Raman spectroscopic library of medieval pigments collected with five different wavelengths for investigation of illuminated manuscripts†

The full-text may be used and/or reproduced, and given to third parties in any format or medium, without prior permission or charge, for personal research or study, educational, or not-for-pro t purposes provided that: • a full bibliographic reference is made to the original source • a link is made to the metadata record in DRO • the full-text is not changed in any way The full-text must not be sold in any format or medium without the formal permission of the copyright holders. Please consult the full DRO policy for further details.


Introduction
The use of analytical techniques for the study of objects of historical and artistic interest has increased in the last thirty years, providing useful information about artists' techniques, the provenance of materials, the nature of degradation processes, 1 authentication and dating.The priceless nature of works of art has driven scientists to employ non-invasive and non-destructive techniques.This, coupled with the high insurance values of the articles, essentially prevents the analysis of objects in host laboratories: work has to be done within the conservation studios of the host institution.Over the past decade, this has been achieved with the advent of portable, instrumentation that can be taken to the host libraries or institutions for in situ investigations on the y.Among the different techniques that are routinely employed for the analysis of artefacts or manuscripts, micro-Raman spectroscopy has proven to be an extremely powerful, due to its portability, specicity, spatial resolution and non-contact, non-destructive nature.  Thest effective application of Raman spectroscopy on cultural heritage objects was on illuminated manuscripts to identify pigments, 32 and it has since been used to study a wide variety of materials including paper, 33 binder media, 34 inks, 35,36 glass, 5,9,10,18 ceramics and pottery, 7,8 gemstones, 20 stones and rocks from archaeological sites, 19,30 degradation products. 38][39] A few portable Raman spectrometers are now commercially available, but these should be used with extreme caution: in many examples, the power of the laser light source is much higher than we would advocate and may cause damage to the artefact.Furthermore, commercially available portable instruments usually have a single laser source, in rare cases two, which are not necessarily the best ones to investigate the wide variety of pigments that can be found on the same artefact, positive identications can oen be difficult. 26,31ndeed, pigments respond differently to laser irradiation according to their nature (dyes or pigments, colour, etc.).In certain cases, for example, where the laser wavelength matches or is close to the absorption bands of the pigment being analysed, the Resonance Raman effect can lead to excellent sensitivity, but, in some cases, some absorption can also lead to large interfering, luminescent background signals which hide the low intensity Raman signals.The selection of the most appropriate wavelength for the identication of pigments then is of great interest and importance.
Portable equipment means compromises have to be made in comparison with xed laboratory equipment.Portable systems tend to sacrice spectral resolution and sensitivity. 31This work provides an updated library of pigments' Raman spectra acquired using different laser wavelengths in order to supply the best spectrum possible for each pigment for comparison to the data collected in situ.The pigments have been chosen as representative of those used in illuminated manuscripts between Vth-XVIth centuries in Europe. 32,36, Expeental Instrumentation Two different Raman spectrometers have been used, equipped with different lasers.The rst was a Horiba Jobin Yvon LabRAM HR confocal Raman microscope, equipped with a Peltier-cooled CCD and 50Â LWD Leica objective.The instrument has four different laser sources available, 488 nm, 532 nm, 632.8 nm and 785 nm providing a diffraction limited laser spot from 1 to 2 mm diameter.Each laser source has different maximum power values that can be reduced using neutral density lter (100% 50%, 25%, 10%, 1%, 0.1% and 0.01%).Reported laser powers were measured aer the objective lens in the sample plane.A 600 l mm À1 grating was used for measurements using the 532 nm, 632.8 nm and 785 nm laser, and an 1800 l mm À1 grating used for the 488 nm laser.The minimum wavenumber for each laser wavelength, dictated by the edge lters in the spectrometer were 200 cm À1 , 120 cm À1 , 70 cm À1 , 100 cm À1 respectively for the 488 nm, 532 nm, 632.8 nm and 785 nm lasers.All the acquisition operations were controlled by Lab Spec 6-Horiba Scientic soware.
For excitation at 830 nm, a Renishaw InVIa micro-Raman spectrometer (Rutherford Appleton Laboratory, Harwell Campus, Didcot), with 830 nm laser source, a silicon CCD detector and Nikon L-PLAN SLWD 50Â/0.45.The resulting laser spot was circa 2 mm in diameter.The spectral range recorded was 70-1800 cm À1 , with a 1200 l mm À1 grating.The maximum laser power was 55 mW, and this could be reduced to lower levels (50%, 25%, 10%, 5%, 1%, 0.5%, 0.1% and 0.05%).Again, the laser power was recorded aer the objective in the sample plane.
A silicon standard sample was used as reference for calibration (520 cm À1 ).The time of acquisitions and number of accumulations was decided on a sample-by-sample basis.
To be able to compare spectra acquired with different devices, spectra were corrected for the instrumental response by comparison to the spectra obtained for a broad-band source. 68,69 stable white light source (HL-2000-CAL Ocean Optics), whose spectral distribution was known (and can be approximated by a black body radiator of 2939 K), was used to generate a correction curve.The lamp emission spectrum was provided as photons/(Dlt), where Dl is the bandwidth detected by the spectrometer, and t the time unit.It had to be multiplied by l 2 (where l is the emission wavelength), in order to obtain the spectrum in terms of photons/t Â D y of Raman shi.The calibration lamp spectrum was recorded for every wavelength laser using the same scan-conditions as used for the sample measurements and covering the same wavenumber range.Using this correction curve, all of the Raman spectra were corrected for instrument response.No background subtraction was performed, since one of the goals of this work was to provide a library that helps to decide which the best wavelength to investigate a certain pigment is, so any luminescent background that may detract from the signal or prove diagnostic was recorded.All the spectra have been normalized to a maximum intensity of 1 in the graphs (Fig. 1-30) available in the ESI (ESI †).In the ESI, † the raw ASCII data of the spectra can be found.Corrected for the instrument response.Thus, relative intensities of peaks may be obtained and compared directly.

Classication of the spectra
A classication of the quality of the spectra collected (Table 1) was made, in order to establish the optimum wavelength for investigating the presence of a specic pigment, and which wavelength provides the maximum number of positive identi-cations.To classify them, a criterion to value the quality of the spectrum had to be established and the signal to noise ratio was considered appropriate.Even though this parameter is meaningful if applied to a single peak or band it does not give an evaluation of the whole spectrum, since to identify a Raman spectrum a "nger printing" approach is oen used; analysing the most intense peak and then the following less intense ones to conrm or refute the identication.In this study only the signal to noise ratio of highest peak of every spectrum was used to evaluate the whole spectrum.However, spectra showing only one peak were not included because a single peak is insufficient to unequivocally identify a pigment. 70In spectroscopy, the signal to noise ratio, SNR, is dened as Fig. 1 Raman spectra of minium or red lead.

SNR ¼ I s
where I is the average intensity (net peak heightbackground) of the signal and s is its standard deviation. 71However, the noise is the result of different sources.So that where s s is the uncertainty of the measurements known as signal shot noise.s b is the noise due to the background, that includes uorescence of the sample and stray light.s d is the noise generated by the dark current of the detector and the s r is the readout noise caused by the conversion from the electronic signal to a digital value in the CCD camera (and subsequent transfer from the detector to the computer). 71The signal shot noise is the square root of the signal intensity according to the Poisson statistic, since the emission and detection of photons are random event. 72,73However, background and the dark noise are detected in the same way as the Raman signal and similarly they are equal to the square root of the background intensity and dark current 71 respectively.Finally, the read out noise is the standard deviation of the numerical value the electrons from the detector device are converted into when digitized. 71Thus, the overall signal to noise ratio is dened as To calculate the signal to noise ratio, the contribution of the dark noise and the read out noise were considered negligible, which is appropriate for the scientic grade CCD camera employed in the spectrometers, while the background noise was the result of the square root of the difference of intensities between the spectra before and aer background correction.A peak may be dened as at least 2 or 3 times the intensity of the noise. 70,72So that the spectra were classied as "very good" spectrum (++) when the SNR > 100; "good spectrum" (+) when 3 < SNR < 100; "spectrum not identiable" (AE) when SNR < 3 and/ or the spectrum presented only one peak; "no spectrum" (À) when no spectrum at all was recorded.

Materials
Pigments and inks investigated were both pigments and dyes, chosen in accordance with the literature, most commonly used in manuscripts between Vth-XVIth centuries, supplied by L.
Cornelissen & Son (London) and Kremer Pigmente GmbH & Co. KG (Aichstetten, Germany).Iron gall ink, Brazil wood and kermes were made following ancient recipes. 74Analysis of pure pigments using the 532 nm and 632.8 nm lasers was made by sampling through the wall of a glass vial containing the pigments.Indeed, the use of a confocal microscope allows collecting radiation coming only from the focal plane, so that there is no signal related to the glass. 72However, using the 785 nm excitation source the spectra presented a large background at around 1400 cm À1 (ref.75) so pellets of pigments were prepared to obtain Raman spectra without glass contribution.The measurements performed with 488 nm and 830 nm excitation were also run on pellets.They were prepared by pressing a mixture of the pigment and a 10% w/w of a wax binder, (BM-0002-1CEREOX® Licowax C Micropowder).To ensure the homogeneity of the samples the mixture of wax and pigment was shaken for 3 minutes with a frequency of 25 s À1 , and pellets were then formed using a hydraulic press, with 9 tonnes per surface pressure.The spectra collected do not show any signals attributable to the wax.

Results and discussion
A total of 32 pigments have been analysed using 5 different incident wavelength laser sources.The spectra are represented in Fig. 1-30 (in the ESI †).They are ordered by observed colour (red, purple, blue, yellow, white, green, black and inks).In Table 1 the positive identications are summarized.Tables 2 to 6 list the wavenumber of the main peaks detected, with references from previous works, [37][38][39][76][77][78][79][80][81] where spectra have been reported for similar conditions. Each tble refers to a single wavelength laser source.The rst column provides the highest observed peak in the measured spectral range for that wavelength.In the second column the other peaks, in decreasing order of intensity, can be found; in the other columns the name and the compounds of the pigment.In the last column, the values of power density are recorded.
To use the library: (1) Select the table pertinent to the laser wavelength used to carry out the measurements; (2) Look for the highest intensity peak in the rst column; (3) Check the other peaks in the second column.The third column provides the name of the pigment.

Red pigments
For convenience, the pigments can be divided into two major groups.The rst is the iron oxide compounds (haematite, red ochre and caput mortuum, which belongs also to the inks group as well), and the others (cinnabar, vermillion, minium/red lead).All the iron oxide (Fig. 2-4) compounds show the main peaks at around 223 cm À1 , between 290 cm À1 , around 407 cm À1 , but only the spectra acquired with 532 nm and the 632.8 nm excitation beams show an intense peak between 1316 cm À1 and 1323 cm À1 .The intensity enhancement is attributed to resonance effects since the absorption edge for Fe 2 O 3 is at 580 nm.Spectra acquired with 488 nm all have poor signals, because of absorption by the pigment.Unfortunately, the red ochre (Fig. 3), which is a mixture of iron oxides, clays and silica, is the one that provided lowest signal to noise ratio with all the laser sources compared to haematite and caput mortuum.The red ochre spectra are indeed affected by uorescence, likely related to the presence of a heterogeneous matrix.In the second group, minium (Fig. 1), or red lead, is a lead oxide, whose the main peak resulting from 632.8 nm excitation is at 121 cm À1 , due to the deformation of the O-Pb-O angle.This was not detectable with the other wavelengths because of the edge lters cutting off low wavenumbers or attenuating the signal, and in literature studies of manuscripts this band is oen omitted for this reason.However, contrary to the report by Burgio et al. 1 who reported sample damage when using 488 nm and 514.5 nm radiations, it was possible to detect the pigment thanks to a low power density at 0.39 mW mm À2 .Cinnabar and vermillion (Fig. 5 and 6) are the mineral and synthetic forms of mercury sulphide, both show indeed the same very strong peak at 251-255 cm À1 .They are both detectable with all the ve wavelengths, and only low laser power is  76,77 169, 76,84 430, 77 220, 77 266, 76,84 1491, 76,84 203sh, 532, 76,77 535sh, 76,77,84 350sh, 76,77 509, 77 1092, 76,84 1060, 76,77 720, 76,84 1459, 77 751, 77,84  required when working with 532 nm and 488 nm lasers to prevent saturation of the detector even at short acquisition times.Realgar is a photosensitive mineral 95 and its transformation to pararealgar occurred using the 532 nm and the 488 nm wavelength laser source.Unfortunately the minimum power level for both the congurations was sufficient for this transformation and good spectra were not obtained without damaging the sample (Fig. 7).

Purple pigments
Cochineal (Fig. 8), orcein (Fig. 9), brazil wood, kermes, purple madder (Fig. 10), alizarin crimson (Fig. 11) and alizarin purple (Fig. 12) are all organic compounds and, these spectra are prone to be affected by uorescence depending on the wavelength of the excitation source.The spectrum collected with the 830 nm and 632.8 nm of cochineal does not show any Raman bands.Orcein, purple madder and alizarin crimson were detectable with all the wavelengths, while alizarin purple did not provide any spectrum with the 532 nm (absorption at 530 nm) and 632.8 nm radiation.For kermes and brazil wood, no spectra could be obtained at any of the ve laser wavelengths.

Blue pigments
Indigo (Fig. 14) provides better spectra with the NIR and IR laser since it possesses a broad absorption band in the visible range. 85However, it is still possible to recognise the pigment, thanks to the peak at 1580 cm À1 circa and 545 cm À1 also with lower wavelengths, which are superimposed upon the weak uorescence from this material.Azurite (Fig. 15) yields a good spectrum with all the wavelengths except the 632.8 nm laser source, attributed to the strong absorption by the pigment at ca. 600 nm.When using the infrared sources at 785 and 830 nm, a very low energy (785 nm and 830 nm, 3.58 and 8.69 mW mm À2 ) was used: at higher levels absorption of the radiation and localised heating of the sample results in its degradation.When View Article Online the particles of azurite do not disperse the heat efficiently and the rate of heat inside the single grain is higher than the rate of heat outside, they thermo-degrade. 86Ultramarine (Fig. 16) can be identied by the strong band at 550 cm À1 using all excitation wavelengths.The absorption band at around 610 nm means that the 532 nm and 632 nm excitation sources, close to the electronic absorption wavelength, benet from strong resonance enhancement and yield a progression of bands due to the bending of S 3 À . 87Indeed, in the spectrum collected with the 488 nm shows clearer a band at 584 cm À1 result of the S 2 À vibration, which with the other sources appears only as a shoulder of the main peak at 550 cm À1 .

Brown pigment
The only brown pigment investigated was raw umber (Fig. 13).
It is a mixture of iron oxides and manganese oxides.Investigation at 830 nm shows bands at 292 cm À1 , 610 cm À1 , 225 cm À1 , 390 cm À1 and 725 cm À1 .Excitation with the green laser presents weak bands at 268 cm À1 , 214 cm À1 and 582 cm À1 .No satisfactory Raman spectra were obtained with the other laser excitation wavelengths, 488 nm, 632.8 nm, 785 nm or 830 nm.

Yellow pigments
Orpiment (Fig. 17) is easily detectable with all the excitation wavelengths, requires only low laser powers to detect, and it may indeed cause saturation of the detector under some conditions.The main chromophore of yellow ochre is limonite, an iron oxy-hydroxide, and this pigment shows the same main  peaks of the mineral at all wavelengths.Ochres come in a range of compositions, due to natural variance of the mineral and the Raman spectrum (Fig. 18) can therefore reect these differences when collecting spectra from real historical artefacts.Lead tin yellow type I (Fig. 19) and massicot (Fig. 20) are both easily identied with the different laser wavelengths, but extra care has to be taken since they are both lead compounds and careful management of the laser power is required to avoid photo degradation: Burgio records alteration at 10 mW with 514.5 nm excitation source. 1 In the eld, this simply requires using a reliable power meter prior to measurement to ensure one knows precisely the light power being delivered at the sample  surface and to work well below the damage thresholds.Lead tin yellow (Fig. 19) and massicot (Fig. 20) were detected with the 488 nm source at 3.9 mW mm À2 and with the 532 nm at 2.06 mW mm À2 for lead tin yellow type I and 0.82 mW mm À2 for massicot, without any degradation.No damage at the sample was noticed with the higher wavelengths and if low values of power (0.37 mW mm À2 with 830 nm laser) were used it was to avoid the detector saturation.The organic nature of gamboge (Fig. 21) causes uorescence in the spectra, especially with higher frequency sources.The 830 nm laser is the one that yields the best spectrum, with an intense peak at 1593 cm À1 .

White pigment
When suspecting the presence of white lead (Fig. 22) it is necessary to employ low laser power (0.39 mW mm À2 with 488 nm and 0.12 mW mm À2 with 830 nm) because like other lead compounds it can easily be photodegraded by the laser. 88,89he highest intensity peak is observed at 104 cm À1 that requires a good laser light rejection cut off lter.As it can be seen from  the spectra presented in Fig. 22 this band is clearly observed in our system when using the 632.8 nm excitation, where the notch lter has a shorter cut off.However, it remains observable using 830 nm excitation source, but lies on top of the sloping laser beam prole.

Green pigments
Verdigris (Fig. 23) and malachite (Fig. 24) both result in good Raman spectra for the shorter wavelengths and do not provide a spectrum with the 785 nm laser because of strong absorption of the light in this region. 84However, the spectra  obtained with the 830 nm do present peaks at 150 cm À1 , 179 cm À1 , 218 cm À1 , 269 cm À1 to identify the pigments unambiguously.

Black pigments and inks
Carbon black, ivory black, lamp black and bistre (Fig. 25, 26, 27 and 28 respectively) are all characterized by broad bands between 1300 cm À1 and 1600 cm À1 due to the amorphous carbon.It is not possible to distinguish one from the others  View Article Online using a Raman spectroscopy. 91In a previous work a band at 965 cm À1 is recorded for ivory black, but it is not present in these spectra.The 785 nm laser seemed to be the one that provided the less intense peaks for all these black pigments.The sepia also shows two broad bands (Fig. 29), but they are generated by melanin, the main constituent of the pigment. 92,93Iron gall ink presents a peak at circa 980 cm À1 in the spectra collected (Fig. 30) with the two NIR sources, then a broad band in between 560 and 570 cm À1 that can be observed in the 532 and 785 nm spectra, as well as one at about 1350 cm À1 .The spectrum acquired with the green laser  has also peaks at 1601 cm À1 and 1644 cm À1 .Historically better spectra have been obtained for this pigment, 94 however we were not able to reproduce these results at the lower power densities.Caput mortuum (Fig. 4) has been already considered among the iron oxide based red pigments.

Conclusions
This work seeks to provide an updated and useful reference handbook for identication by Raman spectroscopy of pigments of the middle ages meeting the needs of researchers working with portable equipment in situ and the increase in the availability of different wavelength lasers.Table 1 helps in choosing the best excitation wavelength to investigate expected pigments on a specic artefact.Indeed, the discussion of the spectra calls attention to the different phenomena that can occur according to the nature of the pigment and the wavelength used to investigate it, for example in considering the resonance Raman condition and how this effects relative peak intensities in comparison to normal Raman effect.The collection of spectra helps to compare the experimental data with references acquired with the same laser source, to unequivocally identify the pigment.The Tables 2-6 provide a practical guide, which in few steps allows the identication of pigments.Since the intensities of peaks vary according to the exciting source, the operator can search references recorded with the same wavelength used during the measurements.New pigment references collected with wavelengths not used in previous works are also provided.This work also concludes that it is evident that the Raman technique alone is not capable of fully characterizing all the pigments selected, especially the uorescent organic dyes that are presenting challenges to curators.This highlights the need to turn to other techniques that can provide complementary information.

Table 2
Characteristic peaks of Raman spectra of pigments acquired with a 488 nm excitation source.Broad bands are labelled with "br", shoulder bands are labelled with "sh" 1222 | Anal.Methods, 2018, 10, 1219-1236 This journal is © The Royal Society of Chemistry 2018

Table 3
Characteristic peaks of Raman spectra of pigments acquired with a 532 nm excitation source (a) range between 100 and 2500 cm À1 , (b) range between 108 and 2500 cm À1 , (c) range between 150-2500 cm À1 .Broad bands are labelled with "br", shoulder bands are labelled with "sh"