XPS Peak Fitting for Defensible Chemical States

A visually smooth fit can still produce the wrong chemical-state assignment. In XPS peak fitting, the central task is not to place enough components under a spectral envelope to minimize residuals. It is to construct a physically credible model of photoelectron emission that remains defensible when a reviewer, colleague, or quality team asks why each peak, width, position, and area is present.

XPS spectra are information-rich but rarely simple. Surface contamination, differential charging, overlapping species, multiplet structure, asymmetric line shapes, plasmon losses, and an imperfect background can all create apparent features. A research-grade workflow treats these factors as model decisions with documented scientific consequences, not as cosmetic adjustments.

Why XPS peak fitting fails so often

The most common failure is treating the fit as a purely mathematical decomposition. Nonlinear optimization can produce a low residual sum of squares with components that have no plausible chemical or physical meaning. Adding unconstrained peaks usually improves the visual overlay. It can also inflate uncertainty, hide a poor background choice, and create chemical states that are unsupported by the sample history or known photoelectron physics.

A second failure begins before fitting: using an energy scale that has not been evaluated for charging and calibration. Binding-energy shifts may represent oxidation-state changes, local bonding environments, or charge referencing errors. If this distinction is unresolved, assigning a shifted component to a new species is premature.

Background selection introduces another source of ambiguity. A simple linear or Shirley-type background may be suitable for some narrow regions and sample types. For systems where inelastic scattering and energy-loss behavior materially shape the spectrum, a Tougaard-type treatment may be more appropriate. Neither choice is universally correct. The appropriate model depends on the spectral region, material, acquisition conditions, and the physical question being asked.

Establish the measurement context before fitting

Peak fitting should begin with the experiment, not the optimizer. Record the X-ray source, pass energy, step size, analyzer configuration, takeoff angle, spot size, acquisition time, and charge-neutralization conditions. These settings influence energy resolution, signal-to-noise ratio, line width, and the degree to which small chemical shifts can be resolved.

Review the survey spectrum before interpreting high-resolution regions. It confirms which elements are present, reveals unexpected contaminants, and helps identify whether a proposed species is chemically credible. For example, assigning an oxidized component in a metal core-level spectrum is much stronger when the corresponding oxygen signal, stoichiometry, and sample-processing history support it.

Charge correction deserves special scrutiny. Adventitious carbon referencing can be useful, but it is not a universal solution. The binding energy selected for the carbon reference must be appropriate to the sample and the carbon chemistry actually present. Conductive standards, internal references, known substrate peaks, and carefully documented neutralization conditions can be more reliable in some workflows. When charging is nonuniform, a single rigid energy shift may not correct every region equally.

Build an XPS peak fitting model from physics

A defensible fit starts by defining the expected spectral architecture. Ask which spin-orbit doublets, multiplets, satellites, loss features, or chemically distinct environments are expected before allowing the software to search for components. Literature values and reference materials are useful starting points, but they should not replace knowledge of the measured system.

For a spin-orbit doublet, the separation and intensity ratio are often governed by atomic physics. Constraining those relationships reduces parameter freedom and prevents the fit from distributing area arbitrarily between paired peaks. Components representing the same chemical state commonly should share a line shape and comparable full width at half maximum, subject to the known behavior of that core level and material.

Line-shape selection also matters. Symmetric Gaussian-Lorentzian product or sum profiles can be appropriate for many insulating or semiconducting compounds. Metallic states may require an asymmetric profile because of the low-binding-energy tail associated with conduction-electron screening. Applying asymmetry indiscriminately, however, can absorb baseline errors or nearby unresolved features. Use it where the physics supports it, and document why.

The same caution applies to satellites. Shake-up satellites in transition-metal oxides and other correlated systems may be chemically diagnostic. They are not generic extra peaks to add when a residual remains. Their expected position, breadth, and relationship to the main line should align with the material class and reference evidence.

Use constraints to reduce false certainty

Constraints are not a way to force data into a preferred conclusion. They are a way to encode prior physical knowledge and limit parameter combinations that are mathematically possible but scientifically implausible. Useful constraints may include fixed or bounded doublet separations, theoretical area ratios, linked widths, shared shape parameters, and realistic limits on chemical shifts.

Overconstraint has its own risk. A sample containing mixed phases, strong disorder, differential charging, or size-dependent effects may genuinely deviate from an ideal reference. The objective is not to make every spectrum conform to a textbook pattern. It is to distinguish justified departures from artifacts introduced by an overly flexible or overly rigid model.

Fit the background and peaks as one analytical problem

Baseline correction is often treated as a preliminary cleanup step, but it directly affects component areas, widths, and even the number of peaks needed. A background that sits too high can erase weak species; one that sits too low can manufacture broad components. The peak model and background therefore need to be evaluated together.

This is where generic graphing tools tend to fall short. They can optimize a set of curves, but they often leave the analyst to manage baseline assumptions, parameter relationships, data provenance, and fit diagnostics manually. A specialized environment such as PeakLab™ is designed to combine baseline-and-peak modeling, nonlinear optimization, scientific constraints, and statistical reporting in a reproducible workflow.

The optimization method should be capable of handling parameter correlation. In a crowded envelope, peak position, width, amplitude, and background parameters can compensate for one another. A stable-looking result may still have poorly determined component areas. Parameter confidence intervals, covariance information, and sensitivity to reasonable starting values reveal whether the interpretation is stable or merely one of several nearly equivalent solutions.

Validate the result beyond the overlay plot

A fit is credible when independent checks point in the same direction. The overlay is necessary, but it is the least demanding test. Examine the residual trace across the entire fitted range. Random, low-amplitude residual structure is generally more acceptable than repeating positive and negative patterns near a shoulder, doublet partner, or background boundary.

Use four practical validation checks:

• Residual behavior: Look for structured residuals that indicate an omitted feature, unsuitable line shape, or biased background.
• Parameter plausibility: Confirm that binding energies, widths, doublet ratios, and separations agree with the proposed chemistry and instrument resolution.
• Cross-region consistency: Test whether assignments agree with related core levels, elemental ratios, Auger parameters, or valence-band evidence where available.
• Model stability: Refit from reasonable alternative starting values and assess whether the same chemical interpretation survives.

Goodness-of-fit statistics have value, but no single statistic proves a model. A high coefficient of determination is particularly weak evidence for an XPS fit because broad envelopes can produce impressive values even when individual components are wrong. Weighted residual metrics, information criteria, parameter uncertainty, and the physical coherence of the model offer a more meaningful basis for comparison.

Avoid interpreting a weak component near the detection limit as a confirmed chemical state simply because the optimization finds it. Signal-to-noise ratio, fitting-window selection, background uncertainty, and the number of free parameters all influence whether that component is distinguishable from noise. State uncertainty plainly when the data do not support a unique assignment.

Make XPS results reproducible and reviewable

A publication-ready or regulated report should preserve more than the final plot. Record the raw data source, preprocessing steps, energy reference, fitting range, background model, peak profile, initial values, constraints, optimized parameters, uncertainty estimates, residuals, and software version. If a component was assigned using a reference spectrum or expected stoichiometry, document that reasoning as well.

Reproducibility becomes especially important for comparative studies, process development, and high-throughput materials screening. Applying the same justified model structure across a series of samples makes trends in oxidation state, surface composition, or treatment response more credible. It also exposes when a sample no longer fits the established model and requires a fresh scientific hypothesis rather than a routine batch fit.

The strongest XPS peak-fitting result is not the one with the most components or the smoothest line. It is the one that connects measurement conditions, spectral physics, constraints, statistics, and chemical reasoning into a model another qualified analyst can inspect, reproduce, and defend.