Appendix E

Equations and Analysis

E.1 Fitting Equations to Data The non-linear Eqs. (E-5), (E-6), and (E-8) can be
transformed by linearizing as follows:
Engineers, technologists, and scientists frequently collect
paired data in order to understand the behavior of the
Y ¼ aX b ln Y ¼ ln a þ b ln X (E-9)
system. Numerical methods of analysis have been
employed to solve a wide range of steady and transient Y ¼ aebX ln Y ¼ ln a þ bX (E-10)
state problems. The fundamentals are essential in the
basic operations of curve fitting, approximation, in- Y ¼ a þ beX ln Y ¼ ln a þ ðln bÞX (E-11)
terpolation, numerical solutions of simultaneous linear
and non-linear equations, numerical differentiation, and
integration. These requirements are greater when new E.2 Linear Regression Analysis
processes are designed.
Various types of software packages are now readily
Regression analysis uses statistical and mathematical
available for scientists and engineers. They must fit
methods to analyze experimental data and to fit mathe-
a function or functions to measure data that fluctuate,
matical models to these data. The least squares provide
which result from random error of measurement. A
the best method for objectively determining the best
Fortran computer program (PROG1) was developed
straight line through a series of points. The method as-
which determines the coefficients that provide the best
sumes that all deviations from the line are the result of
fit for the following equations:
error in the measurement of the dependent variable. The
Y ¼ a þ bX (E-1) method of least squares yields the parameters which
minimize the sum of squares of the residuals (e.g., the
Y ¼ a þ bX 2 (E-2) deviation of each measurement of the dependent variable
Y ¼ a þ b=X (E-3) from its calculated value). If F is the calculated value and
Y is the original value of the dependent variable, then the
Y ¼ a þ bX 0:5 (E-4) residual sum of the squares can be minimized by the
following expression.
Y ¼ aX b (E-5)
bX n
X n
X
Y ¼ ae (E-6) SRS ¼ ri 2 ¼ si 2 ¼ minimum (E-12)
Y ¼ a þ b=X (E-7) i¼1 i¼1

X
Y ¼ a þ be (E-8) where n ¼ number of observations of X–Y data.
 APPENDIX E Equations and Analysis

b
Y i ¼ a þ bXi (E-13) Equations (E-23) and (E-24) apply only when fitting
ri ¼ a þ bXi  Yi (E-14) a straight line to a set of X-Y data points. Both equations
are linear in X, Y, and n and the unknowns a and b.
and Using Cramer’s rule for the simultaneous equations, we
have
n
X n
X P P 
SRS ¼ ri 2 ¼ ða þ bXi  Yi Þ2 (E-15)  Yi Xi 
P P
i¼1 i¼1  Xi Yi Xi 2 
a ¼  P  (E-25)
The values of a and b are found by minimizing Eq. (E-15).  Pn P X2i 
This involves taking the partial derivative of the equation  Xi Xi 
with respect to each variable a and b, and setting the
result to zero. and
P P  P 
v ri 2 v ri 2  n Yi 
¼ 0 and ¼0 (E-16) P P
va vb  Xi Xi Yi 
b ¼  P  (E-26)
Substituting Eq. (E-15) into Eq. (E-16) gives  Pn Xi 
 Xi P Xi 2 
P
v ða þ bXi  Yi Þ2
¼0 (E-17) Solving Eqs. (E-25) and (E-26) gives
va
P P P P
and Xi 2 Yi  Xi Xi Yi
P a¼ P P P (E-27)
v ða þ bXi  Yi Þ2 n Xi 2  Xi Xi
¼0 (E-18)
vb and
This is equivalent to P P P
n Xi Yi  Xi Yi
P P b¼ P 2 P P (E-28)
2 ða þ bXi  Yi Þv ða þ bXi  Yi Þ n Xi  Xi Xi
¼0 (E-19)
va
respectively, where all the sums are taken over all ex-
and perimental observations. The developed computer pro-
P P gram PROG1 computes the constants a and b for Eqs.
2Xi ða þ bXi  Yi Þv ða þ bXi  Yi Þ
¼0 (E-20) (E-1)-(E-8). The program evaluates the correlation co-
vb
efficient, which gives a measure of the accuracy of fit.
Since b, X, and Y are not functions of a, and the partial Alternatively, a table with column heading as Xi, Yi, Xi 2,
derivative of a with respect to itself is unity, Eq. (E-19) and XiYi can be constructed using the Microsoft Excel
reduces to spreadsheet; the constants a and b can be evaluated from
X X X Eqs. (E-27) and (E-28) respectively. The Excel spread-
aþ bXi ¼ Yi (E-21) sheet also employs a trendline through the data after
being plotted into a worksheet. The following are the
Similarly, a, X, and Y are not functions of b. Therefore, steps being employed to fit a straight line to a set of data
Eq. (E-20) becomes in the Excel spreadsheet.
X X X 1. Open a new worksheet and enter the X-data (the
aXi þ bXi 2 ¼ Xi Yi (E-22)
independent variable) in the leftmost column.
where a and b are constants. Equations (E-17) and (E-18) 2. Enter the Y-data (the dependent variable) in the next
are expressed as: column.

n n
3. Plot the data as a line graph (i.e., an X–Y chart) with
X X
an þ b Xi ¼ Yi (E-23) arithmetic coordinates. Do not interconnect the in-
i¼1 i¼1 dividual data points.
4. Activate the graph for editing. Then click on one of
and the plotted data points, thus selecting the data set
n
X n
X n
X as the active editing object (the data points will
a Xi þ b Xi 2 ¼ Xi Yi (E-24) appear highlighted when this step is carried out
i¼1 i¼1 i¼1 correctly).

872
 Equations and Analysis APPENDIX E

5. Choose Trendline from the Insert menu. They The computer results give the constants A and B from
specify the type of curve, and request any pertinent the model Eq. (E-6) as:
options when the Trend-line dialog box appears.
These are, displaying the intercept, equation of the A ¼ 0:7206E þ 15
curve and its associated correlation coefficient (r2) B ¼ 0:9457E þ 04
squared. Extrapolating the curve fit forward, that is E
beyond the rightmost data point or backward Slope: ¼ 0:9457  104
R
beyond the leftmost data point, is another option to
explore. (Generally the equation of the curve and its where
associated r2 value are requested for display on the
graph.) R ¼ 8:314 J=mol K
6. Press the OK button. The curve fitting will then be
E ¼ 78625 J=mol:
carried out and the results are displayed The pre-exponential factor, ko ¼ A ¼ 7:206  104 S1
automatically. The Excel spreadsheet Example E-1.xls shows an-
Commercial curve fitting packages are now available in other example using the Lineweaver-Burk equation to
the market and details can be retrieved from the world determine the constants km and Vmax.
wide web (www) using the Google search engine by
typing linear regression. Typical websites are:

 www.ebicom.net/~dhyams E.3 Polynomial Regression

 www.analyse-it.com
 www.wessa.net. Some engineering data are often poorly represented by
a linear regression. Some form of non-linear regression
Table E-1 shows the computer results of Example E-1. can be developed, if the dependence of Y on X is known,

873
 APPENDIX E Equations and Analysis

Figure E-1 The Excel spreadsheet snapshot of Example E-1.

N
X
Table E-1 Curve Fitting for Two Variables S¼ ½Yj  C0  C1 Xj  C2 Xj2  $$$  Cn Xjn 2
j¼1
X 1/X Y (E-30)
283.0 0.00353 2.08 At the minimum, all the partial derivatives with respect
288.7 0.00346 4.57 to the chosen constants are zero, that is
294.5 0.003396 8.24
300.5 0.003328 15.8 vS vS vS
306.2 0.003266 26.4 ; ; :::; ¼0 (E-31)
311.6 0.00321 46.1 vC0 vC1 vCn
318.7 0.003138 93.5
This gives a system of ðn þ 1Þ linear equations in ðn þ 1Þ
unknowns, C0 ; C1 ; C2 ; .; Cn
XN
vS
although the total convergence of this iterative regression ¼0¼ 2ðYj  C0  C1 Xj  C2 Xj 2
vC0 j¼1
procedure cannot be guaranteed. However, if the form of
dependence is unknown, then Y can be treated as a gen- $$$  Cn Xj n Þð1Þ
eral function of X by trigonometric terms (Fourier anal- vS XN
ysis) or polynomial function. The least squares procedure ¼0¼ 2ðYj  C0  C1 Xj  C2 Xj 2
vC1
can be readily extended to fit the data to an nth-degree j¼1
polynomial: $$$  Cn Xj n ÞðXj Þ
XN
Y ¼ C0 þ C1 X þ C2 X2 þ $$$ þ Cn Xn (E-29) vS
¼0¼ 2ðYj  C0  C1 Xj  C2 Xj 2
where C0 ; C1 ; C2 ; :::; Cn are constants. vC2 j¼1
For this case, the sum of the squares of the residuals is $$$  Cn Xj n ÞðXj 2 Þ
minimized.

874
 Equations and Analysis APPENDIX E

« Linear equations generated by polynomial regression

vS N
X can be ill-conditioned when the coefficients have very
¼0¼ 2ðYj  C0  C1 Xj  C2 Xj 2  $$$ small and very large numbers. This results in smooth
vCn j¼1 curves that fit poorly. Ill-conditioning often happens if
Cn Xj n ÞðXj n Þ the degree of polynomial is large and if the Y values
cover a wide range. The error of the polynomial re-
(E-32)
gression by a standard error of the estimate is shown
The above equations are set to equal zero, and can be below:
rearranged in the following set of normal equations.
P
P P P P 2 SSE
C0 N þ C1 Xj þ C2 Xj 2 þ $$$ þ Cn Xj n ¼ Yj s ¼ (E-34)
Nn1
P P P P
C0 Xj þ C1 Xj 2 þ C2 Xj 3 þ $$$ þ Cn Xj nþ1
P where
¼ Xj Yj n ¼ the degree of polynomial
P P P P
C0 Xj 2 þ C1 Xj 3 þ C2 Xj 4 þ $$$ þ Cn Xj nþ2 N ¼ the number of data points.
P The coefficient of determination can be expressed as:
¼ Xj 2 Yj
« SSE
P P P P r2 ¼ 1 
C0 n
Xj þ C 1 Xjnþ1 þ C2 Xjnþ2 þ $$$ þ Cn Xj 2n SST
P n PN
¼ Xj Yj j¼1 ðYj Y j Þ2
b
¼ 1  PN (E-35)
(E-33) j¼1 ðYj  Y j Þ2
Equation (E-33) in matrix form becomes
The correlation coefficient is given by
UC ¼ V

N
 P
X
P 2
X :::
P n
Xj
 
r¼ 1
SST

SSE 0:5
(E-36)
P P j2 P j3 P nþ1
Xj Xj Xj ::: Xj
P 2 P 3 P 4 P nþ2 The numerator of Eq. (E-34) should continually decrease
U¼ Xj Xj Xj ::: Xj
as the degree of the polynomial is raised. Alternatively,
P« n «
P nþ1 «
P nþ2 « P « 2n the denominator of Eq. (E-34) causes a2 to increase once
Xj Xj Xj ::: Xj
there is departure from the optimum degree.

 C   PP Y 
0 j E.4 Multiple Regression Analysis
C1 XY
P j2 j
C¼ C2 V¼ Xj Yj Inadequate results are sometimes obtained with a single
«
P « n independent variable. This shows that one independent
Cn Xj Yj variable does not provide enough information to predict

875
 APPENDIX E Equations and Analysis

Figure E-2 The Excel spreadsheet snapshot of Example E-2.

the corresponding value of the dependent variable. We These observations are of the form:
can approach this problem, if we use additional in-
dependent variables, and develop a multiple regression
Yj ¼ C0 þ C1 X1j þ C2 X2j þ $$$ þ CK CKj þ xj
analysis to achieve a meaningful relationship. Here, we
can employ a linear regression model in cases where the (E-38)
dependent variable is affected by two or more controlled For j ¼ 1; 2; .; N
variables.
The linear multiple regression equation is expressed as: where
Yj = the jth observation of the dependent variable
Y ¼ C0 þ C1 X1 þ C2 X2 þ $$$ þ CK XK (E-37) X1j,..., Xkj = the jth observation of the X1, X2, ... , Xk
independent variables.
where
We can use a least squares technique to calculate esti-
Y = the dependent variable mates of C b 1 ; .; C
b 0; C b K by minimizing the following
X1, X2 ..., XK = the independent variable equation:
C0, C1, C2 ... ,CK = the unknown regression
coefficients N   
X 2
K = the number of independent variables. S¼ Yj  Cb0þC
b 1 X1j þ $$$ þ C
b K XKj
The unknown coefficients are estimated based on n ob- j¼1
servation for the dependent variable Y, and for each of N
X
the independent variables ¼ xj 2
Xi ’s where i ¼ 1; 2; 3; .; K: j¼1
(E-39)

876
 Equations and Analysis APPENDIX E

Taking the partial derivatives of S with respect to Table E-3 Analysis of Variance Table for Linear Multiple Regression
b 0; C
C b 1 ; .; C
b K , that is vS ; vS ; .; vS and setting them
vb
C 0 vb
C1 vb
CK Source of Degree of Mean
equal to zero, we obtain the following set of equations: Variance Freedom Sum of Squares Squares
    P
NCb 0 þ P X1j C b 1 þ ::: þ P XKj C b K ¼ P Yj Total N 1 SST ¼ ðYj  Y Þ2 MST ¼
P SST=ðN  1Þ
P      SSR ¼ ð Ybj  Y Þ2
b 0 þ P X1j 2 C b 1 þ ::: þ P X1j XKj C
Regression K MSR ¼
X1j C bK SSR=K
P
P Error N K 1 SSE ¼ ðYj  Ybj Þ2 MSE ¼
¼ X1j Yj SSE=ðN  K  1Þ
P     
X2j Cb 0 þ P X1j X2j C b 1 þ ::: þ P X2j XKj C bK
P
¼ X2j Yj
We can calculate the coefficients of the independent
« variables, if Eq. (E-45) is linearized by taking its natural
P     
XKj C b 0 þ P X1j XKj C b 1 þ ::: þ P XK j 2 C bK logarithm to give
P ln Y ¼ ln C0 þ C1 ln X1 þ C2 ln X2 þ :::
¼ XKj Yj
(E-40) þ CK ln Xk
b 0; C
The coefficients C b 1; C
b 2 .; C
b K can then be obtained
Equation (E-40) can be expressed in matrix form as:
by Gaussian elimination. Table E-3 shows the variance
b ¼V
UC table for linear multiple regression. The coefficient of
determination is
where SSE
2 P P 3 r2 ¼ 1  (E-47)
N X ::: XKj SST
P P 1j2 P
6 x1j X1j ::: X1j XKj 7 and the correlation coefficient is
U¼64 «
7
5 (E-41)
P P P  
XKj X1j XKj XKj 2 SSE 0:5
r¼ 1 (E-48)
2b 3 2 P 3 SST
C0 P Yj The test statistic is the F-ratio, which is defined by
6b 7 6 X1j Yj 7
b ¼6C
C 17 6 7
4 « 5V ¼4 5 (E-42)
P « MSR
bK
C XKj Yj F¼
MSE
A computer program PROG3 has been developed to
U is a symmetric matrix.
We can obtain estimates for the coefficients determine the coefficients and correlation coefficient of
b 0; C
C b 1 ; .; C
b K by successive elimination or by solving a multiple regression equation.
for the inverse of U. That is,
E.5 Simultaneous Equations Using
b ¼ U 1 V
C (E-43) the Matrix Methods
where
Matrix solution of sets of linear equations are solved by
U1 ¼ the inverse of U.
b 1 ; .; C
b 0; C b K, the estimates of the Excel as follows.
After solving for C
First, write a set of simultaneous equations in matrix
dependent variable observations Yj can be obtained as
format
follows:
Ax ¼ b (E-51)
b b0þC
Yj ¼ C b 1 X1j þ ::: þ C
b K XKj (E-44)
where A is the coefficient matrix, x is the vector of un-
The power equations have often been derived to calcu- knowns, and b is the result vector. To solve this matrix
late the parameters of experimental data. Such an equation, multiply both sides from the left by the inverse
equation can be expressed in the form: of A:

Y ¼ C0 $X1 $X2 $.$XK (E-45) A1 A x ¼ A1 b (E-52)

877
 APPENDIX E Equations and Analysis

Alternatively, the Excel spreadsheet Prog4a.xls is used

By definition, a matrix times its inverse gives the
identity matrix, and the identity matrix times any vector
gives the vector back. The equation above reduces to the
following solution:
x ¼ A1 b (E-53)
The following example uses the Fortran program PROG4
and Excel spreadsheet Prog4a.xls.
to solve the above example problem. The procedure is as
follows.
1. Start with a new worksheet, and name it Matrix
Method.
2. In cell A1, enter solving sets of equations; Matrix
inversion.
3. In cell B3, enter Ax ¼ b.
Enter the coefficient matrix A and the results
vector b.
4. In cell A5, enter input matrix A.
5. In cells A6:D9, enter the constants of matrix A:

Cell Constant Cell Constant Cell Constant Cell Constant

A6 51.3 B6 43.2 C6 56.4 D6 47.4
A7 11.3 B7 11.5 C7 15.5 D7 8.5
A8 29.4 B8 31.5 C8 22.5 D8 30.4
A9 8 B9 10.3 C9 5.6 D9 13.7

6. In cell F5, enter result vector b

7. In cells F6:F9, enter the constants of the result
vector.
8. In cell A12, enter Inverse matrix (1/A)
9. Select cells A13:D16, then type
¼ MINVERSE ðA6:D9Þ
and press Ctrl + Shift + Enter to insert the formula
into the whole section.
10. In cell F12, enter solution vector x ¼ (1/A)b.
11. Select cells F13:F16, then type
þ MMULT ðA13:D16; F6:F9Þ
and press Ctrl + Shift + Enter to insert the formula
into the whole selection.
12. Turn off the gridelines, and outline the worksheet
cells as shown in Matrix Method.

878
 Equations and Analysis APPENDIX E

Figure E-3 The Excel spreadsheet snapshot of Example E-4.

The worksheet should now look like Matrix Method, the right side of Eq. (E-58) are squares, they will all be
with the solution values 0.11722, 0.378939, 0.211747, greater than or equal to zero. Hence, the only way that y
and 0.305356 for a , b , g, and d in cells F13:F16. This is can be equal to zero is that each of the individual f’s
shown in Figure E-3. should be equal to zero. Therefore, the values of xl, x2, x3,
. , xn that cause y to equal zero will be the solutions to
the given system of equations. The general approach used
E.6 Solving Simultaneous Non- with Solver is to define a target function consisting of the
Linear Equations in Excel using squares of the individual equations, as indicated by Eq.
Solver (E-58), and to then determine the values of x1, x2, x3, . . .
, xn that cause the target function to equal zero. To solve
a system of simultaneous equations with Solver, proceed
Solver can be used to solve a system of simultaneous
as follows.
equations, which can be either linear or non-linear.
Suppose the equations are represented as 1. Enter an initial guess for each independent variable
ðx1 ; x2 ; x3 ; .; xn Þ in a separate cell on the
f1 ðx1 ; x2 ; .; xn Þ ¼ 0 (E-54) worksheet.
f2 ðx1 ; x2 ; .; xn Þ ¼ 0 (E-55) 2. Enter the equations for f1 ; f2 ; f3 ; .; fn and y in
separate cells, expressed as Excel formulas. Within
f3 ðx1 ; x2 ; .; xn Þ ¼ 0 (E-56) these formulas, express the unknown quantities
« x1 ; x2 ; x3 ; .; xn as the addresses of the cells

(E-57) containing the initial guess.
fn x1 ; x2 ; .; xn ¼ 0
3. Select Solver from the Tools menu.
Thus, we have a system of n equations in n unknowns. We
4. When the Solver Parameters dialog box appears,
can find the values of xl, x2, x3, . . . , xn that cause each of enter the following information.
the equations to equal zero, if the function
(a) The address of the cell containing the formula for
y ¼ f1 2 þ f2 2 þ f3 2 þ ::: þ fn 2 (E-58) y in the Set Target Cell location.
(b) Select value of y in the Equal to line. Then enter
is equal to zero; that is, to find the values of xx, x2, x3, . . . , xn
0 within the associated data area (i.e., determine
that cause Eq. (E-58) to equal zero. Since all the terms on

879
 APPENDIX E Equations and Analysis

the values of xx, x2, x3,., xn that will drive the values of the independent variables will appear in the
target function to zero). cells that originally contained the initial values. The cell
(c) Enter the range of cell addresses containing containing the target function will show a value that is
the initial values of x1, x2, x3, . , xn in the area zero or nearly zero.
labeled By Changing Cells.
E.7 Gauss-Seidel Iterative Method
(d) If you wish to restrict the range of the indepen-
dent variables, click on the Add button under
Alternatively, Fortran computer program PROG5 em-
the heading Subject to the Constraints. Then
ploys the Gauss-Seidel method for solving the four ma-
provide the following information within the Add
terial balance equations. The four material balance Eqs.
Constraint dialog box for each of the independent
(E-65)-(E-68) are rearranged to solve for the unknown
variables.
on the diagonal position of each Eq. (E-74).
(i) The cell address containing the initial value of
the independent variable in the Cell Refer- ð1000Þ ð1Þ ¼ 1000CA1 þ ð0:1Þ ðCA1 Þ ð1000Þ
ence location. (E-65)
(ii) The type of constraint (i.e.,  or ) from the 1000CA1 þ 100CA3 ¼ 1100CA2 þ ð0:2Þ ðCA2 Þ ð1500Þ
pull-down menu.
(iii) The limiting value in the constraint data area. (E-66)
(iv) Select OK to return to the Solver Parameters 1100CA2 þ 100CA4 ¼ 1200CA3 þ ð0:4Þ ðCA3 Þð100Þ
dialog box or select Add to add another con-
(E-67)
straint.
Note that you can always change a con- 1000CA3 ¼ 1100CA4 þ ð0:3Þ ðCA4 Þ ð500Þ
straint or delete a constraint after it has been (E-68)
added.
(e) When all the required information has been en- 1100CA1 ¼ 1000
tered correctly, select Solve. This will initiate 1000CA1  1400CA2 þ 100CA3 ¼0
the actual solution procedure. 1100CA2  1240CA3 þ 100CA4 ¼0
1100CA3  1250CA4 ¼0
A new dialog box labeled Solver Results will then appear, (E-74)
telling you whether or not Solver has been able to solve
the problem. If a solution has been obtained, the desired Table E-7 shows the computer results of PROG5.

Table E-7 Solution of Simultaneous Linear Equations by Gauss-Seidel Method

NUMBER OF LINEAR EQUATIONS: 4
MAXIMUM ITERATIONS 15
CONVERGING TOLERANCE: 0.100000E-03
THE COEFFICIENT MATRIX A(1,1) ... A(Nþ1,Nþ1) IS:

1100.0000 0.0000 0.0000 0.0000 1000.0000

1100.0000 1400.0000 100.0000 0.0000 0.0000
0.0000 1100.0000 1240.0000 100.0000 0.0000
0.0000 0.0000 1100.0000 1250.0000 0.0000

THE STARTING VECTOR X(1) ....X(N) IS:

5.000000E-01
5.000000E-01
5.000000E-01
5.000000E-01
PROCEDURE CONVERGED AFTER 5 ITERATIONS
SOLUTION VECTOR X (1) ..... X (N) IS:
9.090909E-01
6.968780E-01
6.654189E-01
5.855687E-01

880
Equations and Analysis APPENDIX E

881
 APPENDIX E Equations and Analysis

CAO uA3
uAO

V1
CA1 k1
CA1 uA4
V2
uAO
CA2 k2

CA2
uAO + uA3
V3 k3
CA3 CA3
V4
k4 CA4 CA4
uAO + uA3 + uA4

uAO + uA4

Figure E-4 Chemical reaction with recycles in four continuous stirred tanks.

Figure E-5 The Excel spreadsheet snapshot of Example E-5.

882
 Equations and Analysis APPENDIX E

Figure E-6 The Excel spreadsheet snapshot of Example E-5.

Figure E-7 The Excel spreadsheet snapshot of Example E-5.

883
 APPENDIX E Equations and Analysis

Figure E-8 The Excel spreadsheet snapshot of Example E-5.

Nomenclature MST ¼ Total mean squares

r ¼ Coefficient of determination
F ¼ F-ratio SSE ¼ Error sum of squares
MSE ¼ Error mean squares SSR ¼ Regression sum of squares
MSR ¼ Regression mean squares SST ¼ Total sum of squares

Further Reading

1. Coker, A.K., Fortran Programs For
Chemical Process Design, Analysis and
Simulation, Gulf Publishing Co.,
Houston, Tx, 1995.
2. Gottfield, B.S., Spreadsheet Tools for
Engineers: Excel 5.0 Version, The
McGraw–Hill Co. Inc., 1996.
3. Billo, E.J., Excel for Chemists – A
Comprehensive Guide, Wiley-VCH, Inc.,
1997.
4. Gottfried, B.S., Spreadsheet Tools for
Engineers: Excel 2000 Version,
McGraw-Hill’s Best – Basic Engineering
Series and Tools, 2000.

884