X-RAY SPECTROMETRY X-Ray Spectrom. 29, 3���17 (2000) X-Ray Fluorescence Spectrometry in Art and Archaeology Michael Mantler1* and Manfred Schreiner2 1 Institute of Applied and Technical Physics, Vienna University of Technology, Wiedner Hauptstrasse 8���10, A-1040 Vienna, Austria 2 Institute of Chemistry, Academy of Fine Arts, Schillerplatz 3, A-1010 Vienna, Austria This paper presents examples of analyses by x-ray fluorescence (XRF) spectrometry in art and archaeology, including pigments in paint layers and illuminated manusripts, of iridescent glasses and of medieval coins. Theoretical aspects of information depths and shielding effects in layered materials are discussed. Element maps were experimentally obtained by a specially designed x-ray spectrometer (1 �� 1 mm pixel resolution) and by electron-excited XRF (electron microprobe). Copyright ��� 2000 John Wiley & Sons, Ltd. INTRODUCTION Art historians, archaeologists and conservators are con- stantly concerned with the questions of where, when and by whom an object was made. Stylistic considerations combined with aesthetic evaluations and comprehensive archive studies can usually provide answers. However, styles were sometimes copied at locations and times completely different from those of their origin, and then investigations of the physical properties and chemical composition of the artifacts are helpful and increasingly applied to allocate an object to a particular historic or prehistoric context, to determine the correctness of the claimed provenance or to explore the technology used for the manufacturing. For example, the deliberate alloying of Cu with Sn, As, Sb and Pb has varied greatly from region to region and from time to time and careful mate- rial analysis combined with such knowledge can be used to presume the geographic origin of an object or at least the origin of the materials of which it was made. Material analysis of our cultural heritage is almost as old as the scientific documentation of objects of art and archaeology. It was Martin H. Klaproth (1743���1817), Professor at the University in Berlin, who reported in 1795 the chemical composition of Roman coins, ancient alloys and glass1 ��� 4 based upon gravimetric analyses and newly developed chemical recipes for the separation of Cu, Pb and Sn, and their quantitative determination. For his studies large amounts of sample material, even whole small coins, had to be dissolved in nitric acid, a procedure which would nowadays not be accepted by the curators or conservators. At the beginning of the 20th century, microchemical techniques5 ��� 11 and spot tests were developed,12,13 which significantly reduced the amount of sample material nec- essary for the analysis. A specific advantage of the clas- sical microchemical analytical tests is that they provide * Correspondence to: M. Mantler, Institute of Applied and Technical Physics, Vienna University of Technology, Wiedner Haupstrasse 8���10, A-1040 Vienna, Austria. information on both inorganic and organic constituents. The greatest disadvantage of requiring a separate sample for each identification could be overcome by using sev- eral separation techniques.13 Therefore, it is not surprising that a number of museum laboratories and scientific labo- ratories specializing in the investigation of materials and techniques used for works of art were established in the first decades of the 20th century. The booming development of electronics in recent decades has brought new analytical instruments, which have opened new horizons with respect to the origin or authenticity, technical conception and preservation of works of art.14 ��� 16 Among the most commonly cited ana- lytical methods in the literature dealing with the investi- gation of inorganic materials such as pigments, glasses, ceramics, and metals are optical emission spectrometry (OES),11,17,18 atomic absorption spectrometry (AAS)17,19 and x-ray fluorescence (XRF) and x-ray diffractometry (XRD). For XRF all excitation methods are used includ- ing photons from x-ray tubes20,21 and synchrotrons, elec- trons [electron probe microanalysis (EPMA), scanning electron microscopy with energy-dispersive x-ray micro- analysis (SEM���EDX)]22,23 and protons [particle-induced x-ray emission (PIXE), particle-induced -ray emission (PIGE)].24 ��� 27 Additionally, neutron activation analysis (NAA) has provided a large number of data for archaeol- ogists interested not only in the chemical composition of objects but also in the geographical and temporal origin of various materials. Studying the ratio of specific isotopes (e.g. 206Pb/204Pb or 16O/18O) by mass spectrometry (MS) has, for example, enabled art historians to assign marbles to certain antique quarries.28,29 THE ROLE OF XRF XRF is a method for the qualitative and quantitative anal- ysis of chemical elements. It is in principle applicable to all elements except the first two (H and He) of the periodic system, thereby covering an energy region from about 50 eV to 100 keV. However, many light elements CCC 0049���8246/2000/010003���15 $17.50 Received 11 May 1999 Copyright ��� 2000 John Wiley & Sons, Ltd. Accepted 10 August 1999

4 M. MANTLER AND M. SCHREINER are difficult to measure and require advanced instrumenta- tion, which often limits practical work to atomic numbers above 13 (A1). On the other hand, owing to costs of equipment and increasingly difficult radiation protection measures at high photon energies, L-lines rather than K- lines are measured at higher atomic numbers (above 50, Sn), so that the measured fluorescent photon energies are generally between 1 and 25 keV and excitation (tube) voltages less than 60 kV. The energy range is not only an analytical characteristic, but also important to estimate the danger of possible radiation damage. Another aspect is radiation protection for personnel involved in field analy- ses or when using specialized equipment with open beams in laboratories. Neither the observed photon energies of fluorescent lines nor their intensities are noticeably affected by the chemical state of the analyzed atoms (except for very long wavelengths and very light elements) so that no destructive dissolution and/or atomization by a flame, arc, spark or plasma is required, as is generally the case in optical emission (and absorption) methods. In fact, not much specimen preparation is required at all, unless highly accurate quantitative analyses or homogenization is required. Homogenization may become an issue as many objects of art and archaeology are rather inhomogeneous by nature, and the question may arise of the extent to which the results from a limited analyzed volume represents the whole artifact. Such considerations are particularly important when corrosion layers, coatings or painted artifacts are analyzed, or when such layers absorb the radiation of an underlying analyzed material. Rarely ever does the analysis of one or even a few elements answer all questions. Often the analyzed element is hoped to represent a pigment or an impurity related to the origin or to provide some other indirect hint. The phase of interest is often an oxide within a matrix of other oxides including the same analyzed element(s), sometimes embedded in organic binders as in the case of paints, and an unambiguous identification and exact quantification of the material of interest is impossible without employing complementary analytical methods. Instrumentation plays an important role. Electron- excited XRF requires a small sample that sustains vacuum and has an electrically conductive surface the method is ideally suited to analyze, for example, cross-sections of paints or element distributions at surfaces. Total reflection XRF is an ultimate tool for trace analyses, but requires in most cases at least microsampling. Only the classical, photon-excited methods allow raw materials of almost any kind and shape to be analyzed (as long as the sample fits into a sample holder or an open beam instrument is used). Both energy-dispersive and wavelength-dispersive modes are employed, but an important difference between these two is the primary radiative intensity. This can differ by a factor of 10 and more and should be taken into account in the case of sensitive materials. Information depth When radiation from a x-ray tube penetrates a specimen, it is absorbed along its path. A major fraction of the energy of the absorbed photons is converted into analytically use- ful fluorescent photons of the various atoms, and some of them reach the surface of the specimen and the detection system. The observed intensity (more correctly, the num- ber of observed photons) is a function of the composition of the specimen and its thickness. A layer of thickness D D 1 .E/ sin 1 C .K��1/ sin 2 emits about 63.2% of the fluorescent intensity of an infinitely thick bulk material, and correspondingly more for thicker layers (Table 1). In this equation, .E/ and .K��1/ are the (total) absorption coefficients of the sam- ple for the incident photons (energy E) and observed fluorescent photons (assuming K��1), is the density and 1 and 2 are the angles of the primary and fluorescent beams to the specimen surface, respectively. Secondary excitation effects may alter the value of D slightly. In most cases, the experimental error in analyses of art objects exceeds 5%. It is then difficult to distinguish between a bulk material and a layer with a thickness of more than 3D and therefore reasonable to classify, arbitrarily, Dinf D 3D as the information depth. This is, however, not a generally agreed definition, and providing a value of D should be preferred. The absorption coefficients in the equation for D are a function of energy (E represents the energy of the tube photon and K��1 that of the fluorescent photons) and the composition of the sample with elements j: .E/ D X j cj j .E/ D ci i .E/ C cM M .E/ .K��1/ D X j cj j.K��1/ D ci i.K��1/ C cM M.K��1/ where the subscript i represents the analyzed element and M the matrix. It is obvious that any large value of one of the coefficients decreases the information depth. Note that the energy E in the case of polychromatic radiation is a mean value within the integration interval and is not trivial to determine. A software utility to compute .E/ for chemical formulae is available from the authors by E-mail to mmantler@xrm.atp.tuwien.ac.at (free of charge for educational and academic institutions). In some cases elements are analyzed where two ana- lytical lines of (very) different energies are available, in most cases a K- and an a L-line (e.g. Cd K�� and Cd L�� in cadmium yellow/red/orange or Ba K�� and Ba L�� in manganese blue). Because of the much higher absorption of the L-lines due to their lower energy, their information depth is much lower this can be sometimes used with advantage for the analysis of layered materials. Table 1. Relative intensities compared with bulk material as a func- tion of layer thickness Thickness % of bulk intensity D 63.21 2D 86.47 3D 95.02 4D 98.17 5D 99.33 Copyright ��� 2000 John Wiley & Sons, Ltd. X-Ray Spectrom. 29, 3���17 (2000)

XRF IN ART AND ARCHAEOLOGY 5 Table 2. Chemical composition of some pigments, usual mixing ratio with oil, absorbance (including binding medium) and density (without binding medium)30 31 Oil (g per 100 g D Density Pigment Formula pigment)a (��m)b (g cm 3 ) Lampblack, soot C 60���90 1 1.65 Black iron oxide FeO��Fe2O3 30���55 15 4.2���4.5 Cobalt blue CoAl2O4 40���75 22 3.25 Ultramarine blue Na6 ��� 8 Al6Si6O24S2 ��� 4 25���55 1 1.9���2.5 Manganese blue Ba3.MnO4/2��BaSO4 10���20 7 4.15 Terra di Siena Fe2O3 30���70 17 2.6���3.3 Umbra Fe2O3 C MnO2 25���70 19 2.5���3.2 Chromium oxide green Cr2O3 15���40 10 4.7 Vermilion HgS 12���30 6 8.0 White lead 2PbCO3��Pb(OH)2 C 9���25 8 6.81 4PbCO3��2Pb(OH)2��PbO Zinc white ZnO 14���45 14 5.5 Titanium white TiO2 16���36 11 3.1���3.3 Cadmium yellow CdS: Cd L��: 30���60 17 4.0 Cd K��: 123 a The lower number for the oil to pigment ratio is for a stiff paste and the higher number for a lighter bodied paint. An average value was used for computations. b Selected line: K�� of heaviest element in pigment, except L�� for Ba, Cd, Hg, Pb, Sn, or as indicated in the case of cadmium yellow. Computed for Rh target end-window tube, 40 kV, 63.5�� and 45�� angles of incident and observed radiation, respectively. A special situation occurs when the absorption edge of a matrix element happens to have an energy between the K�� and K�� lines of an element. The K��/K�� ratio of the element to be analyzed is then a sensitive function of the concentration of the absorbing matrix element. Information depth in paint layers Paints are made by mixing pigments with a liquid binding medium (e.g. drying oil). After applying a layer of paint, it is generally dried and eventually other layers are added as needed. A final coating of varnish often protects paintings. The thickness for the whole structure ranges from a few micrometers up to 1 mm or more. A list of some commonly used pigments and their chemical composition is given in Table 2. Varnish and binding media consist of organic com- pounds (such as vegetable oils, egg yolk, egg white and resins) and thereby predominantly of light elements, which cause only modest absorption for the fluorescent radiation from the heavier elements. For numerical computations the absorption coefficient can be roughly approximated by that of water (Fig. 1). Usually a pigment is analyzed by its heaviest ele- ment, but another selection may be advisable when other pigments containing the same element(s) must be dis- tinguished. As can be seen from Table 3, the heaviest elements in a pigment have usually the highest concen- tration. Computed information depths Dinf D 3D can be obtained from Table 2 and Fig. 2. These values vary con- siderably with the composition of the paint layer and the energy of the measured line. Because of the wide range of possible mixing ratios between pigment and oil and the varying densities of pigments and oils, the number obtained for the geometrical thickness of a layer may be rather uncertain. It should be noted that the drying pro- cess of an oil as a binding medium is a complex chemical Figure 1. Absorption coefficient of water as a function of the photon energy. This can be used to approximate the absorption by the binding medium (oil) or varnish. reaction with the ambient air, which is generally not asso- ciated with the evaporation of the medium, as for example in the case of animal glue and water. Nevertheless, some changes may take place in the density and thereby in the thickness of a layer. Absorption by the matrix and interelement effects Elements in the pigment and those in the medium con- tribute to matrix absorption. Secondary excitation by the binding medium is unlikely, because it consists mainly of light elements, but it is possible within a pigment and between elements of different pigments. As mentioned above, the mixing ratio of pigment and binding medium varies considerably. Naturally, lines with low photon ener- gies (from light elements or L- and M-lines) are more strongly absorbed by the matrix and remain at a low level even for high concentrations of the pigment. As an exam- ple, the theoretical dependence of the Pb L�� intensity from white lead on the concentration of the binding medium is shown in Fig. 3. Copyright ��� 2000 John Wiley & Sons, Ltd. X-Ray Spectrom. 29, 3���17 (2000)