Volume 15 Antibiotics. Isolation, Separation and Purification
edited by M.J. Weinstein and G.H. Wagman
Volume 16 Porous Silica. Its Properties and Use as Support in Column Liquid Chro-
matography
by K.K. Unger
Volume 17 -
7 5 Years of Chromatography A Historical Dialogue
edited by L.S. Ettre and A. Zlatkis
IX
Volume 18A Electrophoresis. A Survey of Techniqu2s and Applications.
Part A: Techniques -
edited by Z. Deyl
Volume 18B Electrophoresis. A Survey of Techniques and Applications.
Part B: Applications
edited by Z. Deyl
Volume 19 Chemical Derivatization in Gas Chromatography
by J. Drozd
Volume 20 Electron Capture. Theory and Practice in Chromatography
edited by A. Zlatkis and C.F. Poole
Volume 21 Environmental Problem Solving using Gas and Liquid Chromatography
by R.L. Grob and M.A. Kaiser
Volume 22A Chromatography. Fundamentals and Applications of Chromatographic
and Electrophoretic Methods. Part A: Fundamentals
edited by E. Heftmann
Volume 22B Chromatography. Fundamentals and Applications of Chromatographic
and Electrophoretic Methods. Part B: Applications
edited by E. Heftmann
Volume 23A Chromatography of Alkaloids. Part A: Thin-Layer Chromatography
by A. Baerheim Svendsen and R. Verpoorte
Volume 23B Chromatography of Alkaloids. Part B: Gas-Liquid ChromatoBaphy and
High-Performance Liquid Chromatography
by R. Verpoorte and A. Baerheim Svendsen
Volume 24 Chemical Methods in Gas Chromatography
by V.G. Berezkin
Volume 25 Modern Liquid Chromatography of Macromolecules
by B.G. Belenkii and L.Z. Vilenchik
Volume 26 Chromatography of Antibiotics
Second, Completely Revised Edition
by G.H. Wagman and M.J. Weinstein
Volume 27 Instrumental Liquid Chromatography. A Practical Manual on High-Per-
formance Liquid Chromatographic Methods
by N.A. Parris
Volume 28 Microcolumn High-Performance Liquid Chromatography
by P. Kucera
Volume 29 Quantitative Column Liquid Chromatography. A Survey of Chemometric
Methods
by S.T.Balke
Volume 30 Microcolumn Separations. Columns, Instrumentation and Ancillary
Techniques
edited by M.V. Novotny and D. Ishii
Volume 31 Gradient Elution in Column Liquid Chromatography. Theory and Practice
by P. Jandera and J. ChudEek
Volume 32 The Science of Chromatography. Lectures Presented at the A.J.P. Martin
Honorary Symposium, Urbino, May 27-31,1985
edited b y F.Bruner
X
Volume 33 Liquid Chromatography Detectors. Second, Completely Revised Edition
by R.P.W.Scott
Volume 34 Polymer Characterizationby Liquid Chromatography

by G. Glockner
Volume 35 Optimization of Chromatographic Selectivity. A Guide to Method

Development
by P.J. Schoenmakers
Volume 36 Selective Gas Chromatographic Detectors

by M. Dressler
XI
The aim of this book is to be of help to those involved in the process of developing
chromatographic methods. I have tried to write a text that is comprehensible and useful
for both chromatographers with some experience, and for novices to the field with a
background in science.
The fundamentals of chromatography are not covered in detail; the reader is referred
to one of the introductory textbooks or courses on the subject.
Method development in chromatography today requires skills, knowledge and, above
all, experience. Therefore, it is a particularly difficult field to enter for newcomers. I feel
that an organized approach to method development, as presented in this book, may shift
the emphasis from experience to knowledge. In this way, it may help newcomers to
understand the process of method development.Also it may open the way for those already
involved in method development to go beyond their personal experience and to apply
different chromatographic techniques and optimization procedures.
The approach followed should be equally beneficial for chromatographers who do not
develop their own methods but wish to improve (optimize) existing ones.
Procedures for developing and optimizing chromatographic separations have attracted
increasing attention not only from researchers, but also from instrument manufacturers.
Already, several of the procedures described are commercially available. The approach
followed does not include describing existing methods. One reason for not doing this is
that the elements that constitute a complete optimization package can be discussed and
understood separately.Therefore, an existingmethod may be good in one respect, but poor
in another. A second reason is that whereas complete optimization packages may be
expected to change a great deal in the next few years, I expect this to be much less true
for the underlying principles; so I would like to think that the material presented here will
still be of value in the years to come.
This book is intended to be a critical assessment of procedures for method development
and selectivity optimization. It is not intended to be a survey of available information,
therefore references to the literature are included only when they are relevant to the text.
Consequently, a number of references have been omitted. No doubt, some may also have
been overlooked.
I am very grateful to a number of people who have reviewed the manuscript of this book
(or parts of it) at various stages during the preparation. Together, they are responsible for
an immense number of corrections, improvements and clarifications.
My gratitude to:
Hugo Billiet (Delft, The Netherlands),
Pieter de Bokx (Eindhoven, The Netherlands),
Cherie Goewie (Utrecht, The Netherlands),
Ernst Lankmayr (Graz, Austria),
Pamela Naish (Cambridge, Great Britain),
Charles Perkins (Cambridge, Great Britain),
Frank Verhoeven (Eindhoven, The Netherlands),
XI11
and especially my teacher for many years
Leo de Galan (Delft, The Netherlands),
from whom I am still learning, and my wife,
Dana Conron,
for correcting the final manuscript and one or two other things.
Eindhoven, February 1986

Peter Schoenmakers
XIV
ACKNOWLEDGEMENTS
I acknowledge permission to reprint previously published material from the following
publishers:
American Chemical Society, Washington D.C., U.S.A.

Figures 3.3, 3.4, 3.20, 5.22, 5.36, 5.38.
Marcel Dekker Inc, New York, NY, U.S.A.

Figure 3.18.
Elsevier Science Publishers, Amsterdam, The Nederlands.

Figures3.2, 3.5, 3.6, 3.13, 3.14,3.15, 3.19, 3.22,3.25, 3.29, 3.30, 3.31, 3.33, 5.1, 5.2, 5.7,
5.8,5.9,5.11,5.13,5.14,5.15,5.16,5.17,5.20,5.21,5.23,5.24,5.25,5.26,5.27,5.37,
6.8, 6.10, 6.11, 6.12, 6.14, 6.15, 6.16.
Preston Publications Inc., Niles, IL, U.S.A.

Figure 2.3.
Royal Chemical Society, London, Great Britain.

Figure 3.1.
The following authors are acknowledged for permission to reprint material originally
published in Chrornatographia by Friedr.Vieweg und Sohn Verlagsgesellschaft mbH,
Wiesbaden, West Germany:
L.de Galan (Delft, The Netherlands)

Figures 3.7, 3.8, 3.16, 3.26, 5.5, 5.29, 5.30, 5.31, 5.32, 5.33, 5.34.
CXiuiochon (Washington D.C., U.S.A.)
Figures 5.18, 5.19.
J.F.K.Huber (Vienna, Austria)
Figure 3.10.
Figure 3.20, 3.23 and 3.24 were adapted from original drawings of J.C.Kraak
(Amsterdam, The Netherlands).
Figure 3.12 was adapted from an original provided by H.M.van den Bogaert (Eindho-
ven, The Netherlands).
Figures 4.4 and 4.5 were taken from unpublished work of A.C.J.H.Drouen (Delft, The
Netherlands).
xv
CHAPTER I
INTRODUCTION
In this chapter the concepts of chromatography, as far as they are relevant to the context
of this book, will be outlined.
The chromatographic system, the column, and the basic fundamentals of chromato-
graphic separations will be briefly discussed.
The extent of separation can be quantified in terms of the resolution obtained between
two consecutivechromatographic peaks. This resolution can be expressed in terms of three
elemental characteristics of chromatographic separation: retention, selectivity and effi-
ciency. The influence of each of these three factors on resolution will be discussed.
1.1 CHROMATOGRAPHY
Chromatography can be defined as the separation of molecules by differential

migration*, i.e. separation is achieved on the basis of different speeds of transportation
for different molecules.
In this book column chromatography will be discussed almost exclusively, although
occasional reference will be made to thin layer chromatography (TLC), the fundamentals
of which are not different from those of column chromatography.
Furthermore, this treatment is limited to those forms of chromatography which involve
two phases (a stationary and a mobile phase) and in which the necessary differences in
speed of migration are caused by differences in chemical interactions between the
molecules of the different sample components (“solutes”) and the two chromatographic
phases, as well as between the solute molecules themselves. Interaction chromatography
is sometimes used as a term to describe such systems.
Separations that are achieved on the basis of the size of the molecules (e.g. size exclusion
chromatography) are not dealt with in this book. Such separations are not selective, and
hence there is no selectivity to be optimized.
I
- - -
Mobile
Phase
~ ~
- Sample
Introduction
l ,
i r q k
~ ~ ~
Detection
~
- Data
Handling
- --
Figure 1.l:Schematic representation of a chromatograph.
A schematic representation of a chromatograph is given in figure 1.I. This figure applies

to all kinds of column chromatography, but the various boxes will have different contents
for different chromatographic techniques, notably for gas chromatograhy (GC) and for
liquid chromatography (LC) (for definitions see section 2.1).
* In this broad definition some techniques which are not usually considered as chromatography are
included, for example field flow fractionation (FFF) techniques and electrophoresis. However,
isotachophoresis is not included.
I
For GC the mobile phase delivery box could consist of a gas cylinder, a reducing valve
and a flow controller. For LC a high pressure pump will be required. In this book the
instrumentation required for chromatography will not be discussed. Only where the
equipment used is relevant to the cause of optimization of selectivity will it feature in the
present text (e.g. sections 5.6 and 7.4).
The rest of this book will focus on the thick box in the centre of figure 1.1, identified
as separation.
1.2 SEPARATION - THE COLUMN
The chromatograph is built around the column, in which the actual separation takes
place. The column accommodates the two chromatographic phases: the stationary phase,
which remains in the column, and the mobile phase, which is transported through it.
Separation is achieved because different sample components (solutes) show different
distributions over the two phases. A solute, having such a high affinity towards the
stationary phase that it resides in this phase exclusively,will stay in the column indefinitely.
A solute, that does not enter the stationary phase at all, will be transported through the
column at the same speed at which the mobile phase is transported. In chromatographic
terms, the latter is called an “unretained solute.
If a column is packed with porous particles, then an unretained solute is assumed to be
swept through the entire volume of the column that is occupied by the mobile phase, either
outside the particles or in the pores. A solute that does not enter any of the pores is called
a (completely) “excluded” solute. Throughout the remainder of this book we will assume
that the solutes will not be (partially or completely) excluded from the pores.
1.2.1 Retention times and capacity factors
The above discussion can be quantified as follows. A solute i distributes itself over the
two phases, resulting in a total quantity qiemto be present in the mobile phase (m), and a
quantity qi.sin the stationary phase (s). The solute molecules which find themselves in the
mobile phase will be transported through the column at the same speed (u ) as the molecules
of the mobile phase. However, this is only a fraction of all the solute molecules, so the
average speed for all solute molecules will be only a fraction of u given by
9i.m
v. = u,
’ qi.m+ 4i.s
where vi is the migration speed, the average speed at which the solute band travels through
the column. The time tR,ineededfor the solute band to elute from the column is determined
by the column length and the average migration speed
tR,i = L / v i .
tR,i is called the retention time of the solute. Similarly, the time which a mobile phase
molecule will spend in the column is
to = L / u
2
to (frequently also denoted by t,,,) is known under different names: the hold-up time,
mobile phase time, or unretained time.
The combination of eqns.( 1.l), (I .2) and (1.3) yields
By definition, the capacity factor (k,)of the solute i is
and hence
c , =~ (1 + k,) t o . (1.6)
Eqn.(l.6) is the fundamental equation for retention in chromatography. Throughout this

book, extensive use will be made of the capacity factor as a convenient means to describe
retention. A major advantage of the use of k for this purpose is the fact that it is a
dimensionless quantity. It follows from eqn.( 1.6) that
0
tlmin - I
Pigure 1 .L: Schematic chromatogram illustrating the meaning ot various retention parameters
3
where f R V i is the net retention time of the solute, i.e. the (average) time which a solute
molecule spends in the stationary phase.
Eqm(l.7) also shows that the capacity factor k can easily be determined from the
chromatogram. This is illustrated in figure 1.2. If a signal at t = t , is obtained in the
chromatogram, then the quantities to, t,, and tkcan all bemeasured directly. The capacity
factor can either be calculated from eqn.(l.7), or determined from a calibration line as
shown in figure 1.2. Two points can be used to construct the line, for instance k= 0 at the
occurance of the unretained peak and (a fictive point) where k = - 1 at the time of injection
( t = 0). The capacity factor of any peak in the chromatogram can be determined very easily
in this way. However, to avoid inaccuracies if high k values occur,
the calibration line may be constructed by using eqn.(l.7) once for a point at a high
value of k.
1.2.2 Distribution coefficients
The quantity q of the solute i in one of the phases is the product of the average
concentration (7) of i in that phase (where the average is taken along the length of the
column) and the volume of that phase. Hence, for the capacity factor (eqn.l.5) we find
The ratio Zi,s/Ci,m is a constant if the distribution isotherm*, i.e. a plot of q svs. qm,is
linear. This is usually the case at high dilutions. Preferably, all (analytical) chromatograp-
hy is performed in this linear region. The distribution coefficient in terms of concentra-
tions (KJ may be defined as
Kc,i = c i , J q m . (1.9)
Since K, may be independent of the solute concentration, but will always be a function
of the temperature (and pressure), the term distribution coefficient is to be preferred to
the alternatives: distribution constant and equilibrium constant. If the distribution
isotherm is linear, K , will also equal the ratio of average concentrations in eqn.(l.8), and
hence
ki = &. Vs/ V , (1.10)
Eqn.(l .lo) relates retention in chromatography (k)to a thermodynamic parameter (KJ.

The so-called phase ratio Vs/Vmis a characteristic of the column**.
* If the stationary phase is asolid surface, then the term adsorption isotherm is more commonly used.
** However, in some kinds of chromatography (e.g. reversed phase liquid chromatography, see
section 3.2) the phase ratio may vary with variations in the mobile phase composition.
4
1.2.3 Selectivity
It was stated at the beginning of this section that solutes are separated in a chromatograp-
hic column on the basis of differences in their speed of migration through the column. We
can define the relative retention (aj,)of two peaks as
a..
I'
= tiJ/ (1.11)
In this equation i represents the first eluting peak of a peak pair a n d j the last eluting peak.
Hence, by definition a is always larger than unity. Sometimes a is called the separation
factor, which is somewhat unfortunate terminology because separation is influenced by
other factors than just a (see section 1.3)*. a is the chromatographic parameter that is most
directly related to the selectivity of the phase system. In this book, therefore, the word
selectivity will often be associated with a. Using eqns.(l.7) and (1.10) we can write two
other equations for aj,:
a..
J'
= kj / ki (1.12)
and
a,.
J'
= K . / K,i
CJ
(1.13)
Eqn(l.12) is very useful in practice, because it expresses a directly in terms of the

capacity factors. We will make frequent use of this equation throughout this book.
Eqn.(l.l3) relates a to the distribution coefficients. Since no phase ratio term appears
in eqn.( 1.13), it is clear that the selectivity (a)of the chromatographic system is determined
only by thermodynamic factors.
The relative retention will be affected only by those factors which affect the distribution
coefficients, i.e.
.- the solute
.- the mobile and the stationary phase (together constituting the phase system)
.- the temperature
.- the pressure.
The effect of the pressure on a and on k is usually negligible. Only in some particular
cases (e.g. in supercritical fluid chromatography, SFC; see section 3.4) will it be a relevant
parameter.
1.2.4 The phase ratio
The phase ratio V,/ V,,,occurs in eqn.(l .lo) as one of the factors that determine retention
(:k)in chromatography. We can influence the phase ratio by varying one or more of several
parameters:
* In chapter 4 we will define a separation factor S which provides a more realistic measure of the
contribution of chromatographic retention to separation.
5
- The type of column
In particular, we can choose between open (capillary) columns and packed columns*.
A wall coated open tubular (WCOT)column has a much smaller phase ratio than a packed
column, due to the small surface area of the wall.
- The column diameter
If open columns are used, then the phase ratio will vary with the column diameter
(provided that the film thickness is kept constant). The cross-sectional area of the column
(and hence the mobile phase volume) is proportional to the square of the column diameter,
while the wall area is proportional to the diameter itself. Hence, the phase ratio is inversely
proportional to the column diameter.
- The sugace area
The area available for the stationary phase will directly affect the phase ratio. If a solid
material is used as the stationary phase in a packed column, if a liquid phase is deposited
on a solid adsorbent with a constant film thickness, or if chemically bonded phases are
employed, the phase ratio (through V,) will be directly proportional to the available
surface area. The surface area of an adsorbent is usually given per unit weight (i.e. the
specific surface area in m2/g). However, it should be noted that the relevant quantity is
the surface area per unit volume (m2/ml) in the packed column.
- The column porosity
This is the fraction of the column volume that remains available for the mobile phase
after packing. There are two contributions to the total column porosity. One part of the
volume available to the mobile phase is in between the particles (interparticle space). For
uniform, spherical particles this is about 40% of the column volume. The second
contribution is due to the very porous structure of materials with large specific surface
areas. This makes a significant part of the intraparticle volume available to the mobile
phase (usually 20 to 30% of the column volume).
- Thefilm thickness of a liquid stationary phase
Clearly, with all other factors constant, V, will increase linearly with the film thickness
(this is also true for the phase ratio V,/ V,, as long as V, < V,,,).For solid adsorbents this
effect does not occur. For chemically bonded phases the (mono-)layer thickness is not as
well defined as the film thickness of a bulk liquid, and neither is the description of variation
in the layer thickness as straightforward as it is for liquids (see section 3.2.2).
In general, the effective volume of a stationary phase (Vb can be increased in a

predictable manner by increasing the surface area, but only for liquids can the same be
said for increasing the film thickness.
Obviously, there are many ways to influence the capacity factors. However, the effects
described above are predictable (see section 4.2.3) and in a sense trivial. It is worth noticing
at this point.that certain parameters do not at all affect the capacity factor and therefore
do not at all affect chromatographic selectivity. These parameters include column length,
flow rate and the diameter of packed columns. This renders these parameters irrelevant
to the selectivity optimization process. In some cases they may be considered as parameters
* For gas chromatography(and for supercriticalfluid chromatography)there is a real choice. Open

columns may theoretically be used in liquid chromatography as well, but their diameter should then
be so small that they do not yet form a realistic alternative to packed columns in practice.
6
during the course of the optimization in conjunction with other parameters, which do
affect the selectivity.For example, a decreased temperature in GC may lead to an increased
selectivity and hence allow an increased flowrate to compensate for the increase in analysis
time. In most cases, however, it is sensible to consider the parameters that do not affect
the selectivity separately after completion of the optimization process. Some comments on
how to choose the values of these parameters will be made at the end of this book (chapter
7).
1.3 RESOLUTION
We have seen that the selectivity in chromatography can be related to the relative
retention of two solutes. However, this parameter does not describe the actual separation
between two chromatographic peaks. There are two factors which determine whether or
not two peaks are completely resolved, as is illustrated in figure 1.3. The relevant
parameters are the distance between the peaks and their width. The distance can be
expressed as the difference in retention times (AtR),while the peak width at the peak base
(usually determined by drawing tangent lines along the slopes of the peaks) can be denoted
by w.
Figure 1.3: Two chromatographic peaks illustrating the definition of resolution (eqn.l.14).
The resolution (R,) between two peaks is now defined as
R, = 2AtR / (w,+ w2). (1.14)
Hence, the resolution is equal to unity if the distance between two peaks equals the average
7
peak width. It should be noted that R, is a dimensionless quantity and that At, and w
should be expressed in the same units (e.g. seconds or mm on a recorder trace).
1.4 EFFICIENCY
Ideally, chromatographic peaks are of Gaussian shape. In practice, because of finite

sample concentrations, inhomogeneities in the stationary phase, dead volumes in the
system and various other factors, they usually are not. In LC peaks tend to be less
symmetrical than in GC.
Nevertheless, to a first approximation the Gaussian peakshape can be assumed for a
chromatographic peak. Iffit) is the signal (detector response) as a function of time and
t, is the retention time of the peak, then a Gaussian peak can be described by
fit) = hexp -1/z (T)

t-tt, 2
(1.15)
In eqm(l.15) h is the height at the peak maximum, A is the peak area, and a i s the standard
deviation of the peak (in time units), a measure of its width.
Some characteristics of a Gaussian peak are summarized in figure 1.4. It can be seen
from this figure that the variance of a genuinely Gaussian peak can be determined by
measuring the peak width at some fixed fraction of the peak height. The peak width at the
t
h/hrnm
0.607
0.5
0.135
Figure 1 . 4 Some characteristicsof a Gaussian peak.
8
base is usually very hard to measure accurately, so that a measurement of the peak width
at half height ( w , / ~ is
) usually considered to be a more practical proposition. However,
most measurements of peak width suffer from lack of precision and accuracy. It is beyond
the scope of this book to discuss other ways for characterizing the peak shape, peak width
and peak symmetry in chromatography.
1.4.1 The plate number
The efficiency of a chromatographic system (i.e. the column plus the instrument) is usually
expressed in terms of the number of theoretical plates (4,which may be defined as
follows:
N = (tR/o)2. (1.16)
Here the peak shape is assumed to be Gaussian, with 0 being the standard deviation. It
follows from figure 1.4 that eqn.(l.l6) can also be written as
N = 5.54 (t,/w,,2)2 (1.1 6a)
or
N z 16(t,/w)2 (1.16b)
where w l I 2 and w are the peak width at half height and the peak width at the baseline,
respectively.
Another convenient equation can be derived for instruments that provide information on
both the peak height (h) and the peak area (A):
N = 2 n(t, h/A)*. (1.17)
In applying eqn.( 1.17) one should be aware of the units involved. N is dimensionless, so
that if t , is expressed in seconds and h in mV, A should be expressed in mV.s.
From the number of plates in the column, the “height equivalent of a theoretical plate”
(HETP), usually abbreviated to “plate height” (H), can easily be calculated:
H = L/N (1.18)
where L is the length of the column.

The plate height will vary with the flow rate ( u ) of the mobile phase through the column.
This variation can be characterized by an H vs. u curve. Such a curve shows a minimum
plate height at some optimum value of u. Again, this will not be discussed any further in
this book and the reader is referred to one of many general textbooks on chromatography.
1.5 OPTIMIZATION
Eqns.( 1.6), (1.12), (1.14) and (1.1 6)can now be combined to yield the key equation for
9
the optimization of resolution in chromatography. If we combine eqm(l.14) and (1.1 6)
we find
R, =
-
- tR,2- ‘R,I .-
vxi (1.19)
(Ol + ‘R.1 + ‘R.2 2
or, in terms of net retention times (eqn.l.7)
R, =
fk.2- tk.1 .-
fi (1.19a)
‘R.1 + tk,2 + 2t0
which upon division of all retention times by to turns into
R, = k2-kl -5 (1.20)
k,+k2+2 2
Eqn.( 1.20) is in itself useful and it will be used later in the book (section 4.2.3). However,
a more generally useful equation can be found by some manipulation:
R,=-.
h-k, k,+k2 fi
.- (1.21)
k,+k2 k , + k , + 2 2
and using eqn.( 1.12) to define aand introducing the average capacity factor E = (k, + k2)/2
for the two peaks we find
R, =
a-I
-.-.- E vxi (1.22)
a+l l + E 2 .
’r
O 1 2 3 L 5
a-
0’ ’ ’0
5,000 10,000
N-
Figure 1.5: Influence of (a) the relative retention (a),(b) the (average) capacity factor(x) and (c) the
number of theoretical plates ( N )on t_he resolution ( R J according to eqn.(1.22). In each case the two
other parameters are kept constant. k and N are assumed not to equal zero, and a not to equal one.
10
Eqn.(1.22) is one of several forms of the same general equation. It is preferred here
because it is symmetrical (towards the two peaks) and exact.
Eqn.( 1.22) shows that there are three factors which together determine the chromato-
graphic resolution. The influence of each of these factors can be discussed independently
x
of the two others, as is shown in figure 1.5. In this figure R, is plotted against a, and N
(from top to bottom). In each of the three plots the two other parameters are kept constant.
Figure 1.5a shows the variation of the factor (a- l ) / ( a +1) with a. Note that by definition
a> 1. It is seen from the figure that this factor, and therefore R , increases more or less
regularly when a increases. If a= 1, then R, equals zero. The value of R, will slowly
increase with increasing a. Even when a= 2 only one third of the maximum value for the
a factor of 1 has been reached. In thermodynamic terms this is already a very large
selectivity, since the two capacity factors (i.e. the distribution coefficients) should differ
by a factor of two (eqns.l.12 and 1.13). Hence, to increase the resolution a is of extreme
importance when its value is close to one, but even at higher values resolution will benifit
substantially from an increase in the thermodynamic selectivity.
x
From figure 1.5b it appears that the variation of R, with is not dissimilar to the
dependence on a. If z=0, then R, will always equal zero (no matter how high the value
of a may be). The factor z/(l
- +x) z.
will increase regularly with increasing However, at
k = 1 already half of the limiting value of 1 has been reached. When % = 9this is 90%, so
that very little can be gained in terms of resolution by increasing the k value further.
Moreover, higher values imply longer analysis times (eqn.l.6) and are therefore less
attractive.
z
The optimum value can be found if we combine eqa(1.22) with eqn.(l.l8) and
eqn.( 1.3):
(1.23)
so that with eqn.(l.6) b e average retention time for a pair of peaks (TR) becomes
t R= t o ( l + k ) = R: (1.24)
z
a can be treated as independent from the average value and if we assume as a first
x, z
approximation that H / u is also independent of then the optimum value would be such
as to minimize the factor.(l + z ) 3 / pThis
. function is plotted in figure 1.6 and it can be
seen from this figure that a minimum value occurs at %=2*. Since eqn(1.24) shows a
shallow minimum with respect to z (see figure 1.6), a range of 1 < k < 5 is usually
considered as the optimum for the separation of mixtures of two components [103]. This
range is indicated in the figure. However, when more than two peaks are present, the range
of capacity factors will also be determined by the peak capacity of the chromatogram
(section 1.6).
* In a more detailed analysis Snyder [loll found a?

optimum around 3 <z< 5 for packed columns.
Guiochon [lo21 found an optimum value around k = 2 in a detailed study of capillary columns,
11
Figure 1.5c illustrates the variation of the resolution with the plate number N.The square
root function initially rises steeply from a value of zero at N =0. The rate of increase in
I6becomes quickly less with increasing N. At N = 2500 I6equals 50. A value of 100
is reached for N = lO,OOO, and 200 for N=40,000.These three values of N may be seen
as typical for three different kinds of chromatography: packed column GC, packed
column LC and capillary column GC. Within the constraints of any of these techniques,
another factor of 2 in resolution by increasing N by a factor of 4 will be hard to achieve.
Improving resolution by increasing the plate number will lead to longer analysis times.
If all other conditions (column or particle diameter, flowrate etc.) remain constant, then
the plate count will be proportional to the column length (eqn.l.18 with H constant), and
so will be the analysis time (eqn.1.2). Hence, R , can be increased by a factor of two at the
expense of a factor of four in analysis time. Usually, the situation is even worse. Merely
increasing the column length to increase N will lead to an increased pressure drop over
the column. Once the pressure drop becomes too high, other parameters will have to be
varied besides the column length, for example the particle diameter in packed columns
may be increased. This results in a decrease in the pressure drop, but also in an increase
in H, leading to a further increase in the required column length and hence to a further
increase in analysis time (eqn.l.2). It can be shown that in such a pressure-limited situation
a twofold increase in R, can be achieved from a fourfold increase in N at the cost of a
sixteenfold increase in analysis time (see also sections 4.4.3 and 7.2.3).
We conclude from the above that the efficiency (flm

in eqm(l.22) is the least rewarding
factor for increasing the resolution.
First, the (average) capacity factor (z)
should be brought into the optimum range, for
instance by varying the temperature (GC) or the mobile phase composition (LC).
Second, The selectivity (a)may be improved. This is extremely rewarding when ais close
to unity, but even at much higher values considerable improvement may be obtained. a
5 - 10
k-
Figure 1.6: Variation of the factcr (1 +E)3/p as a function of z. The minimum observed at E = 2
representstheoptimum value for &in terms ofminimum analysis time(eqn.l.24). Thearrow indicates
the optimum working range (1 < k < 5) for the separation of two peaks.
12
a =11
._
N = 1600
k=O
Figure 1.7: Illustration of the effects of the capacity factor, the selectivity and the plate number on
the appearance of chromatographic peaks. For explanation see text. -
Middlechromatogram: 5 pairs of peaks, all with a= 1.1 and N = 1600. k values increase from left to
right (k=0,1, 3, 5 and 8). Top left: k = 3 , a= l.O_S,N = 1600. Bottom left: k = 3 , a= 1.25, N = 1600.
Top right: k = 8 , a = l . l , N=500. Bottom right: k = 8 , a = l . l , N=5000.
may be varied by changing the nature or the composition of the stationary phase (GC or
L,C) or the mobile phase (LC)*.
- Third, the efficiency ( N )may be adapted to meet the requirements set by the values of
k and a. The value of N is determined by the column characteristics and the flow rate.
While increasing the plate count may require great sacrifices in terms of analysis time, the
reverse is also true. If the values of k and a allow the use of a column with a low N value
to achieve the separation, then the analysis time may be reduced dramatically.
Figure 1.7 illustrates the dependence of R, on a, and N using some actual
chromatographic peaks. The (horizontal) chromatogram in the middle contains a series
* Increasing a is generally advantageous for the separation of two peaks. If a series of solutes needs
to be separated, then an increase in one of the a values may have an adverse effect on the required
analysis time. The problem of how to distribute the peaks over the chromatogram will be discussed
extensively in chapter 4.
13
of peak pairs, for which a (1.1) and N (1,600) are constant. It is seen how the resolution
increases with increasing capacity factors. When z = O the two peaks show complete
overlap. The resolution is then found to increase quickly from the first peak (actually a
pair of overlapping peaks) at x= 0to the third pair around z= 3. R, may still be increased
z
from there by increasing further. However, the improvement in going from x= z=
3 to 8
is not dramatic.
The peak pair at z= 3 is also shown in figure 1.7 for a= 1.05 (top) and a= 1.25 (bottom).
When a= 1.05 only one peak can be observed (R, = 0.37), while a= 1.25 yields an excellent
separation ( R , = 1.67).
The peak pair at I= 8 is shown in figure 1.7 for three different values of N.N = 500 (top
right), N = 1600 (middle) and N = 5000 (bottom right). The improvement in R, for each
factor of about 3 in N is clear from the figure.
1.6 PEAK CAPACITY
The optimal working range in terms of capacity factors will not only be determined by
the considerations of analysis time given in the previous section, but also by the number
of peaks present in the chromatogram. The theoretical peak capacity (n,,) of a chromato-
gram can be found from [lo41
(1.25)
where k, and k, are the capacity factors of the first and the last peak, respectively, and
R, is the required resolution between each pair of successive peaks.
0
log N - 5
Figure 1.8: Variation of the theoretical peak capacity (n& of a chromatogram withihe plate count
(N).for three ranges of capacity factors: (a) 0.2< k < 2 , (b) 1 > k < 10 and (c) 0.5< k<20.
14
In figure 1.8 the theoretical peak capacities are given for some typical ranges of capacity
factors as a function of the number of plates. The resolution values were taken to be equal
to one. In the logarithmic plot of figure 1.8 a straight line is obtained with a slope of 1/2.
A typical packed column with 5000 plates turns out to yield a peak capacity between 17
(case a) and about 50 (case c). An open column with 200,000 plates may accommodate 100
peaks with capacity factors between 0.2 and 2 (case a).
Davis and Giddings [lo51have argued that the theoretical peak capacity is usually not
even approached. Instead, they conclude that in order to provide a 90% probability for
a compound of interest to appear as a pure peak in the chromatogram, the available peak
capacity should exceed the theoretically required value (eqn.1.25) by a factor of 20. If we
consider the same range of capacity factors this results in an excess plate count of about
a factor of 400.
In the treatment of Davis and Giddings the peaks are supposed to be randomly
distributed over the chromatogram. Optimization of selectivity can be seen as the process
to fight statistics and to approach the theoretical peak capacity as closely as possible.
1.7 METHOD DEVELOPMENT
In this section we will discuss a general approach to method development in

chromatography. Developing chromatographic methods is a difficult task. Unlike the
situation for most other methods of instrumental analysis, chemistry is at the heart
of successful method development in chromatography. Good, reliable instrumentation
is a prerequisite for performing sound chromatographic separations. However, the
best possible instrumentation will not provide the required separation without a sensible
choice of the phase system (stationary and mobile phase) and the operating conditions.
Almost this entire book (chapters 2 to 6) is intended to assist in making that sensible
choice.
Usually, those who develop chromatographic methods rely on knowledge, experience
and skill. This makes the field an especially hard one for newcomers to enter. The overview
in this section provides an organized approach to method development, which is intended
to introduce (relative)newcomers to the remainder of the book. Therefore, this section will
contain many references to subsequent chapters and sections.
An organized approach to method development may also be beneficial to the more
experienced chromatographer. It may provide an incentive to go beyond the limits of
personal experience and knowledge. Subsequent chapters in this book are intended to
provide sufficient information to allow the development and optimization of separation
methods in a variety of different chromatographic techniques.
1.7.1 An organized approach
An organized approach to method development may be based on a series of actions as

described below.
1. Getting to know the sample
It will seldom be necessary to analyse samples about which nothing is known, except
15
that they are green, liquid, and smelling. Usually, much more is known about the sample
and it is important to acquire as much information as possible about the sample on
beforehand. This includes obvious questions, such as which are the (expected)components
in the sample, which components are of interest, as well as less direct questions, involving
such characteristics as:
- volatility of the sample (presence of a non-volatile fraction)
- possible solvents
- presence of acids and/or bases
- history of the sample (e.g. details about the synthesis).
Answers to such questions are essential in order to choose the most appropriate
chromatographic technique.
There is a second kind of sample, which is hardly ever found in a real-life method
development laboratory. That is the sample one knows everythingabout. Even if this might
seem to be the case, it is good to be aware of the possibility that unexpected components
may be present in the sample. These may include contaminants, side products, degradation
products, metabolites, etc..
2. Reading up
It is always advisable to search the literature for possible solutions to a paPticular

problem. The task of re-inventing the wheel should not be thought of lightly. Even a brief
scan of the literature may reveal much about the problem. In most cases this will provide
clues as to which chromatographic technique might be used, as well as a starting point with
regard to a possible phase system. Because separations from the literature can often not
be reproduced exactly and because of differences between different samples, this starting
point is usually not the final stage in a method development process and a further
optimization is usually required.
3. Method selection
If the literature does not suggest a particular chromatographic technique to be applied,

then a method should be selected on the basis of the information we have about the sample
and some rules of thumb. In chapter 2 we will discuss methods of going about the selection
of a chromatographic technique. This chapter may help us to select a particular
chromatographic technique (for instance gas-liquid chromatography or reversed phase
chromatography). Chapter 2 also describes some general methods for the characterization
and classification of phase systems and solvents, but it does not tell us how to realize a
separation once an appropriate technique has been selected.
4. Recording a chromatogram
After a particular chromatographic technique has been selected a chromatogram should

be recorded. This is not always straightforward, since recording a chromatogram requires
that the components in the sample show reasonable retention times (capacity factors).
Often, therefore, we have to adjust the chromatographic system in order to get all sample
components to appear as a peak (but not necessarily a well-resolved one) in the
16
chromatogram. This can be done by adapting what we may call the primary parameters
of the separation. The primary parameters for different chromatographic techniques will
be discussed in chapter 3. That chapter is divided into different sections for different
techniques, so that once a particular technique has been selected (using information from
the literature or from chapter 2) the required knowledge about the effects of different
parameters on retention and selectivity is readily available. The most relevant parameters
for each technique are summarized at the end of the chapter (table 3.10) for quick
reference.
In some cases we may speed up the selection of appropriate primary parameters with
the help of programmed analysis, i.e. temperature programming in GC or solvent
programming in LC. Another useful scouting technique may be thin layer chromatograp-
hy (TLC). Possibilities for establishing the appropriate values of the primary parameters
will be discussed in section 5.4.
Ideally, the eventual chromatogram is arrived at under constant conditions (i.e.
isothermal in GC, isocratic in LC). Nevertheless, it may be impossibble to achieve a signal
(peak) for each component in the sample under constant conditions. In that case it may
be necessary to use programmed analysis methods, in which one of the (primary)
parameters is varied (“programmed) during the separation. Programmed analysis will be
discussed in chapter 6.
5 . Optimization of the separation
Once we have recorded a chromatogram, the separation may not be satisfactory. Indeed,
the chances are that not all peaks will be well-resolved.In that case, we will have to optimize
the separation. Eqm(1.22) is the key to this optimization step. The three factors which
together determine the value of R , may to some extent be considered independently. In
fact, in recording the chromatogram we assumed that the retention (capacity factor) may
be optimized independently of the two other factors in the resolution equation. Recording
the chromatogram is equivalent to adapting the capacity factors such that they are neither
too high nor too low. The range of 1 <x< 10 may be considered as optimal (see section
lS), but when the sample contains very few components or when very many plates are
available, a somewhat lower range (e.g. 1 to 5 or 0.5 to 2) may be preferred (see section
1.6).
Another factor that contributes to R, is the plate count N. However, we have seen in
section 1.5 that optimization through an increase in N is expensive, not only in terms of
equipment and columns, but also in terms of analysis time. Therefore, as long as the shape
of the peaks and the plate height (length of the column divided by N) are satisfactory, we
should not rely on the number of plates for optimization, unless as a last resort. Methods
which may be used to optimize the chromatographic system with respect to the required
number of plates will be described in chapter 7.
The third factor in the resolution equation is the most vital one for the optimization of
the separation. Since this factor involves the selectivity (a)we may talk about “Selectivity
optimization”. We have seen in section 1.5 that R, is very sensitive even to small changes
in a if the components are difficult to separate (i.e. a close to 1).
Although the primary parameters will usuallly affect the selectivity to some extent, we
may not use them to vary a,because they will affect the capacity factors to a much greater
17
extent and push them out of the optimum range. Therefore, we usually need to look at
different parameters for selectivity optimization.
The parameters that may be used for optimizing the selectivity in various chromatograp-
hic techniques may be referred to as secondary parameters. They will be discussed in
chapter 3 and summarized in table 3.10.
6 . Optimization of the chromatographic system
Once we have realized optimum capacity factors and optimized the selectivity, we can use
the resolution equation to calculate the number of plates that is required to achieve
baseline resolution (R,=1.5). The required number of plates will to a large extent
determine the kind of column and instrumentation needed to perform the separation. This
will be briefly discussed in chapter 7.
1.7.2 Method development in the laboratory
After we have discussed the development of chromatographic methods, this final section
of chapter 1 will discuss the role of method development in the modem laboratory. We
have seen that in developing methods we should aim at simple, rapid analyses. Program-
med analysis in a routine situation should be avoided whenever possible and a high degree
of automation should be feasible.
To achieve this goal, method developers should ideally find themselves in the opposite
situation, being equipped with flexible, advanced instrumentation, including a variety of
possible injectors and detectors, facilities for temperature or solvent programming, etc..
Multichannel detectors are very useful, as they may be of assistance in recognizing the
different sample components when they move about in the chromatogram during the
selectivity optimization process (section 5.6).
Method Development
Advanced. flexible instrumentation Level 1
Various injectors, detectors
Sophisticated data handling
Multichannel detectors
Figure 1.9: 1llustration.of three instrument levels in a modern laboratory handling a large number of
chromatographic analyses.
18
Figure 1.9 shows a scheme of a large laboratory in which three levels of instrumentation
may be distinguished. The first level is the method development level, where the
instruments are equipped with many features and options. This provides the method
developer with a high degree of flexibility. At level two analyses are performed which
require the use of special instrumentation, such as programming or column-switching
options, sophisticated detection systems, etc.. The third level is that of simple dedicated
equipment, which is ideally used for all routine analysis.
It should be the aim of the method developer to try and develop methods which allow
the use of level 3 instruments instead of level 2 ones as much as possible. If this can be
achieved, the money spent on sophisticated instrumentation for method development will
not be wasted.
REFERENCES
101. L.R.Snyder, J.Chromatogr.Sci. 10 (1972) 369.

102. G.Guiochon, Adv.Chromatogr. 8 (1969) 179.
103. L.R.Snyder and J.J.Kirkland, Introduction to Modern Liquid Chromatography, 2nd
Edition, Wiley, New York, 1979, p.54.
104. J.C.Giddings, AnaLChern. 39 (3967) 1027.
105. J.M.Davis and J.C.Giddings, AnaLChern. 55 (1983) 418.
19
CHAPTER 2
SELECTION OF METHODS
In this chapter we will first classify the different chromatographic techniques on the
basis of the mobile and stationary phases used (section 2.1). This section introduces some
of the terminology and abbreviations used in the remainder of the book.
Section 2.2 provides a rough scheme for the selection of an appropriate chromatograp-
hic technique. This scheme can be used if only a few physical characteristics of the sample
are available and if the literature has not provided a starting point for further optimization.
Section 2.2.1 briefly addresses the possibility to incorporate a scheme for the selection
of chromatographic methods in a computer program, a so-called expert system. This is a
relatively recent proposition, and progress may be expected in this area.
Section 2.3 describes various methods for the characterization and classification of
mobile and stationary phases for chromatography. In section 2.3.1 the solubility
parameter is introduced as a quantitative definition of the word "polarity". Section 2.3.2
describes the characterization of GC stationary phases according to Rohrschneider and
section 2.3.3 the classification of LC solvents according to Snyder. The applicability of the
different methods is summarized in section 2.3.4.
2.1 CLASSIFICATION OF CHROMATOGRAPHIC TECHNIQUES
The most common way to classify the different chromatographic techniques is by the
nature of the two phases involved. The mobile phase can be a gas (gas chromatography,
GC), a liquid (liquid chromatography, LC) or a supercritical fluid (SFC)*.
The nature of the stationary phase can be incorporated in this nomenclature. The
convention hereby is to denote the character of the stationary phase by inserting one or
two letters in the middle. Hence, LSC and LLC are both forms of liquid chromatography
(LC; mobile phase is a liquid), while the stationary phase is a solid (S) and a liquid (L),
respectively.
Chemically bonded phases (CBP's) are very commonly used in LC, and occasionally
also in GC. Such phases cannot be seen as either a solid or a liquid. The common term
I2011 used for LC involving such phases is bonded phase chromatography (BPC)**. To
be consistent, the stationary phase identification should follow that of the mobile phase
in defining t h e chromatographic system. Hence, LBPC should be used for liquid
chromatography using chemically bonded stationary phases.
A summary of the different chromatographic techniques is given in table 2.1.
* A supercritical fluid is a substance above its critical pressure and temperature.

** (L)BPC is not thesame as reversed phase chromatography (RPLC). RPLC can bedefined as a form
of liquid chromatography in which the mobile phase is more polar than the stationary phase, a
situation that does not occur in GC. RPLC will be discussed extensively in section 3.2.2.1.
20
Table 2.1:
Classification of chromatographic techniques
Stationary Mobile Phase

phase
Gas Liquid Supercritical
- (G) (L) Fluid (SF)
Undefined GC LC SFC
Chemically
bonded (BP) GBPC LBPC (S)FBPC (1)
(1) Supercritical fluid solid chromatography (SFSC) may be abbreviated to fluid-solid chromato-
graphy (FSC), etc.
In this book the attention will mainly be focussed on the most popular chromatographic
techniques, i.e. G C and LC. Some comments will be made regarding SFC in section 3.4.
2.2 SELECTION OF CHROMATOGRAPHIC METHODS
Before we can inject a sample into a chromatograph, we should be able to decide upon
which of the above chromatographic techniques is suitable for the separation problem at
hand. Clearly, this requires some information about the sample. Completely unknown
samples may require a combination of different techniques. For example, an unknown
liquid may contain a volatile fraction that can be anaiysed by G C and a non-volatile
fraction, for which LC is required. In many cases, we know something about the sample,
enough to decide which of the many possible analytical techniques can be applied. A
scheme to decide on the appropriate chromatographic technique based on the nature of
the sample is shown in figure 2.1.
The first question to be answered is whether or not the sample is volatile enough to be
analyzed by GC. G C columns are currently available with upper temperature limits of
around 350 "C. Hence, compounds should be sufficiently volatile at this temperature to
be analyzed by GC. A second requirement for the sample, which becomes the more
relevant the higher the temperatures used, is the thermal stability of the sample, both in
the column, as well as in the injector, which in conventional GC is operated at a
temperature slightly above that of the column*. Because of the limited stability of organic
substances at higher temperatures, extremely high temperatures d o not seem to be very
* Cold (on-column) injectors can be used in capillary GC for the analysis of relatively high boiling
(low volatile) solutes, so that only the upper temperature limit of the column itself remains relevant.
21
Y
/I
P
9 *tile?
n t
U U
Figure 2.1: Scheme for the selection of chromatographic techniques. For explanation see text.
useful in GC. The solid adsorbents that may be used up to such very high temperatures
aggravate the problem of thermal lability by their catalytic activity. Therefore, the
majority of all organic substances is not compatible to GC.
For the volatile samples we then have a choice between GSC and GLC. For all but the
permanent gases, which possess a very high intrinsic volatility and need strong specific
adsorption sites to be sufficiently retained, GLC is usually the preferred method.
For those samples that are not compatible with GC, the first question to ask involves
the size (molecular weight) of the solute molecules. Their size should be compared to the
pores of the packing materials that can be used in LC. If the size of the molecules is not
negligible relative to the (average) pore size, then part of the pores and hence part of the
stationary phase present in the column will not be accessible to the solute molecules.
Hence, the simple relationship between chromatographic retention and thermodynamic
distribution (eqn.l.6) loses its significance. To avoid that, wide pore materials can be used
for the separation of large molecules (e.g., proteins) based on their distribution over the
t w o phases [202].
The effect of limited penetration of the pores by the largest molecules may also be
applied beneficially for the separation of very large molecules. Depending on the size of
the molecules (in solution), they will be more ore less excluded from the pores, and hence
the retention times will be affected. This effect is used in size exclusion chromatography
(SEC) or gel permeation chromatography (GPC). In this technique, any interactions
between the solute molecules and the stationary phase are purposefully avoided. The solute
molecules remain exclusively in the mobile phase, but the accessible mobile phase volume,
and hence the retention volume, may vary between the total volume of the mobile phase
and the so-called exclusion volume, which is the total volume of mobile phase outside the
pores. The latter elution volume applies to very large solute molecules (excluded solutes),
22
which will therefore appear first in the chromatogram. SEC (GPC) is the preferred method
if the solute molecular weight exceeds about 3000 (see for instance ref. [201], p.270).
Because SEC (GPC) is a separation technique based on the size of the molecules, rather
than on (thermodynamic) selectivity, it will not receive further attention in this book.
For the samples that will be subjected to other (so-called interactive) LC techniques, the
next question involves the nature of the solvent in which the sample has been or can be
dissolved. If this is a non-polar solvent, such as n-hexane, then the sample solution is
compatible with Normal Phase LC (NPLC), in which mobile phases with a relatively low
polarity are used in combination with more polar stationary phases (see section 3.2.3). In
this form of chromatography solid adsorbents (such as silica or alumina) may be used as
stationary phases (LSC). Alternatively, polar chemically bonded stationary phases may be
used (see section 3.2.2).
If the sample solvent is polar, then the question needs to be answered whether or not
the sample is ionic. If this is not the case, then Reversed Phase LC (RPLC) should be
applied. In this technique polar mobile phases are combined with less polar stationary
phases, typically chemically bonded hydrocarbon chains (section 3.2.2). If the sample is
ionic, then it matters whether the sample consists of strong or of weak ions. Strong ions
will have to be separated on the basis of their behaviour as ions, either by an ion-exchange
mechanism (using a stationary phase that contains ionic groups) or by ion pairing LC (in
which an ionic reagent is added to the mobile phase). These techniques are described in
section 3.3. For weak ions we have a choice. They may be separated as non-charged
molecules by RPLC, if the ionization can be supressed by a suitable choice of the pH.
Alternatively, they may be separated as charged ions, using either of the two methods
described above.
If there is no information on the ionic nature of the sample, it may be subjected to RPLC
first. If the sample consists of ionic solutes under the conditions of the separation, which
is usually manifested by low retention times and therefore insufficient separation, then
positive or negative ion-pairing reagents (counter ions) may be added to enhance the
retention. The advantage of this strategy is that the same column may be used for RPLC
as for ion-pairing chromatography.
The merits of all the different methods involving the distribution of solutes over a
selective system of a mobile and a stationary phase are discussed in the next chapter.
2.2.1 Expert systems
A scheme for the selection of phase systems such as figure 2.1 is based on logical
reasonings and on experience. In order to obtain guidance on how to go about the
chromatographic separation of a new kind of sample, the alternative for going through
such a scheme is to study the literature or to ask an expert. One may try to formalize the
knowledge and experience of the expert and combine it with information from the
literature into a computer program called an “Expert System” [203].
Some steps towards the creation of such an expert system for the selection of phase
systems in chromatography have recently been taken. Karnicky et al. have reported on
attempts to build such a system for LC [204]. The selection of the most appropriate phase
system for a chromatographic separation is a complicated matter. The choice will be
largely determined by the characteristics of the sample (see figure 2.1) and by the analytical
23
demands. However, secondary factors, such as the availability of columns and hardware,
as well as the background, experience and personal preferences of the user may also play
an important role. Because of this complex, partly subjective style of decision making, the
development of an expert system for the purpose of phase selection in chromatography
is a difficult task.
According to Klaessens et al. [203] a program must be able to “solve substantial
problems” in order to deserve the qualification “expert system”. Although the substantial
problems are definitely there to be solved, and hence expert systems may be of use in
chromatography, there is no program good enough yet to deserve the name.
Even if they prove successful, expert systems will not take away the need for
optimization of selectivity. Rather, they may be complementary in that they may provide
the platform for appropriate initial experiments, from which the optimization procedure
may take off.
2.3 CHARACTERIZATION AND CLASSIFICATION METHODS
2.3.1 Polarity; Solubility parameters
Polarity is a key word in many discussions about chromatography, chromatographic

selectivity and chromatographic separations. Nevertheless, it is often unclear what exactly
is meant by the word. According to Rohrschneider, the polarity of a stationary phase in
GC may be measured from its ability to retain polar compounds [205]. Although this
appears to be a good description from the point of view of the practical (gas-)chromato-
grapher, it is not a sound definition in the sense that it leaves us with the question of
defining what is a polar solute.
Clearly, an objective and quantitative measure for the polarity of compounds of
chromatographic interest is needed. Such a quantitative measure may be found in the
solubility parameter introduced by Hildebrand [206]. The solubility parameter is defined
as the square root of the cohesive energy density (c):
S2= C = -E/v (2.1)
where Sis the solubility parameter, E the cohesive energy and v the molar volume. All the
interactive forces between the different molecules of a species contribute to the cohesive
energy. Hence, molecules which exhibit strong interactions give rise to high values of E.
Because ofthe definition of E(relative to the ideal gas phase) it is a negative quantity. Table
2.2 shows some examples of solubility parameters for compounds of chromatographic
interest.
The solubility parameter is commonly expressed in units of ca11/2cm-3’2.One such unit
corresponds to 2.05 lo3 Pa”’. In the remainder of this book, the units of the solubility
parameter will usually be omitted for reasons of convenience.
In table 2.2 values are given for a variety of materials, including both typical solvents
and typical stationary phases. The inclusion of the latter involves some rigorous
assumptions, because the simple definition above (eqn.2.1) bears no relevance for solid
adsorbents. Nevertheless, by looking at table 2.2 the usefulness of the solubility parameter
as a quantity to describe polarity in quantitative terms becomes instantly apparent. The
24
Table 2.2:
Some examples of solubility parameters for compounds of chromatographic interest
Compound 6 ref.
(cal'%m - 3'2)
Water 25.52 207

Alumina
Methanol
-15.85
16 208
207
Pyrocarbon N 14 208
Acetonitrile 13.14 207
Methylene chloride 10.68 207
1,CDioxane 10.65 207
Tetrahydrofuran 9.88 207
Toluene 9.57 207
Ethyl acetate 9.53 207
Alkanes -7 207
Perfluorinated alkanes -5 206
table reflects the chemist's intuition ,with water being the most polar solvent at the top,
and alkanes as common non-polar solvents at the bottom. Only perfluorinated alkanes are
less polar than alkanes according to the solubility parameter model.
An important aspect that deserves attention is the extreme polarity of water. Its
solubility parameter is around 25, while all other compounds in table 2.2, polar or
non-polar, are found between about 5 and 16.
Besides showing the usefulness of the solubility parameter for the quantification of
polarity, table 2.2 also illustrates the shortcomings of the model. On the basis of its
solubility parameter alone, methylene chloride will be expected to behave quite similar to
dioxane, and toluene similar to ethyl acetate. However, in both cases there are considerable
differences between the solvents in practice. For example, dioxane is miscible with water
in all proportions, while methylene chloride is virtually insoluble in water. Clearly, to
account for differences in behaviour between compounds of similar polarity a refinement
of the model is needed.
At the basis of such a refinement should be the different interactions which together
constitute the total cohesive energy. These can be classified as follows:
- Dispersion interaction arises from the mutual interaction of the electron clouds of
interacting molecules. Since every molecule has such a cloud of electrons, all compounds,
polar or non-polar, exhibit dispersion interaction.
- Dipol orientation reflects the energy gain when the negative end of an electric dipole
4h
is surroun ed by positive poles of similar or other molecules. Since a permanent dipole
moment is required, not all molecules can participate in this kind of interaction. For
instance, the non-polar alkanes do not possess a permanent dipole moment.
- Dipole induction interactionoccurs when a permanent dipole induces a temporary dipole
in a neighbouring molecule that does not necessarily possess a dipole moment of its own.
- Acid-baseinteractionscan be seen as a combination of all the processes in which electrons
25
or protons (hydrogen ions) are transferred between molecules. Such interactions can only
occur if both functions are present, i.e. one molecule (or functional group) must act as an
acid, another as a base. Again, completely non-polar molecules cannot take part in any
such interactions.
The different types of interactions that may occur in three different kinds of mixtures
are indicated in table 2.3. Non-polar (or apolar) compounds are those in which dispersion
interaction forms the only contribution to the cohesive energy. Clearly, the interactions
that occur in apure non-polar compound are equivalent to those in a mixture of two such
compounds.
Table 2.3:
Summary of specific interactions occurring in different kinds of mixtures.
Non-polar/non- Polar/ Polar/polar

polar non-polar (polar compounds)
(non-polar
compounds)
Dispersion +
Orientation -
Induction -
Acid-Base -
We may now write the total cohesive energy density as the sum of a series of
contributions, each due to one of the interactions described above:
where the subscripts refer to total (7),dispersion (4,orientation (o), induction (ind), acid
(a) and base (b). Two different kinds of terms appear in eqn.(2.2). Quadratic terms
represent dispersion and orientation interaction. For induction and acid-base interaction
double product terms occur. The first two types of interactions are called symmetrical,
because the two participating molecules are equivalent. In the latter two types of
interactions, the different molecules play diffeFent roles. Either one molecule actively
induces a dipole in another passive molecule, or one acts as an acid and the other as a base.
Table 2.4 lists the individual contributions or partial polarities for the solutes that
appear in table 2.2. From this table, it is clear that a distinction can now be made between
molecules of similar overall polarity. Much of the cohesive energy of toluene is due to
dispersion interaction, whereas dipole orientation is more important in ethyl acetate.
Orientation interaction is of more relevance in methylene chloride than it is in dioxane,’
which shows a considerable contribution from induction interaction.
26
Table 2.4
Some examples of partial solubility parameters (call%m -3'2) for compounds of
chromatographic interest [209].
Compound Partial polarities (cal'%m -3/2)
Water 25.52 7.2 ? ? 21.7 14.2

Alumina N 16 10.8 9.8 ? 11.4 2.5
Methanol 15.85 7.2 3.9 0.1 17.1 5.4
Pyrocarbon - 14 14 small small
0.2
small small
4
Acetonitrile 13.14 7.3 5.8 10
Methylene chloride 10.68 8.0 4.38 1.o 4 1
1,4-Dioxane 10.65 8.10 1.2 3.4 small 2.0
Tetrahydrofuran 9.88 8.0 3.3 0.3 6.2 1.5
Toluene 9.57 8.5 0.8 0.7 8 0.4
Ethyl acetate 9.53 7.6 3.6 1.4 3.5 1
Alkanes -7 N 7 0 0 0 0
Perfluoroalkanes - 5 N 5 0 0 0 0
2.3.2 The Rohrschneider characterization scheme
Rohrschneider [205,210] has developed a scheme for the characterization of stationary

phases for gas chromatography. The scheme is based on the retention index (I). The
retention index is a dimensionlessretention parameter, designed to be independent of flow
rate, column dimensions and phase ratio. The retention index of a solute is defined as 100
times the number of carbon atoms in a hypothetical n-alkane, which shows the same net
retention time as that solute. This definition is illustrated in figure 2.2. By plotting the
logarithm of the net retention time against the number of carbon atoms in n-alkanes, a
straight line is obtained. The net retention time for a solute may then be located on the
vertical axis, and the retention index found on a horizontal scale, which represents 100
times the scale for nc
The definition of the retention index does not necessarily require that the calibration
line in figure 2.2 is linear. However, the observed linear relationship allows us to express
the retention index in a simple mathematical formula:
(2.3)
where ?;,,is the net retention time of the solute x, t i s nthe net retention time of the n-alkane
that elutes just before the solute, and t;,,+, the net retention time of the n-alkane that
elutes just after the solute.
The retention index can be obtained experimentally in a straightforward manner, can
be measured accurately and is relatively robust with regards to small deviations in the
temperature [210]. In the Rohrschneider scheme, Z values are measured at 100 "C.
27
/
I
nc - 1
I
I
I
8
500 600 700 800

I -
Figure 2.2 Illustration of the definition of the retention index in GC. ncis the number of carbon atoms
in n-alkanes, 1 the retention index and t i the net retention time.
Rohrschneider assumed that the polar interaction of stationary phases can be characte-
rized by substracting the retention indices observed on a completely non-polar phase from
those obtained on a polar phase. As a completelynon-polar stationary phase, squalane was
selected. For the polar interactions of a solute on a polar stationary phase we may now
write:
(2.4)
Because squalane is not entirely non-polar, (small) negative values for AZ are sometimes
obtained for very non-polar stationary phases [211].
Rohrschneider further assumed that the polar contribution to the retention index can
be described by a simple linear equation, involving a series of constants which are solely
dependent on the solute (a to e) and a series of constants which depend solely on the
stationary phase (x. y, z, u and s)*. The following equation may now be applied:
AZ=a.x+ b-y+c.z+d.u+e-s (2.5)
* The lack of alphabetical order in the stationary phase parameters is due to the fact that three
parameters (x,y and z) were initially thought to be sufficient [205]. At a later stage, Rohrschneider
found the need for two additional parameters [210].
28
The Rohrschneider scheme has no significance, untii the parameters for five different
solutes (or five different stationary phases) are known. With fewer parameters, the system
is undefined, while more than 25 known parameters create a problem of consistency.
Realizing that the characterization scheme is completely empirical, Rohrschneider made
a very convenient choice for the characterization of stationary phases. The probe solutes
and their parameters are listed in table 2.5.
The stationary phase parameters may now readily be obtained from the retention
indices of the five test probes, using the following equations:
= "methyl ethyl ketone /loo
'= "nitromethane /loo
s = AZpyridine1100 (2.10)
Hence, stationary phases can be characterized very quickly by measuring the retention
indices of the five probe solutes. On the other hand, the characterization of solutes is not
so easy, for a combination of reasons. In the first place, a set of five equations with five
unknowns has to be solved. In the second place, the retention indices of the solute need
to be obtained on five different stationary phases with known Rohrschneider constants,
as well as on squalane. Hence, six different columns are needed. Moreover, in order to
obtain reproducible data, very careful experimentation is required. It is especially difficult
to maintain a squalane column. In this light, the choice of squalane as a reference phase
has been unfortunate. Therefore, the Rohschneider scheme has become extremely popular
for the characterization of stationary phases, and not for the characterization of both
phases and solutes, allowing the prediction of retention indices through equation 2.5.
Some representative examples of GC stationary phases with their Rohschneider
constants are shown in table 2.6.
Table 2.5:
Rohrschneider probes and their parameters. Retention indices on squalane taken from ref.
[212]. Temperature: 100 O C .
Benzene 100 0 0 0 0 649

Ethanol 0 100 0 0 0 3 84
Methyl ethyl ketone 0 0 100 0 0 53 I
Nitromethane 0 0 0 100 0 457
Pyridine 0 0 0 0 100 695
Table 2.6:
Rohrschneider constants for some typical stationary phases for GC. Data taken from
ref. [212].
Stationary phase %Ph (1) x Y Z U S
Silicone SE-30 - 0.16 0.20 0.50 0.85 0.48

Silicone OV-1 0 0.16 0.20 0.50 0.85 0.48
Silicone OV-101 0 0.16 0.20 0.50 0.85 0.48
Apiezon L - 0.32 0.39 0.25 0.48 0.55
Silicone OV-3 10 0.42 0.81 0.85 1.52 0.89
Silicone OV-7 20 0.70 1.12 1.19 1.98 1.34
Dinonylphthalate - 0.84 1.76 1.48 2.70 1.53
Silicone OV-11 35 1.13 1.57 1.69 2.66 1.95
Silicone OV-17 50 1.30 1.66 1.79 2.83 2.47
Silicone OV-22 65 1.58 1.80 2.04 3.27 2.59
Silicone OV-25 75 1.76 2.00 2.15 3.34 2.81
Carbowax 4000 - 3.22 5.46 3.86 7.15 5.17
DEGS(2) ~
- 4.93 7.58 6.14 9.50 8.37
ODPN(3) - 5.88 8.48 8.14 12.58 9.19
TCEP (4) - 6.00 8.71 7.94 11.53 9.40
(1) Percentage of phenyl groups in methylsilicone polymer in OV-series

(2) Diethyleneglycol succinate
(3) /?, /3’ -0xydipropionitrile .
(4) 1,2,3-Tris(2-cyanoethoxypropane)
Table 2.6 shows that some stationary phases show exactly the same Rohschneider
constants. The fact that these phases also show identical chromatographic selectivity in
practice forms an indication for the validity of the Rohrschneider approach. Any of these
phases can be selected, but it would be a waste of time to investigate the selective effect
of more than one of them. The choice will now be based on secondary considerations, such
as stability, temperature range, availability and cost. Indeed, one of the first consequences
of the Rohrschneider characterization scheme was for some manufacturers to reconsider
the program of available stationary phases and to remove obsolete ones (see e.g. ref. [212],
p.62). It is seen that not all non-polar stationary phases are identical, and that minor
differences in selectivity may be anticipated from the use of Apiezon L instead of one of
the silicone polymers.
Table 2.6 shows a series of silicone polymer stationary phases from the OV series. These
are dimethylsiloxane polymers, with a varying percentage of the methyl groups replaced
by phenyl groups. It turns out that the Rohrschneider constants closely follow this increase
in phenyl group percentage. The constants therefore appear to be a reliable indication of
the polarity of the stationary phase.
Finally, some truly polar low molecular weight liquid stationary phases are listed in table
2.6. Initially, it was difficult to synthesize stationary phases of high polarity but yet high
molecular weight (implying low volatility and hence a high temperature limit). In recent
30
years, an increasing number of polar stationary phases with a good temperature stability
has become available.
The Rohrschneider constants obtained for a large number of stationary phases may
serve as the basis for a statistical analysis, from which a limited number of stationary
phases which are most likely to yield different selectivity effects can be selected.
We may conclude that the Rohschneider system allows us to:
1. Classify stationary phases for GC according to their polarity and speczjic interactions.
2.. Compare diyerent stationaryphases, in order to allow a more systematic searchfor the
optimum stationary phase.
3.. Select a limited number of relevant stationary phases for general applications.
The Rohrschneider scheme has been modified by McReynolds [213]. The modifications
include the use of some more convenient test probes (for instance, ethanol often yields very
low retention indices, nitromethane often yields poorly shaped peaks), the use of a
somewhat higher temperature and a factor of 100to avoid decimal points in the stationary
phase parameters. The McReynolds probes are shown in table 2.7. Considering these
modifications, I feel that no justice is done when “McReynolds constants” are tabulated.
It appears to be more appropriate to refer to these constants as “Modified Rohrschneider
constants”, or “Rohrschneider constants, modified according to McReynolds”. An
additional set of five probes may be used according to McReynolds, but the five extra
parameters are not very helpful for characterization purposes.
2.3.3 The Snyder solvent classification scheme
A convenient way to classify solvents of chromatographic interest in terms of their

polarity and the specific chemical interactions is the empirical scheme proposed by Snyder
[214,215]. This scheme is based on experimental (gas chromatographic) distribution
coefficients for three test solutes (“probes”) on a large number of stationary phases, which
were published by Rohrschneider [216]. The probe compounds are ethanol (e),1,Cdioxane
(6) and nitromethane ( n ) . The experimental values for the distribution coefficients
undergo several empirical modifications:
1. The distribution coefficients are corrected for the solvent molecular weight through
multiplication by the molar volume of the solvent (VJ:
Table 2.7:
McReynolds probes and their parameters. Retention indices on squalane taken from ref.
[211]; temperature: 120 “C.
Solute a b C d e Lalane
Benzene 1 0 0 0 0 653
n-Butanol 0 I 0 0 0 590
2-Pentanone 0 0 1 0 0 627
Nitropropane 0 0 0 1 0 652
Pyridine 0 0 0 0 1 699
31
K b = K , . v, (2.1 1)
K , is the experimental distribution coefficient and K b the corrected value. This correction
is required, because any measure for the interactions that occur in certain solvents should
be more related to the ratio of mole fractions than to the ratio of concentrations of the
solute in the liquid phase and in the gas phase. We may assume the molar volume of the
gas phase to be constant and hence irrelevant if our purpose is a classification of solvents.
However, the molar volumes of solvents vary a great deal. The K , values for n-octane in
various hydrocarbon solvents vary up to a factor of 3.9 between cyclohexane and squalane
[216]. The Kb values vary by a more realistic factor of 1.5 [214].
2. A correction is made for the molar volume of the solute:
log KL = log Kh - log K, (2.12)
where K, is the distribution coefficient of a hypothetical n-alkane with the same molar
volume as the solute. K, is approximated by
log K, = (vJ163) log K O (2.13)
where K Ois the K b value for n-octane, vx the molar volume of the solute and the constant
factor 163 represents the molar volume of n-octane. Eq.(2.13) implies that a straight line
through the origin is assumed for a plot of log Kb vs. vx for n-alkanes. This is a reasonable
approximation, but not exactly true. Therefore, the K;Zvalues obtained for n-alkanes other
than n-octane will not be equal to zero. However, the compilation of Rohrschneider [216]
does not allow a more precise estimate of K, to be made.
After the corrections described above to transfer the K , values into KZ values, Snyder
suggests the following definitions for four classification parameters:
Polarity
P' = log K;,, + log KEd + log KZn (2.14)
where e, d and n denote ethanol, dioxane and nitromethane, respectively.
Proton acceptor parameter
x, = log K&jP' (2.15)
Proton donor parameter
xd = log K:,d/P' (2.16)
Strong dipole parameter
x, = log K%,/P' (2.17)
32
Table 2.8:
Solvent classification parameters according to Snyder. Data taken from ref. [215].
P' xf? xd X" Group
Ethanol 4.3 0.52 0.19 0.29 I1

Dioxane 4.8 0.36 0.24 0.40 VIa
Nitromethane 6.0 0.28 0.3 1 0.40 VI I
Methanol 5.1 0.48 0.22 0.3 1

n-Propanol 4.0 0.54 0.19 0.27
i-Propanol 3.9 0.55 0.19 0.27
n-Butanol 3.9 0.59 0.19 0.25
t-Butanol 4.1 0.56 0.20 0.24
i-Pentanol 3.7 0.56 0.19 0.26
n-Octanol 3.4 0.56 0.18 0.25
Benzene 2.7 0.23 0.32 0.45 VII

Toluene 2.4 0.25 0.28 0.47 VII
Chlorobenzene 2.7 0.23 0.33 0.44 VI I
n-Hexane 0.1 (1) (1) (1)

Isopropyl ether 2.4 0.48 0.14 0.38
Dichloromethane 3.1 0.29 0.18 0.53
THF 4.0 0.38 0.20 0.42
Chloroform 4.1 0.25 0.41 0.33
Acetonitrile 5.8 0.3 1 0.27 0.42
Water 10.2 0.37 0.37 0.25
(I) Irrelevant, because of low P' value.

(2) Close to group VIII.
Table 2.8 shows the P' and x values for a number of solvents of chromatographic
interest. The table first shows the values for the three probe solutes. It is clear that the
definitions applied for the x values do not imply that the probes show a value of unity (or
100) for one of the parameters, as was true in the scheme of Rohrschneider (see section
2.3.2). It is therefore wrong to conclude that a compound with a high value for xd closely
resembles dioxane, because in that case dioxane would not resemble itself more closely
than 24%!
It is then clearly demonstrated in the table that the scheme succeeds in classifying
solvents that are chemically similar. Seven additional aliphatic alcohols (besides ethanol)
are shown in the table, and while their P' values vary from 3.4 to 5.1, the values for the
selectivity parameters are virtually identical for all alcohols. Similarly, benzene, chloro-
benzene and toluene appear to show identical selectivity, and hence appear as one class.
The bottom part of table 2.8 shows some solvents of particular interest to LC. It is again
seen that water has a very high polarity. However, the polarity parameter P' ,because it
33
is based on an empirical interpretation of GC data, is less useful as a quantitative measure
for polarity in LC than the solubility parameter S(section 2.3.1). For example, acetonitrile
appears to be more polar than methanol in table 2.8. The reverse order is observed from
the solubility parameters in table 2.2. In the practice of LC, methanol turns out to be more
polar than acetonitrile, and indeed the polarity (solvent strength) of mixed solvents can
be predicted quantitatively from solubility parameters (see section 3.2).
The main strength of the Snyder scheme is for the classification of solvent selectivity.
We have seen from table 2.8 that solvents that are chemically similar yield similar
selectivity parameters. This type of classification can be made on the basis of structural
information alone. However, the Snyder scheme goes one step further, in that it classifies
different chemical classes into a single selectivity group. From the definition equations
(2.14 through 2.17) we see that the three selectivity parameters are correlated by the
equation
x, + X d + x, = 1 (2.18)
and hence, that the polarity (P')plus two interaction parameters characterize a compound
completely. If we disregard the polarity, the selectivity can be adequately depicted in a
two-dimensional figure. This is commonly done in a triangular diagram, as shown in figure
2.3. Figure 2.3a in the top left corner illustrates the interpretation of the figure. For an
arbitrary point in the triangle, it is shown how the values for x, x d and x, can be located
on the axes.
Figure 2.3: Solvent classificationaccording to Snyder.The insert (figure 2.3a) shows how the location
of a solvent in the triangle is related to the valuk for the selectivity parameters x, (eqn.2.15), xd
(eqn.2.16) and x, (eqn.2.17). For identification of classes see table 2.9. Figure taken from ref. [215].
Reprinted with permission.
34
All the common solvents are confined to a limited part of the triangle, as was already
suggested by table 2.8. Figure 2.3 shows eight (partially overlapping) circles, which
represent a classification of 81 studied solventsinto eight classes.The classes are numbered
by Snyder with Roman numerals, and the classification of the solvents is shown in the last
column of table 2.8. Class VI may be subdivided into two subclasses (a and b), but this
subdivision servesno practical purposes [215]. Chloroform is represented by the little circle
in the bottom left of the triangle. Clearly, it is outside the circle for group VIII. Similarly,
the little circle at the top right (representing triethylamine),which is closest to group I, falls
outside the classification. However, apart from these two eccentric ones, all solvents can
readily be classified, and a summary of the resulting classification is given in table 2.9. A
more complete listing of individual solvents into classes is given in ref. [215].
Table 2.9:
Solvent classification according to Snyder [215]. Classes are illustrated in figure 2.3.
Group Solvents
I Aliphatic ethers
I1 Aliphatic alcohols
111 F'yridine derivatives, THF, sulfoxides
IV Glycols, acetic acid
V Dichloromethane, 1,2-dichloroethane
VI a. Aliphatic ketones and esters, dioxane
b. Sulfones, nitriles
VII Aromatic hydrocarbons, halo substituted aromatic hydrocarbons, nitro
compounds, aromatic ethers
VIII Fluoroalkanols, water
It can be seen from table 2.9 that the Snyder classification scheme enables us to classify
a large number of solvents into a limited number of classes. Chemically similar compounds
(e.g., homologues) are found in a single class, but also some very different chemical
functionalities are grouped together. Group VII forms a good example. The important
practical consequence of the classification is that if a certain solvent (e.g., toluene) does
not provide sufficient chromatographic selectivity, it is unlikely that any other solvent in
the same group (VII) will do so. Hence, nitromethane does not need to be tested
experimentally. Instead of choosing a solvent from all 81 entries in Rohschneider's
compilation, the choice is now limited to 8 solvent classes. Hence, what the Rohrschneider
characterization scheme does for the selection of stationary phases for GC is achieved with
the Snyder classification scheme for the selection of solvents for LC.
2.3.4 Summary
We may summarize the conclusions obtained in sections 2.3.1 to 2.3.3 in terms of the
following recommendations:
35
1. The solubility parameter may be used to characterize the overall polarity of compounds
(solutes,mobile and stationaryphases) in chromatography. Zt may also be used to predict
the polarity (solvent strength) of mixtures (see section 3.2).
2. The Rohrschneider scheme may be used to characterize stationary phases for gas
chromatography, to compare direrent stationary phases and to select a limited number
of phases out of a large number.
3. The Snyder scheme may best be used to classifi the many solvents that may potentially
be used in chromatography into a limited number of categories. This classification may
then be used to aid in the selection of solvents which show dizerent selectivity.
REFERENCES
201. L.R.Snyder and J.J.Kirkland, An Zntroduction to Modern Liquid Chromatography,

Second edition, Wiley, New York 1979.
202. F.V.Warren and B.A.Bidlingmeyer, J.Liq.Chromatogr. 8 (1985) 619.
203. J.W.A.Klaessens, G.Kateman and B.G.M.Vandeginste, Trends Anal. Chem.4 (1985)
114.
204. J.Karnicky, T.Schlabach, Sj.van der Wal and S.Abbott, 8th Intern. Symp. Column
Liq. Chromatogr., New York, 1984.
205. L.Rohrschneider, J.Chromatogr. 17 (1965) 1.
206. J.H.Hildebrand, J.M.Prausnitz and R.L.Scott, Regular and Related Solutions, Van
Nostrand Reinhold, New York 1970.
207. R.Tijssen, H.A.H.Billiet and P.J.Schoenmakers, J.Chromatogr. 122 (1976) 185.
208. B.L.Karger, L.R.Snyder and C.Eon, AnaLChem. 50 (1978) 2126.
209. P.J.Schoenmakers, H.A.H.Billiet and L.de Galan, Chromatographia 15 (1982) 205.
210. L-Rohrschneider, J.Chromatogr. 22 (1966) 6.
21 I. The Chrompack Guide to Chromatography; Theory and Applications, Chrompack,
Middelburg, NL, 1981.
212. W.R.Supina, The Packed Column in Gas Chromatography, Supelco Inc., Bellefonte,
PA, 1974.
213. W.O.McReynolds, J.Chromatogr.Sci. 8 (1970) 685.
214. L.R.Snyder, J.Chromatogr. 92 (1974) 223.
215. L.R.Snyder, J.Chromatogr.Sci. 16 (1978) 223.
216. L.Rohrschneider, Anal.Chem. 45 (1973) 1241.
36
CHAPTER 3
PARAMETERS AFFECTING SELECTIVITY

Before starting any optimization process, we should ask ourselves what exactly it is we
want to optimize. In general terms our goal is to obtain the best possible chromatogram
for a particular purpose. In the next chapter we will discuss criteria by which to judge the
quality of a chromatogram. In the present chapter, we will describe the parameters that
influence retention and selectivity, and see which parameters we might consider (or
exploit) during the optimization process.
Where possible, we will derive simple relationships between retention and the relevant
parameters. For reasons of clarity, we will express all equations in terms of the capacity
factor (k).Obviously, the simplest possible equations will be most useful for optimization
purposes. Ideally, we will be looking for linear relationships, since straight lines allow
straightforward interpolation.
At the end of the chapter (section 3.5) we will summarize the relationships that are
recommended for the various parameters in different kinds of chromatography.
3.1 GAS CHROMATOGRAPHY
3.1.1 Gas-liquid chromatography (GLC)
In this form of chromatography retention can readily be expressed in thermodynamic

terms. The definition equation for the capacity factor ( k ) is
where q i is the total quantity of the solute in either phase, Ci the average solute
concentration and Y the total volume of the indicated phase in the column.
If we assume very dilute solutions (as is usually the case in chromatography), we can
write for the concentration in the stationary phase
where xi,sis the mole fraction of the solute in the stationary phase, and p, and M, are the
density and the molecular weight of this phase, respectively.
For the mobile (gaseous) phase we can write
Pi
= l/vi,m = ~
(3.3)
~ R T ’
where vi,,, is the partial molar volume of the solute in the gaseous phase, p i its partial
pressure, and (= p i vi,,, / R T ) the compressibility coefficient. Since we are dealing
with gases at low pressures, we may usually (and definitely for the present purpose) assume
the ideal gas law to be valid. Hence, we may assume to be equal to one.
37
For dilute solutions, we may use Henry’s law to describe the partial pressure of the
solute:
cs
where is the activity coefficient (at infinite dilution) of the solute in the stationary phase
and py is the vapour pressure of the pure solute. Hence, for the ratio of concentrations we
find from eqn~(3.2)to (3.4):
Because this ratio is independent of the concentration of the solute, it equals the ratio of
the average concentrations, and hence
where ns is the total number of moles of the stationary phase present in the column.
From eqn.(3.6) we conclude that there are two solute-dependent factors that affect
retention. In the first place, this is the vapour pressure of the pure solute. p? is a strong
function of the temperature (see below) and therefore, temperature may be used as a
parameter to influence retention. However, the vapour pressure is a pure component
property and it cannot be changed at will. Differences in the vapour pressure of two solutes
(or differences in variation of vapour pressure with temperature) may or may not provide
us with a means to achieve separation. When the vapour pressure is not sufficiently
different, we need to create differences in the second solute-dependent factor.
This is the activity coefficient of the solute in the stationary phase (rf“,). The value of
7 is determined by molecular interactions between the solute and the stationary phase.
Therefore, (chemically) different stationary phases will lead to different values for 7. This
explains the availability of many different stationary phases for GC, many of which show
different selectivity (see section 2.3.2).
Temperature
Eqn.(3.6) provides a good insight into the variation of retention with temperature in
GLC. Both the activity coefficient and the vapour pressure of the solute vary with
temperature in an exponential way. For the activity coefficient we can write
In es= h d R T - s d R , (3.7)
where h , and s E are the (partial molar) “excess” enthalpy and entropy respectively*.
The vapour pressure can be approximated with the common Clausius-Clapeyron
equation:
* The excess quantities measure the deviation of the solution of i in s from the “perfect”solution, for
which Raoult’s law is obeyed, i.e. es = 1.
38
Inpp = h , / R T - s , / R ,
where h, and s, are the molar heat and entropy of vaporization.

If we now combine eqns.(3.6), (3.7) and (3.8) we find
I n k = In T - - h,+ h" + -S E + % + I nRn

s ,
(3.9)
RT R Vrn
i.e. an equation of the form
Ink=lnT+A/T+ B (3.10)
where A and B are constants.

According to this equation, we can expect to obtain a straight line if we plot In (k/ T)
versus 1/ T. This somewhat odd-looking relationship corresponds to the more common
procedure of plotting the logarithm of the specific retention volume ( Vg) against the
reciprocal temperature. Vgis defined as the net retention volume at standard temperature
(0 "C) and per unit weight of stationary phase:
(3.1 1)
Figure 3.1 provides an illustration of the relationship between the specific retention
volume and the temperature.
A great disadvantage of using V, is that ws is not usually known.
Alternatively, we may write eqk(3.10) as
Tln(k/T) = A + BT (3. I Oa)
I I I
1.01
-
L 1 I
26 27 28 29 30 31 32 3
10'IT
Figure 3.1: Example of the linear dependence of the logarithm of the specific retention volume
(eqn.3.11) on the reciprocal temperature. Stationary phase : silicone oil 702. Solutes: n-alcohols
(number of carbon atoms as indicated in the figure). Figure taken from ref. [301]. Reprinted with
permission.
39
(a)
Figure 3.2 Fundamental linear relationship between retention and temperature in GLC according
to eqn.(3.l0b). Stationary phases: (a) SE-30 (silicone oil) and (b) Carbowax 20M (polyethylene-gly-
col). Solutes: n-alkanes (number of carbon atoms as indicated in the figure). Figure taken from ref.
[302]. Reprinted with permission.
or, because k is proportional to the net retention volume Vx,
Tln(VR/T) = A‘ + B’T. (3.10b)
where A’ and B’ are constants.

Figure 3.2 shows an example of a plot according to eqn. (3.10b). The linearity of these
plots is clear.
The stationary phase
The properties of the stationary phase manifest themselves in the activity coefficient in
eqn.(3.6). A very simple expression for the activity coefficient can be obtained from the
concept of solubility parameters (see section 2.3.1). This expression can be seen as a special
form of Hildebrand’s regular mixing rule, and it reads [303].
In cs= ( V / R T )(6i- 6b2 + In ( v i / v s ) + (1 - v / v > , (3.12)
where the subscript i denotes the solute, and s denotes the stationary phase. v is the molar
volume. The first term on the right hand side of eqn.(3.12) is the enthalpic (“regular”)
contribution to the activity coefficient. The other two terms together form the entropic
(“athermic”) part. Unfortunately, this very simple equation does not usually yield
quantitatively reliable results, unless, perhaps, for the interaction of solutes of low-polarity
with phases of low polarity. However, eqn.(3.12) does provide us with a simple means to
explain and understand the role of the “polarity” of solutes and stationary phases in GLC.
Almost always, the enthalpic contribution dominates the right hand side of eqn.(3.12),
in which case we can use the approximate expression
40
In = (vi/R7‘) (Si- SJ’. (3.13)
Clearly, the right-hand side of eqn.(3.13) is always positive. Consequently, it leads to

activity coefficientswhich are above unity, and hence (since the capacity factor is inversely
proportional to y) towards decreasing retention. When the polarity of the solute and the
stationary phase are similar, then only the (small and negative) entropic contribution to
the activity coefficient will remain, y will be close to unity, and retention will mainly be
determined by the pure solute’s vapour pressure.
The vapour pressure, in its turn, is closely related to the boiling point. Indeed, this case
is often encountered in the GLC of low polarity solutes on low polarity stationary phases
(“boiling point separation”).
Because of the occurrence of the excess quantities h , and sEin eqn.(3.9), the coefficients
in eqn.O.10) for the temperature dependence of the retention are a function of the
stationary phase. Hence, every stationary phase may be expected to yield a different
optimum temperature, at which the capacity factors of all sample components fall in the
optimum range. Therefore, to make a fair comparison between two different stationary
phases for a given separation problem*, the (potentially different) optimum temperature
should be established for each of them and the resulting chromatograms should be
compared. The common practice of characterizing (and consequently comparing)
stationary phases at a standard temperature is a very convenient one. Nevertheless, it may
give rise to erroneous conclusions in some cases.
Mixed stationary phases
The choice of a stationary phase is no longer a discrete variable once mixtures of

stationary phases are considered. For binary mixtures the following relationship is usually
observed**
(3.14)
where cp denotes the volume fraction, and the subscripts A and B represent two different
stationary phases. According to eqn.(3.14) a mixture of two stationary phases behaves as
the sum of two individual contributions. The same result may be obtained using a 50150
mixture of the two phases A and B, as with two similar columns (of half the original length)
coated with pure A and pure B, respectively, coupled in series. Indeed, in practice both
methods may have certain advantages.
Figure 3.3 shows an example of a plot of the retention in GLC versus the composition
of the stationary phase. In this and many other cases straight lines are observed. Figure
3.4 shows an exceptional case, in which no straight lines were observed.
Eqn.(3.14) is usually, but not always obeyed in practice. Moreover, it is hard to predict
when deviations will occur. Apparently [306], not only the two stationary liquids involved,
* For some suggestions on how to select different stationary phases see section 2.3.2.
** It should be noted that eqn.(3.14) applies only when the total volume of the stationary phase is
kept constant.
41
'PDNP -
Figure 3.3: Variation of retention (distribution coefficient) with the composition of the stationary
phase in GLC at three different temperatures (indicated in the figure in "C). Stationary phase:
mixtures of squalane and dinonylphthalate (DNP). Solutes: (a) n-octane, (b) cyclohexane, (c)
methylcyclohexane and (d) tetrahydrofuran. Straight lines observe eqn.(3.14). Figure.taken from ref.
500 '\ I'

\'\\
'\f\
L 20
630
K
580 /f ,'
130
0 0.5
PDNP-
0.5
'PDNP -
Figure 3.4 : Variation of retention (distribution coefficient) with the composition of the stationary
phase in GLC. Stationary phase: mixtures of squalane and dinonylphthalate (DNP). Solutes: (a)
cyclohexane, (b) methylcyclohexane, (c) benzene, and (d) toluene. Temperature: 30 OC. Curved lines
are obtained, which do not obey eqn.(3.14). Figure taken from ref. [305]. Reprinted with permission.
42
but also the procedure that is used to form the mixture and to coat the mixture on the
column is relevant in this respect. Nevertheless, eqn.(3.14) may be assumed valid, until
there are indications to the contrary.
After eqk(3.14) turned out to be obeyed by many systems in practice, a model was
developed that could provide a physical picture. This so-called diachoric model [306]
explains the fact that the two components of the mixed phase behave independently by
demixing on a microscopic scale. Hence, the stationary phase is assumed to consist of little
patches or droplets of either pure A or pure B. Obviously, such a model does explain
obeyance of eqn.(3.14), while it also gives a handle to explain deviations from linearity in
terms of complete mixing of the two phases.
It is useful at this point to realize that with the composition of the stationary phase being
a continuous variable and with retention and selectivity being strong functions of
temperature, the optimum composition may also be expected to vary with temperature.
Ideally therefore, temperature and stationary phase composition should be optimized
simultaneously (see section 5.1.1). Moreover, once different lengths of columns with the
individual stationary phases are applied instead of real mixtures, it is in theory feasible
to optimize the temperature of each of the columns, as well as the ratio of column lengths
simultaneously.
3.1.2 Gas-solid chromatography
We can describe the retention behaviour in GSC in similar terms as we did for GLC in
the previous section. However, we have to reconsider our definitions of the concentration
in the stationary phase and of the distribution coefficient. It is common practice to express
the concentration of the solute adsorbed onto the stationary phase in terms of moles per
gram of adsorbent, i.e.
c: = 4JWS 7 (3.1 5)
where w, is the weight of the sorbent present in the column. Therefore, the partition
coefficient
K = -c:- ?-- Vrn
C (3.16)
Cm 4rn Ws
is now expressed in units of ml/g. An expression for Henry's adsorption law can be
formulated that is analogous to eqn.(3.4) [307]:
4, = Hi A , Pi 9
(3.17)
where Hi(units: mol.atm-'.m-*) is Henry's adsorption coefficient for the solute ion the
solid adsorbent. Hi is determined mainly by the interactions between the stationary phase
and the solute, and hence it takes over the role of the activity coefficient in GLC. Note
that the vapour pressure of the pure solute does not appear as an independent entry in
eqn.(3.17). A, is the surface area of the adsorbent. Using eqn.(3.3) with = 1 for the
mobile phase, we find from a combination of eqns.(3.15), (3.16) and (3.17) that
K , = Hi A, R T / w, = HisR T , (3.18)
43
where s is the surface area per unit weight of the adsorbent (m2/g), usually called the
specific surface area. The capacity factor now becomes
(3.19)
and finally for the relative retention:
aI.', . = k / k , = HfH,. (3.20)
Hence, the selectivity in GSC is determined by the values of the Henry adsorption
coefficients of the two solutes.
Eqn.(3.19) describes the ideal case in which the adsorption isotherm of the solute is linear
and the carrier gas does not adsorb onto the stationary phase. This simple situation is not
always encountered, but analytical equations can be derived for many other cases [308].
In fact, the practical conditions in GSC are more often non-ideal than is the case in GLC.
The adsorption isotherm can only be approximated as linear at very low concentrations.
In other words, solute capacities are usually lower in GSC. Surface heterogeneities play
a role, especially on inorganic adsorbents such as silica and alumina. These stationary
phases are also sensitive to contaminations. Consequently, the observed peak shapes and
retention times may be affected by the history of the column ("conditioning") and by the
water content of the carrier gas.
Because the pure component vapour pressure is not one of the determining factors for
retention in GSC (eqn.3.19), GSC is especially useful for the separation of permanent gases
and other solutes of low molecular weight. Even at room temperature the interaction with
the stationary phase may be high enough to retain permanent gases on GSC columns.
Under GLC conditions a reasonable retention for compounds with a very high vapour
pressure is only possible if activity coefficients can be achieved that are very much smaller
than one, and this is not a common situation.
Because of the specific application area of GSC in the world of small molecules, the
number of components to be separated is usually small. For most practical problems,
therefore, specific stationary phases are readily available. Hence, GSC is not the most
fertile soil for selectivity optimization.
Temperature
The temperature of the column is the most important parameter in GSC. Its effect on
retention can be described by the same equations as were used in GLC, as can be seen from
a comparison of eqm(3.6) and (3.19). According to eqn.(3.10) a straight line will be
obtained by plotting In (k/ 7 ) vs. (1 / 7). However, a plot of In k(or In VR)vs. the reciprocal
temperature also yields approximately straight lines, as is illustrated in figure 3.5.
Other parameters
Apart from the temperature, the only other factor that may affect the retention and
selectivity in GSC is the nature and, theoretically,the composition of the stationary phase.
44
Figure 3.5 : Variation of retention (net retention volume per unit surface area of stationary phase)
with the temperature in GSC. Stationary phase: GTCB (carbon). Solutes: o/p-xylene (l), m-xylene
(2), n-octane (3), ethylbenzene(4), toluene (5), n-heptane (6), methylcyclohexane(7), exo-5-methyl-
norbornene (8), endo-5-methylnorborner1e(9),norbornane (10) and norbornene (1 1). Figure taken
from ref. [309]. Reprinted with permission.
The materials that can be used vary from inorganic oxides such as silica and alumina to
organic polymers such as styrene-divinylbenzene copolymers.
The inorganic materials are more stable at higher temperatures. Both silica and alumina
are strongly affected by the water content of the surface. Therefore, the reproducibility and
the repeatability of the separation largely depend on the “conditioning” (equilibration) of
the column. Alumina shows a remarkable affinity to hydrocarbons. For this reason it is
eminently suitable for the separation of volatile hydrocarbon fractions, including alkane
isomers up to about 5 carbon atoms.
Due to both the magnitude of the interactions with the surface, as well as to the surface
heterogeneity, both alumina and silica are less useful for the separation of polar
compounds.
This latter effect is not encountered to the same extent when organic (solid) polymers
are used as the stationary phase. Hence, more or less polar substances with a high volatility
may be eluted from such columns as symmetrical peaks.
Mixed beds of stationary phases in a single column have not found much application
in GSC. However, for the separation of complex gas mixtures, more than a single column
is often used in a configuration that involves switching valves, so that only part of the
sample is subjected to a second column (see also section 6.1).
3.1.3 The use of retention indices
‘The retention index ( I ) was introduced in section 2.3.2 as a reproducible means for
reporting CC retention data. The retention index was therefore found to be highly useful
as the basis for a characterization scheme for stationary phases in GLC. In this chapter,
however, the capacity factor and not the retention index has been used to find expressions,
which describe the influence of the various relevant parameters on the retention. Besides
45
the fact that the use of capacity factors is consistent with other sections in this chapter,
there is also a more fundamental reason behind this choice.
The definition of the retention index in terms of capacity factors is
I = 100 [ n +
In k,-In k,
In k,+ -In k, I- (3.21)
If we use eqn.O.10) for the variation of the capacity factor with the temperature we find
(3.22)
In eqns(3.21) and (3.22) i denotes the solute and n and n + 1 the preceding and the
following n-alkane, respectively (see figure 2.2). It is seen that there is a hyperbolic
relationship between the retention index and the temperature. Although over small
sections of the hyperbola a linear approximation is often used, this is not a sound basis
for temperature optimization, especially not since a straight line can easily be obtained by
plotting In ( k / vs. 11T (eqn.3.10).
An even more complicated result is obtained if we express the retention index as a
function of the stationary phase composition. A combination of eqm(3.14) and (3.21)
(3.23)
(a1
12-
10 -
-
20 LO 60 80 100
% Polybutadiene
0
-
20 LO 60 80
% Polybutadiene
100
Figure 3.6 : Variation of retention with the composition of the stationary phase in GLC. Stationary
phase: styrene-butadiene polymer blends and copolymers, the butadiene fraction is plotted on the
horizontal axis. (a) Specific retention volumes for three n-alkanes and benzene. V is proportional
to the capacity factor. (b) the retention index for benzene. The solid line is calculated from the straight
lines in figure 3.6a. The circles (polymer blends) and triangles (copolymers) represent experimental
data. Figure taken from ref. [310]. Reprinted with permission.
46
Figure 3.6 illustrates the variation of (a) the specific retention volume ( Vg which is
proportional to the capacity factor) with the composition of a mixed stationary phase, and
(b) the variation of the retention index for benzene with the composition. It is clear from
these figures that, whereas straight lines are observed for the variation of the capacity
factor with the composition, the retention index varies in a highly non-linear manner.
Clearly, for the purpose of selectivity optimization, the capacity factor ( k ) is greatly to
be preferred to the retention index (I).
3.2 LIQUID CHROMATOGRAPHY
We will first approach liquid chromatography by assuming that both phases are bulk
fluids (i.e. LLC), and generalize our approach later. For LLC we can define a thermodyna-
mic equilibrium constant ( K t h )as
(3.24)
i.e. the ratio of the solute activities in the two phases. A suitable standard state (at which
by definition a = 1) should now be defined for each phase. As long as we strictly observe
this definition, we are free to choose a convenient one. Hence, we opt for the pure solute
i as standard state for both phases*, since in that case the activities are equal at unit
concentration (1 00% i), and therefore at all concentrations:
(3.25)
and, using the definition for the activity coefficient
(3.26)
For the high dilutions encountered in LC this can be transformed into
(3.27)
where q is the amount of solute i (in number of moles) present in either phase. Finally, by
definition
(3.28)
* 'This definition is equivalent to the one implicitly assumed in Henry's law (section 3.1.1). An
alternative way to arrive at eqm(3.25) is to consider the thermodynamic potential (p).A condition for
equilibrium is that p should be equal in the two phases of a chromatographic system for each
compound. Therefore, we can write for the solute
pi,s = p!s + R T In aLs = pi,m = p!m + R T In ai,m
Eqn.(3.25) follows immediately from this equation, if we again choose for the same standard state
(the pure solute) in both phases (i.e. pys=py,,).
47
The last factor on the right-hand side of eqn.(3.28) is clearly a phase ratio, independent
of the solute. Hence, the activity coefficients determine retention in liquid chromatograp-
hy. In other words, retention is determined by the molecular interactions of the solute with
the stationary and with the mobile phase.
This situation is fundamentally different from the one in GC, where interactions with
the stationary phase and the vapour pressure of the pure solute were the relevant factors
(see section 3.1.1). In LC, both the interactions in the mobile phase and in the stationary
phase can be influenced in order to optimize the selectivity of the system, and neither is
beyond control in the sense that vapour pressure is in GC.
The activity coefficient can be expressed in terms of solubility parameters (eqn.3.12).
Neglecting the (small) entropy correction terms we find
In ki = ( v / R T ) { (ai- 6,J2-(Si - 6J2} + In(n,/n,J

= ( v / R T ) {(6,+6, - 26,)(6,-6,)} + ln(n/n,J. (3.29)
The above equation is very approximate. It involves many assumptions and approxima-
tions [311], and it is not adequate for a quantitative description or prediction of retention
in LC. However, because of its simplicity, it provides us with a very elegant means to
explain many of the features of modem LC in qualitative terms.
Since eqn.(3.29) will not be used quantitatively, we may take some additional liberties
by assuming that it not only provides us with some insight into LLC systems, but may also
be used for the qualitative interpretation of other forms of LC, notably LSC and LBPC
as well. Indeed, this is not a new appproach, and a value for the solubility parameters of
some solid surfaces has been suggested in the literature [303,312].
Retention
According to eqn.(3.29) retention (k) varies exponentially with the polarity of the solute
(ai).It can easily be seen from eqn.(3.29) that as soon as the solubility parameter product
(6, + 6, - 26,)(6, - 6,) becomes significantly positive or negative, the capacity factor
will either be impractically high or uselessly low. Hence, to a first approximation
reasonable values for k will be obtained when the above product is about equal to zero.
This can be achieved in two ways:
1. Choosing the mobile and the stationary phase of roughly the same polarity (i.e. 6, M 6,).
While this has the desired effect on retention, the remedy is equally effective for all
solutes, independent of the value of 6 ,and therefore it creates a very non-selective
phase system (see the discussion on selectivity below).
2. Alternatively, the first factor can be minimized, by taking the solute polarity to be
roughly intermediate between the polarities of the two phases:
Si M (6, + &,)I2 . (3.30)
This very simple rule of thumb for the selection of LC phase systems (a phase system
is a combination of a mobile and a stationary phase) is illustrated in figure 3.7.
In figure 3.7 three horizontal axes have been drawn, representing (from top to bottom)
the solubility parameter of the stationary phase, the solute and the mobile phase. Each
48
1
n-alkylphases
fluoroalkcines
silica
25
calv?cm-3’2
reversed phase straight phase
T metianol 1
alkanes THF acetonitrile water
Figure 3.7: Illustration of the selection of phase systems for LC according to eqn.(3.30). A solute with
a polarity of 12.5 (middle scale) can be eluted from silica (S,= 16; top scale) with a non-polar mobile
phase (Sm=9; bottom scale) or with a polar solvent in a reversed phase system. The shaded areas
indicate the latitude with respect to the selection of the mobile phase. Figure taken from ref. [311].
(non-horizontal) line in figure 3.7 represents a phase system for which eqm(3.30) is obeyed.
Clearly, a vertical line again denotes a system for which the polarities of the two phases
are equal. According tp the above discussion this is a completely non-selective phase
system.
In figure 3.7 two lines have been drawn, which represent two examples of possible phase
systems for the elution of a solute with Si= 12.5 (cal”*.~m-~’~). The line with a positive
slope connects a moderately polar mobile phase (6, = 9) with a polar stationary phase
(a,= 16). Because the polarity of the stationary phase exceeds that of the mobile phase
(a,:> S,), this is by definition a normal (or straight) phase system.
The reverse is true for the line drawn in figure 3.7 with a negative slope (a,< ,a), and
for this reason this system is called a “reversed phase” (RP) system. The particular line in
figure 3.7 connects a typical non-polar (alkane-like) phase with 6,= 7 to a polar mobile
phase with 6,= 18. Such a mobile phase could for instance be created by mixing methanol
( 6 16) ~ with water ( 6 ~ 2 5in) the correct proportions. Since a very wide range of mobile
phase polarities can be covered with mixtures of methanol and water, or even tetrahydro-
furan (THE 6 s 10) and water, the reversed phase system is a very flexible one. Without
changing the column (stationary phase), it can be applied for the elution of a wide variety
of solutes.
Alongside the two lines drawn in figure 3.7 as examples for phase systems, dotted areas
are indicated towards the mobile phase axis. This is done to indicate that, when eqn.(3.30)
is strictly obeyed, a small capacity factor is expected to result. By choosing the mobile
phase polarity slightly further away from that of the solute, the capacity factor can be
moved into the optimum range. However, the margin thus created is usually very small and
49
eqn.(3.30) may serve as a good (albeit qualitative) indication of the behaviour of LC
systems.
Selectivity
From eqn.(3.29) a very simple expression for the selectivity of a phase system may be
derived. For the relative retention (aj.>of two solutes with similar molar volumes ( v i = vj)
we find
In aj.i = In kj - In ki = (2vi/R7')(Si-Sj) (6,-Ss). (3.31)
Hence, if we want to separate a given pair of solutes (i.e. Si and are fixed), the general
approach involves maximizing the only variable left to control in eqn.(3.31), the factor
(6, - as)*.
By definition, the relative retention is larger than (or equal to) 1. Thus, for a normal
phase system, where Ss>S , it followsfrom eqm(3.31) that Sj> 6 ,and hence themore polar
solute will elute last. Again, the reverse is true for a reversed phase system. Becausethe signs
of the two factors in eqn.(3.31) which involve solubility parameters will always be the same,
we may state that it is the absolute difference between the polarities of the two phases that
should be maximized. Therefore, the selectivity of a phase system ( V ) may be defined as
(3.32)
Unfortunately, we cannot just use any combination of phases that would constitute a
very selective phase system. For example, we might want to opt for the combination of an
RPLC column with typically 6,yz7, with pure water (a= , 25.5) as.the mobile phase, which
would result in a selectivity ( V ) of about 18. However, in this particular phase system only
a very polar solute with 6,s16 would satisfy eqm(3.30). It will come as no surprise that
in this particular (highly selective) phase system all but the very polar solutes will have
extremely high capacity factors. In fact, for a given solute, once a given column (stationary
phase) has been selected, the appropriate mobile phase can readily be obtained from
eq~(3.30)(or graphically from figure 3.7). From a substitution of eqm(3.30) in eqn.(3.32)
we find
v = 2 (6, - 6.J . (3.33)
This equation shows that we should ideally select a stationary phase with a polarity that
is very different from that of the solute. Indeed, the recommendation to use normal phase
chromatography (high S,J for non-polar solutes (low Si) and reversed phase chromatograp-
hy (low 6.J for the separation of polar solutes (high Si) is not new. However, this rule of
thumb is much too simple. A complication is caused by the availability of appropriate
mobile phases. For instance, to satisfy eqn.(3.30) for the elution of non-polar solutes
( 4 ~ 7 from
) a silica column (6,yz16), a mobile phase with 6,- - 2 would be required.
* This factor can easily be shown to be of equally great importance for the separation of two molecules
with similar polarities but with different molar volumes.
50
Clearly, this is an impossible proposition. Practical mobile phase polarities will be
restricted to the range between 6,w7 (for alkanes) and 6, = 25.5 (for water).
For some selected stationary phases, the selectivity that can be achieved as a function
of the solute polarity is shown in figure 3.8. Eqn.(3.33) forms the basis for this figure, but
the practical limits for the mobile phase polarity are respected. For example for a reversed
phase column (represented by the line AT in figure 3.8 and denoted by the letters RP) the
maximum solute polarity is just over 16 when pure water is used as the mobile phase. Phases
with intermediate polarity are represented in figure 3.8 by a set of two lines in a V-shape.
The set of lines denoted by LSC represents a typical normal phase adsorption material with
a polarity of around 16 [312]. This stationary phase can be combined with a less polar
mobile phase down to 6 , = 7, yielding the line with a negative slope in figure 3.8. This
branch represents the common application of normal phase adsorption chromatography.
However, the polar adsorbent may also be combined with an even more polar mobile
phase up to a mobile phase polarity of 25.5 and a solute polarity of about 21. Hence, this
line with positive slope in figure 3.8 represents the use of polar adsorbents in the reversed
phase mode for the separation of very polar solutes.
Interestingly enough, according to figure 3.8, the non-polar stationary phase (reversed
phase column) will always lead to a higher selectivity than the more polar stationary phase
(normal phase column), apart from the range for very polar solutes, where the polar
stationary phase is used in the reversed phase mode. Indeed, there has been a recent interest
in separations of this kind on silica, and some polar chemically bonded phases (see section
3.2.2) are especially useful in this respect. The separation of sugars on an amino-type
column is a good example.
The line TW in figure 3.8 is dashed. This is the virtually non-existent situation where
water is being used as the stationary phase. Although this suggestion is not at all practical,
it is clear from figure 3.8 that a very high selectivity could be obtained for polar solutes.
Figure 3.8 :Calculated selectivities according to eqn. (3.33) for various stationary phases as a function
of solute polarity. RPF = perfluorinated reversed phase; RP = reversed phase; PC = pyrocarbon;
LSC = alumina, silica. Figure taken from ref. [31 I]. Reprinted with permission.
51
Therefore, it seems an interesting challenge to try and create stationary phases of a polarity
much higher than that of silica.
There are two other phases indicated in figure 3.8. The first is a so-called pyrocarbon
material. Such a stationary phase is formed by pyrolizing an organic layer on a silica
substrate. The idea is to combine the mechanical strength of silica with the chemical
inertness of carbon. The value of 14 used here can be thought of as typical for
carbonaceous materials. These materials do not seem to behave like non-polar phases in
the tradition of chemicallybonded phases for RPLC, but rather like phases of intermediate
polarity. Hence, as for silica, they may be most useful in the reversed phase mode for the
separation of very polar molecules using aqueous mobile phases.
The thin line in the top left of figure 3.8 denoted by RPF represents a very non-polar
perfluorinated (chemically bonded) stationary phase. Perfluorinated alkanes are known
to behave like even less polar materials than the alkanes themselves 1313). Although it is
theoretically possible to use such materials as a mobile phase (for instance for the
separation of low polarity solutes on a silica column), figure 3.8 suggests that it will be more
rewarding to use perfluorinated materials as the stationary phase. Of course, this
proposition would also be more cost effective. Indeed, such materials have been studied
by several researchers. The general conclusion of these studies turns out to be that there
might be an overall increase in selectivity relative to conventional RPLC systems, but that
this effect is overshadowed by very large specific effects (i.e. selectivity towards specific
solutes)[314].Therefore, perfluorinated materials should be seen as alternative rather than
as superior stationary phases for RPLC.
The behaviour of the perfluorinated phases as discussed above illustrates the fact that
the solubility parameter model, despite its charms, may only be used as a crude
approximation. The occurrence of specific deviations from the general rule may at least
be made plausible by differentiating between different kinds of molecular interactions,
and by introducing so-called partial solubility parameters or partial polarities [303,3121
(see also section 2.3.1). However, such an extension greatly increases the complexity of the
model, without increasing its predictive value correspondingly.
3.2.1 Liquid-liquid chromatography
A liquid-liquid system can be created by coating a particulate matter with a thin layer
of a liquid phase, similar to the way packed columns are used in GLC. To maintain
such an LLC column, the stationary phase should be insoluble in the mobile phase, just
as GLC phases need to be involatile at the temperature of operation. Unfortunately, “in-
solubility” is an absolute demand that can at best be approximated in practice. The
solubility of the stationary phase in the mobile phase becomes even more critical once
some flexibility is desired with regard to the choice of the mobile phase. For example,
mixtures of several pure solvents are usually required in order to adapt the eluotropic
strength (polarity) of the mobile phase such that the capacity factors fall in the
optimum range.
Because complete immiscibility of the two phases cannot usually be accomplished,
practical measures will have to be taken to avoid bleeding of the stationary phase from
the column, for example pre-saturation of the eluent with the stationary phase or the
inclusion of a small coated column (coated with the same stationary phase as the analytical
52
column) in the flow stream before the injector. Basically, both the above remedies are very
similar.
Despite such measures, LLC columns with “insoluble” stationary phases are not very
stable, for instance due to a disturbance of the system every time a sample is injected. The
reproducibility of retention data on these LLC columns is generally unsatisfactory, and
‘‘aging” of the column tends to occur rapidly. Apparently, the stationary phase is not
merely dissolved from the column because of its “solubility” in the mobile phase, but it
may also be eroded from the column because of mechanical processes (shear forces) or
by a solution-precipitation mechanism, which causes the stationary phase to be redistribu-
ted within the column. These effects may be enhanced by temperature changes within the
column, due to viscous ‘heat dissipation and inadequate temperature control. Indeed,
thermostatting of the column (and the eluent reservoir) is vital for the proper operation
of LLC systems.
Another problem associated with LLC is that of mixed retention mechanisms. Ideally,
the solid support in LLC binds the molecules of the stationary phase with strong adsorptive
forces, but it does not exert these forces on solute molecules. Clearly, this ideal situation
can never be realized completely [3151.
For all these reasons, it will be understandable that LLC systems have been virtually
replaced by chemically bonded phases (section 3.2.2) in current LC practice. Consequent-
ly, the various parameters of interest for the optimization of these systems will not be
discussed extensively. With regard to the influence of temperature and mobile phase
composition on retention and selectivity, it is suggested that the same relationships may
be used for “insoluble” LLC stationary phases as are used for LBPC. LLC systems have
been used extensively for the separation of ionic compounds by means of ion-pairing
techniques. Such systems will be discussed in section 3.3.2.
The main parameters in LLC are the polarities of the mobile and the stationary phase.
Increasing the polarity difference between the phases enhances both the selectivity of the
system (figure 3.7) and the stability, due to a reduced mutual solubility of the phases.
In LLC systems there is not a substantial difference between the selectivity characteris-
tics in the normal phase and the reversed phase mode. The choice of either will mainly be
determined by the sample. Polar samples (in polar solvents) will preferably be injected in
a reversed phase system and non-polar samples in a normal phase system.
Within the framework of the given polarity of a phase, its composition may still be varied
for optimization purposes (see the discussion about iso-eluotropic mixtures in section
3.2.2). However, the mutual solubility of the two phases is not only determined by their
polarity, so that changes need to be considered carefully. In conventional LLC systems,
changes to the stationary phase are hard to make, because they may require a lengthy
“re-coating’’ procedure.
Dynamic LLC systems
A promising way to create LLC systems with sufficient stability is the use of immiscible
ternary mixtures to create what is called a “dynamic” (or “solvent-generated”) LLC
system. The principle of such a phase system is illustrated in figure 3.9. This figure shows
an example of a thermodynamic phase diagram of a mixture of three components (A, B
and C). Both the binary mixtures A + B and A + C are miscible in all proportions.
53
However, this is not the case for solvents B and C and therefore there is a range of binary
and ternary compositions at the bottom of figure 3.9 where two liquid phases are formed.
If the three solvents are brought together in proportions that correspond to a composition
that is situated inside this area, such as the point indicated by a dot in figure 3.9, then two
liquid phases will be formed according to the “nodal” line through this point. The position
of the point on the nodal line will determine the ratio of the amounts of the two immiscible
phases formed.
The two liquids thus formed are immiscible, but in thermodynamic equilibrium.
Therefore, we may speak of a dynamic system of two immiscible phases. Figure 3.10 shows
an example of a practical system applied to create a dynamic LLC phase system. A
practical phase system can be created by pumping a mobile phase through a column, the
composition of which corrresponds to a ternary mixture that is in dynamic equilibrium
with another mixture (the two mixtures can be connected by a nodal line). If the mobile
phase is the more polar one of the two ternary mixtures in equilibrium, then a non-polar
(hydrophobic) solid support must be used and a reversed phase system can be generated.
If the mobile phase is the less polar of the two mixtures in equilibrium, a polar support
is required.
The phase ratio of the system will largely be determined by the specific surface area of
the solid support.
Because of the equilibrium between the two phases, dynamic LLC systems are
considerably more stable than the conventional LLC systems. If the equilibrium is
disturbed by the injection of a sample, then it will soon be restored once the sample starts
to move along the column.
LLC systems offer a great flexibilitywith regards to the choice of phasesystems. We have
seen above (figure 3.8) that the choice of available mobile and stationary phases
A
Figure 3.9 : Schematic phase diagram for a ternary system of three liquids, two of which are not
miscible in all proportions. A mixture that corresponds to the composition M in the figure will
“demix” according to the nodal line LN. Two liquid phases are formed that correspond to the
compositions of L and N in the ratio kn.
54
Figure 3.10 Example of a phase diagram for a ternary system used to create a dynamic LLC system.
Components: Ethanol (EtOH), Acetonitrile (ACN) and Iso-octane (2,2,4-trimethylpentane; TMP).
I - V nodal lines. Circles: compositiods determined experimentally by titration (full circles) and GC
(open circles). Figure taken from ref. [315].Reprinted with permission.
determines the possible (general) selectivity that can be achieved in an L C system.

Potentially, LLC systems allow us to use the entire range of the triangle ATW in figure 3.8.
This is neither true for RPLC systems (section 3.2.2), nor for LSC systems (section 3.2.3).
The possibility to form a truly homogeneous, highly polar stationary phase is a real
advantage of LLC systems.
Other advantages of LLC systems include the possibility to form reproducible,
homogeneous stationary phases, a large sample capacity and a large “contamination
capacity” (i.e. LLC columns are not easily polluted by contaminants in the mobile phase
or the samples) [315]. Because the LLC system is generally well-defined, it allows a more
rigorous theoretical treatment than other forms of LC. In particular, LLC retention data
correlate well with liquid-liquid partition coefficients obtained from independent (“sta-
tic”) experiments.
A disadvantage of LLC systems relative to other forms of liquid chromatography (LSC,
LBPC) is the long time it takes to create a phase. For dynamic LLC systems, every new
mobile phase necessarily requires the creation of a different stationary phase. This may
require 50 to 170 (depending on the pore size of the support) times the volume of the mobile
phase in the column to be pumped through the system [315].
The phase diagrams of figures 3.9 and 3.10 will be affected considerably by a change
in temperature. Therefore, the temperature should be controlled very carefully, as indeed
is necessary for all LLC systems.
Summary
1. LLC systems are generally not very stable and not very easy to use in practice.
2. The use of dynamic LLC systems may help to overcome some of these problems.
3. LLCsystems ofler a great degree offrexibility with regard to the possible choice of mobile
and stationary phases.
4. Well-defined LLC phase systems can be made reproducibly.
55
5. LLC systems offer high sample capacities and contamination capacities.
6. Temperature is a critical factor for both the stability and the selectivity of LLC systems.
7. Every change in the mobile or stationary phase requires a long equilibration time.
3.2.2 Liquid bonded phase chromatography
3.2.2.1 Reversed phase chromatography ( RPLC)
RPLC is currently by far the most popular of all LC techniques [316]. Two reasons for
that have already been identified when we discussed LC in terms of solubility parameters.
First, a single RPLC column offers great flexibility for the chromatography of a wide
variety of solutes by using mixtures of water and an organic solvent as the mobile phase
(figure 3.7). Second, the overall selectivity of the RPLC system is almost always superior
to that of other LC systems (figure 3.8). Also, in the previous section, some practical
disadvantages were described for LLC systems, which have resulted in the almost exclusive
use of bonded phases for RPLC. The advantages and disadvantages of RPLC will be
summarized at the end of this section. It suffices here to point out that the emphasis put
on RPLC in this long section is amply warranted, from a theoretical as well as from a
practical point of view.
Even more than in other LC techniques, the exact mechanism of retentionin RPLC is
unclear. Certainly, a simple picture that would enable us to derive unambiguous equations
for the variation of retention with the various parameters of interest cannot yet be drawn.
Unfortunately, there has been too much speculation in the literature throughout the last
decade, often accompanied with insufficient experimental data to justify the conlusions
drawn. Therefore, it is not surprising that there are many different propositions for
expressions to describe the retention behaviour in RPLC.
In the following pages we will discuss the parameters which are relevant for selectivity
optimization and some possible quantitative relationships.
Almost exclusively,chemically bonded phases (CBPs) are now being used in RPLC, the
vast majority of the applications being achieved on silica-basedphases, modified with long
CH3
i-OtH
r-----1
+
I I
CltSi-R
L-----J I
CY
CH3
r------l I
-0fH + R’OtSi-R
L------J I
CH3
Figure 3.1 1: Schematic illustration of the reaction of a silica surface with a monofunctional reagent
(dimethyl-alkyl-ethoxysilane).Figure taken from ref. 1317). Reprinted with permission.
56
R
I B
EtO -Si -OEt HO-& --OH
I I
0 0
I I
Si
/I\
a
/I\
OEt R
\ / OH\ /R
Si Si
/\ / \
0 0
7 9
Si Si
I \
/I\ /I\
R
I
I
0
Si
/
Figure 3.12: Schematic illustration of the possible products formed by the reaction of a silica surface
with a trifunctional reagent (alkyltriethoxysilane).
alkyl chains [316]. Typically, the silica surface, featuring reactive silanol (-SOH) groups
is brought to react with reactive chloro- or alkoxysilanes according to the reaction shown
in figure 3.1 1.
In figure 3.1 1 a mono-ethoxysilane is used as an example. Alternatively, a trifunctional
reagent such as a triethoxy-alkylsilane may be used, to yield what is commonly referred
to as a “polymeric” material. The various possible products from the reaction of
trifunctional silane molecules with the silica surface are shown in figure 3.12.
The term polymeric phases arises from the fact that trifunctional reagents may just as
well react with each other as with the silica surface under the influence of (inevitably
present) traces of water. Hence, the resulting material is not necessarily a well-defined
monomolecular layer. Moreover, for every silanol group that disappears during the
reaction, two new ones are potentially formed once the product is brought in contact with
water. Many of these newly formed silanol groups can subsequently be removed by
reaction with a small monofunctional silane (e.g. trimethylchlorosilane, TMCS)*. Also,
* This so-called end-cappingprocess is common practice in the synthesisof bonded phases for RPLC,
whether mono- or trifunctional reagents are used. This is done in order to keep the number of
remaining silanols to a minimum.
57
many hydroxyl groups can be removed by a heat treatment of the product, because
different ligands attached to the silica surface will be “cross-linked” at elevated
temperatures. However, a number of the silanol groups will definitely remain present. The
presence of these remaining (“residual”) silanols is unwanted, because they may contribute
to the retention process, yielding mixed retention mechanisms and increased band-broade-
ning.
Because the silica surface is effectively shielded by the hydrophobic layer of long chain
silanes, the silanol groups will only exert their influence on solute molecules by long range
(electrostatic) interactions. Hence, the presence of silanol groups will be felt more easily
at higher pH values (pH > 5 ) where the silanol groups become increasingly negatively
charged, and for basic solutes, which may be positively charged at these pH values.
Polyelectrolytemolecules (see below) also tend to be affected by the charge of the surface,
as they are large enough to experiencethe electrostatic forces of a number of ionized silanol
groups.
The resulting CBPs are usually identified by the length of the alkyl chain. For example,
when the number of carbon atoms in the alkyl chain (nJ is equal to 18, by far the most
popular chain length [318], we speak of an octadecylsilica (ODS) or of an RP-18 phase.
The second most popular chain length [318] is the octylsilica or RP-8 (nc= 8). Apart from
the characteristics of the starting material (specificsurface area, pore sizedistribution) and
the reagent used, the alkyl chain length is the only variable to be considered.
Upon increasing the alkyl chain length, the retention (k) will initially increase
exponentially, i.e. In k increases linearly with nc *. However, when the chain length is
increased further, the increase in retention diminishes and the capacity factor becomes
roughly independent of the chain length. This is illustrated in figure 3.13. The “breaking
point” in the In k vs. nc curves was defined by Berendsen and de Galan [319] as the critical
chain length (nz). nr appears to vary with the solute. Tentatively, it increases with the size
of the solute molecules.
The initial increase of In k with nc is usually larger for solutes for which the absolute
retention is larger. Hence, in a plot of In k vs. nc the lines tend to diverge towards larger
nr As a rule, therefore, the relative retention (a) increases with nc until the critical chain
length is reached (for both solutes). Above this point a will become roughly constant.
Although there are exceptions to this rule [319], it does imply that almost always the best
selectivity is obtained with n,values above the critical chain length. This is usually realized
with ODS phases, which is the largest chain length that is commercially available**.
As a conclusion, ODS materials are understandably the most widely applied RPLC
stationary phases, and the stationary phase chain length is a variable that will usually not
be of interest as a single variable for the optimization of selectivity.
* Note that this observation is in conflict with both a liquid-liquid-likebehaviour of the hydrocarbo-
naceous layer (in which case k is expected to increase linearly with nJ, and with adsorption of the
solute on top of this layer (in which case k would be virtually independent of nJ.
** Even larger alkyl chain lengths have been used to synthesize CBPs, but they would definitely be
more expensive if turned into commercial products. Since the critical chain length is usually well
below 18 [319], such expensive materials would not usually have major advantages over the existing
ODS materials.
58
tj o t I
--.----
--.--____ rn-cresol
-.
' ,I I
-
I I
01-' I
1 3 6 10 14 18 2-2
nc
Figure 3.13: Variation of the capacity factor with the length of chemically bonded alkyl chains of the
stationary phase using monomeric phases (nJ. Mobile phase: methanol-water (8020). Solutes:
n-alcohols and aromatic solutes as indicated in the figure. Asterisks indicate the critical chain length.
Figure taken from ref. [319]. Reprinted with permission.
The mobile phase
In the previous discussion the possibility to use mixtures as the mobile phase has already
been mentioned. It was tacitly assumed that a mixture of (for example) methanol and water
has a polarity in between those of the two pure constituents. This will generally be the
case*.
The extent to which retention in RPLC can be made to vary with the composition of
the mobile phase is enormous. For almost all solutes retention will be impractically low
in some pure organic solvent (methanol, THF) and impractically high in pure water.
Hence, to achieve reasonable retention times, a mixture of water and an organic solvent
(a so-called modifier) is usually required.
We have seen before (section 3.2) how this can be explained in terms of solubility
parameters, and it was also concluded that RPLC offers superior selectivity f o r a great
variety of samples.
First of all, let us recall the basic equation for LC retention in terms of solubility
parameters:
In ki = ( v / R T ) { (6, + 6, - 26,)(S, - 6,)} + In (n,/nJ. (3.29)
* Exceptions to this rule will only be observed when two compounds are mixed that exhibit very strong
mutual interactions, for example two compounds that give a chemical reaction, or an acid and a base.
59
To a first approximation, we can write for the solubility parameter of a mixed mobile phase
13201
6, = I: (Pi 4. (3.34)
i
where (P is the volume fraction and the subscript j denotes the individual components of
the mixture. For a binary mixture of water (W) and an organic modifier ( a )eqn.(3.34) reads
and because the sum of the volume fractions must equal one
where cp= cp
,, the volume fraction of the organic modifier. If we combine eqns.(3.29) and
(3.36) we find
In ki = (vi/RT) { 6, + ( ~ ( 6 , - 6 ~ )+ 6, - 2Si} { 6, + cp(6,-6,) - a,} + ln(n,/n,)

= (vi/RTr) { (p2(6,- 6,)' + 2 ( ~ ( 6 ,- 6,) (6, - Si)
+ (6, + 6, - 26,)(6, - as)} + ln(n,/n,). (3.37)
Clearly, eqm(3.37) is a quadratic equation of the form
lnk=Aq?+Bcp+C (3.38)
in which the coefficient A is expected to be positive (see eqn.3.37), B large and negative
(because 6, > 6, and 6, B SJ, and Cis the natural logarithm of the capacity factor in pure
water. Eqn.(3.38), as well as the expectations for the sign and magnitudes of the
coefficients as expressed above, is obeyed very well in practice [321,322]. Only for mobile
phase compositions close to (P=O (mobile phases with 10% or less organic modifier in
water) may considerable deviations be observed [323]. Hence, eqm(3.38) can be used for
the description of the retention as a function of the (binary) mobile phase composition,
but the coefficient Cdoes not necessarily give an accurate estimate for the retention in pure
water [323].
A similar quadratic equation can be used to describe retention in a ternary eluent, where
water is mixed with two organic modifiers, the volume fractions of which are denoted by
(P, and (p2 [324]:
In k = A, (P; + A2 4 + Bl (P1 + B2 (P2 + c+ D (Pl(P2 (3.39)
and for quarternary mixtures the obvious expansion would be:

In k = A , (P,~ + A, P
(: + A, ( P ~ ~
+ B, (P1 + B, (P2 + B3 (P3 + c + D12 (P1 (Pz
+ D13 (P1 (P3 + D23 (P2 9 3 '
(3.40)
The same quadratic equations for retention as a function of composition have been
60
derived by Jandera et al. [325] in other terms (“interaction indices” rather than solubility
parameters) but in a very similar way. Melander and Horvath [326] arrive at an equation
which is quadratic, but they try to describe their results by a linear relationship between
In k and p. Weijland et al. [327]use equivalent quadratic expressions without an underlying
model. Their equations are less practical, because the necessary condition that the sum of
all volume fractions equals one is not implicitly contained in the expressions, so that a set
of two equations remains. They also allow a third order term to allow deviations from the
quadratic model.
Although the quadratic equations for retention (In k) as a function of mobile phase
composition (rp) provide a good description of experimental data, they are inadequate to
describe retention within experimental error. For binary mixtures the standard’deviation
between the quadratic equation (eqn.3.38) and experimental data is typically between 5
and 10% (depending on the solute) [322]. For ternary systems average deviations of 10 to
20% are typical [324]. However, the inclusion of additional (higher order) terms at will is
not an attractive way to improve the description of the experimental data. We will discuss
this more fully in chapter 5 (section 5.5).
Alternative expressions
Several other equations have been suggested for RPLC. Purnell et al. [328] suggested
plotting 1/ k vs. rp. A straight line in such a plot would correspond to the same assumptions
as were used to explain the validity of eqn.(3.14) in GLC. However, straight lines are not
observed, but instead plots of l / k vs. pshow a pronounced curvature towards the (paxis.
Purnell et al. [328] proceeded to describe l / k as a function of p using the following
four-parameter hyperbolic equation:
(3.41)
This equation may be used not only for RPLC, but also for other forms of liquid
chromatography.
Melander and Horvath [329] have suggested a five-parameter hyperbolic equation to
describe k itself as a function of rp:
A ~ ~ +r p +c
k= (3-42)
A’ cp+ B‘
Eqn. (3.42) is an example of a so-called “rational function”. Functions of this kind are
renowned for their flexibility in describing curves without physical modelling.
Lu Peichang and Lu Xiaoming derived a three-parameter parameter equation, which
reads [3301
Ink=a+ Bp+clnq. (3.43)
Obviously, eqn.(3.43) will not be able to yield a successful description of the retention
behaviour in the range of low rp values. Therefore, it was subsequently modified to
Ink = a + B + cln(1 + d q ) . (3.44)
61
None of the equations (3.41) to (3.44) appears to offer a better compromise between the
accuracy of the description and the complexity of the model than does the quadratic
equation (3.38).
Linear approximation for binary mixtures
The quadratic expressions facilitate the description of retention over large ranges of
composition. Such large ranges are usually of limited interest. Both very small (insufficient
resolution) and very large (excessiveretention times) capacity factors are unattractive. The
most useful range of capacity factors is between 1 and 10 (cf. sections 1.5 and 1.6). Over
this limited range eqn. (3.38) for a binary mixture can usually be approximated quite
adequately by a straight line [322,331]:
I n k = In ko - S q , (3.45)
where k, is an imaginary (extrapolated) capacity factor in pure water*. Although a part

of the curve for binary mixtures can be approximated by a straight line, this does not imply
that a part of the retention surface in ternary mixtures can be approximated by a plane.
Straight line approximations can only be used for quasi-binary mixtures, i.e. ternary
mixtures in which the ratio of the volume fractions of the two organic modifiers is constant.
Two of such straight lines (for different ratios) do not usually define a plane.
Initially, Snyder [331] suggested that the slope S in eqn.(3.45) would be independent of
the solute, i.e. S would be a constant for a given stationary phase and two given mobile
phase constituents. For example, for methanol-water mixtures on an ODS column S was
claimed to be about 7 [331]**. Hence, eqn.(3.45) would be approximately valid over a
composition span of 30%( = 2.3 x l O O / S ;there is a span of 2.3 units in In k between k = 1
and k = 10).
Since then, however, it has been shown that the value of S does vary systematically with
the retention behaviour of the solute [322,333]. If binary mixtures of water and methanol
are used as the mobile phase, S tends to increase with an increase in the absolute retention.
This is illustrated by the diverging set of lines in figure 3.14***.
In the methanol-water system, a linear correlation between the coefficients S and In ko
has been observed by several researchers [322,333], and the coefficients p and q describing
this linear relationship as
S = p + qlnk, (3.46)
* What is true for the coefficient C in eqm(3.38) is certainly not less true for In k, in eq~(3.45).Both
coefficients cannot be relied upon to provide accurate estimates of the (logarithmic) capacity factor
in pure water [323,332].
** Snyder used decimal logarithms instead of natural ones, causing a difference of a factor of 2.3
between the value given here and the literature data [331].
*** Note that the simple solubility parameter model (eqn. 3.37) predicts the coefficient B to be
dependent on the (polarity of the) solute and, moreover, predicts the magnitude of B to increase if
the solute polarity decreases. For RPLC this implies an increase of the slope with an increase of the
retention.
62
water methanol
'T
Figure 3.14: Variation of retention with the binary mobile phase composition for methanol-water
mixtures on an ODS column. Solutes: naphthalene ( ), anisole (0)and phenol (x). Thin lines: eqn.
(3.38) for k < 50; thick lines: eqn.(3.45) for 1 < k < 10. The diverging straight lines suggest an increase
of the slope parameter S (eqn.3.45) with increasing capacity factors Figure taken from ref. [322].
have been estimated with remarkable consistency, despite the use of different solutes and
different columns for the evaluation.
The parameters found for the coefficients in eqn.(3.46) are summarized in table 3.1. An
example showing the validity of eqn(3.46) for 32 aromatic solutes is given in figure 3.15.
Although some justifiable comments can be made regarding the use of linear regression
techniques on logarithmic equations [334],the correlation described by eqm(3.46)certainly
appears to be significant, and it may be used in an elegant way to reduce the parameter
space for the optimization of RPLC separations (section 5.4.2).
Eqm(3.46) thus appears to hold reasonably well for the methanol-water system.
However, it is obeyed much less strictly in the system tetrahydrofuran-water and not at
all in the system ACN-water. There appears to be no sensible explanation yet for these
anomalies. Data observed on the correlation between In k, and S for these two binary
mixtures are also included in table 3.1.
The concept of iso-eluotropic mobile phases
As we saw above (eqn.3.34), the solubility parameter concept provides a very simple rule
for approximating the polarity of a mixture. For a binary mobile phase containing water
(W) and methanol (Me) the sum of the two volume fractions should equal 1, hence
(3.47)
is a very simple equation for the polarity of such a mixture. According to eqn.(3.47), a given
mixture of methanol and water will have a given polarity somewhere in between the
polarity of pure methanol and that of pure water. Of course, the same solubility parameter
63
may be obtained with several other mixtures. In general, a mixture of a given polarity may
be formed by mixing two solvents, one with a higher polarity than the required value and
the other with a lower polarity.
.
I
10 -
.
5=2.27+079Ink,
corr coeff. :0.98
5-
/*
C
~
5 Ink, - 10
Figure 3.15: Variation of the slope with the intercept for linear retention vs. composition curves in
RPLC for 32 aromatic solutes on an ODS column using methanol water mixtures as the mobile phase.
Parameters S and In k, correspond to eqn. (3.45). Figure taken from ref. [322]. Reprinted with
permission.
Table 3.1:
Parameters describing the linear relationship between the slope and the intercept of
linear retention vs. composition curves in RPLC (eqn.3.46). Data taken from refs. [322]
and [333] (1).
Stationary phase Modifier p 9 r (2)
Lichrosorb RP-18 MeOH 2.27 0.79 0.98

Hypersil ODS MeOH 3.73 0.74 0.96
Hypersil ODS MeOH 3.62 0.79 0.96
Nucleosil 10-RPI8 MeOH 3.55 0.69 0.94
Nucleosil 10-RP18 MeOH 2.97 0.75 0.93
Lichrosorb RP-18 MeOH 3.50 0.73 0.98
Lichrosorb RP-18 ACN 5.87 (3) - - 0.06
Lichrosorb RP-18 THF 4.33 0.78 0.76
(1) Data in ref. [333]are acompilation from other sources. The values for the intercept were multiplied
by a factor of 2.3 (In 10) to account for the use of natural logarithms in the present text.
(2) Correlation coefficient.
(3) Average S value, since no correlation was observed.
64
To a first approximation (eqn.3.29) we may expect mixtures of the same polarity to yield
the same capacity factors. In other words, mixtures with the same solubility parameters
are expected to have the same eluotropic strength, and therefore they might be called
iso-eluotropic mixtures. If we use T H F (T) instead of methanol in a binary mixture with
water, the following equation relates two iso-eluotropic mixtures
(3.48)
where cpr is the volume fraction of THF. From this equation it follows that
'Me -W
'
cpr= 'PMe ' (3.49)
ST-',
If we use 15.85 for the solubility parameter of methanol, 25.52 for water and 9.88 for T H F
(see table 2.2), we find that
Hence, a mixture of 50% methanol in water is expected to yield roughly the same capacity
factors as a mixture of 31% T H F in water. Similarly, for acetonitrile with a solubility
parameter of 13.14, we find that
(3.51)
These very simple relationships can be verified experimentally as is shown in figure 3.1 6.
The iso-eluotropic compositions of binary mixtures of T H F and acetonitrile with water
have been plotted against the binary methanol-water composition. The thin straight lines
indicate the theoretical relationships from solubility parameter theory (eqns. 3.50 and
3.51). The thick lines correspond to average experimental data over large numbers of
solutes [335]. An (average) experimental data point can be found as follows.
For a particular solute, a capacity factor of 3 may be found in a 50150 mixture of
methanol and water. For the same solute, a mixture of 34 %THF in water may also yield
a capacity factor of 3. For a different solute, the capacity factor in a 50150 methanol/water
mixture may be 30, and the same capacity factor may be observed with a 28/72 THF/water
mixture. The average composition for many solutes that yield the same capacity factor as
the 50/50 methanoVwater mixture yields the (average) experimental point on the solid line
for T H F at cp = 0.5 in figure 3.16.
Due to specific effects, the corresponding compositions of methanol and T H F will not
be exactly the same for all solutes. Conversely, when the iso-eluotropic composition is
taken as the average of that observed for many solutes (or from solubility parameter
theory), some solutes will be eluted later than with the original methanol/water mixture,
and some will be eluted earlier. The relative differences may amount to a factor of two for
certain solutes. This should not be seen as an error in establishing iso-eluotropic mixtures.
Rather, it enables us to exploit iso-eluotropic mixtures to enhance selectivity, whilst
keeping retention roughly constant. This principle is widely used for the optimization of
selectivity in LC.
Figure 3.1 6 shows that there is good agreement between (solubility parameter) theory
65
and experiment for the composition of iso-eluotropic mixtures in RPLC. The great
advantage of this is that the composition of iso-eluotropic mixtures may be predicted for
other organic modifiers than THF and acetonitrile. In table 3.2 a selection of practically
feasible RPLC modifiers is given [336]. The table lists their solubility parameters and their
selectivity group classification according to Snyder (see section 2.3.3). Solvents within a
given group show very similar selectivities in gas chromatography (see table 2.8).
Therefore, it may be expected that the specific effects observed in LC will also be similar
for modifiers in the same group. For each modifier, the percentage that is equivalent to
one percent of methanol in binary mixtures with water (A,) is listed in the table.
Extension to multicomponent mixtures
We can easily extend the above treatment to iso-eluotropic mixtures that contain more
than one modifier. Let us rewrite eqn.(3.51) in a simplified form
(3.52)
which relates the volume fraction of a modifier j in a binary mixture with water to the
volume fraction of methanol in a binary reference mixture ( ( P ~ ~ A,
, ~denotes
~ ~ ) .the ratio
of solubility parameter differences:
(3.53)
According to the simple solubility parameter model any mixture of two iso-eluotropic
mixtures (same value for s) would yield a mixture that is iso-eluotropic to the original two
(eqn.3.34). It then follows from eqa(3.52) that for any ternary mixture of two iso-eluotro-
pic binaries the following equation holds
I' I
Figure 3.16: Iso-eluotropic compositions for binary mixtures of THF and acetonitrile in water,
relative to methanoVwatermixtures. The solid lines represent the average experimentalcompositions
for a large number of solutes. The thin lines represent calculated compositions from solubility
parameter theory (eqns.3.50 and 3.51). Figure taken from ref. [311]. Reprinted with permission.
66
pMe,ref = P M e -k Pj’’j 9 (3.54)
which can be expanded to multicomponent mixtures formed by blending a series of

iso-eluotropic binary mixtures
(3.55)
It follows from eqn.(3.54) that iso-eluotropic ternary mixtures fall on a straight line in
a figure where the two variables pMeand pj constitute the axes. Iso-eluotropic quarternary
mixtures constitute a.plane in a three-dimensional space, and so on.
Eqm(3.54) and (3.55) are very convenient for the definition of iso-eluotropic mixtures
and for the calculation of the eluotropic strength of multicomponent mobile phases, in
terms of a corresponding binary methanol/water mixture.
The solubility parameter model appears to work very well for the prediction of
iso-eluotropic mixtures in LLC and RPLC. However, in LSC the retention mechanism is
very different from the one that was assumed at the outset of this section, and hence a
different model should be applied to allow the description and possibly prediction of the
eluotropic strength in LSC. This model will be described in section 3.2.3.
Temperature
Unlike the relationship between retention and composition, the temperature dependen-
ce of retention in RPLC is beyond dispute. A typical “van ’t Hoff-type” equation may be
used:
Table 3.2:
Iso-eluotropic mixtures for RPLC.
Modifier 6 (1) Aj (2) Selectivity

(in water) (~aI’’*.cm-~’*) relative group (3)
to MeOH
Methanol 15.85 1 I1
Ethanol 13.65 0.81 I1
n-Propanol 12.27 0.73 I1
i-Propanol 12.37 0.74 I1
Acetonitrile 13.14 0.78 VIb
Acetone 10.51 0.64 VIa
THF 9.88 0.62 111
1,4-Dioxane 10.65 0.65 111
DMSO 13.45 0.80 111
(1) ref. [303]

(2) Eqn.(3.53) with Sw=25.52.
(3) See section 2.3.3.
67
In k = A h / R T - A s / R + ln(ns/n,,,), (3.56)
where Ah and As are the (partial molar) enthalpy and entropy effects for the partition of
the solute over the two phases, R is the gas constant and Tthe absolute temperature. n,/ nm
is a phase ratio term (cf. eqn.3.29). Schematically,the temperature effect may be described
by
Ink= A / T + B, (3.57)
where the coefficient A is usually positive, so that retention will decrease when the
temperature is increased.
The effect of changes in temperature on retention and selectivity is not very large.
Certainly, the mobile phase composition (water content) has a much more drastic effect
on the retention. However, what was true for GC (cfsection 3.1) is also true for LC.
Temperature and composition cannot be seen as independent variables, and a different
optimum (mobile phase) composition is likely to be observed at a different temperature
(see section 5.1.1).
Snyder et al. [337] have demonstrated the combined effects of the composition of a
binary mixture and temperature on the retention and selectivity. An increase of the
temperature has the predictable effect of a decrease in retention, with little effect on the
selectivity. Since there is usually only a small margin for which the retention of all the
solutes in a given sample is neither too high nor too low, drastic changes in the temperature
in order to enhance the selectivity cannot be applied. However, a decrease in retention due
to an increase in the temperature can be compensated by increasing the water content of
the mobile phase. In the case in which methanol-water mixtures are used as the mobile
phase, this is likely to result in an increase in the selectivity, because of the regular pattern
of In k vs. cp lines diverging towards cp=O (cf.figure 3.14). Hence, an increase in the
temperature combined with an increase in the water content of the mobile phase will
usually result in an increased selectivity, while retention may be kept constant.
A disadvantage of the operation of RPLC columns at elevated temperatures may be a
more rapid detoriation of the column because the silica is more rapidly dissolved in the
mobile phase. This effect may also lead to a reduced reproducibility of the system
(peakwiths and capacity factors).
A simple but useful equation to express the mutual effects of temperature and mobile
phase composition on retention has been described by Melander et al. [338]:
I n k = A, cp(1- TJT) + A,/T + A,, (3.58)
where A,, A,, and A, are constants for a given solute using a given stationary phase and
two given mobile phase components. T, is the so-called compensation temperature*, at
which the retention is independent of the mobile phase composition. For all practical
* The term compensation temperature arises from the compensation between enthalpy and entropy
at the temperature T, This temperatureturns out to be virtually independent of the solute in an RPLC
system [339]. The magnitude of T, (200 - 300 "C) usually implies that it is a hypothetical rather than
a practical temperature.
68
purposes T, can be considered as an arbitrary coefficient, the value of which may be
determined from experimental retention data.
A minimum of four experimental data points is required to estimate the parameters in
eqn.(3.58), similar to the experimental design employed by Snyder et ul. [337]. Of course,
eqn.(3.58) can only be applied over a limited range of compositions, for example the range
over which 1 < k < 10. To describe retention as a function of both temperature and
composition over wider ranges of the latter, more complicated equations need to be used.
A quadratic equation for the relationship between retention and composition' (eqn.3.38)
can be combined with eq~(3.57)to yield
I n k = A'(l-u/T)q2+ B'(l-b/T)q+ C'(1-c/T), (3.59)
where A', B' , C' , u, b and c are all constants.

At constant temperature this equation reduces to eqn.(3.38), while at constant composi-
tion it turns into eq~(3.57).
The pH of the mobile phase will affect retention in liquid chromatography if the
structure of the solute molecules in solution is affected by the pH. This will clearly be the
case if the solute species may occur in a protonated or a non-protonated form, dependent
on the pH. The pH may also affect the capacity factors of ions and neutral molecules for
which this is not the case, but in this case the effects are usually small.
Retention in RPLC may be expected to be a strong function of the pH if
1. different forms of the solute (e.g. protonated vs. non-protonated) may exist in the
mobile phase, which show different retention times, and
2. the (relative) occurrence of the different forms of the solute changes upon variation of
the pH.
If two different forms of the solute exist, then the first requirement will usually be met,
especially for simple monofunctional solutes. The most obvious example is the dissocia-
tion of a weak acid (HA) in an aqueous environment:
HA + H,O 7t A- + H30+ . (3.60)
Because HA is a neutral (uncharged) molecule and A - is a negatively charged ion, the

retention between the two species of A may be expected to be very different.
The second condition depends on the range of pH variation with respect to the
equilibrium constant of the solute. The dissociation constant of HA is defined as
K, = [A-1 [H,O+l [HA] (3.61)
so that for the ratio between the two different species of A ( r A )
rA = [A-]/[HA] = K,/[H,O+] (3.62)
or
69
Eqn.(3.63) shows that the greatest variation of rA can be expected around pH = pK,,
where rA= 1. When the pH exceeds the p K , value by two units, then rA= 100, so that more
than 99% of the solute is dissociated. When the pH is two units below pK,, then rA= 0.01
and less than 1% dissociation occurs. Hence, the second condition will be met if the pH
is varied in the region of the pK, value of the solute. Since for silica-based RPLC columns
the working range is limited to 2 < pH < 7, solutes with 1 <pK,< 8 may be expected to
show variations in retention upon changing the pH over the entire range allowed. In other
words, for weak acids with a pK, value in the above range the pH is a parameter of interest.
A similar argument can be set up for basic solutes. The followingschematic reaction may
serve as an example:
X + H,O e XH+ + OH- (3.64)
and a dissociation constant may be defined as
so that
/ [ A= K, / [OH-]
rx = [XH+] (3.66)
and
log rx = log K, + log [OH-] = 14-pH-pKb. (3.67)
Hence, the greatest variation in rx is observed for pH values around pH = 14- pK, *.
On silica-based RPLC columns the range of 2 < p H < 7 will correspond to bases with
7<pK,< 12. For these very weak bases pH is a relevant parameter in RPLC. Much
stronger bases @K,< 6) will be protonated completely, or almost completely in RPLC
using silica-based CBPs. On these columns, their ionization cannot be suppressed, and
they should be chromatographed as ions, either on ion-exchange or on ion-pairing systems
(see section 3.3 below).
There has been much recent interest in the development of column packings for RPLC
that can be used over a wider pH range. Potential materials include organic polymers,
carbon packings and (modified) alumina. Clearly, such stationary phases will be most
relevant for the separation of basic solutes.
If a buffer is present in the mobile phase, then the dissociation will be controlled by the
pK, value of the buffer. Because of the pH limitations of silica-based stationary phases,
only weak acids can be used as buffer compounds. As an example, we will consider the
dissociation of a weak acid (HA) in the presence of a buffer acid (HB).
~~
* In an aqueous environment the p K , and p K , values, which correspond to the same reaction, may
be related by p K , + p K , = 1 4 .
70
The following equation can readily be derived [340]:
[A -1
rA =-
HA1
- [A-I[H,O+l
- [HBI .-[B-I
[HA1 [B-I[H,O+l [HBI
(3.68)
where Ka.Ais the dissociation constant of the solute, K a , Bthe dissociation constant of the
buffer, rBthe dissociation ratio of the latter and K , , is a constant. Hence, the dissociation
ratio of A is proportional to the dissociation ratio of the buffer. The ratio rB is a more
practical quantity than the pH in mobile phases other than pure water [341]. The pH is
ill-defined in such mobile phases, but the ratio of the salt and the acid that constitute the
buffer in pure water a known quantity.
An equation for the observed capacity factor (kobs)can be derived if it is assumed that
the acid-base equilibrium kinetics are fast enough to be considered as instantaneous. This
corresponds to the observation of a single sharp peak for the ionizable species in the
chromatogram. In that case it may be assumed [340,342,343] that for the weak acid HA
[ HA1 [A-I
kobs = kHA
[HA1+ [A -1 + kA [HA]+[A-]
-
- HA + k ~ r ~ (3.69)
1 + rA
or with eqn(3.63)
(3.70)
Eqm(3.70) is a three-parameter equation which describes the relationship between

retention and pH for weak acids. If the dissociation ratio of a buffer ( r B = [B-]/[HB])
is used as a variable instead of pH, then a combination of eqm(3.68) and (3.69) leads to
(3.70b)
which again is a three-parameter equation. Similar equations can also be derived for weak
bases.
Figure 3.17 shows the observed capacity factor for a weak acid as a function of the pH.
Arbitrary values of k , = 1 (for the solute ion) and k H A= 9 have been assumed. It is clear
from the figure that only in the vicinity of the p K , value can a large effect of the pH on
retention be observed. If the pH range studied ranges from pH =p K a + 1 upward, then the
variation in kobsis less than 1O0/oof the total variation between k , and k H AA
. similar minor
effect is observed at pH values below p K a - 1. Figure 3.17 shows that complicated
sigmoidal curves are obtained for retention as a function of the pH over a range
encompassing the pK value of the solute. Simple approximations are not possible in this
71
range. Plotting different transformations of the variables will not simplify the picture*.
A practical example of the variation of retention with pH in RPLC can be found in
fig.5.17.
Three experiments are in principle sufficient to establish the three coefficients in
eqn(3.70) for a given solute. In practice this is only true if the three experiments are taken
at such values of the pH (relative to p K J that a sensible estimate of all three coefficients
can be made. This implies one experiment within one pH unit of the p K , value, one
experiment at a higher and one at a lower pH. If the p K , value of a solute is known, then
the retention behaviour can be estimated from a minimum of two experiments. Another
way to reduce the minimum number of required experiments is to assume a negligible
capacity factor for the charged species. Of course, once more experimental data points
become available initial assumptions about the value of any of the coefficients in
eqn.(3.70) can be abandoned.
Multivalent ions
So far we have discussed the behaviour of monoprotic acids and monofunctional bases.
The behaviour of bifunctional or trifunctional acids and bases is not completely different.
0 L, ' I , L
pKa -3 PKO
PH - PKo+3
Figure 3.17: Schematic illustration of the variation of the observed capacity factor (kobs)with the pH
for a weak acidic solute in RPLC according to eqn.(3.70). The retention line is calculated from
eqn.(3.70) assuming k, = 1 and k H A= 9.
* This may only be done if both k , and k,,, are known, for instance form measuments performed
at pH values well above and well below the p K , value. In that case the following transformation of
eqn.(3.70) will result in a straight line:
72
The most important equilibria are those in which one of the two species is ionized, while
the other is neutral. In many situations in RPLC both the monovalent and di- (tri-, etc.)
valent ions have negligible capacity factors in comparison with the neutral molecule.
Therefore, often all but one of the dissociation constants of an oligovalent species can be
neglected. Equations for the influence of the pH on the retention of diprotic and
oligoprotic substances can be found in ref. [316], p.239 et seq..
More value should be assigned to the possible formation of multivalent ions if
separations based on the ionic character of the solute are considered (section 3.3).
A special place is taken by polyelectrolytes. These are molecules containing a large
number of ionizable groups, each of which can be assumed to behave in a manner similar
to a single weak acid or base. However, the combination of a large number of these
functionalities gives rise to a total charge of the molecule that varies continuously from
a large positive number at a low pH to a large negative number at a high pH. Proteins are
good examples of such polyelectrolytes. At some pH, the overall charge of the molecule
will be zero. This does not imply that there is no charge on the molecules, but only that
the total number of positive charges equals the number of negative charges. This point of
zero electric charge is called the isoelectric point, and the pH value at which it occurs is
denoted as PI. Polyelectrolytesshow a variation of retention with pH over the entire range.
However, it should be noted here that their chromatographic behaviour is complicated.
It is not usually possible to isolate different retention mechanisms. In the chromatography
of proteins ionic mechanisms, physico-chemical interactions and size exclusion may all
play a role [344].
Ionic strength
The ionic strength of the eluent will affect the retention of both neutral and ionized
species. For non-charged molecules the effects of increasing the ionic strength of the eluent
can be understood as an increase in the mobile phase polarity, leading to an increase in
retention (“salting-out effect”).
For charged species it was shown by van der Venne et al. [345] and by Otto and
Wegscheider [346] that the so-called Davies equation can be used to describe the effect of
the ionic strength on solute retention quantitatively. This equation reads
(3.71)
where k, is the capacity factor at zero ionic strength, A a temperature-dependent constant

(equal to 0.512 at 25 “C [345]), z the charge of the solute ions and 1 the ionic strength (M).
It can be seen from the above equation that the ionic strength will have a similar effect
on solutes of similar charge. Only if the charges of the components to be separated are
different can the ionic strength be used to vary the chromatographic selectivity.
Summary
At the end of this section we may summarize the advantages and disadvanatages of
RPLC. The parameters that may be used for the optimization of the selectivity will be
summarized in section 3.5.1 (table 3.10~).The major advantages of RPLC are:
73
1. RPLC is a veryflexible chromatographic method. It can be applied to a great variety of
samples (seefigure 3.7).
2. Theoretically, RPLC ofiers a superior selectivityfor all but the very polar samples (see
figure 3.8).
3. RPLCcolumns with chemically bonded (alkyl)stationaryphasesprovide rapid equilibra-
tion, high eflciency and symmetrical peaks.
4. RPLC is compatible with aqueous samples, but also with a number of organic sample
solvents.
5. The aqueous mobile phases used in RPLC allow the use of buffers in the mobile phase.
This may lead to improved selectivity and eficiency. Secundary (ionic) equilibria other
than acid-base dissociation may also be used (see section 3.3.2).
Some disadvantages of RPLC remain. These may be summarized as follows:

1. The current alkyl modified silica stationary phases are not truly non-polar. This leads
to a reduction of the overall selectivity (seefigures 3.7 and 3.8).
2. Residual silanol groups are present on the surface, which may have a negative effect on
the peak shape of basic solutes and poly-electrolytes in particular.
3. Current RPLC stationary phases are only stable over a limited range of pH. A reliable
working range is 2<pH< 7.
4. The mechanism of RPLC is still poorly understood and reliable quantitative theories are
not yet available.
It appears from this summary, that most of the disadvantages of RPLC are not
fundamental, but connected to the use of a particular kind of stationary phases (alkyl
modified silicas). It may therefore be expected that advances in this area will further
enhance the potential of RPLC.
3.2.2.2 Polar bonded phases
Besides the typical RPLC stationary phases with n-alkyl groups, various other
functional groups may be chemically bonded to the surface in a similar way as described
in section 3.2.2.1. A selection of possible bonded bases, arranged roughly in order of
increasing polarity, is given in table 3.3.
Apart from the perfluorinated phases, the polarities of the CBPs in table 3.3 are typically
intermediate between those of the non-polar alkyl phases and polar adsorbents such as
silica. As we saw in figure 3.8, such phases may be operated with a polar eluent in the
Reversed Phase mode, or with a non-polar (or weakly polar) eluent in the Normal Phase
mode. The elution order of the sample components will be reversed in these two cases. A
clear example of this phenomenon has been described by Kirkland [347] for the separation
of some urea herbicides using a CBP with aliphatic ether groups.
We also saw in figure 3.8 that moderately polar stationary phases may be most useful
in the reversed phase mode for the separation of very polar solutes, which cannot be
sufficiently retained on reversed phase (alkyl) materials. A good example of this is the
separation of carbohydrates on amino bonded phases.
Chemically bonded phases may also be tailored to a specific separation problem. A case
in point is the synthesis of chiral stationary phases for the separation of optical isomers.
Another application of polar bonded phases, which is beyond the scope of this book,
74
Table 3.3:
A selection of chemically bonded phases featuring different functional groups.
Type Functional group
RP-F (Perfluorinated) - CnF*n+l
RP-n (n-alkyl) - CnH2n + 1
Cyclohexyl TP
Phenyl -0
Cyano - C = N
Diol - c& - CEL
Amino - NH2
is to be found in size exclusion chromatography, where the chemical modification is aimed

at minimizing rather than enhancing specific interaction between the solutes and the
stationary phase.
In principle, the composition of the stationary phase may be varied by using “mixed
phases”. Phases which incorporate different functional groups in a given ratio have been
synthesized (see for example ref. [348]). However, retention may not be expected to vary
linearly with the composition of such mixed phases in a manner similar to what is observed
with mixed liquid phases in GLC (section 3.1.1). This, combined with the complications
involved in preparing mixed phases and the irreversibility of bonding reactions, excludes
the composition of mixed CBPs as a practical parameter for optimization purposes.
The stability of polar bonded phases is generally considered to be less than that of
n-alkyl phases, because the Si-0-Si-C bonds are less effectively shielded against nucleophi-
lic attacks (ref. [316], p.125).
The mechanism of separation on polar bonded phases is not clear. Due to their limited
proliferation, no theories have been developed solely to descibe this particular form of
liquid chromatography. Instead, descriptions from other fields may be applied. If polar
bonded phases are used in combination with more polar mobile phases in the reversed
phase mode, then the same rules may be applied as in RPLC (section 3.2.2.1) to describe
the effects of the mobile phase, the temperature, the pH, etc..
When used with less polar mobile phases in the normal phase mode, LBPC may be
treated similarly to LSC (section 3.2.3). The same solvents may be recommended for use
in these two forms of LC (ref. [349], p.284). However, the use of particular bonded phases
may impose some constraints on the choice of solvents. For instance, amino type columns
would be modified (reversibly) by mobile phases containing acetic acid and (irreversibly)
by acetone.
75
3.2.3 Liquid-solid chromatography (LSC)
In liquid-solid chromatography (LSC) the solute is distributed between a liquid mobile

phase and a solid surface. A distribution coefficient may be defined in an analogous way
as in GSC (eqn.3.16):
‘a,, = <,s/Ci,m 2 (3.16)
where Ka,,is the adsorption coefficient (ml/g), c:,~the concentration of the solute on the
stationary phase (mole/g) and c, as usual, the concentration in the mobile phase
(mole/ml).
In principle, the adsorption isotherm is non-linear, i.e. K,., varies with varying c , , ~Only
.
at very low concentrations may an approximately linear distribution isotherm be assumed.
Therefore, LSC techniques suffer from a low sample capacity.
Snyder [350] has given an early description and interpretation of the behaviour of LSC
systems. He explained retention on the basis of the so-called “competition model”. In this
model it is assumed that the solid surface is covered with mobile phase molecules and that
solute molecules will have to compete with the molecules in this adsorbed layer to
(temporarily) occupy an adsorption site. Solvents which show a strong adsorption to the
surface are hard to displace and hence are “strong solvents”, which give rise to low
retention times. On the other hand, solvents that show weak interactions with the
stationary surface can easily be replaced and act as “weak solvents”. Clearly, it is the
difference between the affinity of the mobile phase and that of the solute for the stationary
phase that determines retention in LSC according to the competition model.
Snyder [350]formulated the following equation to describe the above effect quantitati-
vely:
log K,., = log V, + a (Sp - A , 8). (3.72)
In this equation is the adsorption energy of the solute on a standard adsorbent. A, is
the adsorption area of the solute molecule. 8 is the adsorption energy of the solvent per
unit area on the same standard adsorbent*, usually referred to as the solvent strength or
eluotropic strength. a i s the activity of the adsorbent and V , is the volume of the adsorbed
solvent per gram of stationary phase. Hence, V, can be seen as a compensation factor for
the dimensions of K,.,.K a / V , is a dimensionless quantity.
The choice of units and standards for the remaining variables is arbitrary. The following
conventions were followed by Snyder [350]:
standard adsorbent (a=1) : dry alumina
standard solvent (8= 0) : n-pentane
standard solute area (A,=6) : benzene**
* 8 is the solvent adsorption energy (S$ equivalent to for the solute) divided by the adsorption
area of the solvent (As). Hence, the competition factor can be rewritten as
sp - A i 8 = / t i ( - g --)
-q
Ai As
This expression shows more clearly that the competition is based on the adsorption energy per unit
area, i.e. e / A , vs. e / A r
** This convention for A, implies that one unit in A, corresponds to approximately 8.5 A*.
76
Using these conventions, values for a (different adsorbents), 8 (different solvents), 9
and A i (different solutes) can be established from chromatographic experiments (for
procedures see ref. [350]).
Eqn. (3.72) describes retention in LSC in terms of separate parameters for the adsorbent
(V,, a), the solute (9,
A i ) and the solvent (8).
As such, it has proved invaluable for the
interpretation of retention and selectivity phenomena in LSC. For example, the effect of
a change in the solvent using the same stationary phase and the same solute can easily be
understood in terms of a variation in EO.
Unfortunately, this is no longer true once several parameters are changed at the same
time. For instance, the value of 8 appears to depend not only on the solvent, but also on
the adsorbent. Hence, different 8 values have been tabulated for different adsorbents.
Some examples of values for common solvents are collected in table 3.4.
Clearly, the values for 8 are different on silica and alumina. They are very different
from those observed on carbon, which is partly, but not entirely due to the different
conventions used in this case by Colin et al. 1351). Likewise, the solute parameters A i and
will be different for different adsorbents. Hence, eqn.(3.72) gives a consistent
description for one particular stationary phase. A new set of parameters will have to be
established for each new adsorbent.
As a consequence, the influence of the stationary phase on retention and selectivity
cannot be explained on the basis of eqn.(3.72).
Influence of the mobile phase
The influence of the solvent in LSC is described by the solvent strength parameter 8.
Table 3.4
Eluotropicstrength (E’, eqn.3.72) of some common solvents for LSC. Data taken from refs.
[349] en [351].
&O 6 Select.
Solvent (1) group
silica alumina carbon (2)
Alkanes 0.01 0.01 0.10-0.17 7

1Chlorobutane 0.20 0.26 0.13 8.42 V
Chloroform 0.26 0.40 0.18 9.87 Vlll
Methylene chloride 0.32 0.42 0.13 10.68 V
Isopropylether 0.34 0.28 - 7.60 I
Ethyl acetate 0.38 0.58 0.13 9.57 VI
THF 0.44 0.57 0.14 9.88 I11
Acetonitrile 0.50 0.65 0.04 13.14 VI
Methanol 0.7 0.95 0 (3) 15.85 I1
(1) Solubility parameter in ca1”2.cm-3’2 [303].

(2) See section 2.3.3.
(3) By definition, see ref. [351].
77
According to eqn.(3.72) a higher value of 8 will result in a lower value for the capacity
factor. It can be concluded from table 3.4 that the solvent strength on silica and alumina
stationary phases roughly increases with increasing polarity (6)of the solvent, but that
there is no quantitative correlation between these two solvent properties. For example,
ethers are much stronger solvents (especially on silica) than can be anticipated on the basis
of their solubility parameters.
Although the number of pure solvents that is compatible with the LSC system is much
larger than it is for RPLC (section 3.2.2.1), mixtures are frequently employed as mobile
phases. Snyder [350] has developed the following equation for the dependence of the
solvent strength (8) on the composition (N, the mole fraction of the stronger solvent B)
in a binary mixture
G B= + (l/an&log { N b 10 anb(G- 4) + 1 - Nb} (3.73)
where n b is the molecular size of the solvent B and a is again the activity of the stationary
phase. Figure 3.18 shows the variation of G Bwith the composition for a series of binary
mixtures in which n-pentane is the weaker of the two solvents. The experimental points
in the figure show that at least for these mixtures eqn.(3.73) provides a good quantitative
description. According to eqn.(3.72)log (1 / K,,J is proportional to G Band the same is true
for log (l/k). Therefore, the variation of retention with composition in LSC appears to
be rather complicated.
I , I I I I I t I J
20 LO 60
% Blvlv) - 80 100
Figure 3.18: Variation of the eluotropic strength (GB)

with the composition of the mobile phase in
LSC (eqn.3.73). Weak solvent is n-pentane; Strong solvents (from bottom): carbon tetrachloride,
n-propylchloride,methylenechloride, acetone, pyridine. Figure taken from ref. [349]. Reprinted with
permission.
78
I 2 I
Figure 3.19: Variation of retention with composition in LSC according to the simplified linear
relationship of eqn.(3.74). Stationary phase: Lichrosorb ALOX T (Alumina). Mobile phase:
n-propanol (rp is volume fraction) in n-heptane. Solutes: lumisterol (l), tachysterol (2), calciferol (3)
and ergosterol (4). Figure taken from ref. [357]. Reprinted with permission.
However, a much simpler relationship between retention and composition may often
be observed in practice. It was suggested by Soczewinski [352,353] that a plot of In k vs.
In X, (where X , is the mole fraction of the stronger solvent B) would yield a straight line,
according to
I n k = c - nInX,, (3.74)
where c and n are both constants. In fact, as pointed out by Jandera and ChuraEek [354],
Soczewinski’s eqn.(3.74) is a simplified version of the treatment of Snyder.
Figure 3.19 shows an example of the linear variation of retention with composition
according to eqm(3.74). In this figure the logarithm of the volume fraction ( q ) of the
stronger solvent is plotted on the horizontal axis. Plotting In X , will lead to a similar linear
plot. The simple equation of Soczewinski (eqn.3.74) often yields a very good description
of experimental data in LSC [353,355,356].
Small percentages of strong solvents (so-called “modulators”) are seen to have a drastic
effect on the value of .z0 (see figure 3.18). This is especially true for the most polar solvent
of all, water. Therefore, the water content of the mobile phase and the extent to which the
79
stationary surface is covered with water are critical parameters in LSC. Besides a drastic
reduction in the retention time, an improved peak shape may be the result of the addition
of water to the mobile phase. Both the column efficiency and the reproducibility of the
analysis may improve as a result.
Isosluotropic mixtures
Figure 3.20 is a graphical representation of the eluotropic strength in LSC. This

nomogram was originally published by Saunders [358].The solvent strength parameter 8'
increases from left to right in the figure as indicated on the top axis. Every other horizontal
line represents a particular binary combination of two solvents with the compositions
indicated. It is clear that the scale division on these lines, which correspond to eqn.(3.73),
is highly non-linear.
The vertical dashed line illustrates how a seriesof iso-eluotropic mixturescan be located.
Mixtures of about 75% methylene chloride in n-hexane, 49% diethyl ether in n-hexane,
50% methylene chloride in 2-chloropropane, 46% diethyl ether in n-hexane, 1.5%
acetonitrile in 2-chloropropane and 0.1% methanol in 2-chloropropane all show similar
solvent strengths (8= 0.30).
As in RPLC, these iso-eluotropic mixtures are expected to yield similar capacity factors,
but may give rise to certain specific effects towards certain (types of) solutes, which may
be exploited to enhance separation.
Of the many different iso-eluotropic solvent mixtures, not all are equally attractive from
solvent strength EO (silica)

0 .05 .10 .15 .20 .25 .30 .35 .LO .L5 .SO .55 .60 .65 .70 .75
0510
i in H~
01 3 5 10 50 :1w
l ' f i ' 1 I I 1 ' 1 1
01 3 5 10 p 1w
: I Il%Et20inHx
1
p
-
ON) 3odm
yz in i Pr CI
7 1 I ,*,I,
I I I I % EtzO in iPrCl
0 .5 1 i2'3 5 I0 30 50100
I I I%ACN~~~P~CI
0 ; -5 1 2 3 5 10 zp :050 100
: , I I I IOhMeOHiniPrCI
' 0 60100
%EtzO in MC
01235 10 lU)
Hx ,hexane I
! I %ACN in MC
7
I
iPrCI.isopropyl chloride ? 5 2; 3p loo

:IH%MeOHinMC
MC,methylene chloride
0 510
Et20,ethyl ether l w ] % A C N in EtzO
ACN. acetonitrile 0 .5 1 2 3 5 10 30
MeOH. methanol !I %MeOH in EtzO
01 3 5 'p , , 'P %MeOHin ACN
10
Figure 3.20: Nomogram illustrating the solvent strength of various binary solvent mixtures for LSC.
Vertical (dashed) line illustrates a series of iso-eluotropic solvents (see text). Figure taken from ref.
80
a practical point of view. Moreover, to allow a reasonably efficient search for optimum
conditions, it is necessary to select a few solvents.
Glajch et al. [359]have suggested the use of methylenechloride, methyl t-butyl ether and
acetonitrile as modifiers in n-hexane (see section 5.5.1).
Eqn.(3.73) suggests that any mixture of two solvents with the same 8 value (iso-eluotro-
pic solvents) will also have the same eluotropic strength. This would allow the application
of a similar strategy for the definition of iso-eluotropic multicomponent mobile phase
mixtures as was used for RPLC in section 3.2.2.1. In practice, the situation in LSC has
proved to be more complicated, because an effect described as “solventlocalization” limits
the validity of eqm(3.72) and (3.73) if polar components (such as acetonitrile or methyl
t-butyl ether) are present in the mobile phase. This makes it difficult to calculate the
composition of iso-eluotropic mixtures for LSC with sufficient accuracy for optimization
purposes [360-3631.
In practice, as a first approximation, it may be assumed that mixtures of iso-eluotropic
solvents can be used. If the resulting solvent strength is either too high or two low, it may
be corrected by the addition of more or less n-hexane.
For many years silica has been the dominant adsorbent used in LSC. Silica has the
advantage of abundant availability. It can be obtained commercially in different particle
sizes (spherical or aspheric materials), different specific surface areas and different pore
size distributions. Because the specific surface area is usually large (up to about 400 m2/g),
the sample capacity of silica is relatively high.
A disadvantage of the availability of many different silicas is the limited reproducibility
of the packing materials. Apart from the factors described above, the chromatographic
behaviour of the silica can be affected by chemical factors such as the structure of the
surface (affected by heat treatments and by washing the column with acidic or basic
solutions), the history of the material (previous usage) and the presence of contaminants
(e.g. metal ions). The water content is another major factor. Physically adsorbed water can
be removed from or added to the surface, but water bound to the surface as silanol groups
‘(chemisorption) cannot be introduced or removed once the silica is packed into the
column.
In general, different silicas from different manufacturers may show large differences in
retention and selectivityand even between different batches of the same (nominal) product
from the same manufacturer the differences may be considerable. For these reasons, it is
often necessary to re-optimize the separation (mobile phase) if a new column is installed
for an existing separation method.
Historically, alumina used to be one of the standard adsorbents for LSC. Snyder (ref.
[350], chapter 11) has compared the chromatographic selectivity of silica and alumina
surfaces extensively. Alumina may offer some advantages over silica, especially for
separations that can be enhanced at high pH values. In recent years therefore, there has
been a revival of interest in alumina and its applications in LC [364].
Another material that has recieved considerable attention in recent years is carbon [365].
Carbon can be used as an adsorbent for LC in one of several forms, such as pyrolytic
carbon (either as such or as a thin layer covering silica particles), glassy carbon and
81
graphitized carbon. An especially interesting carbon material for LC appears to be the
porous graphitized carbon (PGC) described by Gilbert et al. 13661.
The advantages of carbon surfaces are their chemical inertness and stability. However,
it is difficult to prepare carbon packings with the same specific surface area, pore size
(distribution) and pore volume as typical silicas. Moreover, although purely carbonaceous
materials will in theory be highly non-polar reversed phase materials, all materials
prepared so far are found to behave as fairly polar surfaces, requiring fairly non-polar
(high methanol content) mobile phases for the elution of the solutes. Figure 3.8 suggests
that the present carbon-based materials are most useful for the RPLC separation of polar
compounds.
Temperature
The effects of temperature in LSC are similar to those observed in RPLC. Eqn.(3.57)
may be used for a quantitative description. The temperature is usually not considered to
be a relevant parameter. A typical change of 2% in k for 1 "C variation in the temperature
(ref. 13491, p.390) may, however, warrant a careful control of the column temperature.
3.3 SEPARATION OF IONS IN LC
The LC methods discussed before were based mainly on physico-chemicalinteractions

between the solute on the one hand and the two chromatographic phases on the other.
Although we have seen that in RPLC the degree of ionization of weakly acidic or basic
solutes may be a major factor in the control of retention and selectivity, the ionic species
themselves were not exploited purposefully to realize or enhance the separation. In fact,
in a typical RPLC system all fully ionized solutes will show little retention and therefore
little resolution can be achieved between different ions. The methods described in this
section make positive use of the ionic character of solutes to create a chromatographically
selective system.
3.3.1 Ion-exchange chromatography (IEC)
In ion-exchange chromatography (IEC) ionic or, rather, ionizable groups (R) are
permanently present on the surface of the stationary phase. In the absence of the solute,
these groups are all masked by a counterion (0, which is present in the mobile phase in
a constant concentration. Retention is based on an opposite charge between the solute ion
(denoted below as the anion X- or the cation YH+) and the ionic groups on the stationary
phase. The counterion has a charge similar to that of the solute ions. Typically, the
following schematic reactions can be used to describe the IEC process:
E R+U- + X- 3 R + X - + u- (3.75)
or 3 R - U + + YH+ e = R-YH+ + U+. (3.75a)
In the first case we speak of anion-exchange, since the exchanged ions U- and X - are
negatively charged. The second reaction illustrates a cation-exchange process. It is clear
that different stationary phases will typically be used for the two types of IEC. Thus, there
82
fraction
cahon exchanger
strong weak
0 2 L 6 8 1 0 1 2
weak
0 2 L 6 8
-
10 12
PH
Figure 3.21: Variation of the capacity (fraction of charged functional groups) of typical ion-exchange
materials with the pH of the mobile phase. Top: cation-exchangers; bottom: anion exchangers.
Ionized fraction is fraction associated with counterions. Remaining fraction is associated with H30+
or OH- ions.
are anion-exchange columns, with positive groups on the surface, and cation-exchange
columns with negative groups.
A further differentiation can be made from a classification in strong and weak
ion-exchange materials. A strong ion-exchanger possesses functional groups which are
always ionized. This implies that over the practical range of pH values the capacity of the
stationary phase remains unaltered. Weak ion-exchangers are affected by the pH. Weak
cation-exchangers gradually lose their exchange capacity if the pH is decreased. With
weak anion-exchangers this occurs when the pH is increased. The variation of the degree
of ionization of the functional groups on the surface, a quantity that is directly
proportional to the ion-exchange capacity, of some typical stationary phases is illustrated
in figure 3.21. Examples of the different kinds of exchangers are given in table 3.5.
The different stationary phases can also be classified on the basis of their physical
structure. Pellicular materials, the particles of which consist of a hard (glass) core covered
by a thin layer of an ion-exchange resin, may be used if a moderate efficiency and a small
ion-exchange capacity are acceptable, but not if the column is required to have a high
Table 3.5:
Examples of different types of ion-exchange materials.
~~
Cation Anion
Strong Sulfonate - so, Quaternary amines - NR:

Weak Phosphonate - Po;- Tertiary amines - NRZ
Carboxylate -coo- Secondary amines - NR+
83
stability. Glass beads covered with a styrene-divinylbenzene cross-linked copolymer
backbone to which the functional (ion-exchange) groups are chemically bonded may be
employed over the entire range 1 < pH < 13 [3671.
As an alternative to pellicular materials, microparticulate stationary phases may be
used. These are either based on organic resins or on inorganic oxides. The latter class
contains bare oxides, as well as chemically bonded phases, which may be synthesized in
a way similar to that described in section 3.2.2.1, but the functional end group is now an
ionic one.
Resin based materials offer a greater chemical stability (large pH range), whereas silica
based materials are mechanically more stable and allow a wider range of (organic) solvents
to be used. All microparticulate phases offer a high column efficiency and a large
ion-exchange capacity. Therefore, in modem HPLC they are usually preferred to
pellicular packings.
The retention and selectivity in IEC are influenced by a number of parameters, which
we will discuss below.
Counterion concentration
The concentration of the counterion can be used to control the retention in IEC. It plays
a role similar to that of the eluotropic strength of the eluent in RPLC or LSC, in that it
affects retention much more than it does selectivity. The capacity factor can be related to
the distribution coefficient of the solute ( D J
k = D,(V,/V,J. (3.76)
D, is the ratio of the total concentrations of solute ions in the two phases. The total
concentration may involve “free” ions, protonated ions, absorbed molecules and all other
forms of the sofute ion X:
D, = I:[“A?’], / I: rlr.],,,. (3.77)
For the exchange of strong ions on strong exchangers very simple equations result:
D, = [RxJ / [ X - 1 (3.78)
for (monovalent) anions or
D, = [RYH] / [YH+] (3.78a)
for cations. For a weak monovalent anion the situation is slightly more complex and the
distribution coefficient becomes
A similar equation can be written for a weak cation:
(3.79a)
84
If we introduce the exchange equilibrium constant ( Km), which reads for monovalent
anions (see eqn.3.75)
(3.80)
then we find with eqn.(3.78)
(3.81)
and with eqn.(3.79)
(3.82)
where K,,is the acid dissociation constant for solute X(eqn.3.61). A similar equation can
readily be derived for cations. From eqm(3.81) and (3.82) it is clear that for monovalent
counterions, D (and hence the capacity factor k) is inversely proportional to [ U-1, SO that
In k = In [U-]+ In k, , (3.83)
where k, is the capacity factor observed at a (hypothetical) unit concentration of

counterions.
According to eqn.(3.83) a plot of In k vs. In [ U-]will yield a straight line, with a slope
that is independent of both the kind of solute and the kind of counterion. The slope will
only vary if the valence of the solute (a) or the valence of the counterion (b) changes. This
can easily be accounted for in eqn.(3.83), which reads for the general case
In k = ( a / b ) In [vl -t In k, . (3.84)
The validity of eqn.(3.84) is demonstrated in figures 3.22% b and c. In figures 3.22a and
b the retention of some nucleotides in IEC is shown. The counterion is monovalent
potassium dihydrogen phosphate. The figure shows a series of straight lines, the slopes of
which are in good agreement with the predicted values from eqa(3.84): 0.96 for the
monophosphates (solutes 1 to 5), 1.85 for the diphosphate (solute 6) and 3.03 for the
triphosphate (solute 7).
Figure 3 . 2 2 ~illustrates the validity of eqn.(3.84) for the IEC separation of some
inorganic anions. The counterion in this example is a mixture of monovalent and divalent
phthalate ions. At the selected pH of 5.3 the average charge of the phthalate ions is 1.52.
According to eqn.(3.84) this would result in slopes for monovalent solute ions of about 0.66
and about 1.3 for divalent ions. These figures correspond reasonably well with what is
observed in figure 3.22b [3681.
From eqn.(3.84) and figures 3.22a, b and c we conclude that the concentration of
counterions in IEC is a primary parameter which may be used to vary retention, i.e. to
bring the capacity factor into the optimum range. Only the selectivity between solutes of
different valencies will be affected considerably by changes in the concentration of the
counterion.
85
lo1
‘.
Figure 3.22: (a) and (b) Examples of the variation of retention (log k) with counterion concentration
(log c) for nucleotides according to eqn.(3.84). Mobile phase: potassium dihydrogen phosphate in
water, pH=3.15. Stationary phase: Perisorb AN. Solutes: 1 = thymidine 5’-monophosphate, 2 =
ribothymidine 5’-monophosphate, 3 = deoxyuridine 5’-monophosphate, 4 = deoxyguanosine
5’-monophosphate, 5 = guanosine 5’-monophosphate, 6 = guanosine 5’-diphosphate, I = guanosine
5’-triphosphate. Figures taken from ref. [357]. Reprinted with permission. (c) (see opposite page)
Examples of the variation of retention (log k) with counterion concentration (log c) for some
monovalent and divalent inorganic anions according to eqn.(3.84). Mobile phase: Phthalic acid in
water (pH = 5.3). Stationary phase: Vydac 302 IC silica-based ion-exchanger. Solutes as illustrated
in the figure. System peak corresponds to the retention time of the phthalate ion. Figure taken from
ref. [368]. Reprinted with permission.
Mixed retention mechanisms
In the above discussion we have assumed that no other retention mechanisms play a role
in IEC other than an ion-exchange process. For instance, we have assumed in eqm(3.78)
to (3.82) that other forms of the solute, such as the protonated form of an anion ( H X ) , are
not present in the stationary phase. In practice, this assumption is not always correct. For
this reason, different’ packing materials with the same functional groups may show
different selectivities. For example, Rabel[367]has illustrated the differences in selectivity
for the separation of nucleotides on two different strong anion-exchangers.The selectivity
on a pellicular material was markedly different from that on a silica-based stationary
phase. The percentage of cross-linking in an organic resin may also affect the selectivity
[369].
86
I
\* H2 POL
- 0.2
-0.L
-2.7 -26 -2.5 -21, -23 -2.2 -23 -2.0
log -c
Mixed retention mechanisms are most evident in the separation of polyelectrolytes.

These are large, multivalent molecules, which possess polar and non-polar groups (or
"sites") on the surface of the molecule in solution, that may interact physically with the
backbone of the ion-exchanger. The most important examples of polyelectrolytes are
proteins. IEC has long been the major tool for the separation of proteins by HPLC, but
it is being replaced more and more by RPLC [344]. One of the reasons for this is that due
to mixed retention mechanisms broad and non-symmetrical peaks are common for the
IEC separation of proteins.
Type of counterion
The type of counterion used may affect the retention considerably. The eluotropic
strength of the different counterions is usually expressed as an eluotropic series. An
example of this is shown in table 3.6.
Also, the type of counterion may have an effect on selectivity. Especially for anions
dramatic differetices are sometimes observed (see e.g. ref. [370]).
Similarly, the choice of the buffer may have a considerable effect on the selectivity in
IEC. Phosphate has been recommended as the most useful general purpose buffering agent
13671. If other buffers are used, the selectivity needs to be re-optimized.
The pH will affect the capacities of weak ion-exchangers as well as the dissociation of
weak anions and cations (e.g. eqn.3.82). In these cases, the pH may be the most relevant
parameter in the IEC separation process.
Analytical equations can be derived for all combinations of solutes, stationary phases
87
and counterions. However, especiallyif we do not limit ourselves to monovalent ions, these
equations may become quite complex algebraically. For the simple case of a monovalent
anion on a strong anion exchanger using a strong counterion, the relationship is given by
eqn.(3.82). We find that the capacity factor of the solute is proportional to the relative
dissociation, since the last factor in eqn.(3.82) may be rewritten as
(3.85)
This factor will approach unity if [H,O+]4 KQ,x,i.e. pH > pK,,,. Hence, at pH values
well above its p K , . , value, the solute will be completely dissociated. At pH = p K , . , the
dissociation will be SOo/o and it will further decrease when pH < pK,., Some schematic
examples of the effect of the pH on the retention of weak acids and bases on strong
ion-exchangers are shown in figure 3.23.
On weak ion-exchangers the effective relative dissociation is the product of the
dissociation of the solute and that of the stationary phase. The effective relative
dissociation is then
(3.86)
where s denotes the stationary phase (ion-exchanger). This may result in a maximum
retention at a particular pH value, as is illustrated in figure 3.24.
Table 3.6:
Typical eluotropic series for anions and cations in IEC. Ions are listed in order of
increasing elution strength.
Anions Cations
Fluoride Lithium
Hydroxide (OH-) (1) Hydronium (H,O+) (1)
Acetate Natrium
Formiate Ammonium
Chloride Potassium
Thiocyanate Rubidium
Bromide Cesium
Chromate Silver
Nitrate Manganese
Iodide Copper (11)
Oxalate Calcium
Sulfate Strontium
Perchlorate Barium
Citrate Trivalent anions
Hydroxide (OH-) (2) Hydronium (H,O+) (2)
(1) On stong cation and anion exchange columns.

(2) On weak exchangers.
88
. "strong anion exchanger"
-
I
0 2 L 6 8 1 0 1 2 1 L
PH
&ax \m
1 .J "strong cation i
exdanger"\
\
\
\
\
0 2 L 6
-
8 1 0 1 2 1 L
PH
Figure 3.23: Variation of the capacity factor (relative dissociation coefficient, see eqn.3.85) as a
function of the pH for some weak acids and bases on strong ion-exchange materials. p K , values refer
to different solutes.
0 2 L 6 8 1 0 1 2 1 L
"weak cation
exchanger"
0 2 L 6 8 1
-
0 1 2
PH
1 L
Figure 3.24 Variation of the capacity factor (relative dissociation coefficient, see eqn.3.85) as a
function of the pH for some weak acids and bases on weak anion-exchangers @ K b = 6 ; top) and
cation-exchangers @ K , = 7; bottom). p K , values in the figure correspond to different solutes.
Figure 3.25 shows an example of the effects of pH in IEC in practice.
Temperature
Frequently, IEC separations are carried out at elevated temperatures. This is because
the kinetics of the ion-exchange process may be improved dramatically. The efficiency of
the column may be increased by a factor of three if the temperature is increased from 30
to 70 O C [3711. An additional advantage of an increased column temperature is a decrease
in the eluent viscosity and hence a reduced pressure drop over the column.
89
Increasing the temperatureleads to a decrease in retention. However, in some cases this
is accompanied by marked changes in the selectivity (see e.g. ref. [371]).
Organic modifiers
Organic modifiers may be added to the mobile phase in IEC in order to optimize the
6 7 8 9
___)
PH
6 7 8
- 9
PH
Figure3.25: Experimental variation of the retention with pH for somenucleobases (a)and nucleosides
(b) in IEC. Stationary phase: Aminex A-28. Mobile phase: 5 mM citrate - 5 m M phosphate buffer
in 50-50 ethanol-water. Temperature:70 "C. Figure taken from ref. [371]. Reprinted with permission.
90
selectivity for a particular separation. According to Rabel [367], not more than 10% of
modifier may be added, because otherwise the dominant retention mechanism may be
partition or adsorption rather than ion-exchange. However, in practice it is not relevant
what the mechanism behind a separation is. It is only relevant that optimum resolution
is obtained for all solutes.
Therefore, the addition of large amounts of organic modifiers may be considered as a
parameter for the optimization of IEC separations on inorganic stationary phases (with
or without a chemically bonded exchange group). For organic based materials, the
amounts of organic modifiers that can be added to the mobile phase may be limited by
swelling of the polymer.
The addition of organic modifiers may lead to either an increase or a decrease in
retention. Moreover, the effects can differ considerably for different solutes, as is
illustrated in figure 3.26 for some basic alkaloid drugs using unmodified alumina as the
ion-exchange material.
The organic modifier content of the mobile phase can be used to optimize the separation.
This has for example been shown for alkaloids [373,374] and for nucleosides and
nucleobases [372].
Ion chromatography
Ion chromatography is a fashionable phrase used for the IEC separation of inorganic
ions. Initially [375], the term was used exclusively to describe an ion-exchange LC system,
equipped with a special background suppression column and a conductivity detector. The
suppression column is used to reduce the conductivity background of the mobile phase.
If a weak counterion is used (such as the bicarbonate ion), then an exhange of sodium ions
against protons in the suppressor column will lead to a greatly reduced background
conductivity.
Nowadays, however, ion chromatography may describe any of various techniques for
the liquid chromatographic separation of inorganic ions [376]. Therefore, the parameters
that are relevant for IEC in general bear the same kind of relevance in ion chromatograp-
hy. The most prolific application of ion chromatography is the simultaneous determina-
tion of a series of common inorganic anions. These include monovalent ions (fluoride,
chloride, bromide, iodide, nitrate and nitrite) as well as multivalent ones (sulfate, sulfite,
phosphate). Therefore, both the retention and the selectivity are strongly affected by the
concentration of the counterion. This is illustrated in figure 3.22~.
Usually, the stationary phases for ion chromatography have a low capacity, in order to
reduce the background signal in conductivity detection. Selectivity in ion chromatography
can be optimized along the same lines as other chromatographic methods [368].
Gradient elution
Many separations in IEC require the use of gradients, i.e. not all components of the
sample can be eluted at one given composition of the mobile phase, so that this
composition has to be changed during the elution. Since for weak solute ions the pH has
the largest effect on retention, pH gradients are the most effective. However, salt (ionic
strength) gradients are both more general and more predictable in their effects [377].
91
I
. , ,, ,, ,, .
0 20 Lo 60 80
MeOH content/% -
t
i ,, ., ,, ,,
0 20 Lo 60 80
ACN content/% -
Figure 3.26: Variation of the retention of alkaloid drugs in IEC with the concentration of various
organic modifiers in the mobile phase. Conditions: citric acid and trimethylammonium hydroxide;
Figures (a) and (b): 0.01 M, pH=6; Figure (c): (see opposite page) 0.002 M, pH=6. Column:
Spherisorb A 10 Y (alumina). Solutes: cocaine( 0 ), dihydromorphine(*), morphine (El), dihydroco-
deine (A), ephedrine (0)and brucine (x). Figure taken from ref. (3731. Reprinted with permission.
Conclusion
A number of parameters may affect retention and selectivity in IEC. Because of the
subtle differences in selectivity between different ion-exchangers, it is often necessary to
optimize the separation in a specific situation. The common way to do this [367] is by
varying a single parameter at a time, keeping all the others constant. As we will see in
chapter 5 (section 5.1.1), this is not the most appropriateway to approach the optimization
92
I I I
0 20 Lo 60 80
THF content/%-
of chromatographic separations, and therefore IEC is one of the techniques that may
benefit substantially from a systematic approach to method development and optimiza-
tion.
3.3.2 Ion-pair chromatography (IPC)
Ion-pair chromatography (IPC) is based on the principles of ion-pair extraction [378].

The underlying idea is that in a two-phase (liquid-liquid) system of which one phase is
aqueous and the other one organic, ions will predominantly be found in the aqueous layer,
and neutral molecules in the organic one. This will be true for most organic molecules, and
the more so the larger and the less polar the molecules become. It will also be true for most
ions, the more so the smaller and more polar the separate ions (X and r) are. If X- is a
solute anion and if Y+ is a so-called pairing ion present in the aqueous phase, then a simple
equilibrium scheme for ion-pair extraction can be drawn, as is shown in figure 3.27.
In figure 3.27 K%qy is the equilibrium coefficient in the aqueous phase:
(3.87)
and K,, it the usual distribution coefficient.

In a chromatographic system, the capacity factor will again depend on the distribution
coefficient for the solute X, which reads for the example in which the aqueous phase is the
1 mobile one:
(3.88)
organic
Figure 3.27: Simple mechanism for ion-pair extraction. Ion-pair formation occurs in the aqueous
phase. The ion-pair X Y is distributed over the two phases.
93
If the product K”Xq,[ Y+] < 2, which is usually the case at common concentrations of the
pairing ion Y, then eqn.(3.88) simplifies to:
D, = K , , KIqy [ Y+] (3.89)
and hence, with eqm(3.76)
k = K,, Kyy[ Y+](VJ V,,,) (3.90)
which for a given solute Xand pairing ion Yon a given column yields a straight line defined
by
In k = In [Y’] + In k, , (3.91)
where, as in eqm(3.83) k, is the capacity factor at a (hypothetical) unit concentration of

the ion Y. Of course, the above treatment can readily be adapted to account for cationic
solutes and to normal phase ion-pair chromatography (NP-IPC), in which the aqueous
phase is the stationary one.
According to eqn.(3.90), k is proportional to the concentration of the pairing ion, with
the proportionality constant being determined by the distribution coefficient for the
neutral molecule ( K x y ) and by the dissociation constant for this molecule into the two
separate ions X and Y The first factor is affected by the same parameters as
retention in the LC of non-ionic solutes (section 3.2). The latter factor will be determined
by the nature of the solute ion and the pairing ion and by the composition (ionic strength,
pH, modifier content) of the aqueous phase.
Above we have described a very simple mechanism for ion-pair extraction. The
mechanism in practical IPC experiments is usually not quite as simple. There are several
complicating factors.
In the first place, we have not considered the pH influence on the dissociation of weak
solute ions or pairing ions. The effect of the pH in IPC will be addressed below.
aqueous
2“/ XY
J
Figure 3.28: Illustration of the mechanismof IPC.The solute ion X,the pairing ion Y and the ion-pair
X Y are all distributedover the two phases. Ion-pair formation occurs in both phases (reactions along
horizontal lines). “Ion-exchange’’reactions may also occur. These reactions involve solute ions in one
phase and pairing ions in the other. These reactions can be found along diagonal lines in the figure.
94
In the second place, we have assumed that both of the ions ( X and Y) are found
exclusively in the aqueous phase. This is a great simplification of the true mechanism of
IPC. Figure 3.28 shows a more realistic version of figure 3.27.
According to figure 3.28 the solute X can be retained by at least three different
mechanisms:
- Partition of the solute ion between the two phases (characterized by K,)
- Ion-pairing between the solute ion X- and the pairing ion Y+ (characterized by KYy),
followed by partition of the ion-pair XY over the two phases (characterized by K X Y ) .
- Ion-exchange reactions, between solute ions in one phase and pairing ions in the other
(these reactions can be found along diagonal lines in figure 3.28 and can be
characterized by the ion-exchange equilibrium coefficients K g g and Q).
Other reactions, such as the distribution of the pairing ion over the two phases and the two
possible ion-exchange reactions in which the pairing ion and not the solute ion transfers
from one phase into the other can also take place, but do not have a direct effect on the
retention of the solute X.
Figure 3.28 reduces to the simple mechanism of figure 3.27 if both K, and K are very
small. If K , is small (i.e. the solute molecule is mainly in the mobile phase) but K is large
(the pairing ion is mainly absorbed into the stationary phase), then the mechanism of
retention in IPC becomes similar to that of IEC. Typical ion-pairing as well as typical
ion-exchange mechanisms may play a role in practical IPC systems.
Normal phase or reversed phase
Both normal and reversed phase separations are possible in IPC. Normal phase systems
(NP-IPC) are usually LLC systems. The aqueous phase, containing the pairing ion, is
coated onto a silica surface. In order to change the kind or often even the concentration
of the pairing ion it is necessary to coat another column. In some cases the pairing ion may
be dissolved in the (organic) mobile phase (e.g. long chain fatty acids or amines), but the
equilibration of the system will still take a long time. As in all other LLC systems, adequate
thermostatting is vital for the performance of the system.
A major advantage of the use of normal phase systems may be the possibility to use
UV-absorbing (or even fluorescent) pairing ions for the separation of non-UV absorbing
solutes. If a UV absorbing pairing ion is in the (aqueous) stationary phase and if this is
subsequently eluted from the column as an ion-pair in the presence of sample ions, then
a very sensitive detection may be possible (see e.g. ref. [379]).
The mechanism of NP-IPC will be very similar to that of ion-pair extraction, i.e. the
simple mechanism of figure 3.27.
Reversed phase ion-pair systems (RP-IPC) could be of the LLC type, but the use of
chemically bonded (alkyl) phases has become increasingly popular, because of the
increased stability and flexibility of the system. Even if an LLC system is used for RP-IPC,
then a chemical modification of the surface is still required to coat an organic liquid on
the particles of (for example) silica.
The system is usually equilibrated by adding the pairing ion to the mobile phase and
pumping this through until a stable baseline is obtained. Equilibration of reversed phase
95
columns with pairing ions may take up to several hours, depending on the hydrophobicity
(chain length) of the pairing ion and on the flow rate. This is much longer than the
equilibration times in regular RPLC (without the use of pairing ions), but compares
favourably with the effort needed to change the stationary phase in NP-IPC.
The most common way to create an RP-IPC system is to use a genuine chemically
bonded reversed phase column (e.g. C18; see section 3.2.2.1) and to use large pairing ions
with a hydrophobic alkyl chain dissolved in the mobile phase. This technique was
introduced by Knox and Laird, who named it soap chromatography [380]. Because of the
usuatly long alkyl chains of the pairing ions, the use of C18 phases is to be recommended
in order to avoid effects that are related to the critical chain length (see section 3.2.2.1).
Table 3.7 summarizes the advantages and disadvantages of NP-IPC and RP-IPC.
Because of the ease of operation, the latter technique is currently by far the more popular
one. Because of this, most of the following discussion will be focussed on RP-IPC.
Table 3.7:
Comparison of NP-IPC and RP-IPC systems
NP-IPC RP-IPC
Primary parameters Organic phase pH; pairing ion con-

centration and chain
length; counterion
concentration
Organic phase Variable Fixed
Temperature control Critical Not critical
Change kind of pairing ion Difficult Easy (1)

Change concentration of pairing ion Difficult Easy (1)
Prevailing mechanism IP-extraction Dynamic IEC
(1) Equilibration time increases with increasing hydrophobicity (chain length) of the pairing ion.
Effect of pairing ion concentration
Because of the complexity of the IPC mechanism, the effect of the counterion
concentration is not usually as simple as was suggested by eqm(3.91). One reason for this
is that the distribution isotherm for the pairing ion Y is not linear, i.e. K in figure 3.28
is not a constant. Some typical distribution isotherms are shown in figure 3.29.
The figure shows that the initial addition of smalt amounts of pairing ion to the mobile
phase leads to a predominant absorption of these ions in the stationary phase. However,
the stationary phase quickly becomes saturated, the distribution curves flatten and K
decreases. Therefore, the effect of changes in the concentration of the pairing ion is bound
to be very different at low Concentrations than it is at higher ones. For this reason,
96
1 10 pm 100
Figure 3.29 Examples of distribution isotherms of a pairing ion (tetrabutylammonium) in a reversed

phase system. Left: logarithmic representation; right: conventional (Langmuir) representation.
Stationary phase: Lichrosorb RP-18. Mobile phase: indicated percentages of methanol in aqueous
phosphate buffer (25 mM H,PO, and 25 mM NaH2P0,, pH = 2.1-3.4, bromide concentration 200
mM. Temperature: 25 "C. Figure taken from ref. [381]. Reprinted with permission.
Crombeen et al. [382] and Bartha er al. [381] have suggested that the concentration of
pairing ions in the stationary phase be used as the main descriptive parameter in IPC.
However, this parameter is not very convenient in practice, because the distribution
isotherms need to be known.
In practice, the retention (In k) is usually plotted against the logarithm of the
concentration of the pairing ion in the mobile phase (In [ fl). In such plots straight lines
(or slightly curved but monotonic lines) are usually observed over wide ranges in
concentration. An example is shown in figure 3.30.
Kind of pairing ion
Some examples of typical pairing ions are given in tabel 3.8. An extensive list of pairing
ions and their applications can be found in ref. [367].
The kind of pairing ion will mainly be determined by the nature of the sample ions.
Clearly, for anionic samples a cationic pairing ion will be required and vice versa. Also,
if the ions to be separated are small polar ions, then a pairing ion with a large, non-polar
group should be used to enhance the extraction of the ion-pair in the organic phase. If the
sample ions are large, then small pairing ions may be used.
A combination of two large ions is also sometimesused. The resulting ion-pair may have
its charge deeply buried, so that it can be extracted with very non-polar solvents. This
possibility is of more interest for NP-IPC than it is for RP-IPC, because of the flexibility
to choose an appropriate organic solvent in the former technique.
The nature of the pairing ion greatly affects retention, but it will also affect theselectivity
(e.g. ref. [384]). Each pairing ion will require a different concentration to yield capacity
factors in the optimum range.
97
-----
I
0
I
2
I
5
I
10
I
20
RnlmM - I
50
Figure 3.30: Example of variation of retention in IPC with the concentration of the pairing ion
(octanesulfonate) in the stationary phase (P; upper of the two horizontal axes) and in the mobile
lower axis). The axis for P,is linear, while the one for P,,,is not. Stationaryphase: Hypersil
phase (P,,,;
O D s . Mobile phase: lOoh methanol in water.
Solutes: 1 = homo-vanillic acid, 2 = 5-hydroxyindol-3-aceticacid, 3 = 3,4-dihydroxyphenylacetic
acid, 4 = tyrosine, 5 = L-DOPA, 6 = dopamine, 7 = octopamine, 8 = adrenaline, 9 =
3,4-dihydroxymandelicacid, 10 = noradrenaline.
Figure taken from ref. [383]. Reprinted with permission.
Table 3.8: Examples of pairing ions for IPC.
Pairing ion Structure Type
Cations:
Tetra-alkylammonium (1) NR: Strong
Tri-alkylammonium NHR: Weak
A1kylammonium NH,R: Weak
Anions:
Chloride c1- Strong
Bromide Br- Strong
Iodide I- Strong
Perchlorate ClO, Strong
Alkyl sulfonate
Toluene sulfonate RSO, Strong
Naphthalene sulfonate
Alkyl sulfates RSO, Weak
Alkyl phosphates RHPO, Weak
(1) Either symmetrical ions, such as tetrabutylammonium, or non-symmetrical ones, such as

cetyltrimethylammonium,may be used.
98
1.6-
log k
I 0.8 -
0-
0 -
7 9 11 13
chain length -
Figure 3.31 :Example of the effect of the chain length of the pairing ion on retention in IPC. Pairing
ions: alkyl-benzyl-dimethyl ammonium with different lengths of the alkyl chain.
Organic phase: chloroform; Aqueous phase: 0.1 mM of pairing ion in water. Solutes: sodium
cromoglycate (circles) and acid red dye I (triangles). Figure taken from ref. [385]. Reprinted with
permission.
Influence of the chain length of the pairing ion
A way to influence retention without greatly altering the selectivity is to use different
pairing ions from the same homologous series. This offers the possibility to vary the
retention more or less continuously. An example of the effect of the chain length of the
pairing ion is shown in figure 3.31. It is seen in this figure that approximately straight lines
are obtained if the logarithm of the capacity factor is plotted against the chain length of
the pairing ion.
The length of the pairing should be such as to create a stable system with a good capacity.
This implies that one should work on the plateau of the distribution isotherm (figure 3.29).
The required chain length will depend on a series of factors, including the type of the
pairing ion and the modifier content of the mobile phase.
Organic modifiers
Organic modifiers may affect both the retention and the selectivity in RP-IPC. Most of
all, the amount of modifier added determines the distribution isotherm of the pairing ion
and therefore the retention. A secondary factor involves the solubility of the pairing ion
in the mobile phase. Large pairing ions may require large amounts of organic modifier to
allow them to be dissolved sufficiently in the mobile phase.
In principle, the type of organic modifier(s) may be used as a parameter to optimize the
selectivity of the RP-IPC system. However, because of the wide choice of other parameters,
this possibility has not yet been extensively investigated.
99
Interdependent parameters
From the preceding brief discussion it is apparent that three parameters in RP-IPC are
highly intertwined and cannot be chosen independently:
1. the type and concentration of the pairing ion
2. the chain length of the pairing ion
3. the organic modifier content of the mobile phase.
To find a compromise for a mobile phase with neither too large a chain length (because
of slow equilibration) nor too high a modifier content (because of the suppression of
ionization), but yet optimum capacity factors and stable operating conditions is an
optimization problem on its own.
As soon as weak solute ions or weak pairing ions form part of the system, the pH is a
vital parameter in the optimization of IPC separations. Often it is a very selective
parameter, when the different solute ions have different p K , values. The use of silica based
reversed phase packings limits the variations of the pH to the range between about 2 and
7. Because of this limited range of operation, basic solutes often require the use of ion
pairing reagents in RPLC, because they are fully ionized in the practical pH range, and
give rise to highly non-symmetrical peaks in conventional RPLC.
Kind of buffer
The buffer used in RP-IPC has a minor effect on the selectivity. Therefore, the choice
of the buffer will be mainly determined by practical considerations, especially by the
solubility in the mobile phase. Phosphate and citrate buffers allow a wide range of pH to
be used. Acetate buffers are also frequently employed.
The buffer concentration should be sufficient to yield a stable system after injection of
the solute, but has an otherwise negligible effect on retention and selectivity if the
counterion concentration is kept constant (see below).
Counterion concentration
As in IEC, the counterion concentration has a considerable effect on the retention in

IPC. In IPC the counterion is charged similar to the solute molecules, but opposite to the
pairing ion. For example, for the separation of anionic solutes, the pairing agent may be
a sodium sulfonate, in which the sulfonate is the pairing ion and sodium the counterion.
The addition of a buffer salt (e.g. sodium phosphate) and a neutral salt (e.g. sodium
bromide) may also contribute to the concentration of the counterion. Because of the
similar retention mechanism, the counterion concentration has a similar effect on
retention and selectivity in RP-IPC as in IEC.
The practice to keep the total concentration of the counterion constant by adding
varying amount of the neutral salt leads to a more regular behaviour of the RP-IPC system
and is therefore to be recommended [386].
100
Temperature
The temperature wiil affect both retention and efficiency in IPC, but not to the same
extent as it does in IEC. IPC is usually a much more efficient technique (in terms of plate
counts) than is IEC and therefore ambient temperatures usually yield satisfactory results.
Summary
The preferrred form of ion-pairing chromatography is RP-IPC. This is a complex

chromatographic method. Unlike in other kinds of chromatography, there is notone but
a series of primary parameters (type, concentration and chain length of the pairing ion,
pH, organic modifier content) which together determine the capacity factors, but also may
give rise to large variations in the selectivity. As in RPLC, the type of organic modifier may
be used to optimize the selectivity of the system.
3.4 SUPERCRITICAL FLUID CHROMATOGRAPHY (SFC)
In supercritical fluid chromatography (SFC), the mobile phase is neither a gas nor a
liquid. The definition of a supercritical fluid can be illustrated in a classical p- T diagram
such as figure 3.32. In this figure three phases are located in their specific domain: the solid
phase (S), the gas phase (G) and the liquid phase (L). At the triple point (tp) all three phases
may coexist. The line that separates the gas and the liquid phase is the vapour pressure
curve. This line ends at the critical point cp. Above this point, no distinction can be made
between the gaseous and the liquid state. A compound for which either the pressure or the
temperature is above the critical value can therefore not be identified as a liquid or a gas.
The term supercritical fluid is usually reserved for those conditions at which both the
pressure and the temperature exceed the critical value. This area in figure 3.32 can be found
in the top right comer and is bounded by the two dashed lines. These dashed lines do not
correspond to a phase change between conditions on either side. As was shown by Lauer
et al. [387], there is a gradual change in the solvent properties (e.g. density, viscosity) if we
T-
Figure 3 . 3 2 Phase diagram illustratingthe domains of the solid (S), gaseous ( G )and liquid (L) phases
as a function of pressure (p) and temperature (7). tp is the triple point, at which three phases co-exist.
cp is the critical point, which forms the end of the vapour pressure curve (between tp and cp). The
area in the top right corner indicated by SF represents the domain of supercritical fluids.
101
pass the dashed lines under conditions of constant pressure (moving horizontally in figure
3.32) or constant temperature (moving vertically).
Retention in supercritical chromatography is affected by the nature of both the mobile
and the stationary phase. A variety of stationary phases, including high boiling liquids,
polymer films, solid supports and chemically bonded monolayers, has been used.
The choice of possible mobile phases is more limited. The critical properties (critical
pressure pc and temperature TJ should be within practical reach. Moreover, stable
compounds are required, which do not show disintegration at elevated temperatures and
pressures. Also, the mobile phase must not be too agressive towards the materials used in
the column (usually silica-based phases) and the instrumentation (mainly stainless steel).
Therefore, mobile phases that are extremely interesting from a chemical point of view, such
as supercritical ammonia and, especially, supercritical water, have found little use so far.
Table 3.9 lists some possible mobile phases for SFC together with their chemical
properties.
There are several reasons why SFC may gain its place as a separation technique
alongside GC and LC in the years ahead. From a fundamental point of view, the diffusion
coefficients under typical SFC conditions are lower than those typically encountered in
gases, but higher than those found in liquids. The viscosity of supercritical fluids is usually
higher than that of typical gases, but much lower than that of common liquids. At the same
time, supercritical fluids are good solvents for many low-volatile solutes, which are not
compatible with GC. Therefore, SFC may offer the possibility to separate non-volatile
Table 3.9:
Some suggested solvents for SFC. Data taken from refs. [388] and 13891. Asterisks indicate
preferred solvents.
Carbon dioxide* 31.0 72.8

Nitrous oxide 36.4 71.5
Ethane 32.2 48.2
Ethene 9.2 49.7
n-Butane 152.0 373
&Butane 134.9 36.0
n-Pentane* 196.5 33.3
n-Hexane 234.2 29.3
Diethyl ether 193.5 35.9
THF 267.0 51.2
Ethyl acetate 250.1 37.8
Acetonitrile 274.8 47.7
Methanol 239.4 79.9
2-Propanol 235.1 47.0
Ammonia 132.4 111.3
Water 374.1 217.6
102
samples much faster and/or more efficiently (higher numbers of theoretical plates) than
with LC.
From a practical point of view, SFC may allow the use of many different detection
principles, including both typical LC detectors (UV-absorbance, fluorescence)and typical
GC detectors (flame ionization, mass spectrometry). Also, capillary SFC seems to be well
within the posssibilities of current technology, while capillary LC is not.
Promising applications of SFC include group separations (paraffins, olefins and
aromatics) in petrochemical samples, monitoring of supercritical extraction processes
(caffeine from coffee, nicotine from tabacco) and oligomer separations. However, it is in
the field of applications that SFC has yet to prove its value. Unique separations that can
be accomplished with SFC, but not with either GC or LC, have yet to be demonstrated.
The mobile phases that have been used most extensively to date are n-pentane and
carbon dioxide. Pentane has the advantage that it is a liquid under ambient conditions,
so that it can be handled and pumped in the same way as mobile phases for liquid
chromatography. On the other hand, its critical temperature is relatively high (almost 200
"C) and it is a highly inflammable compound.
Carbon dioxide has a vapour pressure of about 50 bar at room temperature. It is
therefore more difficult to handle, and it can only be pumped as a liquid when it is cooled
down to sub-ambient temperatures. However, carbon dioxide is non-flammable and
non-toxic, which makes it very attractive from a practical point of view. Also, the critical
properties of carbon dioxide are very mild.
Mobile phase density
The main factor that influences the retention in SFC is the density of the mobile phase.
For a given eluent, the density is a function of the pressure and the temperature. At a given
temperature, the retention varies with the pressure in a rather coomplicated way, as is
illustrated for the retention (In k) of naphthalene using CO, as the mobile phase in figure
3.33a. Figure 3.33b represents the same data, but now retention is plotted against the
density of the mobile phase. It is seen that smooth curves are obtained, which hardly vary
with the temperature. Of course, to obtain the same mobile phase density, a much higher
pressure is required when the temperature is increased from 35 to 50 OC.
Figure 3.33b is seen to be very similar to a typical plot of retention (In k) vs. composition
(p) in RPLC (see e.g. figure 3.14). Hence, a quadratic equation may be used to describe
the relationship between retention (In k) and density @) in S F C
(3.92)
Composition of the mobile phase
Both pentane and carbon dioxide are solvents of low polarity. The polarity may be
increased by the addition of suitable modifiers to the mobile phase. Such modifiers have
a pronounced effect on the retention. The decrease in retention upon the addition of
modifiers seems to resemble what is observed in LSC (see section 3.2.3). Apart from the
effect on retention, the addition of polar modifiers to the mobile phase also has a marked
effect on the peak shape. Especially in the case of more polar solutes, the addition of
103
modifiers to the mobile phase has become increasingly popular. The nature and
concentration of organic modifier is a parameter that may be used for the optimization
of SFC separations. Some initial work in this direction has been reported by Randall [392].
-2.51
0 30 do 50 60 70 80 90 100 110 120
platm -
3.0-
2.5 -
2.0 -
1 1.5-
-
logk
1.0 -
0.5 -
0.0 -
0.5 -
-1.0 -
-1.5 -
2
-0.-
-2.0 -1.6 -1.2 -0.8 -0.L 0
Figure 3.33: Retention (In k ) as a function of (a) pressure, (b) mobile phase density and (c) the
logarithm of the mobile phase density in SFC at three different temperatures.Mobile phase: carbon
dioxide. Stationary phase: ODS. Solute: naphthalene. Figure taken from ref. [390]. Reprinted with
permission. Experimental data from ref. 13911.
104
Stationary phase effects
It is not yet clear what type of stationary phase will be most useful for SFC. Liquid
stationary phases will almost inevitably be of insufficient stability. Polymeric films of
various thickness and varying degree of cross-linking have been used. Preferably, such
polymeric phases should be covalently bonded to the column wall (open columns) or a
solid support (packed columns). Alternatively, solid adsorbents or chemically bonded
monolayers may be applied.
To a first approximation [393] the selectivity (a) on a given stationary phase may be
expected to be independent of the mobile phase density. Consequently, the problem of
stationary phase selection is similar to that encountered in GC. In GC each stationary
phase will require a given temperature at which the capacity factors are in the optimum
range. In SFC, each stationary phase will require a given mobile phase density. Different
phases may be compared at their individual optimum conditions.
3.5 CLASSIFICATION OF PARAMETERS
In this chapter we have discussed the parameters that affect the selectivity in various
chromatographic methods. The parameters we have encountered can roughly be divided
into three categories:
1. Thermodynamic parameters (T). These include temperature and pressure.
2. Stationary phase parameters (S), which include the nature and composition of the
stationary phase.
3. Mobile phase parameters (M) which include the nature and composition of the mobile
phase, pH, nature and concentration of additives such as buffers, salts, ion-pairing
agents or complexing agents.
Two further categoriesof parameters may be defined, which do not affect the selectivity:
4. Capacity parameters (C) i.e. those parameters which affect the phase ratio: film
thickness, surface area, column diameter (open columns), porosity (packed columns).
5. Physical parameters (P) column length, flowrate, particle size, column diameter
(packed columns).
The capacity parameters do not affect the selectivity (a),but they do have an effect on
the capacity factor ( k ) and hence on the resolution (Rs;see eqn.1.22). The physical
parameters only affect the resolution through the efficiency ( N ) . They also have an effect
on the retention time through the hold-up time to (see eqn.l.6).
Physical parameters may be used to trade off increased resolution against decreased
analysis time. Ideally, this is done separately from the selectivity optimization process,
because the effects are simple, predictable, and independent of the parameters that do
affect the selectivity.
Independent parameters with a simple effect on the resolution may be optimized
sequentially, i.e. one after the other (section 5.1.1). Hence, after the selectivity has been
optimized, the shortest possible column length or the highest possible flowrate may be
established that will provide sufficient resolution. Adapting the length of the column is the
preferred strategy, because it will lead to both faster analysis and lower pressure drops.
105
Some further comments on optimization of the physical parameters will be made in
chapter 7.
The capacity parameters allow a variation of the capacity factor (and hence the
resolution) independent of the selectivity. However, all these parameters are difficult to
vary, since they almost always require new columns to be used. Moreover, the range of
variation offered by these parameters is too limited for them to be generally useful in
optimization schemes (see also section 4.2.3).
Therefore, the thermodynamic parameters (T), the stationary phase parameters (S) and
the mobile phase parameters (M) are the ones we should consider if we wish to select the
most relevant parameters for the optimization of chromatographic selectivity.
Of these three categories the thermodynamic parameters can be varied most easily.
However, temperature has a major effect on retention in GC,but only a minor effect on
the selectivity. In LC its effect is never very large, except from some ionic separations.
Pressure is only relevant as a parameter for SFC.We conclude that temperature should
head the list of optimization parameters in GC,and pressure (possibly in combination with
temperature) in SFC.
Stationary or mobile phase optimization
In GC we cannot use the nature and composition of the mobile phase to vary the
selectivity. Hence, the nature and consequently the composition of the stationary phase
should be used for optimization purposes. There are many more stationary phases
Table 3.10a:
Summary of parameters for selectivity optimization. Asterisks indicate preferred parame-
ters.
Method: Gas Liquid Chromatography (GLC)

Gas Solid Chromatography (GSC)
Section: 3.1.1 and 3.1.2
Primary parameter(s) Suggested relationship
Type Parameter Plot Shape Eqn.no.
Tc Temperature * In(k/T) vs. 1 / T linear 3.10

Cc Film thickness k vs. d,y linear
Cd Surface area k vs. s linear
Secondary parameter(s) Suggested relationship
Sd Stationary phase
Sc Stationary phase composition k vs. cp linear 3.14
106
Table 3.10b:
ters.
Method: Liquid-Liquid Chromatography (LLC)

Section: 3.2.1
Mc Polarity of mobile phase *

Sc Polarity of stationary phase
Cc Phase ratio k vs. V J V , linear 1.10
Md Nature of mobile phase *

Sd Nature of stationary phase
Tc Temperature In k vs. 1/T linear 3.57
Mc PH (1)
Mc Ionic strength (1)
(1) May be used if the more polar phase is aqueous.
available than could possibly (or, indeed, sensibly)be tried. After some different stationary
phases (see section 2.3.1) have been tried, the possibility of using a mixture of two
stationary phases could be considered as a final step in the optimization.
In LC we have a choice between optimizing the stationary phase parameters or the
mobile phase parameters. Obviously, the latter can be changed more readily. Advantages
of mobile phase optimization are:
1. many parameters offer a great flexibility,
2. the mobile phase can easily be changed,
3. there are good possibilities for automation,
4. the investment required for columns is low.
On the other hand, if the stationary phase parameters are being optimized, other
advantages may occur:
1. stable and reproducible operation with simple mobile phases is possible,
2. mobile phases may be selected with a low cost, low viscosity and low toxicity.
Hence, optimization of the mobile phase offers advantages mainly during the optimization
process, while optimization of the stationary phase offers its main advantages after the
optimization. So far, researchers have been more concerned with the optimization process
107
itself than with the usefulness of their results. Hence, the mobile phase parameters have
been optimized almost exclusively. In the future we may see an increased use of the
optimization of stationary phase parameters, especially on the part of column manufactu-
rers. This will result in the availability of reproducible systems for optimized separations
on dedicated stationary phases for GC as well as for LC.
35.1 Summary of parameters for selectivity Optimization
Table 3.10 summarizes the parameters of interest for the various chromatographic
techniques described in this chapter. A distinction is made between primary and secondary
parameters.
Table 3.10~:
ters.
Method: Reversed Phase Liquid Chromatography (RPLC)

Section: 3.2.2
Mc Mobile phase polarity *

(modifier content) In k vs. cp quadr. 3.38
or In k vs. cp (1) linear 3.45
Mc PH (2) In k vs. pH curved 3.631
68/70
Md Nature of modifier(s) (3) *

Mc Ratio of modifier concentrations.*
Tc Temperature In k vs. 1/T linear 3.57
Sd Nature OF stationary phase
Sd Nature of buffer
Ss Stationary phase chain length In k vs. ne linear (4)
Mc Ionic strength In kvs. I hyperboiic 3.71
Md Nature of buffer
( 1 ) Linear approximation for 1 < k < 10.

(2) pH is a primary parameter for ionizable solutes.
(3) Modifier coontent f a - multicomganent mobile phases can be estimated using table 3.1,
(4) Approximately linear up to “critical chain length” (see section 3.2.2.1).
108
Table 3.10d:
ters.
Method Liquid-Solid Chromatography (LSC)

Section: 3.2.3
T v ~ e Parameter Plot Shape Eqn.no.
Mc Eluotropic strength In k vs. In X , linear 3.74
Secondary parameter@) Suggested relationship
Md Nature of modifier(s)*
Md Modulators (1)
Tc Temperature In kvs. 1 / T linear 3.57
(1) Modulators are compounds which can be added to the mobile phase in small quantities to affect
peak shape and/or selectivity (e.g. water, tri-ethylamine).
Primary parameters are those which have a large effect on retention. Usually, these
parameters d o not affect selectivity to the same extent. Therefore, these are the parameters
that can be used to bring the capacity factors of the solute into the optimum range.
Secondary parameters may affect retention, but always affect selectivity. In fact, ideally
the parameters should be selected such that the retention ( k ) is kept roughly constant (i.e.
in the optimum range) while the selectivity (a)can be varied. If the secondary parameters
d o affect retention, then sometimes this ideal situation can be approached by the
simultaneous variation of two (or more) parameters a t the same time. Examples of this may
be found in chapter 5.
Capacity parameters are not often used as primary optimization parameters in
chromatography. Therefore, they are only included in table 3.10 in those cases in which
they are used with some frequency. It should be noted, however, that changing one of the
capacity factors usually involves the use of a completely different column and is therefore
unattractive. Although changing the capacity parameters affects retention in a n essentially
predictable way, changing the column (packing material, film thickness, etc.) may give rise
to unexpected second order phenomena. This is a second reason for which capacity
parameters should not be recommended as primary optimization parameters.
Parameters that affect neither the capacity factors nor the selectivity (such as column
length or flow rate) will not be found in this table.
The parameters are classified by two different letters. The capital letters correspond to
the classification given above, i.e. thermodynamic parameters (T), stationary phase
parameters (S), mobile phase parameters (M), capacity parameters (C) and physical
parameters (P).
The lower case letters indicate a second classification of the parameters. Three different
types of parameters are indicated:
- continuous parameters (c), which can take on all values between given limits (e.g.
temperature),
- discrete parameters (d), which can only take on certain values (e.g. the kind of
stationary phase),
- stepwise parameters (s), which can take on a series of discrete values (e.g. the chain
length of alkylsulfonate ions in ion-pair chromatography).
Table 3.10a lists the parameters that may be used in the two modes of GC discussed in
this chapter (GLC and GSC). Because of the similarity of these two techniques, they have
been combined in one table.
To optimize the capacity factors, the temperature may be adapted. Temperature is the
most commonly used primary parameter in GC. Alternatively,the film thickness (in GLC)
or the surface area of the stationary phase (in GSC) may be used to change the capacity
Table 3.10e:
ters.
Method Ion Exchange Chromatography (IEC)

Section: 3.3.1
Mc Counterion concentration * In k vs.ln[ v] linear 3.84

Mc pH(l)* In k vs. pH , sigm. 3.85/86
Sd Exchange capacity k vs. capacity linear
Secondary parameter@) Suggested relationship

~
Tc Temperature In k vs. l / T linear 3.57

Md Type of modifier(s)
Mc Concentration of modifier(s)
Md Type of counterion
Md Type of buffer
Sd Type of stationary phase
(1) pH is a primary parameter if weak solutes, counterions or ion-exchangers are involved in the
separation process.
110
Table 3.10f:
ters.
Method: Ion Pairing Chromatography (IPC)

Section: 3.3.2
Mc pH(l)* In k vs. pH sigm.

Ms Chain length of pairing ion * In kvs. n, linear
Mc Pairing ion concentration * In k vs.ln [ r] curved
Mc Modifier content * In k vs. cp quadr. 3.38
Md Type of pairing ion *

Md Type of modifier *
Md Type of counterion
Mc Conc.of counterion
Md Type of buffer
Tc Temperature In k vs. 1 / T linear 3.57
Sd Type of statationary phase
factors. However, in order to change either of these parameters another column is required.
Because of the availability of packing materials, the surface area is a discrete parameter.
The most common secondary parameter is the kind of stationary phase used, which is
obviously a discrete parameter. Because the capacity factors will usually differ on different
phases, several parameters will have to be varied at the same time. For example, if another
stationary phase is chosen, the temperature may be adapted to bring the capacity factors
back into the optimum range.
Table 3.10b lists the optimization parameters that may be used in LLC. Clearly, the
polarities of the two phases largely determine retention and selectivity. The exact
composition of (preferably) the mobile phase may be varied to optimize the separation (i.e.
variations in the nature and the concentration of mobile phase components, without
substantial variations in the polarity). Even if the temperature is not a major optimization
parameter, adequate temperature control is required in all LLC experiments. Therefore it
may be experimentally straightforward to exploit temperature as a secondary optimiza-
tion parameter.
Table 3 . 1 0 ~lists the relevant parameters for RPLC. The polarity (modifier content) of
the mobile phase is the main primary parameter for most samples, although for some
111
weakly acidic or basic solutes the pH may have an even larger effect on the capacity
factors.
A series of secondary parameters may be exploited. Changing the nature of the organic
modifier is the most common and probably the most rewarding parameter to use. If ternary
and quaternary mobile phases are considered, then the ratio between the concentrations
of different modifiers becomes a continuous parameter that may be optimized.
Most separations are optimized by considering solely the influence of the concentration
of modifier(s) in the mobile phase and the pH. However, for particular (or particularly
difficult) separation problems there is a series of additional parameters that might be
considered.
Table 3.10d lists the parameters for LSC. Again, most separations may be optimized by
optimizing the eluotropic strength (primary parameter) and the nature (secondary
parameter) of the mobile phase. The latter parameter involves the preparation of different
iso-eluotropic mixtures containing different solvents, or small quantities of very polar
components ("modulators"). As in the case of RPLC, there are several additional
parameters that are not frequently exploited.
Polar chemically bonded stationary phases (section 3.2.2.2) may be used as an
alternative stationary phase for both RPLC and LSC, if variations in the mobile phase do
not result in an adequate separation. If polar CBPs are used in combination with more
polar mobile phases (reversed phase mode), then table 3.10~may be used to find the most
appropriate optimization parameters. If operated in the normal phase mode, table 3.10d
Table 3.10g: t

ters.
Method: Supercritical Fluid Chromatography (SFC)

Section: 3.4
Mc Density (I) * In k vs. p quadr. 3.92

Mc Concentration of modifier In k vs. Q, curved
Md Nature of mobile phase

Md Type of modifier *
(1) The mobile phase density is determined by the combination of the pressure and the temperature.
112
may be used. However, it should be noted that strictly speaking we do not deal with
liquid-solid chromatography. A notable difference is that the water content of the mobile
phase has a much less dramatic influence in normal phase LBPC than it has in LSC.
The parameters that play a role in ion-exchange chromatography (IEC) are summarized
in table 3.10e. A combined optimization of the two primary optimization parameters (pH
and counterion concentration) may already be used to optimize the chromatographic
selectivity of the system. However, different buffers or counterions are often investigated
for their effect on the selectivity. As a rule, elevated temperatures are used to increase the
efficiency rather than the selectivity of the system.
Table 3.10f lists the most relevant parameters for ion-pairing chromatography (IPC).
Here there are four major primary parameters, which cannot be seen as independent.
Hence (see section 5.1.l), these four parameters should preferably be optimized simultane-
ously. Sensible upper and lower limits may be set for each of the parameters and an
optimized separation may result from the process. If this is not the case, there are still many
secondary parameters that could be exploited.
Tables 3.10e and 3.10f suggest that the ionic separation methods IEC and IPC are the
most complicated ones. Both methods involve several mutually dependent primary
parameters and a series of additional secondary optimization parameters. Therefore, these
techniques are bound to be the subject of many optimization studies in the future.
Finally, table 3.10g shows the relevant parameters for SFC. The density of the mobile
phase (determined by the combination of pressure and temperature) is the main parameter
for this technique. Several possible secondary parameters are listed in the table. Because
SFC is not yet a mature technique, the list of secondary parameters may still undergo some
changes.
REFERENCES
301. A.B.Littlewood, G.S.G.Phillips and D.T.Price, J.Chem. SOC.(1955) 1480.

302. J.Takacs, P-Rajcsanyi, L.Kaplar and LOIacsi, J. Chromatogr. 41 (1969) 438.
303. R.Tijssen, H.A.H.Billiet and P.J.Schoenmakers, J.Chromatogr. 122 (1976) 185.
304. C.-F.Chien, M.M.Kopecni and R.J.Laub,. AnaLChem. 52 (1980) 1402.
305. M.W.P.Harbison, R.J.Laub, D.E.Martire, J.H.Purnel1 and P.S. Williams, J. Phys.
Chem. 83 (1979) 1262.
306. R.J.Laub in: Th.Kuwana (ed.), Physical Methods in Modern Chemical Analysis,
Vo1.3, Academic Press, New York, 1983, p.249 et seq.
307. J.A.Barker and D.H.Everett, Trans. Faraday SOC.58 (1962) 1608.
308. D.C.Locke, J.Phys.Chem. 69 (1965) 3768.
309. E.V.Kalashnikova, A.V.Kiselev, D.P.Poshkus and K.D. Shcherbakova, J.Chroma-
togr. 119 (1976) 233.
310. J.Klein and H.Widdecke, J.Chromatogr. 147 (1978) 384.
31 I. P.J.Schoenmakers, H.A.H.Billiet and L.de Galan Chromatographia 15 (1982) 205.
312. B.L.Karger, L.R.Snyder and C.Eon, AnaLChem. 50 (1978) 2126.
313. J.H.Hildebrand, J.M.Prausnitz and R.L.Scott, Regular and Related Solutions, Van
Nostrand Reinhold, New York, 1970.
314. H.A.H.Billiet, P.J.Schoenmakers and L.de Galan, J.Chromatogr. 218 (1981) 443.
315. J.F.K.Huber, M.Pawlowska and P.Mark1, Chromatographia 17 (1983) 653.
113
316. W.R.Melander and Cs.Horvath, in: Cs.Horvath (ed.), HPLC, Advances and Perspec-
tives, V01.2, Academic Press, New York, 1980, p.113.
317. G.E.Berendsen, Preparation and Characterizationofwell-defined Chemically Bonded
Phases for HPLC, Thesis, Delft, 1980.
318. R.E.Majors, H.G.Barth and C.H.Lochmueller, Anal.Chem. 56 (1984) 300R.
319. G.E.Berendsen and L.d.Galan, J.Chromatogr. 1% (1980) 21.
320. A.F.M.Barton, Chem.Rev. 75 (1975) 731.
321. P.J.Schoenmakers, H.A.H.Billiet, R.Tijssen and L.de Galan, J.Chrornatogr. 149
(1978) 519.
322. P.J.Schoenmakers, H.A.H.Billiet and L.de Galan, J.Chromatogr. 185 (1979) 179.
323. P.J.Schoenmakers, H.A.H.Billiet and L.de Galan, J.Chrornatogr. 282 (1983) 107.
325. P.Jandera, H.Colin and G.Guiochon, Anal.Chem. 54 (1982) 435.
326. W.Melander and Cs.Horvath, Chromatographia 18 (1984) 353.
327. J.W.Weyland, C.H.P.Bruins and D.A.Doornbos, J.Chromatogr.Sci. 22 (1984) 31.
328. M.Mcann, J.H.hrnel1 and C.A.Wellington, Faraday SOC.Symp.Series 15 (1980) 82.
329. W.R.Melander and Cs.Horvath, Chromatographia 18 (1984) 353.
330. Lu Peichang and Lu Xiaoming, J.Chromatogr. 292 (1984) 169.
331. L.R.Snyder, J.W.Dolan and J.R.Gant, J.Chromatogr. 165 (1979) 3.
332. C.E.Goewie, U.A.Th.Brinkman and R.W.Frei, Anal.Chem. 3 (1981) 2072.
333. T.L.Hafkenscheid and E.Tomlinson, J.Chromatogr. 264 (1983) 47.
334. S.T.Balke, Quantitative Column Liquid Chromatography, a Survey of Chemometric
Methods, Elsevier, Amsterdam, 1984.
335. P.J.Schoenmakers, H.A.H.BiHiet and L.de Galan, LChromatogr. 205 (1981) 13.
336. P.J.Schoenmakers, A.C.J.H.Drouen, H.A.H.Billiet and L.de Galan, Chromatogra-
phia 15 (1982) 688.
337. J.R.Gant, J.W.Dolan and L.R.Snyder, J.Chromatogr. 185 (1 979) 153.
338. W.R.Melander, B.-K.Chen and Cs.Horvath, J.Chromatogr. 185 (1979) 99.
339. W.R.Melander, D.E.Campbel1 and Cs.Horvath, J.Chromatogr. 158 (1978) 21 5.
340. S.N.Deming and M.L.Turoff, AnaLChem. 50 (1978) 546.
341. C.J.C.M.Laurent, H.A.H.Billiet and L.de Galan, Chromatographia 17 (1983) 394.
342. Cs.Horvath, W.Melander and I.MolnBr, Anal.Chem. 49 (1977) 142.
343. D.J.Pietrzyk and C.H.Chu, AnaLChem. 49 (1977) 860.
344. M.T.W.Hearn, Adv. Chrom. 18 (1980) 93.
345. J.L.M.van d e Venne, J.L.H.M.Hendrikx and R.S.Deelder, J.Chromatogr. 167
(1978) 1.
346. M.Otto and W.Wegscheider, J.Chromatogr. 258 (1983) 11.
347. J.J.Kirkland, Anal.Chem. 43 (1971) 36A.
348. B.Feibush, M.J.Cohen and B.L.Karger, J.Chromatogr. 282 (1983) 3.
349. L.R.Snyder and J.J.Kirkland, Introduction to Modern Liquid Chromatography, 2nd.
Edition, Wiley, New York, 1979.
350. L.R.Snyder, Principles of Adsorption Chromatography, Dekker, New York, 1976.
351. H.Colin, C.Eon and G.Guiochon, J.Chromatogr. 122 (1976) 223.
352. E.Soczewinski, AnaLChem. 41 (1969) 179.
353. E.Soczewihski and W.Golkiewicz, Chromatographia 4 (1971) 501.
354. P.Jandera and J.ChuraEek, J.Chromatogr. 91 (1974) 207.
114
355. ESoczewinski, J.Chromatogr. 130 (1977) 23.
356. L.R.Syder and H.Poppe, J.Chromatogr. 184 (1980) 363.
357. P.Jandera, M.Janderova and J.ChuraEek, J.Chromatogr. 148 (1978) 79.
358. D.L.Saunders, AnaLChem. 46 (1974) 470.
359. J.L.Glajch, J.J.Kirkland and L.R.Snyder, J.Chromatogr. 238 (1982) 269.
360. L.R.Snyder and J.L.Glajch, J.Chromatogr. 214 (1981) 1.
361. J.L.Glajch and L.R.Snyder, J.Chromatogr. 214 (1981) 21.
362. L.R.Snyder, J.L.Glajch and J.J.Kirkland, J.Chromatogr. 218 (1981) 299.
364. C.J.C.M.Laurent, A Reappreciation of Alumina in Liquid Chromatography, Ph.D.
thesis, Delft, 1983.
365. J.H.Knox, K.K.Unger and H.Mueller, J.Liq.Chromatogr. 6 (suppl.1) (1983) 1.
366. M.T.Gilbert, J.H.Knox and B.Kaur, Chromatographia 16 (1982) 138.
367. F.M.Rabel, Advan.Chromatogr. 17 (1979) 53.
368. P.R.Haddad and C.E.Cowie, J.Chromatogr. 303 (1984) 321.
369. P.B.Hamilton, AnaLChem. 35 (1963) 2055.
370. Sj. van der Wal and J.F.K.Huber, J.Chromatogr. 135 (1977) 305.
371. R.Eksteen, P.Linsen and J.C.Kraak, J.Chromatogr. 148 (1978) 413.
372. Sj. van der Wal and J.F.K.Huber, J.Chromatogr. 102 (1974) 353.
373. C.J.C.M.Laurent, H.A.H.Billiet and L.de Galan, Chromatographia 17 (1983) 394.
374. C.J.C.M.Laurent, H.A.H.Billiet and L.de Galan, J.Chromatogr. 285 (1983) 161.
375. E.Sawicki in: E.Sawicki, J.D.Mulik and E.Wittgenstein (eds.), Ion Chromatographic
Analysis of Environmental Pollutants, Ann Arbor Science, Michigan, 1978, p.1.
376. P.R.Haddad and A.L.Heckenberg, J.Chromatogr. 300 (1984) 357.
377. P.Jandera and J.ChuraEek, Gradient Elution in Column Liquid Chromatography,
Elsevier, Amsterdam, 1985.
378. G.Schill in: J.A.Marinsky and Y.Marcus (eds.), Zon Exchange and Solvent Extraction,
Vo1.6, 1974, Chapter 1.
379. J.Crommen, B.Fransson and G.Schill, J.Chromatogr. 142 (1975) 107.
380. J.H.Knox and G.R.Laird, J.Chromatogr. 122 (1976) 17.
381. A.Bartha and Gy.Vigh, J.Chromatogr. 260 (1983) 337.
382. J.P.Crombeen, J.C.Kraak and H.Poppe, J.Chromatogr. 167 (1978) 219.
383. H.A.H.Billiet, A.C.J.H.Drouen and L.de Galan, J.Chromatogr. 316 (1984) 231.
384. R.Modin and G.Schill, Talanta 22 (1975) 1017.
385. E.Tomlinson, C.M.Riley and T.M.Jefferies, J.Chromatogr. 173 (1979) 89.
386. A.Bartha, H.A.H.Billiet, L.de Galan and Gy.Vigh, J.Chromatogr. 291 (1984) 91.
387. H.H.Lauer, D.McManigill and R.D.Board, AndChern. 55 (1983) 1370.
388. W.Asche, Chromatographia 11 (1971) 411.
389. R.C.Reid, J.M.Prausnitz and T.K.Sherwood, The Properties of Gases and Liquids,
Third edition, McGraw-Hill, New York, 1977.
390. P.J.Schoenmakers, J.Chromatogr. 315 (1984) 1.
391. U.van Wasen, I.Swaid and G.M.Schneider, Angew.Chemie 92 (1980) 585.
392. L.G.Randall, Hewlett-Packard Technical Paper no. 102, 1983.
115
CHAPTER 4
OPTIMIZATION CRITERIA
Before any optimization process can be started, the goals of the process should be
defined unambiguously. For chromatographic separations this is not always a straightfor-
ward matter. A chromatogram is more than a simple unique number in one dimension.
Nevertheless, we want to reduce the information contained in the chromatogram to a
single number during the course of the optimization process.
An additional complicating factor is that the goals of the chromatographic optimization
may vary considerably from one case to another. For example, all peaks may need to be
separated, or just some relevant peaks in a complex chromatogram; large series of
presumably identical samples may have to be run in a quality control situation, or a
screening method may need to be developed for a relatively large number of potentially
present drugs or pollutants.
The aim of this chapter is to translate such different analytical goals into objective
functions, i.e. into different criteria which can be the objective goals of an optimization
process.
In the literature many different terms are used for such criteria: (chromatographic)
response functions, objective functions or (chromatographic) optimization functions.
Throughout the rest of this chapter, the neutral term optimization criteria will be used.
4.1 INTRODUCTION
In this introduction the possible goals of an optimization process will be investigated.

In the following sections we will then try to translate these goals into simple mathematical
algorithms. At the end of this chapter the different goals and the recommended
optimization criteria will then be summarized.
4.1.1 Separation of two peaks
The resolution of two chromatographic peaks has been defined in terms of retention
times and bandwidths in chapter 1:
At
R, =
'/2( w, + w2)
and it was shown that for symmetrical (Gaussian) peaks eqn.(4.l) can be transformed into
a very useful equation (see section 1.5):
The fact that resolution as defined by eqm(4.1) can be related so elegantly to the
fundamental parameters of the separation process (i.e. a, and N ) is a great advantage
for the use of resolution (R,) to quantify the extent of separation of a pair of
chromatographic peaks. However, in opting for R, we need to accept all characteristics
116
of this quantity, which may not always be advantageous. The following three characteris-
tics appear to be relevant:
1. The resolution (R,) is independent of the (relative) height(s) of chromatographic peaks.
This is a fair proposition only if we work in the linear range of chromatographic
operation, i.e. if the peak height increases linearly with the injected quantity and the
peak width remains constant. However, even then differences in the ratio of the two
peak heights will lead to differences in the resolution (subjectively) observed by the
chromatographer. A detailed description can be found in ref. [401], pp.34-48.
2. R , may be expressed in terms of fundamental chromatographic quantities (eqn.4.2) for
symmetrical (Gaussian) peaks, but for such peaks only. Ideally, all chromatographic
peaks fall within the above qualification. However, in the practice of GC and in
particular LC, this is usually not true.
3. R, cannot easily be estimated from a chromatogram using eqn.(4.1), since this requires
knowledge of both retention times and peak widths. Establishing the latter from the
chromatogram requires tedious manual measurements or complex mathematical
algorithms for integrators or computers. In either case, the resulting estimates for the
peak widths are not usually very reliable. The use of eqn.(4.2) with a given number of
theoretical plates yields more reproducible results, but again it assumes the peaks to be
Gaussian. In this case only retention times need to be established from the chromato-
gram. N may be obtained from independent measurements or from a “test chromato-
gram”. It may be a fixed number, but also a function of the capacity factor (k). A
disadvantage of the use of eqn.(4.2) may be a variation of N with time, for instance due
to a gradual deterioration of the column. Such a process is not accounted for if the
resolution is characterized without obtaining an up-to-date measurement for the peak
width.
In section 4.2. we will define other criteria that may be used to characterize the
separation between a pair of adjacent peaks in a chromatogram (so-called elemental
criteria).
From the above it is obvious that the following three questions need to be addressed in
this chapter:
1. Should the criteria used to characterize the extent ofseparation of apair of adjacentpeaks
in a chromatogram be affected by the relative peak heights ?
2. Should the shape of the peaks be reflected in the value of the criterion ?
3. Can a criterion be defined which bears relation to fundamental chromatographic
parameters, but is yet conveniently obtained from the chromatogram ?
4.1.2 Separation in a chromatogram
To some extent the quantification of the amount of separation in a chromatogram can

be seen as an expansion of the characterization of the separation achieved for each pair
of successive peaks. However, a straightforward expansion of a criterion for a pair of
peaks, for instance a summation of individual R, values, may easily yield numbers that
do not at all correspond to the chromatographer’s own (subjective) opinion of what
constitutes a good chromatogram.
This is easily illustrated by the two chromatograms shown in figure 4.1. These two
117
(a1 1
(bl
1
0 1 5
k-
Figure 4.1: Two schematic chromatograms,constructed with N = 10,000. The capacity factors of the
peaks in the chromatograms are listed in table 4.1.
Table 4.1:
Resolution data for the chromatograms of figure 4.1
Chromatogram Peak no. k RS = R S
1 1
1.22
4.1.a 2 1.I 2.9
1.72
3 1.25
1 1
1.22
4.1.b 2 1.1 25.3
24.07
3 5
chromatograms are identical, apart from the position of the last peak. Table 4.1 lists the
k values and the resolution factors (calculated from eqn.4.2) which correspond to the two
chromatograms of figure 4.1. Also given in the table is the sum of all resolution factors
in each chromatogram. Due to the improved resolution of the last two peaks, the sum of
the resolution values is much higher for the bottom chromatogram, suggesting this one to
118
be vastly superior. However, as long as all peaks are of equal importance, every
experienced chromatographer would prefer the top chromatogram to the bottom one,
because it would yield a much shorter analysis time.
The condition that all peaks must be of equal importance is relevant in this context. For
example, an analyst who is only interested in the quantization of the last peak in the
chromatogram might well prefer the bottom one. On a very much shorter column this
would yield a very fast resolution of the last peak from the (unresolved) rest of the
chromatogram. In the case that all peaks are (or must be) considered to be of equal
importance, we will speak of the general case. If only a few peaks are of interest, or if some
peaks are of more importance than others, we will speak of specific cases.
The above discussion has led to the following three questions:

1 . How can we expand criteria that measure the resolution between apair of succesivepeaks
to criteria that measure the quality of separation achieved in an entire chromatogram?
2. How should analysis time be reflected in such criteria ?
3. May we use the same criteria for the general case and for specific cases, or how should
we adapt the criteria to serve these dzfferent purposes?
4.2 ELEMENTAL CRITERIA
The resolution between two peaks has been defined in chapter 1 and this definition has
been reviewed in section 4.1.1. In this section we will define and investigate various other
criteria that may be used to quantify the extent of separation between a pair of adjacent
peaks in a chromatogram. We will refer to these criteria as “elemental criteria”. Later in
this chapter the elemental criteria will serve as the basis of criteria for judging the extent
of separation in entire chromatograms.
4.2.1 Peak-valley ratios
Three definitions of peak-valley ratios are illustrated in figure 4.2. All of them express
the extent of separation as some measure of the depth of the valley between two peaks
divided by some measure of the peak height. The first criterion (P) measures the depth of
the valley relative to the interpolated peak height as shown in figure 4.2.a. The
corresponding expression is:
P = f/g (4.3)
where g is the interpolated peak height, i.e. the height to the baseline of the line connecting
two peak tops at the location of the valley, and f i s the depth of the valley relative to the
interpolated baseline. This criterion was suggested by Kaiser [402].
The second peak-valley ratio was suggested by Schupp [403] and is illustrated in figure
4.2.b. In this criterion, which we will refer to as the median peak-valley ratio (Pm),the depth
of the valley at a point midway between two successive peaks urn) is measured relative to
the average peak height (g,) which equals the interpolated peak height at that point. The
corresponding equation is:
119
\ P=flg
Figure 4.2: Three definitions for peak-valley ratios as elemental criteria to quantify the extent of
separation between a pair of adjacent peaks in a chromatogram. (a) Peak-valley ratio ( P ; eqn.4.3)
accordingto Kaiser, (b) median peak-valleyratio (P,; eqn.4.4)according to Schupp and (c) (opposite
page) the valley-to-top ratio ( P ; eqn.4.5) according to Christophe.
120
p, = f , / g , . (4.4)
The third criterion, the valley-to-top ratio ( Pv), was introduced by Christophe [404]. It
measures the height of the valley relative to the height of either of the peak tops. Hence,
for a cluster of two peaks as in figure 4.2, two values of P, can be obtained, one for each
peak. If the ratio of the height of the valley (v) to the peak height (h) is subtracted from
unity, the resulting definition is very similar to the previous two. It is illustrated in figure
4.2.c and the appropriate equation for P, is
P, = 1 - v / h (4.5)
The parameters used in eqns.(4.3), (4.4) and (4.5) are all illustrated in figure 4.2.
All three definitions for peak-valley ratios are very similar. According to the last
definition, a value for the valley-to-top ratio (P,)can be assigned to each peak rather than
to each pair of peaks. However, in the case of two Gaussian peaks of equal heights all three
definitions yield exactly the same results. Even if the relative peak heights vary, the first
two definitions will still yield comparable results. The definition for P, implies that the
value will be higher for the larger peak and lower for the smaller peak (proportionally to
the relative height).
The peak-valley ratios vary from zero for separations where no valley can be detected,
to unity for complete separation. It ought to be noticed that a P value equal to zero does
not necessarily imply that two solutes elute with exactly the same retention time (or k
value). There is a threshold separation below which the presence of two individual bands
in one peak only leads to peak broadening or deformation, without the occurrence of a
valley. In these cases R , values are indeed not equal to zero, because by definition
(eqn.l.14) R, is proportional to the difference in retention times.
121
Three characteristics of peak-valley ratios are:
1. P can readily be estimated from a chromatogram.
2. In theory P will vary with varying relative peak heights (or areas) of the two peaks
involved. A simulation for Gaussianpeaks reveals that this variation is small for
' both P and P,, but it is substantial for P, [405].
3. Because of its pragmatic definition P automatically reflects peak asymmetry and it
can be applied to peaks of all shapes, not exclusively to Gaussian peaks.
For Gaussian peaks of equal height the value of the peak-valley ratio (then the same
according to all three definitions) can readily be expressed in terms of R,. This can be done
by relating the parameters f, g and v (see figure 4.2) to the parameters that describe a
Gaussian peak (ISand h). For the first of a pair of Gaussian peaks (peak A) we can write
(eqn.1.15):
A ( t ) = h, exp -'I2 - (4.6)
while a similar expression holds for peak B.

If we now substitute t = 1/2 ( t , + tB) and assume the values of IS for close peaks to
be approximately equal (cr, z 0, FS 5)we find for the combined signal v (see figure 4.2.c):
' A +'B
v =g-f=A(*)+B(.) t + t ,
= h, exp -'/2 (t;i:)

- +
2
h,exp -'/2
2
2
w (hA + h,) exp -'I2
= (h, + h,)exp -(2 R:)
and given that
g = (h, + h,) / 2
eqns.(4.7) and (4.8) can be combined to yield
(4.10)
122
Eqn.(4.10) gives the relationship between P and R , for two Gaussian peaks, assuming
that oAw a, and assuming that the position of the peak tops is not significantly altered
because of peak overlap. For Gaussian peaks of equal height P= P,= P, and eqn.(dlO)
applies to all three criteria. Calculations performed on simulated Gaussian peaks [405]
confirm that both P and P, closely follow the theoretical curve described by eqn.(4.10),
even when the relative peak heights vary. Clearly, for non-symmetrical peaks (e.g. typical
solvent peaks) the value of P will be affected by the peak heights. However, if the first of
a pair of peaks is a “solvent peak” it may be well-nigh impossible to use any of the
definitions in figure 4.2 to establish either the peak-valley ratio (P) or the median
peak-valley ratio ( P A from the chromatogram. Only a pragmatic ratio between the top
of the observed peak and the height of the valley on the solvent front preceding the peak
may be established from the chromatogram.
Eqm(4.10) also provides insight into the threshold value for P for symmetrical peaks.
According to eqn(4.10) P will be estimated as zero for
or R,y < 0.59 . (4.1 1)
This figure applies to Gaussian peaks, but clearly, for peaks of other shapes there will
also be some threshold value below which changes in the extent of separation will not be
reflected in P.
Wegscheider et al. [406] have modified P so that it will also reflect baseline noise:
P’ =f /(g+2n) (4.12)
where n is the (peak-to-peak) noise level on the baseline. According to eqn.(4.12), P‘ will
decrease when the noise level increases, as well as when the absolute peak heights (reflected
in f and g ) decrease. If noise is a significant factor, eqn.(4.12) may provide a more realistic
evaluation of the merits of the actual separation than does eq~(4.3).Eqm(4.4) and (4.5)
can be modified analogously to account for the influence of baseline noise on P, and P,.
Because of the great similarity between the definitions for P and P,, we will not try to
establish superiority of one over the other. To a large extent, the choice for one of them
will depend on the software that is available to obtain P values from a chromatogram.
The choice between P or P, on the one hand and P, on the other will be determined
by whether or not an influence of the (relative) peak heights is wanted (see discussion
below).
4.2.2 Fractional peak overlap
An obvious criterion by which to judge the extent of separation of chromatographic

peaks, especially for the optimization of a quantitative analysis, is the fraction of the peak
that is free of overlap from adjacent peaks. The definition for this so-called fractional
overlap criterion is illustrated in figure 4.3. An equation to describe the fractional
overlap is
123
(4.13)
where A, is the area of the nth peak and A,,,.-, and A,.,, , are the areas it shares with
the preceding and the following peaks, respectively.
An -An,n-1 -An.n+l
FO =
An
Figure4.3: Illustration of the definition of the fractional overlap (FO) as a criterion for the separation
of a pair of adjacent peaks in a chromatogram.
Clearly, FO gives a good indication of the accuracy with which a peak can be
quantitatively determined in a chromatogram. However, it is not the same as the error
involved in quantitative analysis. The latter is affected not only by the extent of separation
(reflected in FO), but also by the algorithms or programs used to establish the peak area.
If the peaks are assumed to be Gaussian and if the exact peak positions and peak widths
are known (the latter are very difficult to obtain accurately from a chromatogram), then
FO can be calculated. But even then, the calculation is fairly complicated and simple
equations relating FO to the difference in retention times and the standard deviations of
the two peaks cannot be derived. For non-Gaussian, non-symmetrical peaks FO can only
be estimated if the profiles of each of the individual peaks in the chromatogram can be
established. This can be done in a purely mathematical way by “deconvolution”. This
requires some mathematical function that describes the shape of the real peak with some
degree of accuracy, and preferably also knowledge of the number of peaks actually present
in the part of the chromatogram. It also requires complex computer programs.
A more practical way to obtain the profiles of the individual peaks may be a sensible
application of modem multichannel detection techniques (see section 5.6.3). It should be
noted that neither mathematical deconvolution nor multichannel detection can be a
124
substitute for chromatographic separation. They only serve to illustrate that sophisticated
techniques are required if FO is to be used as a criterion by which to judge the separation
between adjacent peaks in a chromatogram. At present, therefore, the FO criterion seems
to be a merely theoretical proposition.
4.2.3 Separation factor
In chapter 1 (eqn.l.20) we have seen that the resolution (R,) can be described as the
product of two factors, one covering the chemical and physical characteristics of the
separation and one reflecting the column efficiency:
k,-k,
.-fi (4.14)
R s = k,+k,+2 2
The first factor in eqn.(4.14) combines the effects of the capacity factor ( k ) and the
selectivity (a)on the resolution. To some extent, k and a can be varied independently for
the purpose of optimization. Notably, k varies with the phase ratio (eqn.l.10) while adoes
not. Hence, if the largest value for a is observed in conditions where k values are either
too high or too low, variations in the phase ratio may be used to realize an optimum
separation.
The parameters that can be used for this purpose have been classified as “capacity
parameters” in section 3.5. These parameters and the ways in which they affect the capacity
factors are summarized in table 4.2.
Table 4.2:
Summary of parameters which affect retention (k), but do not affect selectivity (a). The
proportionalities given assume all other parameters to be constant.
Stationary phase
Liquid Solid; CBP
Open columns Column diameter

( k cc dc-’)
Packed columns Surface area

( k a S,)
I (k a E-’)
For various reasons the “capacity parameters” listed in table 4.2 will not often be used
to optimize k values.
125
In the first place column characteristics will often be determined by practical
conditions, such as availability of columns and materials and instrumental considerations.
In the second place, the parameters listed in table 4.2 cannot always be varied
independently and, moreover, will have side effects on yet other parameters. All the
capacity parameters affect the phase ratio (V,/V,,J. If all other parameters are kept
constant, then the film thickness and the surface area will affect V,, the porosity will affect
V,,, and the diameter of open columns will affect both V,,, and V,. However, it is often
impossible to keep all other parameters constant. For instance, it would be very difficult
to vary the porosity without changing the surface area. An example of the effect of
variations in the capacity parameters on other parameters is the decrease in the number
of theoretical plates in the column that usually accompanies an increase in the stationary
phase film thickness in GLC.
Thirdly, the parameters in table 4.2 turn out to be proportional or inverselyproportional
to k,whereas other parameters which affect both k and a, such as temperature in GC and
mobile phase composition in LC, have an exponential effect on k (see table 3.10). Hence,
even if higher a values can be obtained at some temperature or composition outside the
range where k is optimal, chances are that the parameters listed in table 4.2 do not offer
sufficient flexibility to move k values back into the optimum range.
For all these reasons, it is usually realistic to treat R, as the product of only two
independent factors according to eqn.(4.14). The first factor will depend on the retention
( k ) and the second factor will reflect the efficiency of the chromatographic system (N).
Since we are most interested in small values for R, (i.e. k , M k J , the variation of N with
k can be neglected as a first approximation*, and the two factors can be treated as
independent. Therefore, we can define a separation factor independent of the column
efficiency:
S = h-k, (4.15)
k,+k,+2
This separation factor was first suggested by Ober [407]an it has been used more recently
by Jones and Wellington [408] and by Schoenmakers and Drouen [409,410]. For a given
value of S, the number of plates required to realize a given value of R, (Nne)can easily
be obtained from
N,, = 4 ( R, / S )2. (4.16)
S has the advantagethat it is obtained from the chromatogram much more readily than
is the case for R,. To establish the value of S no estimate of the peak width is required.
Moreover, if we substitute k = (t- @ / t o in eqn.(4.15), we find that
(4.17)
Hence, S can be obtained directly from the retention times of two successive peaks,
* While this may be true for a pair of adjacent peaks in the chromatogram, it may not be quite as valid
an approximation if an entire chromatogram is considered (section 4.3).
126
without the use of an estimate for the hold-up time (to).Conversely, if S is calculated from
k values, the to value used to obtain the latter will not affect the value of S.
We may summarize the advantages of the separation factor S as follows:

1. S is directly related to chromatographic theory ( a s is RJ.
2. S can readily be obtained from the chromatogram.
3. N o estimate for the hold-up time to is required to establish S .
The disadvantages are:
1. Use of S implies the assumption of Gaussian peaks.
2. The plate count N is assumed to be constant throughout (parts ofl the chromato-
gram (i.e. N is independent of k), as well as constant in time.
The second disadvantage can largely be removed by expressing the plate count (N) as
a function of the capacity factor (k). This has been demonstrated by Svoboda [41 I]. If N
is not assumed to be a constant, but some function f(k) of k, then we assume that the
peak-broadening process is determined by the properties of the column and the phase
system and not by the properties of the solute (e.g. its diffusion coefficient). In other words,
if two very different solutes elute with the same capacity factor, we would expect the widths
of the two peaks to be the same. While this may not always be completely true, it appears
that a useful refinement of the elemental criterion is possible in this way. The function f(k)
will add an extra factor to eqn.(4.15):
S’ = k2 - k, * f(k) . (4.1 5a)

k,+k,+2
However, the main reason for preferring S over R, (i.e. that no estimate is required for N)
is now no longer relevant, and therefore, it is more appropriate to introduce the function
f(k) into eqn.(4.14):
k, - k,
R, = .-f( k)1’2 (4.14a)
k ,+k ,+2 2
In order to estimate k (and hence f( k)), an estimate for the hold-up time to is required.
However, this can be avoided if N is expressed as a function f( V,) of the retention volume.
4.2.4 Discussion
A comparison of various elemental criteria has been reported by Knoll and Midgett
[412] and by Debets et al. [413]. Figure 4.4 shows the variation of some of the criteria for
the separation of pairs of chromatographic peaks as a function of the time difference
between the peak tops (At = t2 - t , ) . By definition, R, (and hence S) varies linearly with
At. The peak-valley ratios (P)and the fractional overlap both increase rapidly with
increasing AZat first, but level off towards At =: 4 (T to reach the limiting value of 1. At
high values of At, R , and S will keep increasing, while the other criteria will not.
Figure 4.5 shows the variation of the fractional overlap criterion with At for three
different values of the ratio of peak heights (A). These data were calculated for Gaussian
peaks. It is clear that FO will be lower for peak ratios different from unity. Similar
127
calculations reveal that P and P, are virtually independent of A (for Gaussian peaks), but
that P, varies with A as expected.
FO accurately describes the real extent of quantitative separation obtained in a
chromatogram. If an analysis is optimized on the same column on which it will later be
run as a routine separation, then this is a fair criterion. If however, the analysis will
eventually be run on a different column of a potentially different length (or diameter, etc.),
then it will often be hard to predict the value of FO on that other column. In the case where
2 L 6 8
AL
0
.-*
Figure 4.4: Variation of some elemental criteria as a function of the difference in retention times
between the two solutes. Data calculated for Gaussian peaks of equal height. Courtesy of Anton
Drouen [405].
'I
-0 2 6 8
Figure 4.5: Variation of the fractional overlap criterion (FO) and the resolution ( R J as a function
of the difference in retention times between the two solutes. FO data calculated for Gaussian peaks
of varying peak height ratios ( A = h,/h,). Courtesy of Anton Drouen [405].
128
FOFOequals
equalsunity unitythisthisbecomes
becomesquite quiteimpossible.
impossible.InInother otherwords,
words,givengiventhe thefinal
finalcolumn
columnfor for
routine
routineanalysis,
analysis,very verylargelargevalues
valuesofofAt Atare areunattractive,
unattractive,since sincetheytheydo donot
notincrease
increasethe the
valueofofFO,
value FO,but butdo doleadleadtotoan anincrease
increaseininanalysis
analysistime. time.If, If,however,
however,we wecan cantailor
tailorourour
column
columntotothe theresult
resultofofthe theoptimization
optimizationprocedureprocedure(i.e. (i.e.totothe
thenumber
numberofofplatesplatesrequired),
required),
then
thenlargelargevalues
valuesofofAt Atleading
leadingtotovery verylarge
largevalues
valuesofofR,R,are areindeed
indeedsignificant.
significant.Hence, Hence,
inin the
the case
case where
where the the column
column dimensions
dimensionscan can be be chosen
chosen after after completion
completion ofof the the
optimization
optimizationofofselectivity,
selectivity,the theuseuseofofRR, ,ororSSisispreferred,
preferred,becausebecauseofofthe theclear
clearand andsimple
simple
relationship
relationshipbetween betweenthese thesecriteria
criteriaand andthe therequired
requirednumber numberofoftheoretical
theoreticalplates.
plates.
The
Thesamesameargument
argumentholds holdseven evenmoremorestrongly
stronglywith withrespect
respecttotopeak-valley
peak-valleyratios. ratios.NotNot
only
onlyisisthere
thereaarange
rangeofoflargelargeAt Atvalues
valuesfor forwhich
whichPP==1,1,there thereisisalso
alsoaaconsiderable
considerablethreshold
threshold
range
rangeininwhich
whichAt At>>0,0,butbutthere
thereisisno nodiscernible
discerniblevalley valleybetween
betweenthe thepeaks
peaksand andhence
hencePP==0.0.
For
ForGaussian
Gaussianpeaks peaksofofequal equalheight
heightthisthisthreshold
thresholdrange rangewas wasshown
showntotoequalequalaaresolution
resolution
0.59ororless
ofof0.59 less(eqn.4.11).
(eqn.4.11).
IfIfthe
theanalysis
analysistotobe beoptimized
optimizedinvolvesinvolvesaasample sampleininwhich whichthe therelative
relativepeakpeakareasareasare are
expected
expectedtotobe beconstant
constant(for (forinstance
instanceininaaqualityqualitycontrol
controlsituation),
situation),then thenaacriterion
criterionmay may
bebeused
usedthatthatisisaffected
affectedby bythetherelative
relativepeakpeakheight
height(A), i.e.FO
(A), i.e. FOororPP, ,may maybe beused.
used.IfIfthisthis
isisnot
notthe thecase,
case,then thenaacriterion
criterionshouldshouldbe beselected
selectedthat thatdoes
doesnot notvary
varywithwithAA(R,$ (R,$ ororS;S;PP
ororPPJJ. .This
Thiswill
willavoid
avoidthe thevery
veryunattractive
unattractivesituation
situationininwhich whichthe thelocation
locationofofthe theoptimum
optimum
isisaafunction
functionofofthe the(quantitative)
(quantitative)composition
compositionofofthe sample,so
thesample, thatinintheory
sothat theorytheretheremay may
bebedifferent
differentoptimum
optimumconditionsconditionsfor forevery
everysingle
singlesample!
sample!
This
Thiseffect
effectwillwillbe bemostmostpronounced
pronouncedininthe thecase
casewherewhereaasolvent
solventpeak peakdominates
dominatesthe the
chromatogram
chromatogramand andsolutes
solutesneed needtotobe beanalyzed
analyzedon onthethetail
tailofofthis
thispeak
peak(see
(seesection
section4.6.3).
4.6.3).
AAsimilar
similarargument
argumentholds holdsfor forthetheinfluence
influenceofofthe thepeak
peakshapeshapeon onthetheseparation
separationcriterion.
criterion.
InInthe
thenon-linear
non-linearpart partofofthe thedistribution
distributionisotherm,
isotherm,the theshape
shapeofofthe thepeak
peakwillwillbebeaafunction
function
ofofthe
theinjected
injectedquantity.
quantity.Hence, Hence,once onceagain,
again,the thelocation
locationofofthe theoptimum
optimummay maybe beaffected
affected
bybythe
thecomposition
compositionofofthe thesample.
sample.Also, Also,thetheeffect
effectofofcolumn
columndimensions
dimensionson onthethepeak
peakshape
shape
may
maybe behard
hardtotopredict,
predict,and andthe thepeakpeakshape
shapemay maytotoaalarge largeextent
extentbe bedetermined
determinedby bythethe
characteristics
characteristicsofofthe theinstrument,
instrument,rather ratherthanthanofofthe thecolumn.
column.Therefore,
Therefore,ififthe thecomposition
composition
(or
(orthetheconcentration)
concentration)ofofthe thesample
samplecan canbe beexpected
expectedtotovary varyconsiderably,
considerably,and andififititisis
desirable
desirablethat that thetheresult
resultofof an anoptimization
optimization process processcan can be beextrapolated
extrapolatedtotodifferent different
columns
columns(of (ofthe
thesamesametype)type)and andtotodifferent
differentinstruments,
instruments,then thenititisisadvisable
advisabletotouse usecriteria
criteria
that
thatarearenotnotaffected
affectedby bythe therelative
relativepeak peakareas,
areas,nor norby bythetheshape
shapeofofthe thepeaks.
peaks.
For
Forpractical evaluationFO
practicalevaluation FOisisaavery veryunattractive
unattractivecriterion.
criterion.Its Itsvariation
variationwith withAt Atand
and
with
withthe thepeak
peakarea ratioAAisissimilar
arearatio similartotothatthatofofthe thepeak-valley
peak-valleyratio ratioP,. P,.Pand
PandPP, ,are aresimilar
similar
totoeach
eachotherotherininall respects.PP, , may
allrespects. maybe beobtained
obtainedfrom fromthe thechromatogram
chromatogramslightly slightlymore more
easily
easilythanthanP,P,because
becauseititonly onlyrequires
requireslocation
locationofofthe thepeak
peaktops,
tops,and andnotnotofofthe valleys.To
thevalleys. To
calculateRR, ,from
calculate fromthe thechromatogram
chromatograman estimateofofNNisisrequired.
anestimate required.Scan Scanbe beestimated
estimatedvery very
easily,
easily,using
usingonly onlythe theretention
retentiontimes timesofofindividual
individualpeaks. peaks.
Below
Belowaacertaincertainthreshold
thresholdresolution,
resolution,no novalley
valleycan canbe beobserved
observedbetweenbetweentwo twoadjacent
adjacent
peaks
peaksininaachromatogram.
chromatogram.InInthat thatcase
casethe thevalue
valuefor forany anyofofthethepeak-valley
peak-valleyratios ratioswould
would
equal
equalzero.zero.InIntheory,
theory,the thevalue
valuefor forR,R,andandSSwould wouldexceedexceedzero zerofor foranyanytwotwopeakspeaksthatthat
have
havedifferent
differentretention
retentiontimes times(At (At>>0).0).InInpractice,
practice,this thisdifference
differencevanishes
vanishesififthe thepresence
presence
ofoftwo
twopeakspeakscannot
cannotbe bediscerned
discernedfrom fromthe thechromatogram.
chromatogram.However, However,the theoccurrence
occurrenceofof
ill-resolved
ill-resolvedpeaks peaksininaachromatogram
chromatogrammay maybe berecognized
recognizedvisually
visuallyatatresolutions
resolutionswell wellbelow
below
0.6(the
0.6 (thethreshold
thresholdvalue valuebelow belowwhich whichPPequals equalszero zerofor forGaussian
Gaussianpeaks peaksofofequalequalheight)
height)
(see
(seeref.
ref.[401],
[401],figure
figure2.1 2.11,1,p.38).
p.38).Moreover,
Moreover,there thereare areseveral
severaltechniques
techniqueswhich whichmay maybe beofof
129
129
help in confirming the purity of the peaks obtained in a chromatogram during an
optimization process (see section 5.6).
In some optimization procedures the capacity factor is known as a function of the
parameters. These so-called “interpretive optimization methods” will be described in
section 4.5. From known capacity factors R, and S can be calculated much more easily
than peak-valley ratios and, moreover, from known capacity factors the R, or S values
can be calculated, no matter how small the difference between the two capacity factors
is. In other words, the resolution of a pair of peaks can be calculated in a range in which
it would be very difficult to obtain an estimate for the resolution from an actual
chromatogram. Therefore, the use of R, or S as a criterion to judge the separation in
combination with interpretive optimization methods enables us to appreciate variations
in the resolution in the range of 0 < R,< 0.6. Such variations are very significant because
(i) on a different (more efficient) column the separation with the highest value for R, is
most easily realized and (ii) on the same column, improvements in resolution in the range
0 < R, < 0.6 will help to send the optimization process in the right direction.
Hence, in combination with interpretive methods the use of R, or S as the resolution
criterion appears to be always advantageous.
A further refinement may be sought by incorporating a function f(k) to describe the
dependence of the plate count on the capacity factor (see eqns. 4.14a and 4.15a).
The characteristics of the different criteria are summarized in table 4.3. Table 4.4 lists
the recommendations formulated above for the use of different criteria.
Table 4.3:
Characteristics of different elemental criteria for measuring the extent of separation of a
pair of chromatographic peaks.
Criterion Affected by Reflects Transfer Ease of

actual towards calcu-
Peak area Peak separation other lation
ratio shape columns
- +/- ++ +/-
- +/- ++ ++
P - + + +/- (I) +/-
pm - + + +/- (1) +/-
P“ + + + - +/-
FO + + ++ - --
(1) Indirectly via eqn.(4.10), but only in the range where (approximately) 0.05 < P<0.95.
130
Table 4.4
Recommendations for the use of different elemental criteria for measuring the extent of
separation of a pair of chromatographic peaks. The preferred criteria are given, while
possible alternatives appear in brackets.
Optimization on final analytical column
YES NO
Interpretive method (1)
YES NO
(1) See section 5.5.

(2) The noise level can be incorporated in analogy to eqn. (4.12).
4.3 CHROMATOGRAMS
We will base our discussion about criteria by which to judge the quality of an entire
chromatogram on the elemental criteria for pairs of chromatographic peaks, which have
been defined in chapter 1 (R, and a) and in the previous section . We will look at several
ways of combining the numbers for all individual pairs of peaks into a single number. We
will then discuss the influence of other parameters, such as the analysis time and the
number of peaks on the proposed criteria.
Initially, the discussion will be focussed on the general case (see section 4.1), in which
all peaks in the chromatogram are considered to be of equal importance and all peaks have
to be separated. At the end of this chapter, we will discuss some specific cases, for which
the requirements are different.
4.3.1 Sum criteria
Summation of resolution values has been used by Berridge [414] and summation of
separation factors has been suggested by Jones and Wellington [408].
In the introductory section of this chapter it was shown that a straightforward
summation of resolution values does not yield a satisfactory criterion for the evaluation
of complete chromatograms (see figure 4.1 and table 4.1). A problem that can readily be
appreciated from the example given there, is that the sum of R, values will be determined
mainly by the largest values of R, that occur in the chromatogram. For example, in the
131
chromatogram shown in figure 4.lb, the sum of resolutions turns out to be 25.3. Of this,
24.07 or 95% is due to the excessive separation of the last two peaks. However, in judging
the separation, the chromatographer will immediately refer to the separation of the first
two peaks, rather than to that of the last pair. This is correct, because the first two peaks
determine the efficiency of the chromatographic system that is required to realize the
separation of all three peaks (see also section 4.3.3 below).
Apparently, it is the occurrence of very large R, values that causes problems. Obviously,
this problem can then be avoided by substituting for R, one of the criteria which level off
for very large time differences between the two peaks (see figure 4.4). In this way, the
contribution of the abundantly separated pairs of peaks in a chromatogram is limited. The
resulting sum of FO or P values is to a much smaller extent determined by the largest
values, although in some extreme cases large contributions may still obscure important
changes in small ones. For example, if 20 pairs of peaks were to occur in a chromatogram
and each of these pairs were almost separated to the baseline (P=O.9), then the sum ZP
would equal 18. This is less than if 19 out of the 20 pairs of peaks were amply separated
(P= l), but two peaks showed complete overlap (P= 0), giving rise to a value for ZP of
19. Assuming that all peaks are of equal importance, the latter chromatogram is obviously
inferior. Of course, this is a hypothetical example, but it illustrates a potential limitation
of the criterion ZP.
Figure 4.6 illustrates the dependence of ZP and ZR, on the number of plates in the
50 100
fi-
Figure 4.6: Variation of the sum of peak-valley ratios as a function of the number of plates for the
two chromatograms( a and b) shown in figure 4.1, and for a third chromatogram(c), shown in figure
4.8. P was calculated from eqn.(4.10). Negative values for P were set equal to zero. The sum of
resolution values is shown as a dashed line for chromatogram a only.
132
column. This figure has been calculated from the data in table 4.1, assuming that the
chromatograms of figure 4.1 have been run on a series of columns with different Nvalues.
Obviously (eqn.1.22), ZR, increases linearly with V N . ZS (not indicated in figure 4.6) is
independent of N (see eqn.4.15) The behaviour of ZP is vastly different.
When there are no plates ( N = 0), there is complete overlap of all peaks (ZP= 0). For
chromatogram b this soon changes, since only a handful of plates is necessary to achieve
baseline separation of the last two peaks. For a while, the ZP value for chromatogram b
is then larger than that for chromatogram a, because two peaks are distinguishable instead
of one. However, at still higher values of N the last two peaks in chromatogram a start
to be resolved, soon followed by the first two peaks in both chromatograms. Eventually,
abundant resolution will be achieved for all pairs of peaks, all values of P will equal unity
and the ZP values for both chromatograms will be the same (ZP= 2).
From this point on the use of ZP no longer enables us to quantify the quality of the
chromatogram, because it does not differentiate between chromatograms a and 6.
Moreover, above a certain threshold number of plates (around 10,000 in the example of
figure 4.6) the ZPcriterion becomes very insensitive to the number of plates and to changes
in the relative peak positions, unless these changes have a significant effect on R , values
between about 0.6 and 1.5.
From the above discussion the following five conclusions can be drawn:
1. ZR, is not a useful criterionfor judging the quality of a chromatogram, since its value
is determined largely by the largest values of R, that occur in the chromatogram. i.e. by
thepairs ofpeaks which are the least relevantfor the realization ofa separation. Thesame
is true for the sum of separation factors (ZS).
2. ZP gives a better representation of the actual separation achieved on a given
column, since there is a limit to the contribution of amply resolvedpairs ofpeaks.
3. Above a certain threshold value for the number of plates, a pair of peaks will
become irrelevantfor the determination of ZP. When this threshold is reached for
all pairs of peaks. Z P will have reached its limiting value. Changes in N will no
longer be reflected in ZP, whereas changes in the (relative) retention times become
increasingly irrelevant as N increases.
4. Below a certain threshold value for the number ofplates, all P values will be zero
and again ZP becomes insensitive to changes and provides no information about
the chromatogram.
5. The values of Z R , and ZS on another column with a different plate count can easily be
predicted, since ZR, is proportional to V N and since ZS is independent of the plate
count. For ZP, the situation is less straightforward. The value of ZP is relevant
for the separation on a given column, but cannot be extrapolated from there.
As the main conclusion from this section it appears that ZR,(and hence ZS)is not a
useful criterion, and that ZP may be used for a comparison of chromatograms on a single
column. A problem that remains is the fact that ZP becomes a very insensitive criterion
once the limiting value (equal to n - 1 if n is the number of peaks in the chromatogram)
is approached, as well as in the range in which one or more of the P values become equal
to zero. In section 4.3.4 we will investigate whether composite criteria (involving for
instance the analysis time) can be used to avoid this problem.
133
4.3.2 Product criteria
A second major category of proposed criteria to ress the extent of separation in an

\
entire chromatogram is that in which the product is taken f the values for all pairs of peaks
of one of the elemental criteria defined before. Taking the products of these criteria is
equivalent to taking the sum of the logarithms, for instance
lIR,=exp(ZInR,). (4.18)
The use of the sum of logarithms may have a slight disadvantage in the case where a value
of zero occurs for one of the pairs of peaks. If any of the peak-valley ratios (P,P, or Pv)
is used, then this problem is aggravated because these criteria take on a value of zero below
a certain threshold resolution. The obvious way around this problem, however, is to set
the sum of logarithms equal to minus infinity or to a large negative number once a value
of zero occurs.
A summary of the use of product functions in the literature is given in table 4.5. In some
cases, the products were part of composite criteria involving other factors or terms. We
will come back to these criteria in section 4.3.4. It is clear from table 4.5 that product
criteria have been used more extensively than have sum criteria.
Table 4.5:
Summary of product criteria proposed in the literature to express the extent of separation
achieved in a chromatogram.
Elemental Product criterion proposed by Ref.

criterion
RS Glajch et al., JC 199, (1980),57 415
S Schoenmakers et al., Chr 15, (1 983),688 409
FO Smits et al. ZAC 273, (1975),1 416
P Morgan and Deming, JC 112, (1975),267 417

Watson and Carr, AC51, (1979),1835 418
Spencer and Rogers, AC 52, (1980),950 419
P' Wegscheider et al., Chr 15, (1982),498 406
Explanation of abbreviations:
A C = Analytical Chemistry
Chr = Chromatographia
JC = Journal of Chromatography
ZAC = Fresenius Zeitschrift fur Analytische Chemie
One obvious advantage of product criteria is that the result will be mainly determined
by the smallest values for the elemental criterion, i.e. by the least resolved pairs of peaks.
134
All the criteria that have been discussed in section 4.2 will equal zero if one pair of peaks
shows complete overlap and therefore once such a situation occurs the resulting product
will be zero as well.
I /
/
5
-
~ 1 1 0 ~
10
Figure 4.7: Variation of the product of peak-valley ratios as a function of the number of plates for
the two chromatograms ( a and b) shown in figure 4.1, and for a third chromatogram (c), shown in
figure 4.8. P was calculated from eqm(4.10). Negative values for P were set equal to zero. The product
of resolution values is shown as a dashed line for chromatogram a only.
Figure 4.7 shows the variation of the products of R, and P values as a function of the
number of theoretical plates for the chromatograms of figure 4.1 and an additional
chromatogram shown in figure 4.8. Plotting the products against N yields a straight line
for nR,.In general, l7R, will be proportional to N, where p is the number of pairs of peaks
in the chromatogram. Therefore, the differences between l7R, and l7P will be more
pronounced for chromatograms with large numbers of peaks.
As with the sum criteria, the use of P instead of Rs does not yield a simple relationship
for the variation with N. However, it is very clear from figure 4.7 that the differences
between the l7P values for the chromatograms a and b are much smaller than the
differences in the ZP values (figure 4.6). This illustrates that the value of l 7 P is mainly
determined by the least separated pair of peaks, i.e. the first two peaks, which are the same
in both chromatograms.
All product criteria will be zero if any single pair of peaks is completely unresolved. For
FO, R, and S this situation theoretically only occurs if the retention times of two peaks
are equal. For peak-valley ratios a value of zero is estimated from the chromatogram below
a certain threshold sewration, which for Gaussian peaks corresponds to R, < 0.59
135
(eqn.4.11). Thus, l7P equals zero once any pair of peaks shows no discernable valley. Of
course, there is an important increase in separation if R, is increased from 0 to 0.6. This
is a serious drawback to the use of l7P as an optimization criterion, since it does not
acknowledgedefinite improvements below a certain threshold and it illustrates once more
that the elemental criterion P may be used only if the optimization process is carried out
on the final analytical column.
Even more than ZP, l7Pis a threshold criterion. Its value is zero or one, with only a small
range over which intermediate values occur. Threshold criteria can be used to allocate
areas in which a certain condition is fulfilled. They divide the parameter space (i.e. the
space formed by all the parameters considered in the optimization process, see section
5.1.3) into areas where a certain condition is met and areas where this is not the case. In
the case of l7P there is a diffuse boundary in between the different areas.
So far, all the criteria that have been discussed have suggested either that chromatogram
b is superior to chromatogram a, or that both chromatograms have equal credentials. The
0 k
-
Figure 4.8 Three schematic chromatograms.Constructed for N = 10,000. Capacity factors are listed
in table 4.6.
136
main reason for this is that the capacity factors of the last peaks in the chromatograms
in figure 4.1 are vastly different, i.e. we are comparing two chromatograms which on the
same column under identical conditions would show vastly different analysis times.
Clearly, some correction is required once this is the case.
Figure 4.8 shows the two chromatograms of figure 4.1, together with a third chromato-
gram (c), in which the capacity factors of the first and the last peak equal those observed
in chromatogram b, but the separation of the first two peaks has been improved
dramatically. Table 4.6 lists the data for all three chromatograms. The capacity factors are
Table 4.6:
Data for capacity factors, elemental criteria and for criteria judging the extent of
separation in the entire chromatograms. Chromatograms are shown in figure 4.8.
Criteria for pairs of peaks: separation factor (S, eqn.4.15), resolution ( R , eqn.4.14) and
peak-valley ratio ( P , eqn.4.10).
Criteria for entire chromatograms: sum criteria (section 4.3.1), product criteria (section
4.3.2), normalized resolution product (r, eqn.4.19), calibrated normalized resolution
product (r*, eqn.4.21) and minimum resolution (section 4.3.3).
For discussion see text.
Chroma- Peak k S R S P
togram number
1 1
0.0244 1.22 0.898
a 2 1.1
0.0345 1.72 0,995
3 1.25 r = 0.97
2 0.059 2.9 1.893 r* = 0.13
n 8.4.10-4 2.1 0.893 Rs,min = 1.2
1 1
0.0244 1.22 0.898
b 2 1.1
0.481 24.1 1
3 5 r = 0.18
z: 0.51 25.3 1.898 r* = 0.18
n 1.2.10 29.4 0.898 Rs,min = 1.2
1 1
0.273 13.6 1
C 2 2.5
0.263 13.2 1
3 5 r = 1.00
z: 0.54 26.8 2 r* = 0.98
n 7.2.10 180 1 Rs,min = 13.2
137
, used to calculate S, R, and P. The latter is estimated from eqn.(4.7). Also, the various sum
and product criteria are shown in the table. Since there is a constant factor of 50 (vN12)
between S and R , we will focus on R, and P only.
In all three chromatograms the threshold number of plates for separation appears to
have been approached, so that CP and are only slightly affected by the differences
between the chromatograms. No distinction can be made between the sum criterion CP
and the product criterion nP, The difference between the two is 1 for all three
chromatograms. Both criteria yield marginally higher values for chromatogram b in
comparison to.chromatogram a. Chromatogram c yields the maximum values of 2 for ZP
and 1 for nP.In fact, it is well into the region in which the peak-valley ratio is completely
insensitive to variations in the capacity factors. If the resolution (R,) equals 2, then
eqn.(AlO) yields a Pvalue of 0.999. In chromatogram c of figure 4.8 the resolution between
each pair of peaks is about 13.
ZR, is much higher for chromatograms band c than it is for chromatogram a. However,
it is about equal for the bottom two chromatograms. Hence, ZR, is more sensitive to
changes in the capacity factor of the last peak than it is to changes in the extent of
separation. n R , does yield a much higher value for chromatogram c than it does for
chromatogram b. Hence, nR, can be used for a quantitative comparison of chromato-
grams of similar length (capacity factor of the last peak). When the length of the
chromatogram changes (for instance in going from chromatogram a to chromatogram b),
nR, is not a useful criterion.
Normalized resolution product
Drouen et al. [410] have recognized this problem and proposed a solution by using a
product of normalized resolution values. They divide every value of R, by the average R,
value (x,),where the average is taken over all the pairs of peaks in the chromatogram:
(4.19)
where n is the number of peaks and
(4.20)
The average S value (3 is defined analogously.

Because both R,sand x,$ are proportional to V N , the normalized product of R, values
is equal to that of the S values, and both are independent of the number of plates. The
normalized resolution product ( r ) will vary from zero, in the case where one or more pairs
of peaks show no resolution, to one, if the resolution is equal for all the pairs of peaks in
the chromatogram. Therefore, in choosing r as the optimization criterion the aim is to
achieve an equal distribution of the peaks over the chromatogram.
This aim is realized much better in chromatogram c in figure 4.8 than it is in
chromatogram b. The r values given in table 4.6 illustrate that fact. For chromatogram c
the r value is (almost) equal to 1, while for chromatogram b it is not larger than 0.18.
138
Chromatogram a is very different from chromatogram c, but it also shows good spacing
of the peaks over the chromatogram and hence an r value of about 1.
Clearly, r values do reflect the distribution of the peaks over the chromatogram and are
not seriously affected by the absolute k values.
Chromatogram c appears to show a much longer analysis time than does chromatogram
a. However, if we are free to define the column dimensions after the selectivity
optimization process, chromatogram c can be the basis for a very quick separation on a
very short column.
Calibrated normalized resolution product
In theory, all peak pairs may show equal R , values, but the peaks may occur very late
(with very high capacity factors). A bunch of peaks may move about through the
chromatogram and will yield the same value for any of the criteria discussed so far as long
as the mutual resolution factors between all the different pairs of peaks remain unaltered.
To avoid bunching of the peaks at some high value of k, Drouen et al. have suggested the
inclusion of a hypothetical peak at t = to in the calculation of r [410]*. This yields the
calibrated normalized resolution product (r*),defined as
n-I n-1
r* = ll (R,i,i+,/ R,) = ll (Si,i+l1s) (4.21)
i=O i=O
where
(4.22)
The r* values for the three chromatograms in figure 4.8. are also shown in table 4.6. It
appears from these values that r* is very high (close to the maximum value of 1) for
chromatogram c, but that the bunching of peaks around k = l in chromatogram a is
effectively penalized. The introduction of the hypothetical peak at t = to has the effect of
“calibrating” or “anchoring” the real peaks in the chromatogram to a starting point.
The optimum value of r (or r*) does not correspond to a unique chromatogram, but
rather to a series of chromatograms, each of which has the peaks spread out at constant
resolution intervals in the chromatogram. In other words, the absolute value for R, or S
may vary, but all the normalized values are equal to 1.
r* is higher for chromatogram b than it is for chromatogram a. If we include a peak at
t = to, then the value for S between this peak and the first real peak in the chromatograms
is 0.33 (R, = 16.7; P = 1). This means that for chromatogram a one high value for S occurs
(0.33) in combination with two low values (0.02 and 0.03), while in chromatogram b two
high values (0.33 and 0.48) are combined with one low value (0.02). If the goal is to make
all S values equal, both situations are equally bad. Nevertheless, because of the much
smaller k value for the last peak (and given the equal resolution for the first two real peaks)
chromatogram a may well be preferred to chromatogram b.
* A hypothetical peak assumed at any time after t = to [410] will give rise t o considerable problems,
because peaks may be distributed over the chromatogram before as well as after this imaginary peak.
139
Although r* appears to be very useful as a criterion that strives towards a clear and
objective goal in selectivity optimization, it is still not perfect once two chromatograms
are compared which show very different capacity factors for the last peak, especially when
both r* values are low.
The use of P , in sum and product criteria
The use of P, in sum or product criteria creates a special problem, because two values
for P, can be calculated for each peak in the chromatogram. This applies to isolated as
well as to ill-resolved peaks and it also applies to the first and the last peaks observed in
the chromatogram. The number of P, P,, R,, or S values in an entire chromatogram
usually equals n-1 (where n is the number of peaks). If an imaginary peak is assumed to
be present at t = to, the number of values for the elemental criteria becomes n. We can deal
with this problem in three ways.
1. Use all P, values without correction.
This is a useful approach if no other criteria are to be considered, so that the criteria
based on P, will not have to be compared with other criteria.
2. Use of the lowest value for P, that occurs on either side of a peak.
A considerable disadvantage of this approach is that large improvements in the
resolution may go by unnoticed.
3. Use of an average value for P,.
If two values for P, for each peak are obtained, then this strategy corresponds to the
use of (112) ZP,or VnP,as the optimization criterion. This third approach appears
to be the most correct one, since it creates a common basis for all sum and product
criteria (i.e. those based on P, P , P, R , and S), which may allow a comparison
between the different propositions.
Another peculiar aspect of the use of P, is that its value is by definition proportional
to the height of the peak (see also section 4.2.4). Hence, for a pair of peaks with a certain
valley between them, the P, value will be largest for the largest peak and proportionally
smaller for the smallest one. Since sum criteria are mainly determined by the largest values
for the elemental criteria (see section 4.3.1), we may expect that the value of ZP,will be
affected most by the largest peaks (major components) in the sample. On the other hand,
product criteria are affected most by the smallest values for the elementa1 criteria (see
section 4.3.2) and hence n P , will be determined mainly by the minor components in the
sample, i.e. the smallest peaks which are detected and considered relevant in the
optimization process. Because one very small value (close to zero) for P, will largely
determine the value for the entire product, this emphasis on small peaks is much greater
than the emphasis put on the large peaks by the use of ZP,,.
We might say that if P, is used as the elemental criterion, a weighing factor is
automatically built into the (sum or product) criterion for the entire chromatogram, which
puts the emphasis either on the major or on the minor components in the sample.
4.3.3 Minimum criteria
The lowest value for a which occurs in a chromatogram has been used extensively in
GC [420] and LC [421-4231 (see also section 5.5) as a criterion to quantify the extent of
140
separation achieved in a chromatogram. This so-called minimum a (amin) criterion is set
equal to the lowest value for (I that occurs for any pair of peaks in the chromatogram.
However, the value of a is not a good indication for the separation of a pair of peaks.
For example, if an a value of 1.05 occurs somewhere early in the chromatogram, say
around k = 0.5, then the corresponding S value is 0.008 and 60,000 plates are required to
achieve adequate resolution (R,= 1) of the two peaks. If the same value for awere to occur
around k = 3 , then S would equal 0.018 and to realize an R, value of 1 about 12,000
theoretical plates would be sufficient.
Clearly, it is advisable to substitute R, (a quantity which depends on the plate count)
or S (independent of N) for a. In judging a chromatogram on the basis of the*minimum
value for R, ( Rs.min)or S (Smin), it becomes very easy to estimate the number of plates that
is required to realize the separation with sufficient but not excessive resolution. For
instance, if the final result of a selectivity optimization process is a chromatogram with
an Rs.minvalue of 0.5 on a column with 2,500 theoretical plates, then a column with 10,000
plates will yield an Rs,minvalue of 1 under identical conditions.
During a selectivity optimization process the Rs.minor Smin criterion can be used in two
ways:
1. By setting a threshold value (e.g. Rs.min= I), above which the result is acceptable. If x
is the threshold value, we can describe this criterion as
(4.23)
This criterion may be used during a sequential optimization process (see chapter 5),
leading to an acceptable result and to completion of the optimization process once the
threshold value has been reached. Alternatively, it may be used to establish ranges of
conditions in the parameter space for which the result will be acceptable. This latter
approach has been followed by Glajch et al. 14151, by Haddad et al. [424] and by
Weyland et al. [425] and was referred to as resolution mapping by the former. Within
the permitted area(s) secondary criteria are then required to select the optimum
conditions. For example, the conditions at which the k value of the last peak ( k J is
minimal while the minimum value for R,sexceeds1may be chosen as the optimum. Such
a composite criterion can be described as
min k , f l Rs,min> x . (4.24)
2. A second way to use the Rs,mincriterion is to try and maximize the value for Rs.min.
In other words, one may strive towards conditions at which the lowest value for R,
observed in the chromatogram is as high as possible. We can describe this criterion as
(4.25)
This criterion aims at a chromatogram that can be realized with the lowest possible
number of theoretical plates. Indeed, if the highest possible value of Rs.minhas been
reached, this automatically corresponds to the lowest number of ptates required (see
section 4.4.3).
Although the goat of achieving the separation with a minimum number of plates appears
141
to be clear and unambiguous, the resulting chromatogram is not well-defined.The fact that
Rs,min has the highest possible value reveals nothing about the remainder of the
chromatogram. A very simple example can be found from the two chromatograms in
figure 4.1, which yield the same value for Rs,min. Moreover, merely because of the direct
relationship between R, and k (see section 1S), R, values will tend to be larger for larger
values of k. Hence, in striving towards a maximum value for one may be striving
implicitly towards very high values for k. Therefore, the criterion described by eqn.(4.25)
should only beconsidered as a criterion for a selectivity optimization process if the overall
capacity factors are not expected to change considerably.
In those cases in which the overall capacity factors do vary, it is more realistic to use
the Rs,mincriterion in the way as described under 1. above. Eqn.(4.23) can be used to define
the boundaries within which acceptable resolution can be achieved. It has the advantage
that no implicit aim towards high capacity factors is present in the criterion. A
disadvantage of the use of eqn.(4.23) is that the boundaries defined by the threshold
criterion will change when the acceptance level is changed, i.e. they will be different if x
in eqn.(4.23) is set equal to 1 from when x= 1.5.
Along the same line, the boundaries will change when the number of plates is changed,
i.e. when another column is used, or even when the flow rate is changed on a given column.
If subsequently secondary criteria are used to define a unique set of optimum parameters
rather than an acceptable range (eg. the hierarchic criterion of eqn.4.24), then the location
of the optimum may very well depend on the threshold value (x) selected by the user, or
on the column used during the process of selectivity optimization. For example, if the
following criterion had been used
min k, n Rs,min > 1 (4.26)
and the resulting chromatogram had an Rs,minvalue of 1.3, then necessarily a different
optimum would have been found than if the criterion
min k, n Rs,min > 1.5 (4.26a)
had been used, or equivalently if the original optimization process were repeated on a
column having a factor of 2.25 (1.5*) fewer theoretical plates than the original one.
In theory, the problem of the result being dependent on the number of plates used during
the selectivity optimization process can be circumvented by setting a limiting value for S,
rather than for R,. Nevertheless, the problem that the resulting location of the optimum
will depend on the arbitrarily selected value for the threshold x still remains. There seems
to be no logical way to define a single unambiguous value for Sminwhich can be used in
all cases.
It is interesting at this point to notice the similarity between the use of l 7 P and the
threshold criterion of eqn. (4.23). In figure 4.9 this is illustrated by plotting IlP,n R , and
Rs,min > 1 as a function of N for the two chromatograms of figure 4.1. Clearly, both l7P
and Rs,minserve as threshold criteria, with the boundaries of the acceptable areas being
abrupt when the minimum resolution criterion is used and diffuse for n P . This invites the
use of IlP in a similar way as i.e. analogous to eqn.(4.24). The combination of UP
with other factors to form composite criteria will be discussed in section 4.4.
142
Figure 4.9: Variation of the product of peak-valley ratios, (IIP), the product of resolution values (IIR,)
and the minimum resolution (R,,in) criteria as a function of the number of plates for the
chromatograms shown in figure 4.1. P was calculated from e q ~ ( 4 . 1 0 )Negative
. values for P were
set equal to zero.
Although the value of R , for a given pair of peaks can be quickly transferred from one
column to another by using the proportionality of R, and V N , this is not the case for the
threshold criterion of eqn.(4.23). The problem is that if we know the boundaries of the area
for which Rs.min > 1 using a column of 10,000 plates, we only know the boundaries of
the area for which Rs,min> 0.5 for a column with 2,500 plates. We d o not know what the
boundaries for Rs.min > 1 are in the latter case, because we have no information on how
the value of Rs.minchanges with variations in the parameter settings. Only if the variation
of the capacity factors as a function of the relevant parameters is known, can the
boundaries of the area in which the resolution is adequate be calculated for different
columns with different numbers of theoretical plates. Optimization methods in which this
is the case (so-called “interpretive methods”) will be discussed in section 5.5.
The following conclusions can be drawn from the discussion in this section:
1. The threshold criterion of eqa(4.23) can be used to establish acceptable areas for
adequate separation on a given column.
143
2. The use of this criterion gives results that are similar to the use of IlP.
3. The areas defined by the threshold criterion are not transferable from one column to
another, unless this is done indirectly by means of known capacityfactors.
4. Maximization of the minimum resolution value observed in the chromatogram(eqn.4.25)

corresponds to aiming at the minimum number of plates required to eflectuate the
separation.
5. This criterion can only be used ifthe overall capacityfactors are roughly constant during
the optimization process.
6. This criterion can readily be transferredfrom one column to another.
4.3.4 Other criteria
The first criterion to be suggested for the evaluation of the quality of an entire
chromatogram was defined as the “total overlap” by Giddings in 1960[426]. The definition
reads:
cp = Z: exp -(2 R , ) . (4.27)
In figure 4.10 the function exp -(2 R,) for a single pair of peaks has been plotted as
a function of R,. Also shown in this figure is the theoretical value for P (eqn.4.7).
Figure 4.10 shows that the two functions are roughly complementary, although the
variation of P with R, is more abrupt. This is logical when we consider the theoretical
Figure 4.10: Variation of Giddings’ peak overlap criterion (p) for one pair of peaks and the
peak-valley ratio (P) as a function of the resolution ( R J of the pair. P values were calculated from
eqn.(dlO). Negative values for P were set equal to zero.
144
relationship between P and R, (eqn.4.10) in which the term 2 exp -(2 R:) appears. In
comparison with eqn(4.27) the change of P with R, will be more abrupt than that of cp
(for one pair of peaks) because of the ocurrance of the square of R, in eqn.(AlO).
Roughly speaking, for a complete chromatogram, the criterion Q behaves similarly to
ZP. It functions as a threshold criterion with diffuse (and stepwise) boundaries,
establishing areas for which adequate separation is obtained (cp=O). Because it is based
on R, rather than on P, cp cannot easily be determined from the chromatogram. On the
other hand, cp may more easily be calculated if the capacity factors and the plate number
are known. Both cp and ZP should only be used for optimization processes run on the final
analytical column. In the following discussions cp will not be considered as a separate
criterion. Its merits correspond to those of the ZP criterion.
4.3.5 Summary
In the preceding four sections we have discussed various “sum”, “product” and
“minimum” criteria. A schematic summary of these criteria and an indication of their
applicability are given in table 4.7.
Table 4.7:
Summary of sum, product and minimum criteria.
Glossary:
C Continuous criterion, transferable to other columns
tha,thf Threshold criteria, which locate boundaries at arbitrary (a) or fixed (f) degrees of
separation
X Not useful as an optimization criterion.
Criterion Sum Product Minimum
eqn. 4.25 eqn. 4.23
Rs
S
P
(1) Normalized R , or S values should be used if the capacity factor of the last peak is expected to
vary.
(2) Only to be used if the capacity factor of the last peak is expected to be constant.
(3) Use of l 7 P is to be preferred rather than ZP.
It is seen in the table that three groups of criteria are readily discernible:
1. The useless criteria. This category comprises ZR, and ZS and needs no further
comments.
145
2. The criteria which vary continuously between a low value of zero and a maximum
value located at the optimum conditions. All these criteria allow a transfer of the
resulting optimum from one column to another.
3. The threshold criteria. which may be used to define the boundaries of areas in
which an acceptable result can be achieved. The boundaries can be at some
arbitrary value for R , or S (or at an arbitrary valuefor P< I ) , or at a fued value. The
latter arises naturallyfrom tht. use of ZP or (preferably) l 7 P , or by setting Pmin= 1.
The use of l7P (or ZP)leads to di#ke boundaries.
4.4 COMPOSLTE CRITERIA
In many cases we have seen that the criteria in table 4.7 are adequate for judging the
extent of chromatographic separation, but insufficient to account for the effects of
chromatographic parameters on the separation. Two additional factors readily come to
mind, and both have been used extensively in the literature. These factors are the number
of peaks in the chromatogram and the analysis time.
Table 4.8 summarizes the requirements of the different criteria for these two additional
factors.
Table 4.8:
Requirements for additional parameters in the optimization criteria listed in table 4.7. The
number of peaks present in the sample is asumed to be known.
Glossary:
n Requires additional parameter to account for the number of peaks appearing in the
chromatogram.
t Requires additional time parameter
x Not useful as an optimization criterion
Criterion Sum Product Minimum
eqn. 4.25 eqn.4.23
R S
P t (n) t - t
(1) Time parameter is less necessary when normalized R,yor S values are used.
4.4.1 Number of peaks
One obvious conclusion from table 4.8 is that a provision for the number of peaks in
the chromatogram is only required for the criteria which are not recommended for use in
146
selectivity optimization. All product and minimum criteria become equal to zero once a
peak disappears completely, while CP only approaches its limiting value of n-1 ( n being
the number of peaks) if all pairs of peaks are adequately resolved. If CP is used instead
of nP,then a provision for the number of peaks in the chromatogram can easily be made
by dividing ZP by the number of pairs of peaks in the chromatogram:
c,=-.ZP (4.28)
n-I
In eqn.(4.28) C, is the optimization criterion corrected for the number of peaks in the
chromatogram*.
However, the above is only true if the number of solutes in the sample (i.e. the number
of peaks that should appear in the chromatogram) is known, which is assumed to be the
case in table 4.8. Obviously, if the number of peaks present in the sample is not known,
complete overlap of two peaks may go unnoticed. This problem will affect all criteria,
although not all to the same extent. Product criteria will often be affected more than will
sum criteria. As soon as it is known that the number of peaks in a chromatogram equals
at least the number n, then all the useful criteria in table 4.8 (groups 2 and 3) will
automatically penalize chromatograms which show fewer than n peaks. Hence, if the
number of peaks decreases during the optimization process, there is n o need to correct any
of the useful criteria in table 4.8 for the number of peaks present in the chromatogram.
A different situation may arise if the number of peaks increases during the optimization
process, which will be more frequently the case if the process guides us in the right
direction. In that case the situation may arise in which the calculated value for the criterion
C is lower in a newly obtained chromatogram than it was in previous ones, while the
number of peaks has actually increased. In that case we have clearly interpreted the
previous chromatograms incorrectly. In many cases (for instance simultaneous methods,
section 5.2 o r interpretive methods, section 5.5) it is not necessary to introduce a separate
correction for the criterion values if the observed number of peaks increases during the
optimization process. This is because the calculation of the criterion (response) values is
the final step in such a procedure. In each calculation step the number of peaks can be
taken equal to the largest number of peaks observed in any of the chromatograms. If this
number increases, then the results of the calculations are automatically adapted.
The situation is different if a sequential method of optimization (e.g. Simplex
optimization, section 5.3) is used. In this case a criterion value is assigned to every
chromatogram and the result is compared with previously obtained values. Hence, if the
number of observed peaks increases, this may lead to incorrect comparisons. For example,
if in one chromatogram three fully separated peaks were observed, the value of l7P for that
chromatogram would equal one. However, if in the next chromatogram four peaks were
observed which were not completely resolved (e.g. Pvalues between each pair of successive
peaks of 0.9), then the resulting value for l7P would only be 0.73. However, the second
chromatogram is clearly to be preferred to the first one.
* In this section we will generally use C for some function of the elemental criteria ( R , S or P), for
instance one of the optimization criteria in table 4.8. C , refers to a criterion which has been corrected
for the number of peaks in the chromatogram, while C, refers to a time-corrected criterion.
147
To deal with this problem, it appears to be more correct to update the previously found
criteria values than it is to increase the value of the new one. To do so, it is not only required
to keep track of the criterion values of previous chromatograms, but also of the number
of observed peaks. In the case of Simplex optimization this is especially easy, since only
the criterion values of three chromatograms need to be remembered (see section 5.3).
Hence, for the above example the previous result might be nP= 1 with n = 3. If the new
result is nP=0.73 with n = 4, then the previous result needs to be updated to nP=0 with
n=4.
It is extremely easy to update the old values for the criterion, because all product criteria
become zero for chromatograms which contain less than the highest number of peaks,
whereas all sum criteria remain unaltered. If a composite criterion is used, in which a time
factor occurs, then the previous values for the optimization criterion (C,) may usually only
be updated if the values for the individual contributions (the value of the criterion C and
a time factor) are stored separately.
4.4.2 Analysis time
Analysis time has been incorporated into optimization criteria as a separate term
[414,415,418] or as a separate factor [406,416].
Typically, a separate term for the influence of the retention time appears as follows:
(4.29)
where I, is the retention. time of the last peak (“analysis time”), t,,, is the maximum
allowable analysis time and a and 6 are (positive) weighting factors*. The last three
parameters can all be chosen arbitrarily by the user.
However, the actual influence of the user on the optimization process is limited to one
parameter only, i.e. the ratio between the weighting factors a and 6. This can easily be
understood from eqn.(4.29), once it is rewritten as
C, = a . { C + ( W a ) t,,, - ( b / a )t , }
= a . { C - (b/a)t, + C} (4.29a)
where c is a constant.
Variations in c cause all values of C , to be increased by the same amount, and hence
the location of the optimum and the course of the optimization process are by no means
affected. The same is true for variations in the weighting factor a, which cause all values
of the criterion to be changed by a constant factor. Only variations in the 6 / a ratio that
change the weighting of t, vs. C will affect the selectivity optimization process and the
* Note that for a meaningful summation in eqm(4.29) reciprocal time units (e.g. min- ’) are required
for the parameter b. In this and subsequent equations we will tacitly assume that the correct
dimensions have been assigned to the parameters.
148
location of the optimum. Although users can easily be made to believe that the
optimization process can be influenced by demands with regard to the maximum
allowable analysis time, the fact is that a criterion that corresponds to eqm(4.29)
completely ignores the maximum analysis time selected by the user.
Berridge [414] used a different term to incorporate the analysis time in the optmization
criterion:
C, = ZR, - b It, - tm,,l - d (tmin - t,) (4.30)
where d is another constant, tmin the required “minimum retention time” and t, the
retention time of the first peak. The last term is added in order to avoid overlap of the peaks
of interest with solvent peaks and other signals around t = 1,. For the same reason as
before, the value of tmin is completely irrelevant to the optimization process. However, t,,,
has now become relevant. Eq~(4.30)can be divided into two equations. For t , < t,,,
C, = ZR, + b.t, + d.t, + c’ (4.30a)
where c’ is a constant equal to - (b.t,,, + d.tmin),and for t,> I,,,
C, = ZR, - b.t, + d.t, + C” (4.30b)
where the constant c ” equals (b.t,,, - d.tmin).

According to eqn.(4.30a) the value of the time-dependent criterion C , will increase with
increasing analysis time (tdif the selected maximum analysis time (r,,,,,) has not yet been
reached. According to eqn.(4.30b), an increase of t , above t,,, will result in a negative
contribution to C,. Hence, ,t serves as a desired value for the analysis time, rather than
as a maximum value. The importance of the aim to realize the separation in a time that
equals t,,, can be varied by varying the weighting factor 6.
The criterion C, is always increased if t , increases. In other words, a large value for t ,
will be one of the goals set by the selection of an optimization criterion according to
eqm(4.30). Hence, the two time terms in eq~(4.30)join forces to bunch the peaks between
a retention time of the first peak, which is as high as possible, and the desired maximum
analysis time. It may be expected that the two terms will try to direct the chromatographic
parameters included in the optimization process in opposite directions. The sum of the
resolution factors contributes to this conflict of interest, as ZR, will tend to increase if the
peaks are spread out over larger time intervals. The balance between the different factors
is in principle decided upon by the user in choosing the values of the parameters b and
din eqm(4.30). However, it seems impossible to establish an objective balance between the
importance of resolution on the one hand and retention time on the other. This situation
is disturbing, especially because the course and the result of a selectivity optimization
process will depend on the arbitrarily selected weighting factors.
Smits et al. [416] and Wegscheider et al. [406] incorporated a time correction factor into
their optimization criterion as follows:
c, = c / t, (4.31)
149
where t, again represents the retention time of the last peak*. Essentially, in this way the
obtained separation (expressed in the criterion C ) per unit time becomes the optimization
criterion. Again, a weighting factor may be added, i.e.
c, = c / t: (4.32)
where r is an arbitrary weighting factor. Nevertheless, the choice of r= 1 does seem to be

a natural one.
Unlike the situation with the contribution of a time term (see the discussion above), C,
can readily be expressed in inverse time units (e.g. min-’), so that a dimension problem
will not arise.
Hence, both from the point of view of weighting factors and from that of dimensions
a time factor as in eqn.(4.31) appears to be more appropriate than a time term as in
eqn.(4.28).
A more fundamental advantage of the use of time factors may be that we no longer find
ourselves in a position in which a compromise has to be established between two
conflicting contributions to the optimization criterion. A proper balance between longer
analysis times yielding higher resolution and shorter analysis times yielding lower
resolution may be hard to find. The effect of a change in the chromatographic parameters
will usually be such as to increase one term in eqn.(4.28), but to decrease the other. This
may easily lead to oscillation effects in which the conditions are pushed back and forth,
while the optimum is approached only very slowly. An example where such a problem
seems to occur can be found in ref. [414].
An increase in the retention time accompanied by an increase in the resolution has the
effect of increasing both the numerator and the denominator in eqn.(4.31), so that
oscillations between high and low values are less likely to occur.
Nevertheless, the use of eqn.(4.32) may result in a slower optimization process than if
a simple sum or product criterion were used. It is also unclear at the outset how the process
would respond to chromatograms with the same value for C , but widely different values
for C. In other words, the criterion cannot differentiate between a bad resolution in a short
time and a good resolution in a long time.
This problem can most easily be circumvented by using a threshold criterion for C. If
Cequalled one in acceptable regions of the parameter space and zero outside these regions,
then the result of the use of eqn.(4.32) would correspond to the shortest possible
chromatogram for which the resolution is acceptable (C= 1). A similar situation would
occur if we were to use a threshold criterion with a diffuse change between zero and one,
* In fact, the time needed to elute 95% of the last peak was taken, in which case eqn.(4.31)would read
for Gaussian peaks
c , = - =C C
(4.31 a)
t , + 2 a, t,(l+2/VN)‘
Clearly, the difference between eqm(4.31) and (4.31a) is small. If we assume that genuine
chromatographycolumns have a minimum of 2500 theoretical plates, then the differencebetween the
two equations is always less than 4’/0. Even for non-symmetrical (“tailing”) peaks eqn.(4.31) will
almost always be an adequate approximation of eqn.(4.31a).
150
such as DP. In fact, the criterion
c, = D P / t, (4.3 1b)
is similar to the criterion of eqa(4.24). Because Pdecreases very rapidly once the resolution
of a pair of peaks becomes less than 1, the use of eqn.(4.31b) will not usually result in
chromatograms in which a poor resolution is compensated by a very short analysis time.
The difference between eqns.(4.24) and (4.3 1b) is that in the first case R, and k, are used,
which are useful characteristics if the result of the optimization needs to be transferred t o
another column, while the use of P and t, and eqn.(4.31b) makes this criterion more
appropriate for optimization on the final analytical column.
E q ~ ( 4 . 31b) has the advantage over eqn(4.24) of being a continuous criterion rather
than a combination of two separate ones used hierarchically. We have seen before that the
use of P suffers from the insensitivity of this criterion to changes in the range of high P
values (P = 1)and in the range of badly resolved peaks (P= 0). The use of eqn.(4.31b) will
eliminate the first problem, but the latter problem will remain.
4.4.3 Column-independent time factors
The retention time is determined by three factors:

1. The capacity factor of the solute
2. The column dimensions (hold-up volume)
3. The flow rate.
Only the first factor is influenced by the physico-chemical separation process (the
selectivity), while the other two factors are determined by the column and the operating
conditions, respectively. If C is a continuous criterion (see table 4.7), then both C and C,
can be transferred from one column to another. Both column dimensions and flow rate
have trivial effects on the analysis time t,. However, if the final analysis is to be run on
a different (optimized) column, then it is more logical to use the dimensionless,
column-independent factor (1 + k,) in eqn.(4.31) instead of t,:
c,= C / ( l + kJ. (4.33)
In the case, where the dimensions of the column are to be optimized after completion
of the selectivity optimization process, another time factor may be even more appropriate.
The first step to be taken after the completion of the optimization process is to establish
the required number of plates (N,,). If the lowest resolution value encountered in the
chromatogram is R,,,inr and if the required resolution is R,,,,, then
(4.34)
where N , is the number of plates available on the column used during the optimization
process. If Smin is the lowest value for S in the chromatogram, then
(4.35)
151
Since all continuous criteria (see table 4.7) require knowledge of the R, or S values,
eqn.(4.34) or (4.35) can be used immediately to calculate the required number of plates.
In many cases the required number of plates will only be used to estimate the length of
the final analytical column, while all other parameters are being kept constant.
For example, in GC a capillary column with a given diameter may be used, operated
at a given flow rate, with N being directly proprotional to the column length (L). In GC
or in LC, packed columns with a given particle size may be used at given flow rates. In
all these common cases, the following proportionality series applies:
t , cc to cc Vb cc L cc N,, (4.36)
and hence the final analysis time will be proportional to N,,. Of course, t, is also
proportional to (1 + kd (eqn.l.6), and therefore an appropriate time corrected criterion
is
(4.37)
where f and d denote constant flow rate and diameter (of open columns or of particles in
packed columns), respectively.
The above simple proportionality between t , and N,, (eqn.4.36) is not always obeyed.
For instance, at constant flow rate and particle size, the number of plates that can be
achieved is limited by the maximum allowable column pressure. In that case, we are forced
to vary the flow rate, the particle size, or both. If we do so, the analysis time ( t J will no
longer be proportional to the required number of plates ( N,,).
In chapter 7 (section 7.2.3) we will see that in the case where the pressure drop over a
packed column is kept constant and both the flow rate and the particle size are allowed
to vary in order to realize optimum operating conditions, the retention time will be
proportional to the square of the required number of plates, i.e.
t, = p . N,: . (1 + kJ (4.38)
where pis a constant, the value of which depends on the viscosity of the mobile phase, the
pressure drop and the quality of the packing.
According to eqn.(4.38) a time-corrected optimization criterion under constant pressure
conditions (denoted by the subscript p) may be defined as
(4.39)
In the reality of LC practice, eqn.(4.38), which is based on a different optimum particle

size for a different required number of plates, will not usually be realistic. For LC the truth
may be somewhere in between the two extremes described by eqn.(4.36) (constant flow rate
and particle size) and eqn.(4.38) (constant pressure drop). Relatively long columns with
10 pm particles may be used for difficult separations, requiring relatively large numbers
of plates. For more convenient separations, somewhat shorter 5 pm particle columns may
be used, and for relatively simple separations requiring modest numbers of plates 3 pm
particles packed in very short columns may provide fast analysis.
152
If three or more different particle sizes are to be considered after the completion of the
selectivity optimization process, then this may be an argument for the use of eqn.(4.39)
instead of eqn.(4.37).
The required analysis time itself appears to be both a logical and an elegant choice for
an optimization criterion, Either tnecan be minimized, or, for reasons of consistency, 1I t n e
can be maximized. The criterion of minimum required analysis time then corresponds to
a constant value for C in eqn.(4.37):
4
(4.40)
and with eqn~(4.34)and (4.39, neglecting the constant factors which are irrelevant for
optimization purposes,
(4.41)
An analogous expression can be found for the case of a constant pressure drop over a
packed column with variable particle size from eqn.(4.39):
‘&in
a- g i n (4.42)
CIlP = ( l + k J N ; (I+kJ’
The equations in terms of Smin are especially attractive, since no estimate of the column
plate count is required. If R,is used, an estimate of the peak width or plate count is required
twice, but since R, a VN the plate count cancels in eqm(4.41) and (4.42). This becomes
clear when the two equations are expressed in terms of S.
4.4.4 Time-corrected resolution products
In section 4.3.2 we have seen that the normalized resolution product criterion r aims at
achieving a chromatogram in which all peaks appear at constant resolution intervals from
the first one. If r* is used instead of r, then the regular intervals start at an imaginary peak
at t = c., A chromatogram for which r* = 1 is one of a series for which the constant intervals
can be found. Once the absolute value of S, the number of peaks and the plate number
are known, the chromatogram is defined unambiguously.
The question we can now ask ourselves is whether all.members of the series are equally
good chromatograms. In other words, is the criterion r* on its own sufficient for judging
the quality of a chromatogram. To simplify the discussion, we will investigate the necessity
of time correction factors for chromatograms for which r* = 1, and then extend the result
to include the more realistic chromatograms for which r* < 1.
In the case of a chromatogram for which r* = 1, the value of k, can be calculated if the
absolute value of S (Shas the same value for all pairs of peaks in the chromatogram!) and
the number of peaks is known. The capacity factor for the first peak can be found from
S = - (k, + l)-(ko+ 1)
k, - ko - (4.43)
k0+k,+2 (k,+l)+(k,+l)
and this leads for ko=O to
153
1+s 2s
k, = -( k o + l ) - 1 = -. (4.44)
I-s 1-s
For the second peak we find in a similar way
1+s 1+s *
(4.45)
k“== (k,+l) - 1 = (GI- l
and in general terms
(4.46)
Eqm(4.46) allows us to calculate the k value for the nth peak in an “optimal”
chromatogram (r* = I). Ifwesubstitute S = 2 R,/VNandassume S < 1, then theequation
for the peak capacity (eqn.l.25) follows directly from eqm(4.46). Of course, in practice k
values may also be obtained from the chromatogram instead of from eqr(4.46).
-2 -1 0
log s-
Figure 4.1 1: Calculated characteristics for optimum chromatograms (r* = 1) containing 10 equally
resolved peaks as a function of the separation factor S.Plotted on a logarithmic scale are the capacity
factor of last peak (1 + k, eqn.4.46), the required number of plates (Nne;eqn.4.47), the required
analysis time under conditions of constant flow rate and particle diameter (tnelr.,,; eqn.4.48), and
required analysis time under conditions of constant pressure drop t,&; eqn.4.49). For explanation
see text,
154
In figure 4.11 the function log (1 + k,,) is plotted as a function of log S for
chromatograms with r* = 1. k,, denotes the capacity factor of the 10th peak. Clearly, for
very low values of S all k values are very low. At an S value of 0.03 (log S z - 1.5) all
10 peaks still elute before a k value of 1. Around an S value of 0.12 the 10th peak elutes
at k=10, while at S=O.25 the capacity factor becomes equal to about 100. Roughly,
optimum capacity factors are found around S= 0.1.
The number of plates required for realizing adequate resolution can be found from
eqn.(4.35). If the required resolution (RS.,,) is unity, then
N,, = 4 / smin*. (4.47)
Again, choosing any other value for R,,, is totally irrelevant for the optimization of
selectivity.
In figure 4.11 log N,, is also plotted against log S. In accordance with eqn.(4.47) a
straight line results that has a slope of - 2. The number of required plates quickly decreases
with increasing S.
Two other lines are drawn in figure 4.1 1. These correspond to the time correction factors
in eqm(4.37) and (4.39). Under conditions of constant flow rate and constant diameter
(of particles or open columns) the analysis time, neglecting constant factors, can be
expressed by
(4.48)
while for packed columns under conditions of constant pressure drop and optimal particle
size
(4.49)
Log t,,l,,, and log tnelp are illustrated in figure 4.11. It can be seen that under the
conditions of eqa(4.48) a broad optimum exists around S=O.l. Over the range
0.03 < S < 0.2 the required analysis time varies by about a factor of 2. This range in Scovers
a very large range in k values. For S = 0.03 the capacity factors for ten equally resolved
peaks range from 0.06 to 0.82. For S = 0.2 the capacity factor ranges from 0.50 to 56.7.
Hence, there is a broad optimum range around S = 0.1 in which the required analysis time
does not vary considerably. In this range the criterion r* can be used to try and locate the
best possible chromatogram.
Outside the optimum range this is no longer true. If ten peaks are equally resolved
(r* = 1) with S values of 0.001, then according to eqn.(4.47), four million plates are required
for adequate resolution. Moreover, we can see from figure 4.1 1 that the required analysis
time is a factor of about 600 larger (under constant flow and diameter conditions) than
it would be if S equalled 0.1. If S was 0.5, the analysis time would be a factor of about
200 larger than in the optimum. Hence, we may conclude that for optimization processes
during which the capacity factors may be expected to vary dramatically, a time correction
factor is required even when r* is used as the optimization criterion.
If we consider packed columns under constant pressure conditions, i.e. if we use
eqn.(4.49) instead of eqn.(4.48), then the optimum that corresponds to the shortest analysis
155
time will be observed around S = 0.2. A variation in tneby a factor of two allows operation
in the range 0.1 < S < 0.3. The corresponding ranges in k are 0.22 to 6.44 for ten peaks at
S=O.1 and 0.86 to 487 for ten peaks at S=0.3. Again we see that the optimum range is
quite broad. The range of k values usually considered as optimal, i.e. 1 < k < 10, is well
encompassed in the optimum working ranges of both eqns.(4.48) and (4.49).
It should be noted that in the optimum ranges the number of plates required for “ideal”
chromatograms that show constant resolution intervals throughout is always very small.
The limiting values of S for the optimum ranges correspond to plate numbers of around
4500 (S=0.03) to 45 (S=0.3). When the number of peaks increases, the (1 + k d factor
increases and the optimum shifts towards lower S values (to the left in figure 4.1 1). For
instance, for 15 peaks the optimum S value shifts down from S = 0.1 (400 plates required)
to S=O.O7 (800 plates) if eqn.(4.48) is used, and from S=O.2 (100 plates) to about 0.13
(250 plates) using eqn.(4.49).
Unfortunately, in practice “ideal”.chromatograms showing r* = 1 will be hard to find.
I (C)
0 5 10
k-
Figure 4.12: Three schematicchromatogramswith equal values for the lowest separation factor (Smin.
determined by the first pair of peaks) as well as for the capacity factor of the last peak (kd.
156
Therefore, for a chromatogram with ten peaks and an Smin value of 0.1, the capacity factor
of the tenth peak is bound to be much higher then the value of 6.4 predicted by eqn.(4.46).
In general therefore, the value for Smin in a real chromatogram will be shifted to the left
(lower S values), the number of required plates will be higher and so will the analysis time.
The time correction factors of eqn~(4.48)and (4.49) can readily be used with the lowest
S value (Smi,,)and the largest k value ( k d observed in the chromatogram.
The use of the required analysis time as the optimization criterion (eqns.4.41-42 for
criteria to be maximized, or eqns.4.48-49 for criteria to be minimized) yields a balanced
comparison between the minimum resolution on the one hand and the retentjon on the
other. However, in using the criterion the main disadvantage of the use of Sminas the
function describing the resolution in the entire chromatogram remains. If Smin is used, no
attention is paid to all but one pair of peaks in the chromatogram. By using the required
analysis time as the optimization criterion care is taken that other pairs of peaks do not
extend the length of the chromatogram with impunity. However, even when the Smin
values and the k, values of different chromatograms are the same, these chromatograms
can still be very different.
This is illustrated by the three chromatograms shown in figure 4.12. The Sminvalue,
determined by the first two peaks is the same in all three chromatograms, and so is the k,
value. The top chromatogram shows two peaks early in the chromatogram and a bunch
of peaks between k = 5 and k = 10. The middle chromatogram shows four pairs of peaks
and the bottom chromatogram shows good spacing of the peaks after the first pair.
Table 4.10 lists the required analysis times for the three chromatograms of figure 4.12,
as well as the r* values. Constant flow rate and diameter are chosen as the conditions for
table 4.10. Clearly, the required analysis times are the same for all three chromatograms.
However, r* reveals large differences between the different chromatograms. These changes
in r* are relatively large in comparison with possible changes in the time factor tne. We
concluded from figure 4.1 1 that there were large ranges over which the required analysis
time varied by less than a factor of two. The variation in relative peak positions in the
chromatograms of figure 4.12 gives rise to changes in r* which amount to a factor of 50
between the top and the bottom chromatogram.
Table 4.10:
Required analysis times and time-corrected resolution products for the three chromato-
grams of figure 4.12. Constant flow rate and diameter (of open columns or particles) (i.e.
constant f,d conditions) have been assumed.
Mult. Criterion Eqn. Chromatogram

factor
top middle bottom
103 tne1f.d 4.48 4.4 4.4 4.4

10-4 1’tne1f.d 4.41 2.21 2.27 2.27
10-2 r* 4.21 1.09 7.27 56.4
10-6 rr 4.50 2.47 16.5 128
10-4 ‘“1 4.52 1.30 1.64 2.12
157
The ratio
* *
r, = r / t,, (4.50)
could be used as an optimization criterion to try and accommodate both peak distribution
and analysis time in one criterion. In that case an increase in rf by a factor of 50 would
compensate for an increase in the required analysis time by the same factor. The bottom
chromatogram in figure 4.12 may be more attractive than the top one, but it is quite
obvious that a factor of 50 increase in analysis time is too high a price to pay for the
improved peak distribution.
The reason for this is that changes in resolution are over-emphasizedin the criterion r*,
because n separate resolution factors for n pairs of peaks ( n - 1 if r is used) occur in r*.
In this way, resolution to the nth power is balanced vs. tne. Therefore, a more sensible
criterion would be
rit = V7 / t,, (4.51)
which for the conditions of eqn.(4.48) (constant f,d) becomes
(4.52)
and for the conditions of eqn.(4.49) (constant p)
(4.53)
The values of r&d and r:rlf,d for the three chromatograms of figure 4.12 are included
in table 4.10. The values of r: show an increase of a factor of 50 as discussed before. The
values of rif increase by about 60% in going from the top to the bottom chromatogram.
Hence, the improved peak distribution observed in the bottom chromatogram may
outweigh a 60% increase in analysis time.
4.5 RECOMMENDED CRITERIA FOR THE GENERAL CASE
The final recommendations for optimization criteria to be used in the general case (i.e.
when all peaks are considered to be of equal importance) are summarized in table 4.1 1.
The table shows recommended criteria for four different cases. Below the dashed line, an
alternative criterion (second best choice) is given for each case.
4.6 SPECIFIC PROBLEMS
4.6.1 Limited number of peaks of interest
So far, we have only considered chromatograms in which all peaks were treated as being
of equal importance. Now we will look at a chromatogram in which a number (n)of peaks
appears, but only some peaks @; p < n) are of interest. An example is shown in figure 4.13.
In this chromatogram, eight peaks occur, but only two of these (peaks 3 and 5 ) are of
interest. Seven or eight (if a peak is assumed at t = to) S values can be obtained from the
chromatogram. Four of these involve one of the peaks of interest.
158
Table 4.11:
Recommended criteria for use in selectivity optimization processes in the general case (all
peaks equally important).
ves
I
U Optimization on
final analytical
column?
I
no
sample overall
YFS I no
1 YFS
I no
I
*
max max max r* max rn,
W P , / t, l 7 P / t,
eqn.(4.31b) or max eqm(4.2 1) eqn.(4.52)
LIP, / t" and (4.22) or (4.53)
. . . . . . . . eqn.(4.31
. . . .b). . . . . . . . . . . . . . . . . . . . .
min t , n min t , n max Smin max l/tne
Rs.min > x Rs.min > x
(1) (1)
eqn.(4.24) eqn.(4.24) eqn.(4.25) eqn.(4.41)
or (4.42)
(1) Suggested value for x: 1.5.
The k and S values corresponding to the chromatogram of figure 4.1 3 are given in table
4.12. The relevant values are underlined in this table*.
The chromatogram in figure4.12 can be improved dramatically by using a multicolumn
technique (see also section 6.1). The entire sample can be eluted from a short column to
obtain a rough separation, and only the fraction that contains the relevant peaks (3 and
5) can be passed on to a longer column to realize the entire separation. The first column
may be referred to as a clean-up column or pre-column and the second one as the analytical
column.
However, with or without the use of multicolumn techniques, irrelevant peaks (e.g. peak
no.4) will usually appear in real-life chromatograms and optimization criteria have to be
developed to cope with them.
Most of the criteria used so far can readily be applied to cases where a limited number
of peaks is of interest. The parameters Rs,min, Smin and Pminretain their value, but now
the lowest value should be selected from the relevant pairs of peaks (a pair of peaks is
* The number of relevant S values is never higher than 2p. If two peaks of interest are adjacent in
the chromatogram, then the number of relevant S values is decreased by one.
159
3
N
Figure 4.13: A chromatogramwith eight peaks ( n = 8), of which only two peaks are of interest ( p = 2).
The relevant peaks (3 and 5) are indicated by an asterisk.
Table 4.12:
Retention and resolution data corresponding to the chromatogram of figure 4.13.The
relevant peaks and the relevant values for k and S are underlined.*! was calculated from
eqns.(4.55a) and (4.56a). f,$ was calculated from eqn.(4.57).
Peak no. k 5
0 0
0.20 Smin= 0.067
1 0.5
0.14 Smin
= 0.077
0.20
-
0.25
-
0.091
0.077
0.067
0.16
160
relevant if either of the two peaks is relevant). We distinguish the appropriate criteria for
the specificcase again by underlining, hence13,,in,Smin andLmi,refer to the specificcase.
Even if 0.067 is the lowest separation factor observed in the chromatogram of fig 4.13
(Smin), then 0.077 is still the lowest relevant value (Smin). The analysis time is still
determined by the capacity factor of the last peak (k,= lo), no matter whether or not this
is a relevant peak. However, the lowest relevant value for Rs or S should now be used in
eqn.(4.48) or eqn(4.49).
The criteria which may be used for the specific case in which only a limited number of
peaks is of interest are listed in table 4.13.
Table 4.1 3:
Recommended criteria for use inselectivity optimization processes for the specific case
in which only a limited number of peaks is of interest (see also figure 4.13).
Constant Constant
sample overall
composition? k value?
Yes no Yes no
( 1 ) Suggested value for x: 1.5.

(2) ine can be calculated from eqn.(4.48) or eqn.(4.49), using the lowest relevant value for S emin).
The square root should now be incorporated in all product criteria, i.e. also in the case
in which lTP or I7P, is used instead of nP, This is because it is now now a sensible
convention to incorporate two values for a peak-valley ratio (be it P, P , or P J into the
criterion for every peak of interest. If we did not follow this convention, then a different
situation would exist if two relevant peaks were adjacent (yielding one combined value for
the peak-valley ratio) or separated by an irrelevant peak (which would lead to two different
161
P or P, values). Hence, two values may best be used for every peak of interest and the
resulting product may be “normalized” by taking the square root. For example, we can
write for the product of P values
P
i7P
-= n
i= 1
vPi,i-,.Pi.i+l (4.54)
where Pi,i-, refers to the separation factor between the ith relevant peak and the preceding
one and Pi,i+ that between the ith relevant peak and the following one and where the
number of factors in this product equals the number of relevant peaks @).
If peak-valley ratios are used as elemental criteria, then the separation between the first
peak and the (imaginary) preceding one, as well as the separation between the last peak
and the (imaginary) following one, may readily be characterized by a P value of one. The
retention time of the last peak, which may be used in combination with a product of P
values (see table 4.13) refers to the last appearing peak in the chromatogram, no matter
whether or not this is a relevant peak.
The calculation of (normalized) resolution products if only a limited number of peaks
is of interest requires some additional thought. In calculating normalized resolution
products either R, or S may be used. We will use S in the following discussion. The ideal
situation with regard to the normalized resolution product will be that all relevant S values
are equal, while the irrelevant peaks contribute nothing to the overall length of the
chromatogram (i.e. for the relevant values s=s and and for all other S=O). In that case
the following product should equal unity:
whereSi.i- refers to the separation factor between the ith relevant peak and the preceding
one and &+, that between the ith relevant peak and the following one and where
- n-1
S = l/@-I) c
i= 1
.
Si*i+l (4.56)
When a hypothetical peak is assumed at t = to the calibrated normalized resolution product

becomes:
with
- n-I
s = l / p i =zO si,i+l
. (4.56a)
In these equations, p is again the number of peaks of interest. Hence, the product
includes only the relevant S values, while the sum is taken over all pairs of peaks. If the
sum is divided by the number of relevant peaks (p - 1 or p ) , then a value of r= 1 or r* = 1
can only be reached if all irrelevant peaks appear “nowhere” in the chromatogram, i.e.
coincide with the imaginary peak at t = r,,. If we divided the sum by the total number of
peaks (n), then the resulting values for r would not be restricted to the range O < r< 1, and
we would no longer be able to refer to r as a normalized resolution product.
162
The separation factor between the first peak and the (imaginary) preceding peak, as well
as that between the last peak and the (imaginary) following one, do not take on a natural
value, as was the case for peak-valley ratios (which in both cases could be said to equal
one). In the case of the normalized resolution product the most logical choice is to take
the optimum value, i.e. the average value 3, as the limiting separation factor of the
chromatogram.
The time-corrected normalized resolution product now becomes:
:;t = / t,, (4.57)
where t,, can be found from eqm(4.48) or eqn.(4.49), using the lowest relevant value for
S (Smin) in the calculations.
Weighting factors
The special case described above is in fact a particular example of the use of weighting
factors in the optimization criteria. If some peaks are of interest whereas others are not,
we could use weighting factors of one and zero, respectively, as we have in fact done above.
However, we have seen that even the introduction of these simple weighting factors has
caused some problems regarding the calculation of the various recommended criteria.
These problems will be aggravated if weigthing factors other than one and zero are
allowed.
If a product of peak-valley ratios is used as the optimization criterion, then two values
would need to be used for every peak, one describing its resolution from the preceding peak
in the chromatogram and the other one describing its resolution from the following peak.
Because a product criterion is used, the weighting factors (8) will appear as powers in the
product. If we assume the weighting factors to be positive, we may write
n
nP, = n (Pi.i-, . Pi.i+l)gi’*
i= 1
(4.58)
or, equivalently,
n
nP, = exp { Z ( g / 2 ) In
i= 1
. P i , i + , ) }. (4.59)
This product of peak-valley ratios can be normalized to the sum of weighting factors,
so that the “true” value of the resolution product is less obscured by the arbitrarily selected
values for g:
(4.60)
Eqn.(4.60) is equivalent to an equation suggested by Morgan and Deming [417], apart from
the requirement to use two P values for each peak.
Eqns.(4.58) to (4.60) can be used with P , (if the sample composition is expected to be
constant) or with P or P , if the sample composition is expected to vary.
163
The normalized resolution product (r or r*) can be obtained in an analogous way. In
terms of S the product reads:
(4.61)
- n-1 n
s = x q i + /, x
i= 1 i= 1
gi (4.62)
or, with the inclusion of a (hypothetical peak at t = to)
(4.61a)
with
- n-1 n
s = i =xO si,i+l
/ x
i=O
gi (4.62a)
Table 4.14
Recommended criteria for use in selectivity optimization processes for the specific case
in which weighting factors are used to indicate the importance of each individual peak.
Optimization on
final analytical
column?
Y e no
I i
Constant Constant
sample overall
composition ? k value?
Yes I no
1
Yes I no
I
*
max max max rg max r*,,.g
"p:.g 1 *" "Pi/ t,
eqn.(4.60) or max eqn.(4.61a) eqm(4.63)
W n . g 1t , and (4.62a)
eqn(4.60)
min t, n min t, n
Rs.min x Rs.min 2 x
(1) (1)
eqn.(4.24) eqn.(4.24)
(1) Suggested value for x: 1.5.
164
Finally, the time-corrected normalized resolution product can be found from
I n \
t
= exp In ri / z:
;=o
g; (4.63)
where t,, can again be found from eqn.(4.48) or eqn.(4.49), using the lowest value for S
(Smin)in the calculations. Table 4.14 lists the criteria that may be used in combination with
weighting factors for each peak and refers to the appropriate equations.
The use of Sminor 1/rne as alternative criteria when the optimization does not take place
on the final analytical column (see table 4.1 1 and table 4.13) is not recommended in table
(4.14), because these criteria are not compatible with the use of weighting fqctors.
4.6.2 Programmed analysis
The important aspect of programmed elution techniques with respect to optimization

criteria is that the peak width does not increase with the retention time in a manner
corresponding to eqm(l.16). In programmed analysis a constant peak width is wanted
throughout the chromatogram (see section 6.1).
Because of their pragmatic definitions, the different P values are not at all affected by
changes in the elution mode, i.e. they may be applied under programmed elution
conditions in exactly the same manner as under constant conditions. Resolution (R,)
factors are not fundamentally affected, i.e. the definition given in eqn.(l.l4) can still be
applied. However, the relationship between R, and fundamental chromatographic
parameters given by eqm(l.22) is no longer valid. The separation factor S loses its usual
meaning, since its definition originates from the above eqn.(l.22). A simple solution to this
problem is to use the difference in retention time (At) between two peaks as the sole
criterion for resolution. This is justified by the fact that for ideal elution programmes the
nominator in eqn.(l.l4) has a constant value. Hence, Atmin can be used instead of Rs,min
or Smin.Also, a normalized resolution product may be defined as
n-I
r,, = n
i= 1
- ti) / Z i }
n-l
= n
i= 1
(Ar/z) (4.64)
where
(4.65)
It should be noted that a constant peak width will only be achieved by approximation
in most programmed elution chromatograms. Early eluting peaks (not subjected to
gradient conditions during their migration through the column) may be considerably
narrowed, while late eluting peaks (eluting after completion of the program) may be
considerably broadened.
Harris and Habgood (ref. [427], p.123) have suggested a different definition for a
165
separation factor in programmed temperature gas chromatography. Their definition is
based on the assumption that the width of a peak that is eluted from a gas chromatograph
at a temperature T, during a programmed analysis is the same as it would have been if the
same component had been eluted from the column under isothermal conditions at the
temperature T,, Therefore, the isothermal retention volume at T = T, ( VT) may be used
to characterize the peak width in the denominator of the separation factor, i.e.
- -F
S = VRj- VR,i- Trj- Tr,i (4.66)
‘TPi+ ‘T,J ‘T ‘Tri+ ‘T,J
where V, is the retention volume under programmed elution conditions, Fis the flow rate
(for example expressed in mL/min) and rT the heating rate (e.g. OC/min). Harris and
Habgood proceed to suggest that eqn.(4.66) can be used to explain the influence of the
“programming rate” (r7/ F OC/mL) on the resolution in programmed temperature GC.
However, this is a somewhat simplified picture because the retention temperature T, is
affected by the programming rate. Nevertheless, their conclusion that the resolution in
programmed temperature GC increases with a decrease in the programming rate is
essentially correct.
Snyder [428,429] has suggested a way to relate resolution in solvent programmed
(gradient elution) liquid chromatography to the fundamental parameters of the chromato-
graphic process by defining a median capacity factor under gradient conditions (K),
which is characteristic for the average speed of migration of the solute molecules through
%
the column. In terms of we can use an equation identical to eqn.(1.22) to describe the
resolution R,:
(4.67)
It is important to realize two things in using eqn.(4.67).

1. The median capacity factor 5 is not directly related to the retention time under
gradient conditions. In fact, it can be shown that under someconditionsqhas the same
value for all the peaks in a chromatogram obtained under programmed conditions*.
2. In deriving eqn.(4.67), the relative retention a is assumed to be independent of the
composition. In other words, plots of retention (In k) vs. composition (p)obtained
under isocratic conditions are assumed to yield parallel lines.
For components which are eluted under “ideal” gradient conditions (i.e. those
components that appear neither at the very beginning nor after the end of the actual
gradient in linear solvent strength gradients**, it can be shown that the median capacity
factor is inversely proportional to the gradient steepness parameter, defined as [428]
(4.68)
where S is the slope of the retention (In k) vs. composition (9)plots (see eqn.3.45), V, the
* The suggestion given before that in an “ideal” programmed analysis the p z k width should be the
same for all solutes (see also figure 6.1~)corresponds to the assumption that k, is equal for all peaks.
** For a definition of linear solvent strength (LSS) gradients see section 5.4.
166
hold-up volume of the column and d@dt the programming rate. The latter can be related
directly to the span (Ap) and the duration (1,) of a linear gradient.
It appears from eqn.(4.68) that if the flow rate and the span of the gradient are kept
constant, the gradient steepness parameter ( b ) is inversely proportional to the duration
time (t,) of the gradient, and, hence, that the median capacity factor (k,)
is directly
proportional to t,. Therefore, under these conditions, in gradient elution t, may take the
place of the capacity factor in the resolution equation and eqn.(4.67) may be rewritten
as
(4.67a)
where c is a constant.
Cohen et al. [430] have demonstrated the validity of eqn.(4.67a) in practice for cases in
which a is constant. However, they have also shown that the equation is no longer valid
if a varies with composition under isocratic conditions. Nevertheless, eqn.(4.67a) may
serve as a good rule of thumb for the optimization of gradient duration times (see chapter
6).
Another aspect of programmed elution that will affect the quality of the chromatogram
is the variation (“drift”) of the baseline during the program. Methods to reduce the baseline
drift (or blank signal) and other aspects of programmed analysis will be discussed further
in chapter 6.
4.6.3 Dealing with solvent peaks
In many chromatograms a “solvent peak” will appear. This is typically a large signal
that appears early in the chromatogram. In GC, solvent peaks are usually highly
non-symmetrical (tailing) peaks. In LC they may occur in many different ways, for
instance as large negative and positive signals early in the chromatogram. In some forms
of LC, especially the ionic separation methods (section 3.3) peaks induced by the mobile
phase may occur as genuine (or negative) peaks much later in the chromatogram. This
latter kind of solvent-induced signal can be dealt with as an additional (irrelevant) peak
in the chromatogram, from which the (relevant) peaks need to be separated.
In this section we will discuss some aspects of the more common type of solvent peaks,
i.e. large signals which appear early in the chromatogram. Figure 4.14 shows a typical
chromatogram in which three solute peaks are preceded by a large solvent peak.
There are two fundamentally different ways in which we can deal with solvent peaks.
I. Reduce the solvent signal. This can be done chromatographically, for instance by using
a column switching technique, that prevents the first part of the chromatogram from
entering the analytical column (see also section 6.1). Reduction (or elimination) of the
solvent signal may also be achieved mathematically by subtracting the signal of the
pure solvent (“blank”) from the chromatogram.
2. By modifying the optimization criteria such that an optimum separation of the solutes
from the solvent signal is achieved. This latter method, which is the relevant one in the
context of this chapter, has not received much interest in the literature so far.
Solvent peaks are usually highly non-symmetrical, so that neither R , nor Scan be used.
167
Moreover, neither P nor P, can be established from the chromatogram. The only criterion
which maintains a realistic value is P, As is shown for the first peak in figure 4.14, this
parameter may be estimated from the chromatogram in the usual way (cf. figure 4.2).
Hence, criteria based on P,may be used for chromatograms that resemble figure 4.14. A
great disadvantage of this is that the use of P , has been recommended only for
optimization of samples of constant composition on the final analytical column (tables
4.4 and 4.1 1).
0 k- 5
Figure 4.14 A typical chromatogramcontaining a solvent peak followed by three small peaks. h, and
v, can be used to estimate the peak-valley ratio of the first peak (see figure 4.2.c).
A possible way to deal with solvent peaks if the resolution ( R , ) or the separation factor
(S) is opted for as the elementary criterion (which is to be recommended if the optimization
process does not take place on the final analytical column), is the introduction of (large)
weighting factors for the solvent peak, using for example the criteria described by
eqns.(4.61) to (4.63). For example, if a large weighting factor were assigned to the solvent
peak in figure 4.14 (e.g. g= lo), while small factors were assigned to the (relevant) solute
peaks (e.g. g= l), then the resulting criterion would have the effect of trying to enhance
the separation between the solvent peak and the first peak in the chromatogram. From
there on, a normalized resolution product criterion would again aim at a regular spacing
of peaks in the chromatogram.
168
REFERENCES
401. L.R.Snyder and J.J.Kirkland, An lntroduction to Modern Liquid Chromatography,

Second edition, Wiley, New York, 1979.
402. R.Kaiser, Gas-chromatographie, Geest und Portig, Leibzig, 1960, p.33.
403. 0.E.Schupp 111, Gas Chromatography, Wiley, New York, 1968, p.22.
404. A.B.Cristophe, Chromatographia 4 (1971) 455.
405. A.C.J.H.Drouen, Unpublished results, 1981.
406. W.Wegscheider, E.P.Lankmayr and K.W.Budna, Chromatographia 15 (2982) 498.
407. S.S.Ober in: V.J.Coates, H.J.Noebels and 1.S.Fagerson (eds.), Gas Chromatography,
Academic Press, New York, 1958, pp.41-50.
408. P.Jones and C.A.Wellington, J.Chromatogr. 213 (1981) 357.
409. P.J.Schoenmakers, A.C.J.H.Drouen, H.A.H.Billiet and L.de Galan, Chromarogra-
phia 15 (1982) 688.
410. A.C.J.H.Drouen, P.J.Schoenmakers, H.A.H.Billiet and L.de Galan, Chromatogra-
phia 16 (1982) 48.
411. VSvoboda, J.Chromatogr. 201 (1980) 241.
412. J.E.Knoll and M.R.Midgett, J.Chromatogr.Sci. 20 (1982) 221.
413. H.J.G.Debets, B.L.Bajema and D.A.Doornbos, Anal.Chim. Acra 151 (1983) 131.
414. J.C.Berridge, J.Chromatogr. 244 (1982) 1.
41 5. J.L.Glajch, J.J.Kirkland, K.M.Squire and J.M.Minor, J.Chromatogr. 199 (1980) 57.
416. R.Smits, C.Vanroelen and D.L.Massart, Fresenius Zeitschr.Anal.Chem.273 (1975) 1.
417. S.L.Morgan and S.N.Deming, J.Chromatogr. 112 (1975) 267.
418. M.W.Watson and P.W.Carr, AnaLChem. 51 (1979) 1835.
419. W.A.Spencer and L.B.Rogers, Anal.Chem. 52 (1980) 950.
Vo1.3, Academic Press, New York, 1983, pp.249-341.
421. S.N.Deming and M.L.H.Turoff, AnaLChem. 50 (1978) 546.
422. B.Sachok, J.J.Stranahan and S.N.Deming, Anal.Chern. 53 (1981) 70.
423. J.W.Weyland, H.Rolink and D.A.Doornbos, J.Chromatogr. 247 (1982) 221.
424. P.R.Haddad, A.C.J.H.Drouen, H.A.H.Billiet and L.de Galan, J.Chromatogr. 282
(1983) 71.
425. J.W.Weyland, C.H.P.Bruins and D.A.Doornbos, J.Chrornatogr. Sci. 22 (1984) 31.
426. J.C.Giddings, Anal.Chem. 32 (1960) 1707.
427. W.E.Harris and H. W.Habgood, Programmed Temperature Gas Chromatography,
Wiley, New York, 1966.
428. L.R.Snyder, J.W.Dolan and J.R.Gant, J.Chrornatogr. 165 (1979) 3.
429. L.R.Snyder in: Cs.Horvath (ed.), HPLC, Advances and Perspectives, Vol.1, Academic
Press, New York, 1980, p.207.
430. K.A.Cohen, J.W.Dolan and S.A.Grillo, J.Chromatogr. 316 (1984) 359.
169
CHAPTER 5
- OPTIMIZATION PROCEDURES
The parameters that may be optimized have been discussed in chapter 3 and the criteria
for judging the quality of a chromatogram in chapter 4. In this chapter we will look at the
actual optimization procedures.
A general introduction is given in section 5.1. The specific problems of selectivity
optimization in chromatography are explained in this section and the terminology that will
be used throughout the rest of the chapter is introduced.
Section 5.2 describes simultaneous methods of optimization. In these methods all
experiments are performed according to a pre-planned schedule. After all the experiments
have been performed, the optimum is located.
Section 5.3 describes sequential methods of optimization, in particular the Simplex
method. In sequential methods the optimization procedure starts with some initial
experiments, inspects the data and defines the location of a new data point which is
expected to yield an improved chromatogram. The idea is to approach the optimum step
by step in this way.
In section 5.4 methods are discussed that can be used to reduce the parameter space, i.e.
to restrict the area which is searched for the optimum. Such methods may be used in
conjunction with the optimization procedures described in sections 5.2 and 5.3, but they
are more often used in combination with the optimization procedures described in section
5.5.
This section deals with interpretive optimization methods. In these.methods, the extent
of chromatographic separation is predicted indirectly from the retention behaviour of the
individual solutes. The data are interpreted to locate the optimum in terms of the complete
chromatogram. The interpretive methods may involve a limited number of experiments
according to a pre-planned experimental design (section 5.5.1) or may start with a
minimum number of experiments in order to try and locate the optimum by an iterative
process (section 5.5.2).
For all the interpretive methods described in section 5.5 it is essential to know the
retention data of all the individual components in a sample. Section 5.6 deals with
possibilities to obtain all this chromatographic information.
Finally, the different methods are summarized in section 5.7.
5.1 INTRODUCTION
In this chapter we will describe current procedures that aim at the actual optimization
of selectivity. We will use the information contained in previous chapters. However, full
knowledge of all the information provided in chapters 3 and 4 is by no means necessary.
Clearly, the optimization of selectivity in GC does not require any knowledge of the
parameters that are relevant for instance for ion-pair liquid chromatography. Moreover,
there are some optimization procedures which do not rely on any knowledge of or
information about the parameters to be optimized, nor on how they affect the selectivity.
What the chromatographer needs, however, is sufficient knowledge to decide which
170
parameters he will be optimizing for and to establish sensible upper and lower limits for
these parameters. In principle, this requirement can be circumvented by building
knowledge bases (“expert systems”) into chromatographic instruments (see section 2.2.1).
Although some activity in this area can be observed, it is unlikely that the requirement for
some degree of chromatographic insight for those involved in method development will
soon become obsolete.
The parameters together with their limits define the parameter space. The number of
parameters considered is called the dimensionality of this space. Hence, two parameters
form a two-dimensional parameter space. In this space a two-parameter optimization, or,
alternatively, a two-dimensional optimization may be pursued.
In chapter4 the criteria which can be used for selectivity optimization have been defined
and discussed extensively. In principle, any criterion can be used in combination with any
optimization procedure. Some practical limitations will become obvious in this chapter.
We will generally discuss optimization procedures independent of optimization criteria.
The selected criterion will vary as a function of the selected parameters. This function
is called the response surface. Depending on the selected criterion, the optimization
procedure will be aimed at locating either the maximum or the minimum value of the
response surface. The optimum is defined by those values of the parameters that
correspond to this maximum or minimum.
A very important characteristic of response surfaces in chromatography is their high
degree of complexity. An example is shown in figure 5.1. In this figure a one-dimensional
response surface is shown. The parameter considered is the (binary) composition of the
i”
a min
1.10
1.05
0 0.1 0.2 0.3 0.L

-
0.5 0.6
‘PA
0.7 0.8 0.9 1.0
Figure 5.1 : One-dimensional response surface for the optimization of the stationary phase composi-
tion in gas chromatography. Horizontal axis: composition of binary stationary phase mixture.
Vertical axis: lowest value for a (amin)observed in the chromatogram. For further explanation see
section 5.5.1. Figure taken from ref. [501]. Reprinted with permission.
171
stationary phase for a gas chromatographic separation. The criterion is the lowest value
observed for the relative retention (amin) in the chromatogram.
Typically for this kind of optimization, and, indeed, for all procedures used to optimize
chromatographic selectivity is that at many different points in the parameter space
complete overlap of two components occurs. In figure 5.1 this is the case at all points at
which amin = 1. In two-dimensional optimization processes the conditions at which two
particular components show complete overlap will form a line rather than a point, and
so on.
Selectivity optimization is most useful if parameters can be selected which have a great
effect on the selectivity, and hence complex response surfaces as in figure 5.1 are almost
mandatory. The complexity of response surfaces will be further increased with increasing
complexity of the sample, i.e. with an increasing number of sample components.
Response surfaces in more than one dimension (more than one parameter) are hard to
visualize. Two representations are common for two-dimensional optimization problems,
where the response surface as a function of the two parameters forms a three-dimensional
picture. Figure 5.2 shows a pseudo-isometric three-dimensional plot of such a surface
(figure 5.2a) as well as a contour plot (figure 5.2b).
THF THF
ACN MeOH ACN MeOH
Figure 5.2: (a) Pseudo-isometric three-dimensional response and (b) iso-response contour plot for a
two-parameteroptimization problem. Parameters (in triangular representation):quaternary mobile
phase composition. Criterion:normalized resolution product (see section 4.3.2). 0, is the location of
the optimum. For further details see section 5.5.2. Figure taken from ref. [502]. Reprinted with
permission.
Figure 5.2a is difficult to interpret. Moreover, it may not reveal all the information that
is present. Some information will literally be in the shadow of other parts of the surface
and can only be made visible by turning the surface around (changing the position of
observation). The contour plot on the right is more clear. The use of different tints of grey
or different colours may further enhance the clarity of such a figure.
For three-dimensional optimization processes (leading to four-dimensional response
surfaces) graphical presentation in a single two-dimensional figure is no longer possible.
Any solution to this problem is bound to be difficult to construct and to interpret. A series
172
of contour plots taken along constant intervals of one of the parameters is one of the least
unattractive possibilities.
5.1.1 Univariate optimization
The complexity of the response surface is what makes the optimization of chromato-
graphic selectivity stand out as a particular optimization problem rather than as an
example to which known optimization strategies from other fields can be readily applied.
This is illustrated by the application of univariate optimization. In univariate opti.mization
(or univariate search) methods the parameters of interest are optimized sequentially. An
optimum is located by varying a given parameter and keeping all other parameters
constant. In this way, an “optimum” value is obtained for that particular parameter. From
this moment on the optimum value is assigned to the parameter and another parameter
is varied in order to establish its “optimum value”.
Univariate optimization is a common way of optimizing simple processes, which are
affected by a series of mutually independent parameters. For two parameters such a simple
problem is illustrated in figure 5.3a. In this figure a contour plot corresponding to the
three-dimensional response surface is shown. The independence of the parameters leads
to circular contour lines*. If the value of x is first optimized at some constant value of y
(line 1) and if y is subsequently optimized at the optimum value observed for x, the true
optimum is found in a straightforward way, regardless of the initial choice for the constant
value of y. For this kind of optimization problem univariate optimization clearly is an
attractive method.
However, this is no longer true in other cases. In figure 5.3b the response surface is still
very simple, but because of a mutual interdependence (alternatively called “interaction”)
between the two parameters the contour lines can no longer be transformed into circles.
The simple procedure outlined above no longer results in the real optimum. The latter may
still be reached by repeating the procedure several more times (possibly over smaller ranges
in the parameter space, as indicated by lines 3 to 7 in figure 5.3b), but clearly the
effectiveness and hence the attractiveness of the method are now lost. Moreover, in some
unfortunate cases the variation of a single parameter at the time may not even give an
indication of the location of the optimum [503].
Figure 5 . 3 ~corresponds to a complicated three-dimensional response surface in which
the two parameters x and y are mutually independent. Several optima appear to occur.
Any optimum that occurs on such a response surface is a local optimum. The highest
(lowest, if the absolute minimum is sought) of these optima is the global optimum. If we
apply univariate optimization to this problem (lines 1 and 2 in figure 5.3c), we see that one
of the local optima will be the result of the procedure. This is not necessarily the global
optimum. Indeed, counting the contour lines in figure 5 . 3 ~will reveal that the global
optimum in figure 5 . 3 ~is the one in the bottom left corner. The more the global optimum
dominates the surface, the greater are the chances that it will indeed be found. However,
this is merely a law of statistics, and hence no guarantee is implied.
* The contour lines are taken as circles for reasons of simplicity. Horizontal or vertical ellipses would
yield the same result. These can be transformed into circles by a suitable transformation of the axes.
173
1
-X
Ib)
-X
Figure 5.3: Univariate optimization of individual parameters. (a) A simple response surface with two
mutually independent parameters. (b) A simple surface with dependent parameters. (c) (opposite
page) A complex surface with independent parameters. (d) (opposite page) A complex surface with
dependent parameters.
Drawn lines indicate the course of the optimization procedure. In figure b it is illustrated that for
dependent parameters several reinitiations of the procedure are required to approach the optimum.
174
175
Finally, figure 5.3d illustrates the most difficult optimization problem, that of a
complicated response surface in combination with mutually interdependent parameters.
In this case not even a local optimum may result from the first two steps in the optimization
process.
Unfortunately, for the optimization of chromatographic selectivity we will have to deal
with response surfaces that correspond to figure 5 . 3 or
~ usually figure 5.3d [503],often with
a considerably higher degree of complexity than shown there (compare figure 5.1).
5.1.2 Local vs. global optima
Given the likelihood that many different local optima exist on the response surface, an
important question to ask is how necessary it is to locate the global optimum. It may be
argued that a local optimum may provide an acceptable result in many cases. This will be
true especially when the difference (in response value) between the global optimum and
the local optimum considered is small. In some cases, the local optimum may be more
attractive than the global one because of secondary considerations. For instance, a truly
optimal resolution may be traded off against a lower temperature, a cheaper or less toxic
solvent, or less critical conditions. Due to this last reason a broad local optimum may be
preferred to a somewhat higher sharp global optimum.
There are three prevailing reasons to try and locate the true global optimum. In the first
place, if an analysis needs to be performed a great number of times (e.g. in process
monitoring, quality control), then it will be rewarding to spend time and effort on the
optimization, in order that the analysis can be run quickly and cheaply on a routine basis.
If the expected number of analyses to be run is small, then it is only necessary to reach
sufficient resolution for all components in a reasonable time.
A second reason to aim for the global optimum is given by our inability to tell whether
or not a particular chromatogram constitutes an acceptable result of the optimization
process. Only if we have a sample containing a small number of known solutes, and if we
are sure that no other components (e.g. impurities, degradation products, metabolytes)can
possibly interfere with the chromatogram, only then will we be able to decide readily
whether or not a given chromatogram forms an acceptable result. The danger is that we
accept a chromatogram with say five well separated peaks, while in fact six (or more) major
components were present in the sample. This danger is especially great if the optimization
procedure does not create an impression of the entire response surface. Generally, we are
not able to decide whether or not a local optimum is acceptable, unless we know the global
optimum.
The third reason is that a procedure cannot be designed to find the global optimum only
when it matters. In a number of cases we may give credit to an optimization procedure
for yielding a satisfactory result, even if this result is a local optimum only. However, we
cannot expect the same procedure to yield a satisfactory result in a case in which the global
optimum is the only one at which an acceptable separation may be achieved.
We may conclude that local optima may be more acceptable as a result

1 . the closer the response at the local optimum approaches that at the global one,
2. the fewer the number of analyses that will have to be run on the optimized system,
3. the more information we have about the entire response su$ace, and
176
4. the more we know about the sample.
Finally, it is worth mentioning that we never establish the true global optimum, because
we can never study the effects of all possible parameters simultaneously over the entire
possible range. Because of the parameters we choose and the limits we set for them, the
best we can achieve is the global optimum within the constraints of the parameter space.
5.1.3 Characteristics of optimization procedures
Whatever the optimization procedure may be, it is safe to say that the length of the
procedure, i.e. the number of experiments (chromatograms) required, will tend to increase
with an increase in the number of parameters involved. Hence, increasing the number of
parameters considered leads to
1. increased complexity of the response surface, i.e. more local optima,
2. reduced possibilities for graphical presentation,
3. lengthier optimization processes,
but
4. possibly a better overall result of the optimization.
This latter advantage needs to be balanced against the three considerable disadvantages.
This makes the choice of which parameters to consider during the optimization process
(chapter 3) even more important.
The process may be simplified by defining sensible limits for the parameters. For
instance, GC columns have a specified maximum temperature, but for continuous
operation a practical maximum well below that value is usually observed. Considering the
parameter temperature at a value above the practical maximum is then a waste of time and
effort, which may result in optimum conditions that will not be used in practice (e.g. an
optimum temperature that equals the specified maximum).
After the relevant parameters have been selected and their limits have been defined, the
actual optimization starts with some initial experiments. A sound a priori prediction of
chromatographic retention behaviour is not usually possible for known samples (see
chapter 3) and always impossible for unknown samples. Hence, the initial data for the
optimization process will have to be provided by some sensibly selected experiments. The
initial experiments to be performed will depend on the optimization procedure. They will
be discussed in the following sections.
In some cases, the results of the initial experiments can be used to reduce the number
of parameters considered or to narrow down the limits defined for the selected parameters.
In both cases this can result in great simplifications for the optimization process. This will
be discussed in section 5.4. After the reduction of the parameter space additional
experiments may have to be performed to complete the initial data set.
For each chromatogram the value of the response can be calculated using the selected
criterion (see chapter 4). These values may be used to search for the optimum conditions.
Because of the complexity of the surface, many data points (chromatograms) are needed
to form an impression of the response surface.
Alternatively, the response surface may be calculated indirectly by describing the
capacity factor (or retention time) as a function of the considered parameters, using the
177
data obtained from the initial experiments. We will use the word modelin referring to such
a description of the retention sudace.
There is a distinct advantage in the use of models to describe retention surfaces rather
than response surfaces. The important characteristic of the former is that they are much
more simple than the response surface. Response surfaces form a combination of many
(as many as there are components in the sample) different retention surfaces. In this way,
fewer experiments (chromatograms) may be required to form an overall impression of the
response surface.
The response data are then used to search for the optimum. This may result in the
location of an optimum or in the location of one or more additional experiments. The
decision as to whether or not the final optimum has been located can be incorporated in
a stop-criterion, which decides whether or not additional experiments need to be
performed.
We will refer to methods that try to characterize the response surface indirectly through
the retention surfaces of the individual solutes as interpretive methods. They will be
discussed extensively in section 5.5.
The above discussion can be summarized in a general outline for optimization

procedures. This is shown in figure 5.4. The procedures to be discussed in subsequent
sections can all be assigned a path in this figure.
[Define Liyits I
1
I
2000I 3000 I I
1
I
Calculate Response
Characterize Response Surface1
Search for Optimum

5 - - - - - 7 2
REPORT Criterio
Figure 5.4 General outline for optimization procedures. The numbers shown in the figure can be used
to characterize a particular path in the figure by adding up all numbers found along the way.
178
The numbers shown in figure 5.4 can be used to characterize these paths. A particular
path can be described by adding the numbers that are met along the way (see section 5.7).
At the end of this chapter we will discuss the merits of different optimization procedures.
Some obvious criteria for this discussion have appeared in this introductory section. The
following aspects appear to be relevant:
1. The length of the procedure, i.e. the number of chromatograms required.
2. The result of the procedure, whether it is a local or the global optimum.
3. The possibility to perform multi-parameter optimizations.
4. The complexity of the procedure and the required means of computation.
5. The possibility of unattended automated optimization.
5.1.4 Definitions
We will end section 5.1 with a summary of the most relevant definitions encompassed in
this introductory section.
Parameter Variable to be optimized

Parameter limit Upper or lower limit of a variable
Parameter space All possible combinations of the variables within the parame-
ter limits
Retention surface Retention (capacity factor) as a function of the variables in
the parameter space
Retention line Retention surface for one variable
Model Description of the retention as a function of the variables
Response Criterion value at a particular location in the parameter
space
Response surface Response as a function of the variables in the parameter
space
Response line Response surface for one variable
Initial experiments Chromatograms run at fixed locations in the parameter space
5.2 SIMULTANEOUS METHODS WITHOUT SOLUTE RECOGNITION
Simultaneousmethods are those in which first a decision is made on which experiments

should be performed, then in the second step all these chromatograms are recorded and
in the third step the optimum is located. Of course, not all experiments can be done at the
same time. The word simultaneous can be used because all measurements are done at the
same stage in the optimization process without any calculations or interpretation taking
place in between. In this section we will discuss methods in which the entire chromatogram
is considered. In section 5.5.1 we will look at similar simultaneous methods, in which all
the solutes in the sample are considered individually. In the present section it is not
required to “recognize” the individual solutes in the different chromatograms, i.e. it is not
relevant for the quality of the chromatogram (response value) in which order the peaks
appear and the process is not disturbed by “cross-overs”,i.e. changes in the elution order.
The simplest form of a simultaneous method follows a path in figure 5.4 that can be
characterized by 1011. The term grid search forms an illustrative description of such a
179
process. The parameter space is covered by a grid or raster of experimental conditions
usually at regular intervals. The required number of data points (chromatograms) will be
determined by the complexity of the response surface, the number of parameters
considered and the limits set to these parameters.
The complexity of the response surface will tend to increase with the number of solutes
and with a greater degree of variations in selectivity provided by the selected parameters.
Ironically, we wish to optimize towards those parameters which most affect theselectivity,
hence yield the most convoluted or "rugged" response surfaces and hence require the
greatest numbers of experiments in a grid search approach. A reliable description of the
response surface of figure 5.1 may require some 100 data points at regular intervals along
the composition axis. Because of the selection of the criterion in figure 5.1 the local optima
are most often sharp discontinuities in the response surface. Therefore, they could easily
be missed in a grid search. Moreover, interpolation between the response values at
different data points by means of some (digital) smoothing function around the optima
is not feasible. The selection of a more continuous criterion might enhance the possibilities
for sensible interpolation (see for example figure 5.5).
A greater number of experiments would provide a better description, but would create
a greater experimental work load. A smaller number of points reduces the latter, but also
the chances of locating the global optimum. It will also be clear that an initial coarse grid,
followed by a finer grid located around the highest observed response value (path
characterized by 1012 in figure 5.4) is no solution to this problem. The highest response
found in the initial coarse grid may very well be close to a local optimum, while no data
points happened to be taken in the vicinity of the global optimum.
Especially for the case of figure 5.1, where the composition of the stationary phase is
the continuous parameter along the horizontal axis, a large number of experiments is
extremely unattractive. Fortunately, there are much more effective procedures for this
kind of optimization (see section 5.5.1).
For the important case ofthe Optimizationof the mobile phase composition in reversed
phase LC (RPLC),a typical two-dimensional response surface tends to be much less
rugged, especially if the number of sample components is relatively small ( n < 10). A
typical example is shown in figure 5.5. The seiection of the normalized resolution product
( c eqn.4.19) as the criterion has also contributed to the smoother appearance of figure 5.5
relative to figure 5.1. Note that the criterion r has been recommended in chapter 4 for
optimization processes in which the dimensions of the column are to be optimized after
completion of the procedure (table 4.1 1). Therefore, the grid search approach is more
appropriate for this kind of optimization than for optimization processes on the final
analytical column.
The one parameter in figure 5.5 is the mixing ratio of two iso-eluotropic binary mixtures
(i.e. they are selected to provide roughly the same retention times: see section 3.2). The
compositions are acetonitrile/water 46/54on the left hand side and THF/water 37/63 on
the right hand side of figure 5.5. The resulting mixtures are ternary ones, but there is only
onedegreee of freedom (one parameter). We will discussthe selection of the limiting binary
mixtures extensively in section 5.4.
Some idea of the behaviour of the response surface in figure 5.5 might be obtained by
recording chromatograms at regular intervals of lo%,i.e. 11 experiments along the entire
axis. By interpolation between successive data points there is a reasonable chance that the
180
0 "THF - 0 37
Figure 5 . 5 Example of a response surface for the optimization of the mobile phase in RPLC.
Horizontal axis: ternary mobile phase composition. Drawn line: response surface using the resolution
product as the criterion. Dashed lines: retention surfaces for individual solutes (In k). For further
details see section 5.5.2. Figure taken from ref. [504].Reprinted with permission.
'
0
XI - 4
I
x1-
I
Figure 5.6: Illustration of the data points required for a grid search in a two-parameter space at ten
percent intervals. (a) no constraints; (b) sum of the two parameters not to exceed one.
global optimum will indeed be approached. However, the estimate of 11 experiments is

rather optimistic and the number will have to increase rapidly if the number of components
in the sample increases. For two parameters the number of experiments becomes 11 = 121.
However, if the parameter space is constrained (for instance by the rule that the sum of
the parameters should not exceed unity, which will be the case if x and y are volume
fractions), this number may be reduced to 66. The 66 data points are spread out over an
equilateral triangle rather than a square. This is illustrated in figure 5.6.
181
Table 5.1 gives some figures on the required numbers of experiments for two- and
three-parameter optimizations for cases in which the parameters are constrained or not,
and for a 10% interval as well as for a 5% interval between data points.
Table 5.1:
Number of experiments required for grid search optimization in various cases. The two
numbers given for each situation refer to 10and 5% intervals in the individual parameters.
Number of Constrained (1) Not constrained

parameters
10% 5% 10% 5%
1 - - 11 21
2 66 23 1 121 441
3 286 2925 1331 9261
(1) Sum of the parameters not to exceed one.
The recording of a series of chromatograms under varying conditions will require a

finite amount of time. For the optimization of the major component composition of the
mobile phase in RPLC the situation is relatively favourable due to the fast equilibration
(see section 3.2.2.1). Nevertheless, for a conventional HPLC column with to= 2 min the
actual chromatogram itself would take at least some 20 minutes (O< k < 10). Using ten
minutes as a optimistic estimate for the equilibration time, well over half an hour will be
required for each chromatogram. Of course, small columns with very small particles may
be used for optimization purposes (so-called FAST-LC), and a total time of 20 minutes
for each chromatogram has been claimed. Even if this figure of 20 minutes can be reduced
further, the numbers in table 5.1 do suggest that a grid search approach has a very limited
applicability , especially since many parameters will be much more difficult to control
automatically than the solvent composition in LC.
We can formulate the following conclusions for the grid search approach:
1. The approach is feasible for response surfaces that are not too complicated.
2. Because of 1.,it will be most usefulfor samples containing small numbers of components.
3. Also because of I . , the approach will benefit from the selection of criteria that give rise
to relatively smooth response surfaces. especially in the vicinity of the optima.
4. The criteria suggested by 3. imply that the grid search approach will be less useful for
optimization processes run on thefinal analytical column than it will for cases in which
the column dimensions are optimized last.
5. The length of the procedure will increase rapidly with an increase in the number of
parameters considered. Applications in which more than two parameters are considered
do not appear to be feasible.
6. Once the chromatograms are recorded and the value of the criteria (response)calculated,
very little effort or computation is required.
7. Complete automation of the method is easily possible.
8. Zf the result of a grid search approach is a local optimum rather than the global one, then
182
the result is probably still acceptable, because the procedure provides a good impression
of the entire response sur$ace.
5.3 THE SIMPLEX METHOD
In contrast to the simultaneous optimization procedures described in the previous

section, the Simplex method is a sequential one. A minimum number of initial experiments
is performed, and based on the outcome of these a decision is made on the location of a
subsequent data point. This simplest form of a sequential optimization scheme can be
characterized by the path 1012 in figure 5.4.
The number of initial data points is one more than the number of parameters considered
in the optimization process. These initial experiments define a geometrical figure in the
parameter space which is called a Simplex. A two-dimensional Simplex is a triangle (often
equilateral). A three-dimensional Simplex is a tetrahedron. The description of Simplexes
in more dimensions is somewhat more difficult to envisage, but is mathematically
straightforward.
The initial experiments yield a set of chromatograms, each of which can be assigned a
(criterion) value. Any of the criteria of chapter 4 that yields a single number for each
chromatogram can be used. It will be assumed in the following that a criterion has been
selected for which a maximum value needs to be obtained from the optimization
procedure.
The next step in the Simplex algorithm is to reject the lowest point, i.e. the chromatogram
that yields the lowest response value, and the location of the next data point is found by
reflecting the Simplex in the opposite direction. This process can then be repeated.
Variable 1
Figure 5.7: Illustration of a two-dimensional Simplex optimization. Dotted lines are contour lines:
figures represent the response. Figure adapted from ref. [505]. Reprinted with permission.
183
For a two-dimensional optimization this is illustrated in figure 5.7. The initial Simplex
is the triangle ABC at the bottom of the figure. Because point B yields the lowest response,
it is rejected and the triangle is reflected towards point D. Out of the three remaining data
points, point A is now the lowest and this point is rejected. The Simplex is seen to “walk
up” in the direction of increasing response and will eventually approach the optimum (0).
Around theoptimum, the situation will arise in which the reflection of the triangle results
in a position at which a measurement has already been performed. This will be the case
in the triangle MLN, in which N yields the lowest value. Instead of rejecting N and
returning to measure K, the point with the second lowest response (L) is now rejected and
the triangle is reflected towards point P. This procedure can be repeated until a
measurement has been obtained at point R. Thereafter, no new measurements wilt be
suggested from rejecting either the lowest or the second lowest response value and the
optimization process comes to a halt.
The highest response has been obtained at point M, and hence this is the predicted
optimum. Note that point M has been retained in the Simplex ever since it was first
measured, so that no tables of high values need to be kept and the optimum is one of the
three last remaining points.
It is seen in figure 5.7 that point M is not exactly located at the optimum (point 0).A
decrease in the size of the initial Simplex will imply that the optimum will be approached
more closely, but also that the number of experiments will increase further. Since the
simple optimization described in figure 5.7 has already required 17 experiments, the latter
prospect is not very attractive.
A more rewarding solution to this problem is the use of modified Simplex procedures,
such as first described by Nelder and Mead [507]. Such modified algorithms allow other
operations besides reflecting the triangle, such as contraction or expansion. The manner
in which such a modified Simplex algorithm proceeds is illustrated in figure 5.8 for a
A 20 60 60 80 100
%Water
Figure 5.8: Illustration of a two-dimensional optimization using a modified Simplex algorithm. A

ternary mobile phase for RPLC is being optimized. The third component is acetonitrile.Figure taken
184
two-dimensional chromatographic optimization problem. This example is taken from the
work of Berridge 15051. The two parameters are two of the three volume fractions in a
ternary mobile phase mixture for RPLC. Because the sum of compositions is required to
equal one, only two parameters can be chosen to define the composition of the mixture.
A boundary condition is used to avoid that the sum of the two parameters considered
exceeds one.
Because the initial parameters are taken very close to the edges of the parameter space,
the Simplex necessarily contracts immediately. It typically approaches the optimum area
quickly, but spends much time locating its final optimum around a composition of 50%
water, 20°/0 methanol and (hence) 30% acetonitrile. The advantage of this is that most
measurements are obtained in the location of the optimum. This could also turn into a
disadvantage, because it implies that little information is obtained about the rest of the
response surface, so that no idea can be formed about the merits of the located local
optimum with respect to other optima.
If the response surface is simple, the true optimum can be approached without the need
for a compromisebetween the required accuracy of the resulting optimum and the number
of experiments required, as was the case for the use of a constant step size in figure 5.7.
On the other hand, a stop criterion for the Simplex needs to be defined carefully, because
it will be clear from figure 5.8 that many experiments can easily be wasted in the close
vicinity of the optimum if the requirements are too tight. For example, to locate an
optimum with an accuracy of 0.1% in composition will require much more time (many
more experiments) than if the procedure is stopped when the changes in the composition
in successive steps start to fall well below 1%.
From the point of view of the operation of the Simplex it is advantageous to span the
entire parameter space as much as possible with the initial experiments. From a
chromatographic point of view this is often less attractive. The initial experiments of figure
5.8 had to be performed at mobile phases containing 100% acetonitrile (point A) and 80%
methanol plus 10% acetonitrile (point C). Usually, both these compositions will give rise
to chromatograms with very small k values and hence very little resolution (eqn.1.22).
These chromatograms will be rapid, but hard to characterize by any criterion. At point B,
on the other hand, a mobile phase containing 90% water is likely to yield ample resolution
but with impractically (and maybe even immeasurably) high capacity factors. The choice
for the initial points will have to be made on the basis of considerations regarding the
demands of the Simplex on the one hand and the practical aspects of the chromatography
on the other.
Figure 5.9 shows the result of the optimization procedure illustrated in figure 5.8. Given
that the sample contains four solutes (and not more), the result is reasonable, regardless
of whether it represents a global or a local optimum. A different and perhaps better
chromatogram might have been obtained if a different optimization criterion had been
selected. However, figure 5.9 clearly shows that Simplex optimization may be used with
some success in chromatography.
An important advantage of the Simplex method is that it does not rely on any
chromatographic model and does not require any chromatographic insight. This implies
that a Simplex optimization program can be applied to LSC as well as to RPLC without
any modifications [506]. This is not true for many other methods as will be discussed in
section 5.5.1.
185
1
I , I I I
0 2 L
t lmin - 6 8
Figure 5.9: Resulting chromato ram from the optimization procedure of figure 5.8 obtained at a
d
mobile phase composition of 52 /o water, 21°h methanol and 27% acetonitrile. Figure taken from ref.
The Simplex procedure has some important advantages over other methods:
I. It requires no (chromatographic) information or insight beyond a sensible selection of the
parameters.
2. I t only requires the recording of a chromatogram and the calculation of the response at
each data point. No information regarding the behaviour of the individual solutes is
required.
3. I t can be performed with any number of parameters of any kind. Of course, the number
of required experiments will rapidly increase when an increasing number of parameters
is considered.
However, the Simplex procedure also has some considerable disadvantages, one of
which we have already touched upon. A large number of experiments is usually required
to locate the optimum. Typically, about 40 experiments appear to be required [508,509].
If the parameter space is reduced before the Simplex procedure is started, this number
might be brought down to about 25 (see ref. [510] and section 5.4).
A second, and more serious problem involves the complexity of the response surface.
A simple response surface with one broad optimum, such as the one in figure 5.1, is neither
very common, nor very desirable for the optimization of chromatographic selectivity (see
section 5.1). In the more common case in which the global optimum is the highest of a series
of local optima, the result of the Simplex may very well be one of the latter. Again, it will
be logical that the chances of finding the global optimum.are greatest for simple response
surfaces in which the global optimum is dominant. The fact that Simplex optimization is
most useful for simple response surfaces makes it most useful for simple samples,
containing a limited number of solutes. Also, the inclusion of non-selective parameters,
186
such as the flowrate [508] or to some extent the water content of the mobile phase in
RPLC (figure 5.8) renders the response surface more compatible to Simplex optimi-
zation.
The result of the Simplex procedure is a local or possibly the global optimum. To have
any indication of the relevance of the optimum found, the procedure should ideally be
restarted several times from different sets of initial data points (chromatograms) [5111. This
is the more true since the Simplex procedure provides very little insight into the overall
character of the response surface. However, having to restart the algortithm several times
is in conflict with the large number of experiments required by the Simplex process. This
creates a circle which forms the main objection against the application of Simplex
procedures for the optimization of chromatographic selectivity. This circle is depicted in
figure 5.10.
The disadvantages of the Simplex method can be summarized as follows:

I. A large number of experiments is required.
2. A local optimum may be the result.
3. Little insight into the response surface is obtained.
Restarting the Simplex from different initial experiments will decimate problems 2. and
3. above, but will aggravate point 1. Because of the likelihood that the Simplex
optimization will lead to a local optimum, the use of an initial coarse Simplex in order to
find a suitable area in which an experimental design can be located [512] cannot be
recommended.
Statistical optimization methods other than the Simplex algorithm have only occasio-
nally been used in chromatography. Rafel [513] compared the Simplex method with an
extended Hooke-Jeeves direct search method [514] and the Box-Wilson steepest ascent
path [515] after an initial 23 full factorial design for the parameters methanol-water
composition, temperature and flowrate in RPLC. Although they concluded that the
Hooke-Jeeves method was superior for this particular case, the comparison is neither
representative, nor conclusive.
Drawback of simplex optimization for LC
Many experiments
t o locate
optimum
A 2-l
procedure
Figure 5.10: Figure illustrating the main problem of the use of Simplex procedures for the
optimization of chromatographic selectivity.
187
5.4 REDUCTEON OF THE PARAMETER SPACE
5.4.1 Fall factorial desigm
In section 5.1 we have stressed the importance of selecting the most relevant parameters
for the optimization process and of defining sensible limits for these parameters. The
maximum temperature in GC optimization was discussed as an example of an upper limit
that could be selected sensibly before the optimization is s t a d . Without knowing
anything about the sample, we can guess which parameters will most likely have the
greatest effect on the selectivity. This was done in general terms in chapter 3. The selected
chromatographic system will impose its own limits on the parameter space, e.g. the
maximum temperature of GC columns and the pH limitations of silica-based LC columns.
These constraints are also independent of the sample.
Once we know more about the sample we may be able to narrow down the parameter
space further. Certain parameters may not show any selectivity effects for a certain sample
and may therefore be neglected on the basis of factual knowledge (e.g. pH and ion-pairing
reagents for non-ionizable solutes in LC) or after some initial experiments.
A systematic procedure for the latter has been described by Lindberg et al. [516]. They
suggest the use of a full factorial design to cover the parameter space. If np parameters are
considered and if each parameter is considered to take on lvalues (levels), then the number
of experiments (ne)for the full factorial design is [517]
This exponential relationship prompts us to be careful in the application of full factorial

designs to problems involving many parameters and to minimize the number of levels for
each of them [517). The lowest possible value of 1 is two. If each parameter takes on only
one value (level), then the full factorial design condenses into a single point and loses its
significance. If the number of levels is two, then the study of 3 parameters involves 8
experiments. For 4 parameters this becomes 16 and for 5 parameters 32. Hence, full
factorial designs should only be used if chapter 3 does not provide sufficient insight into
which parameters are the most relevant, and if the optimization problem is important
enough to warrant a very thorough approach.
- ++8 7+++
“1 -.-1 2+--
Figure 5.11: Full factorial design for three parameters at two levels. At each comer the level of the
parameters is indicated. The centre of the cube forms the origin of the design (OOO for parameter
values). The arrows on the left illustrate the three different parameters. Figure taken from ref. [516].
188
Figure 5.1 1 illustrates a full factorial design for three parameters at two levels. The eight
experiments prescribed by eqa(5.1) are positioned on the comers of a cube. The three
parameters each take a high ( + ) and a low (- ) value. Each parameter can be assigned
a direction in the cube.
Figure 5.1 1 shows that parameter 1 ( v , ) will have a low value on the left face of the cube
and a high value on the right face. Parameter 2 (v2) will be low in the front and high in
the back, and parameter 3 (v,) low at the bottom and high at the top. The levels of the three
parameters are indicated at each comer of the cube in figure 5.1 1.
The mean effect of one particular variable can now be estimated by subtracting all
experimental results from the points at which that parameter is low from those results at
which the parameter is high, with all other values equal. Hence, using the design of figure
5.1 1, the function values u> on the left face of the cube may be subtracted from those found
on the right face to yield the mean effect of parameter 1 (e,):
el = cu2-A) + u-f,>+ u,-f,> + U,-f,,} , (5.2)
where4 is the function value obtained at the jth corner of the cube. Besides an estimate
for the mean effect of a parameter, estimates can also be made for the mutual interaction
effects between different parameters, as described in refs. [518] and [519].
Lindberg et al. [516] studied the effects of four parameters for the optimization of a
separation in ion-pairing RPLC. The parameters considered, together with their high and
low values, are given in table 5.2a. Four parameters at two levels lead to 16 data points
(eqn.5.1) and the mean effect for each parameter can be estimated from eight differences
between two data points.
For the optimization of chromatographic separations it is not useful to compare
response(criterion) values at different points in the factorial design. Due to the convoluted
character of the response surface it is unlikely that a good estimate of the influence of one
individual parameter on the response may be obtained from a few data points. If we look
at figure 5.5, the replacement of some methanol by the corresponding amount of THF in
a ternary mixture (i.e. moving from the left to the right in figure 5.5) will sometimes lead
to an increase in the response (r), and sometimes to a decrease, depending on where the
high and the low levels of the parameters are located along the axis in figure 5.5.
For chromatographic separations it is more sensible to compare k values, because
retention surfaces are easier to characterize than response surfaces. Hence, f , in eqm(5.2)
is the capacity factor of the solute at thejth data point. Each solute will have its own values
for kq), and hence a different mean effect can be defined for each sample component and
each parameter. The results for the four solutes and four parameters studied are given in
table 5.2b.
To allow a rapid comparison, the mean relative effects are given, i.e. the difference in
percentage points between two k values at positionsj(high) and h (low) for solute i is found
from
200(k,,- ki.J
Ae, = (5.3)
k i j + ki.h
and the mean effect is found as the average of eight values for Ae,.
It is clear from table 5.2b that the methanol-water ratio and the CSA concentration have
the largest effect on the capacity factor. On this basis Lindberg et al. [516] selected these
189
Table 5.2:
a. Parameters considered for the optimization of a four component mixture in ion-pair
RPLC [516].
Parameter High Low

value value
1. Methanol-water composition (v/v) 38-62 32-68

2. pH 4 2
3. Phosphate buffer concentration (mM) 90 10
4. CSA (1) counter ion concentration (mM) 10 0
(1) Camphor sulfonic acid
b. Calculated mean relative effects of the four parameters of the capacity factors of four
individual solutes. Calculated from data in ref. [516].
Parameter Mean effect on k ('10)
Morphine Codeine Noscapine Papaverine
1. Methanol - 33 - 43 - 80 - 84
2. pH 0 - 1 6 2
3. Buffer - 9 -11 - 15 - 10
4. CSA 81 74 75 66
c. Calculated mean relative effects of the pH and the buffer concentration at different
levels of the CSA concentration. Calculated from data in ref. [516].
Parameter CSA Mean effect on k (%)

conc.
(mM) Morphine Codeine Noscapine Papaverine
2. pH 0 20 17 27 17
2. pH 10 - 21 - 18 - 14 - 13
3. Buffer 0 12 5 - 4 3
3. Buffer 10 - 31 - 28 - 25 - 23
two parameters as the most relevant ones. As expected (see section 3.3), the effect of the
methanol concentration increases with increasing k values. The solutes in table 5.2b are
listed in order of increasing retention, with kMORpH< kcoD 4 k,,,, < kPAp Because
the effect of the CSA concentration is about the same for each solute, an increase in the
190
methanol concentration in combination with a n increase in the CSA concentration can
be used to narrow the big gap between morphine and noscapine in the chromatogram,
while moving the first two solutes away from the solvent front or reducing the analysis
time.
The parameters in table 5.2b do not show considerable specific effects towards the
individual solutes. Ideal parameters for optimization procedures would reveal more
differences along a horizontal line in table 5.2b. The small effect of the pH seems logical
from the selection of a p H range over which the solutes are fully ionized to enhance the
ion-pairing mechanism. However, there is a strong interaction between the p H and the
CSA concentration, which is not revealed in table 5.2b. When the CSA concentration is
0,an increase in the pH has the effect of increasing the retention. If the CSA concentration
is 10 mM, then the p H has the opposite effect. The effect of the buffer concentration is
dependent on the CSA concentration in a similar way. The effects of the pH and the buffer
concentration are shown under the two different conditions with respect to the CSA
concentration in table 5 . 2 ~ Because
. of the averaging, the true effects of the p H and the
buffer concentration on the retention are concealed in table 5.2b. It should be noted that
the differences observed in table 5 . 2 ~would have been much less dramatic if the lower CSA
level had been higher than zero (e.g. lmM), and that pH effects and buffer concentrations
in the absence of a counter ion are not very relevant for ion-pairing RPLC.Nevertheless,
table 5 . 2 ~carries a warning for applying factorial designs in the selection of the most
relevant parameters.
The following conclusions can be formulated:

1. Full factorial designs can be used to select the most relevant parameters.
2. Quite a few experiments are necessary. Therefore, this strategy should only be applied if
the information provided in chapter 3 is insuficient and ifthe analysis to be optimized
warrants a great effort.
3. Individual capacity factors for all solutes need to be measured. This requires individual
injection of each sample component (if known and available) under all conditions, or
advanced detection techniques (see section 5.6).
4. Care must be taken to consider the mutual interaction of individualparameters in order
to avoid an erroneous interpretation of averaged data.
5.4.2 Scouting techniques
In chapter 1 (section 1.5) we have seen that optimum elution conditions require the
solute capacity factor to be in a limited range. Hence, even if we have selected the most
relevant parameters for optimization on the basis of chromatographic knowledge (chapter
3) or a series of carefully selected experiments (section 5.4.1), a large part of the parameter
space may still be irrelevant for optimization purposes, because the capacity factors in
these regions are either too high or too low. Especially those parameters which have a large
effect on retention in chromatography (such as temperature in GC or mobile phase
composition in LC)will show narrow margins. We identified such parameters as “primary
parameters” in chapter 3 (table 3.10). It will be highly beneficial for the efficiency of the
optimization procedure to establish realistic limits for the primary parameters at a n early
stage.
191
This approach is based on eqn.(l.22). It is assumed that the three factors in the resolution
equation can be optimized independently. In this philosophy the retention (k) is first
optimized so that roughly optimal k values are obtained for all solutes. Hence, we are
looking for a few initial experiments,which allow us to narrow down the search to a limited
part of the parameter space. The complicating factor is that we are especially interested
in the primary parameters, i.e. those parameters that have a large effect on the magnitude
of the capacity factors. Therefore, it is not possible to define a fixed set of conditions which
will allow us to elute all samples conveniently. It is highly probable that such a set of fixed
conditions will result in excessively large or small capacity factors.
One solution is therefore to allow a series of conditions to be used, until some idea of
the optimum working range is obtained. For instance, a series of isothermal gas
chromatograms may be recorded, starting at a high temperature and then descending at
regular intervals of (for example) 25 OC, until capacity factors are obtained that are
roughly in the optimum range. A similar method may be used with many other parameters,
such as the mobile phase composition, pH, or concentration of ion-pairing reagent in LC.
It is essential to start a series of such scouting experiments under conditions at which
very low retention may be anticipated for all solutes. In this way, no late eluting peaks will
be overlooked. It is much more practical to increase short capacity factors than it is to
decrease large ones.
The main disadvantage of such a series of sequential scans is the large number of
experiments required to establish the area of optimum capacity factors (see for example
refs. [520] and [521]).
Another possibility for performing a series of isocratic scouting experiments is the

application of thin layer chromatography [522,523]. The data obtained on thin layer plates
may readily be related to capacity factors in column LC (see e.g. ref. [524], p.383). Thin
layer chromatography is especially useful for investigating the possibilities of a series of
stationary phases for the separation of a particular sample, keeping the mobile phase
constant. Different experiments can be run simultaneously on a series of different
stationary phases in TLC. TLC is less attractive when a series of mobile phases has to be
tested on a given stationary phase. An additional advantage of the use of TLC as a
scanning technique is that there is no problem with highly retained compounds. These
would be recognized in TLC as spots around the point of injection. In column
chromatography, they might stay on the column and lead to erroneous interpretation of
the data and eventually to column pollution and degradation.
Programmed analysis
The requirement of a large number of initial experiments can be avoided by using

programmed analysis as a scouting or scanning technique. In GC this is convenientlydone
with a temperature program. This is typically realized by a gradual increase of the oven
temperature after the injection of the sample. Linear temperature programs are almost
exclusively used. The great advantage of such a program is that a large number of solutes
can be made to elute from the column under optimum conditions in one experiment.
Compounds that would yield optimum capacity factors at low temperatures occur early
in the chromatogram, while the components that require a higher temperature elute later
192
(see also section 6.1). Hence, no knowledge of the sample is required to select the initial
conditions.
A convenient rule of thumb is the following [525]:

In order to achieve the same degree of separation in an isothermal analysis as in a
programmed temperature run,a temperature 45 OC below the retention temperature of a pair
of peaks should be selected.
Expressed as a formula, if T,is the retention temperature, i.e. the temperature of the oven
at the time of elution, and To the recommended isothermal temperature, then
To = Tr-45. (5.4)
According to eqn.(5.4), if the result of a programmed temperature scanning experiment

in GC is a bunch of peaks eluted around a column temperature of 195 OC, then a
chromatogram in which all the peaks appear with roughly optimal capacity factors may
be expected to result from an isothermal experiment at 150 OC.
In different forms of LC there may be different primary parameters (see table 3.10). The
term “gradient elution” is generally used for a chromatographic experiment in which the
composition of the mobile phase is varied during the analysis. Salt gradients as well as pH
gradients have been used, especially in IEC [526]. However, the most popular application
of gradient elution involves the composition of the mobile phase. This typically involves
the addition of increasing amounts of a strong solvent (B) to a weak solvent (A). Common
examples involve gradients of water (solvent A) with methanol, acetonitrile or THF
(solvent B) in Reversed Phase LC (RPLC). In Normal Phase LC (NPLC), increasing
amounts of di-isopropyl ether, methylene chloride or chloroform (B) can be added to
n-hexane (A) [527].
A simple way to estimate the appropriate isocratic conditions from the result of a
gradient elution chromatogram is provided by the theory of linear solvent strength (LSS)
gradients of Snyder (for a review, see ref. [528] or [527]). By definition, an LSS gradient
obeys the following relationship:
In this equation kin is the capacity factor, which the solute would show under isocratic
conditions (i.e. an elution at a constant mobile phase composition) corresponding to the
composition at the inlef of the column at the time t that has elapsed since the start of the
gradient. k, is the capacity factor at the start of the gradient ( t = 0), b the gradient steepness
parameter, and to, as usual, the hold-up time of the column.
Clearly, eqn.(5.5) arises as the combination of two effects:
1. The composition of the mobile phase varies as a function of time at the column inlet.’
We refer to this as the gradient program.
2. The capacity factor of the solute varies with the composition of the mobile phase. This
aspect is related to the mechanism of retention (chapter 3).
The shape of the gradient program should be adapted to the mechanism of retention
(i.e. to the particular form of LC) in order to achieve an LSS gradient (eqn.5.5). For
193
example, the variation of retention with mobile phase composition in RPLC can be
approximated by
Ink=Ink,- Sp. (3.45)
Hence, to make the logarithm of the capacity factor linearly dependent on time, the
composition can simply be varied linearly with time, according to
p = A + Bt. (5.6)
The combination of eqm(3.45) and (5.6) yields
log kin = log k, - S ( A + B t)/2.303
SA SBt
= logk, - -+ -* (5.7)
2.303 2.303
A comparison of this equation with eqn.(5.5) shows that the gradient steepness
parameter b is a function of the solute (through S), the gradient program (through B ) and
of the column (through to):
b = S B to / 2.303 . (5.8)
Eq~(5.6)defines a so-called linear gradient. Indeed, linear gradients are most popular
in RPLC [527]. In LSC, retention varies much more strongly with mobile phase
composition than in RPLC, especially when small amounts of organic modifier are added
to the mobile phase (see section 3.2.3). Therefore, concave gradients are to be preferred
[527].
The following is a very convenient rule of thumb for estimating optimum isocratic
conditions from the result of a gradient run [527]:
The mobile phase composition at the column inlet, at a time twice the value of the hold-up
time before the elution of a sample component from the column, may be expected to yield
a capacity factor of three for that component under isocratic conditions.
As an example, we assume a gradient from 100% water to 100% methanol in 20 minutes,
on a column with a to value of 1.5 min. Now a solute that elutes with a retention time t , = 15
min ( t , is the retention time under gradient conditions) is expected to yield k = 3 at the
composition that was reached at the column inlet at t = 15 - 2 x 1.5 = 12 min, which is
60% methanol, 40% water. Assuming that there is no delay time due to instrumental
considerations, this is the composition at the start of the column, but not at the end. One
and a half minutes (to) later, this composition will have reached the end of the column.
If the instrument incorporates a delay, for instance because of the presence of mixing
chambers, this can easily be accounted for. For example, if the delay time is 2 min, then
the composition at the column inlet at t = 12 min. does not equal 60% methanol, but rather
50°/o methanol.
The rule given above is indeed a rule of thumb and not an accurate estimate for the
isocratic behaviour of the solute. It cannot be, since we have seen before that the gradient
194
0 0.5 1
b-
Figure 5.12: Expected capacity factor (k) under isocratic conditions that correspond to the
composition at the column inlet at t = t - 2t,, as a function of the gradient steepness parameter b.
Figure calculated according to ref. [528!.
steepness parameter b will vary with the nature of the solute, because different solutes show
different values of S in eqn.(3.45). However, the rule is a rather robust one, as is illustrated
in figure 5.12. In this figure, the expected capacity factor under isocratic conditions at a
mobile phase composition corresponding to the value at t = t , - 2t, is plotted as a function
of b. It is seen that the resulting capacity factor is indeed around three over a wide range
of b values, incorporating the range of 0.2 < b< 0.4, which Snyder demonstrated to be
optimal [528]. In the range of low b values the simplified model used to construct figure
5.12 is no longer valid, but is is obvious that in the range of very small b values (very slow
gradients) the rule of thumb loses its significance.
If a more accurate prediction of the isocratic elution behaviour is required, then the use
of two [529] or more [531,532] different gradients may provide a possibility. The
disadvantage of this method, besides the need to perform additional experiments, is that
instrumental factors can give rise to quite considerable errors [529,530], so that extreme
precautions may be required.
For the important case of optimizing the solvent eluotropic strength in RPLC, a more
elegant alternative is available. We have seen in chapter 3 (section 3.2.2) that eqn.(3.45)
is a good approximation for the retention behaviour of solutes in RPLC in the range of
optimum capacity factors (1 < k < 10). In chapter 3 we also discussed the validity of the
empirical equation
S =p + q l n k,, (3.46)
which seems to be closely followed, especially for the methanol-water system. If we
195
consider the combination of eqns.(3.45) and (3.46), we observe that knowledge of only one
parameter (i.e. either k, or S) or the capacity factor (1 < k < 10) at one composition is
sufficient to describe retention as a function of composition over the optimum range. A
combination of the two relevant equations yields:
Ink=Ink,-@+ qInk,)p (5.9)
If the value of k has been determined at a certain mobile phase composition, we can obtain
k, from
In ko = (In k + p @/(l - qp) (5.10)
and subsequently S from eqn43.46).

Indeed, eqn. (5.10) has been applied with some success even to extrapolate existing
retention data to find an estimate fdr the retention in pure water, assuming eqn.(3.45) to
be roughly valid even for k + 10 [533]. However, such extrapolations cannot generalIy be
expected to yield reliable results [534].
In RPLC, values for S are usuaIIy between 5 and 10 for small solutes and increase with
the size of the solute molechles [535]. Hence, as a rule of thumb, retention may be expected
to vary by 50 to 100% for a change of 0.1 in p (10% change in composition). This rule
in the range 1 < k < 10. Reckoning with the possibility that nothing
Q
applies to sm 11sol
is known about e pJe, there is only a small range (typically 25-40°/o for small solutes,
but even much sm e for larger molecules) over which optimum k values are observed
for each individual solute. If we consider all solutes in the sample, the working range is
further restricted. Therefore, the one data point that is required to use the combination of
eqns.(5.10), (3.45) and (3.46) to estimate the retention behaviour of a solute may not be
obtained in a straightforward manner. For the same reason as above, therefore, we will
have to rely on a gradient elution run to provide us with the information needed.
The retention behaviour under gradient conditions, assuming both eqm(3.45) and
(3.46) to be valid, can be calculated mathematically 15361. On the basis of such calculations,
we can Fonstruct a diagram that allows us to predict the retention behaviour under
isocraticlconditions from the result of a single gradient run. An example of such a plot is
given in (igure 5.13, for a linear gradient from 1W0/o water to 100°/o methanol in 15minutes.
The to value of the column is relevant in the calculations. For the present case, it was equal
to 125 s. To apply figure 5.13 to other columns, it suffices to adapt the flow rate such that
a similar value for to is obtained. If another gradient or another to value is used, a new plot
should be constructed by the method described in the literature [536].
. Figure 5.13 gives a plot of the volume fraction of the strong solvent that is required to
obtain a given capacity factor for a solute that elutes at a net retention time tk under
gradient conditions. Lines have been drawn that correspond to a series of isocratic
capacity factors (k = 0.5,1,2,5,10 and 20). The optimum range (1 < k < 10) appears grey
in the figure, If we obtain the retention time of a solute under gradient conditions, we can
find from figure 5.13 a series of compositions for different isocratic k values. Hence, figure
5.f3 allows us to estimate the isocratic retention behaviour of a solute from a single
chromatographic experiment.
An example of the application of figure 5.13 is shown in figures 5.14 and 5.15. The
196
k=0.5 1 2 5
s
05
0
0 5 10 15 20
tA/min -25
Figure 5.13: Curves relating the isocratic composition (pJ to the net retention time under gradient
conditions for various values of the isocratic capacity factor. Curves calculated on the basis of
eqm(3.45) and (3.46). Linear gradient 0 - 100°/o methano1 in water. to= 125 s. Figure taken from
ref. [536]. Reprinted with permission.
1 Orcinol 100%MeOH
2 Phenol
3 p-Cresol
L 3.L -Xy lenol
5 3.5-Xylenol
6 2.L-Xylenol
0 5 10
tklmin - 15
Figure 5.14 Gradient elution chromatogram of a mixture of phenolicsolutes. The six numbered peaks
refer to the sample. The remaining signals to the blank. Linear gradient 0 - 100% methanol in water.
ro=125 s. Figure taken from ref. [536]. Reprinted with permission.
former figure shows the result of a gradient elution chromatogram of six phenolic solutes.
Figure 5.15 shows a simplified version of figure 5.13, in which only the curves for k = 1
and k = 10 have been drawn. Both figures correspond to the same linear gradient as figure
5.13. We identify six peaks in the chromatogram under gradient conditions. The other
peaks in the chromatogram correspond to the blank signal*. The gradient program is
* A blank signal is usually inevitable in a gradient run, due to contaminationsin the weaker solvent.
Therefore, a blank experiment needs to be performed and the results of the actual experiment need
to be compared with the blank (see also figure 6.6).
197
\
tktmin -
Figure 5.15: Application of figure 5.13, relating the composition required for k = 1 and k = 10 to the
net retention time under gradient conditions, for the gradient elution chromatogram of figure 5.14.
The first (a) and last ( 0 )peaks that occur in this chromatogram are indicated in the figure. The
required isocratic compositions fall in the range between (pA and cpz Linear gradient 0 - 100 '10
methanol in water. to= 125s. Figure adapted from ref. [536].Reprinted with permission.
illustrated in the figure, by means of the composition at the column outlet. The first peak
elutes at a net retention time of 9 minutes, the last (sixth) after 13.5 minutes. Our task is
now to identify proper isocratic conditions for the bunch of six peaks.
In figure 5.15 we have drawn two vertical lines corresponding to the first peak (a) at
a net retention time of 9 minutes and the last peak ( 0 )after 13.5 minutes. The intersections
of these lines with the lines for k = 1 and k = 10 give us an indication on the optimum
isocratic composition. The first peak is expected to be eluted with k = 1 at a composition
(qA)of 63% methanol in water. If more methanol is used, then the retention of the first
peak will be lower than the optimum range. The last peak is shown to be eluted with k = 10
at a composition (qd that contains 59% methanol in water. If less methanol is present in
the mobile phase, then the last peak in the sample will show a capacity factor that is too
high. Hence, in the range of compositions between qa and qz,coloured grey in figure 5.1 5,
the six peaks are expected to be eluted with capacity factors in the optimum range.
We have learned the following from the application of figure 5.15:
1. The sample of figure 5.14 can be eluted under isocratic conditions, with all capacity
factors in the optimum range.
2. The required compositions contain between 59 and 63% methanol in water.
We may now proceed by investigating either a single composition in the range between
qAand qa or the entire practical range. The former is more commonly done. In that case
the result of a gradient scan is a binary mixtures, which defines the optimal eluotropic
strength of the mobile phase for theelution of the sample. Iso-eluotropic mixtures of other
compositions (using other organic modifiers) may subsequently be exploited for the
optimization of selectivity (see section 5.5).
If only mixtures of a given eluotropic strength are considered as the result of a gradient
scan, then a further optimization of the primary parameter (solvent eluotropic strength)
is not contemplated and the number of parameters involved in the optimization process
is effectively reduced by one. In the optimization of a ternary mobile phase composition
one of the three volume fractions is defined by the two others, as their sum must equal one.
198
If we limit ourselves to iso-eluotropic mixtures, a one-parameter optimization problem
remains. As was described in section 3.2.2, a binary mixture of 60% methanol and 40%
water corresponds approximately to 48% acetonitrile in water or 37% T H F in water. We
may proceed with the optimization procedure by considering these binary mixtures as pure
solvents (e.g. solvent A equals 60140 methanol/water) and refer to them as pseudosolvents
[537] or pseudocomponents [538].
Limitations of the gradient scanning approach
The main disadvantage of the gradient scanning techniques for LC described above, is
the requirement to use selective detectors. Universal detectors, i.e. detectors which register
any solute, will necessarily show a gigantic signal for the change in mobile phase
composition. This background renders the detection of solute molecules impossible.
Hence, selective detectors are required, which do not react to changes in solvent
composition. The UV detector is the most common detector in HPLC. It is compatible with
gradient elution if eluent components are selected which are transparent in the UV.
Unfortunately, a number of solutes will not be detected. For example, aliphatic
hydrocarbons are completely UV-inactive. UV detection can be almost universal,
however, if short wavelengths are selected (e.g. 210 nm). In this case we may talk about
“near universal detection” [543].
The possibility of using short wavelengths will depend on the nature and the purity of
the solvents. From this point of view, acetonitrile may be preferred to methanol for RPLC.
However, as discussed above, methanol-water gradients offer the possibility to estimate
the isocratic retention behaviour fairly accurately from a single gradient run, because of
the validity of eqm(3.46). In mixtures of T H F and water eqn.(3.46) is only approximately
observed, whereas it is completely invalid in mixtures of acetonitrile and water (see table
3.1).
In many cases, selective detection is an advantage rather than a disadvantage. This is
generally the case as long as a detection method is selected which is sensitive to all the
relevant components in the sample.
5.5 INTERPRETIVE METHODS
Interpretive methods of optimization can be described as follows:

1. The chromatographic data is interpreted in terms of the retention surfaces of the
individual components.
2. These surfaces are described by some kind of model. This model may be graphical or
algebraic and based on mathematical or statistical theories, but preferably on
chromatographic insight.
3. Use the model for the retention surfaces of the individual solutes to calculate the
response surface for the complete chromatogram.
4. Search the response surface for the optimum.
Interpretive methods owe their existence to the relative simplicity of the retention
surfaces in comparison to the response surface. Indeed, attempts to describe the latter by
a mathematical model [539,540,541,542] have never been successful. The general idea
behind interpretive methods is is that whereas many experiments are necessary to describe
199
the response surface (see section 5.2), the retention surfaces may be described by an
accurate model on the basis of a small number of experiments.
In the following two sections we will describe two kinds of interpretive methods. In
section 5.5.1 we will discuss simultaneous methods, which involve a fixed experimental
design. In the iterative procedures of section 5.5.2, an initial design that consists of a
minimum number of experiments is used and the location of the next data point is
calculated during the optimization process.
5.5.1 Simultaneous interpretive methods
In this section we will describe several optimization procedures which are simultaneous
in the sense that all experiments are performed according to a pre-planned experimental
design. However, unlike the methods described in section 5.2, the experimentai data are
now interpreted in terms of the individual retention surfaces for all solutes. The “window
diagram” is the best known example of this kind of procedure.
Window diagrams
Window diagrams were developed by h u b and Purnell for the optimization of the
composition of mixed stationary phases for GC (for a review see ref. [Sol] or ref. ISM]).
An example of a window diagram is given in figure 5.16. This figure will be explained
below.
Figure 5.16a is a plot of the retention against the composition. These retention lines
(surfaces) are required for the construction of the actual window diagram (figure 5.16b).
In figure 5.16a the distribution coefficient (0is shown on the vertical axis. If the total
volume of the stationary phase is kept constant, then the phase ratio is constant and K is
directly proportional to the capacity factor k (eqn.l.10). On the horizontal axis is the
mixing ratio of the two components of the stationary phase (9). On the two extremes are
the pure stationary phases S (left) and A (right).
Figure 5.16a can be constructed once the retention data of all solutes have been
measured on the two pure phases. It is assumed that retention (K)varies linearly with
composition (q see section 3.1). Figure 5.16a shows four straight lines, which represent
the (expected)variation of retention with stationary phasecomposition for the four solutes
W,X,Y and Z.
From this figure alone it is possible to get some indication of the optimum conditions
for separation. Any vertical line drawn in figure 5.16a corresponds to a chromatogram that
might be obtained with a particular binary stationary phase mixture. The separation
between individual solutes can be estimated from the intersections of such a vertical line
with the retention lines of the solutes. The two vertical axes represent the chromatograms
on the purestationary phases. On pure S (cp= 0), solute Zis expected to elute first, followed
by Wand Y.Thelatter two solutes will appear as asingle peak. Component X will be eluted
last. If the stationary phase is pure A, then W will elute first, X and Y will co-elute
completely and Z will elute last. At every point where the retention lines for two solutes
intersect in figure 5.16a, the corresponding composition of the stationary phase will give
rise to a complete overlap of two peaks. In figure 5.16a this can be seen to occur around
cp NN 0.2,0.5 and 0.7. At every other composition some separation is predicted, the extent
200
fa1
5 2
0
a2 0.L 06 0.8 1.0
9 A
0 0.2 0.4
‘PA -
0.6 0.8 1.0
Figure 5.16: Example of a window diagram for optimizing the stationary phase composition in GLC.
(a) (top): variation of the retention (distribution coefficient K) with composition for the individual
solutes W,X, Y and Z. (b) (bottom) window diagram showing grey areas (“windows”) at compositions
where all components may be separated. Figure taken from ref. [545]. Reprinted with permission.
of which may be estimated by moving a ruler through figure 5.16a, parallel to the vertical
axis.
This “ruler method” of optimization will suggest that a good separation may be obtained
at compositions around rp z 0.15 or around Q z 0.25. The ruler method is the simplest
and by far the cheapest optimization procedure. It encounters a great deal of scepticism,
because it does not involve the use of a microprocessor. Of course, the ruler method suffers
from severe limitations:
1 . Only a qualitative idea about the optimum chromatogram is formed.
2. I t can only be used for one-parameter optimization problems.
3. The retention lines for the individual solutes need to be known.
Window diagrams, such as figure 5.16b, overcome at least the first of these three
problems. Moreover, they can in principle be expanded to cover two-dimensional
optimization problems (see below). In figure 5.16b lines have been constructed that
201
represent the relative retention (a) of two solutes as a function of composition. For four
solutes a maximum of six values can be defined. If n peaks occur in the chromatogram,
then the number of a values (nJ is generally given by [501,542]
na =
n! - n - (n- 1)
(5.11)
2(n-2)! 2
By definition
aji = kJk, = K/Ki. (5.12)
For the distribution coefficient we can write
K i = Ki.s + rp, A K i , (5.13)
where
A K , = K i , a - Ki,s . (5.14)
Hence, a combination of eqm(5.12) and (5.13) yields the following equation for the
example of the solutes Z and Y:
Kz =
azy = - KZ,s+ P A A K ,
(5.15)
KY KY.s+PAAKY
if K z > K and
- K Y = KY,s+qAAKY
ayz -- (5.1 5a)
KZ KZ..s+qA AKZ
if K , > K ,
Eqns.(5.1 5) and (5.1 5a) describe a hyperbolic function with a discontinuity at the point
where K y = K , i.e. where the retention lines for the two solutes Y and Zintersect in figure
5.16a. This is seen to occur at qA=0.5. Hence, we see the hyperbola for ayz descend
towards a value of 1 at pa = 0.5 in figure 5.16b, and then rise again slowly towards the right
(as azy).Similar hyperbolic lines can be constructed for all five other possible pairs of
solutes and this is done in figure 5.1 6b. Values of asmaller than 1 do not exist by definition.
The final step in the construction of the window diagram is to identify the lowest value
of a which occurs at any composition. In other words, the a,,,,, criterion (see section 4.3.3)
is used to characterize the separation. In figure 5.16b the areas between the axis for a= 1
and the amin value arecoloured grey. Wheregrey areas occur there is a chance of separating
all solutes, provided that the number of plates is sufficiently large. The grey areas,
therefore, are the so-called windows.
It can be seen in figure 5.16b that the highest value of amin is predicted to occur for
q,=O.12 (amin= 1.23). Local optima occur at qAvalues of 0.25, 0.62 and 0.85.
One problem associated with the simple window diagram in figure 5.16b is the use of
aminas the criterion. This was first pointed out by Jones and Wellington [546],who
suggested the use of S,,,,, instead of a,,,. In table 5.3 it is shown that the highest possible
202
value for aminfound from the window diagram does not necessarily correspond to the
highest possible value for Smin,
and hence is not necessarily the global optimum in terms
of the required number of plates (eqn.4.48 or 4.49).
Table 5.3:
Characterization of the optima predicted by the window diagram of figure 5.16b in terms
of several criteria.
-
'PA amin k(1) Smin Nne (2) km tnel,,, (3)
0.12 1.23 0.5 0.034 30,000 1.1 1,800

0.25 1.21 0.7 0.039 24,000 1.25 1,500
0.62 1.05 1.5 0.01 5 168,000 1.6 12,000
0.85 1.04 1.8 0.01 3 227,000 1.85 17,000
(1) For critical pair assuming phase ratio to be 0.001

(2) Eqm(4.35) with R,= 1.5
(3) Eqn.(4.48)
It should be noted that the predicted optima may shift slightly if Sminis used instead
of amin. It is also shown in the last column of table 5.3 that under conditions of constant
flow rate and constant diameter (of open columns or of particles in a packed column) the
optimum at cpA = 0.25 requires a shorter analysis time than the one at cpA = 0.12.
The above discussion illustrates that the window diagram method can be applied with
a variety of criteria. Indeed, the use of amin was not recommended in chapter 4. For the
case of optimization of the stationary phase in GC, a new column will necessarily have
to be prepared either by physically mixing the two stationary phases in the correct
proportion and then coating the column with the mixture, or by combining calculated
lengths of individual columns. In either case, the length of the column should be adapted
to the result of the optimization process. Also, the overall capacity factor may be expected
to vary considerably, as is shown in figure 5.16a. Hence, the recommended criteria are rz,
or l / t n e (see table 4.1 1). If the window diagram method is applied to other problems, for
instance to the optimization of the temperature in GC, then the column may be a fixed
entry and other criteria may be considered.
Laub [544] has suggested the use of l / N n e as a criterion. Noyes [547] suggested that for
optimization on a given column log t , / t , (where retention times and not net retention
times are used) might relate more easily to R, than does amin. However, Smin[546] may
be obtained just as easily from a chromatogram and this quantity is exactly proportional
to R,, as long as the plate count (N) is constant.
The window diagram method also lends itself to the optimization of different
parameters. However, in order to construct the window diagram it is necessary to know
the retention lines or surfaces of the individual solutes. For the optimization of the
stationary phase composition in GC a linear relationship may be assumed between
retention ( Kor k) and composition (volume fraction cp; see section 3.1). Also, the window
diagram method may be very useful for optimizing the stationary phase composition of
203
a mixed phase for LC [548]or for optimizing the temperature (plotting retention vs. 1/ T
15471).Constanzo I5491 applied the method for the optimization of the mobile phase
compositionin ion-pairingLC using a mixture of two pairing-ions (pentane sulfonate and
6or----l
so.
ta
tRlmin
30
20
10
O' if3 3'5 Qo

PH -
Q5 i.0 45 6.0
2.00-
1.75 -
I
a
1.50 -
1.25 -
1.00L
31) 3.5 LO L.5 5.0 ZS 60
- PH-
Figure 5.17: Application of the window diagram method for optimizing the pH in RPLC. Solutes:
S = scopoletin, U = umbelliferone, TF = trans-ferulicacid, TC = trans-p-coumaricacid, CF =
cis-ferutic acid and CC = cis-p-coumaric acid. (a) retention surfaces, (b) window diagram. Figure
taken from ref. 15521. Reprinted with permission.
204
heptane sulfonate). The retention times of the solutes turned out to vary linearly with the
ratio of these two ions.
However, a simple linear relationship does not usually exist. A clear example is the
optimization of the pH in RPLC. The window diagram approach was applied to this
problem by Deming et al. [550,551,552]. They measured the retention of each solute at a
series of pH values (9 in ref. [550], 4 in refs. [551,552]) and fitted the experiments to
eqn.(3.70). This is a three-parameter equation and hence a minimum of three experiments
is required for it to be applied as a description of the retention surface. If more data points
are available, the equation can be fitted to the data by regression analysis.
An example is shown in figure 5.17. In figure 5.17a the retention time is shown as a
function of the pH for six solutes. It can be seen from this figure that four of these solutes
have a p K , value in the pH range studied. Two other solutes do not show a great variation
of retention with pH. Generally speaking, pH optimization is most useful when the
different components in the sample mixture show considerable differences in behaviour,
in other words when the p K , values are different.
From the retention lines in figure 5.17a, the response line (response surface in one
dimension) can be calculated. This is shown in figure 5.17b. Again, a,,,,, has been used as
the criterion. In order to perform the regression analysis on the retention data and to
subsequently calculate the response surface, a computer should be used. This also
facilitates a more efficient calculation of the response. At each value of the pH, the
retention of each solute should be calculated from eqn.(3.70), using the coefficients
obtained from the regression analysis. If the six capacity factors are subsequently arranged
in increasing order, then only five avalues need to be considered to select the value of a,,,,,.
According to eqn.(5.11), the total number of a values for six solutes is 15.
An example of the application of the window diagram method to a similar problem
using Sminas the criterion instead of a,,, may be found in ref. [546].
Hsu et al. [553] applied the window diagram method to the optimization of the
composition of a binary mobile phase in RPLC. However, a straight line was not obtained
by plotting l / k vs. composition and therefore more than two experimental locations were
required.
The latter was also the case for the optimization of the composition of a ternary mobile
phase in RPLC by Issaq et al. [554].The ternary mixture was formed by mixing two limiting
(non iso-eluotropic) binary mixtures and a fourth order polynomial equation was fitted
through five equally spaced data points.
The characteristics of the window diagram method can be summarized as follows:

1. I t is a graphical method, to locate areas (windows) in which allsolutes may be separated.
2, I t can be used in conjunction with a series of optimization criteria. Use of aminis not
recommended.
3. The original window diagram method can be applied to a single parameter only*.
4. The retention surfaces (lines)for all solutes need to be known.
5. Linear relationships are preferred, but not mandatory. For non-linear retention lines
more than two initial (sets of) experiments are required.
* See, however, the discussion below on two-dimensional window diagrams.
205
6. Ifthe number of experimental data points exceeds the minimum requirements to describe
the retention line by a selectedfunction, then the coeficients may be calculated using
regression analysis. The method may then be referred to as a regressive method and the
initial experiments form a regression design.
7. The response surface needs to be calculatedfrom the retention surfaces.
Critical band method
Colin et al. [555] have described a different method to construct a diagram that allows
the prediction of optimum conditions. Their approach is based on the calculation of
so-called critical bands. If the retention surface of a solute j is known, then a forbidden
zone may be defined below the capacity factor ki. If the preceding solute i has a capacity
factor k, which falls in this critical band, then the resolution between iandjis insufficient.
Eqn.(l.20) relates the resolution to the capacity factors of the individual solutes:
kj-ki
.-VN (1.20)
Rs = ki+kj+2 2
We can rewrite this equation and find*
kj(VN-2RJ -4R,
ki = (5.16)
vN+2R,
On a column with a given (average) number of plates, the critical bands can be
calculated for any value of the desired resolution R,. If the retention line (in the case of
a one parameter optimization problem) for solute j is straight, then eqn.(5.16) describes
another straight line. Two applications of this approach are shown in the figures 5.18 and
5.19.
In figure 5.1 8a the retention lines are shown for five aromatic solutes, together with their
shaded critical bands. The vertical axis represents the logarithm of the capacity factor in
RPLC, while the horizontal axis shows the mixing ratio of two (iso-eluotropic) binary
mobile phases, which together constitute a ternary mixture. Separation with all R, values
above 1.6 (the value selected for the construction of figure 5.18a) can be achieved at
compositions at which none of the critical bands overlap. The optimum composition can
be located using the “ruler method (see above), and it is indicated in the figure at a mixing
ratio of 0.83. This corresponds to a mixture containing 33% (0.83 x 40)acetonitrile, 8.5%
(0.17 x 50) methanol and the remaining 58.5% water. The chromatogram obtained with
this mixture is shown in figure 5.18b. All components in the mixture are shown to be well
separated. Clearly, this critical band method is a graphical procedure for the optimization
of a single parameter using a threshold minimum resolution value as the criterion for
separation.
Alternative to the ruler method, other secondary criteria may be applied once separation
windows have been located. For instance, the composition may be selected at which all
* Eqn.(5.16) differs slightly from the original equation [5551, because an alternative equation for the
resolution was used in that work.
206
LO AC N
50 H20 60 H20
1.0 I 1.0
0.5
I
I
1 . 0.0
I
I I I I I 1
11
L
0 1 2 3 L
t/min
L1 1
5
- 6 7 8
Figure 5.18: (a) Figure showing the retention surfaces for some aromatic solutes in RPLC. Critical
bands have been constructed according to eqn.(5.16) below each solute. The dashed line indicates the
optimum ternary mobile phase composition. (b) Chromatogram obtained at the predicted optimum
composition. Figures taken from ref. (5551. Reprinted with permission.
resolution values exceed the threshold (1.6 in the example above), while the capacity factor
of the last eluting peak (kd is the lowest. Another secondary criterion may be a minimum
solvent viscosity (ref. [555], see also ref. [556]).
207
60H20
1.o
I
-J I
log k
I '7
I
QO -
1 1
I
I
I
-0.0
I I , I ,
-
0 1 2 3 L 5
tlmin-6 7 8 9
Figure 5.19: (a) Figure showing the retention surfaces for some aromatic solutes in RPLC. Only solute
nr.11 is assumed to be of interest. Critical bands have been constructed according to eqw(5.16) and
(5.17) below and abovetheretention line for this solute. Thedashed line indicates theoptimum ternary
mobile phase composition. (b) Chromatogram obtained at the predicted optimum composition.
Figures taken from ref. [SSSL Reprinted with permission.
Figure 5.19 shows another example, in which the method is applied t o a specific case
in which only the separation of one solute (3,4-dimethylphenol, no.11) is assumed to be
of interest. In such a case a second line can be drawn above the one for the solute of interest,
which can be characterized by
208
kj ( V N + 2 RJ +4 R ,
k, = (5.17)
VN-2RS
The solute of interest can be separated from all the other solutes at compositions at which
no other retention lines fall within the critical band. This is illustrated in figure 5.19b.
Eqn(5.17) can also be used to allow for the occurance of a solvent peak in the
chromatogram. For example, a large resolution may be demanded between an imaginary
peak at k=O and the first peak in the chromatogram. In that case all solutes may be
assigned a critical band above (eqn.l.17) rather than below (eq.l.16) their retention lines.
Additionally, the method also lends itself to a convenient use of weighting factors in
terms of different resolution requirements for different solutes.
The two applications shown here concern the optimization of the mobile phase
composition in RPLC. However, the method may easily be adapted to other problems. It
is most practical if straight retention lines can be obtained. It should be noted that this is
not usually the case for retention as a function of mobile phase composition in RPLC.In
fact, Colin et al. [555] adapted the value of the hold-up time (to) such as to obtain straight
lines. The fact that they succeeded in doing so for all of 11 solutes considered at the same
time is remarkable, but it may not always be possible. In any case, adapting to in order
to linearize the retention lines will be an awkward practice.
Toon and Rowland [557] used a similar method for the optimization of the composition
of a binary mobile phase in RPLC. However, they did not use eqns.(5.16) and (5.17), but
plotted lines for the observed front and back of the peak. Since these quantities are affected
not only by the capacity factor and the peak width of the solute, but also by the sensitivity
of the detection, the method of Colin et al. is to be preferred.
We may summarize the critical band method of Colin et al. as follows:

1. I t is a simple, graphical method to locate areas where separation may be achieved.
2. Minimum resolution is used as a threshold criterion.
3. The method can be applied to a single parameter only.
4 . The retention surfaces (lines)for all solutes need to be known and preferably linear.
5. Minimal computational means are required.
6. The method lends itself readily to be adapted from the general case to specific
optimization problems.
Extension to multidimensional optimization problems
Window diagrams and related methods may in principle be applied to optimization

problems in more than one dimension. The main difference compared with one-parameter
problems is that graphical procedures become much more difficult and that the role of the
computer becomes more and more important. Deming et al. [558,559]have applied the
window diagram method to the simultaneous optimization of two parameters in RPLC.
The volume fraction of methanol and the concentration of ion-pairing reagent (1 -octane
sulfonic acid) were considered for the optimization of a mixture of 2,6-disubstituted
anilines [558]. A five-parameter model equation was used to describe the retention surface
for each solute. Data were recorded according to a three-level, two-factor experimental
209
design. This implies that the two variables were each assigned three different values (40,
50 and 60% methanol; 0 , 3 and 6 mM ion-pairing reagent).
Experiments were performed at the nine possible combinations of the values of the two
parameters (see eqn.5.1) The model was fitted to the data by regression analysis. From the
retention surfaces, the response surface was calculated.
The same method was applied for optimizing the separation of nine acidic solutes, using
pH (between 3.6 and 6.0) and the concentration of n-octylamine (between 0 and 6 mM)
as the variables [559].The response surface for the latter application is shown in figure 5.20.
The minimum value for the relative retention (amin) was used as the criterion for this figure.
Although the uSe of this criterion has serious disadvantages (see chapter 4), it does not form
an objection for the application of the optimization procedure itself, since once the
retention surfaces are known, the response surface can readily be calculated for a variety
of optimization criteria. The response surface of figure 5.20 suggests a broad global
optimum at a pH close to 5.8 and a reagent concentration around 3.2 mM. The
chromatogram obtained under these conditions is shown in figure 5.21.
I 1.3
1.2
1.1
1.0
Figure 5.20 : Response surface (“two-dimensionalwindow diagram”) for the separation of a mixture
of nine acidic solutes by RPLC. Variables are pH and the concentration of an “ion interaction
reagent” (NOA = n-octylamine). The vertical axis represents the lowest value of a observed for any
combination of two solutes in the sample (amin).Figure taken from ref. [559]. Reprinted with
permission.
Weyland ef al. [560,561] used this method to optimize ternary mobile phase composi-
tions for the separation of sulfonamides by RPLC. They fitted the retention surfaces to
a quadratic model similar to eqn.(3.39), and also used a combination of a threshold
resolution and minimum analysis time (min t , n Rs,min> 1.25; eqn.4.24) [560]. This
criterion may yield a good optimum if the optimization is performed on the final analytical
column (see table 4.1 1).
Otto and Wegscheider [562, 5631 applied the window diagram method for the
simultaneous optimization of the (binary methanol-water) mobile phase composition, the
ionic strength and the pH for the separation of ionic solutes in RPLC. They fitted the
experimental data to a semi-empirical 13-parameter equation based on eqn.(3.45) for the
composition effect, eqn.(3.71) for the effect of the ionic strength and eqn.(3.70) for the
210
0 25 50
tlmin-
Figure 5.21 : Resulting chromatogram at the optimum conditions predicted by figure 5.20. pH = 5.8;
concentration of n-octylamine = 3.2 mM. ODS column; mobile phase methanol-water (20180) with
0.010 M acetate buffer. Solutes: E = phenylethylamine, P = phenylalanine, V = vanillic acid, C =
trans caffeic acid, M = trans p-coumaric acid, F = trans ferulic acid, A = phenylacetic acid, H =
hydrocinnamic acid and N = trans cinnamic acid. Figure taken from ref. [559]. Reprinted with
permission.
effect of the pH*. The solutes included an amino acid, a weak diprotic acid, two dipeptides
(zwitter ions) and three isomeric amino-benzoic acids.
In ref. [562] the data were collected according to a 6 x 3 x 2 factorial design (36
experimental locations at six values for the pH between 2 and 7, three values for the
methanol content between 10 and 30% and ionic strengths of 0.1 and 0.2M). Additionally,
data were taken along a vector in the parameter space for which only the pH varied.
A limited experimental design was used in ref. [563], in which a citrate buffer was used
instead of the phosphate buffer in ref. [562]. The two different buffers yielded markedly
different optimum conditions.
It was found that in order to locate the global optimum the entire parameter space had
to be searched with a computer. A grid with 0.1 unit steps in pH, 2% steps in methanol
concentration and 0.01 M steps in ionic strength [562] meant that over 6000 points had to
be calculated. This indicates that whereas the window diagram method can be applied in
more than one dimension, a considerable price has to be paid in terms of both the number
of experiments (depending on the model) and the computation time required. The validity
of the calculated optimum will mainly depend on the accuracy of the model that is used
to describe the data.
Clearly, the response surface for the three-parameter optimization problems discussed
* For equations for diprotic acids and basis see ref. [316], p.239 et seq..
21 1
above are four-dimensional (hyper-) surfaces, which cannot be visualized. Hence, the term
window diagram refers to a strictly mathematical, rather than a graphical procedure.
Sentinel metbad
A similar method has been developed for the optimization of the mobile phase
composition in RPLC or LSC by Glajch et al. [542,564]. In their scheme, an optimum
quaternary mobile phase composition is the goal of the optimization process. After
reduction of the parameter space to solvents of equal eluotropic strength with k values in
the optimum range, a fixed experimental design, referred to as a Simplex design,* is
applied. This design consists of seven experiments, which are illustrated in figure 5.22 and
listed in table 5.4a. Because the sum of the volume fractions of the three binary solvents
always equals one, we are dealing with a two parameter optimization problem.
Figure 5.22: Experimental design as used in the Sentinel method of Glajch et al. [542]. Figure taken
Praton Acceptor
Proton donor Xn -- Dipole interaction
Figure 5.23: Illustration of preferred modifiers for NPLC (dashed triangle) and for RPLC (dashed
and dotted triangle) in the Snyder selectivity triangle (see section 2.3.3). Figure taken from ref. [542].
* The term Simplex design (or Simplex lattice design) is unfortunate. It creates confusion between
the Simplex method for optimization (section 5.3) and the Sentinel method of Glajch et al., which
is a very different method by all accounts. To avoid further confusion, we will not use the word
Simplex in connection with the Sentinel method.
212
The points on the sides of the triangle (A, B, and C) represent three iso-eluotropic binary
mixtures (solvents A, B and C). The composition of one of the three binary mixtures (i.e.
the appropriate eluotropic strength) should be determined by either a scanning gradient
or a stepwise series of isocratic scans (see section 5.4). Once one of the compositions is
known, the compositions of the two iso-eluotropicbinary mixtures can be calculated using
the conversion factors given in table 5.4b.
The three mixtures A, B and Care mixed in the ratios listed in table 5.4a to yield three
ternary mixtures (in the middle of each of the sides) and one quaternary one (in the centre
of the triangle).
Table 5.4
Summary of mobile phase compositions to be used in the experimental design for the
Sentinel method.
a. Solvent composition at each experimental location.
Experiment no. Volume fractions of binary solvents
'PA 'Pi3 'PC
0 0 0
0 1 0
0 0 1
112 1/2 0
1/2 0 1/2
0 112 1/2
1/3 113 1/3
b. Preferred solvents and required volume fractions in iso-eluotropic binary mixtures

(1).
Column Base solvent Modifiers O/O(V/V)
RPLC Water A Methanol 50 (2)

B Acetonitrile 41
C THF 36
LSC n-Hexane A Diethyl ether (3) 50 (2)

B Chloroform 36
C Methylene chloride 45
(1) Percentages given correspond to section 3.2.2 and not to the original publication ref. [542].
(2) Arbitrary value
(3) Methyl t-butyl ether is to be preferred for practical reasons (ref. [5241, p.366).
213
Glajch ef al. characterize the retention surface by a quadratic model (similar to
eqn.3.39). In such an equation only six coefficients appear, so that the seventh experiment
allows the coefficients to be estimated by regression analysis. An advantage of that may
be that random experimental errors in one or more of the data points become less
significant. There is, however, also a potential disadvantage. If the quadratic equation is
not adequate for the description of the response surface, then the regression analysis
creates errors in the description of the retention surface throughout the parameter space.
Therefore, the model induces an error in the description of the data even at the the
experimental locations.
In general, the following procedure should be recommended. If the main source of error
is in the inaccuracy of the experimentation, then it may be advantageous to fit a model
equation to the data by means of regression analysis. If, however, the limiting factor is the
inaccuracy of the model, then regression analysis should not be applied. An alternative
to regression analysis is a division of the parameter space in segments (see section 5.5.2).
Glajch et al. [542]have used three additional experiments to improve the accuracy of the
optimum predicted by the quadratic model. We will return to the problem of model
inaccpracies in section 5.5.2.
Once the retention surfaces are known, any criterion may in principle be used to
calculate the response surface and to locate the optimum composition. One of the criteria
used by Glajch et al. is the threshold minimum resolution criterion (section 4.3.3). This is
done by means of a graphical procedure, referred to as overlapping resolution mapping or
ORM. This procedure involves the location of areas in the triangle where the resolution
R, exceeds a certain threshold value. This is repeated for all pairs of solutes and the results
are combined to form a single figure.
An example of the procedure and the resulting overlapping resolution map is shown in
figures 5.24 and 5.25. Because of the simple retention surfaces, each pair of peaks yields
2-3 6-7
3-L 7- 8
L- 5 8-9
Figure 5.24 : Overlapping resolution maps for eight relevant solute pairs for the separation of nine
substituted naphthalenes. Figure taken from Glajch et al. [542]. Reprinted with permission.
214
pair
2-3 \
ACN
Figure 5.25 : Overlapping resolution map (ORM) for all nine solutes of the sample in figure 5.24.
Figure 5.25 is the result of superimposing all eight triangles of figure 5.24. Figure taken from Glajch
ef al. [542]. Reprinted with permission.
0.00 2.25 L.50

tlmin - 6.75 9.00
Figure 5.26 :Resulting chromatogram corresponding to the ORM procedure illustrated in figures 5.24
and 5.25. Chromatogram recorded at the optimum composition indicated in figure 5.25 (32'/0 ACN,
15% THF, 53% water and 0% MeOH). Figure taken from Glajch et al. [542]. Reprinted with
permission.
a simple figure in which two areas may be identified. The area where the resolution exceeds
the threshold value is left white, the remaining area in the triangle is grey. Eight different
plots are shown in figure 5.24. Potentially (eqn.5.1 I), there are 36 of such triangles for 9
solutes, but the remaining pairs of peaks are irrelevant (white triangles) and not shown in
the figure. All the different plots may then be combined to form the final ORM (figure
5.25).
There is a clear analogy between this type of figure and a window diagram. In the white
area, which may be called a window, the resolution will be at least 1.5 (for the example
in figure 5.25) for all pairs of peaks. This is illustrated in figure 5.26, which shows the
215
chromatogram that is obtained at a composition within the window. This composition is
indicated with a circled x in figure 5.25.
A disadvantage of the ORM method relative to the window diagram method is that no
idea can be formed of the magnitude of R , within the window area. Hence, the exact
optimum cannot be located. This follows as a logical consequence of the use of R , as a
threshold criterion (see discussion in section 4.3.3). ORM may be used to select areas for
operation in the parameter space, using the same column for the optimization procedure
as for the analysis to be performed later. Because the retention surfaces of the individual
solutes have been characterized during the procedure, a new optimum can be located on
a different column (or different flow rate) by reinitiation of the calculation procedure
using another value for the threshold. The calculation step needs to be repeated
completely, but no new experiments are required. Because the number of solute pairs
increases very quickly with the number of sample components (see the discussion
on window diagrams above), the Sentinel method is most useful for relatively simple
samples.
Issaq et al. [566,567]used the ORM method with a ten-point design, in which additional
experiments were performed at compositions containing the three pseudosolvents in 4 1:1
ratios.
Application of the Sentinel method to LSC
Snyder, Glajch and Kirkland [568,569,570,571] have paid much attention to the
possibilities of using a similar experimental design for optimizing the mobile phase
selectivity in LSC. Unlike the situation in RPLC, it cannot be assumed that any mixture
of two iso-eluotropic mixtures will yield a new mixture which is in turn iso-eluotropic.
Snyder and Glajch [568] conducted a theoretical study on the possibility of calculating the
eluotropic strength for binary solvent mixtures in LSC with a sufficient accuracy. This
approach was expanded by Glajch and Snyder [569] to include ternary and quaternary
mixtures.
Snyder, Glajch and Kirkland [570] introduced two new parameters to describe the
selectivity effects in the optimization triangle for LSC. If methylene chloride (MC),
acetonitrile (ACN) and methyl t-butyl ether (MtBE) are used as the preferred modifiers
in n-hexane, then an empirical solvent selectivity parameter ( m ) can be defined which is
low for methylene chloride and can be made equal for the other two binary solvents. The
latter is achieved by adding the appropriate amount of methylene chloride to the
hexane-ACN binary. Addition of MC is required at any rate, because hexane and ACN
are not miscible in all proportions. By definition we can assume m to equal zero for the
hexane-MC binary mixture and m to equal one for the two other binaries.
A second parameter can be defined as the ratio of the concentrations of the two
localizing solvents:
R = [MtBE]/([MtBE] + [ACN]).
The selectivity parameters for the various solvents in the experimental design of figure
5.22 are listed in table 5.4.5. The combination of pseudosolvents for each experiment is
also listed in the table.
216
Table 5.5:
Summary of mobile phase compositions to be used in the experimental design for the
Sentinel-method as applied to LSC. Selectivity parameters for LSC solvents according to
the experimental design of figure 5.22.
solvent A : Dichloromethane (Methylene chloride; MC)

solvent B : Methyl t-butyl ether (MtBE)
solvent C : Acetonitrile (ACN)
base solvent : n-Hexane
Experiment no. Selectivity parameters Solvent

combination
m R
0 - A
1 0 B
1 1 c (1)
1/2 0 A/ B
1/2 1 A/C
1 1/2 B/ C
213 1/2 A / B/ C
(1) Small amount of MC added to promote miscibility.
A serious disadvantage of using these new selectivity parameters is that they are not
related to volume fractions (or mole fractions) in a straightforward way. Procedures have
been described which can be used to calculate the eluotropic strength of binary 15681and
more complicated mixtures [569] and the selectivity parameters 15691. However, these
already complicated (iterative) procedures are only applicable to solvents of known
composition and calculating the composition once the required values of the solvent
strength (8)and of the selectivity parameters ( mand R) are known is highly complicated.
Therefore, it seems that simplifications are required to create a useful system for the rapid
estimation of iso-eluotropic binary, ternary and quaternary solvents using the preferred
modifiers for LSC.
So far, an empirical approach that neglects the specific problems of LSC has appeared
more feasible. Antle [572] demonstrated the applicability of the Sentinel method to LSC,
using mixed mobile phases corresponding to table 5.4a, i.e. mixing the individual binary
mixtures according to their volume fractions. This yielded some success, although
admittedly not all solvents were iso-eluotropic.
Use of different stationary phases /
Glajch et al. [573] have expanded the Sentinel method to include the “simultaneous”
optimization of the stationary phase. They applied the complete seven-point experimental
217
design in figure 5.22 to three different columns (chemically bonded alkyl, cyano and
phenyl phases). Three separate ORM maps were constructed for the different columns and
the highest optimum was selected. In this way the scope of the method could be expanded
without a dramatic increase in the number of experiments required. However, no attempt
was made to correlate the data obtained with equal mobile phases on different columns.
Only if there is a considerable interaction between the mobile and the stationary phase (e.g.
because of specificabsorption of solvent components) will all 21 data points be significant.
If the reverse is true and the stationary phase effects are independent of the mobile phase,
then only one experiment is required on each additional stationary phase. Hence, nine
experimental locations (7 + 1 + 1) would be sufficient to investigate the behaviour of three
different stationary phases. An optimum number of data points for a complete optimiza-
tion using three modifiers and three stationary phases may be somewhere in between the
minimum number of 9 and the maximum number of 21.
It was claimed [573] that for complicated samples, such as the separation of 20
phenylthiohydantoin (PTH) derivatives of amino acids, the optimization of many
parameters simultaneously is required to achieve sufficient selectivity. However, in ref.
[573] the pH was optimized separately, before starting the complete three-parameter
optimization with two continuous parameters and one discrete one.
A more or less opposite goal was pursued by de Smet et al. [574], who attempted to reduce
the number of stationary phases to a single one, by choosing a cyanopropyl bonded phase
of intermediate polarity, which can be used in both the normal phase and the reversed
phase mode (see figure 3.8). Furthermore, because of a clever choice of modifiers, the total
number of solvents required was restricted to six: n-hexane, dichloromethane, acetonitrile
and THF for NPLC and the latter two plus methanol and water for RPLC. A variety of
drug samples could be separated with a selected number of binary and ternary mobile
phase mixtures.
The advantage of the simplified procedure described by de Smet et al. is the use of only
one column and six solvents, which enhances the possibilities for fully automated
optimization on relatively simple commercial instruments. The disadvantages are that the
column of intermediate polarity could lead to a reduced general selectivity in both modes
(see figure 3.8) and the long equilibration procedure (about 2 hours, involving several
gradients), which is required to switch from the reversed phase to the normal phase mode
and vice versa.
Since it is easily possible to change columns automatically with the aid of selection
valves, it appears that an approach involving a minimum of two colums (one for NPLC
and one for RPLC) is generally to be preferred.
Expanding the parameter space to non-iso-eluotropic solvents
The parameter space in the original Sentinel method is restricted to a series of

iso-eluotropic solvents, which means that only a very small fraction of all possible
quaternary mixtures is considered. This is illustrated in figure 5.27a.
Poyle [575] and d'Agostino et al. [537] have shown that a higher optimum might be
located outside the iso-eluotropic plane in figure 5.27a. The exclusiveuse of iso-eluotropic
solvents may be justified on the following two grounds:
218
(a1
Figure 5.27 : (a) Illustration of the location of the Sentinel experimental design in the tetrahedron
representing all possible quaternary solvents. (b) Illustration of a 12 point experimental design in
which a range of solvent strengths is considered. Figures taken from ref. [537].Reprinted with
permission.
1. The possibilities of varying the eluotropic strength of the eluent, while keeping all k
values in the optimum range, are usually very limited.
2. If an extra parameter is considered, the optimization procedure becomes much more
time-consuming.
d’Agostino et al. [537] modified the method of Glajch et al. [542] so that a total of twelve
experiments is performed in a “truncated pyramid”, i.e. a “slice” of the tetrahedron located
around the iso-eluotropic plane of figure 5.27a. This design is illustrated in figure 5.27b.
On a microcomputer it took d’Agostino et al. one hour of computation time to locate the
optimum with a grid search of the entire response surface at 4% intervals. To find the
optimum with 1% steps (corresponding to steps in the solvent concentrations between 0.1
and 0.7%) took no less than 14 hours of calculation time [537].
Summary
The advantages of the use of fixed regression designs, including multi-dimensional

window diagrams and the Sentinel method, can be summarized as follows:
1. The experimental procedure is straightforward and may easily be automated.
2. A good impression is formed of the entire response surjiace.
3. Any desired criterion may be used for the caldlation of the response surface.
4. Because the retention surjiaces are known, the calculation step of the procedure can be
repeated using diflerent conditions, such as another criterion or another column.
5. The method is relatively fast and simple, because only a limited parameter space is
considered.
The disadvantages are:

1. Experiments are spread out over the entire parameter space. The accuracy of the
219
description may therefore be expected to be roughly equal at each location in the
parameter space. Extra experiments may be required i f a more accurate description of
the response surface around the optimum is required.
Zfthe retention surfaceis characterized by a modelequation, then the accuracy with which
this equation describes the true surface becomes a limiting factor.
To characterize the retention surfaces, individual capacityfactors need to be obtained for
all solutes.
If the parameter space is reduced to a two-dimensional triangle (Sentinel method), then
a better optimum outside this plane may be neglected.
5.5.2 Iterative designs
The first two of the above disadvantages of fixed design methods can be overcome by
the use of iterative designs. These are methods in which an initial design that contains a
minimum number of data points is used, then the results are investigated and the results
of that investigation are used to conclude whether or not one or more new experiments
are required, as well as where these additional experiments should be located in the
parameter space.
The meaning of this complex definition is illustrated in figure 5.28. The procedure starts
with a (small) set of initial experiments. The next step is the application of a model to the
data. This model can be a graphical or a mathematical one, but may also be a simple linear
interpolation between the individual data points. Typically, the model is applied to the
retention surfaces of the individual solutes, and not to the response surface. Alternatively
[537], it may describe relative retentions with respect to a reference component in the
Initial
exps.
Figure 5.28 : Illustration of the operation of iterative designs for the optimization of chromatographic
selectivity.
220
sample. The simplicity of the retention surfaces allows a reasonable approximation to be
made from only a very limited number of experiments.
The model is then used in a calculation step to predict the location of the optimum. This
step involves the calculation of the response surface from the retention surfaces using a
suitable criterion, the location of the (predicted) optimum on this surface, and a decision
about new experiments to be performed. Only this last aspect distinguishes iterative
designs from the previously described fixed experimental designs.
New experiments may not be required if, for instance, the optimum is located at a
position in the parameter space where an experiment has already been performed. If this
is not the case, then the location of one or more additional experiments will be the result
of the calculation step. Subsequently, a new set of experiments is run and added to the
existing database. The model can then be refined using all the available data, and a new
optimum can be predicted.
The procedure may be stopped not only if experiments have already been run at the
suggested locations, but also if the predicted optimum is the same as it was before, or if
it can be established in the calculation step that no further improvement may be expected
from an additional iteration cycle.
The philosophy of iterative designs is to locate the true (global) optimum using a
minimum number of experiments and making maximum use of available insight and
experimental data. Such a philosophy can be justified if
1. the required number of experiments is indeed less than it is using other optimization
procedures,
2. the time and effort needed to analyse the data (calculation step) is small compared to
the task of performing a new set of experiments, and
3. The global optimum is found.
Phase selection diagrams
The method of phase selection diagrams was developed by Schoenmakers et al. [504] for
the optimization of the composition of ternary mobile phases in RPLC. The starting point
of an iterative design may be the same as for a window diagram. We will consider the
optimization of the composition of a ternary mobile phase in RPLC. A very simple
example involving six aromatic solutes is shown in figures 5.29,5.30 and 5.3 1. The ternary
mixture is prepared by mixing two iso-etuotropic binary mixtures (see the discussion on
the Sentinel method in the previous section).
In the present example these mixtures contain 50°/o methanol and 32% THF in water,
respectively. The two chromatograms obtained with the binary mobile phases are shown
in figure 5.29. From the capacity factors observed in these chromatograms, the phase
selection diagram of figure 5.30 can be constructed. On the horizontal axis in figure 5.30
is the mixing ratio between the two limiting binary mixtures. The logarithm of the capacity
factor is plotted on the vertical axis in this figure, and the (dashed) straight lines connect
the two capacity factors observed for each solute. Using this linear interpolation for the
retention lines, the response surface may then be calculated. In figure 5.30 the response
line is drawn using the product resolution criterion.
It can be seen from figure 5.30 that co-elution of three solutes, and hence a product
. resolution of zero, is predicted at compositions of around 35'/0 methanol and 10°/o THF
22 1
1.2!
I 50% MeOH 50%H20

2 Phenol
3 3-Phenyipropanol
L 2.4-Dimethylphenol
5 Benzene
I L 5
0
1
10 20 - tlmin
32% THF 68% H20

5 L
0
I I I
10
I I I
20
I
- I
tlmin
1
Figure 5.29 : Initial chromatograms for the construction of the phase selection diagram shown in
figure 5.30. Figure taken from ref. [504]. Reprinted with permission.
(55% water). Optimum selectivity for all solutes is predicted to occur at a composition of
10% methanol and 25% THF (65% water). The two chromatograms that can be obtained
at these compositions are shown in figure 5.31, and it can be seen that the phase selection
diagram method is very useful in this case.
Figure 5.30 is a so-called phase selection diagram. It is essentially the same as a window
diagram. However, figure 5.30 is the simplest phase selection diagram, which is the starting
point for an iterative procedure, while a window diagram is the final stage of an
optimization procedure using a fixed experimental design. The above example is a very
favourable one, because a phase selection diagram as described above does not usually
give a correct prediction of the optimum composition.
The reason for this is that the linear relationship between In k and composition (mixing
ratio of two iso-eluotropic binary mixtures) is not rigorously valid. A careful examination
showsthat the observed lines for In kvs. composition are slightly and systematically curved
[576,577].
222
0 '4'-THF - 0.32
Figure 5.30 : Phase selection diagram constructed from the chromatograms shown in figure 5.29.
Dashed lines are retention surfaces, the drawn line is the responsesurface. Figure taken from ref. [504].
One way to solve this problem is to choose conditions such that accurate linear
interpolation is possible for all solutes. This approach has been followed by Colin et al.
(ref. (5551;see section 5.5.1), who suggested adapting the value for the hold-up time (to)
as a function of composition. It is questionable whether such an approach can be applied
successfully to a large number of solutes. Moreover, the procedure used to estimate the
appropriate to values may be quite time-consuming.
Alternatively, an iterative method may be applied. The phase selection diagram may be
used to predict the optimum composition. The chromatogram obtained at this composi-
tion may then be compared with the predicted values for the capacity factors. If the
experimental optimum corresponds to the predicted chromatogram (in terms of response
and capacity factors), then apparently the linear interpolation was justified and the result
is a reliable global optimum. If the resulting chromatogram differs from the predicted one,
then the newly obtained set of capacity factors can be used to refine the phase selection
diagram and to predict a new optimum composition. This iterative procedure can be
repeated until the predicted capacity factors no longer differ from the experimental ones.
Such an iterative procedure has been worked out in detail by Drouen et al. [576].
Refinements of the method using the phase selection diagram discussed above include the
use of normalized resolution products (see section 4.3.2), shifted compositions and
confidence ranges.
223
I
0
1
10 20
- tlmin
,
lQ%MeOH 25%THF 65%H20
0 10 20 - t/min
Figure 5.31 :Chromatogramsrun at compositions predicted by the phase selection diagram af figure
5.30(a)to yield oo-elution of three peaks and (b) to yield optimum separation conditions. Figure taken
The use of shifted compositions encourages a good distribution of the experimental data
over the parameter space. The optimization procedure directs the search to a certain area
in the parameter space (around the predicted optimum), but the use of shifted composi-
tions ensures that the maximum amount of new information is obtained from each next
data point. The shift in composition (for a one-parameter optimization problem) can be
described by
X’ =x + 2A { 0.5 - (x-x~)/(x~-x~)} (5.18)
where x‘ is the shifted composition, x the predicted optimum composition, A is a constant,

and x1and x2 are the locations of the closest data point previously measured, below and
above x, respectively. A typical value for A is 0.2 [576], so that
X’ =x + 0.4 (0.5 - (x-x1)/(x2-x1)) . (5.18a)
224
The effects of the shift in composition prescribed by eqn. (5.18a) can be illustrated by
considering the initial phase selectiondiagram, in which two experimentat chromatograms
are incorporated, i.e. at x, = O and at x2= I. Hence, eqn.(5.18a) becomes
X’ =x + 0.4 (0.5 - X) = 0 . 6 ~+ 0.2 . (5.18b)
Eqn.(S.l8b) shows that if the predicted optimum is located at one of the two limiting
binary mixtures, the experimental verification will be performed at a composition 0.2
x-units away, i.e. at a mixing ratio of 4 1 ( A : B )if x = 0 (x’ = 0.2), or at a ratio of 1:4 ( A : @
if x = 1 (x’ =0.8). The further the predicted optimum composition is removed from the
existing data points, the less the shift in composition will amount to. If x = 0.5, no shift in
composition will occur.
Confidence ranges may be defined around each experimental data point by the
following equation:
d = Ax/2 - 1/2 v(A2-46/IAI) (5.19)
where d is the confidence range, Ax the distance between two available data points, 6 the
allowed uncertainty in In k and IAI the (absolute) curvature of a quadratic equation
describing In kas a function of x in the area around the data point. Eqn.(5.19) can be used
if linear interpolation between successive data points is used as a model for the variation
of retention with cornposition. It describes the difference between a linear interpolation
and a quadratic one. A value of 0.025 has been suggested for 6 [576] and in the example
of optimization of a ternary mobile phase composition in RPLC, A is usually smaller
than 1. If we take IAl= 1 and consider the initial situation, then the size of the confidence
intervals that extend above x = 0 and below x = 1 is
d = 1/2 - 112 V0.9 w 0.03.
Hence, in the initial situation the confidence ranges stretch from 0 to about 0.03 and from
about 0.97 to 1.
However, when more data points become available, the size of the confidence ranges
quickly increases. From eqn.(5.19) we see that the confidence interval will equal Ax/2
when the square root equals zero, i.e. when
Ax2 = 46/IAI. (5.20)
Using the same estimates for Sand IAI as before, we find that Axz0.32. Hence, when IAI
equals 1, a total of four data points (x = 0,0.33,0.67 and 1) is suficient to describe the
capacity factor within 2.5% (an error in In k of 6= 0.025 corresponds to an error of about
2.5% in k). When more than two data points are available, a better estimate for A may of
course be obtained from the data. For instance, when the verification of the first predicted
optimum yields exactly the same capacity factors as were predicted, then apparently all
A values are equal to zero and the confidence intervals extend over the entire parameter
space.
225
A considerable advantage of the iterative procedure is that the accuracy of the predicted
retention times is increased at each stage of the procedure. This is not true when a fixed
experimental design is used. For example, d'Agostino et al. [579] obtained a precision of
6% between the predicted and experimental retention times using a fixed experimental
design (corresponding to figure 5.27b). The figure of 6% gives an indication of the
reliability of the result in terms of the predicted optimum conditions. It may be compared
with the present iterative procedure if the value of 6 in eqn.(5.19) is set equal to 0.06.
Confidence intervals may be used to define a stop criterion, i.e. they can be used to judge
whether the optimization process should be continued or halted. If the predicted optimum
falls within one of the existing confidence intervals (calculated for 6= 0.025), then the
experimental capacity factors will be within 2.5% of the predicted values. It should be
noted that an error of 2.5% in k can make a big difference if the relative retention (a)of
a pair of peaks is close to one. It may therefore be required to use a lower value for 6 in
eqn.(5.19).
The full procedure is illustrated in figures 5.32, 5.33 and 5.34 for the separation of a
mixture of five diphenylamines by RPLC. Figure 5.32 contains the three phase selection
diagrams that can be constructed if the initial experiments involve three iso-eluotropic
binary mixtures*. The three initial chromatograms needed to construct this figure are
shown in figure 5.33 (chromatograms a, b and c). The methanol-water (65135) binary
mixture appears at the far left and at the far right of the picture. The two other binary
mixtures (THF-water, 40160 and acetonitrile-water,50150)occur once, on the two vertical
axes in the centre of the figure. The top half of the figure represents the (interpolated) linear
retention lines for the five solutes. The bottom half represents the response surface, using
the normalized resolution product (eqn.4.19) as the criterion.
0 THF- LO -THF 0 MeOH-
Figure 5.32 : Initial phase selection diagrams for three possible ternary mobile phase systems applied
to the separation of five diphenyl amines. Top: (Initial) retention lines. Bottom:(initial) response line.
Criterion:normalized resolution product ( ceqn.4.19;drawn line) Also shown is the response surface
using the product resolution criterion (IIR; eqn.4.18;dashed line). The required chromatogramsare
shown in figure 5.33 (a, b and c). Figure taken from ref. [576]. Reprinted with permission.
* Methods usedtotimatethe correcteluotropicstrength (methanol-waterratio) have been described

in section 5.4. Methods used to calculate corresponding compositions of other (iso-eluotropic)binary
mixtures were discussed in section 3.2.
226
1
26 L THF
17 ACN
r R s 190
@ r 072
LO THF 31 ACN
nRs 149
O 50 ACN N
19.5 MeOH
nRs 15.2
@ , :2: :
I
”;
2L1 MeOH
25 2 THF
nRs 212
r 025 fi
1 I I I I I I
0 200 100 6 o O t l s 80rJ 0 200 LOO 600_tls 800
Figure 5.33 :Chromatograms obtained during the optimization of the composition of a ternary mobile
phase for RPLC for the separation of five substituted diphenyl amines (DPAs). Solutes: (1)
N-nitroso-DPA, (2) 4-nitro-DPA, (3) Z,rl‘-dinitro-DPA,(4) DPA and (5) 2-nitro-DPA. Stationary
phase: Hypersil ODs. Figure taken from ref. [576]. Reprinted with permission.
In figure 5.32 the optimum composition predicted from a combination of all three phase
selection diagrams is a mixture that contains 10.4%methanol and 33.6% THF (x= 0.84).
Eqn.(5.18b) then prescribes a shifted composition of 19.5% methanol and 28% THF
(x’ = 0.7). Obviously, this composition does not fall within one of the (small) confidence
regions, and therefore an experimental chromatogram is recorded at the shifted composi-
tion. This chromatogram is shown in figure 5.33 (chromatogram 6).
Although theseparation is better than in any of the three binary mixtures, it is not nearly
as good as we expected from figure 5.32. The obvious reason for this is curvature of the
retention lines. Therefore, the new data are entered in the phase selection diagram and the
iterative procedure is started. The optimum composition is now predicted to be in a
completely different part of figure 5.32, namely at 30.6% THF and 11.7% acetonitrile
(x= 0.25). According to eqn.(5.18b), this composition is then shifted to 26.4% THF and
17% acetonitrile (x’ = 0.34). The resulting chromatogram is shown as chromatogram e in
figure 5.33. Clearly, the separation is now much improved. In fact, it is better than expected
from figure 5.32, again due to curvature of the retention lines.
227
From here, the iterative procedure takes two more steps, shown as chromatogramsfand
g in figure 5.33. The composition of chromatogram g falls within the confidence range
around the composition used to record chromatogram e. Hence, the procedure has advised
us to stop before recording chromatogram g, but this final chromatogram has been run
to verify the optimum. Chromatogram g is the final result of the procedure, and it does
indeed yield a satisfactory distribution of the peaks over the chromatogram.
The actual retention and response lines, constructed using the data obtained from the
chromatograms in figure 5.33 and from some additional experiments, are shown in figure
5.34.
-30
d
t
c-. !2
-0
0 ACN- 50 -ACN o
0 THF- LO -THF 0 MeOH-
Figure 5.34 :Final phase selection diagrams for the ternary optimization problem illustrated in figures
5.32 and 5.33. Top: retention lines approximated by linear interpolation. Bottom: response lines;
dashed line: resolution product (lTR, eqn.4.18), drawn line: normalized resolution product (r;
eqn.4.19). Figure taken from ref. [576]. Reprinted with permission.
The non-linearity of the retention lines is apparent from this figure. The response lines
have been drawn for two different criteria: the normalized resolution product r (drawn
line; eqn.4.19) and the product resolution function IIR, (dashed line; eqn.4.18). It is seen
that the product resolution criterion would in fact have guided us to a completely different
optimum at a composition of 24.1% methanol and 25.2% THF. The chromatogram that
we would have obtained at this composition is shown in figure 5.33h. Clearly, this
chromatogram is less attractive than the one of figure 5.338. Obviously, the normalized
resolution product is to be preferred to the resolution product itself (see the discussion in
section 4.3.2).
Figure 5.34 also carries a warning. As we saw with chromatogram d (the response of
which was lower in practice than was expected from figure 5.32) and with chromatogram
e (yielding a higher response), the initial predictions of a phase selection diagram should
be approached with some care. The same conclusion can be drawn if we compare the phase
selection diagram of figure 5.32 with the final diagram (figure 5.34). It is seen that the two
figures are markedly different. This is not only true in the two ternary systems which were
considered during the optimization procedure (methanol-THF-water and THF-acetoni-
228
trile-water), but also for the third system (acetonitrile-methanol-water). It is seen from
figure 5.34 that the actual response line in this figure is even worse than was predicted in
figure 5.32. Hence, in retrospect it was quite correct to neglect this system entirely during
the optimization process. However, as in the THF-acetonitrile-water system, the reverse
might also have been the case. If the actual response line in figure 5.34 had been much
higher, instead of much lower than the one predicted from figure 5.32, then it is in principle
possible that a mixture in this phase system would have yielded a higher response than the
optimum THF-acetonitrile-water composition. In other words, this system might have
encompassed the global optimum.
The warning contained in this example is that highly non-linear retention lines may give
rise to global optima that remain unrevealed during the course of an iterative optimization
process. Hence, unlike the situation in which the curvature coefficient A is equal to zero,
a window diagram or a phase selection diagram offers no guarantee that the global
optimum can indeed be located. Usually, however, it it quite simple to verify the result of
an interative optimization procedure by performing one additional experiment.
Billiet et al. [578] have illustrated the importance of the curvature for the optimization
procedure of Drouen et al.. They optimized the pairing-ion concentration for the
separation of a synthetic sample containing both anions and cations. By plotting the
logarithm of the capacity factor against the logarithm of the concentration of the
ion-pairing reagent fairly smooth curves were obtained and the optimization could be
completed within a few chromatograms. However, if the reagent concentration is not
logarithmically transformed, the curves are extremely non-linear (especially in the low
concentration region) and the procedure fails.
The chances that the global optimum will not be found increase when
1. the curvature of the retention lines increases,
2. large areas of the parameter space remain unsearched, and
3. smaller differences exist between the responses at the different (predicted) optima.
Therefore, it should be recommended that one or more additional chromatograms are
recorded after the completion of the optimization process if
1. a local optimum, only slightly inferior to the global one, is predicted to occur, and
2. large areas of the parameter space remain unsearched, in which severe curvature of the
retention lines cannot be excluded.
There appears to be more reason to record extra chromatograms if the result of the
optimization process is not satisfactory. For instance, if the chromatogram of figure 5.338
has been obtained and if it can safely be assumed that there are no more than five solutes
present in the sample, then there is no reason to record an additional chromatogram in
the middle (x= 0.5) of the large unsearched area corresponding to the acetonitrile-metha-
nol-water system.
Linear interpolation vs. model equations
In figure 5.34 the retention lines have been approximated by a series of linear line
segments, rather than by a smooth curve. The alternative is to fit a mathematical equation
to the data, for example a quadratic function for In k vs. the mixing ratio x. If more than
229
three data points are available for one ternary system, then the coefficientsfor the equation
can be found from regression analysis. The same argument holds here as was used in the
discussion of the Sentinel method. If the largest source of deviation from a mathematical
model equation is experimental error, then the use of regression analysis may be beneficial.
If it is lack of fit between the model and the experiments, then it may be detrimental. In
the absence of experimental error, the linear line segments will give rise to interpolation
errors in between data points, but will be correct at those points where experimental data
are available. If a mathematical model equation is used, which does provide an exact
description of the retention behaviour, then the experimental errors are spread out over
the entire parameter space.
In summary, linear interpolation between successive data points should be preferred if
the experimental error in the data points is expected to be small relative to the error
involved in the description of the data with a mathematical equation.
A mathematical model equation is preferred if an equation is available that yields a
quantitatively accurate description of the data within experimental error.
It is clear from the above, that model equations for the description of retention surfaces
have to meet high demands. Preferably, equations should be used that relate to reliable
chromatographic theory, such as the one used to describe the retention behaviour as a
function of pH in RPLC in the window diagram approach described in section 5.5.1. The
use of such a chromatographic equation was clearly better in that case than a statistical
approach using (for example) polynomial equations.
An optimization procedure that involves the use of mathematical equations to model
the retention surfaces within an iterative design has been described by Lankmayr et al.
[580,581]. Using statistical or, preferably, chromatographic model equations, it is straight-
forward to extend an iterative design method to cover more than one parameter. In order
to describe the retention behaviour in RPLC using ternary and quaternary mixtures, a two
parameter quadratic equation may be used. In this case, there is hardly any difference
between a model based on chromatographic theory and a purely mathematical approach.
This becomes more obvious if other parameters are considered, such as the combined
optimization of mobile phase composition and temperature in RPLC, where eqn.(3.58) or
eqn.(3.59) may be used as a (chromatographic) model equation, or the simultaneous
optimization of methanol content, pH and ionic strength [562,563] described above.
Naturally, the number of initial experiments required to start the optimization
procedure will increase if either the number of parameters considered or the complexity
of the model equations increases.As far as the number of parameters is concerned, we have
seen this to be true with any optimization procedure, and hence the number of parameters
should be carefully selected. In order to avoid a large number of initial experiments, the
complexity of the model equations may be increased once more data become available
during the course of the procedure. For example, retention in RPLC may be assumed to
vary linearly with the mixing ratio of two iso-eluotropic binary mixtures at first. When
more experimental data points become available, the model may be expanded to include
quadratic terms. However, complex mathematical equations, which bear no relation to
chromatographic theory (e.g. higher order polynomials [537,579]) are dangerous, because
they may describe a retention surface that is much more complicated than it actually is
in practice. In other words, the complexity of the model may be dictated by experimental
230
error rather than by the underlying retention mechanism. In that case all experimental data
points will be accurately described by the model, but the interpolation between them may
not be correct.
Extension to multi-dimensional optimization problems
Lankmayr and Wegscheider[580,581]have developed a flexible iterative design method

which allows the use of a variable number of parameters. The retention surfaces are
approximated by means of a modified moving least squares algorithm. The procedure is
started with the parameters and their step sizes (e.g. one percent steps in composition) as
selected by the user. During the optimization process the number of parameters may be
decreased by deleting those that appear to be irrelevant, or increased by adding new
parameters. Also, the step sizes can be changed during the optimization process.
Unfortunately, this method has so far not been published in the literature.
The linear segmentation method described by Drouen et al. may also be expanded to
include two-parameter optimization problems. They described the application of an
iterative design method to the optimization of the composition of quaternary mobile phase
mixtures in RPLC [502]. However, the division of a two-dimensional parameter space (in
this case a triangle, similar to the one shown in section 5.5.1) into segments, and the
approximation of the retention surfaces with a series of triangles, is not as straightforward
as the use of a series of linear line segments in a one-parameter optimization problem. In
order to avoid the occurrence of a series of awkward (i.e. long and narrow) triangles,
Drouen et al. established a series of complicated “shift rules” that shift the experimental
location from the predicted optimum towards the edge, along the edge or towards the
centre of a triangular segment in the parameter space [502].
The problem encountered by Drouen et al. is that the procedure has become so
complicated that it requires a large computational effort. Half an hour of computer time
is reported to be needed for each computation step [502]. Another half an hour may be
needed to plot the available information in the form of diagrams. However, this is not
strictly required at intermediate stages of the optimization procedure. If the computer time
becomes excessive, the underlying philosophy of the iterative design, as discussed at the
beginning of this section, comes under pressure. It may then no longer be argued that the
time required for computation is much less than the time required to record another
chromatogram. In that case, several other approaches should be considered:
1. a fixed experimental design might be used, such as the one used in the Sentinel method,
to make the first predicted optimum more reliable;
2. two-parameter optimization procedures should not be used until the possibilities with
single parameter optimization have been fully exploited;
3. the computational procedure should be reconsidered in order to meet the philosophical
requirements of the method.
The first approach may always be followed. A drawback is that it might imply that a
larger number of experiments will be required, some of which may turn out to be of little
value in the end. The required calculations may sill be lengthy, but they only need to be
performed once during the entire optimization procedure. A major advantage is that the
chances of overlooking the global optimum are greatly reduced.
23 1
The second approach is the one followed by Drouen et al. [502]. It is based on the
experience that only in very few cases does the optimization of a quaternary mobile phase
composition in RPLC yield an optimum that is truly quaternary, i.e. contains all four
solvents. Hence, the procedure discussed before for ternary solvents usually leads to the
global optimum. This argument, correct though it may be, only applies to the particular
problem of mobile phase optimization in RPLC,and prohibits the application of the same
method to other two-parameter optimization problems [582].
The third solution to the problem may be found in the use of more efficient computers,
algorithms and computational methods. For instance, if segmentation of the parameter
space (linear interpolation) is used, large parts of the retention surfaces and hence of the
response surface may remain unaltered when a new data point is added to the existing set.
The use of simple model equations instead of linear segmentation may also be more
efficient from a computational point of view. However, such simple equations may only
be used for the description of the retention behaviour in a limited number of cases and
if the model equations become more complex the advantage quickly disappears. For
example, d’Agostino et al. used up to sixth order polynomial equations [537]and their
procedure also led to excessive calculation times.
Another possibility to reduce the computation time for the location of the optimum is
the use of the Simplex algorithm. In section 5.3 we discussed the severe limitations of the
Simplex method as a stepwiseapproach towards the chromatographic optimum. The main
problem associated with the method turned out to be that many experiments were required
to locate an optimum and that this might be a local one, so that the procedure has to be
started repeatedly from different starting points. Hence, the procedure might have to be
run for say ten times with an average number of say 40 data points, thus requiring 400
chromatograms to be recorded.
However, if the Simplex algorithm is only used in the computation step to locate the
optimum in the response surface with all the retention surfaces being known, then 400
chromatograms may be calculated rather than measured. Although the average number
of steps required for each application of the algorithm turned out to be closer to 100 than
to 40 in practice, Svoboda I5171demonstrated that the Simplex apprach may still compare
favourably with a grid search approach during the calculation step, especially for
multi-parameter optimization problems.
In the long run, it seems likely that improvement in instrumentation will speed up the
computation at a faster rate than the chromatography, so that the present discussion will
lose more and more of its relevance. Indeed, it has already turned out that the use of more
efficient computational procedures may succeed in a very considerable reduction of the
calculation time required in the procedure of Drouen et al. [593].
We may summarize the characteristics of iterative design methods as follows:

1. A minimum number of experiments is required.
2. A good idea of the response surface is obtained, especially in the area around the
optimum.
3. A disadvantage of this last aspect is that large areas may remain unsearched. In some
cases, this could imply that the global optimum is overlooked.
4. Linear interpolation between individual data points should be preferred ifthe experimen-
232
tal errors in the data points are small and no reliable equation to describe the data
mathematically is available.
5. The use of model equations should be preferred if the data can be described within
experimental error over the entire parameter space.
6. When several parameters need to be optimized simultaneously, the use of simple model
equations (ifpossible)seems to have advantages over linear interpolation methods. If the
required equations become more complicated, however, this advantage is rapidly losr.
7. Model equations based on chromatographic theory should be preferred to strictly
mathematical ones.
8. The individual capacity factors of all solutes are required to calculate the retention
surfaces.
5.5.3 Summary
Table 5.6 gives a summary of the interpretive methods described in sections 5.5.1 and
5.5.2. The general characteristics of all interpretive methods are the following:
I . The capacityfactors of all the individual solutes need to be obtained at each experimental
location.
2. The retention surfaces must be approximated by some kind of model.
3. A generally small number of experiments is required.
4. A good overall impression of the response surface can be obtained.
5. For single-parameter optimization both graphical and mathematical methods may be
used.
6. In principle, all methods can be adapted to include multi-parameter optimization, but
graphical methods are then no longer possible.
7. The number of required experiments and the computation time will increase when the
number of parameters increases.
In section 5.7 these characteristics will be compared with those of the other optimization
procedures described in this chapter.
5.6 PEAK ASSIGNMENT AND RECOGNITION
By definition, all interpretive methods of optimization require knowledge of the

capacity factors of all individual solutes. This is the fundamental difference between the
simultaneous and sequential methods of optimization (sections 5.2 and 5.3, respectively)
and the interpretive methods of section 5.5. Moreover, in the specific cases in which only
a limited number of components is of interest o r in which weighting factors are assigned
to the individual solutes (see section 4.6.1)* it is also necessary to recognize the individual
peaks (at least the relevant ones) in each chromatogram. In section 5.5 we have tacitly
assumed that it would be possible to obtain the retention data (capacity factors) of all the
individual solutes at each experimental location.
The problem of recognizing the individual solutes in the chromatograms during the
* An exception to this is when weighting factors are only used to deal with a solvent peak in the
chromatogram (see section 4.6.3).
233
optimization process is complicated because of the large amount of overlap that may be
expected to occur between the various peaks in the chromatograms. If this overlap would
not occur, then there is not much need for optimization procedures. In other words,
Table 5.6
Selection of interpretive methods applied for selectivity optimization in chromatography.
Ref. Author(s) No. No. Design Method (1)

par. exp. (1)
501 Laub/Purnell 1 2 f window diagram

549 Constanzo 1 2 f window diagram
550 Deming/Turoff 1 9 f window diagram
552 Price/Deming 1 4 f window diagram
546 Jones/ Wellington 1 6 f window diagram
547 Noyes 1 3/4 f window diagram
554 Issaq et al. I 5 f window diagram
553 Hsu et al. 1 4 f window diagram
555 Colin et al. 1 2 f critical band

557 Toon/Rowland 1 2 f critical band
558 Sachok et al. 2 9 f full factorial

559 Sachok et al. 2 9 f full factorial
560 Weyland et al. 2 9 f full factorial
583 Gant et al. 2 4 f full factorial
542 Glajch et al. 2 7 f Simplex lattice

572 Antle 2 7 f Simplex lattice
512 Weyland et al. 2 7 f Simplex lattice
566 Issaq et al. 2 10 f extended lattice
562 Otto/ Wegscheider 3 36+6 f full factorial

563 Otto/ Wegscheider 3 9+6 f limited factorial
537 d’Agostino et al. 3 12 f modified lattice
517 Svoboda var var f quadratic design
504 Schoenmakers et al. 1 3-10 1 iterative design

576 Drouen et al. 1 3-8 1 iterative design
578 Billiet et al. 1 3-8 1 iterative design
582 Drouen et al. 2 4-10 1 iterative design
502 Haddad et al. 2 4-10 1 iterative design
580 Lankmayr et al. var var 1 sequential global opt.
(1) f = fixed design (simultaneous interpretive method); i = iterative design
234
optimization procedures are most useful when the task of recognizing the individual peaks
is most complicated.
In this section different ways to measure the required data will be discussed.
Table 5.6 (Continued)
Ref. Model Application

(compositions refer to mobile
phase unless stated otherwise)
501 empirical (linear) stationary phase composition GLC

549 empirical (linear) mixed pairing ions LC
550 chrom.mode1 (eqn.3.70) pH opt. RPLC
552 chrom.model (eqn.3.70) pH opt. RPLC
546 linear interpol. pH opt. RPLC
547 linear / curved temperature / bin.comp. RPLC
554 empirical (4th order) ternary comp. RPLC
553 chrom.mode1 binary comp. RPLC
555 linear (to adapted) ternary comp. RPLC

557 empirical (linear) binary comp. RPLC
558 semi-empirical bin.comp. and pairing ion RPLC

559 semi-empirical binary comp. and pH RPLC
560 semi-empir.(2nd order) quaternary comp. RPLC
583 chrom.model (eqn.3.58) temp. and binary comp. RPLC
542 empirical (2nd order) quaternary comp. RPLC and LSC

572 visual comparison quaternary comp. LSC
512 semi-empir.(2nd order) ternary comp. RPLC
566 empirical quaternary comp. RPLC
562 semi empirical 3 param. RPLC ionic solutes

563 semi empirical 3 param. RPLC ionic solutes
537 empirical (to 6th order) quaternary comp. RPLC/LSC
517 empirical (2nd order) pH/comp./pairing reagent RPLC
504 linear interpolation ternary comp. RPLC

576 linear interpolation ternary comp. RPLC
578 linear interpolation pairing-ion concentration
502 linear interpolation quaternary comp. RPLC
582 linear interpolation binary comp. and pH RPLC
580 moving least squares various
235
peak area (counts)
0 , 1 , , 2 , 3 , 4 , 5 K 6 , 1 , 8 , 9 , ~ ,
k
-
Figure 5.35: Three simulated chromatograms in which peaks have been assigned the numbers 1 to 5
(correspondingto the elution order in chromatogram a) on the basis of the peak areas. The areas are
shown in the chromatograms. For chromatogram (c) see opposite page.
5.6.1 Single channel detection
The most obvious solution to the problem described above is the injection of all solute
components separately. Clearly, this allows an accurate determination of the capacity
factors, provided that tbe chromatographic conditions (e.g. flow rate, temperature, mobile
phase composition) are adequately controlled. However, there are two clear disadvantages
of this method:
1. injection of each solute separately greatly increases the number of experiments to be
performed, and
236
I I I I 1 8 1 I I , l a I , I I I , I , I .
0 1 2 3 L 5 6 7 8 9 10
-k
2. in many cases, not all the components in the sample are known, or they are not available
in pure form (or solution).
The first disadvantage implies that lengthier optimization procedures are required.
Hence, one of the most attractive characteristics of interpretive methods, the small number
of experiments required, is sacrificed. However, in fully automated systems little effort is
required from the operator.
In contrast, the second objection is fundamental. In most cases it will be impossible to
inject all sample components separately and hence an alternative method will have to be
found if interpretive methods are to be used.
The problem is illustrated in figure 5.35. In this figure three simulated chromatograms
are shown, which we will presume have been recorded under different conditions. A
chromatogram contains information in two directions. So far, we have almost exclusively
made use of the information in one direction: the retention times (capacity factors) of the
solutes. The information in the other direction, peak height or peak area, may be used to
assist in the assignment of peaks. In figure 5.35a a chromatogram is shown that contains
five peaks. The peaks have been numbered 1 to 5 (circled numbers) and the area of each
peak is indicated in the figure. Figure 5.35b shows another chromatogram, also containing
five peaks. Apparently, however, the elution order has changed in going from chromato-
gram a to chromatogram b. It is seen in the figure that the peaks may be assigned the
numbers 1 to 5 in a different order on the basis of the peak areas.
In figure 5 . 3 5 ~the problem has become more difficult, because now only three peaks
occur in the chromatogram instead of five. On the basis of the areas we may conclude that
the first peak consists of solutes 2 and 3, the small second peak of solute 1 and the third
peak of solutes 4 and 5.
For several reasons figure 5.35 represents a favourable example. In the first place, all
solutes show markedly different areas. The difference in area between each two solutes is
at least 20°/0. In the second place, we have assumed that there are no more than five solutes
present in chromatogram a, an assumption which has not been proved wrong (but neither
237
has it been proved right) in the following two chromatograms. It would clearly have been
wrong’if the chromatograms had been obtained in the reverse order and if we had
concluded from figure 5 . 3 5 ~that only three solutes were present in the sample. The
problem would have been more complicated if none of the chromatograms had shown as
many peaks as there are solutes present in the sample.
There are other major problems with peak assignment on the basis of the areas. These
problems relate to the reproducibility of peak area measurements under widely varying
conditions. Ideally, the area of a peak remains constant even if its capacity factor varies.
However, varying the conditions may affect the peak areas. If the column temperature is
changed in GC, then the flow rate may be affected. Peak areas will change (by a constant
factor) if concentration-sensitive detectors such as the hot wire detector (HWD; katharo-
meter) are used, but not with mass flow sensitive detectors (such as the flame ionization
detector, FID).
When the mobile phase composition is changed in LC, the sensitivity of the detector to
different solutes may be altered. For example, the UV spectra of solutes may shift upon
a change in the composition of the mobile phase [584]. Especially if the detection
wavelength is on the flank of an absorption band, this may easily lead to variations in the
peak area that exceed the 20% difference which we used to discriminate between the
different solutes in figure 5.35.
An additional problem is the measurement of peak areas itself. The integration will give
rise to errors, especially if the peaks are not completely resolved and if the baseline varies
during the analysis.
Hence, we conclude that peak area measurementsmay be of some help in the assignment
of peaks, but there are a series of major limitations, so that only in some very favourable
(simple) cases will a satisfactory result be obtained. Of course, in such simple cases there
may not be much reason to optimize the separation.
Issaq and McNitt [585]published a computer program for peak recognition on the basis
of peak areas. They investigated the reproducibility of the area of some well-separated
peaks for three solutes (anthraquinone, methyl anthraquinone and ethyl anthraquinone)
in the 10 solvents used for their optimization procedure. The solvents inctuded binary,
ternary and quaternary mixtures of water with methanol, acetonitrile and THF. The areas
were found to be reproducible within about 2 percent. The wavelength used for the UV
detector in this study was not reported.
One final comment may be made regarding figure 5.35. This figure forms a good
illustration of the problem of peak assignment as it occurs during optimization procedures.
We tried to assign numbers to peaks, once arbitrary numbers had been assigned in one
chromatogram (figure 5.35a). During this process, none of the peaks was identified. For
the general problem of selectivity optimization, this is quite sufficient. Only in specific
cases, where a limited number of sample components is of interest, are we required to
recognize (but not to identify) the relevant peaks in the chromatogram. To do this,
standard solutions containing one (or more) of the components of interest are required in
the case of single channel detection. If multichannel detectors are used (section 5.6.3), the
components of interest may also be recognized on the basis of information (e.g. spectra)
obtained in independent experiments.
238
5.6.2 Dualchannel detection
The assignment or recognition of peaks in different chromatograms may be aided if

extra qualitative information about the solutes is obtained from the detector. The simplest
way to obtain more information is to combine two detectors in series. Many combinations
are possible, but some limitations arise (see for example ref. [586]):
The time delay between the two detectors should be minimal.
The first detector should be non-destructive and should not contribute significantly to
the band-broadening.
The two detectors should have similar types of sensitivities (e.g. two concentration
sensitive detectors or two mass flow sensitive detectors).
Ideally, the degrees of sensitivity of the two detectors should also be comparable.
Because of this last reason, it is not attractive to combine a hot wire detector with a flame
ionization detector in GC (a combination that also conflicts with the third limitation
above) or a differential refractometer with a UV spectrometer in LC.
For all the above reasons, it is to be preferred to measure two solute properties in one
detector, especially if both measurements can be performed simultaneously. An example
of this is the application of dual-wavelength absorption detection in LC. The application
of this technique for the purpose of selectivity optimization has been investigated by
Drouen et af.[584]. For the purpose of peak assignment or recognition, ratio recording may
be used. The principle of this technique is based on Beer’s law and may be explained from
the following equation for the absorption ratio R,:
(5.21)
In this equation, A , is the absorbance at one wavelength (A,) and A , the absorbance at
the second wavelength (A,). a, and a2 are the respective molar absorptivities (extinction
coefficients) at the two wavelengths, b is the optical path length and cis the concentration
of the solute in the detector cell. It can be seen from eq~(5.21)that the absorbance ratio
R , is independent of the concentration and hence that R , is a constant throughout the
elution of a single peak. The value of R, is determined by the ratio of the two absorptivities
at the different wavelengths and therefore it is characteristic of the solute. In practice, it
is necessary to introduce a threshold for both absorbances (because R, is not defined on
the baseline), whereas it is attractive to use slightly different definitions for the absorbance
ratios, for example [584]:
R A T = A , / A , for A , > A , > A (5.22)
R A T = 2- A , / A , for A, > A , > A (5.23)
R A T = 0 for A , < A n A,<A (5.24)
R A T = -0.1 for A , < A fl A , > A (5.25)
R A T = -0.1 for A , > A n A , < A (5.26)
239
In these equations RAT is the modified absorbance ratio and A is the threshold
absorbance. In fact, it would be beneficial to differentiate between the last two cases. This
may easily be realized by defining RAT= - 0.2 instead of - 0.1 in eqn.(5.26), i.e.
RAT = -0.2 for A , > A n A,<A (5.26a)
An example where the above definitions are used is shown in figure 5.36. In this figure,
three chromatograms are shown, each of them recorded at two different wavelengths
(signals A, and A*). The corresponding “ratiograms” are shown above the chromato-
grams. Figure 5.36a shows the typical block shaped signals that are obtained for pure
peaks. Each of the six peaks yields a specific vaIue for RATand, fortunately, the values
are all different. This allows a rapid assignment of all peaks in figure 5.36b, where again
six peaks are observed in the chromatogram. The peak sequence is indicated by the letters
a to f i n the chromatograms.
The situation is much more complicated in figure 5.36c, in which only four peaks are
observed. On the basis of the RAT values and peak areas, peaks a and d can readily be
assigned in this chromatogram. However, the second peak in figure 5.36~consists of two
partially overlapping peaks (indicated by the deformed shape of the block) and the fourth
peak contains two completely overlapping solutes (indicated by the perfect block shape,
in combination with a not previously observed RAT value). In this particular nasty (but
not necessarily uncommon) example, the problem is almost impossibleto solve using RAT
Figure 5.36: Three chromatograms recorded at two different wavelengths, accompanied by the
ratiograms (top of the figure). For explanation see text. Figure taken from ref. [584]. Reprinted with
permission.
240
values alone. However, the ratiogram reveals the presence of two partially overlapping
peaks in figure 5.36c, which is not revealed in the chromatogram itself. On an only slightly
better column, the two solutes in the second peak in figure 5.36~might have been
sufficiently resolved to allow for a straightforward assignment of all the peaks.
Figure 5.36 illustrates the potential of ratio-recording (chromatogram b) as well as its
shortcomings (chromatogram c) as an aid in the assignment of chromatographic peaks for
the purpose of selectivity optimization. Ratio recording may help to assign peaks that are
well resolved, or at least sufficiently resolved to allow an estimate to be made of the
separate R A T values.
According to Drouen e t d [584], this implies that two Gaussian peaks should be more
than one standard deviation apart ( R , = 0.25). For such a pair of peaks only one peak is
observed in the chromatogram, since a valley is only observed if the R , value exceeds 0.59
(eqn.4.11). In order to make a reliable estimate of the peak area, a much larger resolution
(R, > 1, or preferably R , > 1.5) is required.
Hence, the use of dual-channel detection increases the possibilities for peak assignment
in optimization processes. However, complications are caused by variations in the baseline
[584] and the effect of solvent composition on the UV spectrum may now become even
more serious than in the case of single-channel detection. This may occur for instance if
one wavelength is selected on a rising flank in the UV spectrum, and the second one on
a descending flank. In any case, the selection of the two most suitable wavelengths is one
of the most critical factors. Unfortunately, the wavelength selection is usually quite
arbitrary.
5.6.3 Multichannel detection
The logical next step to consider is the application of multichannel detection. The
combination of more than two detectors in series is unattractive, because of cost
considerations and because of increasing band broadening effects. One possibility is the
combination of several different detection principles into one detector, which has recently
been demonstrated by Cant and Perrone [587], who described a three-channel LC detector
that allows simultaneous monitoring of UV absorbance, fluorescence and conductivity.
However, both of these last two detection principles are fairly specific (i.e. they will be
aplicable to only a limited number of solutes) and hence the true three-channel capability
may only be available in some rare cases. True multichannel detection is obtained by the
combination of the chromatographic separation with a spectroscopic identification
technique. The most successful application of such a “hyphenated method is the use of
a mass spectrometer in the GC-MS combination. The mass spectrometer yields (almost)
universal detection, very high sensitivity and a large amount of qualitative (spectral)
information.
Unfortunately, the LC-MS combination is less successful. In part, this may be due to
technological interfacing problems, but even if these are solved, LC-MS is unlikely to
provide the same degree of universality (large molecules will remain a problem), spectral
information and reproducibility as the GC-MS combination. For the moment, the
combination of LC with a multichannel UV absorption detector is a more realistic
proposition.
Both in GC-MS and in LC-UV true multichannel detection may be obtained, and
241
200 LOO
Wavelengthlnm
(b)
0 100 200 300

timels
Figure 5.37: An example of a spectro-chromatogram recorded with a multichannel UV absorbance

detector in LC. The sample contains a series of dipeptides. (a) (top): pseudo-isomeric three-dimensio-
nal plot; dimensions are time, wavelength and absorption. (b) (bottom): contour plot with constant
absorption lines. Figure taken from ref. 15881. Reprinted with permission.
three-dimensional so-called “spectro-chromatograms” may be recorded. An example of

such a three-dimensional figure is shown in figure 5.37, both as a pseudo-isomeric and as
a contour plot. As was the case for the representation of response surfaces (figure 5.2), the
latter are to be preferred if an objective interpretation of the data from the figure is
required.
The problem of peak recognition may be seen as a simplified version of the problem of
peak identification. The spectra found during the elution of a chromatogram can be
subjected to automated retrieval systems [589]. In this case the peak may be tentatively
identified by comparing its spectrum to a (potentially large) number of reference spectra,
242
contained in a “library”. The spectra which resemble the spectrum of the peak may be listed
in order of decreasing probability. Although many of such library search techniques work
well for GC-MS in particular, there are some limitations. In the first place the peaks are
identified on the basis of probabilities, which should by no means be mistaken for
certainties. Also, only those components which are represented in the library by a reference
spectrum are considered, which limits the search to a finite number of possibilities.
Moreover, such retrieval systems may be too large a tool for the purpose of peak
recognition only.
One possibility is to build a miniature library from the spectra observed during the
chromatograms in an optimization process, and then to assign peak numbers on the basis
of previously stored spectra. Such a library search method requires that representative
spectra of all solutes are obtained at some stage of the optimization process (i.e. ideally
each component should appear as a single, well-resolved peak in at least one of the
chromatograms [590]).Moreover, the method needs to be very flexible in the sense that sum
spectra may occur once two components start to overlap, but also new spectra may be
found when components start to be resolved. MS spectra often yield sufficient information
to allow peaks to be recognized on the basis of their spectra alone, given that only a very
small library is created during the optimization process. However, there are cases (e.g.
geometrical isomers) in which different solutes may show very similar MS spectra.
In general retrieval systems, additional information such as retention data may be
required for a conclusive peak recognition using UV spectra [590]. For obvious reasons,
however, retention data cannot be used as an aid for peak recognition for optimization
purposes. Related solutes, such as isomers and homologues, may show similar spectra,
which are hard to differentiate. In case of doubt, one may have to refer to peak areas for
additional information. Again (see section 4.6.1), this requires very good resolution of the
individual peaks.
Clearly, what is required for a reliable recognition of all the peaks during the
optimization procedure is information on the pure component spectra and the pure
component peaks (elution profiles). A method to obtain both the spectral and the
chromatographic data involves the application of a mathematical technique called
“principal component analysis” (PCA) [592]. This method is based on the additivity of
spectra according to Beer’s law. The absorption (A) at a time t and wavelength A is given
by
n
,qt,n) = I: a p ) b q t ) (5.27)
j= 1
where u,(A) is the molar absorptivity of componentj at the wavelength A,b the optical path
length and c,(t) the concentration of j at time t in the detector cell. n is the total number
of solutes present in the sample.
Eqn.(5.27) shows that the total absorption is the result of the contributions of a series
of factors which depend either exclusively on the wavelength (ajfactors = spectra) or on
the time (cj factors = elution profiles). The individual factors may be obtained with PCA,
but an unambiguous solution for the mathematical problem may only be obtained in a
small part of the chromatogram (“peak cluster”), in which three or fewer components
contribute to the absorption [592].
243
However, the spectra as well as the elution profiles may be very similar (i.e. closely
related, ill-resolved solutes). Figure 5.38 shows an example of a result of PCA applied to
a cluster of three peaks using LC-UV detection. The problem involves the separation of
three proteins with very similar UV spectra. The chromatogram obtained is shown in figure
D,
c
0
f
5:
0
LUJ 250 300 200 250 300 A/nm 3 0
tlmin
Figure 5.38: An example of the application of principal component analysis to obtain the individual
spectra and elution profiles of ill-resolved proteins. (a): Illustration of the chromatogram obtained
at a low wavelength (e.g. 200 nm); (b), (c) and (d): Spectra of the three pure components identified
in the peak cluster by PCA (drawn lines) and true pure component spectra (dashed lines); (e): pure
component elution profiles from PCA (drawn lines) and estimated pure component profiles (dashed
lines). Figure adapted from ref. [592]. Reprinted with permission.
244
5.38a. The three spectra of the individual solutes as obtained by PCA are shown in figures
5.38b to d. The pure component spectra are indicated in these figures by dashed lines.
Figure 5.38e shows the individual elution profiles obtained from PCA, as well as estimated
pure component elution profiles (obtained from separate injections; dashed lines). Clearly,
the potential of the PCA method to obtain both retention and spectral information for
ill-resolved peaks is impressive.
The use of multichannel analysis for the recognition and identification of individual
peaks in a chromatogram is a very rapidly developing area and it may be anticipated that
complex mathematical techniques (such as PCA) will soon become available as a standard
tool for the chromatographer.
A remaining problem for all spectrometric peak recognition methods is the reproducibi-
lity of the spectra recorded under different chromatographic conditions. For example, if
the differences between the UV spectra for a given solute induced by variations in the
mobile phase in RPLC are larger than the differences between the UV spectra of different
solutes recorded under identical conditions, then clearly the application of multichannel
UV detection, with or without the use of PCA techniques, will be of limited use.
5.7 SUMMARY
In the last section of this chapter we will summarize the different optimization
Table 5.7a:
Summary of optimization procedures
Method: Univariate optimization
Includes: -
Section of this book : 5.1.1

Path in Figure 5.4 : 2111
Optimum found : Local (1)

Accuracy of optimum : Low(1)
Number of experiments : Fairly high (1,2)
Criteria to be used : Any
Impression of response surface : Very limited
Knowledge required : Selection of parameters
Multi-dimensional optimization : Straightforward

Computational requirements : Low
Complete automation : Possible (2)
Solute recognition : Not required
(1) Repeated initiation required.

(2) Dependent on procedure used in one dimension.
245
procedures that have been described in the preceding sections. An overview of the methods
is given in table 5.7.
All different optimization methods discussed in this chapter have been classified in five
different parts of this table. The characteristics of the different classes of methods can be
compared using the tables.
Univariate optimization (table 5.7a) is not a good method for the optimization of
chromatographic selectivity. This will be clear from the table, since despite a fairly high
number of experiments only a local optimum will be located on the kinds of response
surfaces typically encountered in chromatography (see figure 5.3). Moreover, the local
optimum may be of little value, because no overall impression of the response surface is
obtained during the process. Once this has been established, the other (favourable)
characteristics of this method are no longer relevant.
Table 5.7b:
Method: Simultaneous methods without solute recognition
Includes: Grid search
Section of this book : 5.2

Path in Figure 5.4 : 1011,1012 (1)
Optimum found : Global (2)

Accuracy of optimum : Moderate (High with (1))
Number of experiments : Very high
Criteria to be used : Any (3)
Impression of response surface : Very good
Multi-dimensional optimization : Impractical above 2 parameters

Computational requirements : Low
Complete automation : Straightforward
(1) Initial coarse grid followed by finer one around the predicted optimum.
(2) For complex samples the global optimum may not be located.
(3) Smooth response surfaces are to be preferred.
Simultaneous optimization methods (table 5.7b) do provide a means of establishing the

global optimum. However, a large number of experiments is required, especially if we wish
to locate the optimum with good accuracy, using a fine grid or an initial coarse grid
followed by a finer one in the area of the optimum. If this latter strategy is employed, or
if the response surface is very complicated (complex samples; inappropriate criteria), the
246
global optimum may not be found. However, because of the very good impression that
is obtained of the overall response surface, it is unlikely that a local optimum, if this results
from the procedure, will be much worse than the global one.
Therefore, the large number of experiments required, which becomes excessive for the
simultaneous optimization of more than two parameters, is the main disadvantage of this
method.
Table 5 . 7 ~ :
Method: Simplex
Includes: Simplex, Modified Simplex, Other statistical search methods
Section of this book : 5.3

Path in Figure 5.4 : 1012,2012
Optimum found : Local (1)

Accuracy of optimum : High
Number of experiments : High
Criteria to be used : Single-value criteria
Impression of response surface : Poor
Multi-dimensional optimization : Straightforward
Computational requirements : Fairly low
Complete automation : Possible
(1) Global optimum may be found after repeated initiation, requiring large numbers of experiments.
The Simplex method (and related sequential search techniques) suffers mainly from the
fact that a local optimum will be found. This will especially be the case if complex samples
are considered. Simplex methods require a large number of experiments (say 25). If the
global optimum needs to be found, then the procedure needs to be repeated a number of
times, and the total number of experiments increases proportionally. A local optimum
resulting from a Simplex optimization procedure may be entirely unacceptable, because
only a poor impression of the response surface is obtained.
On the other hand, the practical characteristics of the Simplex method show that its
application is usually staightforward (even for multi-parameter optimizations) and
requires little knowledge or computational effort. This explains the popularity of the
Simplex methods for the optimization of chromatographic selectivity, despite its obvious
fundamental shortcomings.
247
Table 5.7d:
Method: Simultaneous interpretive methods
Includes: Window diagrams (one or more dimensions)

Critical band method
Sentinel method
Section of this book 5.5.1

Path in Figure 5.4 4021,2121,3221
Optimum found Global

Accuracy of optimum Low (1)
Number of experiments Low
Criteria to be used Any (2)
Impression of response surface Moderate to good (1)
Knowledge required Selection of parameters and model
Multi-dimensional optimization Possible (3)

Computational requirements Variable (4)
Complete automation Complicated (5)
Solute recognition Required
(1) Dependent on accuracy of the model.

(2) Criteria based on R, or S are to be preferred over criteria based on P (see section 4.2.4).
(3) Possible for computational procedures, not possible for graphical ones.
(4) Low for graphical procedures to high for multi-dimensional computational procedures.
(5) Complete automation is complicated because of requirement for solute recognition.
Simultaneous interpretive methods (table 5.7d) provide a way to locate the global
optimum from a relatively low number of experiments. The price that should be paid for
this very important advantage of these methods is an increased effort from the
chromatographer to provide knowledge (to model the retention surfaces), increased
computational requirements and the necessity to recognize all the individual solutes in
each chromatogram. The reliability of the final result will depend on the accuracy of the
model.
Table 5.7d suggests that simultaneous interpretive methods are highly promising for
selectivity optimization, but that there is still much room for improvement if research
effort is directed at
1. the formulation of models based on sound chromatographic theory,
2. improvement of computer techniques to locate the global optimum in a multi-dimen-
sional parameter space,
3. automatic procedures to recognize the individual solutes in each chromatogram.
248
Table 5.7e:
Method: Iterative designs
Includes: Phase selection diagrams
Section of this book 5.5.2

Path in Figure 5.4 4022,2122,3222
Optimum found : Global (1)

Accuracy of optimum : High
Number of experiments : Low
Criteria to be used : Any
Impression of response surface : Moderate
Knowledge required : Selection of parameters and model (3)
Multi-dimensional optimization : Possible (4)

Computational requirements : Variable (5)
Complete automation : Complicated (6)
Solute recognition : Required
(1) Global optimum may be overlooked if large areas remain unsearched.

(2) Criteria based on R,7or S are to be preferred over criteria based on P (see section 4.2.4).
(3) No model required if linear interpolation is used.
(4) Possible for computational procedures, not possible for graphical ones.
(5) Low for graphical procedures to high for multi-dimensionalcomputational procedures. Reduc-
tion of the computation time appears to be possible (see section 5.5.2).
(6) Complete automation is complicated because of requirement for solute recognition.
Iterative designs (table 5.7e) have two main advantages over the simultaneous interpretive
methods described above.
1. The optimum can be located with a high degree of accuracy (defined by the user).
2. The accuracy of the model used to describe the retention surfaces is not the limiting
factor.
However, there is a risk that the global optimum will not be found if large areas remain
unsearched. Therefore, a combination of the two different interpretive methods, a fixed
experimental design followed by an iterative procedure to refine the location of the
optimum, may be the best possible approach, even though a slightly larger number of
experiments may be required than for either of the two methods separately.
Conclusions
Very briefly, the conclusions of this chapter can be summarized as follows:

1. Simultaneous methods (without solute recognition) may be used for selectivity optimiza-
tion in chromatography, but require large numbers of experiments.
249
2. Simplex optimization also requires many experiments, especially if the global optimum
is sought. This is a relatively straightforward method to apply in practice.
3. Simultaneous interpretive methods are a good way to locate the global optimum in a
small number of experiments, but the requirements are shifted towards good models,
computers, and peak recognition methods.
4. Iterative interpretive methods allow a more accurate location of the optimum and do not
rely on the accuracy of models. However, the global optimum may not always be found.
5. An interpretive method, which combines an initial fuced experimental design with an
iterative refinement of the optimum, appears to be the most promising approach.
REFERENCES
Vo1.3, Academic Press, New York, 1983, pp.249-341.
502. A.C.J.H.Drouen, H.A.H.Billiet and L.de Galan, J.Chromatogr. 352 (1986) 127.
503. S.N.Deming, J.G.Bower and K.D.Bower, Adv.Chromatogr. 24 (1984) 35.
504. P.J.Schoenmakers, A.C.J. H. Drouen, H.A. H. Billiet and L.de Galan, Chromatogra-
phia 15 (1982) 688.
505. J.C.Berridge, Trends AnaLChem. 3 (1984) 5.
506. J.C.Berridge, Chromatographia 16 (1982) 172.
507. J.A.Nelder and R.Mead, C0mput.J. 7 (1 965) 308.
508. J.C.Berridge, J.Chromatogr. 244 (1982) 1.
509. J.C.Berridge, Analyst 109 (1984) 291.
510. J.C.Berridge and E.C.Morrisey, J.Chromatogr. 316 (1984) 69.
51 1. F.H.Walters and S.N.Deming, AnaLChim.Acta 167 (1985) 361.
512. J.W.Weyland, H.Rolink and D.A.Doornbos, J.Chromatogr. 247 (1982) 221.
513. J.Rafel, J.Chromatogr. 282 (1983) 287.
514. R.Hooke and T.A.Jeeves, J.Ass.Comput.Mach. 8 (1961) 221.
515. G.E.P.Box and K.B.Wilson, J.R.Stat.Soc.Ser.B 13 (1951) 1.
516. W.Lindberg, E.Johansson and K.Johansson, J.Chromatogr. 211 (1981) 201.
517. V.Svoboda, J.Chromatogr. 201 (1980) 241.
518. G.E.P.Box, W.G.Hunter and J.S.Hunter, Statistics for Experiments; An Introduction
to Design, Data Analysis and Model Building, Wiley, New York, 1978.
519. 0.L.Davis and P.L.Goldsmith (eds.), Statistical Methods in Research and Produc-
tion, 4th edition, Longman, London, 1980.
520. V.V.Berry, J.Chromatogr. 199 (1980) 219.
521. M.P.T.Bradley and DGillen, Spectra Physics Chromatogr.Rev. 10 (1983) 2.
522. W.Golkiewicz, Chrornatographia 14 (1981) 41 1.
523. W.Golkiewicz, Chromatographia 14 (1981) 629.
524. L.R.Snyder and J.J.Kirkland, Introduction to Modern Liquid Chromatography, 2nd
edition, Wiley, New York, 1979.
525. R.Schil1 in R.L.Grob (ed.), Modern Practice of Gas Chromatography, Wiley, New
York, 1977, p.328.
526. P.Jandera and J.ChuraEek, Gradient Elution in Column Liquid Chromatography:
Theory and practice, Elsevier, Amsterdam, 1985.
250
527. L.R.Snyder in: Cs.Horvath (ed.), HPLC, Advances and Perspectives, Vol.1, Academic
529. M.A.Quarry, L.R.Grob and L.R.Snyder, J.Chromatogr. 285 (1984) 1.
533. C.E.Goewie, U.A.Th.Brinkman and R.W.Frei, A n d C h e m . 53 (1981) 2072.
534. P.J.Schoenmakers, H.A.H.Billiet and L.de Galan, J. Chromatogr. 282 (1983) 107.
537. G.d’Agostin6, F.Mitchell, L.Castagnetta and M.J.O’Hare, J.Chromatogr. 305 (1984)
13.
538. H.J.Issaq, Adv.Chromatogr. 24 (1984) 55.
539. S.Turina, M.Trbojevic and M.Kastelan-Macan, AnaLChem. 46 (1974) 988.
540. S.L.Morgan and S.N.Deming, J.Chromatogr. 112 (1975) 267.
541. S.L.Morgan and C.H.Jacques, J.Chromatogr.Sci. 16 (1978) 500.
542. J.L.Glajch, J.J.Kirkland, K.M.Squire and J.M.Minor, J.Chromatogr. 199 (1980) 57.
544. R.J.Laub, Am.Lab. 13 (1981) 47.
545. R.J.Laub and J.H.Purnel1, J.Chromatogr. 112 (1975) 71.
546. P.Jones and C.A.Wellington, J.Chromatogr. 213 (1981) 357.
547. C.M.Noyes, J.Chromatogr. 266 (1983) 451.
548. S.A.Wise, L.C.Sander and W.E.May, J.Liq.Chromatogr.6 (1983) 2709.
549. S.J.Constanzo, J. Chromatogr. 314 (1984) 402.
550. S.N.Deming and M.L.H.Turoff, AnaLChem. 50 (1978) 546.
551. W.P.Price, R.Edens, D.L.Hendrix and S.N.Deming, Anal.Biochem. 93 (1979) 233.
552. W.P.Price and S.N.Deming, Anal.Chim.Acta 108 (1979) 227.
553. A.-J.Hsu, R.J.Laub and S.J.Madden, J.Liq.Chromatogr. 7 (1984) 615.
554. H.J.Issaq, G.M.Muschik and G.M.Janini, J.Liq.Chromatogr. 6 (1983) 259.
555. H.Colin, A.Krstulovic, G.Guiochon and J.P.Bounine, Chromatographia 17 (1983)
209.
556. Sj.van der Wal, Chromatographia 20 (1985) 274.
557. S.Toon and M.Rowland, J.Chromatogr. 208 (1981) 391.
558. B.Sachok, J.J.Stranahan and S.N.Deming, Anal.Chem. 53 (1981) 70.
559. B.Sachok, R.C.Kong and S.N.Deming, J.Chromatogr. 199 (1980) 317.
560. J.\W.Weyland, C.H.P.Bruins and D.A.Doornbos, J.Chromatogr.Sci. 22 (1984) 31.
561. J. W. Weyland, C.H.P.Bruins, H.J.G.Debets, B.L.Bajema and D.A.Doornbos,
Anal.Chim.Acta 153 (1983) 93.
562. M.Otto and W.Wegscheider, J.Chromatogr. 258 (1983) 11.
563. M.Otto and W.Wegscheider, J.Liq.Chrornatogr. 6 (1983) 685.
564. J.L.Glajch, J.J.Kirkland and L.R.Snyder, J.Chromatogr. 238 (1982) 269.
565. J.L.Glajch and J.J.Kirkland, A n d C h e m . 55 (1983) 319A.
566. H.J.Issaq, J.R.Klose, K.L.McNitt, J.E.Haky and G.M.Muschik, J. Liq.Chromatogr.
4 (1981) 2091.
567. I.S.Lurie, A.C.Allen and H.J.Issaq, J. Liq.Chromatogr. 7 (1984) 463.
25 1
569. J.L.Glajch and L-RSnyder, J.Chromatogr. 214 (1981) 21.
570. L.R.Snyder, J.L.Glajch and J.J.Kirkland, J.Chromatogr. 218 (1981) 299.
571. L.R.Snyder and J.L.Glajch, J.Chromarogr. 248 (1982) 165.
572. P.E.Antle, Chromatographia 15 (1982) 277.
573. J.L.Glajch, J.C.Gluckman, J.G.Charikofsky, J.M.Minor and J.J.Kirkland, J.Chro-
matogr. 318 (1985) 23.
574. M.de Smet, G.Hoogewijs, M.Puttemans and D.L.Massart, AnaLChem. 56 (1984)
2662.
575. A.F.Poile and R.D.Conlon, Pittsburg Conference on Analytical Chemistry, 1983,
paper no.142.
576. A.C.J.H.Drouen, P.J.Schoenmakers, H.A.H.Billiet and L.de Galan, Chromatograp-
hia 16 (1982) 48.
577. H.Colin, G.Guiochon and P.Jandera, AnaLChem. 55 (1983) 442.
578. H.A.H.Billiet, A.C.J.H.Drouen and L.de Galan, J.Chromatogr. 316 (1984) 231.
579. G.d’Agostino, L.Castagnetta, F.Mitchel1 and M.J.O’Hare, J.Chromatogr.338 (1985)
1.
580. E.P.Lankmayr and W. Wegscheider, 8th 1nt.Symp.on Column LC, New York, May
21-25, 1984, paper no. 3a.31.
581. E.P.Lankmayr and W.Wegscheider, 9th 1nt.Symp.on Column LC, Edinburgh, July
1-5, 1985, paper no. L3.2.
582. P.R.Haddad, A.C.J.H.Drouen, H.A.H.Billiet and L.de Galan, J.Chrornatogr. 282
(1983) 71.
583. J.R.Gant, J.W.Dolan and L.R.Snyder, J.Chrornatogr. 185 (1979) 153.
584. A.C.J.H.Drouen, H.A.H.Billiet and L.de Galan, AnaLChern. 56 (1984) 971.
585. H.J.Issaq and K.L.McNitt, J.Liq.Chromatogr. 5 (1982) 1771.
586. LS.Krul1, M.Schwartz and J.N.Driscol1, Adv.Chromatogr. 24 (1984) 247.
587. J.R.Gant and P.R.Perrone, Znt.hb. 15 (1985) 24.
588. A.F.Fel1, B.J.Clark and H.P.Scott, J.Chrornatogr. 297 (1984) 203.
589. L.A.Gribov and M-Elyashberg, Crit.Rev.Chem. 8 (1979-80) 111.
590. A.C.J.H.Drouen, H.A.H.Billiet and L.de Galan, AnaLChern. 57 (1985) 962.
591. P.R.Naegli and J.T.Clerc, haLChem. 46 (1974) 739A.
592. B-Vandeginste, R.Essers, Th.Bosman, J.Reijnen and G.Kateman, Anal.Chem. 57
(1985) 971.
593. L.de Galan, Personal communication, 1985.
252
CHAPTER 6
PROGRAMMED ANALYSIS
Programmed analysis can be defined as a chromatographic elution during which the
operation conditions are varied. The parameters that may be varied during the analysis
include temperature, mobile phase composition and flow rate.
In many respects programmed analysis does not differ from chromatography under
constant conditions. Retention is still determined by the distribution of solute molecules
over the two chromatographic phases and the selectivity of the system is still determined
by differences between the distribution coefficients of the solutes. However, if the
operation conditions are changed during the elution, then the distribution coefficients may
change with time, thus affecting both retention and selectivity.
In this chapter we will take a look at some aspects of programmed analysis, particularly
those which bear relation to the chromatographic selectivity. The parameters involved in
the optimization of programmed analysis will be divided into primary or program
parameters and secondary or selectivity parameters. These parameters will be identified
for different chromatographic techniques and procedures will be discussed for the
optimization of both kinds of parameters.
6.1 THE APPLICATION OF PROGRAMMED ANALYSIS
The general elution problem
In chapter 1 (section 1.6) we have seen that only a limited number of sample components
can be eluted with optimum capacity factors in a chromatogram (see eqn.l.25). Real-life
samples often confront us with the problem that some of the components are bunched
together (and ill-resolved) early in the chromatogram, while some other components are
eluted in the optimum range of k values (see figure 6.la). If we change the conditions so
as to increase the capacity factors of the early eluting components, then the later eluting
ones will tend to give rise to impractically high k values. This so-called “general elution
problem” (see ref. [601], pp.54-55) is illustrated in figure 6.1 (chromatograms a and 6).
The idea of programmed analysis is to vary the operating conditions during the analysis,
so that all components of the sample may be eluted under optimum conditions. Such an
ideal situation is illustrated in chromatogram c of figure 6.1. Although such an ideal
situation may not always be realized, figure 6.lc provides a good illustration of the aim
of programmed analysis.
The analysis program may be defined as the function that describes the variation of the
operating conditions (or elution parameters) with time. Most often, only one parameter
is varied during the analysis. Many different programs may be used. The simplest program
is a single step (figure 6.2a) in which the parameter x changes instantaneously at a certain
time t. Other possible elution programs are illustrated in figure 6.2.
253
1
,
0 50 100
k-
Figure 6.1: Illustration of the general elution problem in chromatography. Chromatogramsa and b:
constant elution conditions. Chromatogram c (opposite page): programmed analysis.
When to apply programmed analysis?
In general, programmed analysis may be applied to samples that give rise to the general
elution problem, for example samples with a wide volatility (boiling point) range in GC
or samples with a wide polarity range in LC. The different ways in which programmed
analysis can be applied are summarized in figure 6.3.
The first field of application involves the use of programmed analysis as a scanning or
scouting technique for unknown samples. In this case the (volatility or polarity) range of
254
1 2 3 d 5 6 7 8
(C)
0 10
tlmin
the sample is not necessarily large, but the sample components may fall anywhere in a large
range. This application of programmed analysis has been discussed extensively in section
5.4.
I
t- t-
t- t-
t-
Figure 6.2: Different shapes of elution programs in chromatography. Description of programs: (a)
step; (b) linear; (c) convex; (d) concave; (e) multisegment.
255
The second field of application involves the occasional analysis of wide range samples.
In this category we find samples which only occur in the laboratory occasionally and in
small numbers, so that only a small number of chromatographic analyses have to be
performed. For samples of this kind it is usually sufficient to realize a separation and it
is not rewarding to try and optimize the selectivity, not even if the analysis time is rather
long and the required number of plates high..
The third field of application in figure 6.3 concerns a routine situation, in which a large
number of similar samples needs to be analyzed. It is in this field that it is usually
worthwhile to optimize the program.
The use of programmed analysis in a routine situation is not attractive. The application
of programmed analysis
1. requires more complicated and therefore more vulnerable equipment,
2. leads to reduced analytical reproducibility,
3. leads to increased detection limits because of variations in the baseline*), and
4. will add to the analysis time because of the time required to return to the starting
conditions.
Figure 6.3: Schematic illustration of the fields of application of programmed analysis in chromato-
graphy.
* Detection limits under programmed conditions compare unfavourably to those obtained with
isocratic elution, provided that optimum k values can be obtained in the latter case.
256
Hence, ironically, the best possible result of the optimization of a programmed analysis
is a non-programmed one, i.e. a set of conditions where an optimum separation (or at least
optimum elution of all components) can be achieved without the need to change
parameters during the analysis.
Multicolumn analysis
One way to avoid the need for programmed analysis in a routine situation is to use of
“multicolumn” or ”column-switching” methods. In these techniques more than one
column is used to realize optimum capacity factors (and optimum separation) for all
sample components. For example, if we look at the chromatogram of figure 6.1 a, we may
use a short column to separate the later eluting components, but a column with a higher
phase ratio ( VJ V,J is required to separate the early eluting components. Also, columns
with different stationary phases may be used, as long as the columns are all compatible
with the mobile phase. If different stationary phases are used, then the selectivity may be
optimized using the fixed experimental designs described in section 5.5.1.
Multicolumn analysis requires careful optimization. However, the effects of column
length, phase ratio and particle size are all predictable, so that the separation that will be
achieved on a multicolumn system can be predicted almost exactly. A different set of
columns is usually required for every different analytical problem.
The effort needed to develop and optimize a multicolumn method will become the more
justified the larger the number of analyses that needs to be performed.
More information on theoretical [602] and practical [603] aspects of multicolumn
techniques in GC can be found in the literature. Ref. [604] contains a review of
column-switching methods in LC.
6.2 PARAMETERS AFFECTING SELECTIVITY IN PROGRAMMED ANALYSIS
The effects of changes in a parameter during a programmed elution will generally be

the product of two independent factors:
1. the relationship between the parameter that is being programmed and the retention
under non-programmed conditions, and
2. the variation of the parameter as a function of time.
The relationships referred to in the first factor have been discussed extensively in
chapter 3. Two important examples are the variation of retention with temperature in GC
and with mobile phase composition in LC. If we use programmed analysis to separate wide
range samples, then the parameters which are varied during the elution should have a large
effect on retention. Hence, the most relevant parameters to be considered for programmed
analysis are the primary parameters, which have been listed in table 3.10 for the various
chromatographic techniques. Table 6.1 summarizes programmed analysis techniques for
various forms of chromatography.
An important characteristic of primary optimization parameters is that whereas they
have a large effect on the capacity factors of the solutes, they have a relatively minor effect
on the selectivity (a). This implies that the factors involved in optimizing an analysis
program (the initial and final conditions, programming rate and shape of the program)
do not affect the chromatographic selectivity to a large extent. This will be even more true
257
for parameters which have no effect whatsoever on the selectivity under non-programmed
conditions. Hence, techniques which involve the programming of such parameters (e.g.
.* flow rate programming) will not be discussed in this book.
Table 6.1:
Programmed analysis methods for various forms of chromatography.
Method Primary Programmed analysis

parameters (1)
GC Temperature Temperature programming
RPLC Mobile phase polarity Solvent programming (gradient elu-

tion)
PH pH gradients
LSC Eluotropic strength Solvent programming
I EC Ionic strength Salt gradients

PH pH gradients
I PC Various (2)
SFC Mobile phase density Density programming; pressure pro-

gramming
Mobile phase composition Solvent programming
~ _ _ ~ ~~ ~
(1) See table 3.10.

(2) Not compatible with programmed analysis owing to slow equilibration.
The second factor that determines the effects of programmed analysis, the variation of
the elution parameter(s) with time, is usually referred to as the program, for example a
temperature program in GC. In LC, the program is often referred to as a gradient.
However, we will see below that a programmed analysis in LC involves more than just a
gradient and therefore it is better to speak of a program or a gradient program.
6.2.1 Temperature programming in GC
We have seen in section 3.1 that the primary parameter in both GLC and GSC is the
temperature. We have also seen that retention in GC varies very strongly with the
temperature. The followingequation was found to describe the relationship in quantitative
terms:
Ink=InT+A/T+ B, (3.10)
where k is the capacity factor under isothermal conditions at an absolute temperature T

and A and B are constants.
258
Figure 6.4 shows a schematic example of the variation of retention with temperature in
GC for a number of solutes, which could, for example, form part of a homologous series.
The vertical lines a and b correspond to temperatures at which chromatograms would
be obtained which are similar to the chromatograms a and b in figure 6.1. Hence, we are
confronted with the “general elution problem”.
This is further illustrated by the two (almost) horizontal lines, which enclose the
optimum elution range (1 < k < 10). Apparently, there is no single temperature at which
all components can be eluted from the column under optimal conditions.
Figure 6.5a shows a typical temperature program for GC. The relevant parameters of
the program are also explained in this figure. Temperature programs in GC are almost
exclusively linear programs, i.e. during the actual heating step in the program the
temperature varies linearly with time. Occasionally a program may be comprised of several
linear segments.
Figure 6.5b shows the typical variation of the baseline with time during a programmed
temperature run according to the program of figure 6.5a. The two main sources of baseline
drift in programmed temperature GC are increased bleeding of the stationary phase at
elevated temperatures and variations in the gas flow rate. The use of two identical columns
and two detectors in a parallel configuration (baseline subtraction), of accurate flow
controllers and, especially, the use of stable stationary phases are factors which may be
used to reduce the blank signal.
Harris and Habgood, in their standard work on programmed temperature GC [605]
have shown that the retention time of a component under programmed temperature
conditions is a function of the retention behaviour of the solute under isothermal
conditions and the programming rate. The latter they defined as the heating rate ( r T :
10~1~ - 3
a b
Figure 6.4: Schematic example of the variation of retention with temperature in gas chromatography.
Retention lines are drawn for a group of 8 solutes (e.g. homologues). Vertical dashed lines (a and b)
correspond to chromatograms (a and b) in figure 6.1. “Horizontal” dashed lines indicate the range
of optimum capacity factors.
259
0
I
4
start
( injectI
,I,, ‘b’
ti
- t-
-t
Figure 6.5: (a) Schematicillustration of a temperature program for gas chromatography. The relevant
parameters of the program and the units in which they are typically expressed are as follows: Ti=
initial temperature (“C); 7’’ = final temperature (“C); rT = heating rate (OC/min); ti = initial time
(min); 9 = final time (min). (b) Typical variation of the baseline as a function of time in programmed
temperature GC.
“C/min) divided by the flow rate ( F ; ml/min). Because of this, it may be hard to reproduce
retention data in temperature programmed G C exactly, because whereas it may be
possible to accurately control the heating rate, it may be more difficult to reproduce the
flow rate F within 0.5%.
Resolution in programmed temperature GC is enhanced if the programming rate
( r T / F ) is decreased and if the initial temperature (Ti)is decreased. Giddings [606]
suggested that the first peak in a programmed analysis should not appear within about five
times the hold-up volume of the column. Since the temperature has little effect on the
selectivity in GC (see section 3.1.1), the optimization of temperature programs is a process
that may be seen as resolution optimization rather than as selectivity optimization.
6-22 Gradient elution in LC
We have seen in chapter 3 (table 3.10 b-d) that the composition of the mobile phase is
a primary parameter in various forms of LC (LLC, RPLG, LSC). Gradient elution is only
relevant for the latter two techniques, because the LLC system is not compatible with
mobile phase gradients. Figure 6.6a shows a typical gradient program for LC. The
complete program can be divided into a number of segments.
The program starts and ends at the purge segment (P). The reason for this is related to
the typical baseline observed in a gradient elution LC experiment (figure 6.6b). Unlike the
situation in GC, the main cause of the blank signal in programmed solvent LC* is formed
* By analogy with the term “programmed temperature GC” [605]we will use the term “programmed
solvent LC”, although “solvent programmed LC” is also commonly used.
260
t
Ip
f/ #
t t-
I
I
t-
Figure 6.6:(a) Schematic illustration of a solvent program (or gradient program) for LC. (p = mobile
phase composition: P = purge: R = reverse: E = equilibrate; 1 = inject; G = gradient. (b) Typical
variation of the baseline as a function of time in programmed solvent LC.
by impurities in the mobile phase, especially in the weaker solvent. Because of the high
capacity factors in this solvent, impurities tend to be concentrated at the top of the column
when the weak solvent is run through the column in the equilibrate segment (E). If a
gradient is subsequently applied, then the impurities will be washed from the column and
appear as peaks in the chromatogram. In order to minimize the background signal, the
equilibrate segment (E) should be kept as short as possible.
A second factor that contributes to the baseline variation is the difference in the
background signal (absorption; fluorescence) between the two solvents. This effect causes
the difference in the baseline level between the left and the centre in figure 6.6b. A more
extensive discussion on baseline variations in programmed solvent LC can be found in ref.
[607].
The actual gradient is denoted by G in figure 6.6a. Because large instantaneous
variations in the composition may reduce both the reproducibility of the analysis and the
lifetime of the column, a reverse segment (R) is also necessary in a gradient program. A
reproducible blank signal can only be obtained if the duration of the reverse, equilibrate
and gradient segments, as well as the time of injection (I) and the flow rate, are accurately
controlled. The duration of the purge segment is not relevant in this respect. Therefore,
it is to be recommended that a solvent program in LC is built up from a minimum of four
segments, starting and ending at the purge level.
In RPLC retention varies exponentially with the composition of the mobile phase, i.e.
approximately straight lines are obtained in a plot of In k vs. cp (see section 3.2.2). If we
look at the retention behaviour of each individual solute, then the optimum conditions
(LSS gradient, see section 5.4) correspond to a linear gradient (figure 6.2b). Linear
gradients will indeed be optimal when acetonitrile-water mixtures are used as the mobile
26 1
t I(c' t
-3
log 9 - -2 0.8
cp-.
0.85
Figure 6.7: Schematic illustration of the variation of retention with mobile phase composition in LC.
(a) RPLC with acetonitile-water mixtures; (b) RPLC of small molecules with methanol-water
mixtures; (c) LSC; (d) RPLC of large molecules with methanol-water mixtures.
phase. The typical variation of retention with mobile phase composition for some low
molecular weight solutes in this system is illustrated in figure 6.7a. Linear, roughly parallel
lines are obtained in a plot of In k vs. 'p (see also section 3.2.2).
However, if methanol-water mixtures are used as the mobile phase, the retention lines
for individual solutes tend to diverge towards 'p= 0, as is schematically illustrated in figure
6.7b (see also figure 3.14). In this system, a linear composition gradient would result in a
series of peaks with decreasing intervals. This can easily be understood by following the
horizontal dashed line for which In k = 2.3 (k= 10) from left to right in figure 6.7b. As a
consequence, slightly convex gradients are optimal for RPLC with methanol-water (and
THF-water) mixtures [608]. Nevertheless, for most practical purposes linear gradients are
acceptable for RPLC.
In LSC, an approximately linear behaviour is observed if In k is plotted against In 9.
This is schematically illustrated in figure 6.7~.Hence, in order to obtain the same effect
of the gradient program as in RPLC (figure 6.7a for the simplest case of a linear gradient),
we should aim at a linear variation of In 'p with time, i.e.*
* Eqn.(6.1) arises from the definition equation of LSS gradients (eqn.54, if a linear relationship is
assumed to exist between In k and In cp.
262
ln(cp+d)=ar+ b
or
cp = c exp ( a t ) - d (6.la)
In eqm(6.1) and (6.la) a, 6, c and d are constants. Because cp should increase with time,
a is a positive constant in both equations and hence a concave gradient (figure 6.2d) is
optimal for LSC.
Figure 6.7d shows the variation of retention with composition for some large molecular
weight solutes in RPLC. In this case, the mobile phase composition has a very strong effect
on the retention of the solutes (note the tenfold expansion of the horizontal axis). For low
molecular weight solutes, the slope in the retention vs. composition plots is typically
around 7 (see section 3.2.2) and therefore a typical solute can be eluted with a capacity
factor in the optimum range (1 < k < 10) at mobile phase compositions which span a range
of about 30% (2.3 x 100/7). Hence, for a limited number of low molecular weight solutes,
there is a good chance that an isocratic composition can be established that is within the
optimum range of each individual sample component. An example is given by the vertical
line in figure 6.7a.
The situation is quite different for high molecular weight solutes, as is illustrated in
figure 6.7d. For large, polar molecules that may be eluted with RPLC (e.g. proteins),
retention may be expected to be an exceptionally strong function of mobile phase
composition [609]. In this case, every individual sample component will have a very narrow
composition range over which optimum capacity factors will occur. If a number of
different large molecular weight components are present in the sample it may be almost
impossible to find a constant (isocratic) composition that will give rise to optimum
capacity factors for all sample components, and hence the use of gradient elution may be
hard to avoid.
It turns out [609] that the slope in the In k vs. cp plots is mainly determined by the
molecular weight of the solute. Solutes with very large molecular weight show very steep
lines. The lines denoted by 1 and 2 in figure 6.7d form two examples. Retention, however,
is mainly determined by the polarity of the solute. Therefore, a component with a much
lower molecular weight but also a lower polarity than solutes 1 and 2 in figure 6.7d may
have a similar retention time, but show a much shallower retention vs. composition line
(solute 3 in figure 6.7). Consequently, the regular picture of figure 6.7b is disturbed and
a much less structured pattern remains.
It will be clear from figure 6.7 that the nature of the mobile phase (compare figures 6.7a
and 6.7b) and the stationary phase (compare figure 6 . 7 ~with figures 6.7a and 6.7b) have
a great effect on the character of the retention vs. composition plots and hence on the shape
of the required (optimum) gradient. It will also be clear that, unlike the situation in GC,
the selectivity may be greatly influenced by variations in the mobile phase.
The situation becomes more complex if we realize that figure 6.7 only provides
schematic illustrations of the typical retention behaviour in different forms of LC.
Examples of anomalous behaviour will not be hard to find. For example, figure 6.8 shows
the variation of retention with composition for 23 phenylthiohydantoin (PTH) derivatives
263
k
0 0.10 0.20 0.30 0.U 0.50 0.60

9
-
Figure 6.8: Experimental variation of the retention of 23 phenylthiohydantoin(PTH) derivatives of

amino acids with mobile phase composition in RPLC. Mobile phase: mixtures of acetonitrile and
0.05M aqueous sodium nitrate buffer (pH = 5.81). All mobile phases contain 3’/0 THF. Stationary
phase: ODS silica. Solutes: D = aspartic acid; C-OH = cysteic acid; E = glutamic acid; N =
asparagine; S = serine; T = threonine; G = glycine; H = histidine; Q = glutamine; R = arginine;
A = alanine; METS = methionine sulphone; ABA = a-aminobutyric acid; Y = tyrosine; P =
proline; V = valine; M = methionine; NV = norvaline; I = isoleucine; F = phenylalanine; L =
leucine; W = tryptophan; K = lysine. Figure taken from ref. [610]. Reprinted with permission.
of amino acids in RPLC using acetonitrile-water mixtures that contain a small amount
(3’10) of THF as the mobile phase. The retention behaviour of these solutes under isocratic
and gradient conditions was studied by Cohen et al. [610]. In figure 6.8 we recognize a
rough pattern of parallel In k vs. cp lines, but we also see that many lines intersect
(“cross-over”) due to variations in the slopes for individual solutes.
Another complication may arise if we choose to vary the selectivity of a gradient
program in LC by varying more than one parameter at the same time. For example, the
concentration of two organic modifiers may be varied independently in RPLC (so-called
ternary gradients) or both the mobile phase composition and the temperature may be
programmed.
In figure 6.9 we take a closer look at some ternary gradients in which the composition
(cp) of two solvent components ( B and C) is varied with time*. For simplicity, figure 6.9
has been limited to linear gradients.
In figure 6.9a the concentration of both organic modifiers is seen to increase with time.
* The third and weakest solvent A is assumed to make up the solvent to loO%, i.e. p,, + p ~ pc=
+ 1.
264
t- t-
t
cp
t- t-
Figure 6.9: Examples of linear ternary gradients in which the concentration of two modifiers (Band
C) is varied simultaneously.The concentration of the base solvent ( A ) is not indicated in the figure.
In figure 6.9b we see that one organic modifier is gradually being replaced by another. In
this particular kind of gradient, the two limiting compositions (in figure 6.9b 60% B in
A and 40% C in A ) may be of equal eluotropic strength, so that only the selectivity and
not the eluotropic strength of the eluent is varied during the elution. Glajch and Kirkland
[611] refer to this kind of gradient as “isocratic multi-solvent programming”, because the
elution pattern of the solutes and the resulting chromatogram will represent isocratic
elution much more closely than typical gradient elution. Although it may be possible to
vary the selectivity in different parts of the chromatogram by “isocratic multi-solvent
programming”, it should be noted that this technique features all the disadvantages of
programmed analysis described in section 6.1. Hence, if “simple isocratic mixtures”
(mixtures of constant composition [61 I]) can be used, the use of “isocratic multi-solvent
programming” should be avoided.
a ternary gradient is shown in which a small concentration of the second
In figure 6 . 9 ~
modifier C i s present throughout the elution. Figure 6.9d shows a gradient that runs from
100% A to 100% B and subsequently from 100% B to 100% C. This may be a sensible
program if C is a considerably stronger solvent than B.
Although all the gradients in figure 6.9 are ternary gradients in that two parameters (the
concentrations of two modifiers) are varied at the same time, we may interpret three of
the four gradients as special forms of binary gradients (solvent A‘ -solvent W ) ,in which
A’ and B’ are mixtures of the pure components A, B and C. We may refer to A‘ and B‘
as pseudo-solvents and to the ternary gradients as pseudo-binary gradients. The last
gradient in figure 6.7 (figure d) can be seen as a combination of two binary gradients
265
(solvent A' - solvent B' -,solvent C'). The compositions of the different pseudo-sol-
vents for the gradients shown in figure 6.7 are listed in table 6.2.
Table 6.2:
Compositions of the solventsin pseudo-binary gradients (A'-B' or A'+B'+C' )which
are equivalent to the ternary gradients of figure 6.7.
Gradient Solvent A' Solvent B' Solvent C'

Figure 6.7
'/o B 90 c '/o B 'lo c '/o B O
/' c
0 0 60 40 N/A
60 0 0 40 N/A
0 10 90 10 N/A
0 0 100 0 0 100
From the above we may conclude that many of the ternary gradients which may be used
in LC can be seen as special forms of binary gradients. Of course, this conclusion is no
longer correct if we do not restrict the discussion to linear gradients and allow the shape
of the gradient for one solvent to be different from that for another. However, it may be
difficult to find applications for which such complicated ternary gradients can be proved
to yield better results than the simpler (pseudo-) binary ones.
Summary
In this section we have come to the following conclusions.

I . A solvent program (or gradient program) in LC should be comprised of at least four
segments (purge, reverse, equilibrate and gradient; seefigure 6.6). The program should
begin and end at the purge stage.
2. The pattern of the variation of retention with composition in LC is aflected by the choice
of both the stationary and the mobile phase. The optimum shape of the gradient for
unknown wide range samples is dictated by the phase system. Linear or slightly convex
gradients are optimal for RPLC. Concave gradients are optimal for LSC.
3. For specijk samples the optimum shape may deviatefrom this general rule. The retention
and selectivity under gradient conditions may not follow the expected pattern because of
anomalies in the isocratic retention vs. composition relationships.
4. The selectivity in programmed solvent LC may be varied by varying the solvents used or
by the application of ternary or even more complicated gradients. However, most ternary
gradients can in fact be reduced to binary ones using mixed (pseudo-) solvents.
6.3 OPTIMIZATION OF PROGRAMMED ANALYSIS
There are two aspects involved in the optimization of programmed analysis. The first
one is the optimization of the parameters of the program. These parameters include the
initial and final conditions, the shape of the program (see figure 6.2) and the duration of
266
the program segments, for instance the heating rate (in programmed temperature GC), or
the slope of the gradient in programmed solvent LC. Programmed analysis almost always
involves the variation of primary parameters during the analysis. These parameters (and
others such as the flow rate and the length of the column) will affect the separation, but
the selectivity (a) is only slightly (or not at all) affected. Nevertheless, the program
parameters form a relevant part of the optimization of programmed analysis. For instance,
the choice of the initial conditions will affect the resolution for peaks that appear
early in the chromatogram and the shape of the gradient will determine the overall
distribution of the peaks over the chromatogram and may affect the selectivity for some
pairs of peaks. Multisegment programs (see figure 6.2e) may allow the optimization of the
resolution throughout the entire chromatogram. Examples of this will be given
below.
The second aspect of optimization in programmed analysis involves adapting the
selectivity by variation of secondary parameters. The various secondary parameters listed
in table 3.10 may be used to vary the selectivity of a chromatographic system without
affecting retention to a great extent (see the discussion in section 3.6.1).
The situation in programmed analysis is similar to the one described above for
chromatographic elution under constant conditions, in that retention and selectivity may
be optimized more or less independently. However, under constant elution conditions the
optimization of the retention only involves adapting the primary parameters such that all
capacity factors fall into the optimum range (1 < k < 10). In programmed analysis the
optimization of the retention involves optimizing the characteristics of the program (initial
and final composition, slope and shape) in conjunction with the physical parameters (e.g.
flow rate and column dimensions, see section 3.6).
If the program is optimized so that all sample components are eluted under optimal
conditions, then other (secondary) parameters may be used for the optimization of the
selectivity. However, changes in the secondary parameters may imply that the parameters
of the program need to be re-optimized. For example, if the selectivity in a temperature
programmed GC analysis is insufficient, then another stationary phase may be used to
enhance the separation. However, the optimum program parameters obtained with one
stationary phase cannot be transferred to another column that contains another stationary
phase. The re-optimization of the temperature program for the other column will require
at least one additional experiment to be performed.
The primary and secondary parameters that may be used for the optimization of the
program and the selectivity in programmed analysis, respectively, are listed in table 6.3
for the various chromatographic techniques.
It can be seen in table 6.3 that the optimization of selectivity in programmed temperature
GC involves variation of the (nature or composition) of the stationary phase. To vary this
parameter, a different column and re-optimization of the (primary) program parameters
will be required. This is clearly not a very attractive proposition and therefore the
optimization of programmed temperature GC is usually restricted to optimizing the
program.
In programmed solvent LC the nature of the modifier(s) in the mobile phase is the most
common secondary parameter that may be used for the optimization of the selectivity. This
is an attractive parameter, because different modifiers may be selected and programmed
automatically on various commercial instruments. Therefore, the possibilities for selectivi-
267
Table 6.3:
Parameters for the optimization of programmed analyses in various chromatographic
techniques. Primary parameters may be used to optimize the program parameters (initial
and final conditions, slope and shape). Secondary parameters may be used to optimize the
selectivity.
Method Primary parameter(s) Secondary parameter(s)

(program) (selectivity)
GC Temperature Stationary phase
RPLC Mobile phase polarity; pH Nature of modifier(s);

stationary phase
LSC Eluotropic strength Nature of modifier(s);

stationary phase
I EC Concentration of Nature of modifier(s),

counterion; pH counterion or buffer
SFC Mobile phase density Nature of mobile phase;

stationary phase
Mobile phase composition Nature of modifier(s)
ty optimization in programmed analysis are much greater in programmed solvent LC than

they are in programmed temperature GC.
Selectivity optimization vs. multisegment programs
Two ways are open that may lead to the optimization of the resolution of all pairs of peaks
in the chromatogram. The first is to use the primary (program) parameters in designing
a multisegment gradient, the second relies on the optimization of secondary (selectivity)
parameters. In the first case, the resulting programs will be generally complex and consist
of many segments. In the second case, relatively simple, continuous programs will be
obtained. The latter is generally to be preferred, for the following reasons:
1. With simple, continuous elution programs the elution conditions for the individual
peaks (in terms of peak width and detector sensitivity) will either be constant
throughout the chromatogram, or will vary in a continuous way.
2. Simpler instrumentation may be used and the effect of the quality of instrumentation
on the resulting chromatogram is reduced.
3. The reproducibility of the analysis will be enhanced.
4. Column lifetime will be increased.
5. Selectivity optimization of simple, continuous gradients will be easier than the
optimization of complex multisegment programs, because there are bound to be serious
268
discrepancies between theory and practice, which will prohibit the exact calculation of
programs comprised of many “subtle” segments.
6. Optimization of the primary parameters of the gradient program may only lead to
sufficient separation if the selectivity is sufficiently large. If the a values between one
or more pairs of solutes are low, resolution may be enhanced by a reduction of the
programming rate and by increasing the number of plates. However, this will be at the
expense of increased analysis times and the resolution will never be better than under
constant elution conditions.
Therefore, selectivity optimization, is in principle to be preferred over multisegment
gradients. In GC, where selectivity optimization is not attractive because of the require-
ment to use different columns, one may wish to resort to multisegment gradients for
practical reasons. In LC, where selectivity optimization is readily possible by using
different modifiers in the mobile phase, multisegment gradients are of little practical
interest.
We have seen in section 6.1 that a programmed analysis in chromatography generally

requires more time than an experiment under constant elution conditions. Therefore,
optimization procedures that require large numbers of experiments are the least attractive
for the optimization of programmed analysis. The procedures that were found to require
the largest numbers of experiments under constant elution conditions in chapter 5 were
the simultaneous (“grid search”) optimization methods (see section 5.2). For this reason,
such procedures have not been contemplated for the optimization of programmed analysis
and they will not be discussed in this section. Attention in this section will be focussed on
the choice between sequential methods as described in section 5.3 and interpretive methods
as described in section 5.5.
6.3.1 Optimization of programmed temperature GC
6.3.1.1 Sequential methods
Simplex optimization
Walters and Deming [612] have used a Simplex procedure for the optimization of the
program parameters in programmed temperature GC. We have seen in chapter 5 (section
5.3) that one of the main advantages of Simplex optimization procedures is that no
knowledge is required about the relationships between the parameters considered on the
one hand and the retention and selectivity on the other. Hence, a Simplex program that
can be used for the optimization of chromatographic separations under constant elution
conditions may be used equally well for the optimization of programmed analysis. All that
is necessary to adapt the Simplex program for this purpose is to select an appropriate
optimization criterion for programmed analysis. This subject has been discussed in section
4.6.2.
The two parameters considered by Walters and Deming [612]were the initial temperatu-
re and the heating rate. They used a composite optimization criterion (see section 4.4.2)
and imposed a time constraint of 5 minutes on the system by assigning a very unfavourable
269
(infinite) value to the criterion when the analysis time was longer*. The procedure required
13 experiments, two of which could not be performed because negative heating rates were
suggested by the optimization program.
Because this optimization only concerned program parameters and not selectivity
parameters, the response surface will have been relatively simple. Therefore, the probabili-
ty that the Simplex procedure would arrive at the global optimum rather than at a local
one was greater than it was in section 5.3, where wedescribed the use of the Simplex method
for selectivity optimization.
Systematic sequential optimization
Stan and Steinbach [613] have described a sequential optimization procedure for
programmed temperature GC that searches for an optimum multisegment program in a
systematic way. This procedure can be divided into three different stages:
1. Separation of a maximum number of peaks by adapting the programming rate of each
segment, as well as the length of the preceding isothermal period**.
2. Increasing the resolution (R,; eqn.l.14) values to exceed 1.5 for each pair of peaks by
reducing segment slopes and inserting isothermal periods.
3. Reducing the resolution values to be less than 1.5 for each pair of peaks by increasing
segment slopes and shortening isothermal periods.
The first stage is the actual sequential optimization procedure. It involves the
optimization of the heating rate of each segment followed by the optimization of the
duration of the preceding isothermal period. As an example, a program was described in
ref. [613] that started with (splitless) injection of the sample at 100 O C . The injection period
was followed by a rapid (30 OC/min) heating to the initial program temperature (150 "C).
The total span of the program from 150 to 250 OC was divided in five non-isothermal
segments, each spanning 20 OC. Isothermal segments could be inserted before each of
these, so that a total of ten segments was considered during the first stage of the process.
The procedure starts by recording a first chromatogram in which the maximum heating
rate (30 OC/min) is applied throughout the program. After the injection period, the
temperature is raised from 100 OC (injection temperature) to 250 "C. The resulting
chromatogram is shown in figure 6.10a. In this example, 28 peaks can be registered from
the chromatogram.
The sequential optimization procedure now starts by optimizing the last segment of the
program (230-250 "C)aiming to increase the number of peaks observed in the chromato-
gram. To this end, the slope of this segment is successively reduced from 30 OC/min to 8,
4 and 2.67 OC/min. If reducing the slope does not result in an increase in the observed
number of peaks, then the value is rejected and the previous one is retained. A similar
procedure is followed for the next segment (isothermal at 230 "C). The duration of this
* This time constraint is required because, as we have seen in section 4.4, the incorporation of a
(weighted) time term into the optimization criterion is not an effective way to constrain the analysis
time (see eqn.4.29 and subsequent discussion).
** In this optimization procedure a segment usually refers to a part of the temperature program at
which heating takes place. Such segments may be separated by isothermal periods, during which the
temperature is kept constant. In the present discussion we will refer to the two kinds of segments as
non-isothermal and isothermal, respectively.
270
period is increased in steps from 0 to 1 , 2 or 3 minutes, until there is no further increase
in the observed number of peaks.
For the optimization of each segment a minimum of one and a maximum of three
experiments needs to be performed. Every experiment involves a complete temperature
program from 100 O C (injection) to 250 O C . For the optimization of the entire ten-segment
program, 11 to 31 experiments (including the initial run) are required. Figure 6.10b shows
the resulting chromatogram after 16 experiments were performed following the procedure
described above. The temperature program is shown underneath the chromatogram.
During the procedure, the number of registered peaks has been increased from 28 to 36,
During the second stage of the procedure, each pair of peaks is checked for sufficient
resolution. If R,< 1.5 (eqn.l.l4), then depending on the elution temperature observed for
the peak pair, either an isothermal segment may be inserted in the program, or the slope
of a non-isothermal segment may be reduced. This procedure may be followed simultane-
ously for every ill-resolved pair of peaks. Therefore, few additional experiments are
required*.
Figure 6.1 Oc shows the resulting chromatogram and temperature program after two
more injections. It is seen that the program is now much more complicated and that the
analysis time has been increased from about 25 to about 43 minutes.
During the second stage of the optimization process the number of registered peaks was
increased from 36 to 38. Whereas it was claimed in ref. [613] that 38 is the actual number
of peaks present in the sample, it seems that at this stage of the procedure additional peaks
may only be found accidentally. This will be the case if, in striving for sufficient resolution
of one particular pair of peaks somewhere in the chromatogram, a hidden peak is suddenly
revealed. In a second optimization cycle these peaks may subsequently be resolved with
R,> 1.5. However, if peaks are hidden in parts of the chromatogram in which the
resolution appears to be sufficient for all registered peaks, they will not be found during
stage 2 of the optimization process. The fact that the presence of two more peaks was
revealed during the this stage suggests that additional peaks may be “hidden” in the
chromatogram. Therefore, it may illustrate one of the shortcomings rather than one of the
advantages of the method.
Finally, in the third stage of the process, a procedure similar to that of the second stage
may be followed to reduce the resolution of abundantly resolved pairs of peaks (R,>2).
During this stage, slopes may be increased and isothermal periods shortened, leading to
a reduction of the required analysis time. Figure 6.10d shows the result obtained after an
additional two chromatograms. It is seen that the analysis time has been reduced from
about 43 to about 37 minutes.
The entire procedure illustrated in figure 6.10 involved 21 (1 + 16 + 2 + 2) chromato-
grams and took about 10 hours. Because of the sequential nature and because of the
selection of the criteria, automation is relatively easy. A serious disadvantage of the
method, besides the large number of required experiments and the complexity of the
resulting program, is the dependence of the result on the column used. Possibly, a different
* It may not be possible to achieve sufficient resolution for all the pairs of peaks in the chromatogram
on the particular column. Therefore, a stop criterion is needed in the optimization procedure, for
instance a maximum of two attemptsto separatea particular pair of peaks. If it is difficult to recognize
(pairs of) peaks, then a maximum of two or three optimization cycles each for stage 2 and stage 3 of
the optimization procedure may be considered.
27 1
t lmin -c tlmin-
250 -
0
tlmin -
5 10 15 20 25 30 35 LO L5 50 55 60
Figure 6.10: Application of the systematic sequential optimization procedure of Stan and Steinbach
[613] for the optimization of a temperature program in capillary GC. Column: 25 m x 0.2 mm I.D.
coated with dimethylsilicone bonded phase BP-1 (S.G.E.); Carrier gas: helium; Detector: electron
capture; Sample: halogenated pesticides (residue analysis). (a) Initial chromatogram; (b) Resulting
chromatogram after stage 1 (maximum number of peaks); (c) Resulting chromatogram after stage 2
(increased resolution); (d) (opposite page) Final chromatogram after stage 3 (reduced resolution). For
explanation see text. Figure taken from ref. [613]. Reprinted with permission.
272
-
100,
0 5 10 15 20 25 30 35 LO 15 50
tlmin
result may even be obtained on the same column under different operating conditions (e.g.
flow rate). This is due to the use of the column-dependent R, criterion during the second
and third stages of the optimization process (see discussion in section 4.3.3). Finally, a
reasonable estimate for the initial and final program temperatures should either be made
on beforehand, or established from the first chromatogram.
6.3.1.2 Interpretive methods
The obvious alternative to the sequential optimization methods is the use of an

interpretive optimization method. In such a method a limited number of experiments is
performed and the results are used to estimate (predict) the retention behaviour of all
individual solutes as a function of the parameters considered during the optimization
(retention surfaces). Knowledge of the retention surfaces is then used to calculate the
response surface, which in turn is searched for the global optimum (see the description of
interpretive methods in section 5.5). For programmed temperature GC the framework of
such an interpretive method has been described by Grant and Hollis [614] and by Bartd
[615].
All interpretive optimization methods are by definition required to obtain the retention
data of all sample components at each experimental location. If the sample components
are known and available they may be injected separately (at the cost of a large increase
in the required number of experiments). For unknown samples, for samples of which the
individual components are not available, and in those situations in which we are not
prepared to perform a very large number of experiments (as will usually be the case in the
optimization of programmed analysis) we need to rely on the recognition of all the
individual sample components in each chromatogram (see section 5.6).
273
If many peaks occur in a chromatogram this appears to be a very difficult proposition.
However, the requirement of solute recognition may not give rise to insurmountable
problems in the optimization of programmed temperature GC for the following two
reasons:
1. The optimization procedure may be carried out on the basis of a limited number of
(major) components in the sample [614].
2. Changes in elution order (”component cross-overs”) are unlikely to occur.
For the interpretive optimization of the primary (program) parameters in the programmed
analysis of complex sample mixtures it may well be sufficient to optimize for the major
sample components. This may be done if it is assumed that the primary parameters do not
have a considerable effect on the selectivity, so that if the major sample components are
well spread out over the chromatogram, the minor components in between these peaks will
follow suit automatically, and if it is assumed that the minor peaks are randomly
distributed over the chromatogram. The major chromatographic peaks can be separated
to any desired degree if optimization criteria are selected which allow a transfer of the
result to another column.
Changes in elution order are unlikely to occur, because temperature is not a truly
selective parameter (see section 3.1). To a first approximation, the elution order of the
peaks, and certainly the elution order of the major components, may therefore be assumed
constant.
The retention behaviour of solute moleculesunder programmed temperature conditions

is completely characterized by
1. the parameters of the program, and
2. the variation of the retention with the temperature under isothermal conditions.
If we leave out of account the delay that both the column and the packing material may
cause in the temperature program inside the column relative to the program followed in
the column oven [615], then the program parameters are naturally known.
In principle, the description of the retention vs. temperature relationships requires two
experiments, because a straight line can be obtained by plotting In (k/T) vs. 1 / T (eqn.
3.10).
Grant and Hollis [614] assume a linear relationship between In k and 1/T:
Ink=A+B/T (6.2)
where A and Bare solute-dependent coefficients.They assume that the intercept A remains
“sensibly constant”, and that the slope B is proportional to the (absolute) boiling point
for solutes within a given class. Therefore, the isothermal retention data for some “typical”
solutes from a class at a minimum of two different temperatures is thought to be sufficient
to describe the retention behaviour of all solutes within that class under programmed
temperature conditions. Unlike eqn.(3.10), eqn.(6.2) is not a fundamentally linear
relationship. Since both equations require two experimental data points two establish the
coefficients, the use of the former is to be preferred.
BartQ [615] uses a different relationship to describe the retention vs. temperature
relationship. His equation is also not fundamentally linear and requires a minimum of
three parameters:
274
In(t,-C') =A + B/T. (6.3)
In this equation t , is the retention time under isothermal conditions at the temperature T
and A, B and Care constants. An analogous expression is used to describe the variation
of the peak width at the location of half the peak height ( w , , ~with
) temperature.
The two experiments required to describe the isothermal retention vs. temperature
relationships through eqn.O.10) or eqn.(6.2), or the three required by eqn.(6.3), may either
be performed isothermally or under programmed conditions. However, in the latter case
the calculations to obtain the coefficients A, B, and, if eqn.(6.3) is used, C, will be more
complicated and more than two or three experiments may be required to estimate the
coefficients with sufficient accuracy. The latter aspect suggests the use of an iterative
interpretive method, in which the values of the coefficients are updated after each new
experiment until the accuracy of the predicted optimum turns out to be sufficient.
Neither the procedure described by Grant and Hollis [614], nor that of Bartd [615] is a
complete optimization procedure. They do not provide a generally useful strategy for
unknown or ill-known samples. The application of either approach in practice has not
been described.
6.3.1.3 Discussion
Simplex optimization of the primary (program) parameters in programmed temperatu-

re GC analysis has been demonstrated [612]. A systematic sequential search [613]may be
used as an alternative. The Simplex method may be used to optimize a limited number of
program parameters, whereas the latter approach was developed for the optimization of
multisegment gradients. The use of interpretive methods has so far only been suggested
[614, 6151.
As was the case in its application to the optimization of chromatographic selectivity
under constant conditions, the Simplex algorithm appears to require a rather large number
of experiments. This is also true for a systematic sequential procedure.
If interpretive methods are used, the calculations involved may be complicated and it
is necessary to recognize the individual solutes in each chromatogram. Because the
optimization procedure may be carried out for a limited number of major sample
components and because the elution order is not likely to vary, this will not usually be a
serious problem.
It certainly would not have been a problem in the example for which the Simplex
program was demonstrated in ref. [612]. In this sample only four components were present.
The selection of this particular example to demonstrate the applicability of Simplex
optimization for programmed temperature GC was unfortunate in any case, because a
straightforward isothermal separation of the sample at 70 OC also appeared to be possible.
The example shown in figure 6.10 (ref. [613]) for the optimization of a multisegment
temperature program was more impressive. Unfortunately, the required number of
experiments was large (21). The selection of simple criteria (e.g. maximum number of
peaks) may greatly enhance the possibilities for fully automatic optimization.
If an interpretive method is used, then the number of experiments required to allow an
accurate prediction of the global optimum may be somewhat larger than the theoretical
minimum of two experiments. However, this still compares favourably with the 21
275
experiments performed by Stan and Steinbach [613b who used a systematic sequential
procedure, and to the 11 experiments performed by Walters and Deming [612]to locate
the optimum with a Simplex method. Moreaver, Wdters and Deming performed 8
additional experimentsin the vicinity of the qtimurn to enhance the accuracy oithe result.
6.3.1.4 Selectivity optimization
In order to optimize the selectivity in programmed temperature GC, the parameter to

be varied is the nature or composition of the stationary phase. If this kind of optimization
is to be pursued, then the Simplex procedure will be especially unattractive, because it will
require large numbers of experiments using different stationary phases and, consequently,
different columns. Therefore, interpretive methods are to be preferred for optimizing the
selectivityin programmed temperature GC. Because of the experimentallyobserved linear
relationship between retention and composition in isothermal GC using mixed stationary
phases (eqn.3.14), fixed experimental designs may be used, similar to those employed for
optimizing the stationary phase composition in isothermal GC (window diagrams, see
section 5.5.1).
6.3.1.5 Summary
1. Simplex optimization of the primary parameters in programmed temperature GC

analysis is possible.
2. As with other applications of the Simplex algorithm in chromatography (see section 5.3),
a large number of experiments is required.
3. The response surface for the optimization of the primary (program) parameters in
programmed temperature GC is less convoluted than a typical response surface obtained
in selectivity optimization procedures (see section 5.1). This will increase the possibility
of a Simplex procedure locating the global optimum.
4. A systematic sequential optimization procedure may be used to establish an optimum
multisegment temperature program.
5. Such a procedure requires a large number of experiments, but may readily be automated.
6. For simple samples, in which the individual components can be recognized, interpretive
methods should ailow the prediction of the (global) optimum from a small number of
experiments.
7. For complex samples the separation of the major components may be optimized by an
interpretive method. The resulting optimum program presumably corresponds to the
optimum for the entire sample.
8. Optimization of the selectivity in programmed temperature GC requires the application
of diflerent stationary phases or stationary phase mixtures.
9. In that case, interpretive methods based on f u e d experimental designs (window
diagrams) may be used.
6.3.2 Optimization of programmed solvent LC
The (primary) program parameters may be used to optimize the separation in

programmed solvent LC in a non-selective way. Since this involves optimization of the
276
retention rather than the selectivity, this kind of optimization will only be adressed briefly
in this section. The optimization of the program parameters has been discussed extensively
by Snyder [616,6171 and more recently in an excellent book by Jandera and ChurhEek
[618].
The most useful secondary parameter for the optimization of the selectivity in
programmed solvent LC is the nature of the modifier(s) in the mobile phase. The selectivity
can be varied by selecting various solvents (pure solvents for binary or ternary gradients;
mixed solvents for pseudo-binary gradients). Analogous to the situation in isocratic LC,
it is possible to use different modifiers (and hence to obtain different selectivity),. while
optimum retention conditions are maintained for all solutes. This possibility to optimize
the selectivity in programmed solvent LC will be discussed below.
6.3.2.1 Simplex optimization
As with programmed temperature GC, the application of the Simplex optimization

procedure to programmed solvent LC is relatively straightforward. The same procedure
can be used both for isocratic and for gradient optimization, as long as an appropriate
criterion is selected for each case*.
After earlier applications of the Simplex algorithm for the optimization of programmed
solvent LC by Watson and Carr [619] and by Fast et al. [6201, the possibility of applying
(slightly) different versions of a single Simplex program for the optimization of isocratic
and programmed solvent analysis in LC was demonstrated by Berridge [621]. He used the
Simplex procedure to optimize three program parameters: the initial and the final
composition and the duration of a linear gradient. The convergence of the Simplex
algorithm to the final optimum was said to be rapid, but still 15 experiments were required
to arrive at the optimum. A reason for such a “ r a p i d convergence was suggested to be
the location of the resulting optimum on the edge of the parameter space (final
composition: 100 %B). Another reason may be the relative simplicity of the response
surface in comparison to isocratic optimization in which the selectivity (secondary
parameter: nature and concentration of modifiers) is varied.
An indication of this latter effect can be found in figure 6.1 1, which shows the result of
the Simplex optimization procedure applied to the programmed solvent LC separation of
three antioxidants [621].
The sum of peak-valley ratios was used as the resolution term in a composite
optimization criterion, which otherwise corresponds to eq~(4.30).Berridge also added a
term to describe the contribution of the number of peaks (n). With this, the complete
optimization criterion became
The desired analysis time (t,,,) was set equal to 4 min., whereas the value of the minimum
time ( tmin,which is irrelevant for the optimization process; see section 4.4.2) was taken to
be 1.5 min.
* For criteria based on the peak-valley ratio ( P ) no modification of the criterion used for isocratic
optimization may be necessary (see section 4.6.2).
277
0 1 2 3
tlmin-
Figure 6.1 1: Resulting chromatogram from a Simplex optimization procedure applied to the
separation of three antioxidants. Solvents: 5’h acetonitrile in water (A) and 5Oh water in acetonitrile
(B). Linear gradient 44to 100% B in A in 1.5 min. Column: 10 cm x 5 mm I.D. 5 pm Lichrosorb C-18.
Flow rate: 2.0 mL/min. Solutes: 1 = propyl gallate; 2 = 2-t-butyl-p-methoxyphenol (BHA); 3 =
unknown; 4 = 2,6-di-t-bytul-p-cresol(BHT).Figure taken from ref. [621].Reprinted with permission.
It can be seen in the chromatogram of figure 6.1 1 that four peaks (the three antioxidants
plus an unknown impurity) are amply resolved to the baseline. This implies that all values
for the peak-valley ratio Pare equal to 1 and that the criterion has become very insensitive
to (minor) variations in the resolution between the different peak pairs. In the area of the
parameter space in which four well-resolved peaks are observed, the only remaining aim
of the optimization procedure is to approach the desired analysis time of 4 minutes. The
irrelevance of the “minimum time” tmin is illustrated by the occurrence of the first peak
in figure 4.9 well within the value of 1.5 min chosen for this parameter.
The application of the Simplex procedure for the optimization of the selectivity in
programmed solvent LC (e.g. for the application of ternary gradients) has not yet been
reported. However, there is no apparent obstacle to the applicability of the Simplex
procedure for this purpose.
Of course, the simultaneous optimization of different (primary) program parameters
(initial and final composition, slope and shape of the gradient) and secondary parameters
(nature and relative concentration of modifiers) may involve too many parameters, so that
an excessive number of experiments will be required to locate the optimum. This problem
may be solved by a separate optimization of the program (primary parameters) and the
selectivity (secondary parameters) based on the concept of iso-eluotropic mixtures (see
section 3.2.2). This will be demonstrated below (section 6.3.2.2). However, the transfer of
278
the program parameters optimized with one modifier to an analysis program using another
modifier (or a combination of two modifiers in a ternary gradient) requires more
knowledge and understanding of the relationships between chromatographic retention
and the parameters considered in the optimization procedure than is usual for Simplex
optimization.
6.3.2.2 Systematic optimization of program parameters
Optimization without solute recognition
The concept of linear solvent strength (LSS) gradients developed by Snyder (see also
sections 5.4.2 and 6.2.2) incorporates optimization of both the shape and the slope of
gradient programs. The shape of an LSS gradient is determined by
1. the definition equation of LSS gradients, i.e.
log kin = log k , - b ( t / t o ) ,
where kinis the capacity factor of the solute under the isocratic conditions at the column
inlet at time r, k , the capacity factor under isocratic conditions corresponding to the
initial composition of the gradient program, b the gradient steepness parameter, t the
time elapsed since the start of the gradient (or, more precisely, the time elapsed since
the arrival of the gradient at the inlet of the column) and to the hold-up time of the
column.
2. The relationship between retention and composition under isocratic conditions, i.e. the
function
k = f(q). (6-5)
The combination of these two factors determines the required shape of an LSS gradient.
Linear gradients were shown to result for RPLC in section 5.4, whereas a concave gradient
was found to be optimal for LSC in section 6.2.2.
The optimal slope of the gradient also follows from the LSS concept, since it was shown
by Snyder et al. I6161 that optimum values for the gradient steepness parameter b are in
the range 0.2 < b < 0.4. If the function f(q) is known, then the optimum slope of the gradient
can be calculated. For example, in RPLC the relationship between retention and
composition over the range 1 < k < 10 can be described by
I n k = Ink, - Sq. (3.45)
In RPLC an LSS gradient is a linear gradient that can be described by
q = A + B t . (5.6)
Combination of eq~(3.45)with eqa(5.5) (see also section 5.4) yields

b = S B to / 2.303 . (5.8)
279
Typical S values for small solutes using methanol-water mixtures as the mobile phase
are in the range 5 < Sc 10 [608]. The value of to is determined by the column and the flow
rate. For example, if a column of 15 cm length with an internal diameter of 4.6 mm is used,
the hold-up volume ( Vo)is around 1.5 ml, so at a flow rate of 1.5 ml/min the hold-up time
( f a ) is about 1 min. An optimal gradient with a b value of 0.3 then leads to a range of B
values in eqm(5.6) given by
0.069 < B -= 0.138,

where B is expressed in min-I. The optimum programming rate is seen to be between
about 7 and 14 %/min. For a 0-100% gradient this corresponds to gradient durations (t,)
in the range
where f , is expressed in minutes.

Snyder et al. [616,622] recommend a simple trial-and-error approach for the optimiza-
tion of the remaining two parameters of the program, i.e. the initial and the final
composition. These parameters should be adapted such that solute bands are eluted neither
too early, nor too late in the chromatogram.
If larger solute molecules (e.g. proteins) are to be separated by programmed solvent LC,
then much higher S values may be expected and consequently (eqn.5.8) a lower B value
(shallower gradient) will be required [609].
The Snyder procedure would have led to a quick solution of the separation problem
shown in figure 6.1 1. However, the answer would have been different from that obtained
with the Simplex optimization program. If we assume an S value of about 7 for the solutes
involved and estimate the hold-up volume of the column to be around 1.18 mL (60% of
the volume of the empty column), then we can estimate the b value for the gradient used
in figure 6.1 1
6 = S B to / 2.303 = S 19 V, / (2.303 F) = 7 x 0.373 x 1.18 / (2.303 x 2) = 0.67.
This shows that the very fast gradient (t,= 1.5 min.) used in figure 6.1 1 was indeed two
or three times steeper than the optimum conditions suggested by Snyder.
Following Snyder’s approach, the first experiment could have been a gradient of 0 -
100 %B in A in 6 minutes ( b = 0.30). As a result of this gradient, the initial concentration
could then have been increased to yield (after one or two experiments) an optimum
program with a gradient from about 50 to 100% B in 3 minutes. The overall analysis time
(retention time of the last peak) would not have been much longer than the 3 minutes
observed in figure 6.11, whereas all peaks would have been eluted under optimum
conditions with roughly equal peak widths. The last peak in figure 6.1 1 is considerably
broader than the other ones, because it is eluted after the completion of the gradient
program.
However, the most important difference between the Simplex procedure and a
systematic approach such as the one suggested by Snyder is not in the quality of the
280
resulting chromatogram but is the number of experiments required. For the optimization
of the primary (program) parameters the former required 15 experiments, whereas the
latter would not have required more than 2 or 3.
Optimization with limited solute recognition
In the Snyder approach to gradient optimization the characteristics of the individual

solutesare largely neglected. The optimum shape of the gradient is determined by the phase
system and the optimum slope is usually estimated from simple rules for the retention
behaviour of the solutes (e.g. assuming S = 7 for small solute molecules as we did above).
Only the initial and the final conditions are adapted to the requirements of the sample.
A strategy for the optimization of gradient programs based on the actual retention
behaviour of some sample components has been described by Jandera and Churaeek [623,
6241. This approach relies on the possibility to calculate retention and resolution under
gradient conditions from known retention vs. composition relationships and plate
numbers. Both typical RPLC (eqn.3.45) and LSC (eqn.3.74) relationships can be
accommodated in the calculations and linear, convex and concave gradients are all
possible because of the use of a flexible equation to describe the gradient function. This
equation reads
Q = + B V)
where A is the initial concentration, B the slope of the gradient and V is the volume of
eluent delivered since the start of the gradient. Vis related to the elapsed time t and the
flow rate F by V = Ft. K characterizes the shape of the gradient. If K = 1 the gradient is
linear. K < 1 corresponds to a convex gradient and K > 1 to a concave one.
Optimum gradients were defined by Jandera and ChuraEek [624] to yield
1. a preset (required) value for the resolution (R,) between two arbitrary solutes, and
2. a minimum retention volume for another arbitrary solute.
We can summarize this optimization goal in a way that is consistent with the criteria
described in chapter 4 (section 4.3.3) as follows:
In eqn.(6.7) the pair of solutes for which a minimum resolution of x is required is denoted
+
by i and i 1. j denotes the sample component for which the retention volume under
gradient conditions ( Vg)is to be minimized.
If the retention vs. composition relationships for the solutes i, i + 1 andjare known, then
the gradient parameters A, B and K can readily be calculated for the optimum gradient
according to equation 6.6. Not unexpectedly, the value of the shape parameter K turns out
to be of little significance for an optimization procedure in which only three solutes affect
the result [624]. Therefore, it may be sufficient to optimize the parameters A and B for a
linear gradient ( K = 1).
Figure 6.12a shows the resulting optimal chromatogram for the separation of a mixture
of seven barbiturates by programmed solvent RPLC. This figure was obtained with the
following optimization criterion:
281
Vlml 10
- 5 0 V/mllO - 5 0
Vlml - 10
Figure 6.12 Resulting chromatograms from the Jandera and ChuraEek gradient optimization
method. (a) requiring a minimum resolution ( R 3 between solutes 1 and 2 of 1.7 and minimizing the
retention volume ( Vg)of solute 7 (eqn.6.7a); (b) requiring a minimum resolution between solutes 6
and 7 of 1.75 and minimizing the retention volume of solute 1 (eqn.6.7b); (c) linear gradients used
to obtain the chromatograms a and b (gradient a: 9 = 0.368 + 0.061 V; gradient b (p = 0.523 +
0.0082 V).
Mobile phase components: water (A) and methanol (B). Stationary phase: Lichrosorb ODs. Solutes:
1. barbital; 2. heptobarbital; 3. allobarbital; 4. aprobarbital; 5. butobarbital; 6. hexobarbital; 7.
amobarbital. Figure taken from ref. (6241. Reprinted with permission.
(6.7a)
Figure 6.1 2b shows the resulting chromatogram obtained under the conditions
RJ7,6 > 1.75 n min Vg,,. (6.7b)
The two different linear gradients are shown in figure 6.12~.

It can be seen in figure 6.12 that the two different criteria described by eqns.(6.7a) and
(6.7b) result in different gradient profiles and different chromatograms. In figure 6.12a
the resolution between the last two peaks is clearly insufficient. In figure 6.12b the
resolution of these last two peaks has increased, but at the expense of a decreased
resolution of the first two peaks. In the first chromatogram the gradient is too steep to
obtain sufficient resolution. In the latter chromatogram the initial concentration may be
slightly too high.
Clearly, neither in chromatogram a nor in chromatogram b is the resolution optimized
282
throughout the chromatogram. This is a disadvantage of the procedure. Another
disadvantage is that a choice needs to be made as to which two components will be the
most difficult to separate (“critical pair”) and for which solute the retention volume should
be minimized. The two above chromatograms illustrate that a different choice for the
solutes involved in the optimization criterion will lead to a different result. Apparently,
in order to improve the method other optimization criteria need to be considered. For
example, the resolution could be optimized for both the first two and the last two peaks
in the chromatogram.
Advantages of the procedure are that the calculations are relatively simple and that only
the retention vs. composition relationships of the three solutes involved in the optimization
criterion need to be known.
Complete mathematical optimization
If the retention vs. composition plots of all solutes are known, then it is in principle
possible to calculate the optimum program parameters for a simple, continuous gradient
(figure 6.2a-d). In such a procedure an appropriate optimization criterion can be selected
such that the distribution of all the peaks over the chromatogram, as well as the required
analysis time, can be taken into account (see chapter 4).
However, the calculations required for such an optimization are quite involved. This is
caused by the requirement to calculate the retention times of each solute (and the
resolutions of each pair of adjacent peaks) from the isocratic retention vs. composition
relationships. In order to characterize the response surface, these calculations need to be
performed a number of times. Finally, the optimum needs to be found on the response
surface. If all four program parameters (initial and final concentration, slope and shape)
are considered, the number of calculations would be large, even though the response
surface may be simple compared with those encountered in selectivity optimization (see
the discussion in section 6.3.2.1).
Multisegment gradients
A procedure that avoids the lengthy calculation procedure mentioned above is the one
described by Noyes [625]. She designed a multisegment gradient program on the basis of
visual interpretation of the isocratic retention vs. composition relationships for a number
of phenylthiohydantoin (PTH) amino acids. It was claimed that the mixture could not be
separated by a continuous linear gradient, but no further details on the design of the
multisegment gradient were given.
Issaq et al. [626] have reported on a method for the optimization of a multisegment
gradient program for the optimum resolution of all pairs of peaks in a programmed solvent
LC chromatogram. In their procedure a number of programmed solvent experiments are
performed, either a series of linear gradients between two solvents A and B of variable
duration ( t G) , or a series of linear gradients with constant tG, but a variable final
concentration of B in A. For each pair of adjacent solutes the gradient which yields the
best resolution is then selected and the different linear segments are combined into a
multisegment program.
The exact procedure in which the multisegment gradient is built up from the optimum
283
gradients for the individual pairs of peaks is not clarified, however, and it remains to be
seen whether the calculated program wifl indeed result in optimum separationfor all pairs
of peaks in the chrumatogram. It appears that this goal can only be achieved if the elution
pattern of a pair of peaks through the column is only affected by the particular segment
designed for the optimum resolution of this pair. Unless the different solute pairs are very
far apart in the chromatogram (in which case the overall distribution of the peaks over
the chromatogram would be far from optimal!), the resolution of a pair of peaks is likely
to be much affected by the preceding segments of the program. No examples to
demonstrate the applicability of the method are given in ref. €6261.
6.3.2.3 Znterpretive methodsfor selectivity optimization
Glajch and Kirkland [627]have extended the Sentinel optimization method (seesection
5.5.1) to include the optimization of the selectivity in programmed solvent LC. This
optimization procedure allows the use of linear gradients in RPLC using one or more
organic modifiers in water. The relative concentration of the modifiers does not change
during the analysis (so-called iso-selectivemuiti-solvent gradients (6111, see figure 6.7a).
This allows a straightforward extension of the Sentinel method.
For the optimization of programmed solvent LC the Sentinel method starts by
establishinga suitablebinary methanol-water gradient. The appraa& of Snyderdescribed
above may be used for this purpose. For example 16271, a gradient from 20 to IO@h
methanol (in water) in 20 minutes may be the result.
THF
ACN
Figure 6.13: Figure illustrating the 7 linear gradients used in the Sentinel optimization method for
programmed solvent U=. Initial and final compositions of the gradients are listed in table 6.4.
284
Next, the concept of iso-eluotropic mobile phases is used to determine the binary
acetonitrile-water and THF-water mixtures that correspond to the initial and the final
composition. For example, 20% methanol corresponds [627] to 17% acetonitrile and to
12% THF, whereas 100% methanol corresponds to 84% acetonitrile and to 59% THF.
By analogy with the isocratic Sentinel optimization procedure a series of 7 gradients (all
of the same duration time) can now be defined. These gradients are shown in figure 6.13
and the initial and final compositions are listed in table 6.4.
The individual retention times of all solutes in a 14-component sample mixture were
measured and used to calculate resolution values (eqn.1.14, because eqn.l.22 is invalid)
between each pair of peaks in the chromatogram. The largest value for the limiting
resolution (max Rs.min;eqn.4.25) was used as the optimization criterion.
Table 6.4
Initial and final compositions of the 7 linear gradients shown in figure 6.13. All 7 gradients
have the same duration.
Gradient Mobile phase composition (% v/v)

number
Water MeOH ACN THF
1 Initial 80 20 0 0
Final 0 100 0 0
2 Initial 83 0 17 0
Final 16 0 84 0
3 Initial 88 0 0 12
Final 41 0 0 59
4 Initial 81 10 9 0
Final 8 50 42 0
5 Initial 85 0 9 6
Final 28 0 42 30
6 Initial 84 10 0 6
Final 20 50 0 30
7 Initial 83 7 6 4
Final 19 33 28 20
8 (1) Initial 83 2 14 1
Final 16 10 69 5
(1) Predicted optimum gradient.
285
tlmin -
I
I
(b)
:e
al
In
C
0
::
a,
K
0.0 LO 8.0
tlmin - 12.0 160
Figure 6.14 Result of a Sentinel optimization of programmed solvent LC. Experimental design
according to figure 6.13 and table 6.4. (a) Predicted optimum linear gradient and (b) chromatogram
obtained with the optimum linear gradient. Stationary phase: Zorbax alkylsilica.
Flow rate: 3.0 ml/min. Solutes: A = resorcinol; B = theophylline;C = phenol; D = benzyl alcohol;
E = caffeine; F = methyl paraben; G = benzonitrile; H = nitrobenzene; I = cortisone;J = propyl
paraben; K = ramrod L = butyl paraben; M = chloro-isopropyl-N-(3-chlorophenyl) carbamate
(CIPC); N = progesterone. Figure taken from ref. [627]. Reprinted with permission.
Figure 6.14 shows the resulting quaternary gradient and the resulting chromatogram for
the 14-component mixture to which the 7 gradients described in figure 6.13 and table 6.4
have been applied.
It will be clear that the interpretive procedure described here allows the recalculation
of the resolution surfaces (and the response surface) after the retention times of the
individual solutes have been obtained from the chromatogram at the predicted optimum
(figure 6.14), so that a n iterative optimization procedure, in which the accuracy of the
resulting optimum is improved, is also possible.
The Sentinel gradient optimization method, by analogy with the isocratic Sentinel
method, requires a minimum of 7 chromatograms to be recorded before the optimum
conditions can be predicted and it requires the retention data of all solute components to
be established at each experimental location.
286
Step 3
100 I I
la1
tlrnin-
i:I
t
t b
8.0
tlmin -
Figure 6.1 5: (a) Step-selectivity gradient program designed after “visual interpretation” of the 7
chromatograms obtained during a Sentinel gradient optimization procedure (figure 6.13). (b)
Chromatogram obtained with the step-selectivitygradient of figure 6.15a. Sample and conditions as
in figure 6.14. Figure taken from ref. 16271. Reprinted with permission.
Advantagesare that the selectivity is optimized (secondary parameters) so that optimum

resolution can be obtained and that all components of the sample are considered in the
optimization procedure. Unlike the result of the gradient optimization procedure
suggested by Jandera and ChuraEek, (section 6.3.2.2) the lowest value for the resolution
in the chromatogram is maximized and not the resolution of an arbitrary pair of solutes.
However, because of the selection of the max Rs,min criterion the distribution of the
peaks over the rest of the chromatogram (other than the critical pair of peaks) is not
optimized (see discussion in section 4.3.3). This was realized by Glajch and Kirkland [627]
who therefore tried to optimize a “selective multi-solvent’’ gradient, in which a series of
segments is allowed in order to try and optimize the resolution in various parts of the
chromatogram. They did not describe a formal procedure for the optimization of such
step-selectivity gradients, but used “visual interpretation” of the seven chromatograms
obtained during the optimization procedure described above to design the gradient shown
in figure 6.15a. The chromatogram obtained with this gradient is shown in figure 6.15b.
287
The chromatogram in figure 6.1 5 is only marginally (if at all) better than the one shown
in figure 6.14. However, Glajch and Kirkland correctly state that very few of the
possibilitiesof exploiting various selectivegradients have yet been explored. If the relative
concentrations of the organic modifiers are allowed to vary and if the variation of
composition with time is not restricted to linear relationships, then the distribution of the
peaks over the chromatogram may still be greatly improved. However, the use of simple
continuous gradients is to be preferred to the use of complex multisegment gradients for
a number of reasons outlined in the introduction of section 6.3.
predictive optimization method
The Sentinel method of GIajch and Kirkland described above involves the measurement
of retention data under gradient conditions and the direct optimization of the selectivity,
i.e. the differences between these retention times for different solutes. Jandera et al. [628]
have described a predictive optimization method in which
1. retention vs. composition relationships are obtained under isocratic conditions using
several modifiers,
2. the retention data using ternary gradients are predicted from the isocratic data, and
3. an adequate ternary gradient is selected based on the predicted retention times.
According to Jandera et al. [628], the isocratic retention behaviour of solutes in ternary
solvents in RPLC may be predicted from data obtained with binary mixtures. However,
such predictions are only accurate within about 5%. This accuracy is insufficient for the
purpose of selectivity optimization, where small differences in retention times between
adjacent peaks are of critical importance. Therefore, ideally, binary as well as ternary
mixtures should be used in the isocratic experiments. The selection of an adequate ternary
gradient takes place largely on a trial-and-error basis. However, instead of trial experi-
ments, trial calculations are performed until a satisfactory result is predicted. Only then
will a trial experiment be performed.
Figure 6.16 illustrates the application of the method of Jandera et al. for the selection
of a satisfactory linear gradient for the separation of a mixture of 9 phenolic solutes. It
is seen in figure 6.16a and figure 6.16b that the mixture is not completely separated using
either a binary methanol-water (chromatogram a) or a binary acetonitrile-water gradient
(chromatogram 6). Also,an ”iso-selective”linear gradient, in which the ratio between the
concentrations of methanol and acetonitrile is kept constant, provides insufficient
resolution. Figure 6.16d shows the chromatogram obtained with a linear ternary gradient
which was predicted to provide a satisfactory separation. Indeed, the resolution is better
than in any of the previous chromatograms (a, b and c) and is sufficient with the column
and conditions used in figure 6.16.
Figures 6.16e and 6.16f show two chromatograms using gradients which were predicted
to yield insufficient separation. Using the optimization procedure of Jandera et al., a
number of gradient programs can be tested by calculating the resulting chromatograms,
so that the number of experiments required can be greatly reduced.
It is interesting to note that the gradient predicted by Jandera et al. could not have been
arrived at using the Sentinel method described in figure 6.13.
The predictive optimization method of Jandera et al. is designed to yield an “adequate”
result. In other words, a threshold optimization criterion is used (eqn.4.23). Once a certain
288
8 6 3
60 V/ml 30 20
1 I
10 - 0
I
V/ml 30
I I
20
I
10 - 1
0
LOV/ml 30 20 10 - I
0 Vlml
I
30 20
I
10 - 0
I
8.9
(fl
3
II
8.9
I I I
LOVlml 30 20 10- 0 Vlml30 20 10 - 0
Figure 6.16: Illustration of the predictive optimization method for ternary gradients in RPLC of
Jandera et al. [628]. All figures were recorded with linear gradients from 100°h solvent A to l0O0/o
solvent B in 60 min. Stationary phase: Lichrosorb C18. Flow rate: 1.0 mllmin.
Solutes: 1 = 4cyanophenol; 2 = 2-methoxyphenol; 3 = 4-fluorophenol; 4 = 3-fluorophenol; 5 =
m-cresol; 6 = 4-chlorophenol; 7 = 4-iodophenol; 8 = 2-phenylphenol; 9 = 3-t-butylphenol.
Mobile phase components: (a) solvent A: 20% methanol (in water), solvent B: 100°h MeOH; (b) A:
100% water, B: 100°/o acetonitrile (ACN); (c) A: 100°h water, B 60°h MeOH + 40°/o ACN; (d) A:
20°/0 ACN, B: 100°h MeOH; (e) A: lOoh ACN, B lOOohMeOH; (f) A: 30°h ACN, B lOOohMeOH.
Figure adapted from ref. [628]. Reprinted with permission.
289
minimum resolution is predicted for all the pairs of peaks in the chromatogram this is
said to be an adequate or sufficient result, provided that it can be verified experi-
mentally.
A disadvantage of the method of Jandera et al. is the requirement to know the isocratic
retention vs. composition relationships. If these data are not already known, which will
most likely be the case in the optimization of real-life samples, the experimental effort
needed to obtain sufficient data of sufficient accuracy will be very large.
6.3.2.4 Discussion
We have seen that the primary (program) parameters can be optimized in one of several
ways. If the actual gradient consists of a single segment, four parameters may be
considered, of which two (the slope and the initial composition of the gradient) are most
relevant for the result in terms of resolution. The final composition may affect the required
analysis time (the program should not extend beyond the chromatogram), whereas the
shape of the gradient will have an effect on the overall distribution of the peaks over the
chromatogram.
The Simplex optimization procedure allows different optimization criteria to be used,
so that a good distribution of all the peaks over the chromatogram may be aimed at.
However, the Simplex method does require a large number of experiments, and therefore
seems to be very inefficient for optimization of the primary parameters alone.
Without knowing much about the sample, the Snyder approach may also be used to
optimize the program parameters. This is an empirical approach in which the sample
properties are largely disregarded, but it does lead to the formulation of reasonable
working conditions after only one or two chromatograms have been obtained.
The approach of Jandera and ChuraEek allows the optimization of the resolution of one
given (arbitrary) pair of sample components and the minimization of the retention volume
of another (arbitrary) solute. It requires knowledge of the isocratic retention vs.
composition relationships of these three solutes. The information needed may be acquired
from gradient elution experiments performed as part of the optimization procedure, or
from separate isocratic experiments.The selection of the three arbitrary solutes considered
during the optimization process appears to have a large effect on the result and the
resolution cannot be optimized throughout the chromatogram.
In principle, the retention behaviour of all sample components under gradient
conditions can be calculated once two experimental retention times have been obtained
(either under isocratic or under gradient conditions) [616,618,623,629]. Therefore, in
principle, it ought to be possible to calculate optimum gradient parameters from two
solvent programmed experiments [630]. However, to account for inaccuracies in the
gradient elution data [630,631,632], a few more experiments may be required. Procedures
to obtain isocratic retention vs. composition relationships from a series of gradient
experiments have also been described by Jandera and ChuraEek [633,634]. Determination
of the optimum program parameters based on the retention vs. composition relationships
for all (or all major) sample components will require quite complicated and extensive
calculations. It is the charm of the methods described in section 6.3.2.3 that the required
computational effort is either minimal (i.e. a few computations, which can easily be
performed on a pocket calculator for the Snyder method) or small (i.e. a limited number
290
of computations involving more complex but analytical expressions for the method of
Jandera and ChuraEek).
A second reason not to become involved in extensive calculations for the complete
mathematical optimization of the (primary) program parameters is that a more powerful
way to optimizethe separation of all sample components in the mixture may be to optimize
the selectivity of the gradient by varying the nature of the mobile phase components
(secondary parameters).
Three methods appear to be available for optimizing the selectivity in programmed

solvent LC:
1. the Simplex procedure,
2. interpretive methods, and
3. the predictive optimization method.
The Sentinel method is the outstanding exponent of the group of interpretive methods, as
it has already been applied successfullyfor selectivity optimization in programmed solvent
LC. However, other interpretive methods, based either on fixed experimental designs or
on iterative procedures, can be applied along the same lines. It was seen in section 6.3.2.3
that the extension of the Sentinel method to incorporate gradient optimization was fairly
straightforward.
For the Simplex optimization procedure the common disadvantage of the large number
of required experimentsweighs more heavily for programmed analysis,because more time
is required for each experiment (see section 6.1). Also, the response surfaces encountered
in the optimization of selectivity in programmed solvent LC appear to be no less
convoluted than the ones encountered in isocratic selectivity optimization [627], so that
there is again a large chance that the Simplex algorithm will arrive at a local rather than
the global optimum.
The advantages and disadvantages of interpretive methods are also fully analogous to
those listed in chapter 5 (section 5.5). Fewer experiments are needed, but the recognition
of the different sample components is required in each experiment. Contrary to the
complete optimization of the (primary) program parameters, interpretive methods for the
optimization of the selectivity under programmed conditions do not require more
complicated calculations than do their isocratic analogs. This was amply demonstrated by
Glajch and Kirkland [627], who used the same computer program for the two optimization
processes.
The predictive method of Jandera et al. [628] requires knowledge of the isocratic
retention data of all solute components in binary and (preferably) ternary mobile phase
mixtures. Once these data are available, the method may be very helpful in obtaining an
“adequate” (but not an optimum) separation with a ternary gradient. Unfortunately, the
data required for the application of this predictive method are almost never available, and
hence a large number of experiments need to be performed before any predictions can take
place. When this is the case the method is of very little practical use.
The final question we need to address in this discussion is the general need for gradient
optimization procedures, both for optimizing the program parameters and for optimizing
the selectivity. In section 6.1 several disadvantages of programmed analysis were described
and it was concluded that its application should be avoided if possible. Especially for large
29 I
series of samples, the use of alternative (multicolumn) techniques should be considered.
In isocratic analysis, the general motivation is that the larger the supply of a particular
kind of sample, the more optimization effort is warranted. In programmed analysis this
is not true. In that case, the larger the supply of samples, the larger the urge to look for
alternativemethods. Therefore, gradient optimization procedures are only relevant if they
represent a limited effort. It yet remains to be established just how far the word "limited"
will reach.
6.3.2.5 Summary
The characteristics of the different methods for gradient optimization are summarized
in table 6.5. In table 6Sa, the different methods for the optimization of the program
parameters are compared. Bearing in mind that a large effort is generally not warranted
for the optimization of programmed analysis (seesection 6.3.2.41, we shouldconclude that
the Simplex method is not suitable because of the large experimental effort required, and
Table 6.5:
Summary of the characteristics of gradient optimization methods.
a. Optimization of primary (program) parameters
Simplex Snyder Jandera Complete

method method method mathematical
optimization
No-experiments Large 1 or2 few few
Computational
effort Moderate Minimal Small Large
Resolution YeS No One All solutes

optimization (1) (2) pair
Time Yes YeS One YeS

optimization (3) (4) solute (3)
Recognition None None Three All (major)

requirements solutes solutes
Complete
automation Ea4y Easy Difficult Difficult
(1) Any optimization criterion can be selected that assigns a single criterion value to each
chromatogram.
(2) Optimum slope is selected to provide optimum elution conditions for "average" solutes.
(3) Time factor may be incorporated in optimization criterion.
(4) Initial and final conditions may be adapted to first and last peaks to minimize analysis time.
292
that a complete mathematical optimization is unattractive because of the large computa-
tional effort involved. The method proposed by Jandera and ChuraEek requires somewhat
more effort than that of Snyder. It requires some calculations, the recognition of three
solutes, and knowledge of the isocratic retention vs. composition relationships for these
solutes, obtained either during the optimization procedure or from independent (isocratic)
experiments.
Table 6.5:
Summary of the characteristics of gradient optimization methods.
b. Selectivity optimization.
Simplex Interpretive methods
method Fixed Iterative Predictive

design design optimization
(Sentinel) method
No.experiments Large 7 5-10 Large (1)
Computational
effort Small Moderate Moderate Moderate
Optimum found Local Global Global(2) “Adequate”

(3)
Accuracy of
optimum High Low High -
Impression
of response Poor Good Moderate Poor
surface
Optimization
criterion
Single
value
Any Any Rs.min ’
(eqn.4.23)
x
Recognition
required No Yes Yes Yes (4)
Complete Easy Partly Difficult -

automation easy (5)
(1) Large number of isocratic experiments required.
(2) Global optimum may be overlooked if large areas remain unsearched.
(3) This method aims at achieving an adequate rather than an optimum result.
(4) Recognition of the peaks is required during the isocratic experiments to establish the retention
vs. composition relationships.
( 5 ) Experimental part may easily be automated.
293
On the other hand, this method does take into account the resolution of the most critical
pair of solutes. If this pair can easily and unambiguously be identified, then the method
of Jandera and ChuraiSek may be worth the extra effort.
In table 6.5b the methods for selectivity optimization are compared. Again, the Simplex
method turns out to be unattractive, because of the large number of experiments required.
Also, the resulting optimum may well be a local one.
Interpretive methods will generally arrive at the global optimum after a limited number
of experiments. However, (by definition) the recognition of the individual solutes is
required in each experimental chromatogram. Also, the computational requirements are
relatively high, especially if the simultaneous optimization of several parameters is
considered. For example, (linear) ternary gradients (one parameter) will be much easier
to optimize than quaternary gradients (two parameters).
Interpretive methods may possibly be used for the complete optimization of selectivity
in solvent programmed LC. If any gradient program (multisegment gradients, see figure
6.2e) is allowed, then it may be possible to optimize the resolution of each pair of peaks
in the chromatogram. This possibility has been largely unexploited. However, it also
appears to be of limited practical interest, because of the disadvantages of multisegment
gradients compared with simple, continuous gradients (see introduction section 6.3).
REFERENCES
601. L.R.Snyder and J.J.Kirkland, Introduction to Modern Liquid Chromatography,2nd

edition, Wiley, New York, 1979.
602. J.F.K.Huber, E.Kenndler, W.Nyiri and M.Oreans, J.Chromatogr. 247 (1982) 21 1.
603. W.Blass, K.Riegner and H.Hulpke, J.Chromatogr. 172 (1979) 67.
604. C.J.Little, D.J.Tompkins, O.Stahel, R.W.Frei and C.E.Goewie, J.Chromatogr. 264
(1983) 183.
605. W.E.Harris and H.W.Habgood, Programmed Temperature Gas Chromatography,
Wiley, New York, 1966.
606. JCGiddings in: N.Brenner, J.E.Callen and M.D.Weiss (eds.), Gas chromatography,
Academic Press, New York, 1962, pp.57-77.
608. P.J.Schoenmakers, H.A.H.Billiet and L.de Galan J.Chromatogr. 185 (1979) 179.
609. L.R.Snyder, MStadalius and M.A.Quarry, AnaLChem. 55 (1983) 1421A.
610. K.A.Cohen, J.W.Dolan and S.A.Grillo, J.Chromatogr. 316 (1984) 359.
61 1. J.L.Glajch and J.J.Kirkland, AnaLChern. 54 (1982) 2593.
612. F.H.Walters and S.N.Deming, AnaLLett. 17 (1984) 2197.
613. H.-J.Stan and B.Steinbach, J.Chromatogr. 290 (1984) 31 1.
614. D.W.Grant and M.G.Hollis, J.Chromatogr. 158 (1978) 319.
615. V.BartO, J.Chromatogr. 260 (1983) 255.
61 7. L.R.Snyder in: Cs.Horvath (ed.), HPLC, Advances and Perspectives, Vol.1, Academic
61 8. P.Jandera and J.ChurhEek, Gradient Elution in Column Liquid Chromatography.
Theory and practice, Elsevier, Amsterdam, 1985.
294
619. M.W.Watson and P.W.Carr, Anal.Chem. 51 (1979) 1835.
620. D.M.Fast, P.H.Culbreth and E.J.Sampson, Clin.Chem. 27 (1981) 1055.
621. J.C.Berridge, J.Chrornatogr.244 (1982) 1.
622. J.W.Dolan, J.R.Gant and L.R.Snyder, J.Chromatogr. 165 (1979) 31.
625. C.M.Noyes, J.Chrornatogr. 266 (1983) 451.
626. H.J.Issaq, K.L.McNitt and N.Goldgaber, J.Liq.Chromatogr. 7 (1984) 2535.
627. J.L.Glajch and J.J.Kirkland, J.Chromatogr. 255 (1983) 27.
628. P.Jandera, J.ChuraEek and H.Colin, J.Chrornatogr.214 (1981) 35.
629. P.J.Schoenmakers, H.A.H.Billiet, R.Tijssen and L.de Galan, J.Chrornatogr. 149
(1978) 519.
630. M.A.Quarry, L.R.Grob and L.R.Snyder, J.Chrornatogr. 285 (1984) 1.
633. P.Jandera and J.ChuraEek, J.Chrornatogr.91 (1974) 223.
634. P.Jandera and J.ChuraEek, J.Chrornatogr.93 (1974) 17.
295
CHAPTER 7
SYSTEM OPTIMIZATION
In the preceding chapters we have dealt with the various stages of the process of
developing methods for chromatographic analysis. We discussed the selection of the
appropriate chromatographic method in chapter 2. Chapters 3,4 and 5 described the
parameters, the criteria and the procedures, respectively,that may be used to optimize the
retention and the selectivity. In chapter 6 this approach was extended to include the
optimization of programmed analysis methods.
At the end of the selectivity optimization procedure, we have established the optimum
combination of a mobile and a stationary phase (the optimum phase system). In some
cases, the procedure has been conducted on the column and instrument on which the
analysis will eventually take place (“final analytical column”). For example, if we have
optimized the mobile phase composition for a particular separation of inorganic anions
on a dedicated ion chromatography system, we may not be able to vary the dimensions
of the column or to select different pieces of instrumentation.
Preferably, however, we may still optimize the dimensions of the column after we have
established an optimum phase system. The available instrumentation puts constraints on
the column that may be used and hence, ideally, we should also have the possibility to select
the most appropriate instrumentation for a given application.
In this chapter we will briefly discuss the selection of optimum columns and instruments,
in other words the final optimization of the complete system.
7.1 INTRODUCTION
This chapter describes the final configuration of the chromatographic system (column
and instrument) after the optimization of the phase system (the combination of the
stationary and the mobile phase) has been completed. The entire optimization process is
illustrated in figure 7.1. This figure shows the different stages in the process from the
moment at which it has been decided (either on the basis of literature information or on
the basis of figure2.1) which chromatographic method should be used. For example, it may
have been decided that RPLC is the method of choice. It should also be decided what kind
of detector will be used. For instance, we may choose to use a UV absorption detector.
The “build instrument” stage in figure 7.1 implies that a system should be assembled that
contains the appropriate column and detector. For the optimization of separations with
(capillary) GC we may also have to decide upon the type of injector to be used. However,
at this stage only a workable system (one in which all relevant components can be injected
and detected) and not an optimized system has to be assembled or “built”.
The method development process will be aided if we are able to use sophisticated
instrumentation (see also section 1.7.2). Automated injection and data handling will allow
a number of experiments to be performed without the requirement of an analyst being
present. Moreover, we have seen in chapter 5 (section 5.6) that the use of sophisticated
detection techniques (dual-channel or multi-channel detectors) may be of help in the
optimization process.
296
The inclusion of programming options (temperature programming in GC, solvent
programming in LC) in the instrument may also be helpful, not only if a programmed
analysis may be the result of the Optimization procedure (chapter 6), but also to provide
a scanning (or “scouting”) facility for unknown samples (section 5.4).
In many cases, the “build instrument” stage only involves the insertion of the column
of choice in an existing instrument configuration for method development.
BuiId Instrument
time -
Figure 7.1: Different stages in the process of developing methods for chromatographic analysis. The
individual stages of the process are located on a curve that indicates the sequenceof events (horizontal
axis) and the relative importance of the various steps (vertical axis). See text for further explanation.
The next two stages in figure 7.1 (“optimize k” and “optimize a”)represent the selectivity
optimization process, which we have discussed extensively in previous chapters. It is
assumed in figure 7.1 that most time (and effort) is spent at this stage.
We then come to the final two stages, the optimization of the column and the instrument.
These will be the subject of this chapter. Both of the final stages can be realized in a
relatively short time.
In figure 7.1 a relative importance is assigned to each stage of the process. Because it
is (as yet) impossible to predict the chromatographic behaviour of solutes from structural
information alone and, moreover, because the structure of all sample components is
usually not known, we have to rely on chromatographic experiments for the optimization
of the selectivity. Consequently, the first step is that an instrument should be “built”.
The next most important factor is to bring the capacity factors into the optimum range.
At the same time or immediately thereafter, we should try to optimize the selectivity (a).
Both are very important stages in the method development process, because no separation
will be obtained if either k = 0 or a= 1 (see section 1S),no matter how efficient the column
and how good the instrument. Very large k values should also be eliminated at this stage,
because of both time and sensitivity considerations (see e.g. figure 6.lb).
The column and instrument optimization stages are not of the same degree of
importance as the preceding stages. However, this by no means implies that they are
irrelevant. Analyses may be performed in a fully adequate way on “overdesigned” columns
with large numbers of theoretical plates, but this will usually involve long analysis times
297
and moderate sensitivities. By optimizing the column dimensions and adapting the
characteristics of the instrument to those of the column, the speed and the sensitivity of
the analysis may be greatly enhanced.
The result of one of the optimization procedures described in chapter 5 is a chromato-
gram with the best achievable separation of peaks. Depending on the optimization
criterion (chapter 4), this optimum chromatogram may have been defined in one of several
ways. For example, the separation may be optimized so as to require the shortest possible
analysis time on a given column, or to require the lowest possible number of plates on a
tailor-made column.
The resulting optimum chromatogram may be characterized by two parameters:
1. The lowest value for the resolution ( Rsemin)or for the separation factor (Smin) observed
in the chromatogram.
2. The capacity factor of the last peak ( k J . Once a column and flow rate have been
selected for the analysis, this parameter determines the required analysis time (10).
The lowest value of R, or S observed in the chromatogram determines the number of
plates that is required for the adequate resolution (characterized by R,,,,) of all the peaks
in the chromatogram. This became clear in chapter 4 (section 4.4.3), where we found
(4.35)
or
(4.34)
where N , is the number of plates available on the column used during the optimization
procedure.
The required number of plates (Nnd is the most relevant factor for the selection of the
type of column and the column dimensions. However, there are various other factors
which we need to consider in the selection of the most suitable column for a given analysis:
1. The instrumental constraints, such as the maximum acceptable pressure drop over the
column.
2. The typical size of the sample (injection volume) and the concentrations of the
components to be analyzed.
3. The combination of column and instrument, for instance the selection of injectors and
detectors that allow the optimized separation to be performed.
In packed columns, there are two parameters which may be varied independently in
order to optimize the column characteristics, i.e. the diameter of the column and the
diameter of the particles. In open columns, only the column diameter may be varied.
Additionally, the phase ratio may be varied by changing one of the “capacity parameters”
(see section 3.5). For packed columns these parameters include the surface area of the
packing material, the column porosity and the stationary phase film thickness. For open
columns only the latter parameter is relevant.
In subsequent sections we will discuss the implications of the optimization of the
efficiency and the sensitivity for optimum dimensions of the column. In section 7.4 we will
address the consequences of these optimum dimensions in terms of instrumentation
requirements.
298
7.2 EFFICIENCY OPTIMIZATION
The optimization of the efficiency of the chromatographic system involves the selection
of a column with a sufficient but not excessive number of plates. If a column is used with
twice the required number of plates, then the observed resolution would exceed the
required value by 40%, but both the analysis time and the pressure drop over the column
would be double the required value. Moreover, the sensitivity of the detection would be
decreased by 40% (see section 7.3). It is clear that we should aim to use a column with the
optimum (i.e. the required) number of plates.
7.2.1 Open columns vs. packed columns
The main fundamental difference between open columns and packed columns is in the
required pressure drop, for which Darcy's law provides a general equation:
Ap=qLu/B,
where Ap is the pressure drop over the column, q the viscosity of the mobile phase, u the
(average) linear velocity, and B, a constant representing the specific permeability
coefficient of the column. For an open column*
B , = d: / 32,
whereas for a packed column
B, z d p ' / 1000. (7.2a)
Consequently, if we compare a packed column with an open column for which dp= d , ,
we find a 30 times lower pressure drop over a capillary column of the same length.
Conversely, if we keep the pressure drop constant, much longer capillary columns may be
used, yielding a much higher number of plates.
For the comparison of different kinds of columns it has become increasingly common
to use the reduced plate height (h), defined as
h = H / d (7.3)
and the reduced linear velocity (v), defined as
v = ud/D, (7.4)
where H is the conventional plate height (in units of length) and D, the diffusioq
coefficient of the solute in the mobile phase. In eqm(7.3) and (7.4) the diameter d
represents the diameter of the particles in a packed column (d,) or the diameter of an open
* The constants in eqns.(7.2) and (7.2a) are dimensionless if p is expressed in Pa (N/m2 ), q in Ns/m2,
L, d, and d,, in m and u in m/s.
299
column (dJ An equation for the analysis time (tdcan be found if we combine eqns.( 1.3),
(1.6) and (1.18):
NH
t, = r,fl + kJ = L/u(1 + kJ =-(I
U
+ kd. (7.5)
With eqna(7.3) and (7.4) we find
For packed columns, typical values [701] are h = 3 and v= 3, so that h/ v= 1. For open
columns typically h = 1.5 and v= 5, so that h/v=0.3. Consequently,capillary columns will
lead to analysis times that are about three times shorter (for dp= dJ for thesame separation
( N and k constant). Therefore, in principle, capillary columns are superior to packed
columns. Unfortunately, capillary columns cannot always be used. This arises from the
occurrence of the diffusion coefficient (D,J in eqn.(7.6). Typically, D, is I~,OOOtimes
larger in gases than it is in liquids. This necessitates the use of very small particles (typically
5-10 pm) in HPLC columns. If we compare packed and capillary columns with d,,= d, ,
which is a reasonable assumption for GC [7021, then capilIary columns with very small
internal diameters need to be considered for LC [703]. Such very narrow columns impose
extreme demands on the instrumentation, and at present open tubufar columns cannot be
used for practical LC separations.
7.2.2 Gas chromatography (open columns)
In GC we have a real choice between packed columns (d,= 100-200p;150-65 mesh)

and open columns (d,= 50-500 pm). Capillary columns have the advantage of enhanced
speed of analysis (eqn.7.6). In order to exploit this advantage, “narrow-bore” capillaries
(d,< 100 p)should ideally be used. However, such columns require relatively high inlet
pressures (especially for high plate counts)* and considerable experimental modifications
and have a very low sample capacity [702].
Because of all these reasons, so-called “wide-bore” capillaries (d,w 500 pm) have
recently gained considerable popularity. These columns, which are usually provided with
a thick (about 1 pm) film of stationary phase, behave in a fairly similar way to packed
columns. They show low pressure drops (allowing them to provide a much higher
efficiency than packed columns), may easily be installed in most instruments and have a
high sample capacity. However, they also behave similar to packed columns in terms of
separation speed. Therefore, the current capillary cotumns with diameters between 100
and 300 pm form a reasonable compromise between instrumental limitations and
theoretical promises.
Despite the current popularity of “wide-bore”capiIIary columns, it is to be expected that
advances in instrumentation and column technology, combined with the increased
* For high pressure drops, because of the compressibility of the mobile phase, the favourable effect
of reducing the column diameter is less than that suggested by eqm(7.6). In that case t , a d, is a better
approximation than r,a d,‘ [702].
300
acceptance of capillary columns amongst practical chromatographers, will lead to a
further reduction of the diameter of the column in the future.
Packed columns may still be used in GC as a robust tool to effectuate simple, routine
separations. They have two fundamental disadvantages relative to capillary columns:
1. longer analysis times, and
2. limited efficiency due to a high pressure drop per unit length.
Because of these two reasons the use of packed columns should be limited to simple
separations. In that case there are some practical advantages:
1. Large sample capacity.
2. Because the entire sample can be brought onto the column, the accuracy of quantitative
analysis may be enhanced.
3. High contamination capacity, i.e. “dirty” samples can be injected onto the column
without causing rapid degradation.
4. Low instrumental requirements and easy operation.
5. Detectors which intrinsically require a large volume (e.g. (FT)IR spectrometers) may
be used more readily.
Therefore, despite the theoretical superiority of capillary columns, a place remains for the
use of packed columns in GC.
It is important to notice at this stage that the result of a selectivity optimization

procedure is often a separation that can be realized with a limited number of theoretical
plates. For example, we have seen in chapter 4 that the complete resolution of 10 equally
distributed peaks requires only 400 plates in the optimum situation at which the lowest
analysis time can be achieved (see figure 4.1 1 and related discussion). Large numbers of
theoretical plates are more appropriate for very complex samples, which contain large
numbers of peaks, making selectivity optimization an unrealistic proposition.
If the required number of plates is moderate (say severaI thousands), then short capillary
columns may be used to provide fast analysis of the sample. The required column length
and retention time can easily be calculated from eqm(7.3) and (7.6). For example, if we
operate a capillary column with a diameter of 200 pm at a reduced velocity of 5 with a
reduced plate height of 1.5, then 2000 theoretical plates require a column with
L = N h d, = 2,000 x 1.5 x 0.02 = 60 cm
and, using eqm(7.6) with a diffusion coefficient of 0.1 cm2/s and a capacity factor of 3
for the last solute
1.5 (0.02)2
t, = 3N ( l + k J
VDm
=
5 x 0.1
x 2,000 x 4 = 10 s.
This illustrates the kind of separations that may be realized on conventional capillary
columns (d,= 0.2 mm) if the selectivity has been optimized. The instrumental implications
of such rapid analysis will be addressed briefly in section 7.4.
301
7.2.3 Liquid chromatography (packed columns)
In LC, packed columns are used for practical separations. Eqn.(7.6) shows the speed of
analysis to be proportional to the required number of plates and to the square of the
(particle) diameter. We used this equation in chapter 4 to derive an expression for the
required retention time under conditions of constant flow rate and particle size ( t,,elf,d;
eqn.4.48). However, eqn.(7.6) suggests that the speed of analysis may always be increased
by decreasing the particle size. In other words, it suggests that the smallest available
particles should always be used in packed columns. This interpretation is too simple.
In LC (and to a lesser extent also in GC) the limiting factor is the maximum allowable
pressure drop over the column. As an example, we will look at the dimensions of packed
columns for LC when the operating pressure (Ap) is fixed at the maximum value allowed
by the instrumentation [704].
The pressure drop is given by Darcy’s law (eqn.7.1). Optimum flow rates on columns
with different particle sizes can be related by using the same reduced velocity (v. eqn.7.4)
for each column. Since the diffusion coefficient D, is a constant, we find for the ratio of
the linear velocities on two columns (1 and 2)
U,/U, = dp.2 Idp,, . (7.7)
If the reduced velocities on the two columns are equal, then the reduced plate height (h
eqn.7.3) may also be expected to be equal, and, hence, the column length varies according
to
L = H N,,,= h N,,,d p . (7.8)
Substitution of eqns.(7.2a), (7.4) and (7.8) in eqn.(7.1) yields
All parameters on the right-hand side of eqn.(7.9) are constants, and therefore the optimum
particle size for a given separation is proportional to the square root of the number of
plates required.
Substitution of eqn.(7.9) in eqn.(7.6) now yields
1000 qh2
t, = . N:, (1 + kd= /3 N i e (1 + kJ (7.10)
AP
where /3 is a constant.
An important conclusion from eqn.O.10) is that the analysis time is proportional to the
square of the required number of plates. Eqn.(7.10) also reveals that the analysis time may
be decreased by three factors:
1. A decrease in h. This is achieved by using good column packing techniques.
2. A decrease in q. This can be realized by using low viscosity solvents [705] or elevated
temperatures.
3. An increase in the pressure drop over the column.
302
From eqn.(7.9), we may easily obtain a good estimate for the optimum particle size for
HPLC separations. If the maximum operating pressure of the instrument is 400 bar (40
MPa), the viscosity of the eluent 1 CP ( l o w 3Pas; e.g. water at 20 "C)and the diffusion
coefficient m2/s, then we find around the optimum conditions (v=3,h= 2) from
eqm(7.9)
with d, in m. If we express d, in pm we find
d, = v(Nn,/6500) M K / 8 0
This shows that theoretically a particle size of about 1.2 pm should be used for a separation
that requires 10,000 plates. The associated column length is 2.5 cm (eqn.7.8) and the
analysis time for k = 1 about 20 s.
In practice, we operate LC systems still well above the optimum flow rate. We work well
below the maximum operating pressure of the pump (say 200 bar), at a higher reduced
velocity (e.g. v= 10) and with a correspondingly higher reduced plate height (h = 4). Under
these practical conditions we find:
Nne - 20.106
-- = 0.5.10'5
dg lo3 * - 4 . 1 0 .l o p 9
with d,, in m, and again with d p in pm
dp = l / ( N n e / 5 0 0 ) M K / 2 0
Therefore, under practical conditions a particle size of about 5 pm may be used to realize
10,000theoretical plates. The corresponding column length is 20 cm and the retention time
for k = 1 is 200 s.
Another practical consideration is that only a limited number of particle sizes and
column lengths is available. Figure 7.2 illustrates the practical situation using the same
estimates as for the latter calculation above. In a logarithmic plot, the required column
length as a function of the required number of plates forms a straight line with a slope of
1. Such straight lines are shown for four different particles sizes (3, 5, 10 and 20 pm)
assuming a reduced plate height of 4. The maximum allowable pressure limits the column
length. This is illustrated in figure 7.2 by the thin lines with slope 3/2. The two lines
represent pressure limits of 200 and 400 bar and are calculated for v= 10, D, = m2/s
and q= 1 mPa.s. Eqns.(7.8) and (7.9) readily allow lines to be constructed for other
conditions.
Three columns are indicated with heavy dots in figure 7.2. With these columns
separations may be performed that require up to 10,000 plates (under the conditions of
figure 7.2). The dimensions and characteristics of these columns are listed in table 7.1
(columns 1-111).
A possible fourth column in illustrated in figure 7.2 by an open circle. This 80 cm long
column packed with 10 pm particles is not an attractive column in practice, because of both
its length and the required analysis time. However, it should be noted that shorter columns
303
(L<40 cm) packed with 10 pm particles are of even less practical interest, because the
number of plates is less than that obtained on the 20 cm long column packed with 5 pm
particles (column 111). The only advantage of columns packed with larger particles is the
lower pressure drop (see table 7.1).
t
Id
Llcm
lo2.
10
lo2 lo3 loL lo5

N-
Figure 7.2 Example of the relationship between the required number of plates and the required
column length (h = 4) for four different particle sizes in HPLC. Thin lines indicate the maximum
column lengths for pressure drops of 200 and 400 bar, with v= 10, D,= m2/s and q= 1 mPa.s.
Heavy dots indicate possible columns for separations requiring up to 10,000 plates. Open circle
represents a column with 20,000 plates.
Table 7.1:
Example of a set of columns (1-111) that may be used to realize separations in HPLC
requiring up to 10,000 theoretical plates. Column IVmay be used to extend the set to 20,000
plates. Conditions are given in figure 7.2.
I 3 3 2,500 111 9
I1 5 3 4,200 185 15
304
The set of columns in table 7.1 is a reasonable one for practical purposes. Reasonable
assumptions have been made both for h and for v. However, the pressure drop over the
column may vary quite considerably, depending on the viscosity of the eluent. A viscosity
corresponding to that of water was used for the calculations of figure 7.2 and table 7.1,
but in the practice of RPLC a viscosity that is double that value may be observed for some
mobile phases (e.g. methanol-water 60-40 [705]). For this reason, columns have been
selected for which the pressure drop is below 200 bar under the conditions of table 7.2.
7.2.4 Summary
The following conclusions may be formulated after the discussion in this section.
1. In GC, open (capillary) columns provide faster analyses and allow larger numbers of
plates than do packed columns. However, in some cases packed columns may still have
prevailing advantages. In LC, open columns cannot yet be usedforpractical separations.
2. GCseparations for which the selectivity has been optimized may often be performed very
rapidly on short capillary columns with a conventional diameter (0.2 mm).
3. “Narrow-bore”capillary GC columns will lead to shorter analysis times, but impose
stringent demands on the instrumentation.
4. “Wide-bore” capillary GC columns allow easy operation and have a large sample
capacity. However, they give rise to relatively long analysis times.
5. The optimum particle size for packed columns in HPLC varies with the (square root of
the) number of plates required for the separation.
6. The time of analysis in HPLC may be reduced by using well-packed columns, low viscosity
eluents and high column pressure drops.
7. A small numbpr (e.g. 3) of HPLC columns allows the realization of separations that
require up to 10,000plates under conditions thatform a reasonable compromise between
theory and practice. Relatively large particles (dp> 10 p m ) are of limited practical
importance.
7.3 SENSITIVITY OPTIMIZATION
An important aspect of the chromatographic process is the sensitivity of the detection.

For most common detectors, the recorded signal is directly proportional to the concentra-
tion of the solute in the effluent from the column*. We have seen in chapter 1 (eqn.l.15)
that the observed peak height (h,) can be related to the peak area (A) by
ho= A / c T ~ (7.1 1)
where CT is the standard deviation of a Gaussian peak.
We may express the peak height in concentration units:
ho = cmax (7.12)
* For some detectors (e.g. flame ionization, FID and mass spectroscopy, MS)the detected signal is
proportional to the mass flow of the solute entering the detector, a quantity which equals the product
of the concentration and the volumetric flow rate. Therefore, the detector sensitivity is still directly
related to the solute concentration.
305
where the subscript max indicates the concentration at the peak maximum, and the
standard deviation in volume units (CT= 0,). In that case the area (A) should be expressed
in units of mass (concentration x volume). The law of mass conservation prescribes that
the mass of solute leaving the column should equal the mass of solute entering the column.
The latter can easily be expressed in terms of the concentration of the solute in the sample
solution (ci) and the injected volume ( Vinj):
A = ci Fnj (7.13)
and with eqn~(7.11)and (7.12)
(7.14)
Using the common equations for the plate count and the retention volume we now find
c,,,
- civinj fi
- -.-
G VR
- ci vinj -\rN
--.-.- (7.15)
I&v, ( l + k J
Eqn.(7.15) is the key equation for the optimization of chromatographic sensitivity.
Naturally, the peak height is proportional to the concentration of the solute in the sample
and to the volume of the injected sample. However, this proportionality holds over a
limited range and we cannot increase these two quantities indefinitely without having to
sacrifice another vital characteristic of the system, the linearity of detection. The
proportionality between c,, and the product ciVinjends when N may no longer be
considered as a constant. Consequently, the aforementioned product may be increased
until the plate count starts to be affected.
Injection of large volumes
A series of “tricks” has been devised for the injection of large volumes of samples, all
of which aim at increasing Vinj without affecting N.
In GC the injection may take place at a temperature that is lower than that of the column
oven. The solute bands will be concentrated in a small volume and may be brought into
the column by a subsequent heating of the cold zone. If this zone is part of the column
itself we talk about “cold (on-column) injection”, if it is part of a separate injector unit
we talk about “cold trap” injection. A similar “band compression” effect may be achieved
in a different way by leaving the first part of the capillary column “uncoated” (i.e. no
stationary phase present). The solute band will then be compressed at the point where the
stationary phase starts to be present in the column. This band compression technique is
usually referred to by the unfortunate term “retention gap” [706].
In LC the solute bands may be concentrated on a pre-column, which is eluted with a
weak eluent (low eluotropic strength). A simple way in which the effective injection volume
may be reduced considerably is by dissolving or diluting the sample in a solvent that is
much weaker than the eluent [707]. This has the effect that the initial capacity factor during
306
the injection of the sample is high, so that the solute bands are effectively compressed at
the top of the column (compare “cold” injection in GC).
Effect of column efficiency
Eqn.(7.15) suggests that the sensitivity (cmaX)increases with increasing plate count (N).
However, this is only true if all other factors, in particular V,, are kept constant. A better
way to look at the effect of the column efficiency on the sensitivity is to consider the ratio
(7.16)
where L is the column length, H the plate height, d, the column diameter and E the column
porosity ( ~ = for
l open columns). The implications of eqn.(7.16) become clear if we
introduce the reduced plate height (h).
First for a packed column, where
h = H/d, (7.3a)
and therefore
- -4 1 1 1
---.-.-.- (7.17)
V, nsdh d,‘ d L ddp
Eqn.(7.17) shows that an increase in the column efficiency by way of a decrease in the
reduced plate height (h) has a positive effect on the sensitivity. Therefore, well-packed
columns should be used at (or just above) the optimum flow rate*. If that is the case, then
h may be considered as a constant (2 < h < 3), i.e. considered to be independent of the
column diameter and the particle size.
Eqn.(7.17) shows that an increase in the column efficiency simply by increasing the
column length ( L increases with h constant) has an adverse effect on the sensitivity.
For open columns the reduced plate height may be defined as
h = H/d,, (7.3b)
where, at the optimum flow rate h z 0.7. With E= 1 we find

G -4 1 1
-_-.-.- (7.18)
V, ndh d;l2 dL‘
Again, the sensitivity is enhanced by working at the optimum flow rate (lowest value for
h) and decreased by increasing the column length.
Effect of column diameter
For packed columns eqn.(7.17) suggests a quadratic increase of the sensitivity with a
decreasing column diameter (dJ This is a cause of much confusion, as it suggests that
* The optimum flow rate is the flow rate at which the plate height (H and h) is minimal.
307
columns with a small diameter (“narrow-bore’’or “micro-bore’’ columns) show superior
sensitivity. Generally speaking, this statement is incorrect. This becomes evident if we
return to eqn.(7.15). A combination of this equation and eqn.(7.17) yields:
cmaX a ci Vinj / d: . (7.19)
Therefore, narrow-bore columns yield a much higher sensitivity if, but only if, the amount
of solute injected is kept constant. The amount of stationary phase in a packed column
is proportional to the square of the column diameter (keeping other factors constant).
If we define the reduced mass of the solute injected in the column ( Q,)as
(7.20)
then we find
(7.21)
and
(7.22)
Consequently, if Q, is kept constant, then the sensitivity is independent of the diameter

of a packed column.
The following may serve as an example to illustrate the point. We may inject 5 y1 of a
sample solution into a column with an inner diameter of 5 mm. If we inject the same
amount of sample into a column with a diameter of 1 mm, then (eqn.7.19) the sensitivity
would be increased by a factor of 25. However, the sample loading ( Q,)would be increased
by the same factor and the result may be increased peak broadening, necessarily combined
with loss of detector linearity.
If, on the other hand, we keep the value of Qs constant, then only 0.2 pl of the sample
should be injected on the smaller column. This would lead to the same sensitivity as
obtained on the 5 mm column. In fact, provided that the linear velocity of the mobile phase
and the sensitivity of the detector (defined as the observed signal divided by the
concentration of the solute) remain constant, identical chromatograms may be obtained.
In the case in which no more than 0.2 yl of the sample solution is available for analysis,
the narrow-bore column will give rise to a larger sensitivity. in other words, reduction of
the diameter of a packed column leads to an increased “mass sensitivity” in situations in
which the amount of sample available is the limiting factor.
We may summarize the two real advantages of narrow-bore columns over conventional
columns as follows:
1. Increased “mass sensitivity” in situations in which a limited amount of sample is
available.
2. Decreased consumption of mobile and stationary phase.
For open columns we find from a combination of eqm(7.15) and (7.18) that
cmaX a ci ynj/ d:‘*. (7.23)
308
In open columns the volume of the stationary phase ( Vs)will be proportional to the column
diameter and to the thickness of the stationary phase film (ds). Thus, we find with
eqn.(7.20)
C, ‘inj 0~ Q , , dc ds (7.24)
and hence
According to eqn(7.25) the sensitivity in capillary chromatography increases with

decreasing column diameter.
For several reasons, the drastic effect predicted by eqm(7.25) is not usually observed.
In the first place, the maximum allowable injection volume may not only be determined
by the amount of stationary phase, but also by the amount of mobile phase present in the
column (or, to be precise, the volume of mobile phase that occupies one plate in the
column)*.
In the second place, the film thickness (d,) tends to be increased in the practice of GC
if the column diameter is increased (“thick-film wide-bore” capillary columns), which
leads at constant Qs to an increase in sensitivity (eqn.7.25).
Most importantly, however, detectors which show a signal that is only proportional to
the Concentration and independent of the flow rate (so-called “concentration-sensitive
detectors”) are hardly used in contemporary GC.
Most of today’s popular detectors are so-called “mass-flow sensitive detectors”, i.e. the
recorded signal (h,) is proportional to the mass of solute passing through the detector per
unit time, in other words proportional to the product of the solute concentration (cmax)
and the volumetric flow rate (0.Therefore, we find for “mass-flow sensitive detectors”
with eqm(7.25):
(7.26)
where h, is the observed peak height.

Consequentiy, if “mass-flow sensitive detectors” are used, a reduction of the diameter
of capillary columns will give rise to a decrease in sensitivity. This effect is enhanced by
the tendency to use thicker films of stationary phase (larger d 2 in columns with larger
diameters. The limited sensitivity of the detection is a disadvantage of the use of
“narrow-bore” capillary columns in GC.
Effect of particle size
If we reduce the size of the particles in a packed column, then the column length may
be reduced by the same factor in order to keep the number of plates constant (i.e. Lcc dp;
* Especially in open tubular liquid chromatography(OTLC) the solubility of the sample components
in the mobile phase rather than in the stationary phase may be the limiting factor.
see eqn.7.8). Thus, according to eqm(7.15) and (7.17)
ciy”j Ciyinj
Cmax cc -oc - (7.27)
dLVd, dp
Eqn.(7.27) shows that the sensitivity may be increased by reducing the particle diameter,
if the sample size is kept constant. A constant sample size will correspond to the same
reduced sample size (QJ if the specific surface area is constant. Again, the advantage may
disappear if the solubility in the mobile phase is the limiting factor, because the volume
of mobile phase that occupies one plate decreases proportionaly with the particle size.
In LC practice, we do observe an increase in the sensitivity for columns packed with
small (3 pm) particles (“FAST LC” or “High speed LC columns) in comparison to
conventional columns (5 pm or 10 pm particles).
Summary
1. The sensitivity (observedpeak height) in a chromatographic system can be increased by

the injection of large volumes of samples, as long as this can be done without a decrease
in the observed number of peaks.
2. The sensitivity is increased by using good columns (small values for h) operated at the
optimumflow rate. The sensitivity decreases upon increasing the column length.
3. Reducing the diameter of packed columns leads to enhanced sensitivity ifthe amount of
sample available is the limiting factor.
4. Reducing the diameter of capillary columns leads to an enhanced sensitivity if
“concentration-sensitive’’detectors are used, but to a reduced sensitivity in combination
with “mass-flowsensitive” detectors.
5. If the limiting factor for the maximum allowable sample size is the capacity of the
stationary phase rather than that of the mobile phase, then a reduction of the particle
diameter in a packed column will result in an increase in sensitivity.
7.4 INSTRUMENT OPTIMIZATION
In previous sections we have discussed the optimum dimensions of a column with

respect to efficiency, time of analysis and sensitivity. However, the resulting optimum
dimensions may not be practical. For example, we rejected the theoretically predicted
optimum particle size of 1.2 pm for an HPLC separation requiring 10,000 plates. After
changing the conditions, we arrived at a practical “optimum” particle size of 5 pm.
Although 1.2 pm is the true optimum value, particles of this size cannot (yet) be used.
In this section we will identify other constraints imposed by the instrumentation on the
dimensions of the column. Two factors need to be considered in this respect:
1. the extra-column contributions to the peak broadening, and
2. the time constant of the detection.
Extra-column dispersion
In chapter 1 (section 1.4) we have expressed the width of a (Gaussian) chromatographic

peak in terms of the standard deviation 0. Using the additivity of variances, we may
310
interpret the observed standard deviation (a& as the result of two independent contribu-
tions:
.’,= 4 4- 4,- (7.28)
where 0,is the contribution from dispersive processes in the column and a,, the sum of
all other contributions to the peak width (“extra-column dispersion”).
In general, every piece of instrumentation from the injector upto and including the
detector will contribute to oe..Thus, contributions are included from
1. the injector,
2. the connection between the injector and the column,
3. the connection between the column and the detector, and
4. the detector.
The injector may contribute to the extra-column dispersion in several ways. Ideally, the
sample is injected “instantaneously” onto the column as is illustrated in figure 7.3a. This
ideal situation may only be approached in practice if one of the “band compression”
t- t-
ic)
-t -t
Figure 7.3: Illustration of possible injection profiles. (a) ideal profile (theoretical); (b) ideal situation
(practical): (c) practical profile; (d) practical profile with tail “cut off‘.
31 1
techniques described in section 7.3 is used. Otherwise, it is more realistic to consider the
“block-shaped injection profile shown in figure 7.3b as the ideal situation. In this case
the sample is injected during a time tinj and theconcentration of any solute (cinj)is constant
during this time. tinj is related to the injection volume by
tinj = qnj/ F, (7.29)
where F is the volumetric flow rate.

The less ideal situation illustrated in figure 7 . 3 is
~ a typical injection profile in a practical
situation, although the extent of “tailing” and the characteristic delay time (rinj;see figure
7.3~)will vary for different injection methods and for different forms of chromatography.
An injection profile corresponding to figure 7 . 3 ~can only be tolerated if the contribution
of rinjto the width and to the “non-symmetry” or “tailing” of the chromatographic peaks
can be neglected (see also the discussion on the detection time constant below). If this is
not the case, then the “tail” of the injection profile needs to be cut off (e.g. by switching
the injection valve). The resulting profile is illustrated in figure 7.3d. A disadvantage of
such a profile is that the quantitative accuracy of the chromatographic method is reduced
if the fraction of the total sample that is cut off in the tail is not completely reproducible.
The connections of the column to the injector and to the detector are critical for the
performance of the chromatographic system. They become the more so if the dispersion
of the column itself is decreased (eqn.7.28). All volumes outside the column need to be kept
to a minimum, and those connections in particular that involve diameter changes (e.g. to
and from the column in HPLC) may give significant contributions to sex.
Many detection principles require a finite volume of eluent. For example, a UV
absorption detector yields a signal that is directly proportional to the optical pathlength
(Beer’s law, see eqn.5.21). The volume of the detector flow cell is usually well-defined and
its contribution to a, and hence its effects on the observed dispersion a,, can be discussed
in quantitative terms (see section 7.4.2).
We may assume as a general rule that the extra-column dispersion should not lead to
an increase in the observed peak width of more than 10%. In other words
a, = 1.1 a,. (7.30)
With eqn(7.28) this yields
(7.31)
so that
a,, = v0.21 a, = 0.46 a, (7.32)
Therefore, as a rule of thumb, we may say [707] that a,, should be less than half the value
of a,:
a,, < a y 2 . (7.33)
Eqr~(7.33)can be used as a guideline for the design of chromatographic instruments.
312
Detection time constant
The time constant* ofthe detection is the combined effect of the detector (“detector time
constant”) and the data handling or recorder system. The time constant of the detector may
be partly due to the fundamental kinetics of the detection (e.g. in polarographic detection),
but is usually determined by the amplifier and other electronic components.
The time constant (T)affects both the height and the width of the observed peaks. It can
be shown [708,709] that (for .r<O.2 0,) the observed standard deviation in time units (or)
increases with 7 according to
where the subscript t,c indicates that the column dispersion should also be expressed in
time units. If we again accept a 10°/o increase in the observed peak width (eqn.7.30), then
7 < or,/lO. (7.35)
Eqn.(7.35) should be a guideline not only for the design of chromatographic systems,
but also for the selection of peripherals such as recorders. It can easily be seen that
eqn(7.35) imposes fairly stringent demands. For example, if we generate 10,000 plates in
100 seconds on an HPLC column of conventional dimensions (column I11 in table 7.1),
then or.,=I s and should be 0.1 s or less. Regular recorders have time constants of 0.5 s
or sometimes even 1 s. It is not easy to find recorders which are compatible even with the
conventional LC columns.
The situation is certainly not more favourable in GC, where 80,000 plates may be
generated in 100 s with a capillary column of conventional diameter (0.2 mm). This
corresponds to or,,% 0.4 s and therefore the time constant of the detection system should
be less than 40 ms.
This necessitates the abolition of recorders as the routine practical tool for the
representation of chromatograms. Since analog devices of adequate speed are expensive,
digitization of the data is the logical solution to the problem. In that case, we may assume
as a rule of thumb that at least 20 data points are needed to describe a chromatographic
peak. Using 4or., as a measure of the peak width in time units (seconds), this implies a
sampling frequency cf; expressed in Hz) of
f = 5/ar. (7.36)
According to the above examples this implies that for LC columns of conventional
diameters a sampling frequency of 5 Hz is required, whereas for contemporary capillary
G C the required frequencies are between 10 and 15 Hz.
One final comment regarding the time constant in chromatographic detection limits
concerns the noise level observed on the baseline. Large time constants serve as efficient
filters for (high frequency) noise. Therefore, if a great reduction of the time constant is not
* The time constant of an exponential decay corresponds to the time needed for the signal to decrease
to 37’/0 of its initial value.
313
to result in much increased detection limits, other methods should be used to keep the noise
level down. Because of the trend towards digitization, digital filters seem to be needed to
meet these requirements.
7.4.1 Gas chromatography (open columns)
For the dispersion in volume units in open tubular chromatography we can write
(7.37)
Expressed in the number of plates we find
cr, = ( d 4 ) h (I + k ) d f f i (7.37a)
Clearly, the column dispersion in capillary chromatography is a very strong function of

the diameter of the column. If the column diameter is decreased, then the column
dispersion will decrease strongly and therefore, according to eqn.(7.33), the demands
imposed on the maximum allowable extra-column dispersion will become increasingly
severe.
Schutjes (ref. [702], p.45) has suggested “compensating” for a decrease in d , by an
increase in L. In this way, “narrow-bore’’ columns might be used to effectuate separations
that require very large numbers of plates. However, eqn.(7.37) reveals that in order to keep
the column dispersion constant, a reduction in the column diameter by a factor of 2 would
require an increase in the column length by a factor of V32 w 5.7. The efficiency (N)would
be increased by a factor of 11. Since at constant v the linear velocity is inversely
proportional to the column diameter, the analysis time would only be increased by a factor
of 2.8. However, the pressure drop over the column (eqns. 7.1 and 7.2) would increase in
proportion to L, w and d;*, i.e. by a factor of 5.7 because of the increased column length,
a factor of 2 because of the increased velocity, and a factor of 4 because of the reduced
column diameter, resulting in a total increase by a factor of 45!
Although the above calculation is somewhat oversimplified because the effects of the
compressibility of the gas have been neglected, it serves to illustrate that a reduction of the
column diameter cannot be fully compensated by an increase in the column length to keep
the column dispersion constant. Therefore, when “narrow-bore’’capillary columns are to
be used in GC, the extra-column contribution to band broadening will need to be reduced.
In contrast, the use of “wide-bore’’ capillary columns allows a considerably larger

extra-column dispersion. Consequently,these columns may readily be used in instruments
that are compatible with conventional capillary columns. Moreover, they may be used
instead of packed columns to yield greatly increased plate numbers without the
requirement of major modifications to the instrument. One of the major reasons for the
popularity of “wide-bore’’ capillary columns may therefore be their role as intermediates
in the gradual replacement of packed columns by capillary columns in GC.
Table 7.2 lists the characteristic parameters of various columns for GC. The table shows
the typical characteristics of capillary columns of conventional diameters for two
314
situations: efficient columns for complex samples and short columns for separations for
which the selectivity has been optimized. These columns are compared in the table with
“wide-bore” and “narrow-bore” capillary columns. Typical figures for a packed column
have been added for comparison. The data for the “narrow-bore” columns have been
taken from experiments described in the literature [702]. All other data were calculated
using the equations given in this chapter and the conditions listed above the table.
It is clear from table 7.2 that in terms of extra-column dispersion a “wide-bore’’ capillary
column requires instrumentation similar to that used for the packed column. However, the
capillary column provides eight times as many plates (in a fifteen-fold analysis time).
Conventional capillary columns require a reduction in the dispersion by about an order
of magnitude, whereas “narrow-bore’’ columns require a further reduction by a factor of
about 100. This, combined with the high pressures required, puts “narrow-bore’’ columns
out of reach for current G C instruments.
Table 7.2:
Possible column dimensions for GC. Data calculated from eqns.(7.1), (7.2), (7.6), (7.33),
(7.359, (7.36) and (7.37) for all columns, except “narrow-bore” (literature data; [702]).
Conditions: h = 1.5, v = 5 (open columns); h=3, v = 3 (packed columns); q = 1 mPa.s
(hydrogen); D, = 10 - m2/s.
Parameter/ Packed Open

units
Conventional Wide Narrow
Long Short Long Short

(1) (2)
d/pm 200 250 250 500 65 55

L/m 3 30 3 30 95 22
N / 103 5 80 8 40 1,000 35
Ap/bar 1 3 0.3 0.4 19 10
D”41 1800(3) 170 50 1000 0.2 0.3

Dex/P1 900 (3) 80 25 500 0.1 0.2
t,/s 20 150 15 300 1800 2

a,/ms(4) 280 500 170 1500 1800 10
dms(4) 28 50 17 150 180 1
f/Hz 15 10 30 3 3 500
N,,JI 03 (5) 1.5 7.5 2.25 4.5 54 10
(1) Experimental data from ref. [702], appendix 4, carrier gas: hydrogen.
(2) Experimental data from ref. [702], pp.59-60, carrier gas: helium.
(3) for d,=5 mm and ~ = 0 . 6 .
(4) for k = 0.
( 5 ) Number of datapoints recorded for a digitized isothermal chromatogram for 0 < k < 4 .
315
For each of the columns in table 7.2 the hold-up time (to)can readily be calculated from
eqn.(7.6) using k = 0. The standard deviation in time units follows from to after division
by I6(eqn.l.16). Eqns.(7.35) and (7.36) then provide the maximum allowable detection
time constant (T), the required sample frequency for digital data handling u> and the
required number of datapoints ( Ndat)for recording a chromatogram (0 < k < 4). All these
characteristics are shown for the different columns in the bottom part of table 7.2.
Clearly, both packed column GC and conventional capillary GC already put serious
constraints on the maximum allowable time constants. Ironically, the easiest separations,
requiring the lowest numbers of theoretical plates and hence requiring short columns, are
the most difficult ones to perform in an optimum way. The generation of lo6 plates (in
about an hour) allows the highest value for z (180 ms) in the entire table!
Table 7.2 also lists the required data storage for a typical isothermal chromatogram
(0 < k < 4) recorded on each of the columns. It is seen that conventional capillary columns
already require a five times larger facility than do conventional packed columns. The use
of “long” narrow-bore columns would lead to a further increase in the number of
datapoints by about an order of magnitude.
An increase in the number of datapoints recorded per chromatogram also leads to
increased demands with respect to data handling facilities and computing time. For
instance, 54,000 datapoints require about 100k computer memory (double precision).
The increased computation demands form another obstruction for the application of
narrow-bore capillary columns in routine GC.
7.4.2 Liquid chromatography (packed columns)
Because of the low diffusion coefficients in liquids, the particle size for packed columns
in LC needs to be very small (see section 7.2.1). For the same reason, all external volumes
and diameters need to be minimized. This may easily be understood if we express the
standard deviation in volume units (a,)in the parameters that represent the dimensions
of the column:
a =-=v, VdI+ k)
fi d(L/H)
(7.38)
For conventional HPLC columns with E= 0.6 and h = 4 we find for k = 0

a, = ( d 4 ) x 0.6 x 4 fida d , w 1.8 I6d: d, (7.39)
where a, is expressed in p1 if d , and d , are both expressed in mm.
As an example, a conventional HPLC column of 20 cm length packed with 5 pm (0.005
mm) particles yields 10,000 plates (column I11 in table 7.1). With the conventional column
inner diameter of 5 mm we find
a, = 1.8 x 100 x 25 x 0.005 w 23 pI .
Eqn.(7.33) shows that the total extra-column dispersion for a conventional HPLC system
should be less than about 12 pl.
316
The detector flow-cell, the contribution of which to CJ” is approximately equal to its
volume [707], represents a considerable and recognizablecontribution to the extra-column
band broadening. Typical conventional flow-cells have a volume of 8 pl, which is quite
substantial compared with the maximum allowable extra-column dispersion.
Table 7.3a lists the maximum allowable extra-column-dispersion for the first three
columns listed in table 7.1, using three different internal diameters. It is seen that the
contribution from the detector flow-cell (as well as other contributions) will have to be
reduced considerably if short columns packed with 3 pm particles are to be used. An even
larger reduction in the extra-column dispersion is required for the use of columns with a
reduced inner diameter.
Presently [707], we may be able to reduce the extra-column band broadening by
modifying conventional HPLC equipment to a total of 1 or 2 3.This implies that with these
modifications short “bulky” 3 pm columns or narrow (2 mm) 5 pm columns may be used.
Table 7.3:
a. Maximum allowable extra-column dispersion for the first three columns of table 7.1.
Calculated from eqn~(7.33)and (7.39) for three different column diameters.
Column N dP a, (max) / pl
(mm)
do= 5 d,=3 d,=l mm
I 2,500 0.003 3.4 0.5 0.14

I1 4,200 0.003 4.4 0.7 0.17
I11 10,000 0.005 11.3 1.8 0.5
b. Maximum allowable detection time constants for the first three columns of table
7.1. Calculated from eqm(l.16) and (7.35) for three different k values using the data
from table 7.1. Minimum required sampling frequencies for the same three columns
were calculated from eqn.(7.36) for k = 0. Number of recorded datapoints calculated
for the range 0 < k < 4.
Column N dP t0 T (max) / ms f Ndat

tPm) (9 (HZ) (x 103)
k=O k=l k=4
I 2,500 3 9 18 36 90 30 1.4
I1 4,200 3 15 23 46 116 20 1.5
111 10,000 5 100 loo 200 500 5 2.5
In practice, it may be useful to calculate the minimum required inner diameter for a
given column. This may easily be done by combining eqn.(7.33) with eqn.(7.39):
(7.40)
317
For instance, if the available instrumentation has an extra-column dispersion of 5 p1 and
we have opted for a 5 pm column of 10 cm to realize 5,000 plates, then we find from
eqn.(7.40)
v
I
5 d
x 3.8 m m .
dc ,' v(5.10-3)
Therefore, in this situation a column with an internal diameter of 4 mm or more should
be selected.
Table 7.3b shows the calculated maximum allowable detection time constants for the
first three columns of table 7.1 for three different values of the capacity factor, using the
values for N and to given in this table.
It appears from table 7.3b that modern HPLC columns impose very stringent demands
on the detection (and recording) system. Typical time constants of current LC detectors
are in the range of 0.3 to 0.5 s [710], which is not even sufficient to allow the use of a 20
cm, 5 pm column (column 111 in table 7.3b).
Therefore, a reduction in the time constant of current detection systems, without the
accompanying effect of a great increase in the noise level, should have at least the same
priority as the reduction of the extra-column dispersion in the design of future HPLC
systems.
7.4.3 Summary
In this section we have derived rules of thumbfor the maximum allowable extra-column
dispersion and detection time constant and for the minimum required sample frequency
for digital data handling.
In GC the use of "wide-bore" capillary columns allows the use of instruments designed
to accommodate packed columns (in terms of extra-column dispersion). For capillary
columns of conventional diameter a reduction of the extra-column dispersion by a factor
of 10, and for narrow bore columns a reduction by a factor of 1000, are required.
The extra-column dispersion in HPLC should be further reduced to allow the use of
columns packed with very small particles and/or columns with small internal diameters.
Both in GC and in LC the detection time constant needs to be reduced, even for the
application of conventional columns. A great reduction is required to follow modern
developments in column technology and the use of digital data handling appears to be
unavoidable.
REFERENCES
701. G.Guiochon, AnaLChem. 50 (1978) 1812.

702. C.P.M.Schutjes, Ph.D. Thesis, Eindhoven Technical University, 1983.
703. J.H.Knox and M.T.Gilbert, J.Chromatogr. 186 (1979) 405.
704. J.H.Knox and M.Saleem, J.Chrornatogr.Sci. 7 (1969) 614.
705. %.van der Wal, Chromatographiau ) (1985) 274.
706. K.Grob, Jr., J.Chromatogr. 237 (1982) 15.
318
707. P.J.Naish, D.P.Goulder and C.V.Perkins, Chromatographia 20 (1985) 335.
708. L.J.Schmauch, AnaLChem. 31 (1959) 225.
709. G.McWilliam and H.C.Bolton, AnaLChem. 41 (1969) 1755.
710. G.Guiochon in Cs.Horvath (ed.), High Performance Liquid Chromatography; Ad-
vances and Perspectives, V01.2, Academic Press, New York, 1980, pp.1-56.
319
SYMBOLS AND ABBREVIATIONS
Use of symbols is limited to certain section of the book when indicated.
Symbol Description Sections Introduced

a activity 3.2 eqn.(3.24)
a solute parameter 2.3.2 eqn.(2.5)
a constant various
b gradient steepness parameter eqn.(4.68)
b optical pathlength 5.6 eqn.(5.21)
b solute parameter 2.3.2 eqiL(2.5)
b constant various
b' constant various
c concentration -
c cohesive energy density 2.3.1 eqn42.1)
c solute parameter 2.3.2 eqm(2.5)
c constant various
C continuous criterion chapter 4 table 4.7
C continuous parameters 3.5
c' constant eqn.{4.30a)
d confidence range 5.5.2 eqn.(5.19)
d diameter chapter 7 eqn.Q.3)
d solute parameter 2.3.2 eqn.(2.5)
d constant various
dc column diameter table 4.2
dP particle diameter -
d., film thickness table 4.2
d discrete parameters 3.5
e effect of variable 5.4.1 eqn.( 5.2)
e solute parameter 2.3.2 eqn.(2.5)
f experimental value 5.4,1 eqn45.2)
f measure for peak separation eqn.(4.3)
fm measure for peak separation eqn.(4.4)
f fixed design table 5.6
g measure for peak separation eqn.(4.3)
g weighting factor 4.6.1 eqn.(4.58)
g* measure for peak separation eqn.(4.4)
h partial molar enthalpy chapter 3 eqq.(3.7)
h peak height ,1.4 eqn.(l.l5)
h reduced plate height chapter 7 eqn.(7.3)
ho peak height chapter. 7 eqn.(7.11)
1 iterative design table 5.6
k capacity factor eqn.( 1.5)
k, capacity factor under gradient conditions eqtk(4.67)
321
kin isocratic capacity factor eqn.( 5.5)

kobs observed capacity factor eqn.(3.70b)
ko
- extrapoiated capacity factor eqn.(3.45)
k average capacity factor eqn.(l.22)
1 number of levels 5.4.1 eqn.(5.1)
n noise level 4.2.1 eqn.(4.12)
n number of moles chapter 3 eqn(3.6)
n number various
nb molecular size of solvent 3.2.3 eqn.(3.73)
"e chain length figure3.13
"e number of experiments 5.4.1 eqn.(5.1)
nP number of parameters 5.4.1 eqn.(5.1 )
nP
peak capacity 1.6 eqn.(l.25)
n correction for number of peaks required table 4.8
nf critical chainlength figure3.13
"a number of peakpairs eqn.(5.11)
P number of relevant peaks 4.6.1 eqn.(4.54)
P pressure -
P regression parameter 3.2.2.1 eqn.( 3.46)
P vapour pressure 3.1.1 eqn.(3.3)
'P critical pressure 3.4 -
Po pure component vapour pressure eqn43.4)
4 quantity -
9 regression parameter 3.2.2.1 eqn.(3.46)
r normalized resolution product eqn(4.19)
rA dissociation ratio 3.2.2.1 eqn.(3.62)
rB buffer dissociation ratio 3.2.2.1 eqa(3.68)
TT heating rate eqn.(4.66)
r* calibrated normalized resolution product eqn.(4.21)
S partial molar entropy chapter 3 eqn43.7)
S stationary phase parameter 2.3.2 eqn.(2.5)
S stepwise parameters 3.5
f time -
5 isothermal retention time 6.3.1.2 eqm(6.3)
t ??I hoid-up W eqn.( 1.3)
tG gradient duration time eqn.(4.68)
tR retention time eqn.( 1.2)
t0 hold-up time eqn.(l.3)
t time correction required table 4.8
t i net retention time eqn.( 1.7)
tha threshold criterion, arbitrary boundaries table 4.7
thf threshold criterion, fixed boundaries table 4.7
U mobile phase linear velocity eqn.( 1.I)
U stationary phase parameter 2.3.2 eqn.( 2.5)
322
V measure for peak separation eqn.(4.5)

V migration speed 1.2.1 eqn.(l.l)
V molar volume -
V vector in parameter space figure5.11
W peakwidth -
ws weight of stationary phase eqm(3.11)
w1/2 peakwidth at half height eqn.( 1.16a)
X mole fraction -
X optimization variable 5.1.1 figure 5.3
X stationary phase parameter 2.3.2 eqn(2.5)
X threshold value eq~(4.23)
'd proton donor parameter eqn(2.16)
xf? proton acceptor parameter eq~(2.15)
X" strong dipole parameter eqn.(2.17)
x useless criterion table 4.7
X' shifted composition 5.5.2 eq~(5.18)
Y optimization variable 5.1.1 figure 5.3
Y stationary phase parameter 2.3.2 eqn(2.5)
Z charge of solute ion 3.2.2.1 eqn.(3.71)
Z stationary phase parameter 2.3.2 eqn.(2.5)
A absorption 5.6 eqn.(5.21)
A adsorption area 3.2.3 eq~(3.72)
A curvature coefficient eqm(3.38)
A peak area 1.4 eqn.(l .15)
A constant various
A, adsorption surface area 3.2.1 eqm(3.17)
A' constant various
ACN acetonitrile
B gradient slope eqn.(5.6)
B constant various
Bo specific permeability coefficient eqn.(7.1)
B' constant various
C optimization criterion chapter 4 -
C constant various
C capacity parameters 3.5
C' constant various
CBP chemically bonded phase
D constant various
D distribution coefficient 3.3 eqm(3.76)
E cohesive energy 2.3.1 eqn.(2.1)
F flowrate eqn.(4.66)
FO fractional overlap criterion eqa(4.13)
GC gas chromatography
GLC gas-liquid chromatography
323
GPC gel permeation chromatograph

GSC gas-solid chromatography
H Henry's adsorption coefficient 3.1.2 eqn(l.17)
I ionic strength 3.2.2.1 eqm(3.71)
I retention index eq~(2.3)
IEC ion-exchange chromatography
IPC ion-pairiag chromatography
K distribution coefficient eqn.( 1.9)
K" n-alkane distribution coeffient 2.3.3 eqa(2.13)
Ka acid dissociation constant 3.2.2.1 eqn.(3.61)
'a adsorption coefficient 3.1.2,3.2.3) eqr~(3.16)
Kb base dissociation constant 3.2.2.1 eq~(3.65)
K* distribution coefficient (GC) 2.3.3 eqa(2.11)
KO n-octane distribution coefficient 2.3.3 eqa(2.13)
Kth thermodynamic distribution coefficient eqm(3.24)
corrected distribution coefficient 2.3.3 eqa(2.11)
K, corrected distribution coefficient 2.3.3 eqm(2.12)
L column length -
LBPC liquid-bonded phase chromatography
LC liquid chromatography
LLC liquid-liquid chromatography
LSC liquid-solid chromatography
MS stationary phase molecular weight eqm(3.2)
M mobile phase parameters 3.5
MeOH methanol
MtBE methyl t-butyl ether
MC methylene chloride
N number of plates eqn.(l.16)
Nb mole fraction of strong solvent 3.2.3 eqn(3.73)
number of data points -
Ndat
NP-IPC normal phase IPC
NPLC normal phase LC
ODS octadecyl silica
P peak valley ratio eqm(4.3)
prn median peak valley ratio eqm(4.4)
P" valley-to-top ratio eqn.(4.5)
P physical parameters 3.5
P' polarity parameter 2.3.3 eqn.(2.14)
PCA principai component analysis
PTH phenylthiohydantoin
Qs reduced sample size eqn.(7.20)
R gas constant eqn.(3.3)
'a absorbance ratio 5.6 eqa(5.21)
RS resolution eqn.( 1.f4)
324
RAT absorbance ratio 5.6 eqn(5.21)

RP-IPC reversed phase IPC
RPLC reversed phase LC
S' separation factor corrected number of plates eqn.(4.15a)
S separation factor eqm(4.15)
S slope (RPLC) eqn.(3.45)
surface area table 4.2 '
3
S
adsorption energy
Stationary phase parameters
3.2.3
3.5.
eqn.(3.72)
SEC size exclusion chromatography

SFC supercritical fluid chromatography
T temperature -
Tc compensation temperature 3.2.2.1 eqn.(3.58)
Tc critical temperature 3.4 eqn.(3.58)
To recommended isothermal temperature eqn.(5.4)
Tr isothermal temperature eqn.(4.66)
T thermodynamic parameters 3.5
THF tetrahydrofuran
TLC thin layer chromatography
V volume -
va volume of adsorbed stationary phase 3.2.3 eqm(3.72)
retention volume gradient conditions chapter 6 eqn.( 6.7)
vg
vg specific retention volume chapter 3 eqn.(3.11)
retention volume -
VR
hold-up volume -
vo
VR net retention volume -
Z compressibility coefficient qn43.3)
a constant eqn.(3.59) eqn.( 3.59)

a relative retention (selectivity) eqn.(l.ll)
a adsorbent activity 3.2.3 eqn.(3.72)
B constant eq~(3.59) eqn.(3.59)
B constant 4.4.3 eqn.(4.38)
7 activity coefficient eqn.(3.4)
Y constant eqn.(3.59) eqn.(3.59)
6 solubility parameter eqm(2.1)
6 allowed uncertainty 5.5.2 eqm(5.19)
& column porosity table 4.2
8 eluotropic strength (LSC) eqm(3.72)
tl viscosity eqn.(7.1)
K gradient shape parameter eqn46.6)
L wavelength eqn.(5.27)
c1 thermodynamic potential 3.2 -
V reduced linear velocity eqm(7.4)
325
density eqn.(3.2)
standard deviation eqn.(l .I 6)
observed standard deviation eqn.(7.28)
time constant chapter 7 eqn.(7.34)
weighting factor 4.4.2 eqn.(4.32)
total overlap criterion 4.3.4 eqn.(4.27)
volume fraction -
isocratic volume fraction figure5.13
threshold absorption 5.6 eqm(5.22)
pressure drop eqm(7.1)
eluotropic strength (RPLC) eqn.(3.52)
constant 5.5.2 eqn.( 5.1 8)
D, diffusion coefficient eqn.(7.4)

V selectivity of phase system eqn.(3.32)
SUBSCRIPTS
Symbol Description Sections
a acid 2.3.1
b base 2.3.1
C column
C concentration chapter 3
C number of carbon atoms in chain 3.2.2
d dispersion 2.3.1
d proton donor (dioxane) 2.3.3
e proton acceptor (ethanol) 2.3.3
ex extra-column
f final
f,d conditions of constant flow and diameter
g gradient
g weighted 4.6.1
I solute
1 initial 6.2.1
i+ 1 peak following i
i- 1 peak preceding i
ind induction 2.3.1
inj injection
j solute
j modifier eqn.(3.34)
ji pair of solutes
m mobile phase
max maximum
326
min minimum
n preceding n-alkane 2.3.2
n strong dipole (nitromethane) 2.3.3
n corrected for number of peaks chapter 4
nP non-polar 2.3.2
n+l following n-alkane 2.3.2
n+l peak following n
n-I peak preceding n
ne required
nt corrected for number of peaks and analysis time
0 orientation 2.3.1
0% organic phase
P constant pressure conditions
P polar 2.3.2
P programmed elution 4.6.2
ref reference
t corrected for analysis time
V evaporation 3.1.1
V volume units
C number of carbon atoms in chain

E excess
IE ion-exchange
Me methanol
R retention
T total 2.3.1
T tetrahydrofuran (THF)
W water
a first peak
w last peak
Underlining of symbols indicates relevant peaks (section 4.6).
SUPERSCRIPTS
aq aqueous
org organic
03 infinite dilution
G gas phase
0 standard state 3.1.1
Lines above symbols indicate average or median values.
327
AUTHOR INDEX
Numbers in this index refer to references in the text.
Abbott, S. 204 ChuriEek, J. 354,357,377, 526,531, 532,
d'Agostino, G. 537, 579 618,623,624,628,632,633,634
Allen, A.C. 567 Clark, B.J. 588
Antle, P.E. 572 Clerc, J.T. 591
Asche, W. 388 Coates, V.J. (ed.)407
Bajema, B.L. 413, 561 Cohen, K.A. 430,610
BaYke, S.T. 334 Cohen, M.J. 348
Barker, J.A. 307 Colin H. 325, 351, 555, 577, 628
Barth, H.G. 318 Conlon, R.D. 575
Bartha, A. 381, 386 Constanzo, S.J. 549
Barton, A.F.M. 320 Cowie, C.E. 368
Bartb, V. 615 Cristophe, A.B. 404
Berendsen, G.E. 317, 319 Crombeen, J.P. 382
Berridge, J.C. 414, 505, 506, 508, 509, Crommen, J. 379
510,621 Culbreth, P.H. 620
Berry, V.V. 520, 543, 607 Davis, J.M. 105
Bidlingmeyer, B.A. 202 Davis, O.L. (ed.) 519
Billiet, H.A.H. 207, 209, 303,311, 314, Debets, H.J.G. 413, 561
321, 322, 323,324, 335, 336, 341, 373, Deelder, R.S. 345
374,383,386,409,410,424,502,504, Deming, S.N. 340, 417,421, 422, 503,
534, 535, 536, 576, 578, 582, 584, 590, 511,540,550,551,552,558,559,612
608,629 Dolan, J.W. 331, 337,428, 430, 528, 583,
Blass, W. 603 610,616,622
Board, R.D. 387 Doornbos, D.A. 327,413,423,425,512,
Bolton, H.C. 709 560, 561
Bosman, Th. 592 Driscoll, J.N. 586
Bounine, J.P. 555 Drouen, A.C.J.H. 336,383,405,409,410,
Bower, J.G. 503 424,502,504,576,578,582,584,590
Bower, K.D. 503 Edens, R. 551
Box, G.E.P. 515,518 Eksteen, R. 371
Bradley, M.P.T. 521 Elyashberg, M. 589
Brenner, N. (ed.), 606 Eon, C. 208,312,351
Brinkman, U.A.Th 332,533 Essers, R. 592
Bruins, C.H.P. 327,425,560, 561 Everett, D.H. 307
Budna, K.W. 406 Fagerson, I.S. (ed.) 407
Calfen, J.E. (ed.), 606 Fast, D.M. 620
Campbell, D.E. 339 Feibush, B. 348
Carr, P.W. 418, 619 Fell, A.F. 588
Castagnetta, L. 537, 579 Fransson, B. 379
Charikofsky, J.G. 573 Frei, R.W. 332,533,604
Chen, B.-K. 338 Galan, L.de 209,311,314,319,321, 322,
Chien, C.-F. 304 323, 324, 335, 336,341, 373, 374, 383,
Chu, C.H. 343 386,409,410,424,502,504,534,535,
329
536, 576, 578, 582, 584,590, 593,608, Hunter, W.G. 518
629 Issaq, H.J. 538, 554, 566, 567, 585, 626
Gant, J.R. 331,337,428,528,583,587, Jacques, C.H. 541
616,622 Jandera, P. 325, 354, 357, 377, 526, 531,
Giddings, J.C. 104,105,426,606 532, 577,618,623,624,628,632,633,
Gilbert, M.T. 366,703 634
Gillen, D. 521 Janderovh, M. 357
Glajch, J.L. 359, 360,361,362, 363,415, Janini, G.M.554
542,564,565 568,569,570,571,573, Jeeves, T.A. 514
611,627 Jefferies, T.M. 385
Gluckman, J.C. 573 Johansson, E. 516
Goewie, C.E. 332,533,604 Johansson, K. 516
Goldgaber, N. 626 Jones, P. 408, 546
Goldsmith, P.L. (ed.) 519 Kaiser, R. 402
Golkiewicz, W. 353, 522, 523 Kalashnikova, E.V. 309
Goulder, D.P. 707 Kaplar, L. 302
Grant, D.W. 614 Karger, B.L. 208, 312, 348
Gribov, L.A. 589 Karnicky, J. 204
Grillo, S.A. 430, 610 Kastelan-Macan, M.539
Grob, K., Jr., 706 Kateman, G. 203,592
Grob, L.R. 529,530,630,631 Kaur, B. 366
Grob, R.L. (ed.)525 Kenndler, E. 602
Guiochon, G. 102, 325,351,555,577, Kirkland, J.J. 103,201, 347, 349, 359,
701,710 362,401,415, 524, 542, 564, 565, 570,
Habgood, H.W. 427,605 601,611,627
Haddad, P.R. 368,376,425,582 Kiselev, A.V. 309
Hafkenscheid, T.L. 333 Klaessens, J. W.A. 203
Haky, J.E. 566 Klein, J. 310
Hamilton, P.B., 369 Klose, J.R. 566
Harbison, M.W.P. 305 Knoll, J.E. 412
Harris, W.E. 427 Knox, J.H. 365,366,380,703,704
Harris, W.E. 605 Kong, R.C. 559
Hearn, M.T.W. 344 Kopecni, M.M. 304
Heckenberg, A.L. 376 Kraak, J.C. 371,382
Hendrikx, L.H.M. 345 Krstulovic, A. 555
Hendrix, D.L. 551 Krull, I.S. 586
Hildebrand, J.H. 206,313 Kuwana, Th. (ed.) 306,420,501
Hollis, M.G. 614 Laird, G.R. 380
Hoogewijs, G. 574 Lankmayr, E.P. 406,580,58
Hooke, R. 514 Laub, R.J. 304,305,306,420,501,544,
Horvath, Cs. (ed.) 316,429,527,617, 710 545,553
Horvath, Cs. 316,326,329,339,342 Lauer, H.H. 387
HSU,A.-J. 553 Laurent, C.J.C.M. 341,364,373,374
Huber, J.F.K. 315,370,372,602 Lindberg, W.516
Hulpke, H. 603 Linsen, P. 371
Hunter, J.S. 518 Little, C.J. 604
330
Littlewood, A.B. 301 Phillips, G.S.G. 301
Lochmueller, C.H. 318 Pietnyk, D.J. 343
Locke, D.C. 308 Poile, A.F. 575
Lu Peichang, 330 Poppe, H. 356,382
Lu Xiaoming, 330 Poshkus, D.P. 309
Lurie, I.S. 567 Prausnitz, J.M. 206,313,389
Madden, S.J. 553 Price, D.T. 301
Majors, R.E. 318 Price, W.P. 551, 552
Marcus, Y. (ed.) 378 Purnell, J.H. 305,328, 545
Marinsky, J.A. (ed.) 378 Puttemans, M. 574
Markl, P. 3 15 Quarry, M.A. 529,530,609,630,631
Martire, D.E. 305 Rabel, F.M. 367
Massart, D.L. 416, 574 Rafel, J. 513
May, W.E. 548 Rajcsanyi, P. 302
McCann, M. 328 Randall, L.G. 392
McManigilI, D. 387 Reid, R.C. 389
McNitt, K.L. 566, 585, 626 Reijnen, J. 592
McReynolds, W.O.213 Riegner, K. 603
McWiIliam, G. 709 Riley, C.M. 385
Mead, R. 507 Rogers, L.B. 419
Melander, W.R. 316, 326, 329, 338, 342 Rohrschneider, L. 205, 210, 216
Midgett, M.R. 412 Rolink, H. 423, 512
Minor, J.M. 415, 542, 573 Rowland, M. 557
Minor, J.M. 573 Sachok, B. 422,558,559
Mitchell, F. 537, 579 Saleem, M. 704
Modin, R. 384 Sampson, E.J. 620
Molnar, I. 342 Sander, L.C. 548
Morgan, S.L. 417, 540, 541 Saunders, D.L. 358
Morrisey, E.C. 510 Sawicki, E. (ed.) 375
Mueller, H. 365 Sawicki, E. 375
Mulik J.D. (ed.) 375 Schill, G. 378, 379, 384
Muschik, G.M. 554, 566 Schill, R. 525
Naegli, P.R. 592 Schlabach, T. 204
Naish, P.J. 707 Schmauch, L.J. 708
Nelder, J.A. 507 Schneider, G.M. 391
Noebels, H.J. (ed.) 407 Schoenmakers, P.J. 207, 209, 303, 311,
Noyes, C.M. 547, 625 314,321,322,323,324,335, 390, 409,
Nyiri, W. 602 410, 504, 534, 535, 536, 576,608,629
O’Hare, M.J. 537, 579 Schupp, O.E. 403
Ober, S.S. 407 Schutjes, C.P.M. 702
Olacsi, I. 302 Schwartz, M. 586
Oreans, M. 602 Scott, H.P. 588
Otto, M. 346, 562, 563 Scott, R.L. 206,313
Pawlowska, M. 315 Shcherbakova, K.D. 309
Perkins, C.V. 707 Sherwood, T.K. 389
Perrone, P.R. 587 Smet, M.de 574
331
Smits, R. 416 Turina, S. 539
Snyder, L.R. 101, 103,201,208,214, 215, Turoff, M.L. 340,421, 550
312, 331,337,349, 350, 356, 359, 360, Unger, K.K. 365
361, 362,363,401,428,429, 524, 527, Vandeginste, B.G.M. 203,592
528,529,530,564,568,569,570,571, Vanroelen, C. 416
573,583,601,609,616,617,622,630, Venne, J.L.M. van de 345
63 1 Vigh, Gy. 381, 386
Sonewinski, E. 352,353,355 Wal, Sj.van der 204,370, 372, 556, 705
Spencer, W.A. 419 Walters, F.H. 511,612
Squire, K.M. 415, 542 Warren, F.V. 202
Stadalius, M.609 Wasen, U.van 391
Stahel, 0. 604 Watson, M.W. 418,619
Stan, H.-J. 613 Wegscheider, W. 346,406,562,563,580,
Steinbach, B. 613 581
Stranahan, J.J. 422, 558 Weiss, M.D. (ed.), 606
Supina, W.R. 212 Wellington, C.A. 328,408, 546
Svoboda, V. 411,517 Weyland, J.W. 327, 423,425, 512, 560,
Swaid, I. 391 561
Takacs, J. 302 Widdecke, H. 310
Tijssen, R. 207, 303,321,629 Williams, P.S.305
Tomlinson, E. 333,385 Wilson, K.B. 515
Tompkins, DJ. 604 Wise, S.A. 548
Toon, S. 557 Wittgenstein, E. (ed.)375
Trbojevic, M. 539
332
SUBJECT INDEX
Absorbance ratio - see: ratio recording -, see also: chemically bonded phases
Accuracy of predicted optimum, in itera- Buffer(s)
tive designs 226 -, dissociation ratio 71
Acid-base interactions 25 - 26 -, effect on retention in RPLC 70-71
Activity coefficient -, effect on retention in IPC 100
-, definition 38 Calibrated normalized resolution product
-, effect on retention 153
-, -, in GLC 38 -, correction for analysis time 157- 158
-, -, in LC 48 -, correction for number of peaks 158
Adsorbent activity, in LSC 76 -, definition 139
Adsorption area, of solute in LSC 76 -, for limited number of relevant peaks
Adsorption coefficient 162
-, definition 43 -, need for time correction factors 155
-, in LSC 76 -, use of solute weighting factors 164
Adsorption energy, of solute in LSC 76 Capacity factor
Adsorption isotherm, definition 4 -, definition 3
Alkaloid drugs - , effect on resolution 10- 14
-, retention in IEC 92-93 -, measurement 3
-, retention in I F C 190-191 -, optimum range 11 - 12,16- l7,62,
Alkyl chain length 192,253
-, effect on selectivity in RPLC 58 - 59 - , optimum value 1 1 - 12
-, see also: chain length -, use as retention parameter 37
Alumina 70,77,81 Capacity parameters, definition 105
-, use in GSC 45 Capillary columns
-, use in IEC 92-93 - see: open coiumns
- ,see also: polar adsorbents - see: column(s), capillary
Analysis program Carbon dioxide, as solvent for SFC 103
-, definition 253 Carbon stationary phase(s) 52, 70, 77,
- , possible shapes 253 -255 81 -82
Analysis time, effect on optimization cri- Chain length, of pairing ion in IPC 99
teria 136-137,146, 148-151 Chemically bonded phases (CBPs) 20,53,
Antioxidants, separation by programrned 56-59
solvent LC 278 -, perfluorinated 52, 74
Applications of SFC 103 -, polar 51,74-75
Background signal, in programmed sol- -, -, retention mechanism 75
vent LC 261 -, - ,stability 75
Baseline drift, in programmed temperatu- -, polymeric 57
re GC 259-260 -, see also: end-capping
Binary mixtures 60 -, see also: silica, reaction with silanes
Blank signal, in programmed solvent LC Chromatograph, schematic 1
261 Chromatographic methods
Boiling point separation 41 -, classification 20-21
Bonded phase chromatography 56 - 75 -,nomenclature 20 -21
333
-, selection 16, 21 -23 305 - 306
Chromatography, definition 1 Confidence ranges, in iterative designs
Cold injection in GC 306 225,221 -228
Column diameter Contamination capacity 55
-, effect on phase ratio 6 Continuous parameters, definition 109
- ,effect on sensitivity 307 Contour plot, of response surface 172
Column independent time factors Counterion, definition for IEC 82
151- 153 Counterion concentration
Column porosity, definition 6 -, effect on retention in IPC 100
Column(s) -, effect on retention in IEC 84-87
- ,capillary, for GC 300 -301 Counterion type, effect on retention in
-, -, detection sensitivity 309 IEC 87
-, -,evaluation 315 Critical band method 206 -209
-, -, fast analysis 301 -, application to specific cases 208-209
- , -, narrow-bore 300 - ,characteristics 209, 248
-, -, - ,extra-column dispersion Critical band, calculation of 206, 209
314-315 Critical chain length 58 - 59
- , -, wide-bore 300 Critical point, definition 101
-, -, - , extra-column dispersion Critical properties, of solvents for SFC
314-315 102
-, effect on optimization criteria Darcy’s law 299
145- 146 Density of mobile phase, effect on reten-
- ,effect on overlapping resolution map- tion in SFC 103- 104
ping 216 Dependent variables
- , effect on sum criteria 132- 133 - , effect of temperature and compostion
- , effect on threshold criteria 143 in RPLC 68 -69
- ,factors affecting selection 298 - , general 173
-, importance of optimization 297 -298 -, in IPC 191
-, packed, for GC 301 -, mobile and stationary phase in LC
- ,packed, for LC 302 - 305 218
-, -, evaluation 317-318 -, optimization in IPC 209 -211
-, pzicked vs. open 299 - 300 Depressants, separation by programmed
-, practical dimensions for LC 303 -304 solvent LC 282
- , pressure drop 299 Desired analysis time in optimization cri-
-, temperature limit 21 teria 149
Column-switching - see: multi-column Detection time constant 313 -314
techniques - ,effect on noise level 3 13- 314
Compensation temperature in RPLC 68 -, in GC 315-316
Competition model, in LSC 76 -77 -, in LC 317-318
Complete mathematical optimization of Detection volume, effect on extra-column
progammed solvent LC 283 dispersion 3 12
Composite criteria 146- 158, 277 -278 Detection
Computation time 219,231 -232,290 - , dual detectors in series 239
Concentration sensitive detectors 305, - , dual-wavelength UV 239 -240
309 -, in SFC 103
Concentration of solute in effluent - ,linearity 306
334
Detector flowcell, Excess quantities 38
-, effect on extra-column dispersion in Exclusion 22
LC 317 Experimental design(s) 21 1
Diachoric model -, for Sentinel method 212-213
-, for GLC 41 -43 - , see also: fixed experimental designs
-, for RPLC 61 -, see also: full factorial designs
Differential migration 1 Expert systems 23 - 24,171
Digitization, need for - in chromato- Extra-column dispersion 310- 312
graphy 313 -, in GC 314-315
Dilute solutions 37 - 38 -, in LC 316-317
Diphenyl amines, retention in RPLC 226 Factorial designs - see: full factorial de-
Dipole induction interaction 25 signs
Dipole orientation interaction 25 Film thickness 6
Discrete parameters, definition 110 Fixed experimental design(s) 200 -220
Dispersion interaction 25 - ,characteristics 219 -220
Dissociation constant, definition 69 -, see also: critical band method
Dissociation ratio, definition 69 - ,see also: full factorial designs
Distribution coefficient - ,see also: window diagrams
-, definition 4 - ,see also: Sentinel method
-, in IEC 84 Flexible equation, for gradient shape 281
-, in LSC 76 Fractional overlap - see fractional peak
-, use for solvent classification 32 overlap
Distribution constant - see: distribution Fractional peak overlap 123- 125
coefficient -, definition 124
Distribution isotherm, definition 4 - , evaluation 127- 129
Dual-channel detection, for peak recogni- - , measurement 124- 125
tion 239-241 Full factorial design(s) 188- 191,
Dynamic LLC 53 - 55 209-210
Efficiency optimization 299 - 305 -, evaluation 191
Elemental criteria FAST LC columns 310
- , characteristics 130 Gas-liquid chromatography (GLC)
-, comparison 127 - 131 37 - 43
-, definition 119 - , relevant parameters 106
-, recommendations 131 Gas-solid chromatography (GSC) 43 -45
Eluotropic series, for IEC 87 - ,relevant parameters 106
Eluotropic strength parameter for LSC Gaussian peak
76 -, characteristics 8
-, nomogram 80 -, mathematical description 8
-, values of 77 Gel permeation chromatography 22 -23
-, for binary mixtures 78, 80 General case, definition 119
Eluotropic strength, in LSC 217 General elution problem 253 - 255
Elution program - see: analysis program Global optimum, definition 173
End-capping 57 - 58 Gradient duration times, calculation of
Equilibration time, in LLC 55 optimum range 280
Equilibrium constant - see: distribution Gradient elution 193- 199
coefficient -, blank signal 197
335
-, in IEC 91 -, build instrument stage 2% - 297
-, of proteins 263,280 -,for method development 18- 19,
- ,see also: programmed solvent LC 296 -297
Gradient program 260 - 261 -, requirements in GC 314-316
- ,optimum shape in LSC 262 --requirements in LC 316-318
-, optimum shape in RPLC 261 - 262 Interaction Chromatography, definition 1
-
Gradient scanning 193 199,290 Intermediate polarity, phases for LC 52,
-, for RPLC 195- 199 218
-, -, graphical procedure 197- 198 Interpretive methods 199-235
-, limitations 199 -
- ,characteristics 233,248 249
-, rule of thumb 194 -, definition 178
Gradient shape parameter 281 -, description 199
Grid search -, evaluation 234 - 235
-, as computation method 211, 219 -,for programmed soIvent LC 284 - 290
-, as optimization procedure 179- 181 - , for programmed temperature GC
-, -,evaluation 182- 183 275 - 276,273 - 275
-, required number of points 181- 182 - ,optimization criteria 130
GC-MS combination 241 Ion chromatography 87,91
Height equivalent of theoretical plate Ion exchange chromatography (IEC)
(HETP) 82 -93
- ,see: plate height -, stationary phases 82 - 84
Henry’s adsorption law 43 -, of proteins 87
Henry’s law 38 - ,relevant parameters 1f 0
Hierarchic criteria 141- 142, 206 - 207, - ,retention mechanism 86 -87
210 - , stationary phase
-, for programmed solvent LC 281 -, - ,microparticulate 84
Hold-up time - , -, pelliculars 83- 84
-, definition 2-3 Ion exchange equilibrium constant 85
-, effect on retention surfaces in RPLC Ion-pair chromatography (IPC) 53,
209,223 93 - 101
Hyperbolic equation, for retention in - ,full factorial design 189- 191
RPLC 61 -, normal phase 95 -96
Hyphenated methods 241 - , relevant parameters 111
Ideal chromatograms, calculation of ca- -, retention equation 94
pacity factors t 53 - 154 - ,retention mechanism 94- 95
Immiscibility of phases for LLC 52 - 53 -, reversed phase 95 -96
Initial experiments 177 - ,simple mechanism 93 -94
- ,for iterative designs 220,230 Ion-pair extraction - see: ion pair chro-
Injection delay time 311 - 312 matography
Injection profiles 311 -312 Ionic separation methods 23
Injection time 311 -312 -, see also: ion-pair chromatography
Inorganic anions, retention in IEC 87 -, see also: ion exchange chromatograp-
-, see also: ion chromatography hy
Instrument optimization 310- 318 Ionic strength, effect on retention in
-
- , importance of 297 298 RPLC 73
Instrumentation Iso-eluotropic mixtures 198- 199, 206,
336
/
212-213,218-219,221,226,278, McReynolds constants 31

284 -285 Mean effect of variable, estimation 189
-, in LSC 80-81, 216, 217 Mean relative effect of variable, estima-
- ,in RPLC 63 -67 tion 189
- , -, experimental composition 65 Median peak-valley ratios, definition 121
- , - ,multicomponent 66- 67 Method development
- , -, prediction of composition 65 - 67 -, general approach 15- 18
Isocratic composition -, in laboratory 18- 19
-, optimum range 198 -, instrumentation for 18- 19, 296, 297
-, prediction - see: gradient scanning Micro-bore columns - see: narrow-bore
Isocratic multi-solvent programming 265 columns
Isoelectric point 73 Migration speed, definition 2
Isothermal conditions, prediction from Minimum a criterion 140- 141,
temperature program 193 202-203,210
Iterative design(s) 220-233 Minimum column diameter in LC, calcu-
-, characteristics 232 -233, 249 lation 317-318
-, definition 220 Minimum criteria 140- 144
-, effect of local optima 228 -229 - ,evaluation 143- 144
- ,multi-dimensional 231 - 232 Minimum resolution criterion 141- 142,
Iterative optimization methods - see: ite- 270 - 273,285,287
rative designs -, as threshold criterion 207,214
Library search techniques 242 - 243 Minimum separation factor criterion 141,
Linear equation, for retention in RPLC 142,202 - 203,205
62 Mixed stationary phases
Linear interpolation, of retention surfaces -, for GC 41 -43,200
229 -23 1 -, for LC 75
Linear retention relationships 203 -205, Mobile phase effects
209 -, in LSC 77-78
Linear segmentation, multi-dimensional - , in RPLC 59 -67
231 -232 -, in SFC 103-104
Linear solvent strength (LSS) gradients Mobile phase parameters, definition 105
166,193-195,261-262,279-280 Mobile phase time - see: hold-up time
-, definition 279 Mobile phase(s)
-, optimum slope 280 -, definition 2
-, shape of 279 -, for SFC 102
Liquid-liquid chromatography (LLC) 48, Model(s)
52 - 56 -, equations vs. linear interpolation
-, characteristics 55 - 56 229-231
-, relevant parameters 107 -,for response surface 199- 200
Liquid-solid chromatography 76 - 82 -, for retention surface 178, 199-200,
-, relevant parameters 109 220,214
Literature search 16 -, from chromatographic theory 230
Local optimum, definition 173 -, moving least squares 231
Local vs. global optima 176- 177 -,polynomial 230 -231
LC-MS combination 241 -, regression analysis 230
Mass flow sensitive detectors 305, 309 Modified Simplex optimization procedure
337
-, see Simplex optimization Number of (theoretical) plates - see:
Modifier - see: organic modifier plate number
Modulators in LSC 79-80 Number of datapoints
Molecular interactions 25 - 26 - , for recording gas chromatograms
Monofunctional reagents 56 315-316
Multichannel detection, for peak recogni- - ,for recording liquid chromatograms
tion 241 -245 317-318
Multi-column techniques 159, 167, 257 Number of experiments
Multi-dimensional window diagrams -, for full factorial designs 188
- ,see: window diagrams, multi-dimen- -, for optimizing LSS gradients
sional 280-281
Multisegment gradients 287 - 288 - , for predictive method in programmed
Multisegment programs 268 - 269 solvent LC 288
- ,disadvantages 268 - 269 - , for sequential optimization of pro-
- ,for programmed temperature GC grammed temperature GC 271
270 - 273 - , for Simplex optimization 186
-, in LC 283 - 284 -, -, in programmed solvent LC 277-
- , -, systematic optimization 283 - 284 278
Multivalent ions, retention in RPLC -, initial, for iterative designs 230
72 -73 Number of parameters
Mutual independence of parameters 173 -, in iterative designs 230
-, see also: dependent variables - , in optimization procedures 177
Narrow-bore capillary columns for GC Number of peaks, effect on optimization
300 criteria 1 4 6 - 148
-,extra-column dispersion 3 14 - 3 15 Number of plates
- ,see also: column(s), capillary -, effect on minimum criteria 142- 143
Narrow-bore columns for LC 308 -, effect on product criteria 135
-, advantages 308 -, effect on sensitivity 307
Near-universal UV detection 199 Observed capacity factor, of ionized spe-
Net retention time, definition 4 cies in RPLC 71 -72
Normal phase liquid chromatography Octadecyl silica 58
(NPLC) 23,49,51- 52 Octyl silica 58
-, preferred modifiers 21 2 - 21 3 Open columns 6
Normalized resolution product 153, - , see also: column(s), capillary
223 Optimal particle size in LC 303
-, definition 138 Optimization criteria
-, for limited number of relevant peaks - , comparison 137
162 -, effect on optimum gradient 282 - 283
- , in programmed analysis 165 - , evaluation 145 - 146
-, use of solute weighting factors 164 -, for limited number of relevant peaks
Nucleobases, retention behaviour in IEC 158-163
90 - , - , recommendations 161 - 162
Nucleosides, retention behaviour in IEC - , for programmed analysis 165 - 167
90 - , recommendations for the general case
Nucleotides, retention behaviour in IEC 158-159
86 - , recommendations using solute weigh-
338
ting factors 164 Peak assignment 233 - 245
Optimization procedures - see also: peak recognition
- , characteristics 177- 179, 245 - 249 Peak capacity
- , conclusions 249 - 250 -, statistical 15
-, general outline 178 -, theoretical 14- 15
- , for programmed temperature GC Peak height (relative)
-, -, evaluation 275 - 276 -, effect on elemental criteria 127- 129
Optimization process, overview 296 -298 -, effect on peak-valley ratio(s)
Optimization, of programmed analysis 122-123
266 - 294 - , effect on resolution 117
Optimum temperature, for GC stationary Peak identification 238
phases 41 - ,using spectroscopic techniques
Optimum, definition 171 241 -243
Organic modifier(s) 59 Peak recognition 233 -245
- ,effect in programmed solvent LC 277 -, based on peak areas 236 - 238
-, effect on retention in IEC 90-93 -, -, computer program 238
-, effect on retention in IPC 99 - , in programmed temperature GC
- , effect on retention in SFC 103- 104 273 - 275
Organic polymers - , using principal component analysis
-, use in GSC 45 243 - 245
-, use in RPLC 70 - , using separate injections 236 -237
Overlapping resolution mapping (ORM) - , with dual-channel detection 239 - 241
141,214-215 -, with multichannel detection 241 - 245
Packed columns 6 -, with single channel detection
Packed columns - see also: column(s), 236 - 238
packed Peak shape, effect on elemental criteria
Pairing ion 129
-, chain length 99 Peak width, definition 7
- , concentration Peak-valley ratio(s) 119- 123
-, -, effect on retention in IPC 94, - , correction for baseline noise 123
96 -98 -, definitions 119-121
-, definition 93 -, measurement 122
-, distribution isotherm 97 -, product criteria 135-138, 142
-, examples of 98 - , -, for limited number of relevant pe-
-, nature of, effect on retention in IPC aks 162
97 -99 -, -, time correction factors 151
Parameter limits 177 -, -, use of solute weighting factors 163
Parameter space -, relation to resolution 122- 123
-, definition 171 -, sum criteria 132-133, 137-138
- ,reduction of 188- 199 -, -, correction for number of peaks
Partial polarities 25 - 27 147
Particle size, effect on sensitivity -, theoretical 122
309 - 310 - , threshold value 123
Peak area Permanent gases, analysis of 22,44
-, measurement 238 PH
-, reproducibility 238 -, effect on retention in IEC 87 -90
339
-, effect on retention in IPC 100 266 -268
-, effect on retention in RPLC 69-73 Principal component analysis, for peak
-, -, equation for 71 recognition 243 -245
-, working range, for RPLC 70 Probe solutes
Phase ratio -, McReynolds 31
-,definition 4 -, Rohrschneider 29
-, parameters affecting 5 -, Snyder 32
Phase selection diagram(s) 180- 181, Product criteria 134- 140
221 -231 -, literature 134
-, construction 221 -, see also: peak-valley ratio(s), product
-, two-dimensional 231 -232 criteria
Phenylthiohydantoin - see: PTH Product resolution criteria 134- 135, 226,
Physical parameters, definition 105 228
Pilot techniques - see: scouting techni- -,evaluation 137- 138
ques -, see also: time corrected resolution
Plate count - see:plate number products
Plate height, definition 9 - ,see also: calibrated normalized resolu-
Plate number tion product
-, definition 9 - , see also: normalized resolution pro-
-, effect on resolution 10- 14 duct
-, measurement 9 Program parameters
Polar adsorbents -, definition 266-267
-, use for RPLC 51 - ,optimization 269 - 270,270,273
-, see also: silica Programmed analysis 17,253-295
-, see also: alumina -,advantages of simple programs
Polarity 24-27, 32-33 268 -269
-, see also: solubility parameter(s) -, applications 253 -257
Polarity difference 53 -, as scouting technique 192- 199
- ,see also selectivity of phase systems in -, definition 253
LC -, disadvantages 256
Polarity range of samples 254 - , factors affecting retention 257 -258
Polyelectrolytes 73 - , factors affecting selectivity 257 -266
-, see also: proteins -, in routine situations 256
Pre-column, for sample concentration in -, optimization criteria 165- 167
LC 306 -, primary parameters 257 -258
Predictive optimization method Programmed elution - see: programmed
-, for programmed solvent LC 288 -290 analysis
Pressure limited conditions 155- 156, Programmed solvent LC 260 -266
302 - 303 -, optimization 276 - 294
-, see also: required analysis time, pres- -,optimization of primary parameters
sure limited conditions 292 -293
Pressure, effect on retention in SFC -, optimization procedures
103- 104 -, -,characteristics 292 -294
Primary parameters 17,191 -, -, evaluation 290-294
-,definition 108 -,primary parameters 276 -277
- ,in programmed analysis 257 -258, -, secondary parameters 277
340
- , selectivity optimization 293 - 294 Resolution 116- 117
-, ternary gradients 264 -265 -, definition 7
Programmed temperature GC 258 -260 -, factors affecting 10-14
-, optimization 269 -276 -, fundamental equation 10
- , resolution 260 -, in programmed analysis 165-166
-, retention 259 - 260 -, in programmed temperature GC 260
Proteins 73,263 -, relation to peak-valley ratio(s)
-, analysis by IEC 87 122-123
Proton acceptor parameter 32 -33 -, with solvent peaks 168
Proton donor parameter 32 -33 Resolution criterion
Pseudo-binary gradients 265 -266 -, characteristics 117
Pseudo-components - see: pseudo-sol- -, measurement 117
vents -, see also: resolution
Pseudo-isomeric plot, of response surface Resolution mapping - see: overlapping
172 ~
resolution mapping
~
Pseudo-solvents 265 Response surface

-, definition 199 -, definition 171
PTH amino acids -, representations 172- 173
-, retention behaviour in RPLC 264 Retention gap, in capillary GC 306
-, separation by programmed solvent Retention index
LC 283 -, definition 27 - 28
Quadratic equation(s), for retention in -, polar contribution 28
RPLC 60,214 - ,use for GC optimization 45 -47
Quaternary mixtures 60 -, variation with stationary phase com-
Ratio recording 239 - 241 position 46 -47
-, limitations 241 -, variation with temperature 46
Ratiograms - see: ratio recording Retention line - see: retention surface
Reduced linear velocity, definition 299 Retention mechanism
Reduced plate height, definition 299 -, in IEC 86-87
Reduced sample size, definition 308 -, in IPC 93-94,94-95
Relative retention -, in LSC 76-77
-, definition 5 -, in RPLC 56
- , see also: selectivity Retention surface
Reproducibility -, definition 177-178
-, in programmed solvent LC 261 -, using iso-eluotropic mixtures in
-, of spectral information 245 RPLC 222 -223
Required analysis time 152,203 Retention time, definition 2
- , as optimization criterion 153, Retention
156-157 -, fundamental equation 3
-, calculation 155, 300 -, in programmed temperature GC
-, factors affecting, in LC 302 259-260274-215
-, pressure limited conditions 152, 302 Retrieval systems 243
Required number of plates 126, 203,298 Reversed phase liquid chromatography
-, calculation 151, 155 (RPLC) 20,23,49,51-52,5644
- ,for ideal chromatograms 156 -, characteristics 74
Residual silanols 58 -, fexibility 49, 56
341
-, for large solute molecules 262 - 263 - , application to programmed solvent
-, relevant parameters 108 LC 284 -288
-,preferred modifiers 212 - 213 - , - , experimental design 284- 285
-, selectivity 56 -, application to LSC 216-217
Rohrschneider classification scheme -, characteristics 219 - 220, 248
27-31 - , expansion to non-iso-eluotropic sol-
Routine analysis vents 218-219
-, instrumentation for 18- 19 -, optimization criteria 214
-, with programmed elution 246 Separation factor 125 - 127,153 - 154
Ruler method 200-201, 206 - , characteristics 127
Sample capacity - , correction for plate count 127
-, in GSC 44 - ,definition 126
-, in LLC 55 -, in programmed analysis 166
Sample composition, effect on elemental -, optimum range 154- 157
criteria 129 -, with solvent peaks 168
Sample molecular weight 22 - 23 Sequential methods
Sample solvent 23, 306 -307 - , for optimizing programmed tempera-
Sample volatility 21 -22 ture GC 269 - 273
Samples, information about 15 - 16 Sequential scanning 192
Sampling frequency 3 13 Shape of gradient programs 194
- ,required for GC columns 315 - 316 -, for RPLC 194
- ,required for LC columns 317 -318 Shift rules - see: shifted compositions
Scanning techniques - see: scouting Shifted compositions
techniques - , in iterative designs 224- 225,
Scouting techniques 191- 199 227 -228
- ,graphical procedures for RPLC - ,multi-dimensional 231
197-198 Silica 77
Secondary parameters 18 -, characteristics 81
-, definition' 108 - 109 - ,reaction with silanes 56 -57
- ,in programmed analysis 268 - 269 -, use in GSC 45
Selectivity classes, of solvents 35 -, see also: polar adsorbents
Selectivity optimization . Simplex design 212
- ,definition 17 Simplex lattice design - see: Simplex de-
- , in programmed solvent LC 284 -290 sign
- , - ,evaluation 290 -294 Simplex method, as computation proce-
Selectivity dure 232
-, definition 5 Simplex optimization 183 - 187
-, effect on resolution 10- 14 -,advantages 186
- ,in programmed solvent LC 263 - 264 -,basic method 183 - 184
-,in programmed temperature GC 269, - ,definition 183
276 -, disadvantages 187
-, in LC 50-52 -, initial experiments 185
-,in LLC 54-55 -, modified method 184- 185
-, of phase systems in LC 52 - , number of experiments 186
Sensitivity optimization 305 -310 -, characteristics 247
Sentinel method 212 -220 -, optimization criteria 147 - 148
342
-, for programmed solvent LC 277 -279 Solvent selectivity parameters, for LSC
- ,for programmed temperature GC 216-217
275 - 276,269 - 270 Solvent strength in LSC - see: eluotro-
Simultaneous interpretive methods pic strength
200 - 220 Solvent(s)
-, see also: fixed experimental designs -, classification 33 - 34, 77
Simultaneous optimization procedures -, polarity 32-33
-, definition 179 - , selectivity 32 - 35
- ,without solute recognition 179 - 183 Specific cases, definition 1 19
- , - , characteristics 246 Specific permeability coefficient 299
- , - , see also: grid search Specific retention volume, definition 39
-, with solute recognition Specific surface area 43 -44
- , - ,see: interpretive methods -, definition 6
Single channel detection, for peak recog- Spectro-chromatograms 242
nition 236 - 238 Stability of LLC systems 53
Size exclusion chromatography 22 - 23 Staight phase liquid chromatography
Slope, of retention lines in RPLC 62 - 63 -, see: normal phase liquid chromato-
- , variation with solute 62 -64 graphy (NPLC)
Snyder classification scheme 31 -35, 212 Standard deviation, of Gaussian peak 9
-
Snyder theory for LSC see: competi- Standard state 48
tion model Stationary phase characterization (GC)
Soap chromatography - see: IPC, rever- 27-31
sed phase Stationary phase parameters, definition
Soczewinski equation for retention in 105
- LSC79 Stationary phase(s)
Solubility parameter(s) 24 - 27 -, definition 2
-, definition 24 - , effect on selectivity in programmed
- ,effect on retention temperature GC 276
-, -, in GLC 40-41 -, for GSC 45
-, -, in LC 48-50 -, for SFC 105
- ,effect on selectivity in LC 50 - 52 -, optimization in LC 217-218
-, limitations 52 -, polarity (GC) 27-31
-, of mixtures 60 Stepwise parameters, definition 110
-, relation to eluotropic strength in LSC Stepwise scanning - see: sequential
77 - 78 scanning
-, units 24 Stop criteria
- ,use for selection of LC phase systems - ,definition 178
48 -52 -, for iterative designs 225
Solute, definition 2 - , for Simplex optimization procedure
Solvent generated LLC - see: dynamic 185
LLC Strong dipole parameter 32 33-
Solvent localization in LSC 81 Sum criteria 131 - 133
Solvent peaks, effect on optimization cri- -, evaluation 133
teria 167 - 168 -, see also: peak-valley ratio(s), sum cri-
Solvent program 260 - 261 teria
- ,see also: gradient program -, see also: sum resolution criterion
343
Sum resolution criterion 117 - 119, Threshold separation 121
131- 133,137- 138,149 - ,see also: pea&-valleyratids), thres-
Supercritical fluid chromatography hold value
(SFC)20,101 - 105 Time corrected resolution products
-, applications 103 153-158
-, detection 103 - ,definitions 158
- relevant parameters 112 - ,for limited number of peaks 163
Supercritical fluid, definition 101 Time correction factors in optimization
Surface area, effect on phase ratio 6 criteria 149- 151
Systematic optimization in programmed - ,see also: analysis time, effect on opti-
solvent LC mization criteria
- ,primary paramems 279- 284 -,see also: column independent time
-, with limited solute recognition factors
281 -283 Time correction terms in optimization
-, without solute recognition 279 -281 criteria 148 - 149
Systematic sequential optimization Time required for optimization experi-
-,for programmed temperature GC ments 182
270 -273,275 -276 Total overlap criterion 144- 145
Temperature control in LLC 53,55 Trial calculations 288
Temperature program 259 - 260 Trifunctional reagents 57
-,see also: programmed temperature Two-dimensionai window diagrams 210
GC Type of cdumn, comparison 299 -300
Temperature Univariate optimization 173 - 176
- ,effect on retention in GLC 38 -40 -,characteristics 245
-, effect on retention in GSC 44-45 -, in IEC 92-93
-, effect on retention in IEC 89-90 Unretained time - see: hold-up time
-, effect on retention in IPC 101 Valley to top ratio
-, effect on retention in LSC 82 -,definition 121
-, effect on retention in RPLC 67 -69 - , use in optimization criteria 140
Ternary gradients 264- 265,278 -, with solvent jxaks 168
Ternary mixtures 60,67 Vapour pressure, effect on retention in
- ,prediction of retention from binary GLC 38
data 288 Viscosity, of LC eluents 305
Thermodynamic parameters, definition Volatility range, of samples 254
105 Water, polarity of 25
Thermodynamic phase diagram Weak acids and bases, retention behavio-
-,pure component 101 ur in IEC 88-89
-, ternary mixture 54 Weak acids, effect of pH on retention in
Thin layer chromatography (TLC), as RPLC 204
scouting technique 192 Weighting factors for optimization crite-
Threshoid criteria 136,141 - 143,145, ria 148-149, 150
150-151,207,214,288-290 -, for solutes 163 - 165
Threshold resolution 129 Wide wage samples 254 -255
-,see also: threshold separation Wide-bore capillary columns for GC 300
-, see also: peak-valley ratio@), thres- -, extra-column dispersion 314-315
hold value Window diagram@) 171- 172,200 206 -
344
- , application areas 203 -205
-,characteristics 205 - 206, 219 -220,
248
-, for pH optimization in RPLC
204 -205
- ,multi-dimensional 209 - 21 2
-, -, applications 210 - 21 1
-, optimization criteria for 203
345

Ion of Chromatographic Selectivity - 1986

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Ion of Chromatographic Selectivity - 1986

Încărcat de

Drepturi de autor:

Formate disponibile

JOURNAL OF CHROMATOGRAPHY LIBRARY - volume 35

optimization of chromatographic selectivity

Distributors for the United Statas and C . d a :

ELSEVIER SCIENCE PUBLISHING COMPANY INC.

ISBN 0-444-42681-7 (Vol. 35)

0 Elsevier Science Publishers B.V.. 1986

Printed in The Netherlands

Journal of Chromatography Library (other volumes in the series) IX

CHAPTER 2 SELECTION OF METHODS 20

CHAPTER 3 PARAMETERS AFFECTING SELECTIVITY 37

CHAPTER 4 OPTIMIZATION CRITERIA 116

CHAPTER 5 OPTIMIZATION PROCEDURES 170

CHAPTER 6 PROGRAMMED ANALYSIS 253

CHAPTER 7 SYSTEM OPTIMIZATION 296

LIST OF SYMBOLS AND ABBREVIATIONS 321

AUTHOR INDEX 329

SUBJECT INDEX 333

Other volumes in this series

Volume 1 Chromatography of Antibiotics (see also Volume 26)

Volume 34 Polymer Characterizationby Liquid Chromatography

Volume 35 Optimization of Chromatographic Selectivity. A Guide to Method

Volume 36 Selective Gas Chromatographic Detectors

Eindhoven, February 1986

American Chemical Society, Washington D.C., U.S.A.

Marcel Dekker Inc, New York, NY, U.S.A.

Elsevier Science Publishers, Amsterdam, The Nederlands.

Preston Publications Inc., Niles, IL, U.S.A.

Royal Chemical Society, London, Great Britain.

L.de Galan (Delft, The Netherlands)

Chromatography can be defined as the separation of molecules by differential

A schematic representation of a chromatograph is given in figure 1.I. This figure applies

1.2 SEPARATION - THE COLUMN

1.2.1 Retention times and capacity factors

By definition, the capacity factor (k,)of the solute i is

Eqn.(l.6) is the fundamental equation for retention in chromatography. Throughout this

1.2.2 Distribution coefficients

ki = &. Vs/ V , (1.10)

Eqn.(l .lo) relates retention in chromatography (k)to a thermodynamic parameter (KJ.

Eqn(l.12) is very useful in practice, because it expresses a directly in terms of the

1.2.4 The phase ratio

In general, the effective volume of a stationary phase (Vb can be increased in a

* For gas chromatography(and for supercriticalfluid chromatography)there is a real choice. Open

The resolution (R,) between two peaks is now defined as

R, = 2AtR / (w,+ w2). (1.14)

Ideally, chromatographic peaks are of Gaussian shape. In practice, because of finite

fit) = hexp -1/z (T)

Figure 1 . 4 Some characteristicsof a Gaussian peak.

1.4.1 The plate number

N = 5.54 (t,/w,,2)2 (1.1 6a)

N = 2 n(t, h/A)*. (1.17)

where L is the length of the column.

* In a more detailed analysis Snyder [loll found a?

We conclude from the above that the efficiency (flm

1.6 PEAK CAPACITY

1.7 METHOD DEVELOPMENT

In this section we will discuss a general approach to method development in

1.7.1 An organized approach

An organized approach to method development may be based on a series of actions as

1. Getting to know the sample

It is always advisable to search the literature for possible solutions to a paPticular

If the literature does not suggest a particular chromatographic technique to be applied,

After a particular chromatographic technique has been selected a chromatogram should

5 . Optimization of the separation