Hi Pjotr,

you are right, there cannot be a baseline in ATR spectra as long as the sample can be seen as an infinite medium. In many cases, baselines appear because a ZnSe or Diamond crystal is used, together with an angle of incidence of 45°. This combination gives usually excellent penetration depths, but the combination fails once the refractive index in the visible is slightly higher than the typical organic substances, like e.g. in the case of charcoal or if stronger bands appear in the spectrum, because then the critical angle has to be higher than 45°, and there is no total reflection any more, which could be attenuated. If you want more background, have a look at my lecture notes: Preprint Wave optics in Infrared Spectroscopy

P.S.: Here is a link on a related question in RG: https://www.researchgate.net/post/How-can-I-correct-sloping-baseline-in-FTIR-ATR-biochar-spectra

The sloping is caused by the imperfect contact between the sample and ATR element. As you said, the effective penetration depth of IR is frequency dependent. Thus, if the sample/ATR element contact is not perfect, the IR radiation reflect at the sample surface without obtaining the information from the sample. However, the degree of this occurring for the long wavelength is less. This is the reason why the baseline goes up in the longer wavelength region in a exponential fashion. The supporting evidence of your sample not achieving good contact with the ATR element is your observation of the differential absorption band contour. If the contact is perfect, such contour should not appear. This is caused by the Christiansen effect when the contact at the interface is poor. Similar problem happens when you make a KBr pellet for a very hydrophobic sample. To minimize the contact problem (derivative contour and baseline shift), you may try one of the following. i) Use a IR transparent oupling liquid on the surface of ATR element (you can use only the frequency region the coupling liquid is transparent). For IR transparent material, Nujol mull or Fluorolube can be used.; ii) Use ATR element that is of smaller refractive index, such as KRS-5; iii) If your sample is not heat sensitive, warm up the sample immediately before contacting to ATR element so that the sample forms a good contact.

Thank you Thomas and Hatsuo.

Thomas Mayerhöfer I'm using a Ge crystal at 45° and I'm measuring organic substances with a refractive index of around 1.5. My samples are typically around 1 to several µm thick, so definitely at lower wavenumbers some samples cannot be considered an infinite medium I think? Could this perhaps contribute to baselines in my spectra?

Hatsuo Ishida As my samples are very soft, the contact with the ATR element should in general be relatively good. The baselines in my spectra do indeed go up towards longer wavelengths (lower wavenumbers), but it does not always resemble an exponential function (see figure attached). In this figure the same sample is measured (so the chemistry and sample contact should be nearly identical), but it was deposited on different measurement substrates: ZnSe-window (blue), Au-mirror (red) and Si-wafer (green). Could this, maybe combined with the non-infinite thickness of my samples, perhaps explain the variable baselines?

Pjotr Meyvisch Yes, my guess would be that your problem is related to the different nature of the substrates and that your layers are not thick enough so that you see a substrate effect. Do you get always the same kind of slope for one particular type of substrate? If yes, let me guess: The red one is the spectrum of the substance on ZnSe and the green one on gold?

Thomas Mayerhöfer The red one is actually gold, green is silicon and blue is ZnSe. I usually measure on a Au-mirror, for which the baseline slopes are generally quite consistent/similar, but overall they show the largest similarity with the slope of the green spectrum (about 20°), as you guessed to be the spectrum on gold. The red spectrum on gold in the figure is measured on a different spectrometer in another lab, which could explain the discrepancy. In any case, would you think that this differential substrate effect poses a problem when the goal is to qualitatively intercompare spectra, which are possibly measured on different substrates? Or can a polynomial baseline correction remove most of the variability caused by this effect? Could a substrate effect also cause non-baseline related features in spectra (e.g. peak deformations, horizontal or vertical shifts), which could be problematic in a qualitative analysis?

Please do not use any polynomial baseline correction. I did some quick calculations for PMMA on your three types of substrates (thickness 0.2 microns, naturally polarized light) and some really crazy things can happen, depending on thickness and polarization. If you correct the baseline of the green spectrum with a polynomial baseline, I do not think that any meaningful spectrum can come out of it. On the other hand, in your system, you should mostly be fine - but keep in mind that in general you cannot be sure. Do you know polarization state and layer thickness accurately? If yes, there is a way to correct spectra for such effects.

Thomas Mayerhöfer Unfortunately I don't know the polarization state and have no convenient way of determining the actual thickness of my sample during compression. The latter is just a rough estimate based on how thick the layers on my uncompressed samples are (= microorganisms). From your calculations it seems that mainly the Si spectrum suffers from intense baseline distortions, right? From my data I get the impression that spectra on Au and ZnSe are quite similar, even after polynomial baseline correction which typically requires only one iteration (so that seems not too alarming?). It would be nearly impossible for me to always guarantee an 'infinite' sample thickness, as my method is based on measuring individual microorganisms and some of the smaller species are just not thick enough. Stacking multiple specimens on top of each other would be difficult (not impossible), but then you would also introduce infraspecific variability. The reason I'm mostly measuring in ATR over transmission is that spectra from the latter method in many cases contain quite severe scattering/interference distortions, which are hard to correct via the currently available model-based correction algorithms.

Pjotr Meyvisch Yes, I agree, discard Si as substrate and you should be ok. ZnSe should actually give no baseline at all in the ideal case and gold one that you should be able to handle with the polynomial fit. As long as you do not want to do comparisons of peak height or band area you should be fine. I also would not interpret band shifts that are smaller than about 2-3 cm-1, because those can be a consequence of the different substrates.

Thomas Mayerhöfer Allright, thanks for all the valuable suggestions. Cheers, Pjotr.

A simpler explanation of the base line using an Si crystal is that you are not getting any IR transmission though the crystal beyond about 1400cm-1. The cut off point is about 7-8 microns.

Geoffrey Dent This depends strongly on doping. Undoped Si is transparent.

The comment I made was based an Silicon Window/ATR Crystal data. The general cut off published is around 7-8 microns roughly 1300-1200 cm-1

Geoffrey Dent The ATR crystal is Ge for all spectra Pjotr Meyvisch presented and I simulated. The special thing here is that the samples consist of layers on different substrates, including Si.

One particular problem with ATR is coming from the necessity to correct the absorption for the penetration depth. This correction is refractive index and wavelength dependent and might be done in a number of different ways, but usually it is done automatically by the software. This correction will result in the shift/distortion of the bands and changes to the apparent background of the spectra. You can check the physics in the following links.

Griffith, Infrared spectroscopy

1. P. R. Griffiths, J. A. D. Haseth, Fourier transform infrared spectrometry. (Wiley-Interscience, 2007). Chapter 15.2

2. V. Ruddy, B. D. MacCraith, J. A. Murphy, Evanescent wave absorption spectroscopy using multimode fibers. Journal of Applied Physics 67, 6070-6074 (1990).