MIX_CL_TRUNC + ADMM in cp2k development branch

[Nicholas Winner]

Hello all,

I need to make use of range separated hybrid functionals for periodic systems, so I have pulled the latest development version of cp2k from github which adds this support to admm module (in hfx_admm_utils.F).

To the best of my knowledge, I have implemented it correctly, but the difference in final energy between PBE0 and MIX_CL_TRUNC is ~7 Ha, which seems rather large. When going from GGA -> PBE0 the energy difference <1 Ha. Obviously there should be some change when changing the interaction potential, but this has me considered.

GGA:  ENERGY| Total FORCE_EVAL ( QS ) energy (a.u.):             -983.840366364871670 PBE0 : ENERGY| Total FORCE_EVAL ( QS ) energy (a.u.):             -983.680119471562080

MIX_CL_TRUNC: ENERGY| Total FORCE_EVAL ( QS ) energy [a.u.]:    -974.290021294811368

Here is the XC section I am using to activate MIX_CL_TRUNC with ADMM. Can someone please sanity check and tell me if this is appropriate? 

                &XC

                        &XC_FUNCTIONAL PBE

                                &PBE

                                        PARAMETRIZATION ORIG

                                        SCALE_C 1

                                        SCALE_X 0.665 # 1 - alpha - beta

                                &END PBE

                        &END XC_FUNCTIONAL

                        &HF

                                FRACTION 1

                                &SCREENING

                                        EPS_SCHWARZ 1e-07

                                        EPS_SCHWARZ_FORCES 1e-06

                                        SCREEN_ON_INITIAL_P True

                                        SCREEN_P_FORCES True

                                &END SCREENING

                                &INTERACTION_POTENTIAL

                                        POTENTIAL_TYPE MIX_CL_TRUNC

                                        CUTOFF_RADIUS 1.8

                                        T_C_G_DATA t_c_g.dat

                                        SCALE_LONGRANGE 0.085 # alpha

                                        SCALE_COULOMB 0.25 # beta

                                        OMEGA 0.94486

                                &END INTERACTION_POTENTIAL

                                &LOAD_BALANCE

                                        RANDOMIZE True

                                &END LOAD_BALANCE

                                &MEMORY

                                        EPS_STORAGE_SCALING 0.1

                                        MAX_MEMORY 2000

                                &END MEMORY

                        &END HF

                &END XC

Best,

Nicholas Winner

[augusti...@chem.uzh.ch]

Dear Nicholas,

your XC input looks fine to me, but I have a few remarks. The cutoffs of the truncated Coulomb potential and the long range erf functions are very short. We typically use cutoffs that are around 6 Angstrom, which ensures that results are close enough to a calculation with standard Coulomb potential. Unless you are restarting your calculation from a well converged GGA level calculation, I would refrain from using the SCREEN_ON_INITIAL_P keyword, as the ATOMIC_GUESS starting point might not be good enough for porper screening. Finally, if you are using a range spearated HF potentials, you might want to combine it with a range separated DFT exchange functional as well.

I assume you set up the ADMM section properly and made a sensible choice of primary and auxiliary basis sets. Without your full input file however, this is impossible to say whether there is an issue there.

Hope this helps. Best,

Augustin

----...@googlegroups.com wrote: -----

To: "cp2k" <cp...@googlegroups.com>
From: "Nicholas Winner"
Sent by: cp...@googlegroups.com
Date: 10/28/2020 05:14PM
Subject: [CP2K:14133] MIX_CL_TRUNC + ADMM in cp2k development branch

--
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/06e0d703-46a0-4e05-873a-3852d52f8d66n%40googlegroups.com.

[Nicholas Winner]

Thanks for the reply. I do restart from GGA; I am also using a range separated functional as the hf_admm_utils.F module actually takes the coulomb and longrange scales and uses them to activate PBE_HOLE_TRUNC_LR and XWPBE sections. 

On the point of the short cutoff radius: there is actually a little uncertainty in the ideal cutoff radius. In one sense, we would like to have the truncated PBE0 converged w.r.t full 1/r, but there is some literature evidence that the PBE0 form should be treated as a short-range interaction, and the cutoff should be short. Since the short and long range terms have different ideal mixing fractions, if we extend PBE0 cutoff out too far (like 6-8Å), then while we are reproducing exact 1/r pretty well, we are applying short-range mixing fraction beyond where we should. 

Anyway, that's just an aside if you're interested. Thanks for the assistance.

-Nick