Raman microscopy in art history and conservation science
Gregory D. Smith and Robin J.H. Clark Abstract
Numerous applications of Raman microscopy in art history and conservation science have appeared in the literature, but unfortunately this work has gone largely unrecognized by practitioners in those fields. This article assesses the causes of this situation and seeks to inform conservators, art historians and archaeologists of the role that Raman microscopy is playing in the analysis of historical materials. A brief description of the Raman scattering phenomenon and the instrumentation used to collect Raman spectra is presented, and the capabilities and limitations of the technique are discussed. Importantly, a comprehensive and critical review is provided of Raman microscopy applications in the technical analysis of art and artefacts. Introduction
Dispersive and Fourier transform (FT) Raman microscopy have become established analytical techniques for the identification and study of cultural materials over the past 15 years. Surprisingly, this development has occurred primarily in chemical and materials science departments in academia, rather than in the museum or archaeometry laboratory. Although recently increased interest among conservators, art historians and archaeometrists suggests that this situation may be changing, the widespread acceptance and routine application of Raman microscopy in the fields of art and archaeology have not yet occurred.
The absence of Raman microscopy in conservation and archaeometry laboratories is substantially influenced by two factors. First, the instrumentation is expensive (£150,000/ US$225,000), although this financial obstacle can be overcome through strategic budgeting if the desire to incorporate Raman microscopy is sufficient. The second, and perhaps more fundamental, reason is that conservation scientists are generally unfamiliar with the Raman technique and its widespread application in their fields, despite the existence of a large number of publications on the topic. It is this factor that will be dealt with in the present article.
The lack of awareness of Raman microscopy among conservation scientists is explicable. Although many articles on the topic have appeared in recent years, only a handful has been published in conservation journals, the balance having appeared in specialist spectroscopy and traditional chemistry publications. Because most of this work originates from chemical science departments in academia, there is pressure on the experimenters to justify their non-traditional research to a somewhat indifferent science community. Moreover, funding for such activities is difficult to obtain from traditional sources in the physical sciences. Therefore, scholars working
at the Arts-Science interface are required to defend their work as Science-based rather than Arts-based to both colleagues and funding institutions. This justification is manifested most often by publishing their results in chemical journals that are read and endorsed by the scientists' peers but, unfortunately, not by those in the art world. The end result is that, in most instances, the very practitioners who would benefit from it miss a wealth of valuable information. A better record exists for the appearance of Raman microscopy at recent conferences on conservation and archaeometry, but such meetings rarely leave more than a generally inaccessible book of abstracts.
This article seeks to rectify the problem described above by providing those charged with the care and study of our material culture with a basic understanding of Raman microscopy and its uses in their fields. The Raman scattering phenomenon is treated briefly so as to provide sufficient background information for the discussions that follow. Modern instruments for Raman microscopy are described. The many features that make Raman microscopy so well suited for the analysis of cultural materials are outlined, and the capabilities of the technique are discussed with comparisons to other analytical techniques more familiar to archaeological and conservation scientists, namely infra-red (IR) spectroscopy, optical and polarized light microscopy (PLM), ultraviolet-visible (UV/VIS) spectroscopy, laser-induced breakdown spectroscopy (LIBS), scanning electron microscopy with energy dispersive X-ray analysis (SEM/EDX), X-ray fluorescence (XRF), X-ray photoelectron spectroscopy (XPS), X-ray diffraction (XRD) and particle induced X-ray/γ-ray emission (PIXE/PIGE). Finally, the paper seeks to provide a resource for those interested in Raman microscopy as it applies to art and conservation studies by critically reviewing the relevant literature.
The Raman scattering phenomenon
The theory of Raman scattering is treated here briefly and a further explanation can be found in the general literature on the subject [1-3]. When light impinges upon matter, a small fraction of the incident radiation is scattered, the rest being reflected, absorbed or transmitted by the sample. Photon scattering can occur elastically, called Rayleigh scattering, or inelastically, called Raman scattering. Both phenomena involve the immediate annihilation of an incoming photon of light by the sample and the instantaneous emission of a second photon. In elastic scattering, the incoming photon generates an emitted photon of the same energy, whereas in inelastic scattering, the incoming photon generates a photon of either higher or lower energy. To understand the principles governing Raman scattering, one must examine the momentary interaction of the incoming radiation with the matter on which it is directed.
In Figure 1, the lowest energy levels represent the vibrational states of the ground electronic state of a diatomic molecule, and the upper cluster of lines represents the vibrational states of the molecule when in an excited electronic state. At room temperature, the majority of molecules exist in the lowest vibrational state, while a small proportion occupy higher vibrational states of the ground electronic state. An excitation beam of energy hv0can be considered to excite the molecule into a short-lived 'virtual' state. This higher energy state, caused by the momentary polarization of the electrons in the molecule by the electric field of the incoming photon, is described as 'virtual' because it does not have a fixed energy and is not therefore a true electronic excited state. The instability of the virtual state leads to the immediate emission of a photon as the molecule returns to a lower energy level.
The majority of the emitted photons are at the same energy, hv0, as the excitation beam and return the molecule to the ground vibrational state of the ground electronic state via Rayleigh scattering (Fig. la). However, a small proportion of the oscillating dipoles (approximately 1 in 10') emit photons of either higher or lower energy. Those emitted at the lower energy, h0-hvt, are called Stokes Raman photons (Fig. 1b), the energy difference hv1arising from energy lost from the incoming photon to promote the molecule into an excited vibrational level of the ground electronic state. Anti-Stokes Raman photons (Fig. 1c), appear at the same energy difference in relation to the excitation line, but on the high-energy side of the Rayleigh photons. The scattering centres responsible for these photons are those few molecules that were originally in the first or, less commonly, higher excited vibrational states of the ground electronic state. In these instances, the energy of the excited vibrational state is transferred to the scattered photon as the molecule relaxes from the virtual state to the ground vibrational state of the ground electronic state. The Boltzmann law predicts that the number of vibrationally excited molecules giving rise to anti-Stokes scattering is small at room temperature and so these Raman bands are rarely considered in the type of studies discussed here.
Raman scattering can also arise from excitation of a molecule to virtual states lying near to its first or higher excited electronic states . In fact, excitation directly into one of these electronic absorption bands can generate greatly enhanced Raman scattering in some molecules via a process known as resonance
Fig. 1 An energy level diagram showing the concerted excitation-relaxation phenomena responsible for (a) Rayleigh, (b) Stokes Raman and (c) anti-Stokes Raman scattering.
Raman scattering. It is impossible, however, to predict whether the effect will generate tangible results in strongly absorbing materials since absorption of both the excitation beam and the resulting Raman photons by the sample can negate the resonance enhancement and possibly lead to thermal degradation of the sample. A more detailed description of the effect is outside the scope of this discussion, although its value in the analysis of certain art pigments has been shown previously by a number of authors [5-8].
Figure 1 describes a single vibrational mode of a diatomic molecule. Other photons of potentially different energies, hvt, will be generated for a polyatomic molecule for each of its i normal vibrational modes meeting the selection rules for Raman scattering. By spectrally sorting the Raman scattered photons generated from these Raman-active modes, a highly characteristic vibrational spectrum can be generated for the molecule being examined. As an example, Figure 2 shows the Raman spectrum of vermilion (HgS) generated by a helium-neon (He-Ne) laser operating at 632.8 nm (-15,800 cm-1) and a power of 39 µW. The x-axis is labelled with both an absolute wavenumber scale (top) and the wavenumber shift (bottom) from the excitation line, or Rayleigh peak, whose residual intensity is seen at -15,800 cm-1 after extensive optical filtering. The wavenumber shift for a Raman band is constant, regardless of the excitation line used, and directly relates to the energy of the vibrational level being probed, hvt. It is therefore analogous to the wavenumber provided by IR absorption spectroscopy. Although the vibrational information offered by Raman spectroscopy is analogous to that of IR spectroscopy, it is not identical, but rather complementary, owing to the different selection rules governing Raman scattering (mandatory change in polarizability) and IR absorption (mandatory change in dipole moment).
Modern instrumentation for Raman and FT-Raman microscopy
A modern Raman microscope consists of four main components: an excitation source, an optical microscope, a monochromator or interferometer, and a photon detector .
Fig. 2 The Stokes and anti-Stokes Raman spectra of vermilion (HgS) excited by the 632.8 nm output of a He-Ne laser operating at 39 µW.
Figure 3 depicts a schematic drawing of these modules as combined in the latest line of dispersive Raman microscopes. Many of the earlier limitations of Raman spectroscopy, e.g. poor quality spectra and sample fluorescence, have been largely circumvented by improvements in instrumentation.
The excitation source for Raman spectroscopy is almost universally the monochromatic output of a laser. The frontal view in Figure 3 does not show the compact laser source because it is positioned behind and parallel to the spectrometer unit. The laser beam enters the spectrometer in the bottom right hand corner and is turned by an adjustable mirror to a beam expander, which decreases the power density of the beam so that no optical components are damaged. Like other forms of emission spectroscopy, the intensity of Raman scattering is directly related to the power of the excitation source. However, the highly efficient optical components and sensitive photon detectors available in modern Raman microscopes allow the use of relatively inexpensive, low-power lasers ; the excitation power at the sample is seldom over 5 mW when using visible lasers, a situation that virtually removes all risk of laser-induced degradation of the sample. The excitation line can be of any energy so long as this is above that of the manifold of vibrational levels of the ground electronic state of the material to be analysed. There are, however, several practical considerations when selecting a laser for performing a Raman analysis, and a large selection of lasing lines is desirable when examining a wide range of materials.
The scattering efficiency of a molecule is proportional to the fourth power of the frequency of the scattered photon (v4 dependence), indicating that, other things being equal, shorter wavelength lasers generate the more intense Raman scattering. Unfortunately, selecting an excitation wavelength is not always simple; for instance, coloured materials are best analysed using laser lines of similar colour, thereby limiting the absorption of the laser light, a process that competes with Raman scattering and could, in unskilled hands, lead to sample degradation. Also, the choice of excitation wavelength can be influenced by problematic electronic transitions in the sample, notably fluorescence, which gives rise to red-shifted, broadband emission that is usually thousands of times more intense than Raman scattering. Therefore, if the sample is fluorescent, the Raman spectrum is often swamped by this more intense signal.
Sample fluorescence can be overcome by collecting the higher energy anti-Stokes Raman bands, although this is normally
Fig. 3 A schematic drawing of a modern Raman spectrometer (Renishaw System 1000). The total dimensions of the instrument are 168 cm x 65 cm x 75 cm while the footprint is 100 cm x 65 cm.
not practical due to their low intensities. More commonly, a long wavelength laser, often one in the NIR region, such as a neodymium-doped yttrium-aluminium-garnet laser (Nd 3+:YAG, 1064 nm), is employed as an excitation source. The energy of the Nd3+:YAG laser is too low to excite fluorescence transitions in most samples. However, because NIR laser beams have low Raman scattering efficiency (v4 dependence) and fall outside the sensitive region of Raman detectors, they are normally used in an FT-Raman configuration with pyroelectric detectors so as to benefit from all the advantages of interferometry over dispersive spectroscopy (see below) . A recent development is the availability of long wavelength diode lasers (e.g. the gallium-aluminium-arsenide (GaAlAs) diode laser, 780 nm) whose emission falls within the sensitive region of charge coupled device (CCD) detectors. Using these lasers, the problem of fluorescence can be circumvented without requiring interferometric systems and with increased Raman scattering efficiency in comparison to the Nd3+:YAG laser.
The excitation beam is next passed through an optical microscope where it can be focused to an area of ≤1 µm2. The Raman scattered photons that are generated at the focus are back-collected by the same microscope objective and directed to the spectrometer. The ability of the microscope to focus the radiation is ultimately constrained by the diffraction limit imposed by the excitation wavelength. This can be a limitation when using NIR lasers in FT-Raman microscopy. In these instances, the focus diameter is large, often -20 µm, thereby sacrificing one of the most important advantages of Raman microscopy, that of high spatial resolution. The excitation depth similarly varies with different laser wavelengths. In non-resonance conditions, the sampling depth is normally of the order of the laser wavelength being used, and in resonance or near-resonance conditions, it can be much less than this. Narrowing of the spectrometer slits (see Fig. 3), coupled with a reduction in the active detector area, can allow for tight control over the sampled depth via a confocal effect . In this arrangement, only the Raman scattering generated at a very narrow slice of the laser focus is able to pass through the slit and be properly focused onto the detector. Such spatial discrimination has been used to profile stratified materials, such as crossed ink lines  or paint and glaze layers in ceramics , and can also limit fluorescence interference from material above or below the focal point.
A monochromator or interferometer is next used to sort spectrally the Raman scattered photons after first rejecting the reflected and elastically scattered laser light. Because of their much larger relative intensities, the laser and Rayleigh photons must be removed in order to avoid saturating the detector system. In first-generation dispersive instruments, many of which are still being used to great effect, a fore-monochromator is used to reject the interfering radiation before sending the Raman photons to a second and even a third monochromator for spatial sorting. Although extremely large and slow, these instruments are normally very efficient at eliminating Rayleigh scattering, and spectral data often can be measured to within 10 cm-1 of the Rayleigh line. Portable instruments, such as that shown in Figure 3, make use of small, narrow-band holographic notch filters to reject the Rayleigh scattering. Although considerably smaller and faster, these instruments currently suffer because spectral data cannot be obtained within 50 cm-1 of the laser line. Furthermore, a separate and expensive notch filter assembly is required for each exciting line. A recently announced near excitation tunable (NExT) notch filter should overcome both of these problems by allowing data collection for any excitation wavelength to within 10 cm ' of the Rayleigh line.
Following the Rayleigh rejection filter, a diffraction grating allows the spatial separation of the Raman scattered photons in a dispersive instrument (see Fig. 3). Spectral sorting in FT-Raman instruments, however, is achieved by an interferometer . These instruments often utilize the Michelson interferometer of an existing FTIR spectrometer. The high optical throughput, multiplex capability and high wavenumber accuracy of the interferometer are all retained in FT-Rarnan microscopy. However, an additional complication is introduced in that the inelastically scattered photons from a NIR laser source can fall within the absorption spectrum of water vapour, and so these instruments must be purged with dry air constantly. This requirement cannot be met when examining large objects under a microscope, and so small portions of the Raman spectrum are often inaccessible due to this interference. Moreover, the NIR Raman photons can also be absorbed by the sample via overtone and combination modes, possibly leading to thermal degradation of the sample or to spectral intensity losses.
After rejecting the Rayleigh photons and spatially or interferometrically sorting the Raman photons, the spectrum is recorded through the use of photovoltaic detectors in dispersive Raman microscopy or a pyroelectric detector in FT systems. The former can be either a single photomultiplier tube
that records discrete wavenumber intensities as the monochromator is scanned or an array of detector elements on which a region of the spectrum is dispersed. Two-dimensional detectors are widely used now that inexpensive, thermoelectrically cooled, and highly sensitive CCD cameras are available (see Fig. 3). Because current CCDs are insensitive in the NIR region of the spectrum, FT-Raman microscopy again makes use of existing FTIR instrumentation, germanium and indium-gallium-arsenide (InGaAs) detectors, for recording the interferogram generated as the interferometer is scanned. The need for liquid nitrogen cooling of some NIR detectors is an additional complication of the FT-Raman technique.
Advantages and limitations of Raman microscopy
The characteristics of the Raman microscope that make it so well qualified for the analysis of historical materials include its molecular specificity, speed, non-destructiveness, high spatial (< 1 µm) and spectral (< 1 cm-1) resolution, in situ analysis, applicability to unprepared samples of large or non-uniform shape, and relative immunity to interference [7, 12]. These features are all highly desirable in any analytical tool used to identify and study inhomogeneous and otherwise complex samples such as those represented by museum objects. Although Raman microscopy is not suitable for all forms of conservation and archaeological analysis (e.g. assaying of metal alloys,
Table 1 Strengths and weaknesses of analytical techniques commonly used in conservation science
Raman, visible laser Raman microscopy; IR, mid-infra-red reflectance microscopy; PLM, polarising light microscopy; UV/VIS, ultraviolet/visible reflectance spectroscopy or fibre optic reflectance spectroscopy (FORS); LIBS, laser-induced breakdown spectroscopy; SEM/EDX, scanning electron microscopy with Be-windowed energy dispersive X-ray detection; XRF, X-ray fluorescence spectroscopy; XPS, X-ray photoelectron spectroscopy, also called electron spectroscopy for chemical analysis (ESCA); PIXE/PIGE, external beam proton-induced X-ray emission/proton-induced γ-ray emission; XRD, powder X-ray diffraction.
a Sensitivity can be excellent under resonance conditions.
b Only possible with crystalline materials.
c Only atoms with atomic number Z ≥ 14 (Si) without an evacuated sample chamber.
d X-ray escape depth varies from nm to mm dimensions depending on the material and the element being detected. Can lead
to loss of spatial resolution and to interference from sub-surface layers. e All atoms with Z ≥ 5 (B).
f Depth of sample is normally confined to several nanometres,
g All atoms with Z ≥ 1 l(Na). A proton beam (via PIXE) generates less bremsstrahlung background radiation than other X- ray techniques, providing greater sensitivity. PIXE can also be modified with different detectors to perform nuclear reaction analysis (NRA) and Rutherford back-scattering (RBS) studies,
h Environmental SEM can analyse small, intact artefacts.
isotopic analysis, elemental fingerprinting, etc.) it compares well with other analytical techniques commonly employed in the conservation or archaeometry laboratory for the identification of molecular solids.
Table 1 assesses the relative strengths and weaknesses of Raman microscopy and nine other spectroscopic and microscopic techniques with which the conservation scientist will be familiar. The evaluation of each instrument is necessarily approximate and is based on its typical configuration as found in conservation laboratories across the world, it being well understood that specialized operating conditions and accessories can be used to enhance the performance in each case. The information contained in Table 1 was gleaned from overviews of these analytical methods [13-16]. The comparisons in Table 1 make a strong argument for the role that Raman microscopy can play in the conservation laboratory, and several of the important strengths of the technique are highlighted below. It is worth noting, though, that the complexity of cultural materials inevitably requires the use of several analytical tools for a complete and confident analysis; no single instrument can answer all the questions that may arise in the laboratory. Particularly informative investigations of artefacts have resulted from the powerful combinations of Raman microscopy and total reflection XRF [17, 18], LIBS , PIXE  and FTIR reflection microscopy .
The Raman technique is shown in Table 1 to be highly effective in identifying unknown materials, both organic and inorganic. Molecular specificity is inherent in vibrational spectroscopy, since molecular group frequencies and crystal lattice modes lead to unique vibrational spectra for different materials, and even for materials that are compositionally identical but differ in either connectivity, e.g. realgar versus pararealgar (a and 7-As4S4, respectively) [22, 23], or in their crystal structure, e.g. massicot versus litharge (orthorhombic and tetragonal PbO, respectively) . Comparing the Raman spectrum of a sample to a library of spectra of possible materials usually yields an unambiguous match that leads to the identification of the otherwise unknown compound. Raman spectra databases for minerals [25-28], plant fibres , historical pigments [24, 30], modern synthetic pigments , enamel/glaze pigments , modern ballpoint inks , archaeologically significant gums  and waxes [34, 35], and artistic varnishes, resins, and binders [30, 36] are available in the literature, and databases of Raman spectra of thousands of organic and inorganic materials are offered commercially. Moreover, Raman spectra respond predictably to structural and environmental changes in a molecule in terms of the number of Raman bands observed, their wavenumbers, intensities and bandwidths. This makes the technique appropriate for the study of reactions or of physical changes in materials. Although the technique is well suited to qualitative study, quantitative data are more difficult to obtain due to their dependence on individual instrumental parameters; consequently, only a few researchers examining historical materials have attempted to obtain them [37, 38].
As seen in Table 1, several other techniques also provide superb molecular specificity. IR microscopy, as mentioned previously, generates data analogous to that of Raman spectroscopy. However, as recently as 1994 Mills and White could claim, 'The Raman effect is however a very weak one and good spectra are not easy to obtain. It seems unlikely to be able to compete with FTIR and the infra-red microscope' [39, p. 22]. Even then, Raman instrumentation was improving
rapidly, and today the high spectral quality, ease of use and speed of modern Raman spectrometers is comparable to, and often exceeds, that of IR microscopes. For example, each technique generates a characteristic vibrational spectrum for a material, but the low wavenumber region - critical for the identification of inorganic materials - is more easily studied by Raman microscopy. Furthermore, water, a ubiquitous component adsorbed to or structurally associated with archaeological and artistic materials, is a serious interference in IR spectroscopy but not in Raman spectroscopy. Despite these hindrances, IR microscopy is often capable of detecting thin organic layers that can often be highly fluorescent, such as pigment binders , a task that is typically more difficult for Raman microscopy. As such, IR microscopy is not likely to be replaced in the analysis of artists' materials, but it certainly no longer overshadows Raman spectroscopy.
Powder XRD, a superbly specific technique for identifying compounds based on their atomic structures, is only suitable for highly crystalline samples and is limited, except for the most recent and most expensive diffractometers, to the analysis of ex situ samples, normally material sampled from a small area on the surface of the artefact. By comparison, Raman microscopy can be used successfully to identify even amorphous materials and can be used in situ on large, non-uniformly shaped objects, such as statuary and codices. XRF has been widely used to identify in situ mineral pigments based on their elemental profiles, but it cannot detect elements lighter than silicon without the benefit of an evacuated sample chamber, nor can it distinguish between pigments with identical or even similar elemental compositions. The latter limitation might explain why the yellow mineral pararealgar had never been detected as a pigment on a manuscript illumination until Raman microscopy became widely used in this area ; XRF analysts would have assigned all yellow arsenic sulfide pigments to the commonly used mineral orpiment (As2S3) [22, 23]. Optical microscopy, perhaps the most utilized technique in conservation, requires a highly skilled operator and sometimes considerable sample preparation . Even then, it often fails to provide conclusive evidence for many of the important materials in archaeology and art, especially those that are organic. Although these structural, elemental and optical techniques have a well-deserved place in the conservation laboratory, many of the tasks of identification for which they are normally used can be done equally well and much more quickly by Raman spectroscopy, either as the principal technique or as a confirmatory method.
Raman microscopy offers high spatial resolution for in situ analyses. For complex mixtures, as in paintings where the mixing of several pigments generates the final colour, the ability to resolve spatially and identify pigment grains of different composition is highly desirable. Often unrecognized is the added benefit that high spatial resolution yields in removing interferences in Raman microscopy, such as fluorescence from binders or varnishes. The ability to focus the excitation beam on an isolated, exposed pigment grain effectively reduces the interfering signal from the associated material, often making an apparently hopeless analysis feasible.
Another popular attribute of the Raman microscope is its small size, and therefore its portability. Advances in Rayleigh rejection filters and laser miniaturization are leading to instruments that can be moved easily. This development is allowing more extensive collaboration between academic
research groups and museums that had been hesitant to loan materials to laboratories. The authors have on numerous occasions disassembled, transported and reassembled their research-grade Raman microscope in order to work on-site at London museums and libraries on objects either too large or too valuable to be brought to the laboratory. In one instance, a portable instrument was shipped from England to the United States in order to perform an analysis of the Vinland Map at Yale University . Portability and increasingly rugged instrument design suggest that Raman microscopy is likely to expand into field-based applications in conservation, archaeology and geology in the near future.
Raman microscopy in art analysis and conservation studies
The plenitude of material in art and archaeology to which Raman spectroscopy has contributed precludes a comprehensive treatment of both areas here. Rather, an attempt has been made to focus on those articles concerned with conservation science and the analysis of artists' materials while reserving those dealing with traditionally archaeological objects, e.g. glass and glazes, lithics, ceramics, cave art, etc., for coverage in a separate review. It is well recognized that the separation between these fields is tenuous and somewhat arbitrary, but it is unfortunately necessary in order to treat the material in any way other than that of providing a brief catalogue (see, for example, [42, 43]).
In this review, every effort has been made to cover all of the art and conservation research articles in English from those peer-reviewed journals that are widely available. An attempt is made to assess the work critically and to highlight the historical or scientific importance of the results. Those applications in which Raman spectroscopy was used without the benefit of a microscope have been included, it being understood that the research could have been performed as well or better with a microscope attachment had one been available. Similarly, applications using fibre optic sampling accessories are also reported. This area is recognized as being one that is developing rapidly and that promises to increase the applicability of Raman spectroscopy to extremely large or otherwise inaccessible objects [44, 45]. The review is separated into sections devoted to the analysis of a particular class of material or to a specific research goal. This division is intended to demonstrate the different ways in which Raman analysis is being used and to highlight the various instrumental concerns dictated by different materials.