Dear Michel, First of all thank you very much for your help with this! One small comment: I thought that adding an imaginary part to the potential in EXCHANGE would effectively damp out waves at large distances. So, I was expecting that the XANES would "converge" at a certain cluster size, after which no changes of the fine structure would occur. If that's correct, does it mean that I should choose a larger Vi? Thanks again for you help. Lena Michel Jaouen wrote:

Dear lena,

...

I have a question regarding the convergence of XANES calculations. I'm currently focusing on the O-K edge of perovskite oxides, in particular SrTiO3 and found that there are still fine-structure changes for large (9.5?) clusters. Attached is my *.inp file for the 9.5? cluster and a summary of the XANES for increasing cluster sizes from 6.5? to 9.5?. You can see that there are still significant changes as the cluster size is increased from 8.5 to 9.5?. In particular, I find a double peak feature in the first 5eV after the edge onset, which is not observed experimentally.

It is a normal behavior: when increasing the cluster's size, you are adding more MS paths and then it appears coresponding peaks. It is a general trend for xanes:the measured signal is a probe up to a given distance for MS, a distance you didn't know a priori. Thus the game is to start from a rather small cluster, compute the xanes, compare to the experiment. If some features are missing, increase the size and so on till you obtain (if lucky) all experimental fine structures. On the opposite, as soon as it appears fine structures which don't exist in the experiment reduce the size.

...

Do you have any suggestions how to obtain a better convergence?

Secondly, if you compare the simulated XANES with experimental results (see attachment: EELS STO.png), I'm not able to match all three main peaks, even by scaling the energy scale. Are the deviations I'm seeing within the accuracy of FEFF or do I need to improve my *.inp file?

I have been working a lot on STO and have many exchange about this material with John Rehr. In fact you can obtain a quite good match with the experiment (peaks' positions) if you reduce the lattice parameter by 5% (RMULTIPLIER=0.95, see the attached figure). Of course such a reduction is unphysical but from John's own words :"Also regarding SrTiO3, I'm now thinking that the self-energy might be responsible for the need to use an RMULT 0.95." So either you the Rmultiplier trick or you try the feff8.5 version that includes some improvements to the self-energy (using low-loss and the sum rules for normalization). In any case, feff has to much intensity at the edge onset (sigma*) for light elements like Oxygen.

Two other comments: why adding a real part vr=1.4eV? I don't find this necessary here from my numerous calculations (it is responsible for the hump at the bottom of the pi* which doesn't exist in the experiment: a positive vr induces a red-shift for the spectrum, look at the ldos). I suggest also not to use the default for lmax (-1), but instead lmax=2 for all atoms (spd basis). You can also try to go up to lmax=3 (spdf basis) because Sr is quite heavy and it can help in some case (for pure Ni for instance, to reproduce the K edge, you must go up to f states).

I hope it can help you.

Best regards,

Michel ------------------------------------------------------------------------

_______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit

-- --------------------------------- Lena Fitting Kourkoutis E13 Clark Hall Cornell University Applied and Engineering Physics Ithaca, NY 14853 phone: 607-255-0654 e-mail: lf56@cornell.edu ---------------------------------