[DFTB-Plus-User] Hessian calculations and time limit
Mat Toliday
djlets2004 at yahoo.co.uk
Mon Feb 24 21:39:43 CET 2020
It is hard to say with no knowledge of the machine you are running on, the version of DFTB+ you are using and how it was compiled. You might be able to run it using MPI parallelization instead but you should probably ask someone who knows you computer system and preferably uses DFTB+ on it. It may be working just fine.
Sorry I can't be of more help.
On Monday, 24 February 2020, 17:47:09 GMT, Alessandro LANDI <alelandi1 at unisa.it> wrote:
Thanks for the suggestions.One last question: you said " Check that DFTB is using all the cores, you are using OpenMP (Parallel>UseOMP = Yes) so you might not be."
Should I not use this feature? Should I set something different in the input to be sure all the cores are used?
Thanks again.
Alessandro
Il Lun 24 Feb 2020, 18:16 Mat Toliday via DFTB-Plus-User <dftb-plus-user at mailman.zfn.uni-bremen.de> ha scritto:
That looks quite normal, the Hamiltonian>Mixer mixing parameter in the dftb_in.hsd file might be a bit large, you could try using the default value and you could try reducing the SCCtolerance to 1e-8. It should still work and may be a little faster but I don't think this will speed up calculation enough. Check that DFTB is using all the cores, you are using OpenMP (Parellel>UseOMP = Yes) so you might not be.
Yours,Mat
On Monday, 24 February 2020, 17:02:55 CET, Alessandro LANDI <alelandi1 at unisa.it> wrote:
Hello,thank you for your reply. I have attached the input file I am using for Hessian calculation. I am using the values a senior researcher in my group suggested me (can you please check the attached file and tell me if you think they are too strict?)
In particular I set SCCTolerance = 1e-10. Each geometry step requires 11-15 iSCC cycles, if I understand the output correctly (I reproduce part of the output in the following)
*** Geometry step: 150 iSCC Total electronic Diff electronic SCC error
1 -0.13686329E+04 0.00000000E+00 0.21400324E-04
2 -0.13686329E+04 0.91858965E-10 0.21229086E-04
3 -0.13686329E+04 0.72759576E-11 0.11813509E-04
4 -0.13686329E+04 -0.97543307E-10 0.30138133E-05
5 -0.13686329E+04 -0.21145752E-10 0.74044473E-06
6 -0.13686329E+04 -0.47748472E-11 0.24149976E-06
7 -0.13686329E+04 -0.45474735E-12 0.14577856E-06
8 -0.13686329E+04 -0.68212103E-12 0.38308579E-07
9 -0.13686329E+04 0.68212103E-12 0.16315258E-07
10 -0.13686329E+04 -0.45474735E-12 0.57164407E-08
11 -0.13686329E+04 0.22737368E-12 0.21716895E-08
12 -0.13686329E+04 0.00000000E+00 0.70770501E-09
13 -0.13686329E+04 -0.22737368E-12 0.25188207E-09
14 -0.13686329E+04 -0.22737368E-12 0.10402701E-09
15 -0.13686329E+04 0.45474735E-12 0.31696423E-10Total Energy: -1328.7955553888 H -36158.3668 eV
Total Mermin free energy: -1328.7955553888 H -36158.3668 eV
Volume: 0.752806E+05 au^3 0.111554E+05 A^3
Pressure: -0.398475E-04 au -0.117235E+10 Pa
Best regards,Alessandro
Il giorno lun 24 feb 2020 alle ore 16:51 Mat Toliday via DFTB-Plus-User <dftb-plus-user at mailman.zfn.uni-bremen.de> ha scritto:
Hello,As far as I know you can not restart the second derivatives calculation. However, the Hessian matrix is formed by shifting each atom in the positive and negative Cartesian directions and calculating the force. The matrix equation is: F_i = H_ij * d_j, where F is force, H is a second derivative, d a small displacement and i and j are indices for the degrees of freedom. If you were to add your delta value to an atomic coordinate and then calculated the force vector you could then calculate the Hessian entries yourself using H_ij = F_i/d_j. You should probably use central finite differences, calculating the forces for both + and - d_j values and calculate the difference using the forces from the relaxed unit cell. As far as I can tell this is what the SecondDerivatives driver does.I am a little surprised it is taking too long, what sccTolerance are you using and how many SCCcycles does each step take?Yours,Mat
On Monday, 24 February 2020, 11:17:43 CET, Alessandro LANDI via DFTB-Plus-User <dftb-plus-user at mailman.zfn.uni-bremen.de> wrote:
Dear DFTB+ users,I am trying to calculate the normal modes for a supercell of rubrene crystal, consisting of 1120 atoms. I have optimized the supercell and now I am computing the Hessian using in the input:
Driver = SecondDerivatives {
Atoms = 1:-1
Delta = 1E-5
}
Problem is, I am working on a remote pc where calculations cannot last more than three days and it seems that, even using up to 80 cores, it is not enough to finish the hessian calculation in time. In light of this problem, I am wondering :1) Is it possible to evaluate the second derivatives in more than one run? (Like the geometry optimization where I can run a subsequent calculations starting from the last geometry of the previous calculation...)
2) If this is not possible, can you suggest me another workaround to solve this problem?
Thank you for any help you can give me
Best regards,
Alessandro
--
Alessandro Landi, Assegnista di Ricerca
Dipartimento di Chimica e Biologia "Adolfo Zambelli"
Università degli Studi di Salerno
Via Giovanni Paolo II, 132 - 84084 - Fisciano (SA)
Phone number 089969390_______________________________________________
DFTB-Plus-User mailing list
DFTB-Plus-User at mailman.zfn.uni-bremen.de
https://mailman.zfn.uni-bremen.de/cgi-bin/mailman/listinfo/dftb-plus-user
_______________________________________________
DFTB-Plus-User mailing list
DFTB-Plus-User at mailman.zfn.uni-bremen.de
https://mailman.zfn.uni-bremen.de/cgi-bin/mailman/listinfo/dftb-plus-user
--
Alessandro Landi, Assegnista di Ricerca
Dipartimento di Chimica e Biologia "Adolfo Zambelli"
Università degli Studi di Salerno
Via Giovanni Paolo II, 132 - 84084 - Fisciano (SA)
Phone number 089969390 _______________________________________________
DFTB-Plus-User mailing list
DFTB-Plus-User at mailman.zfn.uni-bremen.de
https://mailman.zfn.uni-bremen.de/cgi-bin/mailman/listinfo/dftb-plus-user
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20200224/3c22f9ad/attachment-0001.htm>
More information about the DFTB-Plus-User
mailing list