Technical Note: 40699

Analytical Method Development in Graphite Furnace Atomic Absorption Spectrometry.

Although the difculties in GFAAS must not be underestimated, the situation is, in fact, not as fraught with problems as it sometimes appears. The key to success is that any analytical method developed for GFAAS must be made as simple as possible, with the minimum addition of reagents and a minimum number of sample handling steps. This paper attempts to lay down such a procedure, based on Thermo's accumulated knowledge gained from a considerable period of involvement in advanced graphite furnace technology and analytical methods development.

Introduction
Graphite furnace atomic absorption spectrometry (GFAAS) has become a popular analytical technique in recent years, mainly because of its enhanced sensitivity that allows measurements in the picogram (10-12 g) range. However, it has also acquired a reputation of being a rather difcult technique to use. Papers describing intractable interferences and excessive background levels have publicised some of the 'real life' problems encountered in GFAAS. As a direct result of the high sensitivity, another practical problem that must be faced is the control of external contamination. Contaminants can be introduced to standards and samples alike and come from a variety of sources, including reagents (such as matrix modiers), solvents (used in dissolution procedures), the laboratory atmosphere and contaminated glassware. Taken together, these problems can give the impression that GFAAS is a technique that requires many complex and poorly understood steps to be made before an analytical method can be successfully developed and proved.

Modern Graphite Furnace Atomizers

Thermo Electron Corporation offers the AA Series range of atomic absorption spectrometers that use Windows-based software to offer high performance and all the exibility required by modern analytical laboratories. The GF95 and GF95Z Graphite Furnaces are major accessories that have been developed to complement the AA Series spectrometers. They are similar designs, with the GF95Z providing a background correction system based on the Zeeman opto-magnetic effect, while the GF95 relies on the continuum source Quadline background correction system included in all the AA Series spectrometers. The design of these furnaces is discussed in detail in (1), but at their heart is a graphite cuvette that is 28 mm long with an internal diameter of 5 mm.

Figure 1. GF95Z Graphite Furnace System

A range of different types of graphite cuvettes is available, so that the optimum type can be selected for the analysis. Some examples are shown in Figure 2.

Figure 2. Thermo Graphite cuvettes.

An analyst new to GFAAS analyses may not fully appreciate why there are

different types of graphite cuvettes available. The normal cuvette is made from a form of graphite known as electrographite. This cuvette is suitable for the determination of volatile elements such as lead and cadmium in simple matrices, such as clean waters. Electrographite is a relatively porous material and allows samples to soak into the graphite lattice. Many elements, including iron, vanadium and molybdenum, then react with the graphite during the furnace program to form stable carbides and so cannot be measured with an electrographite cuvette. Similarly, matrix components such as alkali metal halide salts can penetrate the carbon lattice, to be subsequently released in the high temperature atomization phase, where they will often cause chemical, vapor phase interference effects. For these elements and types of sample, it is necessary to use a modied electrographite cuvette, which is coated with a layer of pyrolytic graphite, generally about 12 - 20 m thick. Pyrolytic graphite is a much denser form of carbon, with very low porosity, so that the sample cannot pass through it to soak into the graphite lattice. Pyrolytically coated electrographite cuvettes are required for the carbide forming, medium-volatile and refractory elements, and for samples with complex, high salt matrices. Extended Lifetime Cuvettes (ELC's) are a unique Thermo development, and have a pyrolytic coating that is up to 10 times thicker than the standard coating. This gives them more stable performance and much longer useful lifetimes than either of the electrographite-based cuvettes. These cuvettes are especially recommended for high throughput analyses with complex samples, and when measuring the most refractory elements. Omega Platform cuvettes have an integral L'Vov platform built into the cuvettes, which makes them particularly suitable for determining the more volatile elements in heavy matrices where vapor phase interferences cause severe problems. Platform atomization and the Omega cuvettes are discussed in more detail in (2). An inert sheathing gas is used to protect the cuvette and graphite enclosure from oxidation. Argon is normally recommended for the purpose, but it is possible to use high purity nitrogen if argon is not available. Some elements, such as vanadium or aluminum, must be measured

using argon because they form stable nitrides with nitrogen. An independent ow of the inert gas is passed through the cuvette during the furnace cycle, to help carry away vapor and smoke produced by the decomposition of the sample. It is also possible to introduce an alternate gas in place of the argon during some or all of the furnace program phases. Oxygen or air can be introduced during the ashing (pyrolysing) phase of the heating program to assist the decomposition of samples containing a high carbon content - for example biological materials or foodstuffs, and can prevent the build-up of a carbonaceous residue inside the cuvette. It is also possible to connect other gases, such as argon/methane for in-situ pyrolytically coating of the cuvette interior, to the alternate gas channel. The nal important accessory is the autosampler, which is nowadays considered an essential part of a GFAAS analyses. Modern graphite furnace autosamplers, such as Thermo's FS95, provide many features and facilities to assist the analyst; some are listed in Table 1.

Figure 3. The FS95 Furnace Autosampler

FS95 FURNACE AUTOSAMPLER FEATURES

Automatic matrix modication wet and dry mixing options available Automatic standard preparation xed or variable volume preparation may be used Automatic sample dilution either a xed dilution ratio for all samples, or intelligent dilution of over-range samples only Automatic re-concentration of samples, using multiple injections Automatic standards addition preparation Automatic QC spike addition Table 1. Some Features of the FS95 Furnace Autosampler

Sample Preparation
Contamination problems are common in GFAAS analyses, but most laboratories do not have easy access to Class 100 clean rooms in which to prepare their samples. However, a careful, systematic approach to sample handling can successfully control contamination even in a normal working laboratory. Glassware, such as volumetric asks, should be lled with 10 % v/v nitric acid and stored until required for use. When needed, the asks are simply rinsed a few times with doubly deionized water. In general, it is better to mark each ask with a particular analyte concentration and use this one ask to contain that concentration of analyte only. Glass volumetric pipettes are often major contributors to contamination. Even when precautions are taken to acid-soak them and rinse them with pure water, they are often left to dry in the open laboratory, dried in contaminated ovens, or rinsed with contaminated acetone. It is preferable to perform volumetric transfers where possible by using automatic pipettes with disposable tips. The tips can be cleaned relatively easily, by dispensing several volumes of 10 % v/v nitric acid followed by several volumes of deionized water before taking up the sample. When using such a pipette for preparing accurate standards and sample dilutions, it is important that the analyst should validate the pipette's accuracy to ensure that it does not introduce signicant volumetric errors. Glassware normally used for ame AAS analysis should not be used for furnace determinations without thorough cleaning, and where possible it is recommended that a specic set of glassware should be reserved for GFAAS solution preparation. All solutions used for GFAAS work should be prepared freshly each working day, and should contain a minimum of 10-2M nitric acid to stabilize the trace metals in solution.

working in GFAAS are the Journal of Analytical Atomic Spectrometry (JAAS), The Analyst, Spectro Chimica Acta, Analytical Chemistry and Anal. Chim. Acta. The journal JAAS from The Royal Society of Chemistry, U.K. contains frequent reviews of all the recent publications in a particular application area, which form an excellent starting point for the research. Consideration should be given to any specic sample collection and transport procedures that may be required to ensure that stable, representative samples arrive in the laboratory for analysis. Solid samples will typically require some form of digestion procedure to bring them into solutions so that they can be analyzed, while liquids may require dilution or concentration to bring the expected analyte concentrations into the measurable range. These sample preparation procedures can dramatically affect the performance of the analytical method, but are outside the scope of this short article.

appreciate that these defaults have been optimized for the measurement of the selected element in clean aqueous standards. When developing a method for analyzing real-world samples, the defaults form a useful starting point, but will invariably require optimization to obtain the best possible results for any particular type of sample.

Setting up the instrument

Before an analysis begins, both the hollow cathode and deuterium lamps should be switched on for about 15 minutes in order to allow them to stabilize properly. It is important to check that the furnace head unit is correctly aligned with respect to the optical beam of the spectrometer. Failure to do so can result in problems such as emission breakthrough, especially noticeable when the analyte resonance line lies above 300 nm. Figure 5 shows the effects of emission breakthrough on the blank signal with a badly mis-aligned furnace head, and the signal when the alignment is correct.

Preliminary Procedures
The starting point when developing a new GFAAS method is to obtain the set of basic instrumental data for the analytes that are to be determined. This information is available from a variety of sources, and is conveniently collated in the On-Line 'Cookbook' that is provided with the SOLAAR Data Station software supplied with all AA Series spectrometers. The Cookbook includes suggested values of the instrument parameters such as wavelength, bandpass, lamp current and whether background correction is required or not. It also includes information on common interferences and suggests some methods for overcoming them. An example of the Cookbook page is shown in Figure 4.

Figure 5. Effects of emission breakthrough on the baseline signal at the barium wavelength of 553.6nm.

Method Development Considerations

Graphite Furnace AAS is a well-established analytical technique, and there is a large body of applications literature available. It is likely that information already exists for most of the common types of analyses, so that information retrieval is important and should be the rst step in developing new method. The journals of interest to analysts
Figure 4. Example of a SOLAAR Cookbook page.

When a new Method is created, the AA Series spectrometers will load default instrument and furnace parameters from the Cookbook for each element that is selected from the software. It is important to

The furnace head is permanently mounted in the right hand sample compartment of the Dual Atomizer M Series instruments, and is correctly aligned during the installation of the instrument. The mounting of the GFS97 Combined Furnace and Autosampler accessory for the single compartment S Series instruments has been designed so that the accessory can be removed and replaced without disturbing the alignment. Nevertheless, it is good practice on all instruments to check the furnace head alignment periodically, and particularly after the furnace head has been disassembled for routine cleaning.

The furnace autosampler has to be aligned with the furnace head, so that the sample can be correctly deposited into the cuvette. Mechanical adjustments are provided on the autosampler body for this purpose, and, as with the furnace head itself, these have been designed to maintain their alignment even when the autosampler is removed from the instrument and replaced. The FS95 provides replaceable capillary tips, which perform the actually sample deposition, and these can become contaminated, as well as bent or distorted, by careless handling. It is important to check the capillary tip alignment whenever a tip is replaced, and it is good practice to examine the tip and conrm that the alignment remains accurate each time the instrument is used. A small dental mirror can be used to examine the position of the capillary tip inside the cuvette, as shown in gure 6, but this requires a certain amount of skill and experience.

suitable for simple aqueous solutions, and is a good starting point for more complex samples.

Capillary tip is too high - sample is not deposited on cuvette oor

Capillary tip is too low - sample runs back up the outside of the capillary tip Capillary tip is correctly positioned sample is deposited correctly on to the oor of the cuvette. Figure 7. GFTV images, showing the effects of autosampler capillary tip positioning.

All AA Series spectrometers are equipped with the Quadline continuum source background correction system. For this to perform to specication, it is important that both the hollow cathode lamp and deuterium background corrector lamp beams are correctly aligned. AA Series instruments provide automatic alignment of the hollow cathode lamp, but the deuterium arc lamp requires manual alignment whenever it is replaced. Alignment procedures are included in the User Manuals. They have also been published in some journals with a more detailed discussion of the effects of misalignment (3,4).

be calculated by simple ratios that a mass of about 700 pg of aluminum should give a signal with a peak height of about 0.8 A, which would be ideal for the top point on the calibration graph. It is normal to use 10 or 20 L injection volumes, so that the top standard concentration should be equivalent to 700 pg in, say, 10 L i.e. 70 g/L or 70 ppb. The lowest standard should give a signal of about 0.1 A, and in this case, a concentration of 10.0 g/L would be suitable. GFAAS calibrations are not usually perfectly linear, and a minimum of three standards is usually used to dene the curve accurately. The mid-point standard concentration should therefore be selected so that the data point falls mid-way between the blank and the top standard, and a concentration of 35 or 40 g/L would be suitable. The published Characteristic Mass gures are measured using aqueous solutions, with the default instrument parameters. Interferences and changes away from the default parameters can cause signicant reduction of the signal and, therefore, the practical concentration range may well be higher in the presence of interferences.

Furnace Heating program

The Furnace Heating Program is critical to the success of a method development program. It is conventionally divided into four distinct phases, and, as with all other parameters, the SOLAAR software provides element dependent default parameters that will serve as a starting point for program optimization.

Calibration range
Figure 6. Using a dental mirror to examine the interior of the cuvette.

Thermo's AA Series instruments offer a better solution, in the form of the Graphite Furnace TeleVision (GFTV) accessory. This is a small CCD camera mounted within the optical system of the spectrometer so that it can produce a live image of the interior of the cuvette on the Data Station VDU display. The height adjustment of the capillary tip (i.e. the distance between the tip and the oor of the cuvette during the injection sequence) is critical, and is dependent on the viscosity characteristics of the sample. Some typical GFTV images are shown in gure 7, showing the effects of incorrect capillary tip height adjustments, but a position 1 - 2 mm from the cuvette oor is

The next step is to calculate the concentration of the analytical standards (taking due account of the injected volume to be employed) that would be required to give a calibration curve in the optimum range of 0.1 to 0.8 A (measured in the peak height mode). These concentrations can be derived from published characteristic mass values (dened as the mass of analyte, in picograms, required to generate a signal of 0.0044 A) or from the furnace check values in the Cookbook, shown as the concentration of analyte which will give an absorbance reading of 0.1 A when a 20 L volume of solution is injected. As an example, the characteristic mass for aluminum is 3.6 pg, so that a sample injection containing 3.6 pg will give a signal with a peak height of 0.0044 A. It can easily

Figure 8. The default furnace program for lead.

The DRY phase

If the furnace autosampler has been aligned correctly, it should be possible to inject a discrete volume of sample into the cuvette, so that it remains as a droplet on the cuvette oor when the autosampler capillary tip is withdrawn. The rst task is to dry this sample pool, so that only the solid material remains. It is important to ensure that the sample is completely dried before moving on to the next, higher temperature phase, but it also important that the cuvette temperature is

not increased to the point at which the liquid boils. The ideal behaviour is that the solvent gently evaporates, leaving the solid material in the same place on the cuvette oor. If the sample does boil, or undergoes any type of vigorous disturbance, it will splash and become dispersed over the entire inner surface of the cuvette. At best, this will degrade the measurement precision, and in extreme cases may prevent any meaningful signal from appearing. The DRY phase temperature is usually set to a few degrees above the boiling point of the main solvent in the sample - 110 C would be a typical for an aqueous sample. The time required to dry the sample depends on the volume of the sample to be used and the nature of the sample, and 30 seconds would be a typical value. Both the time and the temperature of this phase are best optimized by observation, either using the GFTV image, or using a dental mirror to directly view the sample in the cuvette. The DRY phase largely controls the precision and reproducibility of the nal optimized method, and careful optimization of this phase is necessary to obtain the excellent relative standard deviation values of 5 % or better that GFAAS is capable of delivering.

The ASH phase

When the sample has been successfully injected into the cuvette and dried, the solid material that was dissolved in it will remain on the oor of the cuvette, ready for the next stage in the process. If the sample was a clean aqueous standard, this solid material will consist entirely of the analyte element, in the form of the salt of the major anion present, which is usually nitrate. If this is the case, then a discrete ASH phase will not be required. However, more usually the sample contains other metal salts, and possibly organic materials such as proteins - these are collectively known as the sample matrix components. The purpose of the ASH phase is to remove as much of the sample matrix material as possible before the atomic absorption measurement takes place. This will minimize the occurrence of chemical interferences and background absorbance effects derived from the matrix, and so will enable the analyte signal to be measured cleanly and easily. The ASH phase is normally performed at a fairly high temperature, selected to

volatilize the major matrix components. However, the temperature must not be raised to the point where the analyte itself is volatilized, as then the analyte will be removed from the cuvette, and so will not be available for later measurement. The Cookbook provides suggested maximum ash temperatures for each element. These are derived from measurements in simple nitrate solutions of the element, and may not be applicable if the sample matrix contains a different major anion, or is signicantly different in any other way. The time required to complete the ASH phase will typically be in the range 10 - 60 seconds, and will, of course, be dependent on the nature of the sample and the temperature selected. In general, ASH phases that are too long do not cause any problems other than wasted time, and in the initial investigations, it is usually best to set a reasonably long period, typically 30 45 seconds. As the program is rened, it may be possible to reduce the ASH phase time. Running the full program cycle, so that an analytical signal is generated, then systematically varying the ash phase temperature while observing changes in the signal is the best way to optimize the ASH phase temperature. If the ash phase temperature exceeds the maximum ash temperature that applies to the analyte in the particular type of sample under investigation, the signal will sharply decrease above the critical temperature. If a matrix component is responsible for a particular type of chemical interference, then the signal will increase noticeably as temperature is increased so that the component is removed. This type of experiment is time-consuming to perform, and so the SOLAAR software provides an Ash Atomize function to perform the measurements and gather the data automatically. This will be described below. The ASH phase is the most critical phase in the furnace program, and requires careful optimization. As it affects any chemical interferences that may be present, it directly inuences the overall accuracy of the analysis.

The ATOMIZE phase

When the ASH phase has been successfully completed, the only sample material remaining in the cuvette will be the analyte itself, now usually in the form of an oxide, and any refractory matrix components, such

as calcium or magnesium oxides, that have not been removed in the ASH phase. The cuvette is now heated sufciently to atomize the analyte, and the spectrometer records the atomic absorbance signal. This is the ATOMIZE phase. For each analyte element, there is a minimum atomize temperature that must be used, to ensure that all the analyte is atomised, and, like the maximum ash temperatures, these are provided in the Cookbook, and loaded automatically as part of the default parameter set for the analyte element. It is important that the cuvette is heated as fast as possible at the start of the ATOMIZE phase, so that the analyte is atomized as quickly as possible. This generates a dense atom cloud, and maximizes the analytical sensitivity. The Thermo furnaces provide a Temperature Control feature, that uses an optical temperature feedback system to ensure that not only is the atomize temperature accurately controlled, but also that the temperature risetime is always the maximum that is possible. The Temperature Control system is discussed in more detail in (2), and is almost invariably selected for the atomization phase. The ATOMIZE phase time is selected so that all the analyte is atomized, and the atomic absorbance signal returns to the baseline. Because of the fast temperature risetime provided by the use of the Temperature Control feature, this period can be as short as 2 seconds, and is unlikely to be longer than 5-6 seconds. Excessively long ATOMIZE phases may degrade the measurement precision when peak area measurements are used, and will certainly shorten the cuvette lifetime unnecessarily. Optimization of the ATOMIZE phase temperature is usually performed in the same way as optimization of the ASH phase temperature, by performing an experiment in which the temperature is systematically varied and examining the effects of this on the analytical signal. In general, the Cookbook atomization temperatures will be found to be suitable for many types of samples, but when the sample contains large amounts of involatile matrix components that cannot be removed in the ASH phase, it may be necessary to select a temperature that will minimize the amount of matrix volatilized with the analyte atoms, to minimize both background absorption signals and the possibility of vapo phase interferences.

The CLEAN phase

The nal phase in the furnace program is intended to remove any residual matrix material from the cuvette, so that it is ready to measure the next sample. The temperature and time required for the CLEAN phase are obviously dependent on the nature of the sample, but as a rule of thumb, the temperature should be at least 2500 C, or 100 C above the ATOMIZE phase temperature, whichever is greater. However, cuvette temperatures above 2800 C will result in more rapid wear of the cuvette, and will reduce the cuvette lifetime. They should therefore be avoided if possible. It is sometimes difcult to conrm that the CLEAN phase is properly optimised, as the consequences of an incorrectly set up CLEAN phase are associated with memory effects and longer term drift that will occur as the residues build up in the cuvette.

Practical Steps to Method Optimisation

Preliminary experiments
It is usually necessary to perform a few preliminary experiments to conrm that the instrument is set up correctly, and to establish the feasibility of the proposed analysis. The type of cuvette to be used should be selected, based either on the type suggested in the Cookbook, or from previous work and consideration of the nature of the analyte and sample matrix. Using the aqueous calibration standards (including any additional reagents used to stabilize them or matrix match them) it is then important to: Check the blank levels Check the precision of the measurements by analyzing one of the standards several times (at least 6 to 10 measurements are required for a meaningful value) Run all the standards to check for calibration linearity and Calculate the measured characteristic mass for the analyte and compare it with the Cookbook value. The default parameters are normally suitable for these preliminary experiments. Blank signals should normally be less than 0.05 A, and if possible should be <0.01 A. If this is not the case, then steps should be taken to identify the source of the contamination. Typical sources are the reagents used in preparing the standards (i.e. any acids or matrix modiers), glassware and any other apparatus used to carry out the preparation steps. Even the

sample cups used in the furnace autosampler can cause problems (particularly for low level aluminum, for example). The laboratory atmosphere is also a source of contamination and the use of clean boxes or rooms may be required. Once the source is identied then corrective action can be taken to overcome the problem. Measurement precisions of better than 3 to 5 % relative standard deviation should be obtained from the standard solutions with the FS95 autosampler, when the peak height signals are in absorbance range 0.1 to 0.6 A. If the precision values are poorer than this (higher RSD values), there are a variety of possible causes that must be investigated. The sample injection into the cuvette should be observed carefully, using either a dental mirror or the GFTV accessory, and checked for any misalignment, i.e. liquid deposited on the lip of the injection hole rather than inside the cuvette. A common cause of poor injections is contamination on the outer surface of the capillary tip. This is often the result of handling the tip, and can cause the sample to run back up the outside the tip instead of being deposited properly on to the cuvette oor. Wiping the outside of the capillary tip with a tissue moistened with alcohol will usually solve the problem. If the injection is satisfactory, but the precision remains poor, it is likely that the DRY phase is not set up correctly. The GFTV image can show if the sample is boiling, indicating that the drying temperature should be reduced, or that the sample is not drying completely, so that either or both of the phase temperature and time should be increased. It is possible, although difcult, to observe the sample drying with a dental mirror. It is important to appreciate that atomic absorption calibrations are inherently curved, and that the SOLAAR software provides sophisticated calibration algorithms to correct the effects of this curvature. The software will identify calibrations where the curvature is excessive, beyond the correction ability of the algorithm, and will both ag the data and display a suitable error message. If the FS95 is being used to prepare the working standards by automatic dilution of a master standard, then the calibration linearity should normally be typical for the analyte element. Problems here are normally associated with contamination of the diluent solution used, and measuring a sample of the diluent alone should conrm this. It is not possible to correct calibration

data for the effects of a contaminated diluent, and the only effective remedy is to trace the source of the contamination and remove it. If the calibration standards have been prepared manually, it is possible that concentration errors and/or contamination may be present in one standard solution. It is often possible to identify the incorrect standard by visual inspection of the calibration plot displayed by the software, and facilities are provided to allow bad calibration points to be removed from the calibration. Nevertheless, it is preferable to identify and resolve the source of the problem. When the calibration measurements are complete, the SOLAAR software will display the calibration plot and various gures of merit, including the curvature and correlation coefcients. It will also calculate the characteristic concentration (the concentration equivalent to a signal 0.0044 units above the blank measurement). An example is shown in gure 9.

Figure 9. A calibration graph display in SOLAAR software

This characteristic concentration will be reported in the concentration units that have been specied for the calibration standards. To compare the result with the cookbook value, the actual mass of the analyte equivalent to the reported characteristic concentration should be calculated, by taking account of the volume of solution injected into the cuvette. This characteristic mass should typically be close to the Cookbook value for the element if peak height measurements are used. Variations over the range of 0.5 x to 2 x the Cookbook value can be expected, and are due to individual difference between different instruments, hollow cathode lamps and cuvettes, but variations outside this range may require further investigation.

Sample measurements
When satisfactory measurements of the calibration standards have been obtained, investigations using the real sample solutions can begin. If the composition of the samples is similar to that of the calibration standards, then it is likely that useful results will be obtained from the samples immediately, but if there are signicant differences, it may be necessary to repeat the optimization of the injection and the DRY phase. A sample should be chosen that gives a signal that is somewhere near the middle of the calibration graph. If all the samples give very low signals, then it may be necessary to spike a sample with the analyte element so that a measurable signal is produced that can be optimized - it is not practicable to attempt to optimize very small signals, as the effects of the changes are obscured by the noise on the signal. If the signals from the sample are very high, it may be necessary to review the sample preparation procedure, and perhaps increase the sample dilution.

Background correction
Samples analyzed by GFAAS normally exhibit signicant background (non-specic) absorption signals when the wavelength of the analyte element is in the ultra-violet region of the spectrum, below 400 nm. Such background signals can be corrected by the automatic background correction systems provided by the spectrometers. The S Series and the M5 spectrometers are equipped with the Thermo Quadline continuum source background correction system, while the M6 and MQZ spectrometers also provide background correction based on the Zeeman optomagnetic effect. The Zeeman background correction system is capable of correcting for certain types of structured background that cannot be corrected accurately by the Quadline system, but will also degrade both the sensitivity and the noise levels of the measurements, resulting in poorer detection limits than those obtained from the Quadline system. Background correction is discussed in more detail in (6). Selection of the optimum background correction system to use, on spectrometers such as the M6 and MQZ that offer a choice, is sometimes difcult. It is suggested that the Zeeman system should be used for the initial investigations, as it is more likely to give the correct results. Towards the end

of the method development process, an experiment to compare the performance of the two systems should be carried out, as signicantly better sensitivity and detection limits can often be obtained from the Quadline system, if it is capable of correcting accurately for the background signals. The analytical signals derived from the sample solutions should be examined, and the size of the background absorbance signals noted. Both the Quadline and Zeeman background correction systems are capable of correcting for background signals up to 2 A in peak height, but is preferable that the background signals should be less than this if possible. Large background signals often go along with signicant chemical interferences, as they indicate that a large amount of matrix material is being volatilized with the analyte atoms, which can result in gas phase reactions taking place that reduce the analyte atom population. There are various techniques available for reducing the background signals, including the Ash Atomize experiment and the use of matrix modication both described below.

Either or both of the Ash and Atomize parts of the experiments can be investigated in a single run. It is recommended that the ATOMIZE phase temperature should be optimized rst, as the optimum is usually fairly close to the default value. The start temperature, end temperature and temperature increment for the relevant phase can be specied, and the instrument will then proceed to make the measurements, and display the results as an Ash Atomize plot, shown in gure 11.

Figure 11. An Ash/Atomise plot for lead in natural waters

The Ash Atomize experiment

An Ash Atomize experiment permits the effects of changes in the critical ASH and ATOMIZE phase parameters to be investigated in a systematic manner, and as the SOLAAR software provides facilities for performing the experiment automatically, it should be considered as an essential part of any GFAAS method development work. It is necessary to perform the preliminary work described above, so that reasonably stable measurements, with signals magnitudes well within the range covered by the calibration can be obtained from a typical sample solution. The Ash Atomize experiment itself is set up from the Ash Atomize dialogue box shown in gure 10.

Figure 10. The Ash Atomize dialogue box.

It is possible to display the results on an Ash Atomize plot in various ways, showing any combination of the Total, Background and Corrected absorbance signals, with the data presented in peak height or peak area measurements, and it is even possible to add a line showing the effect of the temperature changes on the precision of the measurements. Interpretation of the data presented on the Ash Atomize plot can be difcult for an inexperienced analyst, and so the software attempts to determine the optimum settings. As with all such automatic processes, these optima should be considered as a guide only, and are not a substitute for thoughtful consideration by an experienced analyst. The optimum ATOMIZE phase temperature will be the temperature that gives the maximum Corrected absorbance signal, consistent with the minimum Background absorbance signal and the best measurement precision. It may be necessary to review the peak shape of the analytical signals to conrm that clean signals without distortions or other artefacts are obtained. Finally, the lowest temperature that meets all these criteria should be selected as the optimum, and the Furnace Program should be updated with this value. athan the original sample concentration to allow the differences to easily be measured.

The experiment can then be repeated, this time investigating the ASH phase. The criteria for selecting the optimum temperature are similar to those used for the ATOMIZE phase temperature, but it is possible that the optimum found may be quite different from the default value. If this is the case, it may be necessary to re-optimize the ATOMIZE phase temperature using the new value of the ASH phase temperature. Finally, it is often useful to run the whole experiment again, varying both the ASH and the ATOMIZE phase temperatures over a narrow range around the previously determined optima to rene the choices made and obtain a fully optimized furnace program. At this point, it is worthwhile to re-run the calibration standards again using optimized program, to conrm that the calibration is still satisfactory. It is quite possible, for example, that the furnace program changes have resulted in a slightly different value for the analyte characteristic mass - this should not cause concern, as long as the change is moderate.

Interferences
At this point in the method development process, it will be possible to obtain a good calibration curve and good, reproducible signals from the sample solutions. It is therefore possible to perform a full sample run and obtain an analyte concentration value for the sample solution. The next step is to conrm that this is likely to be the correct concentration value. There are two well-established techniques for conrming the correctness of an analytical result. These are: The analysis of one or more reference materials The analysis of one or more spiked samples Reference materials (often referred to as Certied Reference Materials or CRM's) are stable, carefully characterized material representative of the type of sample being analyzed. They contain known concentrations of the analyte elements that have usually been determined by more than one analytical technique. They may be produced by external organizations, or may be inhouse materials that have been prepared and reserved for this purpose. It is relatively simple to use a CRM to check for interferences that might be present in the proposed method. One or

more CRM's should be selected that are as similar as possible to the actual samples that require analysis, and contain the analytes at approximately the same levels. The CRM's are then prepared and analyzed in the same way as the samples, and the nal concentration results obtained are compared to the certied values provided with the CRM. Most certied values are provided with a range of uncertainty limits about the mean certied concentration, and if the measured result falls within these limits, then the proposed method can be considered to be free of signicant interferences. If uncertainty limits are not provided, measured concentrations within a range of 90 - 110 % of the certied value are usually considered to be acceptable. Spiked samples can be used when suitable a suitable CRM is not available. A sample is selected that is representative of the type of sample being analyzed, and that has a relatively low concentration of analyte. A portion of this sample is taken, and the analyte concentration is increased (spiked) by adding a known amount of the analyte element. The way in which the analyte spike is added is dependent on the type of sample and the sample preparation procedure. If the sample requires extensive preparation procedures, it is preferable to add the spike at the beginning of these procedures. The size of the spike should be chosen so that the concentration of the spiked sample remains within the calibration range of the method, and is also sufciently larger than the original sample concentration to allow the differences to easily be measured. The analyte concentrations in both the original and the spiked samples can then be measured, and the recovery of the spike can be calculated by subtraction. The measured spike concentration can then be compared with the known added value, and the actual recovery can be calculated. The actual recovery of the spike is usually presented as a ratio of the original concentration added. Spike recoveries in the range of 90 - 110 % are usually considered to be excellent, and for many analyses, a wider range may be acceptable. The FS95 autosampler is capable automatically adding a spike to a sample solution, and the SOLAAR software will calculate the spike recovery. This provides a relatively quick and simple check for interferences that affect the actual measurement, but will not identify any

problems that may occur in the sample preparation steps. Analyte recoveries from either CRM's or spiked samples that are greater than 110 % or so of the expected values are usually indicative of a contamination problem somewhere in the process. While there are a few known GFAAS interference effects that result in concentration enhancement (typically those caused by structured background absorbance signals), these are rare, and potential sources of contamination should be investigated rst. Low recoveries are indicative of the presence of interferences. Various strategies are available to enable these to be identied and overcome, which are discussed below, but it is important to recognize that some types of interference cannot be completely suppressed. In these cases, an alternative calibration strategy, such as standard additions calibration, may be appropriate, or changes to the sample preparation procedure, for example, to remove the interfering species before the analysis, may be required.

Interference control
Interferences that can occur in GFAAS measurements fall into one of three possible classes. These are: Spectral interferences, particularly those caused by background absorbance signals that display ne structure in their wavelength dependence Physical interferences, where the physical characteristics of the sample are signicantly different from the characteristics of the standards used Chemical interferences, where residual sample matrix components react with the atomized analyte in the ATOMIZE phase, and reduce the concentration of free atoms available for measurement.

Spectral interferences
True spectral interferences, where a matrix component has an absorption line that overlaps the analyte absorption line, are very rare in atomic absorption spectrometry, and can usually be ignored. Most instances of spectral interference are actually caused by matrix components that exhibit broadband, non-specic absorption with ne structure on the same wavelength scale as the atomic absorption line. Continuum source background correction systems, such as the Quadline system provided by the AA Series spectrometers, are unable to correct

accurately for such this type of non-specic absorption, and correction systems based on the Zeeman effect are required. It is sometimes possible to identify structured background interferences by examination of the Quadline background signals displayed by the SOLAAR software. Negative background signals that go signicantly below the zero absorbance baseline, especially if they then suddenly jump to large positive values, are indicative of problems in this area. Alternatively, the analysis can be repeated at a different wavelength if one is available - if the same or similar concentration results for the same sample are obtained at both wavelengths, then it is unlikely that a spectral interference of any kind is present, and if the background signal shows distortion at one wavelength, but is normal at another, then it is more likely that the second wavelength will give the correct result. If both Zeeman and Quadline background correction systems are available, as they are on the M6 and MQZ spectrometers, then it is an easier matter to identify this type of interference. All that is required is to measure the same sample using each form of background correction. If the concentration results are the same, then it is unlikely that this type of interference is present, and Quadline correction will usually be the best choice. If the concentration results are different, and especially if the Quadline background signal shows unusual distortion and artefacts that are not present on the Zeeman background signal, then it is likely that structured background interference is present, and the Zeeman result is more likely to be correct.

Physical interferences
Physical interferences occur when the physical properties of the samples are very different from the properties of the standards used to generate the calibration. Typical examples are found in clinical analyses, where the viscosity of body uids such as blood serum, is markedly different to that of the aqueous standard solutions. These differences affect both the way in which the sample is deposited in the cuvette, and the way in which it is dried. Low viscosity samples, such as simple aqueous samples, tend to be deposited in to the cuvette in the form of a 'pool', where the liquid wets the graphite surface and spreads out over the cuvette oor until it reaches the internal ridges in the cuvette

that prevent it from moving out of the central hot zone. This is a generally desirable behaviour, as it ensures that maximum surface area of the sample liquid is in contact with the heated cuvette, so that the solvent evaporation (drying) and sample decomposition (ashing) occur smoothly. More viscous samples, and those that do not easily wet the graphite surface, have a tendency to be deposited in the form of a 'drop'. This type of behaviour is less desirable, as thermal contact between the sample and the heated cuvette is less good, which can lead to local over-heating during the DRY phase, and non-reproducible drying and ashing behaviour. Problems of this sort can often be clearly identied from the live GFTV images. The FS95 provides facilities to vary both the speed at which the sample is taken up from the sample cup, and the speed at which it is injected into the cuvette, which can help to alleviate the effects of this type of interference. Adding surfactants to the samples and standards modies the sample viscosity and wetting behaviour and can also help to overcome this type of interference. This technique is so successful that most published clinical analyses now involve the addition of Triton X-100, a readily available non-ionic surfactant, to the sample, typically at concentrations between 0.1 and 0.5 % m/v. As with any procedure that involves addition of reagents to samples, it is important to check the contamination levels of the surfactant before using it. It is also normal to add the surfactant to the liquid used to rinse the autosampler between measurements, to prevent crosscontamination and carryover between samples. Surfactants can exacerbate the tendency of some types of sample to foam as they are dried, which can lead to irreproducible results and the accumulation of carbon residues in the cuvette. The addition of an anti-foam agent, such as a silicone emulsion, has been suggested as a means to overcome the foaming problem, and does work successfully in certain cases. A different type of physical interference can occur when the samples to be analyzed are in a non-aqueous solvent, rather than water or dilute acid. Many such solvents have very low viscosities and are highly mobile. They spread rapidly along the cuvette oor as they are injected, and in some cases will ow right over the internal cuvette ridges and onwards out of the

cuvette altogether. This causes severe problems in the measurement, and can result in signicant contamination of the furnace head itself These highly mobile solvents are also usually highly volatile, and this provides a means to overcome the problem. The GF95 and GF95Z furnace in combination with the FS95 autosampler allow the cuvette to be pre-heated before the sample is injected. If the pre-heat temperature is chosen correctly, it is possible to evaporate the sample as fast as it is deposited on to the hot surface, and so completely prevent the sample from spreading. This technique has the added advantage of concentrating the solid analyte and matrix components in the centre of the cuvette, resulting in excellent sensitivity.

Chemical interferences
Chemical interferences occur when matrix components remain after the ASH phase, and either prevent the analyte from being fully atomized in the ATOMIZE phase (the so-called 'solid phase' interferences), or co-volatilize with the analyte and react with the free atoms as they are formed ('vapor phase' interferences). Both types of chemical interference reduce the free atom concentration in the cuvette during the measurement phase, so that the result is lower than it would be in the absence of the interferent. Chemical interferences are unfortunately common in GFAAS, and there are a few basic ground rules that will help the analyst to predict possible problems. Refractory analyte elements, such as aluminum, which have high atomization temperature also usually have high maximum ash temperatures, so that the interfering matrix components can be ashed away at high temperatures without loss of analyte and, therefore, the determination of these elements is usually free of vapor phase interferences. Vapor phase interferences are more common for the volatile elements, particularly when they are present in sample matrices that contain appreciable quantities of alkali metal salts such as sodium chloride. The twin techniques of matrix modication and platform atomization have been developed to overcome many of these effects, and these are described below. Solid phase interferences are more difcult to predict and to overcome. They are likely to occur when the sample contains a large amount of a matrix component that will decompose to a refractory oxide, such

as magnesium or calcium. The analyte becomes trapped in the crystal lattice of the oxide, and is only released slowly, if at all, in the ATOMIZE phase. Methods for overcoming this type of interference usually involve some form of matrix matching, in which an excess of the interfering component is added to both sample and standards, together with extensive modications of the furnace program in an attempt to break down the crystal lattice that is trapping the analyte. Solid phase interferences can also occur when the matrix contains a large excess of a metal that is less volatile than the analyte. The analyte can then form an alloy or solid solution with the matrix component, and it is necessary to increase the atomization temperature to the point at which the involatile matrix component, not just the analyte itself, is volatilized. This type of behaviour is sometimes exploited deliberately, in the form of matrix modication, to control the behaviour of volatile analytes.

phase and ATOMIZE phase temperatures provided in the Cookbook are almost all measured without using modiers, and so will require optimization if a modier is to be used. The automated Ash Atomize function described above is the best way to perform this optimization, and a full Ash Atomize plot from the sample with the modier addition should always be measured. Many types of modiers have been proposed and examined as the technique has developed, and the uses of the most useful and common specic modiers are discussed below.

it is more usual to add it during the sample preparation and dilution steps.

2. Nickel (10 - 50 g for each injection)

Nickel has been used for many years to stabilize elements such as antimony, arsenic, bismuth, selenium and tellurium in the graphite furnace. Nickel salts are readily reduced to nickel metal by the carbon present in the cuvette, and the analyte elements react with the metal to form thermally stable compounds, such as nickel arsenide. Many arsenic compounds, such as the halides and oxides are volatile, and evaporate at temperatures below 500 C. Nickel arsenide, on the other hand, is a stable compound with a well dened boiling point in the region of 1200 C. By adding nickel to the sample, therefore, the maximum ash temperature can be raised to 1100 C or so, which will allow many troublesome matrix components to be removed without loss of the arsenic analyte. It is, of course, necessary to increase the ATOMIZE phase temperature as well, to ensure that arsenic is fully atomized so that the maximum sensitivity is obtained. Nickel matrix modier is usually added automatically by the FS95 to each discrete sample aliquot as it is injected into the cuvette. Typically, a modier solution is prepared from nickel nitrate that contains approximately 2 % m/v of nickel nitrate hexahydrate salt in deionized water, and a 5 L portion of this is added to each injection. As the nickel is added in large excess, neither the actual concentration nor the volume added are critical. Other transition metal salts can be used in the same way as nickel, and copper has been shown to be an effective modier for the arsenic group elements. However, these modiers do have one major drawback. Because they are added in relatively large quantities, the furnace inevitable becomes contaminated with the modier metal, and this can cause severe problems if it is later required to perform trace analyses of the same metal.

1. Nitric acid and ammonium nitrate (0.1 to 5.0 % v/v)

Both nitric acid and ammonium nitrate have been used to remove halide salts from the graphite cuvette during the ASH phase of the furnace program. Ammonium nitrate reacts with halide salts to form volatile ammonium halides that volatilize easily at temperatures of about 350 C. Nitric acid reacts to form volatile hydrogen halide gas, which is usually volatilized in the DRY phase. The anions remain as the nitrate salts that decompose readily to the oxides as they are heated, and the alkali metal oxides then volatilize. Sodium chloride, for example, will not volatilize until it reaches a temperature of around 1200 C. In the presence of ammonium nitrate, however, an exchange reaction takes place and the ammonium chloride volatilizes at 350 C, followed by the sodium oxide at about 700 C. The ASH phase temperature required to remove the sodium chloride is therefore reduced from 1200 C, well above the maximum ash temperature for volatile elements such as lead, to 700 C, where lead is not lost. Nitric acid is probably the preferred modier since it is simple to handle and can be obtained readily in a puried state. It is also normally added to samples and standards to lower the pH and stabilize the analyte metals in solution. For matrix modication purposes, nitric acid should be added until it is just in excess of the halide concentration present in the sample. Standards should be treated similarly. High concentrations of nitric acid should be optimized carefully, and only the minimum concentration required to achieve the desired effect for a particular sample type should be used. Although the acid can be automatically added to the sample by the FS95 immediately before the sample is injected into the cuvette,

Matrix Modication
The term 'matrix modication' describes the process of adding a reagent to the sample to modify the thermal behaviour of the matrix or, in some cases, the analyte during the furnace program. Specic reagents can be used to: Stabilize the analyte during the ashing stage to permit a higher ashing temperature to be used Convert a matrix into a more volatile form which can be removed at lower temperature during the ashing stage Delay the atomization of the analyte to establish isothermal conditions inside the graphite cuvette, which, especially when used with platform atomization, can overcome many vapor phase interferences. In some situations more than one matrix modier may be necessary to achieve the desired effect, and the whole topic of matrix modication has acquired the reputation of being a poorly understood 'black art'. This, of course, is not true, and a basic understanding of simple chemistry can usually be used to predict the effects of a particular modier. The purpose of a matrix modier is to modify the thermal behaviour of the analyte and sample matrix. This implies that the use of a modier will require changes to the furnace program. The default furnace ASH

3. Ammonium phosphate (50 - 100 g for each injection)

Ammonium phosphate is used to stabilize analytes such as lead, tin and cadmium. The phosphates of these metals are reasonably temperature stable, which permits ASH phase temperatures up to 700 - 800 C to be used without loss of the

analyte. Also, the basic component of the molecule helps to volatilize halides in the sample matrix, in the same way as ammonium nitrate. Ammonium phosphate is available in the monobasic form ([NH4]H2PO4) or the dibasic form ([NH4]2HPO4). Both are equally useful as modiers for the determination of elements such as cadmium, lead and tin. The choice between the mono- and di-basic forms depends largely on the contamination levels present in the commercially supplied materials. Unfortunately, high concentrations of phosphate increase the background signal level, and at some wavelengths exhibit structured background absorption that requires a Zeeman effect background correction system, so care should be exercised when assessing the advantages of this modier.

5. Magnesium nitrate (50 - 500 g for each injection)

Magnesium nitrate is often used in conjunction with palladium, but has also been recommended for use alone. The nitrate anion can provide a source of oxygen to assist in the decomposition of organic matrix components, but the more important mode of stabilization involves the magnesium oxide that remains. Magnesium oxide has a loose, open crystal lattice, and will effectively trap other metals within the lattice. These will not be released until the temperature reaches the point when the crystal lattice itself breaks down. While magnesium nitrate does allow the use of higher ASH phase temperatures, the main benet is to delay the atomization of the analyte, and so it is used extensively with platform atomization, where both the atomization mechanism and modier action work together to slow the atomization of the analyte so that the gas phase temperature inside the cuvette has stabilized, and vapor phase interferences are minimized.

4. Palladium (2 - 50 g for each injection)

Palladium is one of the newer matrix modiers to be employed in GFAAS analyses. It works in the same way as nickel, in that the analyte metals are trapped as solid solutions or alloys in the palladium metal. It is effective because the boiling point of palladium is high, so that the analytes can be stabilized to high temperatures. Several studies have shown that palladium is most effective when it is reduced to the metal as soon as possible in the furnace cycle. It has therefore become common practice to add a mild reducing agent, such as ascorbic acid or hydroxylamine hydrochloride to the palladium solution. These mixed solutions are unstable, and will gradually deposit metallic palladium in the sample containers and autosampler capillary tips. It is possible to use an internal furnace gas that contains hydrogen - typically 5 % hydrogen in argon, that will also effectively reduce the palladium. This technique does not require the addition of reducing agents to the palladium solutions, and so is sometimes preferred. Palladium is often used in conjunction with magnesium nitrate, and this mixture has been proposed as a 'universal' modier, suitable for use in all situations. Given the very wide variety of samples that can be analyzed by GFAAS, it seems most unlikely that a genuinely 'universal' modier exists, and it is generally preferable to investigate and develop the best methodology for each type of sample individually.

Platform Atomization
With a conventional type of cuvette, the cuvette walls are heated by the electric current from the power supply. The gas inside the cuvette is heated by conduction of heat from the hot cuvette walls, so that the temperature of the gas lags behind that of the walls. When an analyte is volatilized from the cuvette wall, it is atomized into a gas phase that is relatively cooler than the surface from which it has been volatilized. If the sample also contains residual matrix components that have not been removed in the ASH phase, then this lower temperature causes the free atoms to re-combine with the matrix salts to form molecular species, which cannot be measured by the atomic absorption spectrometer. Analyte atomic concentration is reduced and hence the interference manifests itself as a signal depression when compared to a simple aqueous standard. Such an effect is known as chemical vapor-phase interference and affects mainly the determination of volatile elements in halide-rich matrices. Platform atomization attempts to reduce or eliminate this type of interference by delaying the atomization of the analyte until the gas phase inside the cuvette has reached the same temperature as the cuvette walls. Platform cuvettes contain a small piece of graphite located inside the

cuvette, on to which the sample is placed. The platform is designed so that it makes very poor thermal contact with the cuvette, and so is heated mainly by radiation from the cuvette wall and conduction from the gas phase. The temperature of the platform therefore lags behind that of the cuvette wall. When the platform does eventually reach the temperature necessary to atomize the analyte, the gas phase temperature has stabilised. The analyte is therefore atomized into a gas phase that is at the same temperature as the atomization surface, which reduces analyte molecular formation and, therefore, reduces vapor phase interferences. It is usually necessary to use a matrix modier such as magnesium nitrate to further delay the atomization of the analyte to obtain the most effective interference control. Platform atomization is discussed in more detail in (3). Figure 12 shows the cross-section of the latest OMEGA platform cuvette that employs ELC technology to produce long lifetimes and good performance with near refractory elements such as chromium.

Figure 12. Cross-section of an OMEGA platform cuvette.

Conclusions
The development of satisfactory GFAAS methods can be made a much simpler task by a well thought out, systematic approach. Using a logical, step-by-step procedure, it is possible to tackle problem areas one at a time, and thus ensure that the nal developed method represents the optimum for that sample. If these simple guidelines are followed, a reliable working method will be obtained and poor quality data avoided.

In addition to these offices, Thermo Electron Corporation maintains a network of representative organizations throughout the world.

References
1. Design Features of the GF95 Graphite Furnace Accessory. Thermo publication PS40701 2. Design Considerations for a new Platform Cuvette for Graphite Furnace Atomic Absorption Spectrometry. Thermo publication PS40704 3. A A Brown and P J Whiteside, Intl. Labmate, 1984, 9 Issue 3, June. 4. A A Brown, Anal. Chim. Acta, 1985, 175, p319. 5. Design Considerations for High Performance Background Correction Systems in atomic Absorption Spectrometry. Thermo publication PS40690
Australia +61 2 8844 9500 Austria +43 1 333 50340 Belgium +32 2 482 30 30 Canada +1 800 532 4752 China +86 10 5850 3588 France +33 1 60 92 48 00 Germany +49 6103 4080 India +91 22 2778 1101 Italy +39 02 950 591 Japan +81 45 453 9100 Netherlands +31 76 587 98 88 Scandinavia +46 8 556 468 00 South Africa +27 11 570 1840 Spain +34 91 657 4930 Switzerland +41 61 48784 00 UK +44 1442 233555 USA +1 800 532 4752

www.thermo.com

FM 09032

Thermo Electron Limited, Cambridge, UK is ISO certied.

2005 Thermo Electron Corporation. All rights reserved. All trademarks are the property of Thermo Electron Corporation and its subsidiaries. Specifications, terms and pricing are subject to change. Not all products are available in all countries. Please consult your local sales representative for details.
TN40699_E 12/05C