Fourier Transform Infrared Analysis of Contamination bySearching Difference Spectra Against Libraries ofDifference Spectra

KENNETH B. LAUGHLIN*Rohm and Haas Company, Spring House, Pennsylvania 19477

A method is presented for quickly identifying contaminants in a material.

Fourier transform infrared (FT-IR) spectra are collected for anomalous

and normal material using a method that gives highly reproducible

absolute intensity, such as attenuated total reflection (ATR) spectroscopy

of liquids or solids that contact the crystal perfectly. The difference

spectrum is calculated as the point-by-point subtraction of absorbance

values without the use of a variable subtraction factor, giving a spectrum

with positive and negative spectral features. This spectrum is then

searched against libraries of difference spectra, such as spectra of possible

contaminants minus the spectrum of normal material. The key advantage

of the method is that it removes subjective judgment in choosing the

subtraction factor. It also provides information about the material that is

depleted along with identifying the contaminant.

Index Headings: Infrared; Spectroscopy; Fourier transform infrared

spectroscopy; FT-IR spectroscopy; Attenuated total reflection; ATR;

Subtraction; Difference; Library; Search; Contamination; Hydrogen

bonds.

INTRODUCTION

Fourier transform infrared (FT-IR) spectroscopy is oftenused to analyze the quality of manufactured materials. FT-IRspectroscopy is well-suited for quality control because spectracan be measured quickly and conveniently and it provides abroad chemical check of product consistency. The spectrum ofthe material can be analyzed to determine whether it matchesspectra of the product normally produced. If an anomalousspectrum is measured, it is very helpful to quickly identify thecompositional difference so that appropriate action can betaken. In a research and development lab, FT-IR spectroscopyis often used to identify compositional differences between twomaterials that have been found to perform differently inapplications testing.

Spectral subtraction has long been used as a tool to analyzecompositional differences.1–3 For example, subtracting twospectra can help reveal the production or loss of functionalgroups in a material as it reacts. It can also be used to isolatethe spectrum of a contaminant by removing the spectrum of thematrix, as described below. The isolated spectrum can then besearched against libraries of conventional FT-IR spectra toidentify the contaminant.

A novel approach is presented here for identifyingcompositional differences, where the difference spectrum issearched against libraries of difference spectra. By using amethod with reproducible optical path length (such asattenuated total reflection (ATR) spectra of liquids, castpolymer films or very soft solids), the method is streamlined

because the subjective choice of subtraction factor can beeliminated.

The method is demonstrated here with examples usingcommercial latex and solution polymers. Several scenarios areconstructed that are directly related to product quality andcould conceivably arise at a manufacturing facility: (1) cross-contamination of two products; (2) dilution of a latex (polymer/water ratio); (3) change in copolymer composition; (4) degreeof neutralization of an acid polymer; and (5) contamination bya solvent.

CONVENTIONAL METHOD OF SPECTRALSUBTRACTION

The difference spectrum is calculated as the point-by-pointsubtraction of the absorbance values of the two spectra. Theconventional method of subtracting FT-IR spectra of twodifferent samples is shown in Eq. 1, where a factor is insertedinto the equation:1

AbsiðdifferenceÞ=AbsiðanomalousÞ� Factor 3 AbsiðreferenceÞ ð1Þ

The subtraction factor is used to compensate for variations infilm thickness and for dilution of the matrix and has long beenrecognized as a critical step in successfully performing FT-IRspectral subtractions between different samples. Many of themost common methods of measuring FT-IR spectra (transmis-sion of thin film, KBr pellet, diffuse reflectance, Nujol mull)are subject to large variations in optical path length. Mostcommercial software packages for analyzing FT-IR spectrainclude a program to vary the factor while continuouslyobserving the subtraction result, which is very useful becausethe choice of the factor is subjective. Expert judgment isrequired to choose the value of the factor, with the goal ofisolating the spectrum of the unknown contaminant in pureform. The choice is often made by observing which peaks gofrom positive to negative as the factor is varied, then setting thefactor such that these features cancel to give the minimumintensity. Because the spectral features usually overlap,choosing the optimum subtraction factor is often difficult. Ifthe spectrum of the contaminant is successfully isolated, thisspectrum can be searched against conventional FT-IR librariesto identify the contaminant.

NEW METHOD OF SPECTRAL SUBTRACTIONWITH SUBTRACTION FACTOR = 1

If the absolute absorbance is very reproducible, a simplerapproach becomes feasible. For example, if the ATR method isused to measure the spectrum for materials that contact theATR crystal perfectly, then the baseline-corrected absolute

Received 20 March 2007; accepted 20 November 2007.* E-mail: klaughlin@rohmhaas.com.

176 Volume 62, Number 2, 2008 APPLIED SPECTROSCOPY0003-7028/08/6202-0176$2.00/0

� 2008 Society for Applied Spectroscopy

absorbance can be reproducible to better than 1%. Thedifference spectrum can then be calculated as the point-by-point subtraction of the absorbance values of the two spectrawithout using a subtraction factor, as shown in Eq. 2:

AbsiðdifferenceÞ=AbsiðanomalousÞ � AbsiðreferenceÞ ð2Þ

With this type of subtraction, positive peaks are observedwhere the contaminant absorbs more strongly than thereference material, while negative peaks are observed wherethe reference material absorbs more strongly than thecontaminant. The key advantage of the subtraction in Eq. 2is that it removes the subjective choice of subtraction factor. Italso provides information about the material being depletedalong with identifying the contaminant.

The method can be used for any experimental technique thatgives a highly reproducible spectral intensity, including ATRspectra of liquids, polymer films cast on an ATR crystal, verymalleable solids pressed against an ATR crystal, andtransmission spectra of solutions in a fixed path length cell.With the ATR method, the effective optical path length isconstant as long as the thickness of the material in contact withthe ATR crystal is much greater than the evanescent wavedepth, which is approximately 1–5 lm for a conventional ATRexperiment.4 This criterion is usually very easy to satisfy,although it can be a problem with polymer films cast on theATR crystal if very thin films are needed to remove the solvent.

EXPERIMENTAL PROCEDURE

The FT-IR spectra are collected with either a MattsonInfinityTM or a Thermo NicoletTM 6700 FT-IR spectrometerusing a 3-bounce diamond ATR accessory (Durasamplir IITM,Smiths Detection). Spectral plots and subtractions are calcu-lated using Thermo Nicolet OMNICTM software. Librarysearches are performed using the mean-centered correlationcoefficient match score.� This match score is insensitive to themagnitude of the spectra being compared and therefore to theamount of contamination.

The diamond ATR crystal is 3 mm in diameter and ismounted nearly flush in a stainless steel plate. The relatively

small surface area of the diamond and the flush mounting makeit easy to cast polymer films directly onto the crystal with adraw-down bar, giving a consistent film thickness of 10–20lm. If necessary, solvent evaporation can be enhanced byheating both the diamond ATR plate and the nitrogen flow overthe sample to 100 8C. Heating the polymer film greatlyincreases the rate of solvent evaporation from polymers with aglass transition temperature (Tg) above room temperature, suchas polymethylmethacrylate (Tg ; 105 8C). Raising thetemperature to near or above the Tg increases the polymermobility and enhances diffusion of the solvent as well asincreasing the vapor pressure. For samples that require heating,the spectra are measured after returning the ATR to roomtemperature, and the background spectrum is measured afterremoving the polymer film to account for slight opticalinstabilities in the ATR accessory and/or spectrometer due tothe heating cycle.

Reference samples of homopolymers of polymethylmeth-acrylate (pMMA), polybutylmethacrylate (pBMA), polybutyl-acrylate (pBA), polyethylacrylate (pEA), poly-2-ethylhexylacrylate (pEHA), polystyrene (pSty), polyvinyl-acetate (pVAc), and polybutadiene (pBD) were purchasedfrom Aldrich Chemical Company. The latex polymersdiscussed below are commercial products synthesized byemulsion polymerization using proprietary methods.

RESULTS AND DISCUSSION

Latex Polymer Blend. Figure 1 (top) shows an overlay ofthe spectra of cast films of a pure polymer Latex A (nominallya copolymer of comparable amounts of MMA and BA), and ablend with 10% Latex B (nominally pBA). For this system,very little chemical interaction is expected between thepolymers because the high molecular weight of the polymerlimits diffusion between the latex particles after the film is cast.The spectra of the blend and pure Latex A are very similar, andit is difficult to distinguish them without expanding the plot.The noisy portion of the spectrum from 1900 to 2700 cm�1 isdue to absorption of the infrared radiation by the diamondcrystal.

Conventional Subtraction. The difference spectrum of castfilms of the blended material minus pure latex A, calculatedwith a subtraction factor of 0.9, matches very closely with thespectrum of pure Latex B, as shown in Fig. 2. The mean-

FIG. 1. Overlay of spectra of cast films of a blend of 90% Latex A/10% LatexB, along with pure Latex A. The spectral differences are very small.

FIG. 2. Conventional method of spectral subtraction of the spectra of castfilms shown in Fig. 1. The spectrum of the contaminant (Latex B) is isolated byusing a subtraction factor of 0.9.

� The formula for the mean-centered correlation coefficient is r = [R(x �xave)( y � yave)]/[R(x � xave)

2R( y � yave)2]0.5.

APPLIED SPECTROSCOPY 177

centered correlation coefficient for the region from 800 to 1500cm�1 is 0.994. For this example, the optimal subtraction factorof 0.9 is known from the blend ratio.

Subtraction with Factor = 1. Figure 3 (top spectrum)shows the difference spectrum of cast films of the latex blendminus the pure Latex A (from Fig. 1), calculated with asubtraction factor equal to one. The difference spectrum of castfilms of pure Latex B minus pure Latex A, shown in the bottomof Fig. 3, is nearly a perfect match. Visual inspection showsthat virtually all spectral features are common to both spectra.The match score for the spectra shown in Fig. 3 is 0.975,calculated for the region from 800 to 1500 cm�1. Furthermore,the difference spectrum gives an accurate quantitative measureof the contamination level, because the 10% latex blend minuspure Latex A is ten times weaker than the difference spectrumof pure materials (absorbance of 0.043 versus 0.42, respec-tively).

Subtraction Balance. In order to obtain reliable results witha subtraction factor of 1, the absolute intensity of the spectramust be very consistent. Figure 4 shows the effect of varyingthe subtraction factor for cast films of the blended minus pureLatex A shown above in Fig. 3. Varying the subtraction factor

is equivalent to changing the absolute intensity of one of thespectra. Figure 5 shows the match scores calculated versus thesubtraction factor over the range from 0.95–1.05. The bestmatch is found for a subtraction factor of 1.01, and a change of0.01 has a substantial effect.

Dilution of Latex with Water. Fourier transform infraredspectra of Latex A are measured before and after diluting withwater from 50% solids to 40% solids. The diluted materialshows stronger water peaks (3300 and 1640 cm�1) and weakerpolymer peaks. The difference spectrum of diluted minusnormal material matches well to the difference spectrum ofpure water minus the normal latex. The match score for theregion 650–1800 cm�1 is 0.982, where the broader region isused to include the water absorption features.

Copolymer Composition. Fourier transform infrared spec-tra are measured for a copolymer latex (nominally a copolymerof MMA and BA), along with a latex synthesized using amonomer mixture where the MMA monomer was blended withEA. The BA content is identical for the two latexes.

The difference spectrum is compared to a library ofdifference spectra of cast films of homopolymers, subtractedin all combinations in both directions (e.g., pEA minus pMMAand also pMMA minus pEA). Seven homopolymers areincluded in the library: pMMA, pBMA, pEA, pBA, pSty,pVAc, and pBD, giving a total of 42 difference spectra. Thetop six matches are shown in Table I. The correct contaminantis identified (pEA minus pMMA), with a match score of 0.948.Visual inspection of the difference spectra in Fig. 6 is even

FIG. 4. Spectral subtraction of cast films of the blend minus pure materialcalculated with subtraction factors 0.96, 1.00, and 1.04, overlaid with thespectral subtraction of pure Latex B minus pure Latex A. A change in thesubtraction factor of 4% strongly distorts the difference spectrum.

FIG. 5. Mean-centered correlation coefficient match score as a function ofsubtraction factor, for difference spectra of cast films of Latex A contaminatedwith 10% Latex B minus Latex A. The match scores are calculated versus PureLatex B minus Pure Latex A for the region from 800–1500 cm�1. The bestmatch is for a factor of 1.01, and a change of 0.01 shows a clear effect.

TABLE I. Mean-centered correlation coefficient match scores for theEA/MMA blend difference spectrum shown in Fig. 6. The top six librarymatch scores are shown, calculated for the region from 800–1500 cm�1.

Library spectrum Match scores

pEA minus pMMA 0.948pBA minus pMMA 0.907pBA minus pBMA 0.852pEHA minus pMMA 0.825pEA minus pBMA 0.788pEHA minus pBMA 0.752

FIG. 3. Spectral subtraction of cast films of the blend minus pure materialusing a factor of 1, with a spectral subtraction of pure Latex B minus pure LatexA. The spectral match is very close, and the magnitude of the spectra can beused to determine the contaminant concentration.

178 Volume 62, Number 2, 2008

more convincing in identifying the contaminant, especiallybased on the similarity of subtle features near 1350, 1100, and1000 cm�1. The spectra in Fig. 6 are measured for the wetlatexes, and the subtraction only works well if the polymer/water ratio is consistent between the two samples.

Neutralization of Carboxylic Acid. Figure 7 shows thespectra of a solution of polyacrylic acid (pAA) in water (5 wt%), neutralized with NaOH to 14% and to 43% molarequivalents. The difference spectrum, shown below, showspositive peaks for the carboxylate salt and negative peaks forthe protonated carboxylic acid.

For investigating the scenario of inconsistent extent ofneutralization, the difference spectrum of anomalous minusnormal material can be searched against a library of differencespectra of acid polymers at high pH minus low pH. Forexample, the library could include polymers containing acrylicacid, methacrylic acid, and maleic acid, neutralized withdifferent bases such as NaOH and NH3.

Solvent Additives. To evaluate the method for detectingsolvents, 3 wt % dipropyleneglycol (DPG) solvent is added to

Latex A. The difference spectrum, obtained by subtracting thepure latex A, is searched against a library of difference spectraof five industrial ether/alcohol solvents minus Latex A. Thecontaminant is correctly identified, with the search resultsshown in Table II. The ‘‘hydrated’’ spectrum of DPG isdiscussed below.

Effect of Chemical Interactions. Chemical interactions arewell known to have a profound effect on vibrational spectra,especially in the case of hydrogen bonding.5,6 For the methodof searching difference spectra to work well, the hydrogenbonding environment must be consistent between the spectra tobe analyzed and the reference spectra. An approach isdemonstrated below for aqueous systems using the exampleof dipropyleneglycol (DPG) added to Latex A. Because theDPG is dissolved mostly in the aqueous phase of the latex, thespectral match will be better if a reference library spectrum isused such that the DPG is measured in an aqueousenvironment.

A ‘‘hydrated’’ reference spectrum of DPG is constructed asfollows: The spectrum of a solution of 10% DPG in water ismeasured, and then the spectrum of pure water is subtractedusing a factor of 0.9. The resulting difference spectrum is thenscaled by a factor of 103 to compensate for the concentration.Due to the change in hydrogen bonding environment of thesolution, the water peaks are distorted and the subtraction givesresidual water features, but the fingerprint region from 900–1500 cm�1 shows only peaks from DPG.

FIG. 6. Spectra of a latex made with BA and MMA, along with a latex madewith BA, MMA and EA (top). The BA content of both latexes is the same. Thedifference spectrum matches closely to the spectrum of homopolymer pEAminus homopolymer pMMA (bottom). Other similar homopolymers show apoorer match.

FIG. 7. Spectra of polyacrylic acid aqueous solutions (5%) neutralized todifferent extents with NaOH. The difference spectrum below shows positivepeaks for the carboxylate salt and negative peaks for the protonated carboxylicacid.

FIG. 8. Difference spectra of a blend of Latex A with 3% DPG minus Latex A(bottom), compared with DPG minus Latex A (top) and ‘‘hydrated DPG’’minus Latex A (middle). The hydrated DPG spectrum matches much better andis generated from a solution of DPG in water, as explained in the text. Spectraof latexes are measured without drying.

TABLE II. Mean-centered correlation coefficient match scores for 3 wt% dipropylene glycol added to latex A minus latex A. The match scoresare calculated for a library of ether/alcohol solvents minus latex A for theregion from 900–1500 cm�1. The DPG (hydrated) spectrum is discussedbelow.

Library spectrum Match scores

Dipropyleneglycol (hydrated) 0.986Dipropyleneglycol 0.970Diethyleneglycolmethylether 0.882Dipropyleneglycolmethylether 0.869Propyleneglycol 0.687Ethyleneglycol 0.667

APPLIED SPECTROSCOPY 179

Figure 8 shows the difference spectrum of a blend of wetLatex A with 3% DPG minus the spectrum of pure Latex A.Also shown are the difference spectra of pure DPG minusLatex A, and the difference spectrum of ‘‘hydrated’’ DPGminus Latex A. The spectral match is visually much better for‘‘hydrated’’ DPG, especially in the region from 1000 to 1100cm�1 assigned to the C–O single bond stretch in DPG. Thematch score calculated from 900 to 1500 cm�1 is 0.986 for thehydrated spectrum versus 0.970 for the pure DPG.

Figure 9 shows the analysis of DPG in Latex A using castfilms rather than wet latexes. The film thickness is about 20lm, and the films are allowed to dry for 15 minutes directly onthe diamond ATR crystal before measuring the spectra. Thedifference spectrum of cast films of Latex A with 3% DPGminus pure Latex A is more similar to the difference spectrumof pure DPG minus the cast film of Latex A than to ‘‘hydrated’’DPG minus the cast film of Latex A, although the match is notperfect for either case. The hydrogen bonding environmentdepends upon the proportion of DPG which is dissolved in thepolymer versus droplets in the cast film. Both environments arevery different from aqueous solution. The portion of DPG thatdissolves into the polymer will predominantly act as hydrogenbond donators to the oxygen atoms in the ester groups of thepolymer rather than forming hydrogen bonds with other DPGmolecules.

Effect of Refractive Index on Spectral Match Index.Spectra measured by the ATR method are known to be subjectto shifts due to changes in the refractive index.4 These peakshifts can have a significant effect on difference spectra,especially in regions where strong absorptions occur. Peaks areshifted to lower frequency in ATR spectra compared withtransmission spectra because of the variation of refractive indexin the vicinity of an absorption peak. The spectral distortionsfrom this effect also degrade the match quality for librarysearches based on conventional subtractions and are not uniqueto the method of analysis presented here.

Implementation of the Method. Currently, commercialFT-IR libraries are normalized so that the absolute intensityinformation is lost, and they are therefore are not applicable forthis method. Because custom libraries are required, this limitsthe applicability of the method to scenarios involving materialsconsidered likely to be encountered. For the application of FT-

IR to quality control of manufacturing, library spectra aretypically available for the products manufactured at a givensite, as well as commonly used ingredients.

For investigating cross-contamination, comparison onlyneeds to be made with difference spectra involving theintended product. A program is written that subtracts each ofthe products or ingredients minus the intended product andcompares them to the difference spectrum of anomalous minusreference material. If 100 products and/or ingredients are used,then 99 difference spectra must be calculated for comparison,which takes less than one minute. The full library of allcombinations of products and ingredients, with 100 3 100 =10 000 spectra, does not need to be calculated becausecontamination of only one product is being considered. Themean-centered correlation coefficient match score is used forcomparison because it is insensitive to the magnitude of thespectra and baseline offsets. Therefore, it will give a goodmatch for the correct contaminant independent of the amountof contamination.

For analysis of variations of monomer composition, FT-IRspectra are measured for homopolymers of all of the monomerstypically used for manufacture. For example, if sevenmonomers are used, the difference spectrum library contains(7 3 7) � 7 = 42 spectra.

Because slight differences in the FT-IR instrument andaccessory can cause changes in absolute absorbance intensityand spectral shifts, optimum subtractions are obtained if thespectra of the anomalous and reference materials are measuredon the same apparatus. Ideally, the library spectra should alsobe measured on the same instrument as the samples beinganalyzed, but this requirement is not as critical because the purematerials typically show large spectral differences. A veryaccurate calibration transfer method may be able to compensatefor instrumental differences by ‘‘correcting’’ the spectra to astandard instrument, but this approach has not been investi-gated yet.

CONCLUSION

A new method of spectral analysis is presented for quicklydiagnosing contamination. FT-IR spectra are measured by atechnique that gives highly reproducible absolute absorbanceintensity, and the difference spectrum is calculated withoutusing an adjustable subtraction factor. The resulting spectrumcan be searched against libraries of difference spectra toidentify the contaminant.

The method has been demonstrated for analyzing composi-tional abnormalities from several types of scenarios: physicallatex blends, dilution of latexes with water, change in monomercomposition used to synthesize a copolymer, different extent ofneutralization of carboxylic acids, and solvents added to alatex.

The method is applicable to situations where one componentor matrix is enhanced at the expense of another. For example, ifcomponents A and B are added, which dilutes component C,the library must contain the difference spectrum of the mixtureof A and B (at the ratio of contaminant material) minus thespectrum of C. If the library contains only A minus C and Bminus C, there will not be a good spectral match. An additionalrequirement is that the effects of chemical interactions in thespectra are consistent. In particular, if there is a significantchange in the hydrogen bonding between the samples and thereference library spectra, the spectral match will not be good. A

FIG. 9. Difference spectra of cast films of a blend of Latex A with 3% DPGminus Latex A (bottom), compared with ‘‘hydrated DPG’’ minus cast film ofLatex A (middle) and DPG minus cast film of Latex A (top). The differencespectrum using pure DPG is more similar, but not extremely close.

180 Volume 62, Number 2, 2008

method is demonstrated for creating library spectra of materialsdissolved in aqueous solution to give better reference spectrathat account for changes in hydrogen bonding environmentversus pure materials. Although the expert judgment inchoosing a subtraction factor is eliminated, the differencespectra should still be examined carefully rather than relying onmatch scores alone.

1. J. L. Koenig, Appl. Spectrosc. 29, 293 (1975).

2. P. R. Griffith and J. A. de Haseth, Fourier Transform Infrared Spectrometry(John Wiley and Sons, 1986), p. 340.

3. B. C. Smith, Fundamentals of Fourier Transform Infrared Spectroscopy(CRC Press, Boca Raton, FL, 1996), p. 56.

4. N. J. Harrick, Internal Reflection Spectroscopy (Harrick ScientificCorporation, New York, 1979, original copyright 1967, John Wiley andSons).

5. G. C. Pimental and A. L. McClellan, The Hydrogen Bond (W. H. Freemanand Company, New York, 1960).

6. M. M. Coleman, J. F. Graf, and P. C. Painter, Specific Interactions and theMiscibility of Polymer Blends (Technomic Publishing Company, Lancaster,PA, 1991).

APPLIED SPECTROSCOPY 181