Dear Giuseppe, All,
Thank you very much for your comments! > 1) Do you think that further effort in the implementation of k_points > coupled with range-reparated (or even optimally-tuned-like) xc > functionals working with ACE and ecutfock in turbo_lanczos/davidson > might provide such a valuable tool? I think yes! But it requires time and effort. Of course, there are issues as Oscar said, but I hope some solutions can be found. This is an interesting problem, and definitively it would be interesting to investigate it. But someone has to work full time on it, and not just during evenings and/or weekends. > 2) Do you think that such kind of calculation might show (after a lot > of work, admittedly) a speedup, with respect to a GW-BSE calculation > of the same system, so spectacular that the juice is worth the squeeze? I believe that yes. In fact, in the past I worked on TDDFT-EELS with (screened) hybrids and some preliminary results were very promising (but I did not finish that project due to lack of time and higher priority of other projects). Best wishes, Iurii -- Dr. Iurii Timrov Postdoctoral Researcher STI - IMX - THEOS and NCCR - MARVEL Swiss Federal Institute of Technology Lausanne (EPFL) CH-1015 Lausanne, Switzerland +41 21 69 34 881 http://people.epfl.ch/265334 ________________________________ From: users <[email protected]> on behalf of Oscar Baseggio <[email protected]> Sent: Thursday, April 16, 2020 8:59:36 AM To: Quantum ESPRESSO users Forum Subject: Re: [QE-users] safe to enable "k-point algorithm" in turbo_lanczos? Dear Giuseppe, I think that I could answer to the first question. In linear response the exx integral are different respect with to ground state calculation, because they contain the response wavefunction, so you need three different ACE operators (without response wfc, with response wfc, with conjugate response wfc); I’m not sure that it is convenient. Second problem, we have to think back an another algorithm to introduce the self-consisted cycle for ACE operator. But if you have some suggestions, we can call each other. Best regards, Oscar Baseggio Il 2020-04-15 23:59 Giuseppe Mattioli ha scritto: > Dear Iurii > The "TDDFT-solid state affair" seems to periodically resurface from > deep waters... Most of it started (at least for me) from a 2008 paper > by Izmaylov&Scuseria (J. Chem. Phys. 2008, 129, 034101) in which some > assumptions on the usage of range-separated functionals such as HSE > were made on the grounds of theory and of a (cumbersome...) example of > an H chain. There is now a vast literature on the tailoring of > appropriated range-separated hybrid functionals for solid-state TDDFT, > because we are stubborn and do not basically cope with the fact that > you can do (almost) whatever you want with TDDFT in the case of 0D > system, and that all these wonderful things are reduced to some sort > of poor "joint density of states" in the case of 3D systems! :-D > > I would be very happy to read a comment from you: after years of > painful applications of EXX to QE calculations, which I've pioneered, > with ACE and ecutfock we have now found a very honorable balance > between speed and accuracy of ground state calculations (with NC PPs, > in my experience, but this is another matter). The quest for a > possible missing rung of the Jacob's excited-state ladder between > epsilon.x and GW-BSE might be at hand. Thus: > > 1) Do you think that further effort in the implementation of k_points > coupled with range-reparated (or even optimally-tuned-like) xc > functionals working with ACE and ecutfock in turbo_lanczos/davidson > might provide such a valuable tool? > > 2) Do you think that such kind of calculation might show (after a lot > of work, admittedly) a speedup, with respect to a GW-BSE calculation > of the same system, so spectacular that the juice is worth the > squeeze? > > Thanks in advance to you and to everybody for any feedback > Giuseppe > > Quoting Timrov Iurii <[email protected]>: > >> Dear Michal, >> >> >>> @Iurii: the motivation (or rather - hope) is that TDDFT (or >>> perhaps even epsilon.x) will give us the correct trend, whilst being >>> much cheaper than GW/BSE (which we are trying as well). >> >> >> You can try to use the turboEELS code with a very small transferred >> momentum q. The code prints at the end the loss function >> -Im[1/eps(w,q)] but also Im[eps(w,q)]. In the long wavelength limit >> Im[eps(w,q->0)] gives you the optical absorption spectrum at the >> TDDFT level in the adiabatic approximation with a LDA/GGA functional >> (see e.g. Fig. 3 in Comput. Phys. Commun. 196, 460 (2015)). >> >> >> Best, >> >> Iurii >> >> >> -- >> Dr. Iurii Timrov >> Postdoctoral Researcher >> STI - IMX - THEOS and NCCR - MARVEL >> Swiss Federal Institute of Technology Lausanne (EPFL) >> CH-1015 Lausanne, Switzerland >> +41 21 69 34 881 >> http://people.epfl.ch/265334 >> ________________________________ >> From: users <[email protected]> on behalf of >> Michal Krompiec <[email protected]> >> Sent: Wednesday, April 15, 2020 7:54:32 PM >> To: Quantum ESPRESSO users Forum >> Subject: Re: [QE-users] safe to enable "k-point algorithm" in >> turbo_lanczos? >> >> Dear Oleksandr, Dear Iurii, >> Thank you very much, I understand that this code should rather stay >> disabled. @Iurii: the motivation (or rather - hope) is that TDDFT (or >> perhaps even epsilon.x) will give us the correct trend, whilst being >> much cheaper than GW/BSE (which we are trying as well). >> Best, >> Michal >> >> On Wed, 15 Apr 2020 at 18:39, Oleksandr Motornyi >> <[email protected]> wrote: >>> >>> Dear Michal, >>> >>> >>> The k-points algorithm of turbo_lanczos.x as it is does not give the >>> correct results if there is more than one single k-point. However >>> you can look up the Lanczos implementation in the thermo_pw module, >>> which works fine with k-points as far as I remember (of course bare >>> in mind what Iurii said about Adiabatic TDDFT). >>> >>> >>> Regards, >>> >>> Oleksandr >>> >>> On 15/04/2020 19:35, Timrov Iurii wrote: >>> >>> Dear Michal, >>> >>> >>> The k-points implementation of turbo_lanczos.x was not tested and >>> hence it is disabled. But most importantly, very likely there are >>> bugs (I think I know a few). Someone has to work on it and fix bugs. >>> But is there a motivation to do so? Adiabatic TDDFT with LDA/GGA is >>> known to be not accurate for the optical absorption spectroscopy of >>> solids. >>> >>> >>> Greetings, >>> >>> Iurii >>> >>> >>> -- >>> Dr. Iurii Timrov >>> Postdoctoral Researcher >>> STI - IMX - THEOS and NCCR - MARVEL >>> Swiss Federal Institute of Technology Lausanne (EPFL) >>> CH-1015 Lausanne, Switzerland >>> +41 21 69 34 881 >>> http://people.epfl.ch/265334 >>> ________________________________ >>> From: users <[email protected]> on behalf of >>> Michal Krompiec <[email protected]> >>> Sent: Wednesday, April 15, 2020 7:26:09 PM >>> To: Quantum Espresso users Forum >>> Subject: [QE-users] safe to enable "k-point algorithm" in >>> turbo_lanczos? >>> >>> Hello, >>> I'm trying to compute the spectrum of a crystalline insulator using >>> turbo_lanczos. I got the following error: >>> >>> Error in routine lr_readin (1): >>> k-point algorithm is not tested yet >>> >>> >>> >>> I guess I could just comment line 477 out in TDDFPT/src/lr_readin.f90 >>> to disable this error: >>> >>> ! K-points are implemented but still unsupported (use at your >>> own risk!) >>> ! >>> IF (.NOT. gamma_only ) CALL errore('lr_readin', 'k-point >>> algorithm is not tested yet',1) >>> >>> But how unstable is this code? Any reason not to try this? >>> >>> Thanks, >>> >>> Michal Krompiec >>> Merck KGaA >>> _______________________________________________ >>> Quantum ESPRESSO is supported by MaX >>> (www.max-centre.eu/quantum-espresso<http://www.max-centre.eu/quantum-espresso<http://www.max-centre.eu/quantum-espresso<http://www.max-centre.eu/quantum-espresso>>) >>> users mailing list [email protected] >>> https://lists.quantum-espresso.org/mailman/listinfo/users >>> >>> _______________________________________________ >>> Quantum ESPRESSO is supported by MaX >>> (www.max-centre.eu/quantum-espresso<http://www.max-centre.eu/quantum-espresso<http://www.max-centre.eu/quantum-espresso<http://www.max-centre.eu/quantum-espresso>>) >>> users mailing list [email protected] >>> https://lists.quantum-espresso.org/mailman/listinfo/users >>> >>> -- >>> Oleksandr Motornyi >>> PhD candidate >>> >>> Laboratoire de Solides Irradies >>> Ecole Polytechnique (Palaiseau, France) >>> >>> _______________________________________________ >>> Quantum ESPRESSO is supported by MaX >>> (www.max-centre.eu/quantum-espresso<http://www.max-centre.eu/quantum-espresso<http://www.max-centre.eu/quantum-espresso<http://www.max-centre.eu/quantum-espresso>>) >>> users mailing list [email protected] >>> https://lists.quantum-espresso.org/mailman/listinfo/users >> _______________________________________________ >> Quantum ESPRESSO is supported by MaX >> (www.max-centre.eu/quantum-espresso<http://www.max-centre.eu/quantum-espresso<http://www.max-centre.eu/quantum-espresso<http://www.max-centre.eu/quantum-espresso>>) >> users mailing list [email protected] >> https://lists.quantum-espresso.org/mailman/listinfo/users > > > > GIUSEPPE MATTIOLI > CNR - ISTITUTO DI STRUTTURA DELLA MATERIA > Via Salaria Km 29,300 - C.P. 10 > I-00015 - Monterotondo Scalo (RM) > Mob (*preferred*) +39 373 7305625 > Tel + 39 06 90672342 - Fax +39 06 90672316 > E-mail: <[email protected]> > > _______________________________________________ > Quantum ESPRESSO is supported by MaX > (www.max-centre.eu/quantum-espresso<http://www.max-centre.eu/quantum-espresso>) > users mailing list [email protected] > https://lists.quantum-espresso.org/mailman/listinfo/users _______________________________________________ Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso<http://www.max-centre.eu/quantum-espresso>) users mailing list [email protected] https://lists.quantum-espresso.org/mailman/listinfo/users
_______________________________________________ Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso) users mailing list [email protected] https://lists.quantum-espresso.org/mailman/listinfo/users
