Multi-Peak Curve Fitting for Spectroscopy Data in SciDAVis
Real spectra are rarely a single clean peak. Raman bands overlap, XPS envelopes hide two oxidation states under one hump, and chromatograms merge shoulders into their neighbors. Deconvolving them means fitting a sum of peak functions and reading the positions, widths, and areas of the components. SciDAVis gives you two honest tools for this: a guided multi-peak quick fit and the full Fit Wizard for mixed or complex models.
Step 1 — Prepare the spectrum: baseline first
A drifting baseline corrupts every peak parameter, especially areas. Handle it before or during the fit:
- Quick route: every built-in peak function in SciDAVis carries a
y0offset parameter, so a flat baseline is absorbed automatically during the fit. - Better route: if the baseline slopes or curves, fit it together with the peaks in the Fit Wizard — e.g. a sum of a linear term plus your peaks — rather than trying to subtract it by hand beforehand. Subtracting an imperfect baseline by eye and then fitting the residue is the most common source of biased areas.
- Sanity check: zoom into peak-free regions of the spectrum. If the fitted baseline misses them visibly, the model is wrong — add a slope term or restrict the fit range before trusting any area.
Step 2 — Pick the right peak model
The two workhorse line shapes, exactly as defined in the official Handbook:
| Model | Parameters | Use it when… |
|---|---|---|
| Gaussian | A = height, w = width, xc = center, y0 = Y offset | Broadening is dominated by instrumental resolution or Doppler/thermal effects — typical for Raman bands, XRD reflections, and chromatography peaks. |
| Lorentzian | A = area, w = width, xc = center, y0 = Y offset | Broadening is dominated by lifetime/pressure (collisional) effects — typical for IR absorption lines and NMR resonances. Note the A here means area, not height. |
Rule of thumb: start with Gaussian for condensed-phase vibrational spectra and chromatography, Lorentzian for gas-phase and magnetic-resonance lines. When both broadening mechanisms matter, the rigorous answer is a Voigt profile — which SciDAVis does not ship built-in; approximate it in the Fit Wizard with a custom expression or decide which single shape dominates (see FAQ).
Step 3 — The multi-peak quick fit
For the common case — N peaks of the same shape — SciDAVis has a guided path:
- Plot your spectrum and make the plot window active.
- Go to
Quick Fit → Fit Multipeak → Gaussian(or→ Lorentzian). - Enter the number of peaks you expect.
- For each peak: click one data point near the peak top on the plot, then press ENTER to validate the choice. These clicks are only initial guesses for the centers — they do not need to be exact.
- SciDAVis runs the least-squares fit, writes the results to the log window, draws the fitted curve on the plot, and creates a table storing the fit (find it in the project explorer — it is hidden by default).
This is the fastest route from “overlapping hump” to “two centers and two areas,” and for many routine spectra it is all you need.
Step 4 — Mixed models in the Fit Wizard
When peaks need different shapes — say a Gaussian for one band and a Lorentzian for its neighbor — or you need a baseline term included, open the Fit Wizard (Analysis → Fit Wizard, or Ctrl+Y). Per the Handbook, the wizard works in two stages:
- Stage 1 — the model: pick functions from the selector (built-in set, your saved functions) and add their expressions, or type your own analytical expression in the bottom half of the dialog. Then list the parameters to fit as a comma-separated list.
- Stage 2 — the parameters: give an initial guess for every parameter. This is the single most important step: quick-fit commands use default guesses and can fail to converge if they start far from the answer; the wizard lets you seed each center, width, and height from what you see on the plot.
- Weighting: the second tab offers No weighting (default), Instrumental (uses your Y-error bars — add them to the curve first), Statistical (weights from the square roots of the data), and Arbitrary Dataset (your own weight column). Use Instrumental whenever you have real error bars; it is the statistically correct choice.
- Output: in the Custom Output tab choose Uniform X Function for a smooth model curve or Same X as Fitting Data for a table of fitted values at your actual X points — the latter is what you want for residual analysis.
Step 5 — Judge the fit with residuals
A pretty overlay curve proves nothing. Open the fit table (project explorer), add a column with data − fit at each X, and plot it:
- Good fit: residuals scatter randomly around zero with no visible structure — the model has captured everything systematic.
- Wavy or humped residuals: a component is missing or the wrong line shape is in use. Add a peak or switch Gaussian ↔ Lorentzian for the suspect band.
- Residuals fanning out with intensity: your weighting is wrong — switch to Instrumental weighting with real error bars.
- One region stubbornly off: restrict the fit range to exclude it, or add a baseline term (Step 1) rather than forcing peaks to absorb the error.
Worked example — two overlapping Raman bands
Imagine a Raman spectrum with a broad envelope around 1,350–1,450 cm−1 that you suspect hides two bands (a common D/G-region situation in carbon materials):
- Plot the spectrum; check the baseline in the flat regions on both sides of the envelope. It is roughly flat, so the built-in
y0offset will do. Quick Fit → Fit Multipeak → Gaussian, number of peaks = 2. Click near each visible shoulder top, ENTER each time.- Read the log window: two centers (
xc), widths (w), heights (A). Plot residuals from the hidden fit table — they scatter randomly. Done in under two minutes. - If residuals showed structure, you would switch to the Fit Wizard: same two Gaussians plus a linear baseline term, seeded with the quick-fit results as initial guesses, with Same X as Fitting Data output for a second residual check.
FAQ
How do I know how many peaks to fit?
Start from chemistry, not statistics: how many components does the sample plausibly contain? Then check the residuals — systematic structure after fitting N peaks argues for N+1. Fitting more peaks than the data supports gives beautiful overlays with meaningless, highly correlated parameters.
Gaussian or Lorentzian — how do I choose?
Match the broadening physics (Step 2). If unsure, fit both and compare residuals and parameter uncertainties in the log window — but do not mix shapes without a physical reason, or you are just adding knobs.
Can I constrain a parameter to a range?
SciDAVis 2.9.x does not offer min/max bounds in the Fit Wizard. Control the fit through good initial guesses, staged fitting (strong peaks first), and sensible weighting instead.
Does SciDAVis have a Voigt profile built in?
No. Approximate it with a custom expression in the Fit Wizard, or decide which broadening mechanism dominates and use the corresponding single shape — for most routine work the difference is smaller than the experimental noise.
Next steps
- Curve Fitting in SciDAVis — single-function fits and the Fit Wizard in depth
- FFT Frequency Analysis in SciDAVis — find hidden periodicities before fitting
- Importing Data into SciDAVis — get spectra from instrument files into tables
- Exporting Publication-Quality Figures — turn fitted spectra into journal figures
- All SciDAVis Tutorials — back to the tutorial hub