94 Volume 59, Number 1, 2005 APPLIED SPECTROSCOPY0003-7028 / 05 / 5901-0094$2.00 / 0q 2005 Society for Applied Spectroscopy

Optical Characterization of Varnish Films by SpectroscopicEllipsometry for Application in Artwork Conservation

KYRIAKI POLIKRETI, ANDREAS OTHONOS, andCONSTANTINOS CHRISTOFIDES*Photonics and Optoelectronics Research Laboratory, Department of Physics, University of Cyprus, P.O. Box 20537, Nicosia 1678,Cyprus

The specific aim of this paper is to measure the optical constants offresh varnish layers up to a thickness of 10 mm by spectroscopicellipsometry. It is the first time that this technique has been usedin artwork conservation and it may prove very promising due toits nondestructive character. Samples of fresh dammar varnish(natural resin) and Paraloid B72 (synthetic resin) applied on glassand carbon black acrylic paint were analyzed. Both varnishes wereconsidered as perfect dielectrics, and the real part of their refractiveindex was described by the Cauchy model: n(l) 5 A 1 B/l2 1 C/l4. The Cauchy coefficients for dammar varnish and Paraloid B72were then determined for layers of known thickness. The ellipso-metric data were fitted to a model, which includes a mixed varnish–air layer. The optical properties of this layer were calculated by theBruggeman effective medium approximation. In the case of carbonblack acrylic paint, another mixed layer (paint–varnish) was addedto the model. The results are very close to the values given in theliterature. Given the measurement reproducibility, the results showthat ellipsometry can discriminate between dammar varnish andParaloid B72. This is very important in artwork conservation stud-ies, because it has been done by time-consuming, destructive tech-niques up to now. Future work includes measurements of othertypes of natural and synthetic varnishes, in an attempt to introducea nondestructive method for picture varnish identification and ag-ing studies.

Index Headings: Spectroscopic ellipsometry; Refractive index; Pic-ture varnish; Artwork diagnostics.

INTRODUCTION

During the last decades, the research towards usingnondestructive techniques in artwork diagnostics and re-vealing the original ‘‘fresh’’ appearance of paintings con-centrates numerous attempts from various fields. One ofthe numerous problems is the identification of paintingmedia1,2 and the study of their aging.3,4

A typical painting consists of the substrate (canvas,wood, etc.), a first layer of a thick primer, followed byseveral paint layers and an outer varnish layer. Varnishforms a protective coating over the paint layer but alsohas an aesthetic function. Various varnish recipes are de-scribed in the literature, but the main constituent is al-ways a resin, natural or synthetic, boiled with a dryingoil or diluted in a volatile solvent.1 Natural resins areexcretions or secretions of certain plants but their com-position is not entirely known yet, except that they arebased on terpenoids.5

Varnish identification is a difficult task and can be doneonly by using destructive techniques, like application ofdifferent solvents,6 gas chromatography mass spectrom-

Received 30 December 2003; accepted 14 September 2004.* Author to whom correspondence should be sent.

etry,1 Fourier transform infrared spectroscopy,2 micro-Raman spectroscopy,7 and pyrolysis gas chromatographymass spectrometry.4 As for their optical properties, mostof the measurements reported in the literature on picturevarnish refractive indices were done on an Abbe Refrac-tometer (at 589.3 nm). According to these results, varnishoptical indices vary, with natural resins showing higherrefractive indices (1.525–1.539) than synthetic resins(1.467–1.487, except Laropal K80 with 1.529).8,9

Spectroscopic ellipsometry has been proved a veryconvenient and accurate, nondestructive technique for themeasurement of thicknesses and optical constants of thinfilm layers on solid surfaces.10,11 Although it has beenused in the characterization of thin polymer films used inchemical sensors,12 protective silane films for aluminumsurfaces,13 latex14 and polystyrene15 films, and generallyin several types of polymers and other organic coatings,it has never been applied to paint layers.

This work investigates the potential of spectroscopicellipsometry to characterize picture varnishes and dis-criminate between natural and synthetic varnishes non-destructively. In future stages, this project intends to useellipsometry in order to monitor varnish aging3–5 (i.e.,yellowing caused mainly by exposure to UV irradiation).This work is carried out under the working frame of theEuropean RTD project (IST) ‘‘Conservation-RestorationInnovation Systems for image capture and digital Ar-chiving to enhance Training Education and lifelongLearning’’ (CRISATEL). The ultimate project of CRIS-ATEL is to develop techniques and software to predictthe original non-aged appearance of the paintings (i.e.,when the varnish was still fresh).

EXPERIMENTAL

Materials. The varnishes used here are ‘‘oil varnish-es’’ of two types: Dammar varnish, consisting of 43%dammar gum in turpentine oil, and Paraloid B72 (50%in toluene). The final dry film consists, in the first case,only of dammar gum and in the second case, of an ethylmethacrylate (70%) and methyl acrylate (30%) copoly-mer.

The acrylic paint used in this work is Pebeo Extra-Fine Acrylic. The paint mixture recipe contains 20% car-bon black pigment, 40% binder (acrylic polymer emul-sion), 30.7% water, 4.5% thickeners, 3.5% plasticizer,0.9% dispersing agent, 0.2% ammonia, and 0.2% preser-vatives.

Varnishes are applied on glass or paint. N and K valuesfor the glass substrates are given in Fig. 1. N and K rep-resent the real and the imaginary parts of the complex

APPLIED SPECTROSCOPY 95

FIG. 1. N and K values for the glass substrate.

FIG. 2. Ellipsometric data for dammar varnish on glass (incidence an-gle of 758). Measured tanC and cosD values (circles) are compared tothe model (a) without and (b) with a surface layer. Fitting parametersare given in Table I.

refractive index n (n 5 N 1 iK). The data shown in Fig.1 were obtained by considering glass as a bulk material(at 708). The absorption coefficient of the glass substrateis zero. Reflection from the backside of the glass slideswas suppressed by abrading the backside surface.

For the paints used in this work, the binder-to-pigmentvolume ratio is greater than one, creating a continuousand water/liquid tight paint film, which prevents capil-larity or wetting phenomena. Consequently, the mixingof varnish and paint at their interface is only of the ‘‘me-chanical’’ type without any physicochemical or chemicalphenomena.

The application of the films was done with an auto-matic film applicator (filmograph). The filmograph is abar coater that applies a grooved film on a flat surfacewith a speed of 100 mm/s. The film becomes flat bygravity while drying. The samples were dried in a hori-zontal position at room temperature (24 8C) and 30–50%humidity for two weeks. Film thickness was measured byan electronic gauge (DeFelsko, PosiTector 6000) usingmagnetic and eddy current principles, with an accuracyof 60.5 mm.

Methods. The varnish ability to level roughened paintsurfaces and its refractive index are the most critical fac-tors in comparing the appearance of a painting to its un-varnished appearance.16 But in our case, we are interestedin comparing the appearance of a painting to the non-aged varnished appearance and the varnish refractive in-dex becomes a very important factor. Small changes inthickness (,2 nm) and refractive index (,0.005) can beeasily detected by spectroscopic ellipsometry. The tech-nique measures the change in polarization of a linearlypolarized, collimated beam, due to reflection from a sam-ple surface. The polarization change is described by twovalues, C and D, which are related to the ratio of Fresnelreflection coefficients, rp and r s for parallel (p) and per-pendicular (s) light components, respectively:10,11

priD5 tan C·e (1)

sr

For the ellipsometric measurements carried out in thiswork, a scanning spectroscopic ellipsometer (SOPRAmodel GES5, SOPRA, France) was used. All measure-ments were done at four angles of incidence, 60, 65, 70,and 758, and in the range from 300 to 800 nm. Most ofthe data analysis was done by WinElli Analysis and Sim-ulation Software (Version 3.04, SOPRA SA). The differ-

ent film-stack models used for varnish layers on differentsubstrates will be described in detail for each case in thefollowing. The Levenberg–Marquardt algorithm17 wasused for minimizing the difference between experimentaland model values and consequently calculating the filmparameters. The surface microroughness effect was ap-proximated by a thin film of an effective medium. TheBruggeman effective medium approximation (EMA) wasused in this work.

RESULTS AND DISCUSSION

Varnish on Glass. The ellipsometric parameters ofdammar varnish and Paraloid B72 applied on glass aregiven in Figs. 2 and 3, for thicknesses of approximately8 and 7 mm, respectively. The oscillations are interfer-ence figures due to the perfectly homogenous thicknessof the varnish layer.

Figure 4 shows the values of cosD for dammar varnishcompared to those for the glass substrate. Values above1.000 for cosD indicate low signal-to-noise ratio and areconsidered equal to 1.000. Therefore, our varnish samplescan be considered as perfect dielectrics. These resultsagree with the varnish transmittance measurements by Dela Rie,3 according to which the absorbance is practicallyzero even above 300 nm for non-aged varnishes. In thefollowing, we are going to neglect the imaginary part ofthe refractive index in the wavelength range of 300 to800 nm. Accordingly, the dispersion of the varnish re-fractive index can be described by the Cauchy parame-

96 Volume 59, Number 1, 2005

FIG. 3. Ellipsometric data for Paraloid B72 on glass (incidence angleof 758). Measured tanC and cosD values (circles) are compared to themodel (a) without and (b) with a surface layer. Fitting parameters aregiven in Table I.

FIG. 4. CosD values for dammar varnish (circles) and glass substrate(line). Values above 1.000 for cosD indicate low signal-to-noise ratioand are considered equal to 1.000.

FIG. 5. Structure models used in the analysis of (a) varnish on glass and (b) varnish on paint.

terization model (N(l) 5 A 1 B/l2 1 C/l4). This modelwill be used to fit our ellipsometric data.

It is well known that spectroscopic ellipsometry is verysensitive to film surface quality. In our case, the standarddeviation of the heights (s) and correlation length (T) ofthe varnish surface irregularities are expected to be lessthan the wavelength of light (s K l and T K sl). Thisis the typical case of ‘‘microroughness’’. The simplestmodel of a microrough surface is based on an equivalentthin film, consisting of a homogenous mixture of materialand voids. The optical properties «ef of this effective layer,according to the Bruggeman effective medium approxi-mation (EMA), which will be used here, are given by theformula:18

1 2 « « 2 «ef efp 1 (1 2 p) 5 01 1 2« « 1 2«ef ef

where « is the dielectric function of the material and p isthe volume fraction of voids.

Figure 5a shows a simple sketch of the model struc-ture. The real parts of the refractive index of glass overthe range of 300 to 800 nm were calculated by consid-ering it as a bulk material, and are given in Fig. 1.

The ellipsometric data of Figs. 2 and 3 were fitted tothe model by minimizing the error function:

2 2n C 2 C D 2 Dexp expth th2x 5 1O 1 2 1 2[ ]G Gi51 C Di i

where G values represent the error bars on each measureddata point. Minimization was done by the Levenberg–Marquardt algorithm, which shows fast convergence butgets trapped in local minima. In order to obtain a globalminimum we executed multiple runs (around 100) of thealgorithm, starting from randomly chosen initial param-eter values from reasonable intervals. The data for allfour incidence angles (60, 65, 70, and 758) were fittedsimultaneously.

Figures 2 and 3 show the best fitting of the data withand without the surface layer. The relevant Cauchy co-efficients are shown in Table I. Standard errors of theobtained parameters were estimated from an error ma-trix.17 It is evident that the introduction of the surface

APPLIED SPECTROSCOPY 97

TA

BL

EI.

Bes

tfi

tm

odel

par

amet

ers

for

the

varn

ish

film

s,as

sum

ing

zero

exti

nct

ion

coef

fici

ent:

A,

B,

and

Car

eth

eC

auch

yco

effi

cien

ts;

Ts

and

Tin

tar

eth

em

ixed

laye

rth

ick

nes

ses

for

the

surf

ace

and

inte

rfac

em

ixed

laye

rs;

Fs

isth

evo

idfr

acti

onfo

rth

esu

rfac

ela

yer;

and

Fin

tis

the

varn

ish

volu

me

frac

tion

for

the

varn

ish

–pai

nt

inte

rfac

ela

yer.

Var

nish

type

Sub

stra

teT

v(n

m)

EM

AA

B(m

m2 )

C(m

m4 )

Fs

(%)

Ts

(nm

)F

int

(%)

Tin

t(n

m)

Dam

mar

Dam

mar

Par

.B

72P

ar.

B72

Dam

mar

Par

.B

72

Gla

ssG

lass

Gla

ssG

lass

Acr

ylic

Acr

ylic

8176

62

8305

63

7463

69

7523

610

6869

630

7028

617

none

Sa

none

Sa

S1

Ib

S1

Ib

1.52

496

3E2

41.

5267

69E

24

1.46

256

7E2

41.

4733

66E

24

1.51

316

2E2

41.

4673

66E

24

0.00

536

1E2

40.

0051

61E

24

0.00

896

2E2

40.

0076

61E

24

0.00

566

3E2

40.

0048

62E

24

2.8E

25

69E

26

2.2E

25

68E

26

21E

25

60.

5E2

68.

3E2

66

1E2

76E

25

63E

25

7E2

56

4E2

5

—99

.16

0.1

—99

.06

0.1

98.9

60.

198

.86

0.1

—14

66

11—

342

617

294

65

158

69

— — — —29

610

326

6

— — — —11

326

3212

156

18

aS

urfa

cela

yer.

bS

urfa

cean

din

terf

ace

laye

rs.

FIG. 6. TanC (open circles) and cosD (black squares) for carbon blackacrylic covered by dammar varnish and Paraloid B72 (;6 mm). Thevalues for plain carbon black acrylic are given for comparison. Arrowsindicate corresponding axes.

FIG. 7. N and K values for carbon black acrylic.

layer in our model improves the fit. The effective mediumwas found to have a thickness of 146 nm in the case ofdammar and 342 nm in the case of Paraloid B72, and acomposition of 99% varnish and 1% air. The calculatedvalue for the real part of the refractive index at 589 nmis 1.539 for dammar and 1.494 for Paraloid B72. Thesevalues agree with the values given in the literature at thesame wavelength: 1.539 for dammar and 1.487 for Par-aloid B72.8,9

Varnish on Carbon Black Acrylic Paint. TanC andcosD for dammar varnish and Paraloid B72 on carbonblack acrylic paint (7 mm) are given in Fig. 6. The intenseoscillations occurring on paint are not as intense as thosein Figs. 2 and 3. As the varnish follows the outer unevensubstrate surface, its thickness fluctuates and causesdamping of the interference peaks.19

The model used for varnish on paint is given in Fig.5b. In this case, the surface of the substrate is also rough,so we include in the structure model another effectivemedium layer, which substitutes the varnish–paint inter-face layer.

The refractive indices for the carbon black acrylicpaint were measured for a layer 20 mm thick applied onglass, considering the paint as a bulk material, and aregiven in Fig. 7. The real part of the refractive index rang-es from 1.53 to 1.46 and the imaginary part from 0.03 to0.09. To our knowledge, the optical constants of picture-acrylic paints are not available in the literature, and wecould thus not cross-check our results. However, with theknowledge of the refractive index of the acrylic binderand the complex refractive index of the carbon black par-

98 Volume 59, Number 1, 2005

FIG. 8. Measured tanC values and fitting for dammar varnish and Par-aloid B72 (;7 mm) on carbon black acrylic. Fitting parameters aregiven in Table I.

FIG. 9. The shaded areas represent the confidence limits (95%) for themean dispersion curve obtained by ten measurements for varnish onpaint. The solid lines represent the values calculated for varnish on glass(Table I).

ticles, we can roughly estimate the expected values byusing the effective media theory. If we assume that thepowder used in our samples consists of amorphous car-bon particles like those used for solar absorbing paints,we expect a refractive index around 2 and an extinctioncoefficient around 0.25.20 The diameters of the carbonparticles in our samples vary between 40 and 50 nm,where absorption (and not scattering) dominates. As forthe acrylic polymer, we can consider a representative re-fractive index value around 1.48 (for ethyl and methylmethacrylates).21 For a mixed material, consisting of car-bon black (10%, N1 5 2, K1 5 0.25) and acrylic polymer(90%, N2 5 1.48, K2 5 0), the effective medium theorygives Neff 5 1.529 and Keff 5 0.022, which are close tothe measured values.

Figure 8 shows the best fitting of the data to the model(including the surface and the interface layer). The rele-vant Cauchy coefficients are shown in Table I. The fittingis not as good as in the case of varnish on glass mainlybecause the model does not take the thickness non-uni-formity into account. Besides, the errors in thickness cal-culation are much higher and the values in Cauchy co-efficients are lower. The volume fraction of varnish at thevarnish–paint interface was found to be around 30% andthe thickness of the interface layer was just over 1 mm.The average thickness of the interface layer was alsomeasured by scanning electron microscopy. Thin sectionsof the samples were examined and the thickness wasfound to be more than 1 mm and less than 1.5 mm.

As for the measurement reproducibility, it was testedby successive measurements on different areas of thesame sample. In the case of varnish or paint on glass theellipsometric data are almost identical, but in the case ofvarnish on paint, the error was calculated at 0.5% fortanC and at 0.5% for cosD. This corresponds to 1% inindex and includes the precision of the measurement to-gether with the variations in homogeneity and isotropyof the applied layer (paint and varnish). The relevant dis-persion curves are expected to lie in the shaded areas ofFig. 9. The values extracted from varnish on glass werealso plotted for comparison. It is evident that, even with-out using a variable thickness model, ellipsometry candiscriminate dammar varnish from Paraloid B72. All nat-ural varnishes show refractive indices higher than 1.50,while most of the synthetic varnishes (except LaropalK80) used in painting show lower values (at 589.3 nm),

which seem to have 1.49 as an upper limit.8,9 This meansthat ellipsometry can discriminate between natural andmost synthetic varnishes, which is very promising sinceit is the first time that a nondestructive technique hasproven effective in varnish discrimination. Varnish iden-tification, an absolutely necessary procedure for conser-vation planning, can be done today only by using destruc-tive and time-consuming techniques.

We have to note here that we chose to present here theresults for carbon black acrylic paint because for othermineral pigments used in painting, the effective mediumapproximation is not valid. For highly scattering titaniumoxide, for example (with typical particle sizes on the or-der of 1 mm), the bulk calculation used here is not enoughand additional work involving diffraction theories22 willbe presented in another paper.

CONCLUSION

The technique of spectroscopic ellipsometry was usedfor the first time on picture varnishes, giving very prom-ising results. The varnish refractive indices were calcu-lated for layers applied on glass and carbon black acrylicpaint. The calculated indices are very close to literaturevalues, proving that spectroscopic ellipsometry can beused for optical characterization of varnish layers appliedon paintings. The main advantage of the technique is itsnondestructive and no-contact character, which is of greatvalue to conservators and museum curators, who have toremove varnish samples in order to make optical char-acterization of varnish layers. Besides, spectroscopic el-lipsometry can be used as a tool for discriminating be-tween natural and synthetic varnishes, a procedure thatuntil now could only be done by using laborious andtime-consuming techniques.

These first results of ellipsometry applied to artworkwill be continued with measurements on other types ofpicture varnishes and substrates, together with varnishesaged by heat and ultraviolet irradiation.

ACKNOWLEDGMENTS

The support of the European Union Program CRISATEL (IST—1999—20163) is gratefully acknowledged. We are also indebted to Mr.Christophe de Deyne from PEBEO for the preparation of dammar andacrylic paint samples and the PosiTector measurement of thicknesses.

APPLIED SPECTROSCOPY 99

1. M. P. Colombini, F. Modugno, S. Giannarelli, R. Fuoco, and M.Matteini, Microchem. J. 67, 385 (2000).

2. M. T. Domenech-Carbo, A. Domenech-Carbo, J. V. Gimeno-Ade-lantado, and F. Bosch-Reig, Appl. Spectrosc. 55, 1590 (2001).

3. R. E. De la Rie, Stud. Conserv. 33, 53 (1988).4. D. Scalarone, M. Lazzari, and O. Chiantore, J. Anal. Appl. Pyrol.

64, 345 (2002).5. G. A. Van der Doelen, Molecular studies of fresh and aged triter-

penoid varnishes, Ph.D. Thesis, University of Amsterdam, Amster-dam, The Netherlands (1999).

6. A. Burnstock and T. Learner, Stud. Conserv. 37, 165 (1992).7. P. Vandenabeele, B. Wehling, L. Moens, H. Edwards, M. De Reu,

and G. Van Hooydonk, Anal. Chim. Acta 407, 261 (2000).8. R. E. De la Rie, Anal. Chem. 61, 1229A (1989).9. J. H. Townsend, ‘‘The refractive index of 19th-century paint media:

a preliminary study’’, in ICOM Committee for Conservation 10thTriennial Meeting, J. Bridgland, Ed. (International Council of Mu-seums Committee for Conservation, Paris, 1993), p. 586.

10. R. M. A. Azzam and N. M. Bashara, Ellipsometry and PolarizedLight (North Holland, Elsevier, Amsterdam, 1987).

11. H. G. Tompkins, A user’s guide to ellipsometry (Academic Press,San Diego, 1993).

12. K. Spaeth, G. Kraus, and G. Gauglitz, Fresenius’ J. Anal. Chem.357, 292 (1997).

13. A. Franquet, H. Terryn, P. Bertrand, and J. Vereecken, Surf. Inter-face Anal. 34, 25 (2002).

14. A. F. Routh and W. B. Russel, Ind. Eng. Chem. Res. 40, 4302(2001).

15. J. Bai, C. M. Snively, W. N. Delgass, and J. Lauterbach, Macro-molecules 34, 1214 (2001).

16. R. S. Berns and R. De la Rie, Stud. Conserv. 48, 73 (2003).17. W. H. Press, B. P. Flannery, S. A. Teukolsky, and W. T. Vetterling,

Numerical Recipes: The Art of Scientific Computing (CambridgeUniversity Press, Cambridge, 1988).

18. D. Franta and I. Ohlıdal, J. Mod. Opt. 45, 903 (1998).19. S. Pittal, P. G. Snyder, and N. J. Ianno, Thin Solid Films 233, 286

(1993).20. N. Etherden, T. Tesfamichael, G. A. Niklasson, and E. Wackelgard,

J. Phys. D: Appl. Phys. 37, 115 (2004).21. http://www.texloc.com/closet/clprefractiveindex.html.22. I. Ohlıdal and F. Lukes, Opt. Acta 19, 817 (1972).