ISSN:0003-2654    

DOI:10.1039/AN98712BP009

出版商:RSC    

年代:1987

数据来源: RSC

摘要:

The AnalystThe Analytical Journal of The Royal Society of ChemistryAdvisory Board*Chairman: J. D. R. Thomas (Cardiff, UK)J. F. Alder (Manchester, UK)D. Betteridge (Sunbury-on-Thames, UK)E. Bishop (Exeter, UK)*C. Burgess (Ware, UK)D. T. Burns (Belfast, UK)G. D. Christian (USA)*M. S. Cresser (Aberdeen, UK)L. de Galan (The Netherlands)*A. G. Fogg (Loughborough, UK)*C. W. Fuller (Notfingham, UK)V. D. Goldberg (London, UK)T. P. Hadjiioannou (Greece)A. Hulanicki (Poland*C, J. Jackson (London, UK)*P. M. Maitlis (Sheffield, UK)E. J. Newman (foole, UK)T. B. Pierce (Harwell, UK)E. Pungor (Hungary)J. RSiiEka (Denmark)R. M. Smith (Loughborough, UK)W. I. Stephen (Birmingham, UK)M. Stoeppler (federal Republic of Germany)K. C. Thompson (Sheffield, UK)*A.M. Ure (Aberdeen, UK)A. Walsh, K.B. (Australia)G. Werner (German Democratic Republic)T. S. West (Aberdeen, UK)*P. C. Weston (London, UK)J. D. Winefordner (USA)Yu. A. Zolotov (USSR)P. Zuman (USA)*Members of the Board serving on the Analytical Editorial BoardRegional Advisory EditorsFor advice and help to authors outside the UKDr. J. Aggett, Department of Chemistry, University of Auckland, Private Bag, Auckland, NEWDoz. Dr. sc. K. Dittrich, Analytisches Zentrum, Sektion Chemie, Karl-Marx-Universitat, Talstr.Professor L. Gierst, Universitb Libre de Bruxelles, Faculte des Sciences, Avenue F.-D.Professor H. M. N. H. Irving, Department of Analytical Science, University of Cape Town,Dr. 0. Osibanjo, Department of Chemistry, University of Ibadan, Ibadan, NIGERIA.Dr.G. Rossi, Chemistry Division, Spectroscopy Sector, CEC Joint Research Centre,Dr. 1. RubeSka, Geological Survey of Czechoslovakia, Malostranske 19, 118 21 Prague 1,Professor K. Saito, Coordination Chemistry Laboratories, Institute for Molecular Science,Professor M. Thompson, Department of Chemistry, University of Toronto, 80 St. GeorgeProfessor P. C. Uden, Department of Chemistry, University of Massachusetts, Amherst,Professor Dr. M. Valchrcel, Departamento de Quimica Analitica, Facultad de Ciencias,Professor Yu Ru-Qin, Department of Chemistry and Chemical Engineering, Hunan University,ZEALAND.35, DDR-7010 Leipzig, GERMAN DEMOCRATIC REPUBLIC.Roosevelt 50, Bruxelles, BELGIUM .Rondebosch 7700, SOUTH AFRICA.EURATOM, lspra Establishment, 21020 lspra (Varese), ITALY.CZECHOSLOVAKIA.Myodaiji, Okazaki 444, JAPAN.Street, Toronto, Ontario M5S 1A1, CANADA.MA 01003, USA.Universidad de Cordoba, 14005 Cordoba, SPAIN.Changsha, PEOPLES REPUBLIC OF CHINA.Editor, The Analyst:Philip C.WestonSenior Assistant E ditors:Judith Brew, Roger A. YoungAssistant Editors:Anne Horscrof?, Pamil SehmiEditorial Office: The Royal Society of Chemistry, Burlington House,Piccadilly, London, WIV OBN. Telephone 01-734 9864. Telex No. 268001.dvertisements: Advertisement Department, The Royal Society of Chemistry, BurlingtonHouse, Piccadilly, London, WIV OBN. Telephone 01-437 8656. Telex No. 268001The Analyst (ISSN 0003-2654) is published monthly by The Royal Society of ChemistryBurlington House, London W1V OBN, England.All orders accompanied with payment shoulcbe sent directly to The Royal Society of Chemistry, The Distribution Centre, Blackhorse RoadLetchworth, Herts. SG6 IHN, England. 1987 Annual subscription rate UK f160.00, Rest oWorld f179.00, USA $315.00. Purchased with Analytical Abstracts UK f364.00, Rest of Workf403.00, USA $709.00. Purchased with Analytical Abstracts plus Analytical Proceedings Ukf411 .OO, Rest of World f455.00, USA $801 .OO. Purchased with Analytical Proceedings Ukf200.00, Rest of World €224.00, USA $394.00. Air freight and mailing in the USA b)Publications Expediting Inc., 200 Meacham Avenue, Elmont, NY 11 003.USA Postmaster: Send address changes to: The Analyst, Publications Expediting Inc., 20(Meacham Avenue, Elmont, NY 11003.Second class postage paid at Jamaica, NY 11431. Alother despatches outside the UK by Bulk Airmail within Europe, Accelerated Surface Posloutside Europe. PRINTED IN THE UK.Information for AuthorsFull detai1.s of how to submit material forpublication in The Analyst are given in theInstructions to Authors in the January issue.Separate copies are available on request.The Analyst publishes papers on all aspects ofthe theory and practice of analytical chemistry,fundamental and applied, inorganic andorganic, including chemical, physical, biochem-ical, clinical, pharmaceutical, biological, auto-matic and computer-based methods. Papers onnew approaches to existing methods, newtechniques and instrumentation, detectors andsensors, and new areas of application with dueattention to overcoming limitations and to un-derlying principles are all equally welcome.There is no page charge.The following types of papers will be con-sidered:Full papers, describing original work.Short papers: the criteria regarding origin-ality are the same as for full papers, but shortpapers generally report less extensive investi-gations or are of limited breadth of subjectmatter.Communications, which must be on anurgent matter and be of obvious scientificimportance. Rapidity of publication is enhancedif diagrams are omitted, but tables and formulaecan be included.Communications receive pri-ority and are usually published within 5 8weeks of receipt.They are intended for briefdescriptions of work that has progressed to astage at which it is likely to be valuable toworkers faced with similar problems. A fullerpaper may be offered subsequently, if justifiedby later work.Reviews, which must be a critical evaluationof the existing state of knowledge on a par-ticular facet of analytical chemistry.Every paper (except Communica,tions) will besubmitted to at least two referees, by whoseadvice the Editorial Board of The Analystwill beguided as to its acceptance or rejection. Papersthat are accepted must not be published else-where except by permission. Submission of amanuscript will be regarded as an undertakingthat the same material is not being consideredfor publication by another journal.Regional Advisory Editors.For the benefit ofpotential contributors outside the United King-dom, a Panel of Regional Advisory Editorsexists. Requests for help or advice on anymatter related to the preparation of papers andtheir submission for publication in The Analystcan be sent to the nearest member of the Panel.Currently serving Regional Advisory Editors arelisted in each issue of The AnalystManuscripts (three copies typed in double spac-ing) should be addressed to:The Editor, The Analyst,Royal Society of Chemistry,Burlington House,Piccadilly,LONDON W1V OBN, UKParticular attention should be paid to the use ofstandard methods of literature citation, includingthe journal abbreviations defined in ChemicalAbstracts Service Source Index. Wherever pos-sible, the nomenclature employed should fol-low IUPAC recommendations, and units andsymbols should be those associated with SI.All queries relating to the presentation andsubmission of papers, and any correspondenceregarding accepted papers and proofs, shouldbe directed to the Editor, The Analyst (addressas above). Members of the Analytical EditorialBoard (who may be contacted directly or via theEditorial Office) would welcome comments,suggestions and advice on general policy mat-ters concerning The Analyst.Fifty reprints of each published contribution aresupplied free of charge, and further copies canbe purchased.@ The Royal Society of Chemistry, 1987. Allrights reserved. No part of this publication maybe reproduced, stored in a retrieval system, ortransmitted in any form, or by any means,electronic, mechanical, photographic, record-ing, or otherwise, without the prior permissionof the publishers

ISSN:0003-2654    

DOI:10.1039/AN98712FX013

出版商:RSC    

年代:1987

数据来源: RSC

摘要:

ANALAO 1 12(4) 345-556 (I 987)The AnalystApril 1987The Analytical Journal of The Royal Society of Chemistry34534735536537738539 139740741 141 7423427433437441445449455459463467473477481485489CONTENTS7TH SAC INTERNATIONAL CONFERENCE ON ANALYTICAL CHEMISTRY, BRISTOL, UK, 2&26 JULY, 1986EDITORIALConcepts for Improved Automated Laboratory Productivity. Plenary Lecturct-M. Bonner DentonRecent Advances in the Theory of Atomisation in Graphite Furnace Atomic Absorption Spectrometry: the Oxygen -Extreme Trace Analysis of the ElementsThe State of the Art Today and Tomorrow. Plenary Lectur-Gunther TolgRegression Techniques for the Detection of Analytical Bias-Brian D. Ripley, Michael ThompsonChemometrics in Pharmaceutical Analysis-John C.BerridgeElucidation of Olive Oil Classification by Chemometrics-Omar Eddib, Graham NicklessDetermination of Morphine in Body Fluids by High-performance Liquid Chromatography with ChemiluminescenceDetection-Richard W. Abbott, Alan Townshend, Richard GillRapid High-performance Liquid Chromatographic Method for the Determination of Propranolol in Plasm-Keith A.Smith, Simon Wood, Magda CrousReversed-phase High-performance Liquid Chromatographic Retention Behaviour of Benzylpenicillin and Its Acid - BaseDegradation Products-Andrew M. LipczynskiUse of Macroporous Polymeric High-performance Liquid Chromatographic Columns in Pharmaceutical Analysis-Michael J. Cope, Ian E. DavidsonCombined Capillary Column Gas Chromatography - Molecular Fluorescence Spectrometry-Colin S.Creaser, AndrewStaffordRoutine Determination of Organochlorine Pesticides and Polychlorinated Biphenyls in Human Milk Using Capillary GasChromatography - Mass Spectrometry-Mark P. Seymour, Terry M. Jefferies, Arthur J. Floyd, Lidia J. NotarianniUse of Fast Atom Bombardment Mass Spectrometry t o Identify Materials Separated on High-performance Thin-layerChromatography Plates-Kenneth J. Bare, Harry ReadApplication of Directly Coupled Flame Atomic Absorption Spectrometry - Fast Protein Liquid Chromatography t o theDetermination of Protein-bound Metals-Les Ebdon, Steve Hill, Philip JonesA Comparison of Isotope Dilution Mass Spectrometric Methods for the Assay of Copper in Copper Ore ReferenceMaterials-Ellyn S.Beary, Karen A. Brletic, Paul J. Paulsen, John R. MoodyDetermination of Degradation Products in Tributyl Phosphate Using 31P Fourier Transform NMR Spectrometry-Naohito UetakeSelective Separation of Zinc from Other Elements on the Amphoteric Resin Retardion 11A8 and its Use for theDetermination of Zinc in Biological Materials by Neutron Activation Analysis-Rajmund Dybczyriski, Shatha S.Aldabbag hCombination of Impurity Pre-concentration in Compound Semiconductors with Different Methods of Analysis-Svetlana S. Grazhulene, Yuri I. Popandopulo, Vassili K. Karandashev, Natalya I. Zolotaryova, Nina I. ChaplyginaPolyurethane Foams as Sorbents for Noble Metals. Selective Pre-concentration of Small Amounts of Platinum fromHydrochloric Acid Containing Tin(ll) Chloride Followed by Flame Atomic Absorption Analysis-Karim F.G.Brackenbury, Lynn Jones, Klaus R. KochAnalysis of Chemical Forms of Iridium. Spectrophotometric Determination of [lr(N02)6]s, [lr(NO2)CI5]3- and[Ir(NO)CI,]--Yi-bi n Qu, Ji ng-mi ng WangSpectrophotometric Determination of Zinc with 1 -[Di(2-pyridyl)methylene)-5-salicylidenethiocarbonohydrazid~M~.Teresa Morales, Ma. Teresa Montaiia, Guillermina Galan, Jose L.Gomez ArizaExtraction - Spectrophotometric Determination of Molybdenum with Toluene-3,4-dithioCMa. Pilar Bermejo-Barrera,Jose Francisco Vazquez-Gonzalez, Francisco Bermejo-MartinezDetermination of Molybdenum with Gallic Acid and Hydroxylamin-Pilar Bermejo-Barrera, Jose F. Vazquez-Gonzales,Ma Carmen Pazos- Navei ra, Francisco Bermejo- Ma rt i nezSimultaneous Derivative Spectrophotometric Determination of Iron(lll) and Bismuth(ll1) with EDTA-Adela Bermejo-Barrera, Ma.Pilar Bermejo-Barrera, Ma. Manuela Guisasola-Escudero, Francisco Bermejo-MartinezIndirect Spectrophotometric Method for the Microdetermination of Chlorine or Bromine in Organic Compounds Using1,5-Diphenylcarbazide-Mouayed 0. Al-Abachi, E. S. SalihIndirect Spectrophotometric Microdetermination of Some Carboxylic Acids and Unsaturated Organic Compounds inAqueous Solution-Thabit S. Al-Ghabsha, S. M. Abdul AzizCarbon Alternative. Plenary Lecture-Boris V. L'vovcontinued inside back coverTypeset and printed by Black Bear Press Limited, Cambridge, Englan49349950150751 151 552352753 1535539543545549553Metal Chelate Fluorescence Enhancement in Micellar Media: Mechanisms of Surfactant Action-Alfred0 Sanz-Medel,Rosario Fernandez de la Cam pa, Jose lgnacio Garcia AlonsoInfrared Spectrophotometric Determination of the Alkalinity of Overbased Petroleum Sulphonates-El ham 2.Said,lmad H. Al-Wahaib, Hikmet H. NimaApplications of Attenuated Total Reflection in the Infrared Analysis of Carbohydrates and Biological Whole CellSamples in Aqueous Solution-John H. Hopkinson, Catherine Moustou, Nicola Reynolds, John E. NewberyDetermination of Ascorbic Acid in Multivitamin Tablets by Thermometric Titrimetry with Cerium(1VtAnthony R.Mayers, Colin G. TaylorSome Titrimetric Applications of Manganese(lll1 Salt Solutions: Analysis of Higher Valent Complexes of Manganeseand NickeCMounir A.MalatiModels for Dispersion in Flow Injection Analysis. Part 1. Basic Requirements and Study of Factors AffectingDispersion-David C. Stone, Julian F. TysonAnalytical Information from Doublet Peaks in Flow Injection Analysis. Part 1. Basic Equation and Applications t o FlowInjection Titrations-Julian F. TysonAnalytical Information from Doublet Peaks in Flow Injection Analysis. Part II. Determination of Stability Constants-Julian F. TysonFlow Injection Procedures for the Determination of Ethanol and Alcohol Dehydrogenase Using Co-immobilisedBacterial Luciferase and Oxidoreductase-Abdul Nabi, Paul J. WorsfoldSimultaneous Determination of Organic Isomers in Mixtures by Flow Injection Analysis with a Diode ArrayPhotodetector-Baldomero Bermudez, Fernando Lazaro, Maria Dolores Luque de Castro, Miguel ValcarcelCopper-deposited Wire Ion-selective Electrode for the Determination of Copper(l1)-Borislava D.VuEuroviC, Milo5 6.Raj kovicKnown Addition Potentiometry Using an Ion-selective Electrode as an Indicator Electrode for the Determination ofMicro-amounts of Fluoride in Standard Water Samples-Sheng-Luo XieApplication of a Periodate Liquid Membrane Electrode t o the Determination of a-Diols, Carbohydrates andHydrazines-Saad S. M. Hassan, M. M. ElsaiedPolarographic Determination of Antibilharzial Organic Antimony Compounds-Youssef A. Gawargious, Nabil 6.Tadros, Amir Besada, Louis F. lbrahimElectrochemical Behaviour of Polarised Carbon Fibre Micro-electrodes-V.J. Jennings, J. E. MorganNEWCertified Reference MaterialsPublicationsBureau of Analysed SamplesCatalogue No. 550Overseas Reference Materials List No. 555for copies of these publications pleasewrite, telex or telephone to:BAS Ltd., Newham Hall, Newby,Middlesbrough, Cleveland, TS8 9EATelex: 587765 BASRIDTelephone: (0642) 317216-Understanding-Our EnvironmentEdited by R. E. Hester, University of YorkThis book provides a wide-ranging and authoritative coverage of topicswhich are fundamental to our understanding and appreciation of thenature of our environment. The six chapters are written by authors whoare acknowledged experts in their particular fields, providing a sufficientdepth of treatment to satisfy the needs of serious students of the subjectmatter but in a style and level which will be found comprehensible andinteresting by the generally concerned, involved, and educated layman.Brief Contents:Monitoring; The Air; Water; Land Contamination and Reclamation;Assessing the Ecological and Health Effects of Pollution; Regulation andthe Economics of Pollution Control; Appendix; Information Retrieval.Hardcover 348pp ISBN 0 85186 901 6 (1986) Price E42.50 ($11.00)RSC members ElZ.50 (RSC members are entitled to a 20% discounton bulk orders for 15 or more copies of Understanding OurEn vironrnen 1.)Ordering:Non-RSC members should send their orders to: The Royal Society of Chemistry,Distribution Centre, Blackhorse Road, Letchworth. Herts SG6 1HN. UK.RSC members should send their orders to: Membership Manager. The RoyalSociety of Chemistry, 30 Russell Square, London WClB 5DT. UK.A001 for further information. See pag

ISSN:0003-2654    

DOI:10.1039/AN98712BX015

出版商:RSC    

年代:1987

数据来源: RSC

摘要:

ANALYST, APRIL 1987, VOL. 112 345 Editorial SAC 86-Bristo1, UK, July 20-26, 1986 In keeping with the example set by the SAC 83 Conference, particularly under the strong influence of the late Professor John Ottaway, some of the papers presented at the 7th SAC International Conference have been collected together to produce this Special Issue of The Analyst, and it is hoped that this will act as a valuable reference volume on the wide range of topics and current developments in analytical chemistry that were presented at the conference in Bristol. On this occasion the SAC Conference was twinned with the 3rd BNASS Symposium. This decision was taken by the RSC Analytical Division Council during 1984, and turned out to be the correct course of action, as an extremely successful joint conference resulted.Four Plenary Lectures, the 3rd BNASS opening lecture, 115 invited and contributed lectures, 120 posters, seven workshops and four update courses comprised the programme. The full scientific programme was published in the June 1986 issue of Analytical Proceedings (1986, 23, 173-248) (Handbook Issue), which also contained the abstracts of all the conference papers. The papers in this issue include three of the Plenary Lectures, including that by Professor L'vov, who unfortu- nately was unable to be present at the conference. Of the 52 papers submitted for publication in The Analyst, 41 appear in this Special Issue. Some have been rejected by the normal refereeing procedure, and others, which were submitted late, will appear in later issues. The Editorial Board is grateful to the Plenary Lecturers and others for the production of their manuscripts for this issue of the journal.The SAC 86 Conference was sponsored by the International Union of Pure and Applied Chemistry and the Federation of European Chemical Societies. Again, the Editorial Board is grateful for the general support given by both IUPAC and FECS that made the Conference a successs and this issue of The Analyst possible. Publication of this Special Issue is a valuable contribution to the wider dissemination of the research that was presented in Bristol last July. Additional copies of this issue are available for personal purchase from the Royal Society of Chemistry, Distribution Centre, Blackhorse Road, Letchworth, Herts, SG6 1HN (price &16/$32). Its publication was made possible through the contributions of many individuals and due to particularly conscientious efforts on the part of the editorial staff.The SAC 86 Executive Committee is grateful for the tireless efforts of everyone concerned with the organisation of the conference and the production of this Special Issue of The Analyst. Readers may also like to know that many of the papers presented at the 3rd BNASS part of the joint conference have been published in March as a Special Issue of the RSC's successful new journal, Journal of Analytical Atomic Spec- trometry, which is also available for individual purchase (price f25/$50). Together these two Special Issues of The Analyst and J A A S contain over 70 papers of direct interest to all analytical scientists and provide a wide-ranging cross-section of research work at the forefront of our discipline. The two Special Issues can be purchased for a total price of &29/$58. J. Gareth Jones Chairman, SAC 86l3rd BNASS Executive Committee

ISSN:0003-2654    

DOI:10.1039/AN9871200345

出版商:RSC    

年代:1987

数据来源: RSC

摘要:

ANALYST, APRIL 1987, VOL. 112 347 Concepts for Improved Automated Laboratory Productivity* Plenary Lecture M. Bonner Denton Department of Chemistry, University of Arizona, Tucson, AZ 85721, USA The use of automated laboratory techniques is rapidly increasing. Significant changes are occurring both in how tasks are accomplished and in which tasks are practical and cost effective. Although the automation of a given task does not inherently dictate the use of some form of computer, the greater system flexibility achieved through software control, coupled with the recent drastic reduction in computer hardware costs, has already made this approach to automation extremely popular. The vast proliferation of computational hardware does not solve all of the problems in laboratory automation-far from it.Two major problem areas arise, development of suitable function systems to conduct the desired chemistry and development of the proper software. Today, in many instances workers have resorted to mimicking human manipulation of samples through the use of robotics. Although this approach is viable for some situations, it is far from optimal for many other applications. Laboratory automation today often involves the use of instruments designed to perform a specific task (e.g., sample preparation and analysis) on a high work load. However, there is a trend toward increasing flexibility through multi-task capability. This concept can be implemented through several means. One example would be an instrument which is configured in such a manner that it can or does obtain a wide range of data.Software quickly sifts through the results and displays the requested information to the user. This approach allows a great deal of flexibility, as different information can be obtained merely by changing the software. Additionally, the presence of possible interferences, unusual results on species not requested and even over-all system performance can be constantly monitored and presented to the user. Many of these concepts will be considered while describing a new generation of intelligent atomic spectrometric instrumentation. The ultimate goal is an automated system capable of accepting any type of sample and performing any analysis such that all desired information would be obtained. Ideally, following analysis, the sample would be returned unharmed.Such a highly flexible, non-destructive instrument is “science fiction“ today, but much more limited systems based on present technology are not out of the question. Keywords: Automated laboratory productivity; computers; atomic spectrometry As analytical chemists we are interested in qualitative analysis, composition, quantitative analysis-how much is really there and, in many instances, speciation-how those various components are combined. The tool that all of us as analytical chemists really dream about having available is what I refer to in concept as the “Mark I Magic Analyser.” In this instance, the Mark I Magic Analyser will accept any sample, .whatever it might be. We merely tell the Mark I what we would like to know about the sample, and the Mark I performs that analysis while even telling us other interesting and important aspects of that particular sample’s composition.Finally, the Mark I would return our sample. Clearly, such technology is not available today. However, some of the concepts in the Mark I certainly can be applied when we consider automation and future automation in our own laboratories. We would like to have the ability to handle samples with little or no sample preparation, or the ability to perform sample preparation automatically in a manner transparent to the user. Fig. 1 shows some of these concepts: automatic sampling, automatic sample preparation if required, automated analysis, automated data reduction and, finally, data presentation, data correlation and trend pattern analysis. Notice that there is feedback from all of these different steps to the control step.Additionally, we can learn from the trend-pattern analysis, allowing the automated system to actually become smarter as time progresses. Recently there has been a proliferation in the concept of using robotics merely to replace humans, that is, to have the * Presented at SAC 86, the 7th SAC International Conference on Analytical Chemistry, Bristol, UK, 20-26 July, 1986. robot actually go through all the steps of an analysis that would have been done conventionally by human beings. In some instances this is appropriate; in others it would be far better to look over the whole situation, and possibly approach the analysis with robotic and/or other systems in a totally different manner than one would with human laboratory technicians.I would like to consider two approaches to the problem of laboratory automation through the use of improved sample handling and with more intelligent instrumentation. In both instances, atomic spectrometry will be the technology that will be utilised to gain insight into the sample composition. Although far from the desired capabilities of the Mark I Magic Analyser, atomic spectrometry does provide an in-depth knowledge of the elemental composition of the sample. We shall be considering flames and d.c. and inductively coupled plasmas for actually analysing samples. Until recently, most samples were analysed in the liquid form and converted into an aerosol, which was introduced into either the flame or the plasma with the Bernoulli principle or cross-capillary types of nebulisers.In both instances the sample passes through a small capillary that limits the nebuliser in handling highly viscous materials, materials that have very high levels of suspended solids or even samples with very high levels of dissolved solids, as the end of the capillary can become encrusted by salt crystals and the rate of nebulisation altered. New Technology to Minimise Sample Preparation Several years ago, my research group first introduced the Babington geometry nebuliser to atomic spectrometry112 and explored a variety of configurations for such systems.3 At that time we demonstrated the ability to nebulise motor oils, from348 ANALYST, APRIL 1987, VOL.112 Data presentation Automatic - Automatic’ Data Automatic- Automatic sampling sample prep.- analyser data reduction--ccorrelation \ I/+ Trend-pattern a na I ysi s Cont ro I Fig. 1. Automated chemical analysis normally involves several steps. Currently numerous different spectroscopic techniques are employed in the analysis step Table 1. Zinc determination (pg ml-1) Total diges- Matrix Babington tion* Condensed milk . . . , . . . . 7.5 7.9 Grapefruit juice, homogenised . . 0.39 0.41 Pineapple syrup . . . . . . . . 0.54 0.58 Haemolysedwholeblood . . . . 12 13 Tomatosauce . . . . . . . . 0.62 -4 Pickled beet juice . . . . . . 0.72 0.81 * Acid-digested samples were analysed using AAS. Centrifuge, supernatant digested 1.6 0.25 0.56 0.64 0.70 14 c 0 v) .- .- E UJ SAE 5 W to 90 W transmission grease, condensed milk, whole blood, hydraulic fluid, urine, orange juice, pineapple syrup and even tomato sauce with absolutely no sample preparation and no problems associated with nebuliser clogging? We had a great deal of fun showing a slide depicting the spooning of tomato sauce into the nebuliser and a nice red cloud issuing from a burner base.We also demonstrated that Babington principle nebulisers had essentially the same sensitivities for conventional aqueous samples at low salt concentrations, etc., whilst still being able to handle these very difficult types of samples2 (Table 1). Babington principle nebulisers were found to provide reliable analyses of such complex and difficult samples as condensed milk, grapefruit juice and blood, showing that the same answers were obtainable directly with the Babington method that were observed with conventional acid digestion procedures.We also established that in most analyses the results reflected analysis of the entire sample matrix and not just the supernatant. However, with the tomato sauce, where very large fibrous materials are present, the data indicated that a transport problem caused the results to correlate closely with those obtained from acid digestion of the supernatant. Considering the geometry that was being employed at that time, this was not surprising. Other observations made when analysing motor oil indi- cated that when the viscosity of the motor oil increased above approximately SAE 25 W, the actual sample flow across the nebuliser changed.After this point the viscosity of the standards had to be matched with that of the unknown to ensure valid results. Although this seems to be a trivial matter, in practice it is a significant limitation because, even though an oil of known viscosity can be put into a particular engine, the viscosity after a period of engine operation can be very different. Fuel dilution can decrease the viscosity and sludge formation can increase the viscosity. Therefore, to make valid analysis, a viscosity measurement was necessary in order to allow matching of the standards with the unknown. Sub- sequent studies”5 demonstrated that if the nebuliser tip is heated to approximately 70 “C, the emission intensities for Pennzoil SAE 20 W, 30 W, 40 W and 50 W motor oil all converged.Additionally, the sensitivity of the system was increased by approximately an order of magnitude (Fig. 2). Fig. 3 shows the divergence in intensities for a variety of spiked oils from that obtained with Pennzoil30 W. A greater deviation is observed between various different brands of 30 W motor oil than between the Pennzoil20,30,40 and 50 W, 0 20 40 60 80 0 i I t em perat u re ‘“C Fig. 2. Emission intensity as a function of temperature for various grades of Pennzoil HD motor oil, each spiked with 100 pg g-1 of Conostan oil-soluble Fe standard. Oil grade: A, 20 W; B, 30 W; C, 40 W; and D, 50 W indicating that viscosity is not the major cause of deviation but that other factors in the composition of the oil, such as antioxidants and additives, also contribute.These studies indicated the capability of the Babington principle nebuliser for determining dissolved metals in oils. But what about real-world samples? Real-world samples also contain particulates of sizes below the cutoff of the filtration system in the particular engine. As with the tomato sauce, where the large particulates tended to settle out, these relatively heavy metal particles, particularly with the lower flow-rates associated with the sample injection in an induc- tively coupled plasma (ICP), can be selectively removed in the spray chamber , preventing them from being introduced into the plasma discharge. Analysis of real samples actually showed that in the conventional configuration the results tended to be low. This led to the investigation of an inverted plasma configuration, originally used by Reedbin the first ICP system where powder and gas were introduced to grow crystals.Subsequently, Greenfield et aZ.7 used a similar configuration for the first analytical studies in an ICP. In later studies the plasma was turned “right side up” and operated in the mode that we normally use today. A comparison of the results obtained with direct injection ICP, solvent dilution atomic absorption spectrometry (AAS), where the sample was diluted 1 + 9 with isobutyl methyl ketone, the inverted ICP and finally a conventional acid digestion ashed AAS, which we shall consider is the actual correct value for the following series of studies, is shown in Table 2. Data from the analysis of several samples by total digestion AAS and then direct injection ICP are shown in Table 3.Although this problem could have been automated using robotics to carry out conventional digestion procedures, the inverted ICP equipped with a heated Babington principle nebuliser provides a much simpler approach, allowing a sample to be drawn out of a running engine, taken directly to the system and run in a period of a few minutes. Clearly this indicates a route to improving automation and sample throughput by rethinking the method of analysis, incorporat- ing new technology and, in this instance, totally eliminating the sample preparation steps.ANALYST, APRIL 1987, VOL. 112 1.0 349 0 - - - A 0 X X . m . 0 X T - 0.0 I I I 1 I 1 I 20 30 40 50 60 70 80 Temperati ire/”C Fig. 3. 20 W; 0, Pennzoil40 W; A, Penmoil50 W; x, Ray Lube 30 W; 0, Quaker State 30 W; M, Valvoline 30 W; 0, Kendall GT-140 W Ratios of emission intensities given bv various oils to the intensity measured for Pennzoil30 W at the same temperature: A, Pennzoil Table 2.Comparison of four methods for the determination of iron in a 30 W motor oil sampled from an automobile Concentration of iron/ Method pg ml-1 Direct injection, upright ICP . . . . . . 56 Solvent dilution, AAS . . . . . . . . . . 36 Direct injection, inverted ICP . . . . . . 117 Ashing, AAS . . . . . . . . . . . . 106 ~~~ ~ Table 3. Comparison of the results of analysis of four motor oil samples by direct injection and inverted ICP and acid digestion followed by determination by AAS Total digestion, AASI Direct injection, Sample No.pg ml-1 inverted ICP/pg ml-1 1 140 141 (+22) 2 404 (+35) 406 (k 14) 3 205 (+22) 242 (+22) 4 106(+15) 117 (f15) Utilising Information Provided by a Technique Another example of improved automation involves better utilisation of the information actually available from an analytical technique. We shall focus on atomic emission spectrometry, where for any given element there can be a large number of emission lines. Those who have experienced the pain and drudgery of using photographic emulsions also readily appreciate the tremendous amount of spectral infor- mation this technique actually provides. Unfortunately, quantification, using photographic emulsions is extremely laborious and generally limited in accuracy. Two alternative approaches are commercially available for actually measuring this information.One involves a slew scan type of read-out where one wavelength at a time is observed. This has a major disadvantage because plasma, nebulisers and read-out systems are never totally stable. Also, the time involved can be substantial if one wants to look at several lines for each element. The second approach is to use a direct reader or polychromator, which places a series of slits and photomulti- plier tubes on the focal plane. This popular technique suffers from a number of significant limitations. It is very expensive, large and bulky and is not generally considered to be a portable type of instrument. Each of the slits and photomulti- plier tubes requires alignment to a specific line. This can represent a substantial amount of initial setup time and can require periodic re-alignment.By far the most serious limitation of the direct reader is the very limited amount of spectral data observed, as one photomultipler is required per line observed. What is really needed is some type of electronic read-out which will measure the photon flux at all wavelengths simultaneously, that is, an electronic equivalent of a photo- graphic emulsion. This has been dreamed of for years and a wide variety of researchers have expended considerable effort investigating a variety of “camera devices, ” including vidi- cons, intensified target vidicons, plumbicons, orthicons, image dissectors, photodiode arrays and a variety of other imaging devices.819 The “camera techniques” previously explored have suffered from one or more problems, including poor dynamic range, insufficient spectral range, poor repro- ducibility between detector elements, cross-talk between elements (blooming, smearing, etc.) , an insufficient number of resolution elements, poor sensitivity (quantum efficiency), high dark and/or read current, inability to provide integration of photon flux, inability to access detector elements randomly, high cost per detector element and poor reliability.Astronomers have also been faced with the problems associated with photographic emulsions. In fact, modern astronomy is carried out to a large extent with a variety of new high technology, solid-state imaging devices , including charge coupled devices (CCDs) and charge injection devices (CIDs).lO As the emphasis in astronomy is generally on extremely low light detection, charge coupled devices are widely utilised.11912 However, atomic spectrometry has an350 ~ C 3 3000 c .- * i 2000 4- .- fn (u ANALYST, APRIL 1987, VOL. - - Focusina mirror Cassegrain image reducer Solid-state High-order Cross-dispersing dchelle grati& first-order grating Folding mirror Collimating mirror w Entrance slit Fig. 4. An optical system employing an Cchelle grating, first-order r t i n g and Cassegrain image reducer to create a two-dimensional ocal plane suitable for use with charge injection device detectors c .- f ! rh V V Horizontal Horizontal I ).I //I * .- e E l .- ill Horizontal Fig. 5. (a) Plot of a s ectrum showing analytical lines with background from the lasma; (6 plot of the background from the plasma; and (c) plot ofthe background substracted from the analytical signal unusual condition not often encountered by astronomers.In atomic emission spectrometry, it is necessary to be able to detect parts per billion of a particular component, while also being able to quantify very high levels of a component without having problems associated with blooming or smearing from lines associated with major constituents in the sample. This necessitates an extremely wide dynamic range. Of all the charge transfer devices available today, the CID has the unique capability of reading out a given detector element either destructively or non-destructively . As the CID, unlike the photomultipler tube, has no internal gain, consider- able gain must be added by outside amplifiers.Unfortunately, associated with all amplifiers is a certain degree of noise, principally from the very first stage of the amplifier, which is amplified by the entire gain of the amplifier string. Aikens et aZ.13 at Kitt Peak National Laboratories demonstrated the improved signal to noise ratio inherent by summing a number of non-destructive read-outs from the CID. As the noise from the video pre-amplifier is essentially white, summation of multiple non-destructive read-outs can eliminate this com- ponent. Unfortunately, CIDs are not configured in the ideal geometry. The ideal geometry of approximately 0.5 m long, the ability to be curved along the focal plane, with individual detector elements 10 pm wide by 2 mm tall, is not practical using today’s technology.In fact, these devices range from ._ c cn 4000 5000 3 a 5 1000 0.00 0 1 2 3 4 5 6 7 1 Timeih 112 Fig. 6. Plot of observed signal for four analytical lines of different intensity read non-destructively at a rate of one read per second for 8 h. Note that the observed line intensities are not affected by over 2.8 x lo5 read-outs ‘ Exa m i nation ” ‘ ”Read” window CID imager Fig. 7. Read-out during quantitative analysis involves non-des- tructively sampling a three-by-three detector “examination” matrix to determine when a selected line has reached the desired intensity, and then reading a three-by-thirteen array multiple times to acquire accurately both the line and background intensity approximately 8 x 8 mm containing 1.64 x 104 detector ele- ments to 6.56 X 8.78 mm containing over 3.6 x 105 elements- hardly the desired geometry for incorporation into a conven- tional direct reading spectrometer.As these are X Y devices, some approach must be pursued which will utilise this format. One choice would be that of an Cchelle grating spectrometer similar to those manufactured by Beckman Instruments, Leeman Laboratories, etc. With the Beckman Cchelle spectrometer, on an area of 10.2 x 12.6 cm one actually generates the equivalent of a linear focal plane that would be approximately 10 m long, achieving high resolution with a respectable degree of total light throughput. Unfortunately, 10.2 x 12.6 cm is a far cry from the actual size of the devices in question. Previous investigators have faced similar problems with the use of other types of XY detectors.Felkel and Parduel4 studied a vidicon detection system on a modified Spectrametrics Beckman kchelle spectrometer, where a Cassegrain telescope was used as an image reducer. Fig. 4 shows one approach using an Cchelle grating in conjunction with a first-order grating to sort out the orders vertically; a Cassegrain image-reducing system reduces the image to provide a focal plane suitable for use with the CID detector. Clearly there are a number of tradeoffs in the design of an Cchelle spectrometer system for use with one of the CID detectors. These include wavelength coverage, resolution and light throughput. Unlike electron beam types of read-out systems such as the vidicon, these solid-state devices are directly digitally addressable.Therefore, with a properly designed stabilised optical system, extremely accurate back- ground subtractions can be readily performed. Fig. 5(a) depicts a small portion of three orders. One might guess that the very large peak is a signal due to the analyte; however,ANALYST, APRIL 1987, VOL. 112 351 \"Read" window Fig. 8. Representation of how a "read" window is selected to contain a spectral line and adjacent background Fig. 9. Actual example of a read window showin the observed intensity for the iron 297.32 line and surrounding bacfground when the background is observed [Fig. 5(b)], the large peak is confirmed to be the analyte, whereas the medium peak just to the right was in fact due to the background. Subsequently, when these two are subtracted, Fig.5(c) shows the simplified spectra, clearly indicating that the large peak was the sample, but also showing a number of smaller peaks clearly vastly above the signal to noise background readily usable for chemical analysis. While the current trend in CID technology is toward a larger number of total detector elements, unfortunately there is also a trend toward reducing the actual size of the wafer and hence greatly reducing the total area of each detector element. This necessitates even more stringent optical designs. Current devices under investigation include the CID 17 and CID 20. The CID 17 is composed of 248 rows and 388 columns. Each detector element is 23 x 27 pm. The CID 20 employs 488 rows by 388 columns with each detector element being 11 X 27 pm. To appreciate fully the capabilities of CIDs, one must refer to the quantum efficiencies available in current state-of-the- art photomultiplier tubes, where quantum efficiencies above 10% and certainly above 20% are extremely rare. In contrast to the photomultiplier tubes, CIDs such as the CID 17 and 20 have quantum efficiencies approaching 50% (550 nm) ranging from 7% at 200 nm, 13% at 225 nm, to 18% at 800 nm.When this is coupled with the fact that the devices have essentially zero dark current when properly operated, one realises that this is a truly remarkable detector. The unique capability of the CID to mix both destructive and non-destructive read-outs gives the atomic spectroscopist ~~ Table 4. Comparison of detection limits observed with the d.c.plasma source and CID spectrometer with those obtained with a commercial d.c. plasma - Cchelle system, showing competitive detection limits Detection limit/pg ml-1 CID 17, d.c. Beckman d.c. Element Wavelengthhrn plasma - Cchelle plasma - Cchelle Al . . Ca . . Cr . . c u . . Fe . . In . . Mn . . Ni . . Pb . . . . . . 308.22 394.40 396.15 . . . . 393.37 396.85 422.67 . . . . 357.87 359.35 360.53 427.48 428.97 . . . . 324.75 327.40 . . . . 358.12 371.99 373.49 373.71 386.00 . . . . 303.94 325.86 410.18 451.13 . . . . 259.37 293.93 403.08 403.31 403.45 . . . . 352.45 341.48 . . . . 363.96 405.78 368.35 7 3 1 2 0.4 2 4 6 4 5 3 3 6 1 3 15 15 7 12 2 2 5 2 4 5 6 7 11 6 - - - 2 0.7 2 2 2 5 20 10 4 2 10 2 20 10 Table 5. Detection limits observed in a lo00 pg ml-1 gadolinium matrix demonstrating a maximum of a one order of magnitude decrease when analysing a matrix producing a very complex emission pattern Detection limit/ Element Wavelengtldnm pg ml-1 Ca .. . . . . . . . . . . 422.67 396.85 393.7 Cr . . . . . . . . . . . . 425.43 427.48 Fe . . . . . . . . . . . . 373.49 280.27 285.21 Ni . . . . . . . . . . . . 341 -48 Mg . . . . . . . . . . . . 279.55 3 5 3 2 14 2 0.2 0.4 0.6 70 another very powerful capability, that is, to be able to vary the integration time from one wavelength to another depending on the actual amount of light observed at each wavelength. Hence it is possible to integrate very intense lines for short periods of time until good signal to noise ratios are obtained and, while the system is still in its linear dynamic range, measure those very intense lines.Subsequently, those intense wavelengths can be destructively read out while integration continues on wavelengths associated with very weak emission lines. The net result is that strong lines are integrated for a short period of time, whereas weak lines are integrated for extended periods of time, allowing good signal to noise ratios to be obtained in both instances without problems associated with overloading the device. This method requires that non-destructive readings be truly non-destructive. The signals observed from re-reading four different lines at the rate of one read-out per second for 8 h is shown in Fig. 6.352 1 nterface Memory sE::ie ANALYST, APRIL 1987, VOL. 112 Camera head electronics I Liquid nitrogen cryostat \ Fig.10. Schematic diagram of the CID 17 camera system used for this work. Array detector sequencing is provided by a Photometrics camera controller which receives instructions from a host Motorola 68 000 based computer One approach for implementing variable integration time detection is shown in Fig. 7. In this example, there is a combination of a read window, which is 13 detector elements wide by 3 detector elements long, and an examination window, which is the centre array of 3 by 3 detector elements, where the actual emission line falls. The 3 by 5 adjacent areas are used for background subtraction as shown in Fig. 8. An actual example of the iron 297.32 nm line is shown in Fig. 9. For the purpose of determining appropriate integration times, it is necessary only to look at the detector elements associated with the examination window.Finally, when the signal present in the examination window is of sufficient intensity to allow a good signal to noise ratio, the entire array associated with the read window is recorded so that appropriate background correction techniques can be applied. As these devices have essentially zero dark current when properly operated, the ultimate detection limit is limited merely by the background and drift characteristics of the source being studied and the full charge capacity of a given detector element. The detection limits, defined as twice the standard deviation of the blank, observed with the CID 17 d.c. plasma- University of Arizona Cchelle system in comparison with those published in the literature for the Beckman d.c.plasma - Cchelle spectrometer are presented in Table 4. It should be noted that the detection limits in the charge injection device system were actually run under one set of conditions for all elements concerned and not optimised from element to element. On comparing the observed with the published detection limits, one sees that certainly the system is very competitive, sometimes beating and in most instances at least matching the commercial system. Additionally, the dynamic ranges for the CID system extend to well over 10 000 p.p.m., one to two orders of magnitude better than those available from the commercial instrument. Additionally, the system is able to perform in very complex matrix systems, as shown in Table 5.Note that the detection limits are degraded only slightly for iron, not at all for chromium and by a factor of 10 for calcium and nickel. This is minimal when one considers the complex line structure presented by an element such as gadolinium at such a high concentration. An over-all schematic diagram of the system is shown in Fig. 10. As mentioned previously, when properly operated, CIDs have essentially zero dark current. This means that the devices must be maintained at or near liquid nitrogen temperature, necessitating a liquid nitrogen Dewar system. The over-all instrument is controlled by a Motorola 68 000 base processor hosted in a multibus configuration with a special camera controller developed by Photometrics (Tucson, AZ, USA). This camera controller uses 2901 bit slice bipolar processors operated in a pipeline mode to achieve the necessary high-speed control of the array detector. Realisation of an Intelligent Spectrometer Now that we have presented an overview of the spectrometer system, let us go back and ask a few simple questions regarding the analysis of individual samples: (1) Do we want an answer, or do we want a valid analysis? This is not a trivial question.All of us realise that many modern-day instruments only provide a number and not necessarily a valid analysis. (2) Do we always analyse very similar samples, or do we receive widely varying samples, regarding matrix, the elements of interest, etc.? (3) Do we have unlimited time to analyse each sample? Unfortunately, time is money.To achieve a valid analysis using current technology requires a combination of the following. Initially we must analyse the sample to determine the potential matrix effects, then we must select the appropriate wavelengths for each analyte species. We must validate the procedure using synthetic or NBS standards in as similar a matrix as possible. Next we actually analyse the sample. Note that we had to analyse the sample to find out what type of matrix was present, so that later we could actually analyse the sample in a valid manner. Finally, we must check by analysing the sample by various other techniques and/or the use of standard additions, etc. This entire process can be greatly simplified by properly employing the capabilities of the previously discussed spec- trometer.In fact, with this spectrometer’s capabilities we are able, for the first time, to achieve a truly intelligent instrument that can make intelligent decisions on the sample as it isANALYST, APRIL 1987, VOL. 112 353 actually being analysed. What do we mean by intelligent instrumentation? Webster defines intelligence as “the ability to learn or understand or to deal with new or trying situations.” In the role of the modern chemical laboratory, an intelligent instrument should have the ability to reason and apply logic at a level normally associated with the human mind. Through proper utilisation of the CID Cchelle spec- trometer’s capabilities, it is possible to observe rapidly and non-destructively the entire emission spectrum of a particular sample during the early stages of the analysis (milliseconds).This provides the data necessary for the instrument to make intelligent decisions, choosing the following parameters: Which wavelengths are appropriate for use with this particular sample based on the observed concentra- tions of each analyte? The observed concentrations of other potentially interfering analytes? The effects of the host matrix, solvent, etc. Which read-out modes are most suitable? How many detector elements are to be read for each wavelength for signal and background correction? Which data reduction modes will be employed? We have a wide variety of data for carrying out sloped base-line corrections, etc. Which diagnostic procedures are desirable? We may want to monitor the argon emission from the plasma at several different wavelengths indicating changes in actual plasma excitation conditions.We may want to monitor hydrogen emission associated with the sol- vent; this would give us information concerning drifts and changes in the nebuliser. We may want to observe carbon emission; this can tell us if the solvent has been changed in a particular sample from aqueous to organic solvent material. If aqueous solutions were used for standardisation, the nebuliser will change the nebulisation rate and we would have an invalid analysis. During analysis the system chooses the optimum integration time for each analytical wavelength, remembering that there might be a number of analytical wavelengths for each element sought, collects the appropriate background data for each analytical wavelength and monitors the various diagnostic parameters.Following the analysis, the system reduces and presents the data, compares results obtained for each element at each wavelength employed for that element and, using this, can estimate the accuracy and precision for each element. Additionally, the system notes any unusual circumstances, i.e., the example of an organic solvent suddenly substituted for an aqueous set of solutions when the standardisations had been carried out with aqueous standards. employed to improve laboratory productivity. Automation of conventional procedures is not always the most appropriate mechanism to achieve optimum results. In oil analysis, new nebulisation approaches completely sidestep the need for sample preparation, eliminating a whole series of complex, intricate procedures.Although these procedures could cer- tainly be automated through the use of robotics, why bother? The second example illustrates how proper utilisation of the information content provided by a technique can significantly reduce the amount of labour necessary to achieve reliable valid analyses. This example also in the view of the author, points toward the next major breakthrough in laboratory automation-intelligent instrumentation, often referred to as artifical intelligence. These studies were supported in part by the Office of Naval Research and Dow Chemical. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. References Denton, M. B., Fry, R. C., and Windsor, D. L., “Advances in Nebulization Techniques for Spectrochemical Analysis,” Paper No. 73, 1976 Pacific Conference on Chemistry and Spectro- scopy. Fry, R. C., and Denton, M. B., Anal. Chem., 1977,49, 1413. Fry, R. C., and Denton, M. B., Appl. Spectrosc., 1979,33,393. Denton, M. B., Algeo, J. D., Sims, G. R., Phillips, H. A., and Hoek, F. B., “Optimization of Instrumentation for Oil Analysis,” Paper No. 314, FACSS National Meeting, Septem- ber 22, 1982. Algeo, J. D., Heine, D. R., Phillips, H. A., Hoek, F. B. G., Schneider, M. R., Freelin, J. M., and Denton, M. B., Spectrochirn. Acta, Part B , 1985, 40, 1447. Reed, T. B., J . Appl. Phys., 1961, 32, 821. Greenfield, S., Jones, I. L., and Berry, C. T., Analyst, 1964, 89, 713. Talmi, Y., Anal. Chem., 1975,47, 658A. Talmi, Y., Anal. Chem., l975,47,699A. Dereniak, E. L., and Crowe, P. G., “Optical Radiation Detectors,” Wiley, New York, 1984. Clary, M. C., Klassen, K. P., Snyder, L. M., and Wang, P. K., Proc. SPIE, 1979,203, 98. Weimer, P. K., and Cope, A. D., in Kazan, B., Editor, “Advances in Image Pickup and Display,” Volume 6, Academic Press, New York, 1983, p. 177. Aikens, R. S., Lynds, C. R., and Nelson, R. E., Proc. SPZE, 1976, 78, 65. Felkel, H. L., Jr., and Pardue, H. L., Anal. Chem., 1977,49, 1112. Conclusion Two different concepts have been presented in the hope that they serve as good examples of how new technology can be Paper A61447 Received November 21st, 1986

ISSN:0003-2654    

DOI:10.1039/AN9871200347

出版商:RSC    

年代:1987

数据来源: RSC

摘要:

ANALYST, APRIL 1987, VOL. 112 Recent Advances in the Theory of Atomic Absorption Spectrometry: Plenary Lecture Boris V. L'vov Atomisation the Oxygen 355 in Graphite Furnace - Carbon Alternative* Department of AnalyticaKhemistry, Polytechnical Institute, Leningrad 79525 7, USSR Two approaches to the problem of incomplete analyte atomisation in graphite furnace atomic absorption spectrometry based on the formation of either monoxides or carbides in the gas phase are compared. The discussion of the first approach provides an explanation of an apparent discrepancy between the comparatively high (10-8-10-7 atm) free oxygen partial pressure in the sheath gas and the absence of its effect on atomisation under typical analytical conditions. Within the second approach we have revealed, experimentally studied and theoretically interpreted the major features of the appearance in graphite furnaces of gaseous carbon at enhanced levels compared with equilibrium concentrations.The theory of gaseous carbide formation developed on this basis has provided an explanation for many unusual effects caused by the furnace material, sample vaporisation techniques and sheath gas (Ar vs. N2) on the shape and magnitude of analytical signals, and theoretically substantiated the advantages of the stabilised temperature platform furnace. Keywords: Graphite furnace atomic absorption spectrometry; graphite activation; metal - O2 interaction kinetics; non-equilibrium gaseous carbon; gaseous carbides State of the Art Today, almost 30 years after the birth of graphite furnace atomic absorption spectrometry (GFAAS) ,1 and about 15 years after the advent of the first commercial electrothermal atomisers, we are in a position to maintain that this method has won the recognition of analytical chemists all over the world as one of the most sensitive, selective and simple analytical techniques.The automation and computerisation of instrumentation, the refinement of the atomisation technique and the use of the Zeeman effect to take into account spectral interferences have made AA spectrometers a reliable tool for completely automated analysis. The concept of the stabilised temperature platform furnace (STPF) introduced by Slavin2J in the early 1980s, and presently enjoying widespread use, has brought us close to the development of an absolute method of analysis.4 A possibility has emerged for the first time in the history of instrumental methods of excluding from analytical procedures the use of standards and even regular calibration using pure analyte solutions.Against the background of these truly revolutionary5 technical and methodological achievements one would con- sider all the more anachronistic the slow progress in the theory of atomisation, the absence of consensus and, sometimes, even of any opinion whatsoever regarding the variety of unusual effects in GFAAS, some of which are outlined below. The reason for the anomalously low sensitivity for a number of elements in graphite furnaces, in particular for boron and the lanthanides, remains unclear. The reasons for many of the effects of furnace material on atomic absorption are also unknown.6.7 For example, when evaporated from the wall of an uncoated graphite tube, the characteristic masses of Al, Ge, Si and Sn are smaller than those obtained with pyrolytic graphite coated tubes, whereas for V, Mo and Ti the reverse is true.6 Nothing can be said about the reasons for the enhancement of the signal when analytes are vaporised from the platform and lanthanides in a Ta-foil lined furnace.8 An explanation cannot yet be suggested for the enhanced ~~ *Plenary lecture at SAC 86, the 7th SAC International Conference on Analytical Chemistry, and BNASS, the Third Biennial National Atomic Spectroscopy Symposium, Bristol, UK, 2&26 July, 1986. (Unable to be presented.) sensitivity for some elements (Pd, Ge) in nitrogen compared with argon, and for differences in the vaporisation rates of B, Ba, Sr, Er, Eu, Sc, Ti, V and some other elements in nitrogen and argon.9 There is no consensus as to the interpretation of the molecular spectra obtained with graphite furnaces.Some authors10J believe monoxides to be responsible for the spectra observed in the vaporisation of oxygen-containing compounds of Al, Ba, Sr and Ca, whereas others12313 attribute them to monocyanides and carbides. The purpose of this paper is to discuss the status of the problem and to explain these and some other unusual phenomena based on the results of our recent studies. In agreement with most researchers, we believe that the key to understanding the mechanism of these phenomena lies in the interaction of the analyte with oxygen and carbon.Let us consider this problem from the viewpoints held by the proponents of the oxygen and carbon hypotheses. Oxygen Differences in Evaluating the O2 Partial Pressure The partial pressure of free oxygen present in graphite furnaces has recently become a subject of heated debate between several research groups. Three different approaches to this problem have surfaced, namely, an equilibrium approach,14-16 based on assuming thermodynamic equilibrium to exist between the carbon and the sheath gas; a hypothetical approach17.18 involving an ( a priori) assumption of a linear variation of the quantity -loglop(02, atm) from 10 to 20 in the range 1500-2500 K, and an empirical approach's21 based on direct and indirect measurements of the oxygen partial pressure. An analysis of the results and arguments presented in some of these papers has revealed a number of aspects that are either misunderstood or left out of the considerations.Total Oxygen Pressure in Graphite Furnaces Some comparisons18.21 of the hypothetical with the empirical p02 scales neglect the fact that in both direct and indirect measurements the total pressure of oxygen is determined, i.e., N 2 ) r = P(02) + 0.5 p ( 0 ) * * . * (1)356 ANALYST, APRIL 1987, VOL. 112 rather than p(02). At high temperatures these quantities may differ substantially. Indeed, at 2500 K, p(02) = 10-20 atm corresponds t o p ( 0 ) = 1.4 x 10-12 atm, so that ~ ( 0 ~ ) ~ = 0.7 X 10-12 atm. Thus the difference between the empirical and hypothetical scales at 2500 K constitutes only five orders of magnitude rather than twelve18.21 (Table 1).Note that this difference remains approximately constant over the range 1500-2500 K. Arguments Against the Empirical ~ ( 0 2 ) Scale and their Criticism The main argument advanced against the empirical scale in which p(02) varies from 10-6 to 10-7 atm over the range 1500-2500 K is the thesis18 that at such oxygen pressures the atomisation of a number of elements (B, Si) forming thermally stable metal - oxygen molecules is practically impossible. In order to estimate the degree of dissociation of the monoxides, the existence of the following equilibrium is tacitly assumed: MO(g) =M(g) + 0 . . . . . It has, however, been pointed out19 that as we are speaking here about the excess oxygen pressure above the equilibrium level, “this conclusion is valid only in the case where the partial pressure of element in the gas phase at each moment of time is much less than that of oxygen, i.e., p(M) << pz (02).In the opposite case, practically all of the oxygen present will be bound in a gaseous oxide MO, the purge gas becoming ‘pure’ for the excess amount of the element.” It is this which accounts for the noticeable absorption of silicon and boron18 when they are introduced into the furnace in amounts (2 X 10-10 and 5 X 10-9 mol, respectively) exceeding by far the total amount of free oxygen present in the furnace under these .conditions (7 x 10-12 mol, 2500 K, argon flow 0.8 cm3 s-1).13 Another argument put forward against the empirical scale is the so-called buffering effect of the excess CO that is assumed18.21 to maintain a low and constant concentration of oxygen in the furnace, irrespective of its original content in the sheath gas.This buffering effect, however, is based on the same heterogeneous equilibrium C(S) + 0 . 5 0 2 = CO . . . . * * (3) that underlies the purely thermodynamic approach. 14916 As shown convincingly by Sturgeon et al. ,21 this equilibrium is not reached. Moreoever, contrary to the conclusions drawn by the above workers,z1 their data on the partial pressures of CO and C02 prove persuasively that the partial pressure of free oxygen corresponding to the equilibrium C 0 2 = C O + O . . . . . . ( 4 ) setting in at temperatures above 2200 K is in good agreement with earlier measurements.19J0 This is obvious from Table 2, which presents the values of p(02)x calculated by us.Thus all methods for the direct or indirect determination of the O2 partial pressure in furnaces with even a low argon flow (ca. 0.4 cm3 s-I), without exception, yield results that agree well with one another. This supports the validity of the empirical scale. Kinetic Limitations of the M - O2 Interaction In evaluating the equilibrium M(g) + 0 2 = MO(g) + 0 . . . the finite time (tR) of atom residence in the furnace, which restricts substantially the number of collisions between interacting particles, is not taken into account. The impor- tance of this factor in gas-phase equilibrium studies has been pointed out by Holcombe et a1.22 as early as 1979, although this investigation was carried out using rod rather than tube atomisers.Table 3 presents the lowest concentrations and partial pressure of oxygen required for a single collision of the metal atom with the O2 molecule to occur, calculated by us for different experimental conditions. The number of collisons was calculated by well known relationships,23 assuming for the mean radius and mean molar mass of the colliding particles the values 1.5 x 10-10 m and 0.05 kg mol-1 for T = 2000 K, p = 1 atm. As seen from Table 3, the minimump(02) that still can be of significance for process (5) is ca. 10-9 atm under gas-stop conditions, ca. 10-8 atm with a flow of 0.8 cm3 s-1 and ca. 10-7 atm with a flow of 5 cm3 s-1. At still lower values ofp(02) the Table 1. Total oxygen pressure in graphite furnace Argon p(02)datm flow-rate/ Scale Reference cm3 s-l 1500 K 2000 K 2500 K Equilibrium .. . . 14 - 2 x 10-18 5 x 10-15 5 x 10-13 Hypothetical, . . . 17 - 2 x 10-1’ 1 x 10-11 5 x 10-13 Empirical . . . . 19 0.8 5 x 10-7 2 x 10-7 4 x 10-8 20 0.4-2.1 1 x 10-6 2 x 10-7 4 x 10-8 Table 2. Calculation of p ( 0 2 ) , from measurementsz1 of p(CO,)/p(CO) ratio. In a T-shaped pyrocoated tube, at stationary tem- perature, and under an argon flow-rate of 0.4 cm3 s-1 P(CO2) T/K p ( C 0 ) K*latm p(0)latm p(0,)latm p(Oz),/atm 2118 5.1 x 10-2 5.07 x 10-6 2.6 X lo-’ 2.9 X 10-8 1.6 x 10-7 2378 5.9 x 10-3 1.42 x 10-4 8.4 x 10-7 1.2 x 10-8 3.3 x 10-7 2620 3.6 X 10-4 1.68 X 10-3 6.0 X lo-’ 5.7 X 10-lo 3.0 x 10-7 * Equilibrium constant for reaction (4). ~~~~~ Table 3.Minimum values of concentration (nmln) and partial pressure (pm,,,) of gaseous reagent for a single collision of analyte atom with a reagent molecule in an HGA-type furnace Sheath gas Gas flow-rate/ nrnd flow mode cm3s-1 tR/s molecule cm-3 pminlatm Gasstop . , . . - 0.25 1.2 x 1010 3.3 x 10-9 Mini-flow . . 0.8 0.05 6.2 x 1Ol0 1.7 X Fullflow.. . . 5 .O 0.01 3.1 x 10” 8.3 x 10-8ANALYST, APRIL 1987, VOL. 112 357 metal atoms leave the furnace without undergoing a single collision with the O2 molecules. Taking into account the kinetic limitations of reaction (5) determined by the rate constants of the metals involved, the minimum values ofp(02) at which chemical equilibrium can be obtained may be higher still. Hence, any debate on the reasons for the variation of p ( 0 2 ) in the sheath gas below the above levels has no practical significance.Attempts at a thermodynamic evaluation of reaction (5) under these conditions appear to be just as useless. Conclusions Thus, the residual partial pressure of oxygen in a pure sheath gas (<1 x lO-3% 02) with even a low flow-rate (0.4 cm3 s-1) through the graphite furnace and for temperatures up to 2500 K is not less than 10-7 atm. When measuring metal vapour pressures less than 10-7 atm this may result in a reduced senstivity for some elements because of the formation of MO molecules. The effect of O2 becomes more pronounced for higher sheath gas flow-rates or for higher O2 partial pressures in the sheath gas. At low furnace temperatures typical of the vaporisation of volatile elements (Zn, Bi, Pb) this may affect not only the peak area but also its position. This is connected with the effect of O2 on the process of oxide vaporisation.19 At the higher temperatures required for the vaporisation of In, Ga, Sn and Si the concentration of O2 near the furnace wall decreases, with the result that the effect of O2 is confined to the furnace axis and manifests itself in a reduced peak area.This conclusion is in accord with the experiments of Byrne et al. 16 At metal vapour partial pressures considerably in excess of 10-6 atm the influence of O2 on the peak area is negligible, even with intense sheath gas flow-rates, owing to the purification effect.19 The only consequence of this effect is the appearance of a curvature in the calibration graph when the amounts of O2 and M in the gas phase become comparable.For the typical analytical conditions of temperatures above 2200 K and the gas-stop mode, the residual oxygen pressure in pure sheath gas in a pyrocoated tube is below 10-8 atm.19 Its effect on the monoxide formation may be neglected even at the furnace axis as firstly, the analyte partial pressure in the furnace, as a rule, is greater than 10-8 atm and, secondly, the number of the metal atom collisons with 0 2 molecules becomes insufficient for equilibrium ( 5 ) to occur. Thus free oxygen pressure variations in the sheath gas cannot account for the numerous effects of the furnace material and sheath gas on the sensitivity and vaporisation rates of analytes under typical analytical conditions.Other possibilities should be explored, Carbon Historical Background Our interest in the formation of stable gaseous compounds of metals with carbon as a possible reason for the incomplete atomisation of elements in carbon-containing media began more than 10 years ago. In particular, in a paper24 published in 1976, we suggested that the decrease of the integrated absorbance signal for a large number of elements vaporised from a graphite rod in a hydrogen flame with an addition of 5 8 % C2H2 is due to the formation of MC,-type compounds. Later, in 1979, in a report12 dealing with the problem of monocyanide formation in AAS, we stated that: “For a longer period of time we have been living in an ‘oxidising’ world and have got used to the idea that monoxides and hydroxides represent the only obstacles in our way to solving the problem of complete and over-all atomisation. With the advent of high-temperature reducing flames and graphite furnaces a hope began to dawn that free carbon present in atomisers of these kinds would help to solve this problem.However, this did not happen. Having eleminated the previous obstacles, we stumbled upon others. The reducing medium did not remain inert with respect to free atoms and ‘issued,’ in place of monoxides and hydroxides, their carbon analogues, i. e. , dicarbides and monocyanides. In this respect, the reducing world turned out to be a symmetric image of the ‘oxidising’ one. ” There was sufficiently firm ground for drawing such a conclusion. It may be recalled that systematic studies of gaseous carbides by high-temperature mass spectrometry started more than 25 years ag0.25 During this period the structural and thermodynamic characteristics of gaseous carbides of many elements were thoroughly investigated.As seen from Table 4,26 carbide molecules were found to be formed with elements of all sub-groups of the Periodic Table, with the exception of IIB and VIIIA. Apart from the compounds listed in Table 4, some elements are known to form other carbides, e.g., MC3. Of particular interest are MC2 molecules. According to the view first put forward by Chupka et al.25 the C2 radical in MC2 may be regarded as pseudo-oxygen. Therefore, the MC2 molecules should be considered as typical compounds of the elements as are their gaseous monoxides.In most instances these molecules are the most thermally stable. An exception is the VIIIB sub-group elements, for which the most thermally stable are the MC molecules, and also some element of the IVB sub-group, for which the most stable molecules of the MC4 type. At the same time it has to be admitted that for most elements the fraction of carbide molecules under thermo- dynamic equilibrium is very small, with the exception of elements with energies of dissociation Do (M-C2) 3 660 kJ mol-1 (Ce, Hf, La, Pr, Si, Th and U). In the example of aluminium, for instance, at 2000 K the equilibrium partial Table 4. Known gaseous carbides.26 The symbols indicate that the carbide exists Carbide Periodic Table sub-group Element* MC MC2 MC4 M2C M2C2 IA . . .. H + + + + IB _ . . . Cu + + IIA . . . . Be + Ba + IIIA . . . . B + + + A1 + + + + Ga + IIIB . . . . Sc + + + Y + + + Ln + + + An + + + IVA . . . . Si + + + + Ge + + + IVB . . . . Ti + + + Zr + + + Hf + + + VA . . . . P + + + + VB _ . , . V + + + Nb + + VIA . . . . S + + Se + + + VIB . . . . Cr + Mo + + VIIA . . , . Hal + + + + VIIB . . . . Tc + VIIIB . . . . Ru + + Rh + + 0 s + Ir + + + Pt + + + * Ln = lanthanides; An = actinides; Hal = halogens.358 ANALYST, APRIL 1987, VOL. 112 pressure ratio p(MC2)/p(M) is 5 x 10-6 (Do = 477 kJ mol-1) and for phosphorus it is 5 x 10-4 (Do = 510 kJ mol-1). It is apparently for this reason that workers in the field of pyrometallurgy and carbon chemistry studying the mechan- isms of carbon interaction with metals disregard the possibility of gaseous carbides being involved in these processes.It is for the same reason also that in the late 1970s we stopped halfway in the development of this version. Fortunately, in our studies in the early 1980s of the mechanism by which yg amounts of A1203 are vaporised in graphite furnaces we have succeeded in revealing27 a remark- able effect consisting in the appearance of fast spikes of signals against the main smooth pulse, which originate from the thermal dissociation of A1203 (Fig. 1). In a subsequent investigation of this effect it was shown28-30 to be connected with the reduction of the oxide by carbon, and established that the transfer of carbon from the furnace walls to the oxide involves A12C2 molecules the concentration of which is comparable to that of the A1 atoms.Thus, contrary to the above-mentioned thermodynamic calculations, we came to the same conclusions as when studying the reasons for incomplete vapour atomisation in electrothermal atomisers. Thus, all the data suggest that during atomisation carbon in no way behaves by the rules, and that it is more active toward metals than is so under thermodynamic equilibrium. This implies that it should also vaporise more easily than graphite, which is at thermodynamic equilibrium. These arguments have been put forward by us26 in a preliminary form in 1984. In the section dealing with the reasons for the remarkably high activity of carbon it was stated that: “For the ideal single crystal of graphite, a hexagonal lamellar structure is charac- teristic. Each of the carbon atoms in the plane of a layer is bonded to three immediately neighbouring atoms.The bonds to the atoms of the adjacent layers are considerably weaker, such that they can be neglected. In practice, as a consequence of the numerous defects in the crystal structure in connection with the small dimensions of the grains of polycrystalline graphite, the presence of impurities, and particularly the 1.6 I 1.4 - 1.2 - g 1.0 - a C m 0.8 n a 0.6 0.4 - - - J 0 20 40 60 Tirneis Fig. 1. Vaporisation of 1 pg of A1 as AI(NO& from the wall of the pyrolytic graphite coated tube heated at a rate of 17 K s-1 in the temperature range 1800-2800 K under a stopped-flow of Ar activation of the surface by foreign gases and metal vapours, a certain proportion of the atoms on the surface of the tube will prove to be bonded not to the three neighbouring atoms, but to a smaller number, such that their detachment from the defective layer becomes far more probable.Of course, if activated graphite is heated for a sufficiently long period, the number of defects will decrease as a result of recrystallisation, such that the vapour pressure of the carbon will gradually approach the equilibrium value. Therefore, the greatest excess of the pressure over the equilibrium pressure should be expected either with continuous activation of the surface by foreign gases (for example, O2 present as an impurity in the argon) or at the initial moment of heating activated graphite, even under vacuum conditions. ” In order to account for the formation MC2-type gaseous carbides in amounts comparable to the free atom concentra- tion it was assumed26 that the C2 partial pressure during sample atomisation in the graphite furnace at 2200 K exceeds by four to five orders of magnitude the tabulated (thermo- dynamic equilibrium) value.Despite the fact that these estimates had support from some of the observations of other workers, including mass spectrometric measurements, many workers considered this hypothesis with extreme scepticism. Thus, the behaviour of carbon in the course of graphite furnace operation became the key problem in further studies into the mechanism of the atomisation of analytes in GFAAS. Its solution has been found to be an easier task than we had anticipated. Determination of Gaseous Carbon in Graphite Furnaces Experimental We used three different methods to measure the content of the carbon vapour present (Table 5): (i) direct determination31 of C2 molecule concentration, n(C2), from the absorption and emission of the (0,O) Swan band head; (ii) indirect determi- nation of the concentration of n(C2) in a nitrogen-sheathed furnace from the absorption and emission of the (0,O) CN band-head based on the equilibrium C2.g) + N2=2CN(g) .. . . . and (iii) indirect determination32 of the concentration of n(C2) by successively measuring the absorption of the A12C2 band maximum and of the A1 auto-ionisation line at 193.6 nm. The C2 concentration was calculated from the equilibrium . . . . (7) The aluminium was introduced in the graphite furnace either as Al(N03)3 solution or in the form of metal granules of about 1 mg.In contrast to the preceding measurements,32 we have taken account of the background absorption near the 206 nm band and 193.6 nm line. Although the method of direct determination of the C2 content is preferable over the indirect techniques, its use is restricted to temperatures above 2700 K. Using all three methods permitted the whole temperature range of interest to be covered (1800-3000 K) for both sheath gases, namely, argon and nitrogen. The application of independent methods improved the reliability of the data obtained. ~~ Table 5. Methods of gaseous carbon (G) determination Method of Background measurement Band or linehm correctionhm Absorption . . . . C2516.5 - Absorption . . . . CN388.3 - Absorption .. . . A1193.6 200 A12C2 206 202 Emission . . . . G516.5 516.8 Emission . . . . CN388.3 388.6 Sheath Ar - N2 3000 0.04 A r - N 2 2700 0.04 N2 2500 0.07 N2 2400 0.07 Ar 1800 0.07 Ar 1800 0.07 gas T,,,/K Slit widthlnmANALYST, APRIL 1987, VOL. 112 I 359 Equipment. All measurements were carried out on a Perkin-Elmer Model HGA-500 atomiser fitted with a Model 5000 AA spectrometer. The absorption and emission signals were recorded with a 3500 data station and a graphics plotter. Standard graphite tubes both with and without pyrolytic graphite coating and platforms made of anisotropic pyrolytic graphite were used in the experiments. High-purity argon and nitrogen, containing not more than lO-3% 0 2 , served as the sheath gases. Procedure. The signals were measured in the fast tempera- ture ramp mode (ca.2000 K s-1) under stopped-flow conditions. The absorbance and emission signals were measured at a constant furnace temperature for 60 s. At temperatures above 2600 K the recording time was limited to 40 s because of overheating of the atomiser. In the emission studies the signal measurement at the band maximum was followed by determining the background emission from the heated walls of the furnace, primarily from its ends. Fig. 2 illustrates absorbance and emission traces of the C2 band in Ar at 3000 K, Fig. 3 displays similar records for the CN band in N2 at 2600 K and Fig. 4 shows absorbance traces of the A12C2 206-nm band and A1 193.6-nm line, together with the background in Ar at 1800 K. The aluminium was introduced into the furnace before heating in the form of metal granules.Results and discussion The traces in Figs. 2 and 3 reveal clearly that in all instances the signals connected directly or indirectly with the gaseous carbon concentration in the furnace are at a maximum at the onset of constant furnace temperature. Subsequently, these signals decay gradually, tending to constant values. Let us assume that the signals obtained in pyrolytic graphite coated furnaces reach their equilibrium values characteristic of the given temperature within 60 s (or 40 s for T 3 2700 K). The excess concentration of the constituent in question can then be quantitatively evaluated for each moment of time by relating the running values of the signals to their minimum values for the tails.In the emission measurements we took the background into account by preliminarily subtracting the background signal from each running value. The resulting graphs plotted from the data in Figs. 2 and 3 and characterising the variation of the excess concentration of the C2 molecules with heating time are displayed in Figs. 5 and 6. It can be 0 1 2 3 4 5 6 7 8 Ti rn els Fig. 2. function of time in a pyrolytic graphite coated tube at 3000 K in Ar (a) Absorption and ( b ) emission of the C2 516.5-nm band as a readily seen that the absorption and emission measurements yield close results when carried out under the same conditions. The results displayed in Fig. 7 would at first glance appear somewhat unexpected. In contrast to the behaviour of the C, lines in Figs.5 and 6, the calculated excess concentration of the C2 molecules at 1800 K in the presence of A1 vapour continues to increase slightly after the stationary furnace temperature has been reached. The ratio n(Cz)lr~(C2)~~ in this instance was calculated by the expression The numerical coefficient in equation (8) was determined at 2500 K for the ratio n(C2)/n(C2),, derived from Fig. 6. The 0.07 .s 0.06 2 0.05 I .= 0.04 +.? ? 0.03 0 3 *; 0.02 .- E LIJ 0.01 0 L , 0 2 4 6 8 10 12 14 16 18 Time/s (a) Absorption and ( b ) emission of CN 388.3-nm band as a Fig. 3. function of time in a pyrolytic graphite coated tube at 2600 K in N2 0.7 - (a) 0.6 0.5 0.4 0.3 0.2 - - - - - m +? 0.1 - I I 0 20 40 60 Timels Fig. 4. Absorption of (a) AI2C2 206-nm band and (b) A1 193.6-nm auto-ionisation line as a function of time in a pyrolytic graphite coated tube at 1800 K in Ar360 ANALYST, APRIL 1987, VOL.112 I 2 1 $ 0 . 2 1 Absorbance I I 8 I 0 1 2 3 4 Time/s Fig. 5. Time dependence of the ratio n(&)ln(CJ,, calculated from (a) absorption and (b) emission data of Fig. 2 equilibrium constant K(A12C2) for reaction (7) and the equilibrium partial pressure p ( Q , , were taken from tables.33 Apart from obtaining the dependence of n/neq as a function of time and temperature of the pyrolytic graphite coated furnace, we measured the radial distribution of the ratio nlneq and also the dependence of n/neq on furnace type ( i . e . , with and without a pyrolytic graphite coating) and sheath gas used. Bearing in mind the higher sensitivity of the emission compared with the absorption method, the major part of these experiments consisted in emission studies of the CN band.An analysis of the totality of the data obtained brings us to the following conclusions: (i) In the initial stage of furnace heating the concentration of the gaseous carbon (C2 molecules) is much higher than the equilibrium value. As the temperature increases, the maxi- mum value of this excess (n/neq) drops from about 105 at 1800 K to a factor of 2-3 at 3000 K (Fig. 8). (ii) At a constant A1 vapour pressure (ca. 10-3 atm at 1800 K) in the furnace the relative excess of C2 concentration, n/neq, increases slightly in time from about 1 x 104 to 4 x l o 4 (Fig. 7). (iii) In the absence of metal vapour (a “clean” furnace) the maximum absolute concentration of the C2 molecules remains approximately constant above 2400 K (Fig.9). (iv) The excess of C, molecules increases rapidly from the axis to the wall of a clean furnace (Fig. 10). As the height of the zone viewed by the monochromator is about 2.5 mm, the true variation of n/neq along the furnace radius is probably much stronger than would follow from Fig. 10. (v) The ratio nin,, behaves differently in time for clean furnaces with and without pyrolytic graphite coating at equal temperatures (Fig. 11). Initially the ratio n/neq in the pyrolytic graphite coated furnace is greater than that for the uncoated furnace. A few seconds later the pattern changes, namely, the ratio n/neq in the pyrolytic graphite coated furnace is greater than that for the uncoated furnace.The total excess carbon, f(n/n,,)dt, measured during the signal recording time (60 s) 700 600 Absorbance - 0 1 2 3 4 5 Ti meis Fig. 6. Time dependence of the ratio n(C,)/n(&),, calculated from (a) absorption and (6) emission data of Fig. 3 60 000 50 000 40 000 U -$ 30000 20 000 10 000 0 20 40 I Timels Fig. 7. Time dependence of the ratio n(Cz)/n(C2)eq in the presence of A1 metal calculated using equation (8) from data of Fig. 4 for the uncoated furnace is a few times greater than that for the coated furnace. (vi) For the same clean furnace, the value of n/neq obtained in nitrogen is less than that for argon (Fig. 12). Insignificant initially, this difference reaches a factor of 1.5-2 about 2-5 s after the heating is switched on. Causes of the Appearance of Non-equilibrium Gaseous Carbon Let us turn back to the reasons for such a remarkable behaviour of carbon in graphite furnaces and attempt to relate it with other much more natural and universally adopted ideas concerning the properties of graphite.Among these is, in particular, the active chemisorption of oxygen that is revealed at low temperatures. Subsequent heating results in a desorp- tion of oxygen, not in the form of 02, but rather as CO molecules produced in the dissociative chemisorption of 0 2 and, to a lesser extent, of C02 molecules. The desorption ofANALYST, APRIL 1987, VOL. 112 14 361 4 / - / 0‘ I 1 I I I I 1800 2000 2200 2400 2600 2800 3000 TemperaturelK Fig. 8. pyrolytic graphite coated tube as a function of temperature Maximum relative concentration of C2 molecules in a / / / I , 9 1800 2000 2200 2400 2600 2800 3000 I I I 1 I ~- 1 Temperature/K y Fig.9. Maximum absolute concentration of C2 molecules in a pyrolytic graphite coated tube as a function of temperature. The dashed line corresponds to a probable change in concentration these molecules starts at 60&700 K and is complete above 1300-1500 K. The chemisorption of O2 on graphite occurs over a relatively small fraction of surface area, which does not exceed 1OY0.34 Therefore, it may be assumed that the desorption of CO and C02 involves the detachment from the surface of the crystal lattice of only single atoms (in the form of C02 molecules) or of pairs of neighbouring atoms (in the form of CO molecules). The detachment of single carbon atoms (or of pairs of neighbour carbon atoms) results in a rupture of three (or five) bonds.If we recall that the mean number of bonds per carbon atom is initially (for a thermodynamically perfect lattice) 1.5, it becomes clear that the desorption of chemi- sorbed oxygen brings about a decrease of this number. The detachment of C 0 2 molecules creates in the lattice three times, and that of CO molecules twice, the number of “defective” carbon atoms with one rather than 1.5 bonds per atom. As the enthalpy of vaporisation of “defect-free” graphite is 716 kJ mol-1 at 2000 K,33 the difference in the detachment energy (AE) between a normal and defective carbon atom should be 239 kJ mol-1. This entails a difference in the carbon partial pressure of a factor exp(AE/RT), which is 1 X lo5 for 2500 K, 1.7 x 106 for 2000 K and 2.1 x 108 for 1500 K.Hence, the presence of defects in the graphite crystal structure should eventually increase the volatility of carbon compared with the perfect structure. These arguments are valid for any type of carbon-containing materials, both for a perfect single crystal, which may be 4.0 - 3.9 - 3.8 - 3.7 - x 3.6 - ,E 3.5 c“ 3.4 - 5 3.3 - - U . 3.2 - 3.1 - 3.0 - 2.9 - 2.8 ,b 2.7 I 1 I I I 0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 Distance from axis/mm Fig. 10. n(CN),, in a pyrolytic graphite coated tube at 2600 K in N 2 Radial dependence of the maximum value of n(CN)I 20 U c“ . 10 0 1 Coated 1 0 1 2 3 4 5 Timeis Fig. 11. Time dependence of the ratio n(CN)In(CN),, on the tube used at 2600 K in N2 0.3 r 1 v) C 4- .- = 0.2 2 P +? - 4- .- C 0 .- : 0.1 E - .- w 0 1 2 3 4 5 6 7 8 9 1 0 Timels Fig.12. Emission of C, 516.5-nm band in Ar and N2 at 3000 K considered to approximate the surface of the pyrolytic graphite coating, and for ordinary polycrystalline graphite. The only difference between them, other conditions being equal, is that the degree of surface activation of polycrystalline graphite is higher than that of pyrolytic graphite coated material. The reason for this is, mainly, the greater real surface area of polycrystalline material. When pre-activated material is heated, the first atoms to vaporise are the carbon atoms in the defective structure. As the defect concentration decreases gradually, the gaseous carbon content approaches the equilibrium level.362 ANALYST, APRIL 1987, VOL.112 The effect of the activation of carbon by oxygen on its subsequent behaviour and, in particular, on its vaporisation deserves further theoretical and experimental study. However, even at the present time, some of the results already obtained by us together with available literature data may be considered as lending support to the above reasoning. In particular, in complete agreement with the above, the total absolute amount of carbon released from furnace walls remains practically constant irrespective of the furnace temperature (item iii). The total amount of surface carbon for a standard furnace with an inner and outer surface area of 12.3 cm* is 4 x 1016 atoms, corresponding to a mass of 1 pg. According to the measurements of Sturgeon,35 a single heating of the furnace for 6 s to 2000 K reduces its mass by about 0.2 pg.The major loss of carbon under these conditions is due to the desorption of C02 and CO. When heated to 2650 K, the mass loss is 0.9 pg for a pyrolytic graphite coated furnace (in agreement with the calculation) and 1.5 pg for an uncoated furnace. In the latter instance the area of the activated surface exceeds clearly its geometric size. As the desorption of C02 and CO starts at 600 K,36it should be expected that the activation of graphite above this temperature is not as efficient as it is below 600 K. In order to elucidate the effect of the graphite activation temperature on the subsequent vaporisation of defective carbon, we carried out the following experiments.The same furnace was period- ically heated and cooled (with the power supply set in the automatic mode) while continuously recording the CN band absorbance. The heating conditions were constant (2600 K, 10 s), whereas the cooling temperature was varied from 2300 to 600 K (10 s). When in the cooling stage, nitrogen (5 cm3 s-1) was passed through the furnace, whereas during the heating this nitrogen flow was stopped. As follows from Fig. 13, at 2300 K there is practically no surface activation, at 1300 K the activation efficiency increases and only at 600 K does it reach a maximum. Thus an analysis of the available theoretical and experi- mental data, combined with our own measurements, suggests that the appearance of excess gaseous carbon over the equilibrium content during the heating of graphite tubes originates directly from the preliminary activation of the graphite surface by oxygen.An additional contribution to graphite surface activation may come from metal vapours, particularly at sufficiently high partial pressures. This is supported by the experiments with aluminium granules. Interpretation of Unusual Effects in GFAAS The enhanced content of gaseous carbon in graphite tubes provides favourable conditions for the formation of gaseous carbides in concentrations above the equilibrium level. This is the most substantial conclusion that can be drawn from the experiments. Only now can the hypothesis of the formation of gaseous carbides in amounts comparable to the free metal atom concentration be transferred from the realm of fantasy and fiction to that of solid fact.Only now can it be used as a basis to explain the above-mentioned observations of the effect of furnace material, sample vaporisation techniques and sheath gas on the vaporisation rate and the degree of analyte atomisation and thereby transfer them from the category of “puzzling” phenomena to that of “obvious” facts. Consider some of these effects. (i) As shown in Fig. 11, the gaseous carbon content in the initial stage of heating of the pyrolytic graphite coated furnace is somewhat greater, and a few seconds after the maximum temperature has been reached, smaller than that of the uncoated furnace. This accounts for the difference in the effects of the pyrolytic graphite coating on the sensitivity for elements of different volatility when vaporised from the 0 20 40 60 Time/s Fig.13. Effect of cooling temperature on absorption of CN band at 2600 K furnace wall. For more volatile elements (Al, Sn, Si and Ge) the degree of gaseous carbide dissociation in a pyrolytic graphite coated furnace is lower, and for low-volatility elements (Mo, V and Ti), which take time to vaporise, higher than that for uncoated furnaces.617 (ii) The use of the platform delays the vaporisation of the analyte compared with its vaporisation from the wall. At the moment of analyte vaporisation from the platform the content of gaseous carbon that is predominantly released from the wall is substantially lower than that released during the ramp heating. As a result, the degree of dissocation of gaseous carbides increases considerably, approaching unity for most elements.4 This accounts for the noticeable enhancement of sensitivity provided by the platform (Table 6).For Co, Cu, K, Li, Mg, Mn, Na, Ni, Pb, Pd and Rb the same increase in the sensitivity by a factor of 1.4 k 0.1 corresponds to the increase in the vapour density at the optical axis expected to occur if the platform is used. For the other 20 elements this effect is complemented by an enhanced degree of gaseous carbide dissociation. Note the absence of any correlation between the magnitude of the effect and the energy of monoxide dissocia- tion, D,(MO), which is still more evidence for the non-oxygen nature of the phenomenon. (iii) The three- to nine-fold discrepancy between the calculated and experimental values of characteristic mass, mo, for Er, Eu, Si and Ti4 when vaporised from the wall is likewise accounted for by the incomplete dissocation of the gaseous carbides. The dissociation energies37 of these molecules range from 540 (EuC2) to 661 kJ mol-1 (Sic2).The discrepancy between the calculated and experimental values of mo reaching a factor twelve for Ba and three for Sr4 when evaporated from the wall, is primarily due to the formation of gaseous monocyanide molecules, which are fairly stable for these metals. Although the amount of free nitrogen present in an argon-sheathed furnace does not exceed 0.1% during the vaporisation of these elements, the concentration of CN is noticeable owing to the enhanced concentration of gaseous carbon.Our experiments9 showed the degree of dissociation of BaCN and SrCN under these conditions to be about 0.2 and 0.5, respectively. (iv) Elements forming unstable monocyanides and stable gaseous carbides may have higher sensitivities with nitrogen as the sheath gas, as the concentration of excess carbon in nitrogen due to the formation of CN molecules is somewhat lower than that in argon. Our measurements suggest that Pd and Ge should be among such elements. Even when vaporised from the platform, the difference in their sensitivity is 1.3- to 1.5-fold (Fig. 14). (v) The formation of gaseous carbides during sample atomisation may be accompanied by their partial decomposi- tion on the surface of the sample particles involving the appearance of a carbon film.38 This process becomes particu- larly noticeable if the sample particles become colder than the furnace walls because of intense vaporisation of the material.ANALYST, APRIL 1987, VOL.112 363 Table 6 . Effect of platform on sensitivity. Atomisation at maximum power heating in Ar under stopped-flow conditions mo(W> * mo(P) 1.3 1.4 1.5 1.4 1.5 1.5 1.4 1.3 2.1 2.4 1.6 2.8 2.1 2.8 2.0 Do(MO)37/ mow) * Do( MO)V Element kJ mol-1 T/K rno(P) Element kJ mol-1 T/K Ag . . . . 220 2000 3.3 Li . . . . 340 2700 A1 . . . . 500 2700 2.0 Mg . . . . 360 2400 As . . . . 480 2500 1.8 Mn . . . . 410 2500 Au . . . . 220 2100 2.2 Na . . . . 250 2100 Be . . . . 440 2700 5.2 Ni . . . . 360 2700 Bi . . . . 340 2100 3.0 Pb . . . . 370 2200 Cd . . . . 280 1800 4.8 Pd . . . . 280 2700 co I... . 360 2600 1.3 Rb . . . . 250 2100 Cr . . . . 390 2700 1.7 Sb . . . . 380 2600 c s . . . . 290 2200 3.1 Set . . . . 420 2200 c u . . . . 260 2600 1.4 Si . . . . 790 2900 Fe . . . . 405 2600 1.6 Sn . . . . 530 2500 Ga . . . . 380 2600 5.5 Tet . . . . 390 2200 Ge . . . . 650 2700 7.5 T1 . . . . 310 2100 In . . . . 320 2300 2.5 Zn . . . . 270 2000 K . . . . 280 2100 1.4 * mo(W) and rno(P): characteristic masses for analyte vaporised from the wall and from the platform. t 1 pg of Ni was added. I Nitrogen A 0.8 0.7 0.6 0.5 0.4 2 0.3 0.2 0.1 0 a 0 z 0.3 a, U K 0.2 e 0 n Q 0.1 0 1 2 3 4 5 6 7 8 9 Timels 0 1 2 3 4 Ti me/s Fig. 15. Absorption signals for 30 ng of Ba vaporised from the wall of the pyrolytic graphite coated tube in Ar and N2. The signal in N2 was enlarged 3.3-fold Fig.14. Absorption signals for 10 ng of Ge vaporised from the wall of the uncoated tube at 2750 K in Ar and N2 still continue to interpret these and some other spectra recorded in graphite furnaces as due to monoxides. Our repeated attempts to obtain any of the well known44 monoxide spectra in the graphite furnace did not meet with success. (vii) BaC2 and SeC2 molecules have been observed in noticeable concentrations by Styris45>46 in a mass spectro- metric study of the products of vaporisation of the oxygen- containing Ba and Se salts from the graphite surface under vacuum. The temperatures at which the signals appear (1200 K for BaC245 and 40S500 K for SeC246) agree with the onset of the carbothermal reduction of BaO and Se02 calculated by the scheme in reference 47.General Conclusions Our analysis of the two approaches to the problem of atomisation of elements in graphite furnaces, which assume the formation of either oxygen- or carbon-containing com- pounds in the gas phase, shows that under typical analytical conditions only the second approach is capable of providing a satisfactory explanation of the incomplete atomisation of elements. The consideration of the effect of oxygen has substantiated the validity of the empirical scale for the free oxygen partial pressure, which is supported by all direct and indirect measurements. It is explained why free oxygen does not affect the atomisation of elements under typical con- ditions, and also why its effect on atomisation becomes The appearance of the carbon film results, in its turn, in a reduced rate of sample vaporisation. For this reason the higher concentration of the gaseous carbides observed in glassy carbon tubes compared with pyrolytic graphite coated furnaces (which is supported by a difference in the characteris- tic masses) is accompanied also by a reduced vaporisation rate of these elements.' We believe that this could explain also the reduced vaporisation rate of many elements in graphite tubes compared with Ta-lined furnaces,* and also the differences between the vaporisation rates of the same elements ob- served9 in graphite furnaces in Ar and N2.In nitrogen the formation of the carbon film occurs less intensely than in argon, other conditions being equal, because of a slightly lower concentration of gaseous carbon and, hence, of gaseous carbides.This effect is illustrated in Fig. 15 for Ba vaporisa- tion, for pyrolysis and atomisation temperatures of 1500 and 2900 K and a 1-s ramp time. (vi) The decisive role of gaseous carbides and, for a number of elements, of monocyanides is in good accord with our12J3 interpretation of the molecular spectra observed in graphite furnaces during the vaporisation of Al, Ba, Sr and Ca as originating from A12C2, AlCN, BaCN, SrCN and CaCN. A similar spectrum to that of A12C2 with a long-wavelength maximum at 255 nm is observed39 for GaC2, InC2 and TIC2, with long-wavelength maxima at 252, 272 and 269 nm, respe~tively.~~ Strange as it might seem, some workers4c43364 ANALYST, APRIL 1987, VOL. 112 noticeable under intense sheath gas flow-rate or when there is a high content of 0 2 in sheath gas, particularly at moderate temperatures (below 1500 K).As for the effect of carbon, we have for the first time revealed , experimentally studied and theoretically interpreted the major features observed in the appearance and distribu- tion over the furnace of gaseous carbon in enhanced compared with equilibrium concentrations. This provided a solid basis to substantiate the decisive role of gaseous carbon-containing compounds such as carbides and monocyanides in the atomisation of elements in graphite furnaces, and to explain numerious instances of the effect of furnace material, sample vaporisation techniques and sheath gas on the analytical signal. An interpretation has been found of the positive effect obtained from the use of platforms and pyrolytic graphite coated furnaces as means for suppressing many of these drawbacks and favouring the highest sensitivity.Some other aspects of the atomisation problem have not been touched upon in this paper, in particular, the major mechanisms of sample vaporisation (thermal dissociation4~50 and carbothermal reduction27-30 of oxides) where the above concepts prove to be of paramount importance. Also outside the scope of this paper are the applications of these concepts to the theory of some industrial high-temperature processes involving the carbothermal production of metals from 0res,~~>51 specifically of iron, and the catalytic graphitisation, oxidation and gasification of carbon-containing materials.38 However, the very list of the processes reflected on a microscale in the phenomena observed in graphite tubes leaves no doubt that GFAAS has presented to the researcher not only a perfect method of analysis but an elegant tool for exploring the world.The world we live in. Note Added on the Proofs The recent scanning electron microscopic studies by Welz et al.52 showed the formation of carbon hollow shells (blisters) after atomisation of 20-25 pg of La from the pyrolytic graphite platforms. This fact is in accordance with our explanation of the reduced vaporisation rate of many elements in graphite tubes. The author is grateful to Martin Mojica who carried out most of the experiments on gaseous carbon and to Leonid Polzik for fruitful comments on the manuscript.1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. References L’VOV, B. V., Inzh. Fiz. Zh., 1959, 2(2), 44; 1959, 2(11), 56. Slavin, W., and Manning, D. C., Anal. Chem., 1979, 51,261. Slavin, W., and Carnrick, G. R., Spectrochim. Acta, Part B, 1984,39, 271. L’vov, B. V., Nikolaev, V. G., Norman, E. A., Polzik, L. K., and Mojica, M., Spectrochim. Acta, Part B , 1986, 41, 1043. Ottaway, J. M., At. Spectrosc., 1982,3, 89. De Loos-Vollebregt, M. T. C., and de Galan, L., Spectrochim. Acta, Part B, 1984, 39, 449. Schlemmer, G., and Welz, B., Fresenius 2. Anal. Chern., 1986, 323, 703. L’vov, B. V., and Pelieva, L. A., Can. J. Spectrosc., 1978,23, 1. L’vov, B. V., and Mojica, M., paper presented at the 2nd Regina1 Conference of Analytical Chemistry, Krasnoyarsk, June 16-20, 1986.Hutton, R. C., Ottaway, J. M., Epstein, M. S . , and Rains, T. C., Analyst, 1977, 102, 658. Tsunoda, K., Fujiwara, K., and Fuwa, K., Anal. Chem., 1978, 50, 861. L’vov, B. V., in Kirkbright, G. F., Editor, “XXI Colloquium Spectroscopicum Internationale and 8th International Confer- ence on Atomic Spectroscopy, Keynote Lectures,” Heyden, London, 1979, p. 152. L’vov, B. V., and Ryabchuk, G. N., Zh. Prikl. Spektrosk., 1980,33, 1013. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38, 39. 40. 41. 42. 43, 44. 45. 46. 47. 48. 49. 50. 51. 52. Chakrabarti, C. L., Chang, S. B., and Roy, S . E., Spectrochim. Acta, Part B, 1983, 38, 447. Chang, S. B., Chakrabarti, C. L., Huston, T. J., and Byrne, J. P., Fresenius Z .Anal. Chem., 1985, 322, 567. Byrne, J. P., Chakrabarti, C. L., Chang, S. B., Tan, C. K., and Delgano, A. H., Fresenius Z. Anal. Chem., 1986,324, 448. Frech, W., Persson, J. A., and Cedergren, A., Prog. Anal. At. Spectrosc., 1980, 5 , 279. Cedergren, A., Frech, W., and Lundberg, E., Anal. Chern., 1984, 56, 1382. L’VOV, B. V., and Ryabchuk, G. N., Spectrochim. Acta, Part B, 1982, 37,673. Sturgeon, R. E., Siu, K. W. M., and Berman, S . S . , Spectrochim. Acta, Part B, 1984,39, 213. Sturgeon, R. E., Siu, K. W. M., Gardner, G . J., and Berman, S. S., Anal. Chem., 1986, 58, 42. Holcombe, J. A., Eklund, R. H., and Smith, J. E., Anal. Chem., 1979, 51, 1205. Kikoin, A. K., and Kikoin, I. K., “Molecular Physics” (in Russian), Nauka, Moscow, 1976, p. 137. Katskov, D. A., Kruglikova, L. P., L’vov, B. V., and Polzik, L. K., Zh. Prikl. Spektrosk., 1976,25, 918. Chupka, W. A., Berkowitz, J., Giese, C. F., and Inghram, M. G., J. Phys. Chem., 1958, 62, 611. L’vov, B. V., Zh. Anal. Khim., 1984,39, 1953. L’vov, B. V., and Savin, A. S . , Zh. Anal. Khim., 1982, 37, 2116. L’VOV, B. V., Dokl. Akad. Nauk SSSR, 1983, 271, 119. L’vov, B. V., and Savin, A. S . , Zh. Anal. Khim., 1983, 38, 1925. L’vov, B. V., and Savin, A. S., Zh. Anal. Khim., 1983, 38, 1933. L’vov, B. V., Novotny, I . , and Pelieva, L. A., Zh. Prikl. Spektrosk., 1980,32, 965. L’vov, B. V., and Yatsenko, L. F., Zh. Anal. Khim., 1985,40, 626. Gurvich, L. V., Khachkurusov, G. A., Medvedev, V. A., et al., “Thermodynamic Properties of Individual Substances” (in Russian), Nauka, Moscow, 1978-1982. Walker, P. L., Rusinko, F., and Austin, L. G., Adv. Catal., 1959, 11, 133. Sturgeon, R. E., personal communication. Dollimore, J., Freedman, C. M., and Harrison, B. H., Carbon, 1970, 8, 587. Krasnov, K. S., Editor, “Molecular Constants for Inorganic Compounds” (in Russian), Khimiya, Leningrad, 1977. L’vov, B. V., Dokl. Akad. Nauk SSSR, 1985,283, 1415. L’vov, B. V., Norman, E. A., and Polzik, L. K., Zh. Prikl. Spektrosk., 1987, in the press. Fuwa, K., Haraguchi, H., and l’sunoda, K., in Fuwa, K., Editor, “Recent Advances in Analytical Spectroscopy, Proceedings of the 9th International Conference on Atomic Spectroscopy,” Pergamon Press, Oxford, 1982, p. 119. Dittrich, K., Vorberg, B., Funk, J., and Beyer, V., Spectro- chim. Acta, Part B, 1984,39, 349. Dittrich, K., Spivakov, B. Ya., Shkinev, V. M., and Vorob’eva, G. A., Talanta, 1984,31,341. Sedykh, E. M., and Belyaev, Yu. I., Prog. Anal. At., Spectrosc., 1984, 7, 373. Pearse, R. W. B., and Gaydon, A. G . , “The Identification of Molecular Spectra,” Chapman and Hall, London, 1976. Styris, D. L., Anal. Chem., 1984, 56, 1070. Styris, D. L., Fresenim Z . Anal. Chem., 1986, 323, 710. L’VOV, B. V., I z v . Vuzov. Chern. Metallurgiya, 1986, NO. 1,4. L’vov, B. V., and Rybchuk, G. N., Zh. Anal. Khim., 1981,36, 2085. L’vov, B. V., and Fernandez, G. J. A., Zh. Anal. Khim., 1984, 39,221. L’vov, B. V., Ryabchuk, G. N., and Fernandez, G. J. A., Zh. Anal. Khim., 1984, 39, 1206. L’vov, B. V., and Yatsenko, L. F., Zzv. Vuzov. Chern. Metallurgiya, 1986, No. 5, 1. Welz, B., Curtius, A. J . , Schlemmer. G., Ortner, H. M., and Birzer, W., Spectrochim. Acta, Part B, 1986, 41, 1175. Paper A61381 Received July 14th, I986

ISSN:0003-2654    

DOI:10.1039/AN9871200355

出版商:RSC    

年代:1987

数据来源: RSC

摘要:

ANALYST, APRIL 1987, VOL. 112 365 Extreme Trace Analysis of the Elements-the State of the Art Today and Tomorrow* Plenary Lecture Gunther Tolg lnstitut fur Spektrochemie und angewandte Spektroskopie in Dortmund und Laboratorium of Reinststoffanalytik des Max-Planck-Institutes fur Metallforschung in Stuttgart, Bunsen-KirchhoK -Str. 11/13, 0-4600 Dortmund I , FRG The main aim of innovative elemental analysis is the improvement of analytical methods in terms of power of detection, reliability and economy. In routine analysis, there is a definite trend towards fully computerised, simultaneous or sequential multi-element methods that offer savings in both manpower and time. Extreme trace analysis, on the other hand, which involves ng g-1 and pg g-1 levels, usually cannot be achieved by routine cost-saving methods owing to the inadequate power of detection and the growing uncertainty of the results as lower and lower concentration levels are reached.The costs of erroneous decisions based on unreliable or wrong analytical results are often greater than the savings in manpower costs achieved by employment of direct instrumental methods. It is therefore the responsibility of the analyst to find the right strategy, taking into consideration the economic viewpoints in each instance and realising that expert knowledge, experience and critical capability are often to be valued more highly than the employment of the newest types of apparatus that are currently so fashionable. Improvements in both detectability and reliability in the determination of very low concentrations are usually achieved by first carrying out basic research, often involving contributions from techniques based on a wide range of principles, particularly with regard to more sensitive excitation sources with lower inter-element interferences.After a brief systematic discussion of the most important principles of atomic spectrometric methods-those based on the absorption, emission and fluorescence of electromagnetic radiation and also on emitted ions and atoms-a critical comparison is given of new developments in determination methods, e.g., atomic absorption spectrometry, optical emission spectrometry, coherent forward scattering, laser-enhanced ionisation spectrometry, atomic fluorescence spectrometry, X-ray fluorescence analysis and mass spectrometry.Keywords: Trace analysis; elemental analysis A report on the status quo and trends in the field of extreme trace analysis of the elements can only be presented here in a drastically shortened form with generalisation of the extensive subject matter. What this report can do is to refer to basic developments in some very powerful methods as far as they affect the determination of very low absolute amounts of elements, which could be of prime importance for analytical work in forthcoming years. There are two main kinds of motivation for lowering the limits of detection: those related to the characterisation of technological high-purity materials and those related to the determination of trace elements at their natural concentra- tions in biotic and environmental materials, so as to gain more knowledge about their complex biochemistry and the levels at which they are still essential or becoming toxic.Progress in methods of bulk analysis is one goal, but of much greater interest is the analysis of the micro-distribution of the elements, also at trace levels. This important new area of microtrace analysis constitutes the main motive for developing more sensitive methods of elemental analysis. Most scientists involved these days in the development of more powerful methods of determination are highly special- ised in only one analytical technique and may thus easily lose the broader perspective of the multitude of analytical approaches and the limitations of the many other methods of concurrence-which can hardly be grasped because there are so many-but which are nevertheless all essential for optimum solutions of the many sophisticated analytical problems of today and tomorrow.* Presented at SAC 86, the 7th SAC International Conference on Analytical Chemistry, Bristol, UK, 20-26 July, 1986. One of the most important rules in practical extreme trace analysis states, “one method is no method,” and therefore only a multi-method concept can avoid the danger of developing a new method that does not meet the requirements of the actual problem. Further, one must not forget that in practical extreme trace analysis, not only the instrumentation for the determination of the elements but also sample pre-treatment and chemical aspects1 play important roles in achieving the progress briefly commented upon above.On the other hand, the problem-orientated user of analytical methods with today’s state of instrumentation is unlikely to be capable of substantially improving any analytical method as far as the instrumentation is concerned. Therefore, the common aim will only be achieved by close partnership of those developing the methods and those producing the apparatus on the one hand, and the analytical strategists or the users of highly specialised analytical instruments on the other. With the support of many colleagues who are working in both of these directions, I have been encouraged to attempt this difficult inter-method comparison in an independent and critical way. The innovative analytical research we have to consider in this context is directed, above all, towards three ultimate aims: the improvement of the analytical power of detection, the improvement of the reliability of the results gained and the improvement of the economy of the methods employed.These three essential criteria, subjected to continuous evolu- tion, are very closely correlated. They also depend on the question of whether they are discussed in terms of routine analysis or from the point of view of still unsolved problems in extreme trace analysis. Indeed, one should be aware that each improvement in the power of detection in trace analysis re-imposes the question of the reliability of the results in each instance. In turn, dependence on the reliability is then an366 1 Cs I B a l ANALYST, APRIL 1987, VOL.112 Hf Ta W Re 0 s lr Pt I Au pyBF 16Pb l8Eh 1 0 6 0 6.D B E B E economic aspect. The relative importance of the economic aspects is very different in routine analysis at the microgram level using well established procedures to that in the far from routine field of trace analysis at nanogram and picogram levels. In routine analysis, one aims at a high and cost- effective sample throughput with commercially available instruments, whereas in extreme trace analysis the reliability of results is much more the cost-determining factor. Indeed, if we take into consideration the possible consequences of basing decisions on unsatisfactory or even wrong results, we find that the real cost of “cheap” analyses could well be much greater than that of the highly specialised and labour-intensive analytical techniques.When methods are compared in practical extreme trace analysis, the power of detection and the reliability are the most important criteria to be taken into account. Both, however, depend on the element to be determined, on the matrix, on the boundary concentrations of concomitant components in the system and on many other factors. These factors, therefore, hardly permit generalised statements. Methods can only be compared objectively with one another for a very specific analytical problem, based on the experience and sound critical attitude of the analyst. Whilst being aware of these major problems, one should nevertheless establish, at least in an over-all sense, the power of detection, the susceptibility to interferences and also the economic aspects of the methods with respect to the determi- nation of individual elements or elemental groups, in order to know the trends for quality criteria of methods-but under no circumstances more than that.One difficulty in making such comparisons arises because different workers may be using different techniques based on similar analytical principles so that their judgements can hardly be compared. Moreover, they frequently make rather idealised statements in order to propagate the methods they employ. For instance, different excitation sources employed in the same analytical principle, such as inductively coupled (ICP) and d.c. plasmas (DCP) in optical emission spec- trometry (OES), can only really be compared objectively in terms of power of detection and reliability when measure- ments are made with identical samples and spectrometers.Such circumstances are, however, rarely found in comparisons of methods in the literature. There is another very important point that should be emphasised in practical extreme trace analysis. The physical conditions that lead to a high power of detection and analytical reliability, e.g., the lowest possible spectral interferences, La Ce Pr Nd Srn Eu Gd Tb Dy Ho Er Tm Yb * A A A A A A A A A eZa FAAS 32 elements ET-AAS 53 elements = ICP-OES 59 elements DCP-OES 29 elements AFS 31 elements Lu Fig. 1. Elements that can be determined at hi her concentrations by flame A A S (FAAS), electrothermal AAS &T-AAS), ICP-OES, DCP-OES and atomic fluorescence spectrometry (AFS) .Compded from Parsons et a1.2 LD C 10 ng ml-1 Th u - - good signal to noise ratios and high quantum yields, often go together with more important chemical aspects such as low contamination risks, defined chemical counter actions during the signal excitation step and error-free calibration pro- cedures. Indeed, these individual analytical steps lead, if not yet well established, to systematic errors that are very difficult to detect and these finally determine the accuracy of the analytical procedure. This must be seen before the back- ground of another very important rule in extreme trace analysis, namely that the best power of detection and the highest reliability can only be attained when an element is determined in its isolated form enriched on a target as small as possible.In the following systematic evaluation, power of detection and reliability are used as criteria. Economic aspects can only be included in a very general manner. The power of detection differs considerably not only from one method to another, but also from one element to another. This can be shown impressively in a comparison of limits of detection given by Parsons.2 For concentrations above 10 ng ml-1 in Fig. 1, any of the optical methods listed can be used. However, if the limit is set at 0.01 ng ml-1, as shown in Fig. 2, only graphite furnace AAS and, for some elements, namely those with thermally stable oxides, ICP-OES are still relevant. Now let us continue by considering the different principles that are made use of in atomic spectroscopy and their excitation sources (Table 1).This scheme shows many similarities and cross-connections with regard to the excitation sources, which can be very helpful in this survey. Other principles such as radiochemical or electrochemical methods have not been considered here, but they are indispensable in many instances and sometimes cannot be bettered by other methods in extreme elemental analysis. In a selective manner, and including mainly our own activities (for a more detailed review see reference 3), I shall now attempt to show which innovations have been introduced in the different methods, starting with AAS. Atomic Absorption Spectrometry Flame AAS is inferior to furnace AAS in terms of power of detection by 2-3 orders of magnitude.4 On the other hand, one finds less cross-interferences and related systematic errors.Therefore, flame AAS methods should always be given preference to furnace AAS methods if the concentra- tions to be determined are high enough. Over the past few [zzd FAAS 2 elements 0 ET-AAS 21 elements = ICP-OES 5 elements DCP-OES 0 elements AFS 3 elements . . . . . . . . . . . . . . . . . . . .‘ . . . . . . . .ANALYST, APRIL 1987, VOL. 112 367 years, the absolute power of detection of flame AAS has been increased by reducing the sample consumption, which was usually 1-2 ml. This can be done as follows: 1. By the injection technique, introduced by Berndt and Jackwerths; the absolute power of detection is improved by a factor of 10-20. 2. By sample vaporisation on a heated wire loop made of, e.g., iridium or tungsten (Fig.3) as shown by Berndt and Messerschmidt ,6 especially for microsamples, which in many instances improves the power of detection together with other advantages that cannot be referred to in detail here. Additional advantages can be achieved by using the “slotted- tube atom trap” (collection tube) te~hniques.7~8 3. It is also possible to carry out a direct analysis of solid-state samples, especially when relatively readily volatile elements (e.g., Cd, Zn, Pb and Tl) are to be determined. With biotic matrices, the microsample is decomposed in a separate decomposition device and subsequently the analyte is evapor- ated and fed in the form of an aerosol directly into the flame gases (Fig. 4). With this technique, a wet decomposition step can be circumvented and the analyte losses that usually occur during solution evaporation can be avoided.9 5 cm burner / / /- Fig.3. Loop technique in FAA9 When L’vov introduced electrothermal atomisation in a graphite furnace and the late Dr. Massmann of this Institute was able to improve the power of detection still further by introducing the micro-furnace technique, AAS became the most sensitive technique for the direct and indirect determina- tion of many elements, except carbide-forming elements and several non-metals, e.g., C, N, 0, P, S and the halogens. The excellent power of detection, going far down into the picogram region, is surpassed only for a few elements by other methods, such as activation analysis, inverse voltammetry and chelate GC methods. This, together with the relatively low cost, has promoted graphite furnace AAS (GFAAS) to the method of choice in extreme trace analysis.Soon after the initial, extremely euphoric period, one began to discover a series of grave systematic errors, due mainly to the very complex chemical reactions occurring during the periods of evaporation, ashing and atomisation, and to radiation absorption processes within the furnace. Significant progress in the reduction of systematic errors has been realised by improving the graphite furnace surfaces, e.g., by coating with pyro-graphite or metal oxides, glassy carbon tubes or by controlled and constant temperature conditions during atomi- sation by, e.g., platform techniques, the tube-in-tube tech- nique and, which seems to be very promising, by twin-furnace techniques (Fig.5 ) . Partial success was also achieved by the empirical use of matrix modifiers. The most important progress in analytical Burner Sucking unit Fig. 4. Solid sample pre-burning technique in FAAS9 Table 1. A selection of atomic spectroscopic methods in elemental trace analysis. CFS, coherent forward scattering; GDL, glow discharge lamp; HC, hollow-cathode; HG, hydride generation; HRRS, high repetition rate spark; LRRS, low repetition rate spark; LM, laser micro; SN, sputter neutral; TR, totally reflecting; TI, thermionisation; and TOF, time-of-flight Technique of excitation ~~ ET Analytical . . furnace, H.f./u.h.f. Sample method Flame wire-loop Arc Spark plasma GDL,HC Laser X-ray Electrons Ions form* AAS . . . . + ZAAS .. . . HG-AAS . . OES . . . . + HG-OES . . AFS . . . . + CFS . . . . + XRE . . . . XRF . . . . Optogalvanic . . + MS . . . . + SN-MS . . . . TOF-MS . . + + + + + + DCP + + TI + ICP + ICP HRRS CMP LRRS MIP + ICP MIP + ICPMIP + + + GDL LM HC GDL/HC + exc. + exc. + GDL + GDL + + &l 1, s Volatile g, 1, s 1, s Volatile 1, s hydrides hydrides 1 WD/ED EMPt PIXE 1,s TR 1, s + 1, s + S SIMS/FAB g, 1, s S * g = Gaseous; 1 = liquid; s = solid. t EMP = Electron microprobe.368 1P > > - - ANALYST, APRIL 1987, VOL. 112 1 reliability, however, was achived by physical means of coping with spectral interferences and spectral background. Apart from background correction with the aid of the deuterium discharge lamp and the pulsed hollow-cathode lamps according to Smith and Hieftje,lo the use of line splitting in the magnetic field due to the Zeeman effect (Fig.6)11J2 is the most effective method. The magnetic field may be applied to the primary radiation source or to the sample atom source. The first variation ( b ) , brings certain advantages with respect to solid-state sample analysis, but requires the use of special discharge lamps. Therefore, the second variation permits, especially with alternating magnetic fields ( c ) , high sensitivity when using conventional hollow-cathode lamps. Owing to the improved signal to noise ratio, the power of detection may under certain conditions be improved by up to one order of magnitude compared with the compensation by a deuterium lamp. However, examples of considerable over- compensation that could lead to serious systematic errors have become known.13 By the introduction of Zeeman compensation, the direct furnace AAS for solid-state samples has become very attrac- tive and has been satisfactorily confirmed for a series of -.I A - - - 7 3 Fig. 5. Isothermal atomisation techniques in GFAAS. (a) Graphite tube with L’vov platform and ( b constant temperature graphite furnace (Lundber g/Fr ec WLindber g ] elements with unproblematic chemical behaviour during atomisation (e.g., for Cd, Zn, Cu and Sb and, with reserva- tions, for T1 and Pb also) in biological matrices.14-16 Direct solid sampling in AAS has several advantages. Chemical sample preparation involving the risk of contamina- tion is avoided and the sample throughput can be extensively increased.The only time-consuming steps are those of sample homogenisation and the weighing of the sample for analysis. As with the present commercially available instruments, the sample amount, however, is restricted from the upper microgram to the lower milligram region; therefore, the power of detection is rather low. The small amounts of sample used, on the one hand, require sample homogenisation for particle diameters of a100 pm, in order to facilitate represen- tative analysis of larger samples.17 On the other hand, they make possible micro-distribution analytical investigations (e.g., for the medical examination of histological tissue specimens). A further possibility for the direct analysis of solids (e.g., glasses, biological samples) is offered by laser techniques18 similar to laser micro-OES, but only for single-element investigations.Laser micro- AAS is most useful for fundamen- tal investigations of laser shots as used in OES and mass spectrometry (MS). It is a simple means of studying systemat- ically sample vaporisation, aerosol transport and atomisation. In addition to the cold vapour technique as used for the determination of Hg, hydride generation AAS (HG-AAS) represents a special combination of chemical separation and pre-concentration with AAS determination. l 9 It results in an improved power of detection and accuracy for all those elements which can be reduced quantitatively to volatile hydrides (e.g., As, Se, Sb and Sn). The determination of selenium may serve as an example to demonstrate the efficiency of this technique.The Se hydride produced is frozen out within an absorber tube after drying and subsequently swept into the measuring cell of the AA spectrometer by heating (Fig. 7). With this technique, Se may be determined down to the 1 pg region with excellent Furnace in a.c.-fed Static polariser Monochromator (c) Pulsed hollow-cathode magnetic field and filter and signal & processing 1 c7 or electrodeless discharge lamp - - A Fig. 6. Main principles of Zeeman effect AASANALYST, APRIL 1987, VOL. 112 369 reliability. For instance, using this technique the Se content in segments of a single hair can be reliably determined with a relative standard deviation of S5% .20 However , this particular success should not give the impres- sion that AAS has already reached the stage of a well established method for extreme trace analysis. The systematic errors may be considerable; they are due not only to instrumental hazards, but also to chemical changes occurring during signal generation. The latter can only be lowered by complementary techniques: by separation of the vaporisation and atomisation steps with the aid of twin-oven techniques,21 by detailed systematic investigations of time-resolved spectra permitting the differentiation of signal and background peaks22.23 or investigations with radiotracers, which provide information on losses during sample preparation and vapori- sation steps in the furnace (elemental volatilisation)24 or, for instance, during hydride formation .20 v1 Furnace position B: desorption 100 "C 2' Fig.7. HGC-AAS device for the determination of Se in the pg ml- range.20 1, Quartz vessel; 2, reduction solution; 3, cold trap (-70 "C); 4, quartz adsorption tube; 5 , heating system; 6, adsorption region, Chromosorb W, 30-60 mesh; 7, liquid N2 cooled AP block; 8, insulation; 9, quartz cuvette; 10, furnace; 11, EDL; 12, AA spectrometer; and 13, recorder Direct current plasma 2 x 7 A (d.c.1 70-80 V Argon Liquids - suspensions - hydrides - (solids) (m icrosa m ples) Detection limits: < 100 ng ml-1 Moderate matrix effects by alkali metals With echelle system: good for line-rich elemental matrices Moderate costs Microwave induced plasma 2.45 GHz 40-200 W Argon - helium Dry aerosols - hydrides (liquids), microsamples Absolute detection limits: ng-pg High matrix effects by alkali metals Optical Emission Spectrometry As the oldest multi-element atomic spectrometric method, optical emission spectrometry (OES), which originally used flame, electric arc or spark excitation, has earned its well established position.As the power of detection for numerous elements is relatively good, it was the classical spectroscopic method of choice until AAS became available. Even today, it is still indispensable , particularly for fast, roughly qualitative overview analyses in the entire 1-18 8-1 region. Especially in the field of metal analysis, this technique has been developed further into a fast , precise, simultaneous multi-component analytical procedure for process control, but has met strong competition from XRF for elements with atomic numbers 310.25 Indeed, the latter technique has the advantage of spectra with few lines, and it is simpler and faster.This classical period of OES has been excellently reviewed by Laqua,26 and need not be discussed further here. Therefore, we shall move on to excitation with electric plasmas, as preferentially applied in multi-element solution spectrometry. These sources differ from the classical excitation sources and also from XRF in the high linear dynamic range and therefore require relatively simple calibration. The various plasma sources available are shown in Fig. 83; the ICP and DCP, because of their universal applicability and the relative freedom from interferences, find especially wide application. However, this does not mean that the other sources, such as the capacitively coupled microwave (CMP) or the microwave-induced plasmas (MIP) are unimportant. This can be illustrated by a few examples.Much experience has been gathered in the use of the ICP. With pneumatic nebulisation, the power of detection lies somewhere between flame and furnace AAS.4 Especially for elements with thermally stable oxides (e.g., lanthanides), the detection limits are low, in contrast to elements with a high ionisation and excitation potential for the most sensitive analytical lines. Accordingly, Cd, T1 and Pb can be deter- mined by AAS with a considerably higher power of detection. The power of detection of ICP-OES and of DCP-OES techniques is almost identical and is similar to those of spark, Inductively coupled plasma Capacitatively coupled microwave plasma 5-50 MHz Argon - nitrogen-(air) Liquids - suspensions hydrides - (solids) mi crosa m p les Detection I i m its: < 100 ng ml-l Low matrix effects by alkali metals 1-10 kW 2.45 GHz 0.4-1 kW Argon - nitrogen Liquids (low salt contents) - (solids) - hydrides Detection limits: (1 pg-ml-l High matrix effects by alkali metals Low costs (Coupling with echelle monochromator) High costs Low costs Fig.8. Plasma sources for OES370 ANALYST, APRIL 1987, VOL. 112 GD-OES and conventional XRF. According to empirical results obtained, alkali metal effects are greater in the DCP technique than in the ICP technique.3 An enhancement of the power of detection by one order of magnitude could be realised in both instances by ultrasonic nebulisation, which, however, has so far not been brought into general routine use.A reduction in the sample size to 50 p1 can be achieved by the injection technique, which is only a special form, in fact, of flow injection.3 A suitable coupling of electrothermal atomisation (ETA) and OES results in an improvement in the detection power of about one order of magnitude. Other techniques, e.g., hydride ICP-OES, have also been developed. For the analysis of solid-state material, ETA and also spark ablation in combination with ICP-OES are useful.3 None of these specialised techniques (Fig. 9), however, has yet outgrown its development phase. Presumably, the power of detection will not surpass that of the solution techniques, but the matrix effects are considerably lower.Conventional microwave-induced plasma (MIP) techniques generally do not reach the power of detection of the ICP or DCP and the interferences are higher because of the lower gas temperatures. Here, further developments of one-filament plasmas and of the resonator introduced by Beenakker27 have led to multi-filament and toroidal cylinder plasmas with superior discharge stability. With these discharges, the spec- trometric analysis of wet aerosols can be carried out at a microwave power of <200 W.28 In contrast to ICP excitation, it is a micro-method with a low solution throughput and accordingly low Ar or He consumption. In addition, micro- wave generators are three to four times cheaper than ICP generators. Because of the excellent absolute power of detection of the MIP-OES technique, its combination with a sample vaporisa- - Pneumatic nebulisation Conc'entric Babinnton Fritted-disc Cross-flow - Ultrasonic nebulisation microchannel plate - Electrothermal evaporation Graphite furnace Carbon Cup Metal filament - Hydride generation rod ==k- - Direct sampling of compact solids Electroerosion Laser ablation -Direct sample insertion device k d- 11 Fig.9. Sample introduction systems for ICP-OES tion technique (ETA-MIP-OES) (Fig. 10) preceded by chemical element pre-concentration is of particular interest.29.30 The absolute limits of detection for Hg are less than 1 pg when a suitable combination of decomposition, separation and determination methods is used.31 The applica- tion of the MIP-OES technique as a simultaneous, element- specific detector in GC and HPLC procedures is even more interesting.After HPLC separation, we were able to detect different organomercury compounds via the Hg signal with a high power of detection compared with conventional UV photometry.32 In addition to the plasma sources some discharges under reduced pressure became of particular interest .3 The glow discharge lamp (GDL), as reviewed chiefly by Laqua and Ko,33 is a very stable and reliable excitation source for bulk analyses of electrically conducting solid-state samples and can also be used for in-depth profiling in surface layers of electrically conducting samples. The erosion rates of the GDL are up to 50 nm s-1, which is much higher than in ion sputtering techniques combined with SIMS or AES.Accord- ingly, the method finds many practical applications, e.g., in the control of steel coatings.34 GDL excitation also offers extremely interesting possibili- ties for further developments in OES when a furnace for evaporation is combined with a glow discharge for excitation, as shown by Falk et aZ.35 (Fig. 11). Indeed, here separate optimisation of the volatilisation and excitation conditions first became possible, in contrast to earlier work by Littlejohn and Ottaway36 on graphite furnace OES (GFOES). In conventional GFOES, the detection limits are 2-3 orders of magnitude poorer than in GFAAS, owing to the background continuum-despite improvements resulting from platform techniques, wavelength modulation and metal tube furnaces in place of graphite furnaces. It is significant that all the conventional excitation sources in OES, such as arc, spark, chemical flames or ICP (or DCP), operate under local thermodynamic equilibrium conditions and do not permit the separate optimisation of each analytical step, i.e., vaporisation, atomisation and excitation during To Cooling Microtron 200 \I Micropipette HGA 74 IT 0.9 m Czerny-Turner monochromator Fig. 10. and 3, tuning Graphite furnace MIP-OES. 1, Plasma gas; 2, carrier gas; Anode voltage Tube atomiser = cathode ,Negative glow -Sample Anode - \\\ .\\\\\\\\\\\\\ ' . \\ -1 Fig. 11. Principle of the FANES source from Falk et ~ 1 . 3 ~ANALYST, APRIL 1987, VOL. 112 371 signal production. However, the low-pressure glow discharge is an excitation source operating in non-thermal equilibrium.In the negative glow, gas ions form and are accelerated on to the cathode. When the sample is taken as the cathode, it is sputtered by the ionic bombardment and atomised. The excitation of the atomic vapour occurs in the spatially separated negative glow , where the electron temperature is considerably higher, but where the gas temperature and the degree of ionisation are considerably lower than in the usual plasmas. Accordingly, a considerably lower background intensity and a better power of detection are obtained. During GF-OES procedures with furnace temperatures below 3000 K, the degree of ionisation, but also the degree of excitation, are low. Now with furnace atomic non-thermal excitation spectrometry (FANES) the power of detection is comparable to that of GF-AAS and it includes the possibility of use in multi-element analysis with OES.The question of whether this very promising combination will become important for applied analysis cannot yet be answered-only time will tell. Coherent Forward Scattering In coherent forward scattering (CFS), also termed stimulated emission of resonance radiation, a primary and a secondary radiation source serving as an atom reservoir are arranged in sequence along one optical axis (Fig. l2).37 As a result of magneto-optical effects (Faraday and Voigt effects) , the stimulated emission of resonance radiation is polarised circularly or linearly and can be observed. This is accom- plished by placing the secondary source between crossed polarisers (polarisation prisms).The CFS spectrum has only a few resonance lines of the element to be determined. For sequential determinations, a hollow-cathode lamp, which radiates the light of the element to be determined, is suitable as a primary radiation source. For simultaneous determina- tions, a xenon arc source can be used. For a sequential multi-element system, tunable laser beam radiation can be used as well. Atomic vaporisation in the analytical excitation source may be produced in flames, furnaces and also in an ICP. The advantages of CFS lie in the simple spectra. The loss of intensity due to polarisation, however, is a disadvantage. As the power of detection up to now has been restricted to the ng ml-1 region, and as the set-up is rather complex, this technique is so far not a strong competitor in extreme trace analysis.Laser-enhanced Ionisation Doppler-free laser-enhanced ionisation spectroscopy (LEI) is a spectroscopic method from which an extremely good power of detection can be expected. Resonance radiation from a tunable laser can ionise elements within an atom reservoir very selectively. The ions can be detected by sequential optogalvanic means or with the aid of a thermionic diode, the latter being more sensitive by an order of magnitude or more, as Niemax and co-workers at this Institute were able to The principle of a thermionic diode (Fig. 13) is simple. A directly heated tungsten cathode, surrounded by a cylindrical anode, is positioned in the atom reservoir evacuated to about 100 mTorr. Ions or molecular ions produced by photoionisa- show .383 Scatterina L tor Polariser B-' H, ' 'I' Analyser Fig.12. Principle of coherent forward scattering spectroscopy. H , , Voigt configuration; Hil, Faraday configuration tion having a lifetime of about 1-10 ms arrive in the space charge surrounding the cathode and recombine with the electrons. The presence of the diffused ions in the space charge trap lowers the potential threshold for the electrons and increases the diode current. The detector responds to all ions in the system, and is thus element non-specific. On the production of atomic vapour within the detector by, e.g., heating up of the detector (detector functioning as a furnace) and sending monochromatic laser beams of suitable wavelengths through the detector (Fig. 14), only those atoms of which the absorption wavelengths correspond to those of the laser beams will be excited.The excited atoms are secondarily excited by impact with atoms and ionise. In this manner, it is possible to detect up to 103-104 atoms cm-3 in the presence of 1013 atoms cm-3 within the detector. Owing to the Doppler-free nature of the method, the isotopes of an element produce completely distinguishable signals, so that calibration by isotope dilution, as in MS, can be carried out. In principle, all elements that can be evaporated during the heating cycle of the detector can be determined. As the detector temperature is restricted owing to thermal stability of the equipment material (about 1000 K), only elements of relatively high volatility (e.g., Hg, Cd, Zn, T1, Pb, Bi, alkali and alkaline earth metals, in addition to gases) produce sufficient vapour pressures.At present, the most serious disadvantage in the practical utilisation of the method is the very high price of the gas laser, which, however, may be substituted soon by far less expensive semiconductor lasers (laser diodes). Compared with laser spectrometry, the OES techniques, which use a laser for the vaporisation of samples while exciting an atomic vapour in a spark, a high-frequency plasma or a glow discharge (Fig. 15), find their main applications in micro- analysis. 18 They suffer from relatively poor spatial resolution, namely about 210 pm, compared with SIMS, PIXE and other probe methods. Perhaps the laser vaporisation in micro-OES will experience a revival in combination with LEI.BignaiL& ................... ++ t Buffer gas Fig. 13. (a), Thermionic diode principle and (b), thermionic heat-pipe diode. A, B, sample boat; C, W filament cathode; D, shielding grid; E, water cooling I Tunable laser c- Recorder element introduction Tunable laser Fig. 14. Principle for isotope-selective Doppler-free laser-enhanced ionisation spectrometry with the thermionic diode. After Niemax et a1.39372 ANALYST, APRIL 1987, VOL. 112 Atomic Fluorescence Spectrometry The application of tunable lasers has also revived AFS in combination with widely differing atomisation techniques (e.g. , flame , ICP and furnace techniques) .3 In laser-induced fluorescence (LIF) , the furnace technique has the best chance of improving the detection limit to the pg ml-1 region, but it is more susceptible to interferences than the ICP technique.At present, the power of detection of ICP-LIF for elements such as Se, As and Cd, which require a high excitation potential, is even better than in ICP-OES.40 X-ray Spectrometry X-ray fluorescence spectrometry (XRF) calls for little com- ment, as both conventional wavelength and energy-dispersive XRF are well developed. Their absolute power of detection is highest for elements with an atomic number around 30. Here, the detection limits are about 0.1 pg and they become poorer 1. Direct observation I FLaser beam Spectral apparatus - Sample ' I I Auxiliary 2. Cross spark I electrodes j + - l 3. H.f. discharges (a) Radio- frequency R.f. coil 4 1 'Quartz tube (b) Microwave Cavity 4.Inductively A coupled plasma :*-3 II Sample ablation chamber Fig. 15. OES with laser atomisers for higher and lower atomic numbers. The main limiting factor for the power of detection is the Compton background. Further, elemental cross-interferences make it necessary to calibrate with standards that must be very similar in composi- tion to the sample. The limitation of the power of detection due to the Compton background may be avoided in the following three ways. 1. The total reflection XRF (TXRF) technique (Fig. 16), introduced by Wobrauschek and Aiginger41 and methodolog- ically improved by Schwenke and Knoth,42 can be used. The angle of incidence of the exciting beam is so small that the latter is totally reflected at the target surface.If the target surface has very little surface roughness (as is possible, e.g., with quartz) and when it is homogeneously covered with a very thin film of sample material, TXRF results in an improvement in the power of detection by a factor of up to 1000 in comparison with conventional energy-dispersive techniques.43 A serious problem lies in optimum sample preparation on the micro-scale, a problem that we are working on intensively in this Institute. Difficulties start with cleaning of the quartz-type sample carriers and homogeneous spread- ing of the sample (e.g., by condensation of solutions, cathodic sputtering or electrolytic deposition). Thin histological tissue specimens can be directly examined, as we showed recentlye44 Another thin-film sample technique with detection power in the picogram range was recently de~cribed.~5 2.With respect to the power of detection, TXRF is superior to the proton-induced X-ray emission (PIXE) technique46 by one or two orders of magnitude; PIXE, however, has a higher absolute power of detection than conventional XRF (Fig. 17). The advantage of the PIXE technique lies in the fact that the proton beam can be focused to about 1 pm in diameter, so that multi-element microdistribution analyses can be carried out. For instance, elemental determinations at a concentration level of about 1 pg g-1 in the diameter of a human hair are possible (Fig. 18). 3. Synchrotron radiation (SYXRF) can also be employed as an excitation s0urce.49~50 It can be focused to 1-10 pm in Si - Li detector Reflection unit X-ray tube L.-.L Fig. 16. Principle of TXRF r- f 102 4- 8 in-i \ (E0 = 16.5 keV) I I I 40 I " 10 20 30 Atomic number Fig. 17. Extrapolated limits of detection for conventional energy dispersive XRF, PIXE and synchrotron-XRF (SYXRF). After Bos et al.47ANALYST, APRIL 1987, VOL. 112 373 diameter and still possesses sufficient intensity to penetrate through and excite thin samples, as was demonstrated by Kelelsen et aZ.49 recently. The expected powers of detection lie at approximately 100 pg. Better detection powers can be expected under conditions of a totally reflecting excitation beam.51 These three excitation sources open up useful routes in the direction of multi-element determination, high detection power and very simple evaluation modes. Mass Spectrometry Elemental mass spectrometry, based on its principle, takes up a special position when compared with the methods discussed above, as follows.1. It is the only universal method for multi-element determinations that permits a sequential or simultaneous determination of all elements and their isotopes in solid-state materials and solutions. 2. For all elements, limits of detection below 1 pg g-1 are easily obtained. 3. The conditions for an “absolute procedure” are particu- larly favourable as, in contrast to other spectroscopic methods, the atomic components of an analyte are detected directly and not via the lengthy means of emitted light quanta. 4. The spectra have relatively few lines. The resolution is high enough to separate the lines of all isotopes.Accordingly, a very precise quantitative evaluation is possible by isotope dilution analysis (except for single-isotope elements). Until now, the discrepancy between the methodological capabilities and their realisation was disappointingly great, because the conventional double-focusing sector field instru- ments with arc and spark sources and ion-sensitive photo-plate detectors were very expensive, slow with respect to recording of spectra, time consuming and too detailed in the photo- metric evaluation of spectra. Moreover, the precision and accuracy were unsatisfactory. The use of conventional solid- state source mass spectrometry was therefore restricted to high-purity materials (metals, semiconductors) where very low contents of impurities in relatively simple and constant matrices have to be determined.In recent years, a very promising change in this bad image of the elemental mass spectrometry has taken place. This change can be attributed mainly to two rapid developments: 1. In the field of quadrupole MS, more efficient and inexpensive instruments have steadily invaded the market. They have replaced the expensive sector field instruments, at least for routine analysis. The financial investment required Zn distributions Se distributions Hair E 2 300 $ 200 g 100 .- .I- 0, G O 0” 40 80 120 40 80 120” Positionium Fig. 18. of hair segments by PIXE from Bos ef af.48 Microdistribution analysis of Zn and Se across the diameter now lies at the same level as for powerful spectrometers for OES or XRF. 2. The new excitation sources such as h.f.plasmas (ICP, MIP), glow discharges and lasers will replace the conventional spark techniques, which have serious systematic errors. Vast experience in OES is transferrable to MS and permits the expectation that rapid progress can be made in terms of reliability and economic aspects. The older thermionic source, which has an absolute power of detection in the picogram range, has been utilised in analysis very impressively in Germany.52 It can be predicted that ICP-MS3 (Fig. 19) will lead to new lines of approach to the routine analysis of solutions. By using the isotopic dilution technique, the calibration is improved in both instances. In ICP-MS, the power of detection for most elements is at the level of 0.1-1 ng ml-1, which is about one order of magnitude better than in ICP-OES.The limits of detection of the latter method, moreover, differ over a range of nearly five orders of magnitude for the specific elements, which is not the case in ICP-MS. Further, in ICP-MS, the term “multi-element analysis” may be applied in a truer sense than in ICP-OES. These very optimistic perspectives concerning ICP-MS will have to be confirmed, however, in routine applications that are just about to start. For the direct analysis of solid-state materials, the glow discharge (GD-MS) and the sputter neutral (SN-MS) tech- niques promise considerable progress. With GD-MS (Fig, 20)54 we have been able to gather very positive experience recently at this Institute. We have combined quadrupole MS with glow discharge techniques, which have been further developed as excitation sources for bulk and depth profile analysis with OES by Laqua and co-workers in this Institute with great success.By avoiding Differentially pumped region . . . Quadrupole n Skimmer Nebuliser ICP\ I I /I / CoTputer w / \ \ R.f. Mecianical Ciyo Quad supply supply pump system Fig. 19. Principle of ICP-MS Quadrupole Ion masslfilter detyctor Sample \ Pump 1 Pump 2 Quadrupole Computer supply Fig. 20. Principle of GD-MS nass374 ANALYST, APRIL 1987, VOL. 112 molecular ions and clusters, one arrives at the stage shown by the spectrum obtained with a copper sample (Fig. 21). Excellent dynamics and a very low background almost free from molecular ions and clusters can be obtained and, at the first attempt, evaluatable peaks for main and minor com- ponents.We believe that we shall be able to achieve a power of detection for nearly all elements in the range of 0.1 pg g-l and below, with a total recording time of about 3 min. SN-MS as developed by Oechsner and co-workers5~58 also proved to be a very promising technique for problem-orien- tated applications. In each sputtering process with ions, there are neutral particles that also are sputtered with good yields. They are ionised in a high-frequency discharge and isolated and detected in a quadrupole mass spectrometer. According to the mode of excitation applied, either in one step with an h.f. plasma (Fig. 22), or by sputtering of the sample with an ion beam and subsequent ionisation of the emitted neutral particles in an h.f.plasma (Fig. 23), bulk or depth profile analyses with depth resolutions to approximately 1 nm can be carded out preferentially. The power of detection is uniform for all elements and is in the upper ng g-1 region. There are hardly any cross-interferences. Because all of the above-mentioned new excitation modes in MS are still at the development stage, any comparative evaluations would be premature. The same is also true for LAMMA and SIMS. The latter are probe techniques and are finding increasing experimental and practical application, especially for in situ micro-distribution analysis of the ele- ments. However, in general, they are not yet usable for routine quantitative applications. Conclusion There is a great risk, in trying to review so wide a field in so brief a space, that the broad message will become lost in the detail, but I do wish to draw attention to the multitude of atomic spectroscopic methods that are still undergoing further development.It should be stressed that only a selection of the analytical tools available, and their limits, could be described here. Most of the determination methods listed here form the final step in extreme trace analytical combined procedures, and this will be so until suitable standard reference samples for the calibration of direct instrumental procedures are available in all fields. As up to now most known instrumental analysis methods are not universally applicable to all elements or, as the power of detection differs very much from one element to another, mass spectrometric methods, especially SN-MS, are very promising.Extremely good powers of detection can be expected in the future from LIF and LEI. TXRF, PIXE and SYXRF are helpful techniques under development for micro- distribution trace analysis. Further, it should be stressed that, apart from SN-MS and activation analysis, the determination of each element is subject to strong concentration-dependent cross-interferences from concomitant elements and the opti- mum power of detection and reliability can only be achieved when the elements to be determined are first isolated from the sample (Fig. 24). Therefore, we must pay equal attention to Ar+ I 100 8 1 .: 6o [ 40 20 40 60 80 100 120 140 mlz 1 x 30000 Ion gun Multiplier U I I H.f. Dlasma I 110 115 120 125 rnlz I Sn+ 1 Target / Electrical diaphragm mass spectrometer Fig.23. Principle of SN-MS for surface analysis ion; bombardment: separate ion gun; postionisation: h.f. plasma SamDle 1 I I 115 116 117 118 0' rnlz Fig. 21. S ectrum of copper sample by GD-MS, with the ranges rnlz 100-148and 114-119 expanded. Ag = 14.5, Sn = 4.8, Sb = 1.8, Te = 10.6 pmol mol-I. From Jakubowski el af.55 Multi-stage procedure Sample Decomposition Multiplier H.f. Dlasma I Separation Determination && 0.00.0 neter - mass spectron Fig. 22. pg g-l range. Ion bombardment, ionising h.f. plasma Principle of SN-MS for bulk and thin film analysis in the Fig. 24. Schematic diagram of direct- and multi-stage proceduresANALYST, APRIL 1987, VOL. 112 375 Table 2. Capability of determination methods in elemental trace analysis Limit of determination (order of magnitude)*/ Matrix Multi-element Speciation analysis Method Technique ng ml-1 effectst determination 0 .. . . . . . . 0 0 . . AAS Flame (W-loop) (0) Furnace HG (HGC) 0 (0) ZAAS . . . . . . Furnace 0 0 OES-DCP . . . . . . Pneum. nebulis. OES-ICP . . . . . . Pneum. nebulis. H 0 ETV H 0 H (0) . . ETV H (0) . . HG (HGC) H (0) . . OES-HC H 0 . . . . . . . Volatilisation FANES 0 0 . . AFS H 0 . . . . . . . . . Laser ICP/ICP H 0 . . 0 . . HG (HGC) OES-MIP . . . . . . Pneum. nebulis. H . . . . . . HCACP 0 Laserhrnace 0 . . CFS H 0 . . . . . . . . . . Furnace H 0 . . XRS . . . . . . . . XRF PIXE H 0 . . Optogalvanic . . . . TR-XRF 0 0 MS 0 0 . . . . . . . . . ID-TI ID-ICP H 0 ID-FD 0 0 . NAA . . . . . . .. INAA H ( 0 ) RNAA cl ( 0 ) Voltammetry . . . . DPASV 0 0 . . DPCSV 0 0 . . ChelateGC.. . . . . ECD 0 0 . . ChelateHPLC . . . . UVD H 0 . . Spectrophotometry H Fluonmetry . . . . 0 0 . 0 . * H, For a high number of elements; 0, only for special elements. t 0 , Low; 0 , medium; 0 , high; and 0 , very high matrix effect. No No No No Yes Yes Yes Yes Limited Limited Limited Limited Limited Yes Yes Yes Yes Limited 2 > 6 Z> 14 Z> 14 Limited Yes Limited Yes Limited Limited Limited Limited Limited No No Possible Possible Po s s i b 1 e Limited Yes Possible Yes Yes Yes Yes Yes Yes determination methods and to methods of sample prepara- tion, such as the special techniques for sampling, the methods of sample dissolution and separation and pre-concentration methods in extreme trace analysis, which have not been treated in this paper.This should explain why, in our contributions so far, we have placed much emphasis on the subject of sample preparation. The problems encountered therein can, in principle, be overcome today1.59; however, most of the time they are not taken into account sufficiently. This neglect is the main cause of the widespread uncertainties in trace analysis at the ng g-1 and pg g-1 levels. Therefore, efforts in the field of sample pre-treatment are just as important as, if not more important than, the development of powerful determination methods. To conclude, I should again point out clearly that numerous other determination methods, including the classical chemical methods, have their firm place in the toolbox of the extreme trace analyst, not only with respect to the power of detection, but also in view of other significant merits, illustrated in summary form in Table 2.References 1. Tolg, G., in Malissa, H., Grasserbauer, M., and Belcher, R., Editors, “Nature, Aim and Methods of Microchemistry,” Springer Verlag, New York, 1981, pp. 203-230. 2. Parsons, M. L., Major, S., and Forster, A. R., Appl. Spectrosc., 1983, 37, 411. 3. Broekaert, J. A. C., and Tolg, G., Fresenius 2. Anal. Chem., in the press. 4. Slavin, W., Anal. Chem., 1986, 58, 589A. 5. Berndt, H., and Jackwerth, E., Spectrochim. Acta, Part B, 1975,30, 169. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. Berndt, H., and Messerschmidt, J., Anal. Chim. Acta, 1982, 136, 407. Brown, A.A., Milner, B. A., and Taylor, A., Analyst, 1985, 110,501. Berndt, H., and Messerschmidt, J., Anal. Chim. Acta, 1982, 136,407. Berndt, H., Spectrochim. Acta, Part B, 1984, 39, 1121. Smith, S. B., Jr., and Hieftje, G. M., Appl. Spectrosc., 1983, 37, 419. Broekaert, J. A. C., Spectrochim. Acta, Part B, 1982, 37, 65. Stephens, R., CRC Crit. Rev. Anal. Chem., 1980, 167. Wibetoe, G., and Langmyhr, F. J., Anal. Chim. Acta, 1984, 165, 87. Langmyhr, F. J., and Wibetoe, G., Prog. Anal. At. Spectrosc., 1985, 8, 193. Vollkopf, U., Grobenski, Z., Tamm, R., and Welz, B., Analyst, 1985, 110, 573. Colloquium of the State-of-the-Art of Solid Sampling AAS, Wetzlar, 1984, Fresenius 2. Anal. Chem., 1985, 322, 654. Stoeppler, M., Kurfiirst, U., and Grobecker, K. H., Fresenius 2.Anal. Chem., 1985,322,687. Laqua, K . , in Omenetto, N., Editor, “Analytical Laser Spectroscopy,” Wiley, London, 1979, pp. 47-118. Welz, B., “Atomabsorptionsspektrometrie,” Third Edition, Verlag Chemie, Weinheim, Deerfield Beach, Bade, 1983. Piwonka, J., Kaiser, G., and Tolg, G., Fresenius 2. Anal. Chem., 1985, 321, 225. Baxter, D. C., Frech, W., and Lundberg, E., Analyst, 1985, 110,475. Barnett, W. B., Bohler, W., Carnick, G. R., and Slavin, W., Spectrochim Acta, Part B, 1985,40, 1689. Baasner, J., Berndt, H., and Eiermann, R., in Web, B., Editor, “Fortschritte in der Atomspektrometrischen Spuren- analytik,” Band 2, VCH Verlagsgesellschaft, Weinheim, 1986, Schmid, W., and Krivan, V., Anal. Chem., 1985, 57, 30. pp. 387-395.376 ANALYST, APRIL 1987, VOL.112 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. 41. 42. Koch, K. H., Spectrochim. Acta, Part B, 1984, 39, 1067. Laqua, K. in Kelker, H., Editor, “Ullmanns Encyklopadie der Technischen Chemie,” Band 5 , Verlag Chemie, Weinheim, Beenakker, C. I. M., Spectrochim. Acta, Part B, 1977,32,173. Kollotzek, D., Tschopel, P., and Tolg, G., Spectrochim. Acta, Part B, 1984,39, 625. Volland, G., Tschopel, P., and Tolg, G., Spectrochim. Acta, Part B, 1981, 31, 901. Broekaert, J. A. C., and Leis, F., Mikrochim. Acta, 1985, 11, 261. Kaiser, G., Gotz, D., Tolg, G . , Knapp, G., Maichin, B., and Spitzy, H., Fresenius 2. Anal. Chem., 1978, 291, 278. Kolletzek, D., Oeschsle, D., Kaiser, G., Tschopel, P., and Tolg, G., Fresenius 2. Anal. Chem., 1984,318,485. Laqua, K., and KO, J. B., in “Analytikertreffen 1982,” Wiss. Beitrage in der Karl-Man-Universitat, Karl-Man-Stadt , 1983, p. 142. Quentmeier, A., and Laqua, K., in Koch, K. H., and Massmann, H., Editors, “13. Spektrometertagung,” Walter de Gruyter, Berlin, 1981, p. 37. Falk, H., Hoffmann, E., and Liidke, Ch., Spectrochim. Acta, Part B, 1984, 39,283. Littlejohn, D., and Ottaway, J. M., Analyst, 1979, 104, 208. Wirz, P., Debus, H. Hanle, W., and Scharmann, A., Spectro- chim. Acta, Part B, 1982, 37, 1013. Niemax, K., Appl. Phys., 1985,38, 1. Niemax, K., Lawrenz, J., Obrebski, A . , and Weber, K.-H., Anal. Chem., 1986,58, 1566. Omenetto, N., and Human, H. G. C., Spectrochim. Acta, Part B, 1984,39, 115. Wobrauschek, P., and Aiginger, H., Anal. Chem., 1975, 47, 852. Schwenke, H., and Knoth, J., in Bratter, P., and Schramel, P., Editors, “Trace Element Analytical Chemistry in Medicine and Biology,” Walter de Gruyter, Berlin, 1980, p. 307. 1980, pp. 441-500. 43. 44. 45. 46. 47. 48. 49. 50. 51. 52. 53. 54. 55. 56. 57. 58. 59 Stossel, R.-P., and Prange, A., Anal. Chem., 1985,57,2880. Baumgardt, B., Klockenkamper, R., and Tolg, G., in “Abstracts, XXIV Colloquium Spectroscopicurn Inter- nationale Garmisch-Partenkirchen,” 1985, p. 738. Ruch, C., Rastegar, F., Heimburger, R., Maier, E. A., and Leroy, M. J. F., Anal. Chem., 1985, 57, 1691. Garten, R. P. H., in Fresenius, W., Gunzler, H., Huber, W., Lunderwald, I., and Tolg, G., Editors, “Analytiker Taschen- buch,” Band 4, Springer Verlag, Berlin, 1984, pp. 259-286. Bos, A. J. J., Vis, R. D., Verneul, H., Prins, M., Davies, S . T., Bowen, D. K., Makjanic, J., and ValkoviE, V., Nucl. Instrum. Methods, 1984, B3,232. Bos, A. J. J., van der Stap, C. C . A. H., Vis, R. D., and Valkovit, V., Spectrochim. Acta, Part B, 1983,38, 1209. Kelelsen, P., Knochel, A., and Petersen, W., Fresenius 2. Anal. Chem., 1986, 323, 807. Giauque, R. D., Jaklevic, J. M., and Thompson, A. C., Anal. Chem., 1986, 58, 940. Iida, A., Yoshinaga, A , , Sakurai, K., and Gohsi, Y . , Anal. Chem., 1986,58, 394. Heumann, K. G., Beer, F., and Weiss, H., Mikrochim. Acta, 1983, I, 95. Houk, R. S., Anal. Chem., 1986,58,97A. Harrison, W. W., Hess, K. R., Marcus, R. K., and King, F. L., Anal. Chem., 1986, 58,341A. Jakubowski, N., Stuwer, D., and Tolg, G., Znt. J . Mass Spectrom. Ion Proc., 1986, 71, 183. Oechsner, H., and Stumpe, E., Appl. Phys., 1977, 14, 43. Muller, K. H., and Oechsner, H., Mikrochim. Acta, 1983, Muller, K. H., Seifert, K., and Wilmers, M., J. Vac. Sci. Technol., 1985, A3, 1367. Tschopel, P., and Tolg, G., J. Trace Microprobe Tech., 1982, 1, 1. Paper A61251 Received July 2.5th, I986 Suppl. 10, 51.

ISSN:0003-2654    

DOI:10.1039/AN9871200365

出版商:RSC    

年代:1987

数据来源: RSC

摘要:

ANALYST, APRIL 1987, VOL. 112 377 Regression Techniques for the Detection of Analytical Bias* Brian D. Ripley Department of Mathematics, University of Strathclyde, Glasgow GI IXH, UK and Michael Thompson Applied Geochemistry Research Group, Department of Geology, Imperial College, London S W7 ZBP, UK Regression techniques are commonly applied to compare two analytical methods at several concentrations and to test the biases of one method relative to another. However, regression is strictly applicable only when one method is without error, for example in comparisons with reference materials. A regression-like technique, maximum-likelihood fitting of a functional relationship (MLFR), is explained and is demonstrated to work well. Under some conditions weighted regression provides a good approximation to MLFR, and so can be used if more convenient. Keywords : Bias; regression analysis; functional relationship; chemom e trics When a new analytical method is to be tested in a laboratory it may be used on samples of suitable reference material, but more often it is compared with an existing method on a range of suitable materials whose concentration levels are not known at all precisely.It is important that the samples chosen cover the range of concentrations expected in future use, as the bias of the new method can vary with concentration (Thompson1 discussed the possible forms of this variation in some detail). Regression techniques are commonly used to estimate the bias of the new method. It is not commonly appreciated that these techniques are theoretically wrong and can be practically misleading when used to compare two methods, both of which are subject to error.This will be so in all comparisons of analytical methods except where precisely known reference materials are used. Consider Fig. 1. Here we simulate the effect of taking ten reference samples with concentrations well spread throughout the range 0-1, and measuring them with a new method which actually has no analytical bias (the known reference values are the abscissa and the measured values the ordinate). Applying linear regression, we obtain a fitted line near y = x [Fig. l(b)] and would reasonably conclude that the new method has no appreciable bias. Note that regression is not a symmetrical technique; we talk of regressing y on x.If we interchange the roles of x and y we can still fit a regression line, shown in Fig. l(c). This has a steeper slope and might mislead us to suggest that analytical bias was present. (It can be shown always to have a steeper slope.) Suppose now that we did not know the reference values for these samples, but had measured them by an existing method. We would then have the observations shown in Fig. 2. If we regress the results of the new method on those of the old method we will usually find a smaller slope and might suggest that the new method is biased, high at low concentrations and low at high concentrations. Even if we force the regression line to pass through the origin we will still (usually) obtain a slope of less than 45". The lack of symmetry could also be an embarassment here.Suppose the only difference between the methods was the analysts who carried them out. Then each analyst would feel that his results should be the reference (abscissa) results. Taking the point of view of each analyst in turn, we might conclude that each analyst felt the other was obtaining too low results at high concentrations! Fig. 3 gives a set of results leading to this conclusion, even though no biases are present. *Presented at SAC 86, the 7th SAC International Conference on Analytical Chemistry, Bristol, UK, 20-26 July, 1986. In the results of this paper we examine why these problems occur and give a method that avoids them, in particular giving the same logical conclusion for each analyst. Linear Regression It is now necessary to introduce some notation.Let n = the number of materials analysed; xi = observation by old method on material i; yi = observation by new method on material i; and ui = reference value for material i. Throughout we assume that the old method is not analytically biased. This is really a matter of definition unless universally accepted reference values are available, when we would use these for xi. The assumption is that the average value Of X i over repeated measurements would be ui. The basic assumption underlying regression is that the average value of y depends on x, so that y = f(x) + measurement error For linear regression we suppose that f( ) is a linear function, so that y = a + px + measurement error * * (1) Already we have a problem, as y ought to depend on the true value u rather than the observed value x .This might be ignored if the observed values are very accurate, but not otherwise. The slope p and intercept a of equation (1) express the way in which the analytical bias of the new method varies with concentration. If no bias exists then a = 0 and p = 1, the point of the tests being to see if this is so. Linear regression chooses values a = a and p = b to minimise +cyi - a - pxiy following the principle of least squares. Fig. 4(a) illustrates geometrically that in regressing y on x we work with vertical distances. However, in regressing x on y we use horizontal distances [Fig. 4(b)], so there is no reason to expect to obtain the same fitted line. This has been recognised as a problem for over 100 years.For example, Adcock2 suggested in 1878 minimising the sum of squared perpendicular distances to the fitted line [Fig. 4(c)]. This has the immediate advantage of symmetry in the two methods, and is a special case of the MLFR technique developed later in this paper.378 ANALYST, APRIL 1987, VOL. 112 C U 3 a -4- C C .- t T C (I: C C c ( 0 Reference values Fig. 1. (a) Simulated set of results of a comparison of a new method with reference values. True relationship shown. (b) Linear regression of y on x . (c) Linear regression of x on y 0 Concentration by existing method Fig. 2. values. The lines shown are the regression of y on x and y = x As Fig. 1, but with measurement errors on the reference Weighted Regression In linear regression we are implicitly assuming that the magnitude of measurement error will be the same at all concentrations (without this there is no rationale for ordinary Concentration found by analyst A Fig.3. A set of results with regression lines of y on x (bottom line) and x on y (top line) x-va I u e Fig. 4. on x , (b) regression of x on y and (c) Adcock’s solution An illustration of the differences between (a) regression of y least squares). However, this is an unrealistic assumption in applications to analytical chemistry, where it is recognised that the standard deviation increases with concentration (but the coefficient of variation decreases). A widely applicable assumption is thatANALYST, APRIL 1987, VOL. 112 379 u(u) = a0 + el4 . . . . * * (2) so that the variability of the measurement error depends linearly on concentration.This implies that CV(U) = 0 + uo/u which is consistent with the evidence of Horwitz et a1.3 and others.4 We shall assume from now on that the variances of the measurement errors are known. Thus K~ = variance of xi hi = variance of y i There is a slight problem here, as equation (2) supposes that the variance depends on the unknown true concentration. However, u varies Slowly with U, SO Ki = U(Ui)2 z U(Xi)2. The standard technique if the true values are known is weighted regression, which minimises or n F(yi - a - pxi)2/var(yi) n F wi(yi- a- Px~)’ . . . . . . (3) where wi = l/var(yi) are the “weights.” Weighted regression is generally less well known than linear regression but is available as an option in many statistical packages.If the x values are without error it will on average give the same line as linear regression but estimate a and p more precisely. Weighted regression is still a regression technique, that is, it assumes that the x values are true, and so has all the problems outlined in the Introduction. Functional Relationships Evidently, the correct assumption in comparing two analytical methods is that xi = ui + error yi = vi + error and vi = f(ui), the central value for the new method. Analytical bias is deviation from vi = ui. Such prescription is known to statisticians as a functional relationship (see, e.g., Sprents). We are concerned with a linear functional relation- ship: to assess the variation of bias with concentration. This formulation of the problem is intrinsically symmetric inp the old and new methods.To estimate a and p we suppose that the “errors” (yi - vi) and (xi - ui) are independent and normally distributed. The method of maximum likelihood tells us to choose &,b,&, . . . , f i n to maximise L = constant - gZ(xi - u ~ ) ~ / K ~ - QE(yi - vj)2/hi . . ( 5 ) with vi = a + pui from equation (4). This is again symmetric. Maximum likelihood is a widely favoured method in statistics for the solution of general statistical problems. In this instance it reduces to Adcock’s suggestion for equal measure- ~ ment error variances (hi = K~ = K). Then v i = a + p u i . . . . * - (4) 1 L = constant - -Z [distance from (xi,yi) to (ui,vi)]2 using Pythagoras’s theorem. Choosing the nearest point on the line gives the perpendicular.distances. Hence we can regard the MLFR technique as generalising Adcock’s idea to precision varying between the two methods and with concentration. It remains actually to maximise L to find a and p. This is described in the Appendix. In general, some iterative numerical technique has to be used, but the technique is simple to program and well within the capabilities of a microcomputer. The MLFR technique is nearly statistically unbiased. 2K Iterated Weighted Least Squares Let us write bi = xi - ui and ci = y i - vi for the measurement errors in the old and new methods, respectively. Then our assumptions can be expressed as xi = ui + 6i yi = Vi + Ei = a + pUi + Ei = + p ( X i - 6i) + Ei = a + pxi + ( E j - pa,> showing the dependence of yi on xi.The error term qi = E~ - has variance This suggests using weighted least squares with weights wi(f3) = l/var(qi) = l/(hi + ~ P K ~ ) . This is a very similar procedure to MLFR but still maintains all the disadvantages of regression techniques. Nevertheless, we can ask if it provides a good approximation to MLFR. To use this estimator we have to replace p by the weighted regression estimate b in wi. Thus the fitting procedure Is iterative, using the current value of kuntil the process stabilises. Let us call the resulting estimate p, and the technique IWLS. It can be shown6 that on average a underestimates p. This corresponds to the flattening of the slope discussed in the Introduction. The size of tke effect can be seen in the special case QI = 0, p = 1, fii = v~~ = hi with the line constrained to go through the origin.Then the bias in the estimation of p is about 82, whereas the standard deviation of B == O m . This suggests that the weighted regression will give misleading results if no2 > 1/2 [when bias > 4 (standard deviation)], that is for large numbers of samples or poor precision. A general expression is given in the Appendix that can be used to check if it is safe to use weighted regression, or to correct it to a better approximation to MLFR. Testing for Bias The main purpose of these techniques in comparing analytical methods is to test a = 0, p = 1, and perhaps if this is rejected to use the fitted line to re-calibrate the new method. Systematic statistical biases in estimating will affect the testing of (Y = 0 as well as f3 = 1.Hence we assume that MLFR or bias- corrected iterated weighted least squares has been used to avoid such problems. To test p = 1 we compute using wi = wi(B), whereas for a = 0 we use (B - 1)VEWi ( X i -X)2 &//2WiXi2EWi2Wi(Xi - 9 2 in both instances using a 2-test for significance (so a 5% significance test rejects values outside [ -1.96, 1.961). Here x = ZwixiEwi is the weighted mean. If these tests reject there is another possibility that we should check, namely that the variance specifications are realistic. If the measurement errors are much larger than expected then & and fi will be more variable and so bias might be detected in error. A good check is to compute the scaled residuals ri = (yi - & - Pxi)- The sum of r,2 should be around (n - 2), and a plot of ri against lii should not show any noticeable pattern, such as a U-shaped plot or greater variation at one end than the other.These indicate more complex problems and a need to consult a statistician. Examples Our first example is a set of 30 pairs of determinations of arsenate(V) ion in natural river waters7 (Table 1). The x values are determined by selective reduction and atomic absorption spectrometry, whereas the y values came from cold trapping and atomic emission spectrometry. The quoted380 ANALYST, APRIL 1987, VOL. 112 Table 1. Arsenic(V) in 30 natural waters determined by two methods: (i) continuous selective reduction and atomic absorption spectrometry; and (ii) non-selective reduction, cold trapping, and atomic emission spectrometry. Results in pg 1-l.x, y are the mean concentrations, s ( x ) , sCy) are the standard errors of the mean AASIselective reduction AESkold trapping X 8.71 7.01 3.28 5.60 1.55 1.75 0.73 3.66 0.90 9.39 4.39 3.69 0.34 1.94 2.07 1.38 1.81 1.27 0.82 1.88 5.66 0.00 0.00 0.40 0.00 1.98 10.21 4.64 5.66 19.25 44 1.92 1.56 0.76 1.26 0.39 0.43 0.22 0.84 0.25 2.07 1 .oo 0.84 0.13 0.47 0.50 0.36 0.45 0.33 0.23 0.46 1.27 0.06 0.06 0.15 0.06 0.48 2.24 1.05 1.27 4.18 Y 7.35 7.92 3.40 5.44 2.07 2.29 0.66 3.43 1.25 6.58 3.31 2.72 2.32 1 S O 3 S O 1.17 2.31 1.88 0.44 1.37 7.04 0.00 0.49 1.29 0.37 2.16 12.53 3.90 4.66 15.86 S b ) 2.07 2.23 0.96 1.53 0.59 0.65 0.19 0.97 0.36 1.85 0.93 0.77 0.66 0.43 0.99 0.33 0.66 0.54 0.13 0.40 1.98 0.01 0.15 0.37 0.12 0.62 3.51 1.10 1.31 4.45 values are concentrations varying from 0 to 20 pg 1-1.The following fitted lines were obtained: linear regression: y = 0.544 + 0.845~ x = -0.299 + 1.089~ y = 0.005 + 0.890~ weighted regression: x = -0.167 + 0.631~ y = 0.109 + 0.904~ x = -0.083 + 0.871~ MLFR: y = 0.106 + 0.973~ (standard errors) (0.048) (0.076) IWLS: These lines are illustrated in Fig. 5. Note the serious statistical bias of both (unweighted) linear regression lines, and that in this instance IWLS is not a good approximation to MLFR. For this example ZrF/(n - 2) == 1.27, so the specified variances can be accepted, and a plot of ri versus xi indicates no problems with the fit. The second example is determinations of beryllium in rock and soil reference samples. The x values were obtained by an inductively coupled plasma atomic emission spectrometric (ICP-AES) method after fusion with lithium metaborate and dissolution in dilute nitric acids; the y values were determined by an AAS method after acid decomposition and solvent extraction9 (Table 2).The first batch of specimens with concentrations in the range 0.1-2.5 pg g-1 gave results 0 5 10 15 20 As" by AASlpg I - Fig. 5. Linear fits to 30 data points of concentration of arsenate ion in river water, in pg 1-1. The x values were determined by selective reduction, the y values by cold trapping. In all instances the solid line is the MLFR fit. (a) Linear regression of y on x (A) and of x on y (B). ( b ) Weighted linear regressions. (c) Iterated weighted linear re- gression solutions illustrated in Fig.6. The fitted lines were as follows: linear regression: y = -0.162 + 1.063~ x = 0.203 + 0.899~ weighted regression: y = -0.081 + 0.839~ x = 0.171 + 0.918~ IWLS: y = -0.135 + 1.048~ x = 0.161 + 0.925~ MLFR: (standard errors) (0.053) (0.038) Here the IWLS results are a good approximation to MLFR. We would accept a unit slope but might suspect a translational bias, with a # 0. There is no obvious pattern in the scaled y = -0.167 + 1.076~ANALYST, APRIL 1987, VOL. 112 381 Table 2. Beryllium in rock and soil reference materials determined by two methods: (i) ICP-AES after fusion with lithium metaborate and dissolution in dilute nitric acid; and (ii) AAS after acid decomposition and solvent extraction. Results in pg g-1. x, y are mean con- centrations, s@), sCy) are the standard errors of the mean ICP-AES result AAS result Sample USGS G2* GSP-1* AGV-1* DTS-1* PCC- 1 * GXR-l* GXR-2* GXR-3 GXR-4* GXR-5 * GXR-6* CCRMP SO-1 so-2 SO-3 SO-4 SY-1 SY-2 SY-3 NIM NIM-L NIM-N NIM-G NIM-P NIM-S NIM-D ANRT DR-N FK-N GS-N UB-N MA-Nt BE-N AN-G CRPG GH BR X 2.34 1.20 1.88 0.08 0.12 1.12 1.60 2.16 1.34 1.35 2.04 1.97 1.02 1.45 22.4 28.2 22.6 22.37 27.0 0.38 7.27 0.28 1.55 0.06 1.50 1.06 5.19 0.33 1.96 0.33 5.54 1.85 300 GA 3.40 MICA-FE 4.19 MICA-MG 0.04 BCS BCS-375 3.02 BCS-376 1.33 4x1 0.08 0.09 0.08 0.07 0.07 0.10 0.10 0.52 0.08 0.09 0.09 0.11 0.08 0.06 0.07 0.63 0.37 0.23 0.61 0.07 0.15 0.05 0.05 0.07 0.10 0.09 0.12 0.07 6.0 0.11 0.09 0.12 0.11 0.06 0.11 0.07 0.09 0.09 Y 2.48 1.22 2.14 0.0026 0.0023 1.05 1.42 1.99 1.06 1.04 1.89 1.90 0.75 1.16 26.3 29.4 23.7 23.3 29.5 0.35 7.75 0.14 1.61 0.013 1.63 1.05 5.50 0.054 1.80 0.23 5.21 1.76 3.62 4.07 0.048 3.06 1.35 317 s(Y> 0.05 0.04 0.045 0.0006 0.0006 0.035 0.04 0.29 0.045 0.04 0.04 0.045 0.045 0.03 0.03 0.31 0.30 0.26 0.32 0.04 0.10 0.026 0.04 0,001 0.04 0.035 0.08 0.0016 3.1 0.045 0.03 0.07 0.04 0.06 0.065 0.0010 0.055 0.04 * Specimens selected for separate treatment in the “first batch.” t Specimen not included in any calculations.residuals (ri) but 2ri2/(n - 2) = 2.52, an indication that the specified variances are too small. We therefore conclude from this sample that there is no difference between the methods at these concentrations. The techniques were then applied to the 37 specimens with concentrations in the range 0.0-30.The fitted lines were as follows: linear regression: weighted regression: y = -0.194 + 1.084~ x = 0.193 + 0.920~ y = -0.066 + 0.795~ x = 0.148 + 0.937~ IWLS: y = -0.153 + 1.062 x = 0.152 + 0.936~ MLFR: (standard errors) (0.020) (0.008) y = -0.161 + 1.068~ 2.5 2.0 1.5 1 .o 0.E 0 2.E 2.c c I ul 9 1.5 B 1.c 3 0) m 0.: ( 2.! 2.1 1 .t 1 .( O.! I I I / / / / / / / 0.5 1.0 1.5 2.0 2.5 0 Be by ICP-AESIpg g-1 Fig. 6. Determination of beryllium in rock and soil materials, in pg g-1. The x values were obtained by ICP-AES and the y values b AAS (see text). (a) Linear regression of y on x (A) and x on y (B). (by Weighted linear regressions of y on x (A) and x on y (B). (c) MLFR line (-) with IWLS y on x (A) and x on y (B) Here the results are better. The two linear regression lines are almost coincident but both are statistically biased.Weighted regression of y on x is worse, but the remaining four lines are indistinguishable on the scale of Fig. 7. We found 2rFl(n - 2) = 2.78, again indicating some optimism in the variance specification. Even allowing for this by increasing the standard errors by we can still conclude that both translational and rotational analytical biases exist between the two methods. The MLFR line accords well with the thirty-eighth value (300,317) which was omitted from the calculations as it would have dominated the fitted lines. Conclusions Both the theory and the examples show that the unthinking use of linear regression can lead to seriously misleading conclusions about the relative bias of two analytical methods.Weighted regression, especially with weights wi = l/(var yi + p2 var xi), can be better. However, a maximum likelihood382 ANALYST, APRIL 1987, VOL. 112 This is evidently minimised by to give 0 3 0 W m Be by ICP-AES/kg 8-1 Fig. 7. As Fig. 6, for 37 samples from all sources. The solid line is MLFR, the dashed lines are linear regressions of y on x and x on y (almost coincident lines) functional relationship line should be used if possible. [A FORTRAN 77 program for MLFR is available from one of the authors (B. D. R.)]. Appendix: Finding the MLFR Line General Case Our task is to maximise L given in expression (5) and subject to equation (4). This is equivalent to minimising Q = Z [ ( X ~ - ~ i ) 2 / ~ i + bi - (Y - @ ~ i ) ~ / h i ] First minimise over ui. The only term involving Ui is Qi = [(Xi -.t ~ i ) ~ / K i + Cyi - CY - PUi)2/hi] Note that wi depends on p but not a. For a line through the origin we take & = 0. Otherwise, we find xwi ( j i - PXi) & = 2 W i which depends on p. Finally, we must minimise Qm(&,P) = x W i ( P ) l y i - &(P) - b i I 2 by numerical means to find the MLFR estimate of p. A suitable algorithm is Nash’s algorithm 17.10 A Special Case In the special case of hi cc K~ (which includes variances that do not depend on concentration) we can proceed further. Let h i = C K ~ . Then so does not depend on p, and where If we seek a stationary point of Qm(&,P), we find 2 [SxyP2 + ( C L - SyJP - CSxyI (c + P2I2 This is a quadratic equation with two solutions (Syy-CSxx) * V(CSXY - SYY)2 + 4csx,2 2 s x y 8 = of opposite signs.The minimum occurs when has the sign of Sxy. Hence in this instance we can compute B exactly. (If & = 0 takes f = jj = 0). In this special instance the iterated weighted least-squares solution b reduces to the weighted regression of y on x with weights (Ki-’), as wiEwi = K~-KZK~-~. Now -2bQ < O -- - (c + b 2 ) + (yi - a - Pxi>2) provided b > 0. Hence > b whenever the latter is positive.ANALYST, APRIL 1987, VOL. 112 383 Distribution of In general, the bias of B is downwards. We have & = O & free mean of F: variance of l/XwiuF l E W i ( U i - qz p(1-&ViKiEWiUF) P[I - X W ~ K ~ E W ~ ( U ~ - it)21 Hence the mean of is p X W i K i I I h W i ( U i - Z)2 standard deviations too low, and the underestimation will be important if this exceeds one half. An alternative is to correct the bias by using N p[1 + XWiKiEWi(Ui - q2] again taking Z = 0 if the line is constrained to pass through the origin. To use these results we must replace ui by xi. References 1. Thompson, M., Analyst, 1982, 107, 1169. 2. Adcock, R. J., The Analyst (Des Moines, Zowa), 1878, 5 , 53. 3. 4. 5. 6. . 7. 8. 9. 10. Horwitz, W., Kamps, L. R., and Boyer, K. W., J. Assoc. Ofl. Anal. Chem., 1980,63, 1344. Zitter, H., Fresenius 2. Anal. Chem., 1971, 255, 1. Sprent, P., “Models in Regression and Related Topics,” Methuen, London, 1969. Ripley, B. D., unpublished work. Anderson, R. K., Thompson, M., and Culbard, E., Analyst, 1986, 111, 1153. Watkins, P. J., andThompson, M., Geostand. Newsl., 1983,7, 273. Terashima, S . , Geostand. News., 1983, 7 , 295. Nash, J. C., “Compact Numerical Methods for Computers: Linear Algebra and Function Minimisation,” Adam Hilger, Bristol, 1979. Papers A61237 Received July 21st, 1986 Accepted October 7th, 1986

ISSN:0003-2654    

DOI:10.1039/AN9871200377

出版商:RSC    

年代:1987

数据来源: RSC

摘要:

ANALYST, APRIL 1987, VOL. 112 385 Chemometrics in Pharmaceutical Analysis* John C. Berridge Analytical Chemistry Department, Pfizer Central Research, Sandwich, Kent CT13 9NJ, UK In this introduction to the subject of chemometrics, attention is focused on experimental design, considering both manual and automated applications of chemometric methods to systems of pharmaceutical interest. Additionally, the use of robotics and expert systems is briefly discussed. Keywords: Chemometrics; pharmaceutical analysis; robotics; expert systems; automated analysis Chemometrics is defined1 as the application of mathematical and statistical methods to design or select optimum pro- cedures and experiments and to provide maximum chemical information by analysing chemical data. This is perhaps too restrictive a definition as the use of computational techniques now also deserves specific mention.Additionally, it should not be thought that chemometrics is restricted simply to chemistry as there will be applications in any area where experiments have to be defined and the data they produce analysed. The philosophies behind this rapidly growing sub-discipline are two-fold. Firstly, that chemometric methods are applied to the design and implementation of analyses so that the most efficient and informative experi- ments are carried out. Secondly, that the experimental data should then be examined by appropriate methods to ensure that all available information is extracted and that these data are presented in formats amenable to rapid assimilation and interpretation. Both aspects are paramount to pharmaceutical analysis.The need for economic efficiency coupled with the increasing regulatory requirements behoves every phar- maceutical analyst to ensure that experiments are both well designed and interpreted. It will not be possible in this introduction to the use of chemometrics to consider all possible uses of mathematical and statistical methods. Attention has thus been directed towards a number of the more successful and general techniques that can be easily adapted to a wide range of problem types. As a robot is simply a computer able to carry out physical manipulations, it is appropriate to examine briefly the uses of robotic techniques in pharmaceutical analysis. Finally, as expertise in analysis continues to expand at an ever increasing rate, the use of expert systems to guide experimental design will also become increasingly common.The Impact of Chemometrics on Experimental Design In all experimental situations it is vital to follow four basic steps: (i) Understand the problem: what information is needed and which variables are available for investigation? Which criteria will be used to judge a satisfactory outcome? (ii) Devise a plan for solving the problem: ideally, use all options available to investigate the problem (but see below!). (iii) Carry out the experiments. (iv) Review the data: what were the less important variables? What information is redundant? Are there any other techniques to be used in acquiring data or reviewing results? *Presented at SAC 86, the 7th SAC International Conference on Analytical Chemistry, Bristol, UK, 20-26 July, 1986. An excellent example of how chemometrics has been applied to assist the first step is in the standardised strategy devised by Massart and co-workers2-8 for the extraction and determination of drugs in formulations and biological samples. A standardised extraction strategy, using either sodium n-octylsulphonate or di(2-ethylhexy1)phosphoric acid is coupled with HPLC determination using a single stationary phase (cyanopropyl) with a restricted set of mobile phase components. The selection of the preferred components was made9 on the basis of information theory, thus feeding back into new experimental designs the results from previous experiments.These studies have been extended to demon- strate the applicability of this approach to acidic, basic and neutral substances (mostly drugs) .8 Such standardised pro- cedures greatly simplify the choice of experimental paramet- ers.Step (ii) is the selection of an experimental design. Experimental designs can be classified conveniently into two broad categories: firstly simultaneous designs where all experiments could be conducted at the same time but where they must all be completed before data analysis can be carried out, and secondly sequential designs where the results from the last experiment(s) are used to determine the conditions for the next. Thus factorial designs, arguably one of the most useful formats for efficient experimentation, will be first discussed together with a related experimental design known as the simplex lattice design.Both methods require all the experimentation to be carried out prior to the evaluation of results. From the second category, a method that uses experimental results to guide the procedure towards optimum values is the sequential simplex procedure-a guided search procedure which is comparatively easy to automate. Step (iii) demands more thought than many perhaps devote to it. To truly assess the significance of variables, it is important that the influences of noise and drift are minimised. Simultaneous designs offer the advantage that the experimen- tal order can be randomised. This is not possible in sequential experimental procedures where it is also important to ensure that the direction of progress has not been biased by the presence of noise.Fortunately, the sequential simplex proce- dure has checks built into it to minimise the effects of noise: experiments can also be duplicated if noise is a problem. It is worth noting that one of the uses of the simplex procedure is in the constant tracking and optimisation of experimental systems and instruments that are prone to drift. Step (iv), the analysis of data, is perhaps where chemo- metrics is making its largest impact. Unfortunately, the subject is outside the scope of this introduction and requires separate consideration. Simultaneous Experimental Designs In the development of a new analytical method, it is imperative to have information about the independence and386 ANALYST, APRIL 1987, VOL. 112 Table 1.Worksheet for analysis of factorial designs A 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 A 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 B C D E F G H I J Worksheet for a 23 factorial design with interactions Factor Expt.No. A B C 1 2 + - - 3 - + - 4 + + - + 5 6 + - + 7 - + + 8 + + + - - - - - B C D E Result 23 49 96 74 63 99 96 99 1 72 170 162 195 26 - 22 36 3 2 242 357 4 39 98 33 - 48 - 33 F G H Worksheet for a 23 factorial design with interactions 3 599 43 131 -81 115 35 - 65 15 I Effect Sum A B AB C AC BC ABC J Factor Expt.No. A B C 1 2 + - - 3 - + - 4 + + - + 5 6 + - + 7 - + + 8 + + + - - - - - Result 23 49 96 74 63 99 96 99 1 +FlO+Fll +F12+F13 +F14+F15 +F16+F17 +F11 -F10 +F13-F12 + F15 - F14 +F17-F16 2 +GlO+Gll +G12+G13 +G14+G15 +G16+G17 +G11-G10 +G13-G12 +G15-G14 +G17-G16 3 +HlO+H11 +H12+H13 +H14+H15 +H16+H17 +H11-H10 +H13-H12 +H15-H14 +H17-H16 Effect Sum A B AB C AC BC ABC interdependence of the parameters available for variation.An example might be in the development of a solid-phase extraction method using the disposable extraction cartridges that are now generally available. If the molecule of interest is ionic, pH is obviously a variable requiring investigation. The solvents used in the initial extraction (possibly a mixture of buffer with an organic modifier such as methanol) and the subsequent elution of the molecule may also need investigation. However, the success of the method is likely to depend on the correct selection of both pH and solvent composition-they cannot be investi- gated individually.The fallacy of the “one at a time” approach is illustrated in the hypothetical example shown in Fig. 1. Working at a constant pH of 3.0, the optimum level of organic solvent that should be added is 60% (point Y). Working at a constant level of organic solvent (20%), the optimum pH is now 6.5 (point X). However, the most efficient extraction is actually achieved at a pH of 6 with 50% organic solvent (point Z ) . That it is vital that more than one parameter should be varied at one time may come as a surprise, but chemometrics is about just this philosophy-if possible, vary as many things at once as can be varied and then use mathematical techniques to sort out the significance of these variables. If may be found that more than these two variables actually need consideration; perhaps temperature is important too.Factorial experimental designs are ideal for investigating the relative importance of variables and for establishing their dependences. For relatively small numbers of variables, factorial experimental designs are also efficient in terms of experimentation: the number of experiments required for a full factorial design on n variables is simply 2n. The analysis of factorial experiments is also relatively simple. Computer programs can be written or purchased to assist with their interpretation but most laboratories now possess a “spread- sheet” program that can be very quickly set up to analyse factorial experiments. Table 1 is a worksheet for a three- variable factorial design, showing the interactions and how data analysis can be carried out using Yates’ method.10 There are variants of the basic factorial design, such as the central composite design,ll intended to permit easier fitting of polynomials to the data obtained and reducing the number of experiments required.Where large numbers of variables are being investigated, partial factorial designs can be used. 10 For even larger studies, where in excess of perhaps ten variables are under investigation, an experimental design known as the Plackett - Burman design is useful and this again is easy to analyse using a spreadsheet.12J3 The fact that the number of experiments rises exponentially with the number of factors investigated is one of the drawbacks of using factorial designs, this perhaps being no better illustrated than by a recent (somewhat controversial) proposal for process validation in which it is suggested that five basic parameters be investigated at three different conditions, i.e., 35 = 243 experiments.14J5 Further validation of a single raw material that might have five analytical parameters itself, leads to the need to carry out ( F ) 3 = 32 768 experiments.If three ingredients are involved, some eight million experiments are required for a complete study! Clearly, points (i) and (ii) above require very careful consideration before any experimentation commences. The potential uses of factorial designs in pharmaceutical analysis are many yet the number of literature examples is surprisingly small. Bolton16 demonstrated their use in the design and interpretation of pharmaceutical stability studies.Factorial designs have also been used in the development of pharmaceutical dosage forms such as tablets17 where multipleANALYST, APRIL 1987, VOL. 112 387 2 3 4 5 6 7 8 9 PH Fig. 1. Problems associated with “one at a time” optimisation: true optimum is located at Z 100 Fig. 2. Relationship of mixture design to factorial design MeOH 1 (33.3,33.3,33.3) - 3 (0,50,50) 2 MeCN 5 TH F Fig. 3. Sim lex lattice design: experiments 1-7 are used for estimation o! coefficients, experiments 8-10 are used to check goodness of fit analytical criteria were combined to provide the response from each experiment. HPLC method development has been an area where factorial designs have found use.18 Cotton and Down19 carried out a comprehensive investigation into the factors affecting the reversed-phase separation of sulindac and related compounds, whereas the optimisation of pH and surface-active ion concentration for the separation of weak acids, weak bases and zwitterionic compounds was described by Kong et aZ.20 A central composite design was used by Lindberg et al.21 in the optimisation of the separation of a mixture of morphine, codeine and papaverine by reversed- phase ion-pair chromatography.A sub-section of factorial designs is the simplex lattice design, also referred to as a mixture design.22 From Fig. 2 it can be seen that a mixture design is a factorial design restricted by the requirement that the variables must add up to loo%, i . e . , they are not freely variable. An obvious area for exploitation of mixture designs is in the development of liquid chromatographic methods, both thin-layer and high-perfor- mance liquid chromatography.18 The sum of components in a liquid chromatography mobile phase is 100% and the mixture design permits the simple investigation of three variables.A minimum of seven experiments is conducted under the conditions shown in Fig. 3 and the results of these experiments are then used in the fitting of a polynomial model to the response surface. The form of the polynomial is usually y = a1x1 + a2x2 + a3x3 + a12x1x2 + alfllx3 + aH2x3 + a123xlXg3 The coefficients are estimated using the equations al = rl, a2 = r2, a3 = r3, a12 = 4r12 - 2(rl + r2), a13 = 4r13 - 2(rl + r3), a23 r2 + r3) where ri is the average result for each of the seven experiments.Once again a spreadsheet is useful for the calculation although it is also a relatively simple task to write a suitable computer program. For such a simple experimental design, it is surprising to find that only in chromatography has its use been more fully exploited and it has been possible to use the simplex lattice design as the basis for automated HPLC separation 0ptimisation.23~24 There are other examples of pharmaceutical interest, however, especially in the design of formulations and dosage forms.25.26 = 4r23 - 2(r2 + r3), a123 = 27r123 - 12(r12 + r13 + r23) -k 3(rl + Sequential Experimental Designs A drawback of the use of simultaneous experimental designs is that no data analysis is possible until all experimentation is completed.Sequential designs overcome this problem by feeding back the results of the current experiment in order to determine the conditions of the next. Rapidly becoming the most frequently used method is the sequential simplex procedure.27 A simplex is simply described as a geometrical figure, described by one more point than the number of variables being optimised. The co-ordinates of the simplex at which the worst result was obtained are rejected and reflected through the centroid of the remaining points to provide a new point at which the next experiment is conducted. Fig. 4 shows the movements of simplexes across a response surface guided by the rules of the modified simplex procedure. Here, the simplex size is expanded as moves are made in favourable directions or contracted (and reflected) away from poor response areas.It is beyond the scope of this paper to describe the simplex procedure in detail, its operation being well detailed in the literature.1OJsJ7 The mathematics of the simplex procedure are straightfor- ward, if a little complicated to fully implement on a computer. There are commercially available simplex programs28J9 but, once again, the spreadsheet can be commissioned to carry out the necessary calculations. Table 2 shows a simplex work sheet for three variables: less than one hour was required to set this up from scratch. The simplex procedure is a very powerful method, applic- able in many areas of pharmaceutical and related analysis. As well as being useful in the optimisation of analytical proce- dures, it also has uses in establishing optimum instrument settings.30Jl The procedure is also relatively easy to automate and has found particular use in the fully automated optimisa- tion of a wide variety of HPLC separations.18132 Other areas of successful application of the simplex procedure include the design of capsule formulations,33 clinical chemistry,34 includ- ing the determination of glucose in serum and urine,35 and in general method development .36 The simplex procedure is not without its problems, however, and these must be borne in mind before it is used.It is necessary to devise some response function or criterion that can be used to express the results of experiments in a simple numerical form as input to the procedure. A consequence of using such a function (be it a minimisation or maximisation) is that the procedure may locate a local rather than the true, global optimum.Nevertheless, the ease of implementation of the procedure, its suitability for automation and its need to start with only one more experiment than the number of variables being investigated make the method very attractive.388 ANALYST, APRIL 1987, VOL. 112 Table 2. Worksheet for a three variable simplex procedure. Selection of the new vertex is made according to the rules of the procedure A 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 B C D E F G A B C D E F G 2 3 5 6 Experimentnumber:-. . . . . 7 8 Simplex optimisation worksheet (3-variable) 4 Simplex optimisation worksheet (3-variable) Experiment number:- .. . . . . Vertex X Y Z Result 9 Vertex X Y Z Result 1 10 10 10 3.5 Worst 10 1 10.00 10.00 10.00 3.50 Worst 2 20 10 10 4.67 12 2 20.00 10.00 10.00 4.67 3 10 20 10 6.32 13 3 10.00 20.00 10.00 6.32 4 10 10 20 3.81 14 4 10.00 10.00 20.00 3.81 11 15 16 Sum Centroid P Distance P-w Distance * .5 (P-W) * .5 Reflection R Contraction towards R Contraction towards W Expansion +C12+C13+C14 +D12+D13+D14 +E12+E13+E14 +C17/3 +D17/3 +El713 + c20 - c10 +D20-D10 + E20-El0 + c23/2 + D23/2 + E23/2 +C20+C23 + D20 + D23 + E20 + E23 + C20 + C26 +D20+D26 + E20 + E26 +C20-C26 +D20- D26 + E20 - E26 + C29 + C23 +D29+D23 + E29 + E23 17 Sum 18 19 20 Centroid 21 P 22 23 Distance 24 P-W 25 26 Distance * .5 28 29 Reflection 30 R 31 32 Contraction 33 towards R 34 35 Contraction 36 towards W 37 38 Expansion 39 27 (P-W)*.5 40.00 40.00 40.00 13.33 13.33 13.33 3.33 3.33 3.33 1.67 1.67 1.67 16.67 16.67 16.67 15.00 15.00 15.00 11.67 11.67 11.67 20.00 20.00 20.00 L I Level of variable 1 Fig.4. Modified simplex procedure. The initial simplex is ABC. Point A is rejected to give point D after which the simplex size is doubled Expert Systems and Robotics In its simplest definition, an expert system is just a computer program designed to function as though it were an expert on the subject in question. The key features of the program are that it uses sets of rules to operate upon a knowledge base (a data base) and that is has a “natural language interface” through which the user can ask questions such as “how” and “why.” Expert systems are still in their infancy in pharmaceuf- ical analysis but a promising area is in the assistance with experimental design.37.38 Expert systems are under develop- ment that will allow the analyst to debate the experimental conditions required for a particular analysis and that can then be used to determine appropriate machine configurations, settings or conditions .3 8 3 HPLC method development, often characterised more by luck than judgement, is also benefiting from the application of expert systems.39,40 Robotics are also legitimately classified as a section of chemometrics as a robot can be defined as a computer capable of physical manipulations. Robots are rapidly gaining a place in all aspects of pharmaceutical and clinical analysis, from the assay of biological samples, through routine assay procedures, to the fully automated testing of dissolution profiles41- including washing up the apparatus afterwards! Robots are also being coupled with analytical instrumentation and optimi- sation software to carry out unattended method develop- ment.42 An additional advantage of the use of robots is the possibility of on-line feedback and continuous validation of a procedure.43 No doubt the complete automation of analytical method development will soon be realised by the linking of a robot with an expert system: what role for the analyst then? Conclusions The careful and efficient design of experiments is paramount for successful and economical pharmaceutical analysis.Chemometric principles have a vital role to play in the acceptance of the need to investigate the interdependence ofANALYST, APRIL 1987, VOL.112 389 variables and to abolish the philosophy that it is bad practice to vary more than one parameter at a time. The formalisation of the discipline of chemometrics heralds a new era in analysis as robotics and expert systems become a practical reality-the literature will no doubt expand appropriately44.45 to keep pace with these exciting developments. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. References Kowalski, B. R., Trends Anal. Chem., 1981, 1, 71. Hoogewijs, G., and Massart, D. L., J. Pharm. Biomed. Anal., 1983, 1, 321. Detaevernier, M. R., Hoogewijs, G., and Massart, D. L., J. Pharm. Biomed. Anal., 1983, 1, 331. Hoojewijs, G., and Massart, D. L., J.Pharm. Biomed. Anal., 1984, 2, 449. Hoogewijs, G., and Massart, D. L., J. Liq. Chromatogr., 1983, 6, 2521. Hoogewijs, G., and Massart, D. L., J. Pharm. Belg., 1983,38, 76. Hoogewijs, G., and Massart, D. L., J. Chromatogr., 1984,309, 329. De Smet, M., Hoogewijs, G., Puttemans, M., and Massart, D. L., Anal. Chem., 1984, 56, 2662. Massart, D. L., and Hoogewijs, G., Pure Appl. Chem., 1983, 55, 1861. Massart, D. L., Dijkstra, A., and Kaufman, L., “Evaluation and Optimisation of Laboratory Methods and Analytical Procedures,” Elsevier, Amsterdam, 1978. Deming, S. N., and Morgan, S. L., Anal. Chim. Acta, 1983, 150, 183. Abell, M., Trends Anal. Chem., 1984, 3, (3), VII. Kateman, G., Trends Anal. Chem., 1986, 5 , (3), IV. Berry, I. R., in Loftus, B.T., and Nash, R. A., Editors, “Pharmaceutical Process Validation,” Marcel Dekker, New York, 1984, pp. 203-249. Sharp, J. R., Pharm. J . , 1986, 236, 43. Bolton, S. J., J. Pharm. Sci., 1983, 72, 362. Founer, D. E., Buck, J. R., and Banker, G. S . , J. Pharm. Sci., 1970, 59, 1587. Berridge, J. C., “Techniques for the Automated Optimisation of HPLC Separations,” Wiley, Chichester, 1985. Cotton, M. L., and Down, G. R. B., J . Chromatogr., 1983, 259, 17. Kong, R. C., Sachok, B., and Deming, S. N., J. Chromatogr., 1980, 199, 307. Lindberg, W., Johansson, E., and Johansson, K., J. Chromat- ogr., 1981, 211, 201. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. 41. 42. 43. 44. 45 * Snee, R. D., Chemtech, 1979,9,702. “SENTINEL System,” Du Pont, Wilmington, DE.Leher, R., Ind. Res. Dev., 1983, 25, 116. Takayama, K., Imaizumi, H., Nambu, N., and Nagai, T., Chem. Pharm. Bull., 1985, 33,292. Huisman, R, Van Kamp, H. V., Weyland, J. W., Doornbos, D. A., Bolhuis, G. K . , and Lerk, C . F., Pharm. Weekbl. Sci. Ed., 1984,6, 185. Deming, S. N., and Morgan, S . L., Anal. Chem., 1973, 45, 278A. Van der Wiel, P. F. A., Kateman, G., and Vandeginste, B. G. M., “CHEOPS-Chemometrical Optimisation by Simplex,” Elsevier, Amsterdam. Deming, S . N., and Morgan, S . L., “Simplex-V, Statistical Programs,” Rowlett, Houston, TX. Parker, L. R., Morgan, S. L., and Deming, S. N., Appl. Spectrosc., 1975, 29, 429. Siegel, M. M., Anal. Chim. Acta, 1981, 133, 103. Berridge, J. C., J. Chromatogr., 1982, 244, 1. Shek, E., Ghani, M., and Jones, R. E., J. Pharm. Sci., 1980, 69, 1135. Krause, R. D., and Lott, J. A., Clin. Chem., 1974, 20, 775. Lott, J. A., and Turner, K., Clin. Chem., 1975, 21, 1754. Dols, T. J., and Armbrecht, B. H., J. Assoc. Off. Anal. Chem., 1976, 59, 1204. Borman, S. A., Anal. Chem., 1985, 57, 983A. Detaevernier, M. R., Michotte, Y . , Buydens, L., Derde, M. P., De Smet, M., Kaufman, L., Musch, G., Smeyers-Verbeke, J., Thielmans, A., Dryon, L., and Massart, D. L., J . Pharm. Biomed. Anal., 1986, 4, 297. Musch, G., De Smet, M., and Massart, D. L., J. Chromatogr., 1985, 348, 97. Karnicky, J., Anal. Chem., 1984, 56, 1312A. Strimaitis, J. R., and Hawk, G. L., Editors, “Advances in Laboratory Automation Robotics,” Zymark Corporation, Hopkinton, MA, 1985. Lochmuller, C. H., Lung, K. R., and Cushman, M. R., J. Chromatogr. Sci., 1985, 23, 429. Johnston, G., Lab. Scz. Tech., 1986,2(2), 22. “Chemometrics and Intelligent Laboratory Systems, ” Elsevier, Amsterdam. “Journal of Chemometrics, ” Wiley, Chichester. Paper A61256 Received July 31st, 1986 Accepted October 2nd, 1986

ISSN:0003-2654    

DOI:10.1039/AN9871200385

出版商:RSC    

年代:1987

数据来源: RSC

摘要:

ANALYST, APRIL 1987, VOL. 112 391 Elucidation of Olive Oil Classification by Chemometrics* Omar Eddib Quality Control Laboratory, Ministry of Health, Tripoli, Libya and Graham Nicklesst Department of Inorganic Chemistry, University of Bristol, Cantock’s Close, Bristol BS8 I TS, UK Olive oils from Libya, Tunisia and Turkey were analysed by gas chromatography using capillary columns (Silar 1OC - Si02-treated surface) with flame-ionisation detection. Data on Italian olive oils were abstracted from the literature. Both sets were stored and converted into a computer-compatible format. The data were statistically treated by FASCLUS and SCATER methods contained within the G3D procedures of the SAS computer software package. After suitable processing, canonical variables were statistically established from the data sets for the tested oils, leading to the reliable discrimination of virgin and refined oils.It appears that the G3D system is far better than PLOT representation at discrimination. Five unknowns were examined for their authenticity and were classified as expected. The fatty acid methyl ester database was convenient for chemometric techniques and is suitable for sample identification. Keywords: Olive oil; gas chromatography; chemometrics The availability of reliable but fast analytical procedures has progressed very markedly and, as a result, a flood of information can be generated from a series of samples including the examination of a large number of variables. However, it is important to produce as much information as possible so that the best and most reliable conclusions can be achieved.The analytical chemist must, therefore, devote much time to the difficult process of extracting the relevant information for the problem under study from a variety of sources. Computerised data analysis offers a practical means of interpreting such large data sets. Various multi-variate statis- tical and pattern recognition techniques have been developed for extracting chemical information from raw data. Chemometric techniques have been defined as the utilisa- tion of mathematical and statistical methods for handling, interpreting and predicting chemical data.1 The aim is possibly wider, to include: programming the measurements of more rational chemical information; showing that multi-variate techniques also allow the extraction of hidden information; and characterising a redundant information source that often completely masks the relevant information contained in the chemical composition.2 Among the many separation techniques, gas chromato- graphy (GC) is a well recognised method for the analysis of oils and fats.3?4 Data from such analyses are frequently used as indicators of quality, identity and authenticity.Accurate identification and quantification of the basic components of the samples can be useful in the detection of adulteration and counterfeiting.”7 The chromatographic profiles usually pro- vide the necessary degree of resolution of compounds of interest within a reasonable period of time. Information on changes in quality can conveniently be obtained by analysis of the chromatogram traces of samples.839 The term “chemomat- ographic profile” means that the over-all visual appearance of a chromatogram is the basic element from which the useful information is obtained.This is in contrast to data presenta- tion in the form of a table that may contain the same information but that could not be applied directly for the same purpose. For a limited number of samples, containing certain specific features, the human sensor is often a satisfactory * Presented at SAC 86, the 7th SAC International Conference on Analytical Chemistry, Bristol, UK, 20-26 July, 1986. t To whom correspondence should be addressed. recogniser. However, in situations involving a large number of samples and several variables, the resulting chromatograms are often complex to interpret.Hence several statistical packages have been used for chemometric evaluations, including the correlation of chromatographic profiles of olive oils,1OJ1 orange juices12 and coffees.13 Chemometric tech- niques have also been used to make a variety classification of wine8.9 and milk fats.14 Correlation studies using two-dimensional distributions (PLOT) of cluster analysis for olive oils of different harvesting years are now well established.15 Correlations using all three dimensions in a graphical format are not often carried out because of the difficulty of representation. The purpose of this work was to demonstrate a practical application of a statistical software package, the Statistical Analysis Systems (SAS), in particular the subroutine G3D (graphical three dimensions)l6 using capillary column GC data.Experimental Various olive oil samples were examined by capillary column GC using a flame-ionisation detector (FID). Attainment of the maximum separation of the various fatty acid methyl esters in a limited time led to the development of soda-glass capillary columns previously treated with colloidal silicic acid prepara- tions followed by coating with various polar phases, in particular Silar lOC.17 The data (Table l), processed by pattern recognition techniques, sought to discriminate between oils of different origins (different countries and various locations in the same country). Table 1. Olive oil samples used for cluster analysis Oil typekountry Year Code 1. Virgin oils .. . . . . 1983 c1 c 2 A1 A. Libya . . . . . . . . 1983 Lb 1984 Lb B. Tunisia . . . . . . 1983 T 1984 T 2. Refined oils C. Turkey . . . . . . 1983 R D. Italy . . . . . . . . It No. of samples 15 20 15 22 28 19 25 14 42392 ANALYST, APRIL 1987, VOL. 112 Samples were collected from the Libyan market. The products represent full growing harvests from Libya, Tunisia and Turkey. The Libyan oils included both virgin and refined oils. The virgin oils were divided into three types according to their geographical origin. Those oils originating in the area bounded by 11-14"E and between 30.5" and 32.5"N were called mountain oils. The name is appropriate because the majority of this area lies above 1000 m. The olive oils which originated from the coastal strip of Libya occur between 12.5" and 18"E.They were further subdivided into two types, those oils from 12.5" to 13.5"E were called Coastal 1 and those from 13.5" to 18"E Coastal 2. The height of this area is usually less than 300 m above sea lev 1. However, the Tunisian and oil that has been extracted from the seed without pre- treatment using a cold pressing system. Refined oil, on the other hand, is an oil that has been subjected to chemical treatment in order to improve its quality, e.g., acidity. All samples were packed in sample vials at the factory (Olive Oil Refinery, Tripoli, Libya). The Italian olive oil data were obtained for fatty acid methyl esters from the literature.2 Methyl esters of the samples were prepared by esterification using the boron trifluoride - methanol procedure.18 The esters were examined using a Carlo Erba Fractovap Series 220 gas chromatograph equipped with a capillary inlet system and an FID detector. A single soda-glass capillary column was installed in the capillary inlet system by means of a Vaspel ferrule. A 20 m x 0.30 mm i.d. column coated with ca. 0.25 pm Silar 1OC (Phase Separations) on a colloidal Si02-treated surface was employed. The procedure for column preparation is described elsewhere. 17 Additional conditions are listed in Table 2. The Trivector 2000 data system included a Trilab computer, a Type 2002 VDU terminal with full graphics, 160K RAM, twin floppy disk drive system and a Type 2012 plotter - printer with four-channel connection units and associated software. Turkish samples were only 0.f refined oils. Virgin olive oil is an Table 2. GC conditions Samples . . . . . . Instrument . . . . . . Samplesize. . . . . . Column . . . . . . Temperature programme Detector . . . . . . Carriergas . . . . . . . . 193 olive oils . . . . . . . . 7 "C min-1 . . FID at 300°C . . H2 at 1 ml min-l Carlo Erba Fractovap Series 220 1 pl at a splitting ratio of 1 : 60 20 m x 0.30 mm i.d. Silar 1OC - SiOz 70 "C (5 min)-240 "C (5 min) at 30 15 0 30 15 0 30 15 0 Ti me/m i n Fig. 1. Separation of olive oil samples using a soda-glass column treated with silicic acid and coated with Silar 1OC. (a) Libyan oil; (b) Turkish oil; and (c) Tunisian oil. 1 = 16 : 0; 2 = 16 : 1; 3 = 17 : 0; 4 = 18:O; 5 = 18: 1; 6 = 18:2; 7 = 18:3; and 8 = 20:O The standard basic system allowed for data collection and storage via an asynchronous communication interface compat- ible with the FID signal output, and then stored on the disk provided by the manufacturer.For quantitative analysis of the methyl esters prepared by the BF3 - MeOH procedure the percentage by mass recovered was calculated from the peak area of the sample measured by the Trivector 2000 database system. Peak-area correction factors were required when the detector response was different for each component. A calibration graph was constructed using known concen- trations for each pure standard compound. The correction factors were calculated by comparing the response of each component to a reference. In any analysis run, the area of each component was multiplied by its correction factor value to determine the real concentration.For this purpose, hepta- decenoic acid (17 : 0) was added to each sample as an internal standard. The procedure is laborious and time consuming. The data analysis of the chromatogram of the GC - FID output yielded seven variables for each run. The full set of features were identified as palmitic (16 : 0) , palmitoleic (16 : 1) , stearic (18 : 0) , oleic (18 : 1) , linoleic (18 : 2) , linolenic (18 : 3) and arachidic (20 : 0) acids. The data sets were stored on magnetic disk in a form compatible with the SAS chemometric package compatible with the Statistical Analysis System (SAS) software system, Version 5.03, which was run on a VAX minicomputer operating under the VMS system (SAS Software, Medmen- ham, Marlow, Bucks SL7 2EB).A variety of pre-processing, classification and display methods are available. Various scaling, feature selection, discrimination and display methods were used to determine the potential of the data. The application of the SAS package using a G~D/SCATTER procedure improved the separating power for similar samples owing to the rotation and tilt abilities of the program. Hence different angles could be examined for the best clustering presentation. Rotation involves the application of a non- singular transformation to the common factors to turn the picture in the horizontal ( x - y) plane to aid interpretation, and tilt specifies the angle at which the graphical representa- tion may be tilted about the y-axis also to aid interpretation. To examine the separation power of the system, samples of known origin were added to the file, which had previously been used to provide a good clustering distribution of the tested oils.These samples were identified as unknowns (1-5). The results show an exceptional classification of the imposed oils corresponding to their original source. Results and Discussion Typical chromatograms of the Libyan, Tunisian and Turkish refined olive oils are shown in Fig. 1, which depicts the major components of the analysed samples. Although the oils examined showed some differences in their features and the levels present, a fairly large number of oils needed to be examined in order to determine whether the differences were statistically significant, or simply reflected variations between individual refining processes amongst the different batches of oils.To examine this aspect, a series of virgin oils were collected from the coastal and mountain regions of Libya. The samples were analysed using the same analytical procedure as for refined Libyan oils. The SCATTER procedure that is included in the G3D program was applied to display the data sets as class means on canonical variables. The variables selected to perform clustering were palmitic (16 : 0), palmitoleic (16 : 1), stearic (18 : 0) and linoleic (18 : 3) acids. The procedure also allowed the selection of the best of best canonical variables according to their separation power. No classification of raw data was produced and discrimination of the data appeared after the necessary statistical steps, which ended with the display of the canonical readout.ANALYST, APRIL 1987, VOL. 112 Fig.2. shows the classification of virgin Libyan olive oils using CAN1, CAN2 and CAN3. CANl is the best canonical variable, being a linear combination of the quantitative variables that summarise the between-class variation (in absolute values). CAN2 is the next highest canonical variable. It appears that the same three factors were chosen in each situation. The results indicate the clear discrimination between the coastal and mountain regions. The efficient separation power of the system was evident because the coastal region was also distributed according to the geography of the Libyan coastal territories. The samples were clearly divided into three separate clusters.The FASCLUS procedure, on the other hand, provided the distribution table for the oils examined. The classification was based on the pseudo-F statistics, R-squared and cubic clustering criteria. As the total number of clusters required was specified in the file, the frequency percentages must be within the limit of the specified frequency number (ten). A typical pattern recognition using the FASCLUS procedure is given in Table 3. Fig. 2 also illustrates the performance of the G3D - SAS system in which the coastal samples were separated into two clusters, yet some areas of overlap are still evident. No mis-classification was observed when the coastal samples were compared with the mountain region series. The corresponding data obtained for the Italian samples were processed in an identical manner to the Libyan oils.The specified number of clusters was four; the SCAITER method of data distribution is shown in Fig. 3. As expected, the greater the number of similar samples encountered in the analysis stages, the less efficient was the discrimination ability. However, the standard of the SAS system is still sufficient to allow a complete discrimination between the different types of oils. The results demonstrate that CAN3 affords the critical 2.54 - 3. 28 Kc_% 51 -4.88 Fig. 2. Discrimination between Libyan virgin olive oil samples. Plot of canonical discriminant functions using fatty acid content (as methyl esters) as basic data. Locations: pyramid = eastern coastal zone of Libya (coastal 1); diamond = western coastal zone of Libya (coastal 2); and cube = mountain zone of Libya 393 separation between Umbrian and Apulian oils and between Calabrian and Sassarian oils, whereas CANl and CAN2 show a good distribution of Umbrian and Apulian oils on the one hand and Calabrian and Sassarian oils on the other.FASCLUS confirmed the classification ability of the data (Table 4). The results allow a direct comparison of the data for the Italian oils, which is considered to be satisfactory and is in agreement with the SCA'ITER procedure results. Libyan, Tunisian, Turkish and Italian samples of olive oils were also studied using the same system. The results given in Fig. 4 show considerable inter-species discrimination, but this is not unexpected as the dimensionalities of the examined sets were high.Fig. 4 refers to refined oils from the various countries concerned. The data probably have a sufficient discriminatory power to show sub-clusters indicative of area type within country type where the oil samples are sufficiently different. -8. 9e 8. 7 -2.00 aL -5.91 Fig. 3. Discrimination between olive oil samples from four zones in Italy. Plot of canonical discriminant functions using fatty acid (as methyl esters) as basic data. Locations: pyramid = Apulia; heart = Sassari (Sardinia); cube = Calabria; and diamond = Umbria 1.93 -1.54 a. -5.00 %. 6 4 -6.20 Fig. 4. Discrimination between refined olive oil samples from four different countries. Plot of canonical discriminant functions using fatty acid (methyl esters) as basic data. Pyramid = Turkey; heart = Libya; cube = Italy; and diamond = Tunisia ~ Table 3.Classification of Libyan olive oils using the FASCLUS procedure (based on a 2D plot) Species Coastal Coastal Mountain Cluster No. 1 2 Total 1 Cases present 0 3 23 26 Cases as a percentage 0.00 3.19 24.47 27.66 Percentages in the row 0.00 11.54 88.46 Percentages in the column 0.00 7.32 100.00 Percentages in the row 12.20 87.80 0.00 Percentages in the column 16.67 87.80 0.00 Percentages in the row 92.59 7.41 0.00 Percentages in the column 83.33 4.88 0.00 2 Cases present 5 36 0 41 Cases as a percentage 5.32 38.30 0.00 43.62 3 Cases present 25 2 0 27 Cases as a percentage 26.60 2.13 0.00 28.72 Total . . . . . . . . . . . . . . . . . . 30 41 23 94 31.91 43.62 24.47 100.00394 ANALYST, APRIL 1987, VOL.112 Table 4. Classification of Italian olive oils using the FASCLUS procedure Species Cluster No. Apulia Calabria Sassari Umbria Total 1 Cases present 25 4 0 36 65 Cases as a percentage 11.06 1.77 0.00 15.93 28.76 Percentages in the row 38.46 6.15 0.00 55.38 2 Cases present 28 1 0 0 29 Cases as a percentage 12.39 0.44 0.00 0.00 12.83 Percentages in the column 44.64 4.08 0.00 100.00 Percentages in the row 96.55 3.45 0.00 0.00 Percentages in the column 50.00 1.02 0.00 0.00 Percentages in the row 3.13 96.88 0.00 0.00 Percentages in the column 5.36 94.90 0.00 0.00 3 Cases present 3 93 0 0 96 Cases as a percentage 1.33 41.15 0.00 0.00 42.48 4 Total . . . Cases present Cases as a percentage Percentages in the row Percentages in the column . . . . . . . . . . . . .. 0 0 36 0 36 0.00 0.00 15.93 0.00 15.93 0.00 0.00 100.00 0.00 0.00 0.00 100.00 0.00 56 98 36 36 226 24.78 43.36 15.93 15.93 100.00 Table 5. Classification of unknown olive oil samples using the FASCLUS procedure Species Cluster No. 1 2 3 4 Total Cases present Cases as a percentage Percentages in the row Percentages in the column Cases present Cases as a percentage Percentages in the row Percentages in the column Cases present Cases as a percentage Percentages in the row Percentages in the column Cases present Cases as a percentage Percentages in the row Percentages in the column . . . . . . . . . . . Italian 0 0.00 0.00 0.00 0 0.00 0.00 0.00 0 0.00 0.00 0.00 45 23.08 100.00 100.00 . 45 23.08 Libyan 0 0.00 0.00 0.00 43 22.05 71.67 100.00 0 0.00 0.00 0.00 0 0.00 0.00 0.00 43 22.05 Tunisian 0 0.00 0.00 0.00 0 0.00 0.00 0.00 42 21.54 55.26 100.00 0 0.00 0.00 0.00 42 21.54 Unknown Unknown Unknown Unknown Unknown Turkish 1 2 3 4 5 0 0 0 0 0 0 7.18 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 0.00 0.00 0 0 0 0 0 17 0.00 0.00 0.00 0.00 0.00 8.72 0.00 0.00 0.00 0.00 0.00 28.35 0.00 0.00 0.00 0.00 0.00 100.00 0 10 8 10 6 0 0.00 5.13 4.10 5.13 3.08 0.00 0.00 13.16 10.53 13.16 7.89 0.00 0.00 100.00 100.00 100.00 100.00 0.00 0 0 0 0 0 0 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 14 10 8 10 6 17 7.18 5.13 4.10 5.13 8.72 8.72 ~~~ Total 14 7.18 60 30.77 76 38.97 45 23.08 195 100.00 For example, the Italian data are clearly split into two sub-clusters, but less discrimination can be seen for the Libyan oils.The application of the fatty acid methyl ester data as variables seems to provide a reasonable indication of the original source of the oil, not only for a different region but also for the same location. The quality of the data has a significant influence on the discrimination ability. The involvement of G3D improved the possibility of extracting the hidden characteristic of the variables used, which was less evident when only using the two-dimensional displays. An attempt was made to identify five unknowns. About ten samples of each group were analysed by the same GC procedures, and the data were included in the same file as the known samples (Libyan, Tunisian, Turkish and Italian; see Fig. 4). The classification results are gven in Fig.5 . The distribution of the unknowns was as expected. Unknowns 1-4 were selected from Tunisian sources, whereas unknown 5 was collected from Libyan oil. All specimens were pooled to their cluster indicating their origins. Chemometrics and GC - FID show clearly the ability to discriminate between similar or different origins. The present 4.95 2. 4s net -2.44 7. Fig. 5. Discrimination between refined olive oils of known origin from four countries and five unknown oils. Cylinder = Turkey; cube = Italy; pyramid = Libya; square = Tunisia; heart = unknown 1; club = unknown 2; diamond = unknown 3; spade = unknown 4; and circle = unknown 5. Each symbol for the unknowns represents a separate determination method provided a good means for the identification and the authenticity of the tested unknowns (see Table 5).To demonstrate the superiority of G3D over PLOT pro- cedures, a simple comparison was attempted and is shown in Figs. 4 and 6 for the four-country set.ANALYST, APRIL 1987, VOL. 112 395 m * I I I I a e * I I a I a a I a e ’ a r a s I a s a n a a a 1 a a m I a n s a a a I 8 m I 8 I a I I 1 a * I I Wl I I I O r I I 1 4 4 4444 I - 8 - 4 2 I 1 111 4 I 1 111 I 1 1 1 I I 1 1 I - 4 * I 111 nt I 11 I 2 a a s a s 4 . 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 u u 4 u 4 4 4 4 4 4 4 11 11 1 1 1 11 111 111111 1 al 1 2 1 - e * I I 1 I I * * ~------.---------.-------*--.---.-------*-----.--------*- 4 - a - l - 1 0 1 a a 4 1 VII Fig. 6. Discrimination between refined olive oil samples from four countries (to be compared with Fig.4). Plot of canonical discriminant function in two dimensions; the two best canonical variables were taken, i.e., CAN1 and CAN2.1 = Italy; 2 = Turkey; 3 = Tunisia; and 4 = Libya In conclusion, the application of the fatty acid methyl ester information to classify olive oils has been confirmed. The results obtained were utilised to identify some olive oil samples. The method is now being actively investigated to determine the adulteration of olive oils with other seed oils, e.g., safflower, at low levels of addition. The use of the SAS system with the G3D - SCATTER procedure has provided the extra hyperspace for the successful classification of a number of olive oil samples. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. References Bertsch, M., Mayfield, H.T., and Thomason, M. M., in Kaiser, R. E., Editor, “Proceedings of the Fourth International Symposium on Capillary Chromatography, Hindelang, FRG, 1981,” Huthig, Heidelberg, 1981, p. 313. Forina, M., and Lanteri, S . , in Kowalski, B. R., Editor, “Chemometrics, Mathematics and Statistics in Chemistry,” Reidel, Dordrecht, 1984, p. 305. Conacher, H. B. S . , J. Chromatogr. Sci., 1976, 14,405. Ening, M. G., Pallanch, L. A., Sampugna, J., and Keeney, M. , J. Am. Chem. Soc., 1983,60, 1788. Lifshitz, A., Stepak, Y., and Brown, M. B., J. Assoc. 08. Anal. Chem., 1974, 57, 1169. Shatski, T. F., and VanderCook, C. E., J. Assoc. 08. Anal. Chem., 1978, 61, 911. Prager, M. J., and Micklewica, M., 1. Assoc. 08. Anal. Chem., 1982, 65, 166. Kwan, W. O., and Kowalski, B. R., Anal. Chim. Acta, 1980, 122, 215. Van der Voet, H., and Doornbos, D. A., Anal. Chim. Acta, 1984, 161, 115. Forina, M., Armanino, C., Lanteri, S., and Tiscorina, E., in Martens, H., and Harris, J. M. Editors, “Food Research and Data Analysis,” Applied Science, Barking, 1983, p. 189. Kryger, L., Talanta, 1981,28,871. Gostechnick, G. F., and Zlatkis, A., J. Chromatogr., 1975, 106,73. Biggers, R. E., Hilton, J. J., and Gianturco, M. A., J . Chromatogr. Sci., 1969, 7 , 453. Massart-Leen, A. M., and Massart, D. L., Biochem. J . , 1981, 196, 611. Forina, M., and Armanino, C., Ann. Chim., 1982, 72, 127. Lee, J., and Hales, C., “The G3D Procedure in SAS/Graph User’s Guide, Version 5 Edition,” SAS, Gary, NC, 1985, pp. 399-412. Eddib, 0. A., Nickless, G., and Cooke, M., J. Chromatogr., 1986, 368, 370. Hensarling, T. R., and Jacks, T. J., J. Am. Oil Chem. Soc., 1983,60,743. Paper A61455 Received December Ist, 1986 Accepted December 19th, I986

ISSN:0003-2654    

DOI:10.1039/AN9871200391

出版商:RSC    

年代:1987

数据来源: RSC