The 284.8 eV you didn't measure

My own research is microwave spectroscopy, so I am not an XPS specialist. But the failure mode here is one I recognize immediately, because it is the same one I wrote about in the last Lab Science report: an inherited convention that starts as a practical shortcut, hardens into a default, and eventually gets treated as a measurement when nobody ever measured it. The 285 eV sp3 label was one of those. The 284.8 eV reference is a bigger one, because it does not just mislabel a peak. It sets the position of every peak in the spectrum.

Why the number exists at all

X-ray photoelectron spectroscopy works by knocking electrons out of a sample and measuring their kinetic energy. Subtract that from the photon energy and you get a binding energy, which is what you actually want, because binding energy tells you about chemical state.

That arithmetic only closes if you know the sample's electrical potential. For a metal bolted to the spectrometer, you do: the sample and the instrument share a Fermi level, and the scale is defined. For an insulator, you do not. As photoelectrons leave, the surface goes positive, and the entire spectrum slides to higher apparent binding energy by an amount nobody knows. Most labs fight this with a flood gun that sprays low-energy electrons back at the surface, which stabilizes the charging but does not eliminate the offset. You are still left with an unknown constant added to every peak.

So you need something in the spectrum whose true position you already know, and you shift everything until that thing lands where it belongs. Almost every sample that has touched air carries a thin film of hydrocarbon contamination, called adventitious carbon. It is everywhere, it always gives a C 1s signal, and it is free. The convention is to declare its main C-C/C-H component to sit at 284.8 eV, and shift accordingly.

The assumption buried in that move is that adventitious carbon appears at a fixed binding energy regardless of what it is sitting on. That assumption is what the argument is about.

The case against: 360 samples and a 2.89 eV spread

Grzegorz Greczynski at Linköping University has been making the case against for years. His most recent large dataset, published in Applied Surface Science in 2024, tests it on 360 thin-film specimens spanning metals, nitrides, carbides, borides, oxides, carbonitrides and oxynitrides.

The result is not subtle. Measured against the spectrometer's Fermi level, the C 1s position of adventitious carbon ranged over 2.89 eV across the sample set. In his words, that is "seven times more than the range specified in ISO guidelines, and more than many chemical shifts." The extremes are instructive. Adventitious carbon on magnesium sat at 286.90 eV. On molybdenum nitride it sat at 284.08 eV. Both are ordinary materials, and the two numbers are 2.8 eV apart.

The pattern is not random. Greczynski measured each sample's work function by ultraviolet photoelectron spectroscopy immediately after the XPS run, and found that the C 1s binding energy tracks it. Add the two together and the scatter collapses:

In fact, the sum of the C 1s binding energy and the sample work function is equal to 289.58 ± 0.12 eV.

A standard deviation of 0.12 eV, which he notes is smaller than the instrumental resolution. That is a tight relationship, and it has a physical reading: adventitious carbon is not electronically equilibrating with the sample. It is aligning to the vacuum level, so its apparent binding energy floats with whatever the substrate's work function happens to be. Low work function pushes the C 1s peak up, high work function pulls it down. Magnesium has a low work function. Molybdenum nitride has a high one. The numbers land exactly where that model says they should.

He also forecloses the obvious objection. If the spread were caused by differential charging in surface oxide layers, shifts should go one direction, toward higher binding energy. Many of his samples sit below 284.8 eV, which differential charging cannot produce. His conclusion is that referencing to adventitious carbon requires measuring the work function every time, and that this holds "irrespective of whether samples are measured grounded or insulated from the spectrometer."

The case for: 1,237 samples and 5 years of routine work

Against that sit two large datasets from multi-user analytical facilities, which is to say from labs that run other people's samples all day.

Mark Biesinger at Surface Science Western reviewed 1,237 samples submitted over five years. For the 117 of them that also carried an independent secondary reference, letting him check the answer, he found an average adventitious carbon C 1s binding energy of 284.91 eV with a standard deviation of 0.25 eV. Across 522 assessed cases, using adventitious carbon "gave satisfactory and meaningful results in 95%" of them, provided it was combined with sanity checks such as the Auger parameter and known peak-fitting routines.

David Morgan at Cardiff repeated the exercise independently at a second facility, on two different spectrometers, with five years of his own data. His conclusion:

Ultimately, for the majority of XPS analysis where samples are electrically isolated from the spectrometer, charge referencing using a value of 284.8 eV for adventitious carbon is still the most appropriate choice.

These are not casual defenses of a lazy habit. Both are large, traceable datasets with independent cross-checks, and Morgan's peak-fitting model was developed in-house yet came out close enough to Biesinger's that he calls the similarity striking.

So: 0.25 eV of scatter, or 2.89 eV? Both, apparently.

The methods sections explain most of it

Read how each lab mounts a sample and the picture changes.

Greczynski's samples are thin films on conducting substrates, deliberately grounded. He states plainly that "the charge neutralizer was not used in any of the reported experiments," and that valence band spectra were taken to confirm the Fermi level cutoff sits at zero, "thus proving that the samples are in good electrical contact to the spectrometer."

Morgan does the opposite on purpose. His samples are "mounted on a double-sided adhesive tape attached to a glass slide to float them from the spectrometer," with the surface potential then controlled by a charge compensation source. Biesinger reaches the same conclusion explicitly, writing that his work "demonstrates that electrical isolation (floating) of mixed insulating/semi-conducting/conducting samples significantly improves outcomes by mitigating differential charging issues."

These are two different electrical configurations, and they are not measuring the same quantity. On a grounded conducting film, the sample's own work function is well defined and varies enormously between magnesium and molybdenum nitride, so a contamination layer that aligns to the vacuum level will visibly slide around. On a floating insulator under a flood gun, the surface potential is being set by the neutralizer rather than by the substrate, and much of that variation is suppressed.

That is the practical resolution, and it is more useful than picking a winner: the validity of 284.8 eV depends on how your sample was mounted, and that is the one detail methods sections routinely omit.

Where it stays genuinely unresolved

I want to be careful not to tidy this up more than the literature does.

Greczynski does not accept the grounded-versus-floating split as a get-out. His stated conclusion covers both configurations explicitly. His argument is that vacuum-level alignment is a property of the adventitious carbon interface itself, not an artifact of how the sample is clamped, and that a flood gun setting the surface potential does not make the reference physically meaningful, it just hides the variation behind a different unknown. Nothing in the Biesinger or Morgan datasets directly refutes that, because neither measured work functions.

What Morgan offers instead is a practicality objection, and it is a fair one. Of the capping-layer and work-function approaches Greczynski proposes, he writes that "such sample preparation and measurements are impractical for a modern high-throughput analysis laboratory." A facility running dozens of external samples a week is not going to add a UPS work-function measurement to every job.

So the honest state of play is: for grounded conducting samples the 284.8 eV convention is demonstrably unsafe and the size of the error is material-dependent and can exceed a real chemical shift. For floating insulators analyzed with charge compensation and cross-checked against a secondary reference, two large facility datasets say it works about 95 percent of the time. Whether the underlying physics makes the second case merely lucky is still being argued in print.

What this costs when it goes wrong

The reason this is worth caring about outside the XPS community is that mis-referencing does not announce itself. It produces a clean spectrum with confident peak labels that are all shifted together, and the shift then propagates into databases that everyone else cites.

Greczynski points at exactly that, noting the spread of binding energies reported for identical chemical states in the literature: 2.5 eV for the Zr 3d_5/2 peak of zirconia, 2.3 eV for Al 2p in alumina, 1.6 eV for Na 1s in sodium chloride. Those are supposed to be single, well-defined values. A database range of 2.5 eV for a common oxide is not chemistry, it is accumulated referencing error.

Morgan's data shows the same problem from the other side. On a series of ZSM-5 zeolites, calibrating to the C 1s signal produced variability "greater than an expected experimental uncertainty of ca. 0.2 eV," while calibrating to the silicon signal instead moved the answers around by up to 0.9 eV. Same spectra, different reference choice, different conclusions about the material.

What to do about it

The three papers converge on more practice than their disagreement suggests.

Report the configuration. Whether the sample was grounded or floated, and whether charge compensation was on, is not housekeeping detail. It determines how much your reference is worth, and it is usually missing.

Use a second anchor whenever one exists. Biesinger's 95 percent figure is explicitly conditional on cross-checks, not on adventitious carbon alone. The Auger parameter is the strongest option because it is independent of both charging and reference level. A stable substrate peak or a well-characterized support works too.

Know the exceptions. Morgan lists them: polymers are better referenced at 285 eV for comparison against Beamson and Briggs, siloxane-contaminated surfaces put C 1s near 284.4 eV, heavily graphitic material sits near 284.5 eV, and for a genuinely conducting sample with a visible Fermi edge you should be using the Fermi edge.

Calibrate the instrument separately from referencing the sample, and do not confuse the two. Both Morgan and Greczynski calibrate against sputter-cleaned metal standards per ISO 15472, with metallic gold 4f_7/2 at 83.96 eV and copper 2p_3/2 at 932.62 eV. That proves the energy scale is linear and correct. It says nothing about the unknown offset on the next insulating sample you load, which is a separate problem and the one this whole argument is about. It is the same distinction I drew in the report on what NIST traceable actually buys you.

And if you are reading rather than running the experiment: when a paper reports a binding energy shift of a few tenths of an eV as evidence of a chemical change, check what it referenced to and how the sample was mounted. That shift may be real. It may also be the substrate's work function.

What I could not confirm

I read Morgan's paper and Biesinger's paper in full as published PDFs. For Greczynski, the version I read in full is the author's arXiv preprint, arXiv:2405.10919v1, deposited 17 May 2024. The abstract matches the published Applied Surface Science article word for word, but I did not access the publisher's version of record, so I cannot rule out changes made in production.

I have not read the ISO 15472 standard itself. The calibration values quoted here are as reported in Morgan's methods section.

I did not inspect the supplementary sample list accompanying Greczynski's paper, so the 360-specimen composition is as described in the text. I also have not seen a study that directly tests the central question raised here, which is whether work-function tracking of adventitious carbon persists on deliberately floated insulators under active charge compensation. As far as I can tell that experiment has not been published, and it is the one that would settle this. If it appears, this report needs updating and I will note the change here with a date.