Three standards that look alike can trade places in a fit. The mean valence still comes out right.
The height of an edge tells you how much of an element is there. Its shape, the energy-loss near-edge structure (ELNES), tells you the oxidation state. Here you build a Mn L2,3 spectrum from states you choose, then read the valence back three ways: fit it with references (MLLS), take the L3/L2 ratio, or find the L3 peak position. Change the counts, the energy calibration or the references, and watch which answer breaks.
Counts per energy channel at the top of the white line, above background. Energy shift: how far the unknown's energy axis sits from the references' (drift between sessions). Worse resolution: the unknown recorded with a wider zero-loss peak (ZLP FWHM) than the references.
DualEELS correction on: the unknown is shifted back using the zero-loss peak (ZLP) recorded at the same time as the edge. The references are always on a calibrated axis.
| Step | What you do | Knob on this page | What goes wrong |
|---|---|---|---|
| 1 Record | Record the unknown and your references at the same energy, dispersion, resolution and collection angle. | counts, worse resolution (ZLP) | Too few counts: the answer wobbles. Different resolution: fractions shift. |
| 2 Energy shift | Remove the energy shift so every spectrum sits on the same energy axis, ideally with a zero-loss peak (ZLP) recorded at the same time (DualEELS). | energy shift, DualEELS correction | 0.5 eV of drift moves the L3 position by half a valence. |
| 3 Background | Fit A·E−r in a background fit window (Γ) just before the edge and subtract it. | background fit window (Γ) | A window that touches the edge rise, or is very short, gives a wrong exponent. |
| 4 MLLS | Fit the edge as a non-negative sum of your references over a window that covers both white lines. | MLLS fit window, MLLS references | A missing reference cannot be fitted. Watch the residual. |
| 5 Ratio | Remove a 2:1 step continuum, integrate L3 and L2, compare the ratio with your references'. | (automatic) | The ratio depends on the step model and windows. Use the same ones for references and unknown. |
| 6 Report | Report the mean valence with its spread, the residual, and how the references were recorded. | table above | Separate fractions of three similar states are poorly fixed, even with no noise. |
The three references here are made-up shapes with white-line ratios (about 4.1, 3.2 and 2.2 after processing) and L3 positions (640, 641 and 642 eV) in the range reported for manganese oxides. They are not a copy of one paper's data. For real work the references must be your own, recorded like the unknown. The edge model has no multiplet structure, no crystal-field splitting and no oxygen K edge nearby. There is no plural scattering, so thickness effects are not shown. The noise is counting noise only; real detectors add readout noise and gain variation. MLLS here uses non-negative least squares with no extra terms; some people also fit a small linear background term or the first derivative of each reference to absorb a small energy shift.
On this site: Reading the EELS spectrum · EELS core-loss quantification · EELS thickness calculator · EFTEM elemental mapping
None of them alone. MLLS uses the whole shape and gives the smallest spread, if your references are right. The L3/L2 ratio is simpler and less sensitive to a small energy shift. The L3 position is quick but needs a perfect energy axis. If two methods agree within their spread, you can report the valence with some confidence.
Mn2+, Mn3+ and Mn4+ spectra look alike. A mix of Mn2+ and Mn4+ can look almost the same as Mn3+ (lesson 5). The mean valence is fixed much better than the separate fractions. Report the mean valence, or fit only two states if you know the third is absent.
Use lesson 2: set the counts, read the spread in the table. The spread falls as one over the square root of the counts. Then check that your sample survives that dose (see the beam damage page).
That page uses the edge area to find how many atoms of an element are there. This page uses the edge shape to find their oxidation state. You can run both on one spectrum.
@misc{tripathy2026eelsfinestructure,
author = {Tripathy, Manisha},
title = {EELS Fine Structure and Oxidation State},
year = {2026},
howpublished = {\url{https://untetheredatom.com/tem/eels-fine-structure-oxidation-state}},
note = {Interactive web tool}
}