Analysis of IR data

Infrared spectroscopy for catalysis: what the bands mean,
and how to get from raw data to numbers you can defend

Anibal Boscoboinik · Center for Functional Nanomaterials, Brookhaven National Laboratory
Synchrotron Catalysis Consortium course · November 19, 2026

Why you should care

You have heard how to look at the catalyst. IR looks at the molecules on it.

TechniqueWhat it sees bestWhat it misses
XASOxidation state and local coordination of the metal, averaged over the sampleAdsorbates, light elements
XPSSurface composition and chemical statesMolecular identity of adsorbates is indirect
TEM / STEMSize, shape, structure of individual particlesWhat is adsorbed, statistics
IRWhich molecules are on the surface, on which sites, and how that changes under reaction. Works at pressure, in the lab, on a seconds timescaleThe solid itself (mostly), absolute amounts (with effort), anything without a dipole change

The strongest papers combine them: IR tells you the adsorbate, XAS and XPS tell you what it is sitting on.

Plan

Thirty minutes, three blocks

1. What a band means

8 min

  • A bond is a spring, the selection rule
  • What the instrument records
  • Four IR techniques, plus Raman and HREELS
  • CO, the probe molecule

2. Raw data to numbers

15 min

  • Single beam to absorbance
  • Gas-phase subtraction
  • Baseline
  • Peak fitting
  • Time series

3. Staying honest

7 min

  • Assignment
  • Is intensity proportional to coverage?
  • Spectator or intermediate?
  • Pitfall checklist

Everything interactive here runs in a browser. Open it and play along: catir.upton.club

A bond is a spring

ν̃ = (1/2πc) √(k/μ). Stiffer bond, higher wavenumber. Heavier atoms, lower. Isotopes change μ and nothing else.

The selection rule: the dipole moment has to change

Not every vibration absorbs. This is why N₂ and O₂ are invisible, and why CO₂ and water are in every spectrum you will ever take.

The detector never sees a spectrum

It records an interferogram. Resolution, apodization, zero-filling and phase correction are analysis choices already made before you open the file.

Four techniques, and where to find them at BNL

IRRAS

also called RAIRS or IRAS

Grazing reflection from a flat surface: single crystals, thin films, model catalysts.

+ Sub-monolayer, orientation. − Flat reflective samples only.

CFN: AP-XPS + IRRAS on one sample, UHV to ~20 mbar. PM-IRRAS on the Vertex 80v

Transmission

 

Beam through a pressed wafer, or powder held on a metal mesh.

+ Beer-Lambert holds: the one to use for counting sites.

− Opaque below about 1100 cm⁻¹ on oxides. Mass transfer.

CFN: Vertex 80v, powder on a tungsten mesh, UHV or gas, down to −120 °C

O-PTIR + Raman

optical photothermal IR

IR laser heats where the sample absorbs, a visible beam reads it out.

+ Sub-micron IR maps, Raman from the same spot.

− Laser range excludes the CO stretch. Not for monolayers.

CFN: mIRage, 758 to 1806 cm⁻¹, < 500 nm. In situ in development

DRIFTS

diffuse reflectance

Loose powder in a flow cell: the operando workhorse.

+ Real powders, no pressing, easy to couple with XAS or MS.

− Quantification is model-dependent. Bed gradients.

Not at the CFN. Elsewhere at BNL: Chemistry, some NSLS-II beamlines

O-PTIR: infrared chemistry at visible-light resolution

A conventional IR microscope cannot resolve anything smaller than several microns. O-PTIR reads the IR absorption with a visible beam, so resolution stops depending on the IR wavelength.

Which technique reaches which bonds?

Infrared on supported catalysts goes blind below about 1000 cm⁻¹, exactly where metal–carbon, metal–oxygen and metal–hydrogen bonds vibrate. Raman and HREELS reach them.

CO: one probe molecule, many answers

More back-donation from the metal into CO 2π* weakens the C–O bond: lower wavenumber. Position reports the site. Cations push it above the gas value.

Block 2

From raw data to numbers: the workflow

STEP 1RatioSample single beam over background. Choose the background on purpose.
STEP 2Remove the gasSubtract gas-phase and vapour references, scaled where no adsorbate absorbs.
STEP 3BaselineAnchor only where nothing absorbs. Same recipe for every spectrum.
STEP 4FitFewest components that leave a noise-like residual. Each one needs a name.
STEP 5Follow in timeBand areas against time, temperature or gas composition.

Every step is a model. Write down what you did, apply it identically to the whole series, and keep the raw single beams.

You just heard the same message for XPS. It is the same disease: a fit will always converge, and the software will never tell you it is meaningless.

Demonstration: an operando dataset, start to finish

Simulated: CO adsorption, then CO oxidation light-off, then purge, on a supported Pt catalyst. 30 spectra, one per minute.

Assignment: a band position is a hypothesis

A lookup table gives you candidates. These give you evidence: isotope labelling, reference compounds, coverage and temperature dependence, calculation.

Is band intensity proportional to coverage?

In DRIFTS, neither Kubelka-Munk nor log(1/R) is linear in general. And on metals, neighbouring CO molecules share intensity.

IRRAS on flat model catalysts: the surface selection rule

Only the dipole component perpendicular to the metal is seen, only with p-polarized light, best near grazing incidence. A missing band can mean a flat-lying molecule.

Spectator or intermediate?

A band that is present under reaction conditions is not therefore part of the mechanism. The largest bands usually belong to the most stable species, which are the least reactive ones.

  • Transient experiments: switch the gas, compare how fast the band decays with how fast the product forms.
  • Isotope switching (SSITKA): ¹²CO to ¹³CO at steady state. An intermediate exchanges at the rate of product labelling.
  • Modulation excitation: periodic perturbation plus phase-sensitive detection picks out species that respond, and rejects the static ones.

From the demo

Adsorbed CO: falls when CO₂ rises. Kinetically relevant.

Carbonate / formate: grows regardless. On the support, watching.

The second set gave the biggest bands in the fingerprint region.

A real example

Placeholder.
Drop one of your own datasets here: raw single beams, the processed spectra, and the one number that came out of it.
Put the image in slides/img/ and replace this block. Or delete the slide.

Pitfall checklist

  • Background taken at a different temperature than the sample spectrum
  • Gas reference recorded at a different resolution, pressure or temperature
  • Baseline anchors that move from spectrum to spectrum
  • More fit components than you can name
  • Reading a 10 cm⁻¹ shift as an oxidation-state change without checking coverage
  • Quoting ratios of band areas as ratios of sites
  • Trusting absorbance above about 1: too little light is left and the scale stops being linear
  • Assigning from position alone, with no isotope or reference check
  • Calling the biggest band the intermediate
  • Throwing away the single beams and keeping only processed spectra
Take home

Three things to remember

1. A spectrum is a ratio, and every processing step is a model. Record the recipe and apply it to the whole series.

2. Position suggests, isotopes prove. Coverage moves bands as much as chemistry does.

3. Present under reaction is not the same as involved in reaction. Test it with a transient.

All tools: catir.upton.club · open, free, no install.