Pourquoi ne pas utiliser simplement la somme des carrés des écarts, éventuellement pondérée par les incertitudes point par point, si seul le spectre théorique change cela donnera une idée assez fiable.

Savoir si un modèle est _significativement_ meilleur que l'autre est en revanche plus difficile.

Pour toi il est difficile de savoir quel serait le meilleur spectre thlorique malgrès que l'écart pour l'un spectre theorique par rapport au spectre theorique soit plus faible qu'un autre théorique?

Tout dépend à quel point plus faible, bien sûr, mais sur le principe oui : ne serait-ce que parce que plus il y a de paramètres dans le modèle, mieux on ajuste, même si les paramètres ajustés supplémentaires n'ont aucun intérêt.

De façon plus générale, comment s'assurer que le modèle ne marche pas mieux simplement du fait des hasards du bruit expérimental qui tombent bien avec ce modèle et pas avec l'autre, alors qu'ils sont en fait aussi bons ?

Fluo, transmission, détection d'électrons ?

Chers amis, pourquoi faisez ce discussion en francais? Au fait, je voudrais proposer ...  I would like to propose to keep the English language so that all others may follow your discussions easier.

In context to the question by Jean-Baptiste: I always find it very difficult to judge the qualtiy of a XANES calculation. Of course one can try to apply some least square fitting routines in order to optimize the fit - but I would propose always to keep in mind the minimization of the energy of the atomic arrangement e.g. by DFT calculations. Just by moving around the atoms and calculating the XANES until the spectrum "looks good" is certainly not the alternative ... enjoy the week-end, Dirk

Theoretical XANES spectra are particularly calculated for qualitative analysis, and comparison on relative scale. Multiple scattering formalisms embedded in FEFF8 or later versions can be used for XANES calculation. However, you need to model the geometry of studied material for FEFF calculations. For modeling and refining your model, you can use ATOMS or XFIT. I would prefer XFIT that allows to refine coordinates taking your experimental XAFS into consideration. For quantitative comparison, the FDMNES program (modified version: Bourke, J. D., C. T. Chantler, and C. Witte. "Finite difference method calculations of X-ray absorption fine structure for copper." Physics Letters A360.6 (2007): 702-706. ) works more reliably.

The question is interesting and challenging and in principle the answer is quite easy.

You can run a fit as already possible with some SW packages. However, because at variance of EXAFS, for the XANES technique there are no equation associated, to answer to your question properly I need to know why you need a single parameter to evaluate XANES simulations. Can you clarify what is the problem you need to solve with a comparison based on a single parameter? 

Take care that just changing the range of the XANES you can weight different contributions (e.g., electronic vs. structural) and get a different feeling about the comparison.  

I agree with Dirk, the structure should have physical/chemical sense.

A few comments:

Though some molecular modeling may be useful (see the links), but it has to be well justified.

DFT modeling can be used with prudence. Something so simple as water, or metal hydration, has proven to be not so simple, and it's an example of the complexities in guessing "what is the structure" of something.

The danger in  computational chemistry vs chemical synthesis, is that any input coordinates will give an output spectrum, no matter if the molecule can ever exists.

Article The Catalytic Conversion of Thiophenes over Large H-ZSM-5 Cr...

Article XANES Study of the Carboxylate Binding Mode in Two Pterin Hydroxylases

Article Simulating the XANES of metalloenzymes - a case study

Interesting question.  This would ideally mean an analysis based on a Chi calculation and I don't think the accuracy of the various calculations is as advanced as say the EXAFS packages i.e simply,. different calculations of the same molecule different results. 

  Perhaps one could use different measures e.g. a differential or the Pendry (EXAFS Theory PRB 11 (1975) 2795-2811, EXCURVE package ) reliability factor based on the logarithmic derivative of the intensity used in low energy electron diffraction (https://en.wikipedia.org/wiki/Low-energy_electron_diffraction and references therein). 

I agree with Ana, and Dirk and the model needs to make chemical and physics sense e.g. the variation and number of parameters that make sense. 

Of course it is possible! There is a FitIt program specially designed for XANES ( see the first reference). It allows you to quantitavely compare calculated and experimental spectra. Moreover, you can use it to determine the optimal atomic and electronic parameters using simulated XANES. For this you can look through the corresponding part of our recent work (second reference).

Article FitIt: New Software to Extract Structural Information on the...

Article Temperature- and Pressure-Dependent Hydrogen Concentration i...