Optimisation of the Experimental Conditions of a New Method, Based on a Quartz Crystal Microbalance, for the Determination of Cyanide M. Teresa S. R. Gomes*, A. Alexandre F. Silva, Armando C. Duarte and Jo�ao A. B. P. Oliveira Department of Chemistry, University of Aveiro, 3810 Aveiro, Portugal A new method based on a quartz crystal microbalance was developed for the determination of cyanide. As the sensitivity depends on pH, temperature and nitrogen flow rate, a modified simplex was used to optimise these experimental parameters.Two different versions of the proposed method were optimised. For the first version a sensitivity increase of 1.5 was observed after 27 runs, whereas for the second version a sensitivity increase of 1.7 was observed after 12 runs. Keywords: Simplex optimisation; quartz crystal microbalance; piezoelectric crystals; cyanide A new method based on a quartz crystal microbalance (QCM) was developed for the determination of cyanide.The method is based on the fact that the cyanide promotes the disproportionation of HgI:1 Hg2 2+ " Hg2+ + Hg0 (1) The addition of cyanide ion forces reaction (1) to the right, as it forms a strong complex with HgII. The complete reaction can be described by Hg2 2+ + 2CN2 " Hg0 + Hg(CN)2 (2) The amalgamation of the mercury vapour on the gold electrodes of a piezoelectric quartz crystal leads to a frequency decrease,2–5 which is a linear function of the cyanide content of the sample.If acid is added to the mercury solution, it suppresses the hydrolysis of HgII ions, as shown in reaction (3), which otherwise would drive the mercury disproportionation reaction to the right, creating a high background level of Hg0.1 Hg2+ + H2O " HgO + H+ (3) The frequency decrease for a specific sample depends on several experimental parameters such as the carrier gas flow rate, pH and temperature in the reaction cell. In order to reduce experimental errors, it is important to maximise the frequency changes.A modified simplex algorithm6 was used to optimise the experimental conditions for a solution with a cyanide concentration close to the limit established for industrial waste waters7 and to the centroid of the linear calibration curve. Experimental Apparatus The gas flow rate was controlled with a variable area flow meter (Cole Parmer, Chicago, IL, USA). The piezoelectric quartz crystals were 10 MHz (HC49/U; Euroquartz, Crewkerne, Somerset, UK) and the frequency was monitored with a frequency counter (PM6680, Philips, Eindhoven, The Netherlands).All the other equipment was laboratory made and has been described elsewhere.5 A nitrogen flow, controlled by a flow meter, enters the bottom of a thermostated glass cell, through a sintered glass plate. This cell contains an acidified mercury solution, and allows the introduction of the cyanide solution through a silicone-rubber septum at the top. The mercury vapour flows through a 3A molecular sieve column, where it is dried, and impacts both faces of a piezoelectric crystal with gold electrodes.The oscillation frequency of the crystal is monitored with a frequency counter and a frequency decrease, proportional to the added mass, is observed during the mercury amalgamation. In a second version, an extra cell was inserted between the existing one and the flow meter, for reasons explained in the Procedure section. Reagents Mercury nitrate (Panreac, Barcelona, Spain), nitric acid (Riedelde Ha�en, Hanover, Germany), potassium cyanide, (Merck, Darmstadt, Germany) and phosphoric acid (Fluka, Buchs, Switzerland) were all of analytical-reagent grade. Nitrogen was of R grade from ArL�ýquido (Porto, Portugal).Procedure Two different versions of the new method based on a QCM were developed for cyanide determination. In a first approach, 10.0 cm3 of a 4.0 31025 m HgNO3·H2O solution, acidified with HNO3, were introduced into a glass cell.Nitrogen flowing through the bottom of the cell bubbled through a sintered glass plate and carried the mercury vapour on to the gold electrodes of a piezoelectric crystal. After an initial decrease, the frequency stabilised, and then 5.0 cm3 of a cyanide solution were injected. A new frequency decrease, which was proportional to the cyanide content, was observed, because the formation of HgII complexes promotes mercury disproportionation. Chlorides and thiocyanides are often present in industrial waste waters and can also form strong complexes with HgII.Furthermore, the responses are strongly influenced by sample pH. Therefore, for applications where chlorides and thiocyanides are present, or low cyanide concentrations necessitate the introduction of large volumes of sample, the method needed to be changed. Another glass cell, inserted before the one that contains the mercury solution, allows the sample introduction over phosphoric acid and the formation of hydrocyanic acid.The hydrocyanic acid is then carried by the nitrogen flow into the mercury solution cell and, as before, a frequency decrease is observed. Samples with cyanide concentrations bracketing the Portuguese legal limit7 for industrial waste waters (0.5 ppm [CN2]) were analysed by both versions of the method, and a sample volume of 5.0 cm3 was experimentally found to be adequate and selected for all subsequent experiments. The experimental parameters were then optimised for the solution with a Analyst, October 1997, Vol. 122 (1139–1141) 1139concentration close to the centroid of the linear calibration curve, 0.389 ppm [CN2] and 0.761 ppm [CN2] for the first and second versions, respectively. After the evaluation of the responses of the first four sets of conditions, a computer program written in FORTRAN 77 calculated the next set of conditions to be investigated. The algorithm followed the rules of the modified simplex method of Nelder and Mead.6 Each measurement was performed just once, to keep the number of experiments to a minimum.However, if a vertex was retained in 3 + 1 simplexes, the response was re-evaluated. If a vertex corresponded to a negative quantity, or the frequency stabilisation before the cyanide introduction took more than 25 min, an arbitrary 0 Hz response was assigned. Before the optimisation procedure, the estimated relative standard deviation of the concentration corresponding to the centroid of the linear calibration curve of the method was 6%.Therefore, the search was halted when, for the latest simplex, the tolerance (defined as the ratio of the difference between the greater and smaller responses over their mean) was less than 0.060. The optimisation procedure for the second version started from the optimum conditions found for the first version. Results and Discussion Fig. 1 shows, for both versions, the evolution of the response during the experiments.For the first version, the first simplex vertex corresponded to a temperature of 26.9 °C, a flow rate of 103 cm3 min21 and a volume of 3.0 cm3 of HNO3 in 100 cm3 of the mercury solution. The observed frequency decrease with the introduction of 5.0 cm3 of a 0.389 ppm [CN2] solution was 1978 Hz. After the optimisation procedure, the frequency decrease obtained with the introduction of the same amount of cyanide solution increased to 3150 Hz, with a temperature of 35.0 °C, a flow rate of 60 cm3 min21 and a volume of 1.0 cm3 of HNO3 in the mercury solution.For the second version, and using the former optimum conditions, the introduction of 5.0 cm3 of a solution 0.761 ppm [CN2] produced a frequency decrease of 253 Hz, which was increased to 562 Hz with change in the parameters to temperature 36.0 °C, flow rate 31 cm3 min21 and 2.8 cm3 of HNO3. Fig. 2 shows that, for both versions, the response increased as the flow rate of the carrier gas decreased.For practical reasons, and as already mentioned, when the frequency stabilisation took longer than 25 min, the frequency decrease was set to zero. The time for attaining frequency stabilisation depends not only on the carrier gas flow rate but also on the quantity of HNO3. Fig. 3 shows, for the first and second versions, the frequency decrease observed versus the volume of 0.1 m HNO3 used in the preparation of 100.0 cm3 of Holution.For the first version, the optimum was obtained with small volumes of acid, whereas for the second version it corresponded to large volumes. This contradictory behaviour must result from the different way of supplying cyanide into the mercury solution: an alkaline solution of CN2 in the first version and hydrocyanic acid in the second one. Increasing the amount of HNO3 decreases the mercury disproportionation before sample introduction, according to eqn. (3), in the same way for both versions of the method.This contributes to an increase in the response to cyanide, as there is an increase in the amount of HgI present at the moment of sample introduction, in addition to a smaller amount of mercury already amalgamated on to the crystal electrodes. The introduction of the alkaline cyanide solution directly into the mercury solution, in the first version, contributes to an increase in Fig. 1 Evolution of the response during the experiments. Fig. 2 Frequency decrease versus nitrogen flow rate for the first and second versions of the method.The points with a zero frequency decrease correspond to a frequency stabilisation longer than 25 min. Fig. 3 Frequency decrease versus volume of HNO3 for the first and second versions of the method. The points with a zero frequency decrease correspond to a frequency stabilisation longer than 25 min. 1140 Analyst, October 1997, Vol. 122mercury disproportionation and a higher frequency decrease than in the second version, where no alkali was added to the mercury solution.Moreover, the CN2/HCN ratio, which depends on the acid concentration, can also influence the amount of mercury vapour that reaches the crystal, in each version of the method. The temperature seems to be a less important factor than the nitrogen flow rate or the amount of acid. As a general conclusion, the simplex method, with a small number of experiments (27 in the first version and 12 in the second), leads to a signal increase and a general improvement in the slope of the linear portion of the calibraiton curve.The sensitivity increased from 4897 to 7430 Hz ppm21 for the first version and from 562.4 to 969.2 Hz ppm21 for the second version. The linear dynamic working range for the first version was 0.10–0.78 and 0.05–0.50 ppm before and after optimization, respectively. For the second version it was 0.32–1.02 and 0.24–0.86 ppm, respectively. The first version of the method becomes the obvious choice in the absence of chlorides, since it has the highest sensitivity of the two versions.The equation of the calibration curve is DF = 7430 [CN2] + 89.9 (r2 = 0.997) for cyanide concentrations in the range 0.05–0.50 ppm for the first version and DF = 969.2[CN2] 2 184.0 (r2 = 0.990) for cyanide concentrations in the range 0.24–0.86 ppm for the second version, where DF is the observed frequency decrease in Hz and [CN2] is the concentration of cyanide in the analysed standard solutions in ppm.In addition to chlorides, the most common compounds in the waste waters from the electroplating industry, suspected possibly to interfere in determination of cyanide, were added to standard cyanide solutions with concentrations within the linear working range for both versions of the method. In the first version of the method, Cu, Zn and Ni were not suspected to interfere, as the stability constants of their cyanide complexes were smaller than that of [Hg2+(CN)4]22.8 As the formation constant of [Fe3+(CN)6]32 was larger than that for [Hg2+(CN)4]22, a solution of K3Fe(CN)6, with a concentration of cyanide close to the centroid of the linear calibration curve, was introduced and cyanide could not be detected, as no frequency decrease was observed.Although the formation constant of [Fe2+(CN)6]42 was smaller than that of the [Hg2+(CN)4]22 complex, a solution of K4Fe(CN)6, with a concentration of cyanide close to the centroid of the linear calibration curve, was introduced into the sample cell and no interference was observed, as the signal had a magnitude corresponding to the cyanide content.For the second version of the method, a solution of NaCl with a chloride concentration of 144.6 ppm was introduced into the second cell and, as expected, no response was observed. Solutions of Cu, Zn and Ni, with cyanide concentration within the linear working range, were prepared, and no interference was observed with concentration of those ions of 12.87, 1.25 and 39.4 ppm, respectively.With the introduction of solutions of K4Fe(CN)6 or K3Fe(CN)6, with a concentration of cyanide close to the centroid of the linear calibration curve, no signal was observed. The inability of the second version of the method to detect hexacyanoferrate(ii) or hexacyanoferrate(iii) was not considered to be an important disadvantage, as the cyanide in these chemical forms is not considered to be toxic. As H2O2 is usually added in electroplating plants to destroy cyanide, 5 cm3 of a solution 1% in H2O2 was introduced into the cell in the second version of the method, and no frequency decrease was observed.We are grateful to the Junta Nacional de Investigaç�ao Cient�ýfica e Tecnol�ogica for the financial support of A. A. F. Silva through a scholarship (BM/926/94). References 1 Marshall, G., and Midgley, D., Anal. Chem., 1981, 53, 1760. 2 Bristow, Q., J.Geochem. Explor., 1972, 1, 55. 3 Sheide, E. P., and Taylor, J. K., Environ. Sci. Technol., 1974, 8, 1087. 4 Suleiman, A. A., and Guilbault, G. G., Anal. Chem., 1984, 56, 2964. 5 Gomes, M. T., Rocha, T. A., Duarte, A. C., and Oliveira, J. O., Anal. Chem., 1996, 68, 1561. 6 Nelder, J. A., and Mead, R., Comput. J., 1965, 7, 308. 7 Portuguese Water Quality Policy Act, Decreto-Lei No. 74/90, Di�ario da Rep�ublica–I S�erie, Imprensa Nacional-Casa da Moeda, EP., Portugal, 1990 (in Portuguese). 8 Morel, F. M. M., Principles of Aquatic Chemistry, Wiley, New York, 1983. Paper 7/02660I Received April 18, 1997 Accepted August 5, 1997 Analyst, October 1997, Vol. 122 1141 Optimisation of the Experimental Conditions of a New Method, Based on a Quartz Crystal Microbalance, for the Determination of Cyanide M. Teresa S. R. Gomes*, A. Alexandre F. Silva, Armando C. Duarte and Jo�ao A. B. P. Oliveira Department of Chemistry, University of Aveiro, 3810 Aveiro, Portugal A new method based on a quartz crystal microbalance was developed for the determination of cyanide.As the sensitivity depends on pH, temperature and nitrogen flow rate, a modified simplex was used to optimise these experimental parameters. Two different versions of the proposed method were optimised. For the first version a sensitivity increase of 1.5 was observed after 27 runs, whereas for the second version a sensitivity increase of 1.7 was observed after 12 runs.Keywords: Simplex optimisation; quartz crystal microbalance; piezoelectric crystals; cyanide A new method based on a quartz crystal microbalance (QCM) was developed for the determination of cyanide. The method is based on the fact that the cyanide promotes the disproportionation of HgI:1 Hg2 2+ " Hg2+ + Hg0 (1) The addition of cyanide ion forces reaction (1) to the right, as it forms a strong complex with HgII. The complete reaction can be described by Hg2 2+ + 2CN2 " Hg0 + Hg(CN)2 (2) The amalgamation of the mercury vapour on the gold electrodes of a piezoelectric quartz crystal leads to a frequency decrease,2–5 which is a linear function of the cyanide content of the sample.If acid is added to the mercury solution, it suppresses the hydrolysis of HgII ions, as shown in reaction (3), which otherwise would drive the mercury disproportionation reaction to the right, creating a high background level of Hg0.1 Hg2+ + H2O " HgO + H+ (3) The frequency decrease for a specific sample depends on several experimental parameters such as the carrier gas flow rate, pH and temperature in the reaction cell.In order to reduce experimental errors, it is important to maximise the frequency changes. A modified simplex algorithm6 was used to optimise the experimental conditions for a solution with a cyanide concentration close to the limit established for industrial waste waters7 and d of the linear calibration curve.Experimental Apparatus The gas flow rate was controlled with a variable area flow meter (Cole Parmer, Chicago, IL, USA). The piezoelectric quartz crystals were 10 MHz (HC49/U; Euroquartz, Crewkerne, Somerset, UK) and the frequency was monitored with a frequency counter (PM6680, Philips, Eindhoven, The Netherlands). All the other equipment was laboratory made and has been described elsewhere.5 A nitrogen flow, controlled by a flow meter, enters the bottom of a thermostated glass cell, through a sintered glass plate.This cell contains an acidified mercury solution, and allows the introduction of the cyanide solution through a silicone-rubber septum at the top. The mercury vapour flows through a 3A molecular sieve column, where it is dried, and impacts both faces of a piezoelectric crystal with gold electrodes. The oscillation frequency of the crystal is monitored with a frequency counter and a frequency decrease, proportional to the added mass, is observed during the mercury amalgamation.In a second version, an extra cell was inserted between the existing one and the flow meter, for reasons explained in the Procedure section. Reagents Mercury nitrate (Panreac, Barcelona, Spain), nitric acid (Riedelde Ha�en, Hanover, Germany), potassium cyanide, (Merck, Darmstadt, Germany) and phosphoric acid (Fluka, Buchs, Switzerland) were all of analytical-reagent grade. Nitrogen was of R grade from ArL�ýquido (Porto, Portugal).Procedure Two different versions of the new method based on a QCM were developed for cyanide determination. In a first approach, 10.0 cm3 of a 4.0 31025 m HgNO3·H2O solution, acidified with HNO3, were introduced into a glass cell. Nitrogen flowing through the bottom of the cell bubbled through a sintered glass plate and carried the mercury vapour on to the gold electrodes of a piezoelectric crystal. After an initial decrease, the frequency stabilised, and then 5.0 cm3 of a cyanide solution were injected.A new frequency decrease, which was proportional to the cyanide content, was observed, because the formation of HgII complexes promotes mercury disproportionation. Chlorides and thiocyanides are often present in industrial waste waters and can also form strong complexes with HgII. Furthermore, the responses are strongly influenced by sample pH. Therefore, for applications where chlorides and thiocyanides are present, or low cyanide concentrations necessitate the introduction of large volumes of sample, the method needed to be changed.Another glass cell, inserted before the one that contains the mercury solution, allows the sample introduction over phosphoric acid and the formation of hydrocyanic acid. The hydrocyanic acid is then carried by the nitrogen flow into the mercury solution cell and, as before, a frequency decrease is observed. Samples with cyanide concentrations bracketing the Portuguese legal limit7 for industrial waste waters (0.5 ppm [CN2]) were analysed by both versions of the method, and a sample volume of 5.0 cm3 was experimentally found to be adequate and selected for all subsequent experiments.The experimental parameters were then optimised for the solution with a Analyst, October 1997, Vol. 122 (1139–1141) 1139concentration close to the centroid of the linear calibration curve, 0.389 ppm [CN2] and 0.761 ppm [CN2] for the first and second versions, respectively.After the evaluation of the responses of the first four sets of conditions, a computer program written in FORTRAN 77 calculated the next set of conditions to be investigated. The algorithm followed the rules of the modified simplex method of Nelder and Mead.6 Each measurement was performed just once, to keep the number of experiments to a minimum. However, if a vertex was retained in 3 + 1 simplexes, the response was re-evaluated. If a vertex corresponded to a negative quantity, or the frequency stabilisation before the cyanide introduction took more than 25 min, an arbitrary 0 Hz response was assigned.Before the optimisation procedure, the estimated relative standard deviation of the concentration corresponding to the centroid of the linear calibration curve of the method was 6%. Therefore, the search was halted when, for the latest simplex, the tolerance (defined as the ratio of the difference between the greater and smaller responses over their mean) was less than 0.060.The optimisation procedure for the second version started from the optimum conditions found for the first version. Results and Discussion Fig. 1 shows, for both versions, the evolution of the response during the experiments. For the first version, the first simplex vertex corresponded to a temperature of 26.9 °C, a flow rate of 103 cm3 min21 and a volume of 3.0 cm3 of HNO3 in 100 cm3 of the mercury solution. The observed frequency decrease with the introduction of 5.0 cm3 of a 0.389 ppm [CN2] solution was 1978 Hz.After the optimisation procedure, the frequency decrease obtained with the introduction of the same amount of cyanide solution increased to 3150 Hz, with a temperature of 35.0 °C, a flow rate of 60 cm3 min21 and a volume of 1.0 cm3 of HNO3 in the mercury solution. For the second version, and using the former optimum conditions, the introduction of 5.0 cm3 of a solution 0.761 ppm [CN2] produced a frequency decrease of 253 Hz, which was increased to 562 Hz with change in the parameters to temperature 36.0 °C, flow rate 31 cm3 min21 and 2.8 cm3 of HNO3.Fig. 2 shows that, for both versions, the response increased as the flow rate of the carrier gas decreased. For practical reasons, and as already mentioned, when the frequency stabilisation took longer than 25 min, the frequency decrease was set to zero. The time for attaining frequency stabilisation depends not only on the carrier gas flow rate but also on the quantity of HNO3.Fig. 3 shows, for the first and second versions, the frequency decrease observed versus the volume of 0.1 m HNO3 used in the preparation of 100.0 cm3 of HgNO3 solution. For the first version, the optimum was obtained with small volumes of acid, whereas for the second version it corresponded to large volumes. This contradictory behaviour must result from the different way of supplying cyanide into the mercury solution: an alkaline solution of CN2 in the first version and hydrocyanic acid in the second one.Increasing the amount of HNO3 decreases the mercury disproportionation before sample introduction, according to eqn. (3), in the same way for both versions of the method. This contributes to an increase in the response to cyanide, as there is an increase in the amount of HgI present at the moment of sample introduction, in addition to a smaller amount of mercury already amalgamated on to the crystal electrodes.The introduction of the alkaline cyanide solution directly into the mercury solution, in the first version, contributes to an increase in Fig. 1 Evolution of the response during the experiments. Fig. 2 Frequency decrease versus nitrogen flow rate for the first and second versions of the method. The points with a zero frequency decrease correspond to a frequency stabilisation longer than 25 min. Fig. 3 Frequency decrease versus volume of HNO3 for the first and second versions of the method.The points with a zero frequency decrease correspond to a frequency stabilisation longer than 25 min. 1140 Analyst, October 1997, Vol. 122mercury disproportionation and a higher frequency decrease than in the second version, where no alkali was added to the mercury solution. Moreover, the CN2/HCN ratio, which depends on the acid concentration, can also influence the amount of mercury vapour that reaches the crystal, in each version of the method.The temperature seems to be a less important factor than the nitrogen flow rate or the amount of acid. As a general conclusion, the simplex method, with a small number of experiments (27 in the first version and 12 in the second), leads to a signal increase and a general improvement in the slope of the linear portion of the calibraiton curve. The sensitivity increased from 4897 to 7430 Hz ppm21 for the first version and from 562.4 to 969.2 Hz ppm21 for the se version.The linear dynamic working range for the first version was 0.10–0.78 and 0.05–0.50 ppm before and after optimization, respectively. For the second version it was 0.32–1.02 and 0.24–0.86 ppm, respectively. The first version of the method becomes the obvious choice in the absence of chlorides, since it has the highest sensitivity of the two versions. The equation of the calibration curve is DF = 7430 [CN2] + 89.9 (r2 = 0.997) for cyanide concentrations in the range 0.05–0.50 ppm for the first version and DF = 969.2[CN2] 2 184.0 (r2 = 0.990) for cyanide concentrations in the range 0.24–0.86 ppm for the second version, where DF is the observed frequency decrease in Hz and [CN2] is the concentration of cyanide in the analysed standard solutions in ppm.In addition to chlorides, the most common compounds in the waste waters from the electroplating industry, suspected possibly to interfere in determination of cyanide, were added to standard cyanide solutions with concentrations within the linear working range for both versions of the method. In the first version of the method, Cu, Zn and Ni were not suspected to interfere, as the stability constants of their cyanide complexes were smaller than that of [Hg2+(CN)4]22.8 As the formation constant of [Fe3+(CN)6]32 was larger than that for [Hg2+(CN)4]22, a solution of K3Fe(CN)6, with a concentration of cyanide close to the centroid of the linear calibration curve, was introduced and cyanide could not be detected, as no frequency decrease was observed.Although the formation constant of [Fe2+(CN)6]42 was smaller than that of the [Hg2+(CN)4]22 complex, a solution of K4Fe(CN)6, with a concentration of cyanide close to the centroid of the linear calibration curve, was introduced into the sample cell and no interference was observed, as the signal had a magnitude corresponding to the cyanide content. For the second version of the method, a solution of NaCl with a chloride concentration of 144.6 ppm was introduced into the second cell and, as expected, no response was observed.Solutions of Cu, Zn and Ni, with cyanide concentration within the linear working range, were prepared, and no interference was observed with concentration of those ions of 12.87, 1.25 and 39.4 ppm, respectively. With the introduction of solutions of K4Fe(CN)6 or K3Fe(CN)6, with a concentration of cyanide close to the centroid of the linear calibration curve, no signal was observed. The inability of the second version of the method to detect hexacyanoferrate(ii) or hexacyanoferrate(iii) was not considered to be an important disadvantage, as the cyanide in these chemical forms is not considered to be toxic. As H2O2 is usually added in electroplating plants to destroy cyanide, 5 cm3 of a solution 1% in H2O2 was introduced into the cell in the second version of the method, and no frequency decrease was observed. We are grateful to the Junta Nacional de Investigaç�ao Cient�ýfica e Tecnol�ogica for the financial support of A. A. F. Silva through a scholarship (BM/926/94). References 1 Marshall, G., and Midgley, D., Anal. Chem., 1981, 53, 1760. 2 Bristow, Q., J. Geochem. Explor., 1972, 1, 55. 3 Sheide, E. P., and Taylor, J. K., Environ. Sci. Technol., 1974, 8, 1087. 4 Suleiman, A. A., and Guilbault, G. G., Anal. Chem., 1984, 56, 2964. 5 Gomes, M. T., Rocha, T. A., Duarte, A. C., and Oliveira, J. O., Anal. Chem., 1996, 68, 1561. 6 Nelder, J. A., and Mead, R., Comput. J., 1965, 7, 308. 7 Portuguese Water Quality Policy Act, Decreto-Lei No. 74/90, Di�ario da Rep�ublica–I S�erie, Imprensa Nacional-Casa da Moeda, EP., Portugal, 1990 (in Portuguese). 8 Morel, F. M. M., Principles of Aquatic Chemistry, Wiley, New York, 1983. Paper 7/02660I Received April 18, 1997 Accepted August 5, 1997 Analyst, October 1997, V