No agreement between the LR and RT absorption spectra

[Natalia K]

Dear community,

I am calculating the absorption spectra for a silver chain using both linear response (TDDFPT) and real-time propagation implemented in CP2K 9.1. Within the RT propagation, I apply a delta kick in all three directions and then take the Fourier transform of the dipole moment and get the average (over three directions) spectrum that is in a very good agreement with the literature. In particular, with another GGA linear-response calculation with the Gaussian code (image attached, top - my results, bottom - reference). Within the TDDFPT approach, I calculate the oscillator strength which should give the same frequencies as the real-time approach, but it doesn't. In the figure attached, the oscillator strength is plotted with impulses. I tested many different parameters but LR never agrees with RT. I also tested a simpler system, a water molecule, and got a better agreement between the two methods, but still not as perfect as here, for instance: https://nwchemgit.github.io/RT-TDDFT.html#absorption-spectrum-of-water

I would appreciate any help or suggestion. I attach both RT and LR input files used to obtain the results on the figure.

Best regards,

Natalia

[Victor Volkov]

Dear Natalia:

If I understand right, under the

&GLOBAL
  PROJECT Ag6
  RUN_TYPE RT_PROPAGATION
  PRINT_LEVEL LOW
&END GLOBAL

you do not compute properties of UV optical transitions 

but a dielectric response in the ground state.

Victor

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/c1c50acf-05be-49c1-80d3-da0dc60210c1n%40googlegroups.com.

[Beliz Sertcan]

Hello Natalia,

Regarding the mismatch between CP2K and Gaussian TDDFT results, once should use the TDA keyword in Gaussian instead of TD to get an exact match as CP2K employs the Tamm-Dancoff approximation. Furthermore, I would highly recommend turning on the Poisson section and setting the solver to MT instead of analytical.

    &POISSON

       PERIODIC NONE

       POISSON_SOLVER MT

    &END

Since you are working with a heavy element I would also consider using a bigger basis such as TZVP maybe even TZV2P. With these additions I think you will be able to match the RTP spectrum better, but I would not expect a perfect match. In my opinion what you have with the positions of the sticks is not so bad.

Best,

Beliz

[Natalia K]

Dear Beliz,

thank you for your answer. As I said, I tried changing many different parameters in TDA-CP2K. For instance, MT Poisson solver changes the spectrum by a tiny bit, as does the TZP basis set. The parameter change affects more the higher frequency and not the low frequencies. The XC functional does affect a lot the results, but still, the difference remains huge when comparing to RTP+Fourier transform (the latter agrees with the literature, so I trust it) with the same parameters.

I do not perform Gaussian calculations myself, I just compare with the literature that employs Gaussian, so I cannot check the TD vs TDA.

(Just a side note that with SIESTA, I do get agreement between the linear response and real-time propagation with delta kick for the absorption spectra of the same system. So, is it the Tamm Dankoff approximation that does something weird??)

Natalia