Skip to first unread message
Martín Taccone
May 4, 2020, 7:23:57 PM5/4/20
to ipi-users
Hi everyone!
I'm trying to run a REMD with i-PI together with CP2K using PM6 semi-empirical method of a 122 atoms system. I'm using i-PI 2.0 and CP2K 7.1 versions. I'm having problems converging the scf wavefunction. I'm using pretty much default settings for the scf but the convergence value in the CP2K calculation doesn't seem to get down:
-----
SCF WAVEFUNCTION OPTIMIZATION
Step Update method Time Convergence Total energy Change
------------------------------------------------------------------------------
1 P_Mix/Diag. 0.40E+00 46.8 1.90017930 -404.6539254312 -4.05E+02
2 P_Mix/Diag. 0.40E+00 4.9 1.87393308 -424.9920521068 -2.03E+01
3 P_Mix/Diag. 0.40E+00 4.7 1.94613893 -161.6114774763 2.63E+02
4 P_Mix/Diag. 0.40E+00 4.9 1.75568543 -293.4195908037 -1.32E+02
5 P_Mix/Diag. 0.40E+00 3.5 1.92213422 -165.4224690531 1.28E+02
6 P_Mix/Diag. 0.40E+00 4.3 1.84401007 -233.3222524911 -6.79E+01
7 P_Mix/Diag. 0.40E+00 3.1 1.88584111 -234.2643488910 -9.42E-01
...
98 P_Mix/Diag. 0.40E+00 1.9 1.81931388 -72.1544623211 7.82E+00
99 P_Mix/Diag. 0.40E+00 1.7 1.80765244 -158.1586533059 -8.60E+01
100 P_Mix/Diag. 0.40E+00 1.9 1.80050542 -174.0201778781 -1.59E+01
Leaving inner SCF loop after reaching 100 steps.
-----
The thing is that when running the exact same input file in C2PK with ENERGY_FORCE run type (single point energy and forces calculation) and without the MOTION section, that connects it to i-PI, the scf cycle converges:
-----
SCF WAVEFUNCTION OPTIMIZATION
Step Update method Time Convergence Total energy Change
------------------------------------------------------------------------------
1 P_Mix/Diag. 0.40E+00 0.3 1.72151646 -409.1021350678 -4.09E+02
2 P_Mix/Diag. 0.40E+00 0.3 1.11796544 -442.3497706606 -3.32E+01
3 P_Mix/Diag. 0.40E+00 0.3 0.67150872 -466.4870457424 -2.41E+01
4 P_Mix/Diag. 0.40E+00 0.2 0.40117847 -482.4872620237 -1.60E+01
5 P_Mix/Diag. 0.40E+00 0.1 0.23924267 -492.6660070957 -1.02E+01
6 P_Mix/Diag. 0.40E+00 0.1 0.14708233 -498.9955145976 -6.33E+00
7 P_Mix/Diag. 0.40E+00 0.2 0.09109772 -502.8800229628 -3.88E+00
...
59 DIIS/Diag. 0.25E-05 0.2 0.00003984 -508.8626515084 1.46E-11
60 DIIS/Diag. 0.26E-05 0.2 0.00002457 -508.8626515084 -5.00E-11
61 DIIS/Diag. 0.28E-05 0.1 0.00000856 -508.8626515085 -4.82E-11
*** SCF run converged in 61 steps ***
-----
Also the time needed for each scf step when using i-PI is much higher than the ENERGY_FORCE calculation.
I thought that maybe when using DRIVER run type in CP2K some default values change for some reason, but I printed the explicit input settings in both cases and there are no differences. I have also tried changing some scf settings in the CP2K input but it didn't worked, so I'm kind of lost here. Could be a CP2K compilation problem? Or that the latest version of CP2K is not supported by i-PI?
I attach both CP2K input and output files and i-PI input file as well.
Thanks in advance and try to stay at home!
Martín.
Michele Ceriotti
May 6, 2020, 3:09:54 PM5/6/20
to ipi-users
I can't tell for sure, as you do not provide input.xyz, but my guess is that you provide input.xyz in angstrom, without specifying the units in the header.
For sure, you provide the cell parameters as if they were angstrom, but in i-PI *everything is in atomic units unless explicitly specified*
(exception being PDB where there's a clear standard that says units are angstrom)
Reply all
Reply to author
Forward
0 new messages