Computer-assisted SIMPLEX optimisation of an on-line preconcentration system for determination of nickel in sea-water by electrothermal atomic absorption spectrometry Ma Teresa Siles Cordero, Elisa I. Vereda Alonso, Pedro Can�ada Rudner, Amparo Garcý�a de Torres and Jose� M. Cano Pavo�n* Department of Analytical Chemistry, Faculty of Sciences, University of Ma�laga, E-29071, Ma�laga, Spain Received 22nd February 1999, Accepted 28th April 1999 An original on-line preconcentration system for the determination of nickel by electrothermal atomic absorption spectrometry is achieved by replacing the sample tip of the autosampler arm by a microcolumn packed with a silica gel chelating resin functionalised with 1-(di-2-pyridyl )methylene thiocarbonohydrazide.The modification of the autosampler in the tubing line and circuit allowed either the flow of the sample through the column or the operation of the autosampler in the normal mode, where microlitres of HNO3 2 M, which acts as the elution agent, pass through the microcolumn, eluting Ni(II), which is directly deposited in the graphite tube as a drop of a precisely defined volume.Optimum graphite furnace operating conditions were sought by software which integrates the SIMPLEX optimisation method and quality requirements such as sensitivity and precision. The optimised method has two linear calibration ranges, between 0 and 1 ng ml-1 and between 1 and 5 ng ml-1, with a detection limit of 0.06 ng ml-1 and a throughput of 36 samples h-1 for 60 s of preconcentration time.The accuracy of the method was examined by the analysis of a certified reference material and by determining the analyte content in spiked seawater and synthetic sea-water. The results show suYciently high recoveries. The accurate determination of trace and ultratrace elements resins as packing materials is simpler and less time consuming than other options. Chelating resins, such as Chelex-100,4–7 in sea-water is one of the most important and challenging tasks in analytical chemistry. The determination of heavy Muromac A-1,8–10 quinolin-8-ol,5,11–14 poly(dithiocarbamate) 15,16 and diethyl dithiophosphate on C18 bonded silica,17 metals is often required for the routine monitoring of marine environmental pollution and for ocean modelling programmes.have been used for the enrichment of natural waters and biological materials prior to their measurement by flame Although electrothermal atomic absorption spectrometry (ETAAS) has very low detection limits for trace metals in atomic absorption spectrometry (AAS), inductively coupled plasma atomic emission spectrometry (ICP-AES) and oV-line aqueous solution,1 the direct determination of trace metals in sea-water by ETAAS is diYcult even with sophisticated back- ETAAS.On-line flow injection column preconcentration in atomic spectrometry was reviewed by Fang et al.18 Based on ground correction and chemical modification.This is due to the low concentrations and strong interference from the sample solid-phase extraction with C18 silica gel, Sperling and co-workers2,3,19,20 modified the on-line flow injection system matrix. ETAAS with on-line sorbent extraction separation and preconcentration can solve these problems and lead to easy of a flame atomic absorption spectrometer to achieve feasible determinations with ETAAS. Liu and Huang used C18 silica determination.Flow injection (FI ) as a microsample introduction system gel and ammonium pyrrolidine dithiocarbamate (APDC) for the determination of Cu, Cd and Pb21,22 and Muromac A-1 oVers some distinct advantages over manual batch-type procedures, such as fully automated sample management and for Cu and Mo in sea-water.23 Yan et al. employed APDC for the determination of Sb in sea-water24 and diethyl dithi- operation in a closed system. Although FI, as the name implies, handles flowing streams, FI preconcentration and ophosphate for Pb in biological and environmental samples,25 both using a knotted reactor. matrix separation on solid sorbents is a discontinuous process which is appropriate for the discrete, non-flow-through nature In developing an ETAAS method, one needs to adjust many variables (injection volumes, drying, ashing and atomisation of ETAAS.2 While with flow-through detectors the loading time is wasted in waiting for elution, this time can be fitted times and temperatures) in order to establish optimum conditions for the analysis.This can be very time-consuming if a perfectly into the cycle of a furnace programme. The closed nature of the FI system, together with the direct coupling to conventional univariate optimisation is undertaken manually and, where interactions exist between the variables, one is the sample introduction capillary of an autosampler, avoiding any collection of samples in vessels exposed to the environ- unlikely to find the true optimum.26 The problem is greater if the optimisation procedure must be in accordance with the ment, reduces contamination problems to a minimum. As the analyte element is eVectively separated from the matrix and quality policy of a particular laboratory.Of the direct optimisation methods, the SIMPLEX method (first introduced by an almost pure solution is injected into the furnace, chemical modifiers are no longer required, which further reduces con- S.N. Deming and co-workers27,28 in Analytical Chemistry) has been the most commonly used in analytical chemistry tamination problems. On-line reagent purification is another feature of FI which helps to keep blank values at a low level, applications. A deduced alternative that has been widely accepted by analytical chemists was devised by Spendley et al.29 permitting determinations in the ng l-1 range for a large number of elements.3 and modified by Nelder and Mead.30 This useful algorithm is known as the modified SIMPLEX method, MSM; however, Of these systems, column preconcentration using chelating J.Anal. At. Spectrom., 1999, 14, 1033–1037 1033Table 1 Graphite furnace temperature programme (Vi=20 ml ) it optimises a single parameter. Successful use requires the selection of a response function which weights a number of Temperature/ Ramp Hold Argon flow rate/ variables. Recently,31 the possibility of coupling a mathemat- Step °C time/s time/s mlmin-1 ical model with optimisation algorithms was evaluated following an insight into the response function.In this paper, a 1 110 1 13 250 2 130 5 21 250 methodology that can alleviate the above-mentioned diYculties 3 1060 11 22 250 is proposed for the optimisation of the graphite furnace 4 2300 0 5 0 (read 5 s) programme; it is based on the coupling of a descriptive 5 2400 1 4 250 mathematical model of the response function developed by Vereda et al.32 with a SIMPLEX algorithm, which are programmed into software, using quality criteria previously selecby soaking in 0.1 M hydrochloric acid) from coastal surface ted.With this software, seven variables were optimised towards water of La Cala del Moral beach, Ma�laga, Spain, and quality requirements such as sensitivity and precision. analysed in less than 24 h. In this paper, a simple on-line preconcentration system, achieved by replacing the sample tip of the autosampler arm Instrumentation and procedure by a microcolumn packed with a silica gel chelating resin functionalised with 1-(di-2-pyridyl )methylene thiocarbonohy- A Perkin-Elmer (Norwalk, CT, USA) Zeeman/4100 Z1 atomic drazide (DPTH-gel ), is developed for the determination of absorption spectrometer equipped with an AS-70 furnace nickel by ETAAS.The modification of the AS-70 autosampler autosampler was used throughout. Pyrolytic graphite coated in the tubing line and circuit allowed either the flow of the tubes with pyrolytic graphite platforms were used in all sample through the column or the operation of the autosampler experiments.The light source was a nickel hollow cathode in the normal mode, where microlitres of HNO3 2 M, which lamp operated at 25 mA; the wavelength was set to 232.0 nm acts as the elution agent, pass through the microcolumn, with a spectral slit width of 0.2 nm. The optimised graphite eluting Ni(II), which is directly de graphite tube furnace programme, by the SIMPLEX method, is shown in as a drop of a precisely defined volume.The accuracy of the Table 1. The software of the instrument permits the pipetting method was examined by the analysis of a certified reference speed of the pump of the autosampler to be varied from 40% material and by determining the analyte content in spiked sea- to 100%; thus a pipetting speed for the autosampler of 40% water and synthetic sea-water. The results showed good was programmed to increase the contact time of the eluent agreement with the certified value and suYciently high with the microcolumn, which produces a greater eYciency recoveries.of elution. The microcolumn containing the DPTH-gel was a glass tube (3 cm×3 mm id) packed to a height of 0.5 cm; at both ends Experimental of the microcolumn, polyethylene frits (Omnifit, Cambridge, Reagents and samples UK) were fixed to prevent material losses. On the end of this column was placed a piece of sample capillary of the sampler High-purity reagents were employed in all experiments.For arm, in imitation of the sample tip of the sampler arm. Details the synthesis of DPTH-gel, the following were used: silica gel on the design of this microcolumn are given in Fig. 2. Thus (particle size, 0.2–0.3 mm), 3-aminopropyltriethoxysilane and the sample tip of the sampler arm was replaced with this diglutaric aldehyde were purchased from Fluka (Buchs, microcolumn, permitting normal working of the sampler.Switzerland), thiocarbohydrazide and di-2-pyridyl ketone were A peristaltic pump, P (Gilson Minipuls 3, Villiers, France), supplied by Aldrich Chemie (Steinheim, Germany), ethanol fitted with a vinyl pump tube (1.65 mm id), was used for and toluene were obtained from Carlo Erba (Milan, Italy) loading of the sample. A Rheodyne (Cotati, CA, USA) Type and hydrochloric acid was supplied by Merck (Darnstadt, 50 six-port rotary valve was used as a switching valve.Germany). The synthesis and characterisation of DPTH-gel Transport lines were made using 0.5 mm id Teflon tubing. The were described in a previous paper.33 The structure of DPTHFI manifold is shown in Fig. 2. It operated as follows: during gel is shown in Fig. 1. the 1 min sample loading period, a 3.4 ml min-1 flow of sample A stock solution of Ni(II) was prepared from the nitrate (standard or blank) at pH 9.0, buVered with boric acid–NaOH, (Merck) and standardised gravimetrically; standards of is pumped (via P) through the microcolumn ( located in the working strength were made by appropriate dilution as sampler arm); the metal ion is adsorbed on the sorbent required, immediately prior to use.A pH 9 buVer was prepared by mixing boric acid (Merck) 0.2M with NaOH (Merck) until pH 9 was reached. HNO3 (Merck) 2 M was used as eluent. Glycine (Merck) 0.2M was used to mask interferent ions. De-ionised water (18 MV cm-1) was used throughout. The certified reference material was Community Bureau of Reference (BCR) CRM 505 estuarine water.The composition of the synthetic sea-water was (in g l-1): 27.9 of NaCl, 1.4 of KCl, 2.8 of MgCl2, 0.5 of NaBr and 2.0 of MgSO4. Sea-water was collected in poly(propylene) bottles (previously cleaned Fig. 2 Schematic diagram of FI-ETAAS system: P, peristaltic pump; N N N N H N N H H N Si S H O O O Fig. 1 Structure of DPTH-gel. PAAS, AS-70 autosampler pumps; Vs, switching valve. 1034 J. Anal. At. Spectrom., 1999, 14, 1033–1037microcolumn and the sample matrix is sent to waste; then, the precision and sample throughput) calculated as a function of two variables (except for the sample throughput that depends switching valve (Vs) is actuated and the pumps of the AS-70 furnace autosampler, PAAS, are connected, permitting the on a single variable): S=f(slope of the calibration curve, detection limit); C=f(consumption of sample, consumption operation of the autosampler in the normal mode; a wash step takes place with de-ionised water and, immediately after, the of reagent); A=f(relative standard deviation of the slope of the calibration curve, relative standard deviation of the deter- sampler arm lowers the sample capillary into an autosampler cup (filled with eluent) aspirating 20 ml HNO3 2 M; then, the mination); and R=f(time of analysis).Each value of the diVerent variables is subject to preliminary normalisation.sampler arm swings over to the graphite furnace and the tip of the sampler capillary is inserted into the dosing hole of the The algorithm used for the SIMPLEX optimisation was based on the modified SIMPLEX method of Nelder and graphite tube where the eluted Ni(II) is deposited as a drop; the sampler arm then returns to its initial position and the Mead.30 The convergence criterion used in the SIMPLEX algorithm stated that the standard deviation between the cycle of the furnace operation commences (Table 1); while the temperature programme is running, the switching valve is values of the response function should be less than a preset value, fixed by the user at the beginning of the optimisation again turned to start a new loading of the sample (standard or blank); thus, when the spectrometer gives the measurement, procedure.In this study, the value was fixed at 0.12, and the values of the weighting coeYcients were: s=100%; c=75%; the microcolumn is ready for a new injection of eluent.The eluent injection system (normal operation of auto- a=100%; and r=75%. The values of the response function were evaluated by sampler) provokes a flow through the column in both directions, up (when it is aspired) and down (when it is injected means of the construction of a calibration curve with four points [a blank and three standard solutions of 2, 5 and into the graphite tube), avoiding the continuous increase in column compactness. 8 ng ml-1 of Ni(II)].Six blank replicates were measured in order to calculate the detection limit and six replicates of the intermediate standard of 5 ng ml-1 for the calculation of the Software and mathematical model of response function for the optimisation of the ETAAS program relative standard deviation. For the consumption of the sample and reagent, cost (C), the value of 1.0 was given because the The system software, written in C++, is menu-driven and consumption of the sample (optimised in a univariate manner; interactive (Fig. 3). It includes a modified SIMPLEX optimis- 3.4 ml for 60 s loading) and the consumption of the reagent ation procedure and a mathematical model of our response (HNO3 2 M) (varied between 12 and 40 ml ) cannot be function that permits automated optimisation of analytical considered to be important in the cost of the analysis. methods, including all the parameters that define the overall The experimental parameters optimised were: eluent analytical quality, in a weighted way.injection volume (Vi/ml ); hold drying time 1 (td1/s); hold The response function, developed by Vereda et al.32 drying time 2 (td2/s); ashing temperature (Tm/°C); ramp ashing time (rm/s); hold ashing time (tm/s); and atomisation time A= (sS+cC) (aA+rR) 20000 (1) (ta/s). Therefore, a SIMPLEX with eight initial vertices was established. The normalisation conditions and the initial vertiis a weighted linear combination of variables, guiding the ces of the SIMPLEX were determined in preliminary experioptimisation towards diVerent compromise quality require- ments by using a number of systems with extreme values of ments, where s, c, a and r are the weighting coeYcients, called the variables to be optimised.quality coeYcients, which can vary between 0 and 100% and S, C, A and R are analytical properties (sensitivity, cost, Results and discussion Optimisation of the chemical variables In the initial study, the stability of the DPTH-gel resin was studied experimentally in acidic, neutral and basic media by observing any physical change occurring in the material; the results obtained showed that the resin was stable over a wide pH range, namely pH 0–13. Since the solution pH aVects the extent of complexation, which in turn determines the percentage of metal retained by the resin, the preconcentration of nickel from solutions buV- ered at diVerent pH was studied.The pH was adjusted from pH 2.0 to 5.0 using sodium acetate–acetic acid buVer and from pH 5.0 to 11.0 using boric acid–sodium hydroxide buVer.The results (Fig. 4) indicated that the optimum pH range was around pH 6.0 to 9.0. All subsequent studies were carried out at pH 9.0. Nitric acid was chosen as the eluent owing to its eVective elution of the adsorbed analyte complex. The eVect of eluent concentration on the emission signal of 1 ng ml-1 Ni, using a constant volume of injection of eluent of 15 ml, was examined.The signal increased as the HNO3 concentration increased up to 1 M. An HNO3 concentration of 2 M was chosen for subsequent studies. Preliminary tests showed that the sample volume was not an important factor when the mass of analyte arriving at the column was kept constant. The influence of the sample flow Fig. 3 Schematic representation of software. Boxes indicate separate rate was studied using a constant volume of injection of eluent programs; items within a box indicate functions available from that program’s menu.of 15 ml. For this purpose, 17 ng of nickel were brought to J. Anal. At. Spectrom., 1999, 14, 1033–1037 1035Table 2 Performance of the method Linear Detection Calibration range/ limit/ Determination equation r ng ml-1 ng ml-1 limit/ng ml-1 ya=0.047+0.095x 0.992 0–1 0.09 0.45 yp=0.064+0.116x 0.992 0–1 0.06 0.36 ya=-0.01+0.150x 0.998 1–5 yp=0.188+0.078x 0.999 1–5 ya, Absorbance as peak-area; yp, absorbance as peak-height; x, concentration; r, regression coeYcient.function versus the number of experiments at any time of the optimisation. The experimental responses were measured using the peakheight and peak-area. The optimum ranges optimised for the Fig. 4 Influence of pH on the preconcentration of nickel. furnace programme variables using the peak-height were: 15–30 ml for Vi; 10–20 s for td1; 21–30 s for td2; 1000–1300 °C for Tm ; 10–15 s for rm ; 20–30 s for tm; and 4–6 s for ta; using pH 9.0 and passed through the column at diVerent flow rates the peak-area: 12–30 ml for Vi; 9–25 s for td1; 20–30 s for td2; (the flow rate was varied by changing the speed of the sample 981–1300 °C for Tm; 9–15 s for rm; 19–30 s for tm and 4–6 s pump). Changes in the flow rate of the sample were studied for ta.Thus an average value of these variables was taken for between 1.8 and 6.3 ml min-1, resulting in an optimum sample subsequent experiments (Table 1).flow rate of 3.4 ml min-1 with the best signal-to-blank ratio. The eVect of the sample loading time on the emission signal Performance of the method of 1 ng ml-1 Ni was studied at a sample flow rate of Under the optimum conditions, with the use of a 60 s 3.4 ml min-1. The signal increased linearly up to a 5 min preconcentration time, a sample flow rate of 3.4 ml min-1 and preconcentration time using the peak-height signal and up to an injection volume of eluent of 20 ml, two linear calibration a 6 min preconcentration time using the peak-area signal after graphs were obtained in the ranges 0–1 and 1–5 ng ml-1 of which the slope decreased gradually.The sensitivity was Ni(II). The figures of merit of these calibration graphs are increased by increasing the sample loading time; however, a summarised in Table 2. The detection and determination limits, loading time of 60 s was selected in order to fit this time into defined as the concentrations of analyte giving signals equival- the cycle of the furnace programme and thus achieve a high ent to three and ten times, respectively, the standard deviation sampling frequency with a reasonable degree of sensitivity.A of the blank plus the net blank intensity, were calculated. longer loading time can be employed for samples with low These values are also given in Table 2. concentrations of nickel. The signal appeared 83 s after sample injection plus 15 s for washing the microcolumn before the injection, giving a sample Software and optimisation procedure for the ETAAS program throughput of about 36 h-1.The structure of the software, shown in Fig. 3, is based on a The precision of the method for aqueous standards, screen menu system, which adds great flexibility to the system evaluated as the relative standard deviation (RSD, n=6), was besides simplifying its use. Adaptation of the system to the 5.8% for 0.8 ng ml-1 of Ni(II) and 2.6% for 3 ng ml-1.quality requirements of a particular laboratory is performed The enrichment factor (EF), defined as the ratio of the by the user changing the quality coeYcients: Change quality detection limits before and after preconcentration, was 58. coeYcients from the Main menu. The values of the coeYcients The concentration eYciency (CE), defined as the product of for this work were: s=100%; c=75%; a=100%; and r=75%. the EF and the sampling frequency in number of samples The normalisation conditions were introduced from the item analysed per hour, was 35 min-1.The consumptive index, Values Max-Min. The ranges used for the normalisation were: defined as the volume of sample, in millilitres, consumed to 0.010–0.200 for the slopes; 0–3 ng ml-1 for the detection limit; achieve a unit EF, was 0.06 ml. 1.0–10.7% for the relative standard deviation of the slope of the calibration curve and for the relative standard deviation Interferences of the analysis; 1–2 for the consumption of the reagent sample; The eVect of various ions that commonly occur with nickel in and 1.0–2.5 min for the time of analysis.sea-water samples was examined under the optimum working For access to the SIMPLEX optimisation, the user has to conditions. For this study, diVerent amounts of the ionic strike the tab key and then the computer demands new species tested were added to a 3.0 ng ml-1 solution of Ni(II). experimental parameters. This process is repeated until the response function is optimised.During the optimisation pro- Table 3 Interferences cess, the software oVers the advantage to the user of modifying the values of the experimental conditions or of giving the Foreign species Tolerated ratio/m/m response function the value of 0.0. These possibilities impede the occurrence of impossible experimental conditions for the Ca(II ), Mg(II ), Ba(II), Mn(II ), K(I), system (for example, a negative injection volume) and establish Na(I ), Pb(II), Mo(VI), Se(IV), acetate, Cl-, NO3-, I-, F-, CO32-, glycine >4000 the boundaries given by the user, assigning to the response Hg(I ), Ag(I), PO43-, SO42- 3000 function the value of 0.00 when these are overcome.In this Cd(II), Al(III ), Fe(III )a, Co(II )a 2000 work, a maximum cycle time of furnace operation of 2.5 min Cu(II), Hg(II), Zn(II ), Cr(III)a, Sn(II)a 1000 was used, and the relative standard deviations of the slope of Bi(III), Mn(II ) 750 the calibration curve and of the analysis should not be higher Co(II), Fe(III), Sn(II ) 200 than 10.7%.The program also permits the user to save the aWith 2.5 ml of glycine 0.2 M. response function and display the graph of the response 1036 J. Anal. At. Spectrom., 1999, 14, 1033–1037Table 4 Determination of nickel in certified and spiked sea-water experimental parameters to optimise, adapting the system to samples his/her particular quality requirements. Certified Added/ Founda/ Recovery Sample value/ng ml.1 ng ml.1 ng ml.1 (%) Acknowledgements CRM 505 1.41¡¾0.12 1.57¡¾0.17 111.3 The authors thank the Direccio¢¥n General de Investigacio¢¥n Synthetic sea-water 0.40 0.42¡¾0.18 105.0 Cient©¥¢¥fica y Te¢¥ cnica of Spain (DGICYT) for supporting this Synthetic sea-water 0.70 0.71¡¾0.22 101.0 study (Project PB 96-0702) and also the Junta de Andaluc©¥¢¥a.Synthetic sea-water 1.00 1.02¡¾0.18 102.0 Sea-water . 2.34¡¾0.13 Sea-water 1.00 3.45¡¾0.08 111.0 References Sea-water 2.00 4.46¡¾0.07 106.0 Sea-water 4.00 6.26¡¾0.15 98.0 1 D.L. Tsalev, V. I. Slaveykova and P. B. Mandjukov, Spectrochim. Acta Rev., 1990, 13, 225. aMean¡¾standard deviation for four replicate measurements of four 2 Z. Fang, M. Sperling and B. Welz, J. Anal. At. Spectrom., 1990, individual preparations. 5, 639. 3 M. Sperling, X. Yin and B. Welz, J. Anal. At. Spectrom., 1991, 6, 295. The starting point was an interference-to-nickel ratio of 4000 4 S. Olsen, L.C. R. Pessenda, J. Ru¡Æz¢§ic¢§ka and E. H.Hansen, Analyst, m/m; if any interference occurred, the ratio was gradually 1983, 108, 905. lowered until the interference disappeared. A given species 5 Z. Fang, J. Ru¡Æz¢§ ic¢§ka and E. H. Hansen, Anal. Chim. Acta, 1984, 164, 23. was considered to interfere if it resulted in a ¡¾5% variation 6 K. Vermeiren, C. Vandecasteele and R. Dams, Analyst, 1990, 115, of the AAS signal. The results obtained are given in Table 3. 17. The interference of Fe(III), Co(II), Cr(III ) and Sn(II) can be 7 R.Rattray and E. D. Salin, J. Anal. At. Spectrom., 1995, 10, 1053. significantly lowered by adding glycine 0.2M to the medium. 8 S. Hirata, Y. Umezaki and M. Ikeda, Anal. Chem., 1986, 58, 2602. 9 S. Hirata, K. Honda and T. Kumamaru, Anal. Chim. Acta, 1989, Determination of nickel in certified and spiked sea-water 221, 65. 10 D. B. Taylor, H. M. Kingston, D. J. Nogay, D. Koller and samples R. Hutton, J. Anal. At. Spectrom., 1996, 11, 187.In order to test the accuracy and applicability of the proposed 11 V. Porta, O. Abollino, E. Mentalti and C. Sarzanini, J. Anal. At. method for the analysis of real samples, a certified reference Spectrom., 1991, 6, 119. 12 F. Malamas, M. Bengtsson and G. Johansson, Anal. Chim. Acta, material CRM 505 and spiked sea-water and synthetic sea- 1984, 160, 1. water samples were analysed. CRM 505 shows a pH of 1.6, 13 A. Fang and B.Welz, J. Anal. At. Spectrom., 1989, 4, 543.so that a previous neutralisation with diluted NaOH is rec- 14 K. Akatsuka, J. W. McLaren, J. W. H. Lam and S. S. Berman, ommended. The slopes of the calibration graphs and those J. Anal. At. Spectrom., 1992, 7, 889. using the standard additions method were very similar for all 15 X. Wang and R. M. Barnes, J. Anal. At. Spectrom., 1989, 4, 509. samples, except for CRM 505 which had to be evaluated with 16 S. Arpadjan, L. Vuchkova and E. Kostadinova, Analyst, 1997, 122, 243.the standard additions method. The results given in Table 4, 17 D. Pozebon, V. L. Dressler, J. A. Gomes Neto and A. J. Curtius, as the average of four individual preparations, show good Talanta, 1998, 45, 1167. agreement with the certified value and suYciently high 18 Z. Fang, S. Xu and G. Tao, J. Anal. At. Spectrom., 1996, 11, 1. recoveries. 19 M. Sperling, X. Yin and B. Welz, J. Anal. At. Spectrom., 1991, 6, 615. 20 M. Sperling, X. Yin and B.Welz, Spectrochim. Acta, Part B, 1991, Conclusions 46, 1789. 21 Z. Liu and S. Huang, Spectrochim. Acta, Part B, 1995, 50, 197. The most frequently stated advantage of ETAAS is its low 22 Z. Liu and S. Huang, Anal. Chim. Acta, 1993, 281, 185. detection limit. However, in certain instances, the detection 23 Y. Sung, Z. Liu and S. Huang, J. Anal. At. Spectrom., 1997, limits are still inadequate, especially when the sample has a 12, 841. complex matrix. In these circumstances, a preliminary precon- 24 X. Yan, W. VanMol and F. Adams, Analyst, 1996, 121, 1061. centration and/or separation is required. Conventional oV-line 25 X. Yan and F. Adams, J. Anal. At. Spectrom., 1997, 12, 459. 26 D. Betteridge, T. J. Sly, A. P. Wade and J. E. W. Tillman, Anal. procedures for separation and preconcentration, although Chem., 1983, 55, 1292. eVective, are usually time-consuming and tedious, and are 27 S. N. Deming and S. L. Morgan, Anal. Chem., 1973, 45, 278A. vulnerable to contamination and analyte loss. This study has 28 S. N. Deming and L. Parker, CRC Crit. Rev. Anal. Chem., 1978, shown that the FI-ETAAS method described allows the rapid 7, 187. determination of nickel. The analytical scheme of the proposed 29 W. Spendley, G. R. Hext and F. R. Himsworth, Technometrics, system is much simpler than other oV-line and on-line pro- 1962, 4, 441. 30 J. A. Nelder and R. Mead, Comput. J., 1965, 7, 308. cedures because it combines trace enrichment, derivatisation 31 M. Poch, J. L. Montesinos, M. del Valle, J. Alonso, A. N. Arau¢¥ jo and detection in a single analytical set-up. and J. L. F. C. Lima, Analusis, 1992, 20, 319. For a detailed optimisation of systems with four or more 32 E. Vereda, A. R©¥¢¥os and M. Valca¢¥rcel, Anal. Chim. Acta, 1997, variables, SIMPLEX may still provide the most viable option, 348, 129. as the computerised optimisation is faster and much less labour 33 P. Can.ada Rudner, A. Garc©¥¢¥a de Torres, J. M. Cano Pavo¢¥n and intensive. The interactive optimisation procedure can be E. Rodriguez Castello¢¥ n, J. Anal. At. Spectrom., 1998, 13, 243. applied to any analytical method because the user chooses the Paper 9/01451I J. Anal. At. Spectrom., 1999, 14, 1033.1037 1037