[CPMD-list] Energy components in PIMD
Axel Kohlmeyer
akohlmey at cmm.chem.upenn.edu
Wed May 14 18:35:58 CEST 2008
dear lukasz,
first off, let me commend you for being so careful
and persistent. this is giving a great example to
other people and it has been very educational to me
(and i hope for the many other "lurkers" on this list
as well).
LW> > since the machine i have available for these kinds of tests,
LW> > i cannot run that long. please see the first attached graph.
LW> > it does show a little slant now.
LW>
LW> Going down to 200 a.u. with e-mass seems to stabilize the drift in EKINC
LW> more or less (with the timestep of 3 a.u) - see attached graph. Do you
LW> think it's reasonable ? Would it be sane to go even to 150 a.u. and
LW> possibly reducing the time step to 2 a.u. (having the future PIMD
LW> simulations in mind) ?
hard to tell. your system is definitely a bit more sensitive
that the typical "simple-bulk-where-many-errors-cancel-out"
setup that people prefer to study for convergence analysis...
one thing that came to my mind, is that you may improve the
situation by turning the "outside" hydrogens into deuterium.
their bonded vibrations seem to be of pretty high frequency and
thus slowing them down by increasing their mass may help improving
the stability w.r.t. timestep and ficitious mass.
LW> > BTW: based on its "spectrum" i have changed coupling frequency
LW> > for the N-H chains to 2500 cm-1, if you look at the second
LW> > attached graph this seems a much more reasonable choice as it
LW>
LW> What do you call "spectrum" in this case ? - with FFT(EKINC) I get
LW> something different...
this was run though the fourier.x code and "temperature corrected".
you could generate the same with something like:
awk '{print $1, $2, 0.0, 0.0, $2, 0.0, 0.0}' ENERGIES > out
(echo out ; echo spec.dat ; echo 3.0 ; echo 300.0 ; echo 20000.0 ; echo 1 ) | fourier.x
and then plot columns 1 and 5 (columns 1 and 2 are what you probably see).
LW> > when you turn on path-integrals, you _will_ have to reduce the time step,
LW> > anyways (unless you run BO-dynamics).
LW>
LW> I will try to reduce the time step to 3 a.u. and the e-mass to 200 a.u.
LW> but increase ecutoff to 35 Ry (or even to more) in the case of VDB
LW> pseudopotentials - hopefully this will allow for longer PIMD run...
if you look at the timings then you should have seen that VDB potentials
are of no use to speed up the calculation. with 35ry or more and DUAL of
6 or more, you'll be faster using norm conserving pseudopotentials _and_
you avoid all potential problems from USPPs. that part of the code is
not as well optimized and debugged. we are still occasionally running
into USPP-related problems where code broke during the last major
rewrite of CPMD and nobody cared much to update the USPP part accordingly.
BTW: i made a test with BO dynamics. when using wavefunction
extrapolation and some other tricks (that are not yet in 3.11.)
i get very good energy conservation with the attached input,
which would bypass the problem of "tuning" the fictitious kinetic
energy alltogether. when running those simulations side-by-side
and fourier transforming the energy components one can also nicely
see how the "drag" from the fictitious electron-dynamics and the
deuteration of the "rim-hydrogens" affect the various vibrational
modes.
LW> There is one more issue regarding the COM. The second attached plot
LW> shows the conserved energy, that exhibits jumps every 2500 steps - that
LW> is every time I SUBTRACT COMVEL ROTVEL. The jumps occur on the fifth
LW> decimal place in Ha. Would you accept such behavior ? (I know one should
LW> not rescale velocities when using NH thermostat but the manual is
LW> equivocal on what is already a bad thing.)
you can rescale more often, say every 25th step? that would make
the disruptions smaller (but more frequent).
cheers,
axel.
p.s.: after going through so many tests and checking out so many
details. what do you think about summarizing your experiences and
adding it to the CPMD manual as a guide for other CPMD users?
LW>
LW> Regards,
LW> Lukasz
LW>
LW>
--
=======================================================================
Axel Kohlmeyer akohlmey at cmm.chem.upenn.edu http://www.cmm.upenn.edu
Center for Molecular Modeling -- University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582, fax: 1-215-573-6233, office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.
-------------- next part --------------
OPTIMIZE WAVEFUNCTION
RESTART WAVEFUNCTION COORDINATES VELOCITIES
RESCALE OLD VELOCITIES
QUENCH BO ELECTRONS
NOSE IONS MASSIVE
300.0 2500.0
&CPMD
MOLECULAR DYNAMICS BO
RESTART WAVEFUNCTION COORDINATES VELOCITIES ACCUMULATORS
RESTART EXTRAP LATEST
REAL SPACE WFN KEEP
CONVERGENCE ORBITALS
1.0e-6
EXTRAPOLATE WFN STORE
4
TIMESTEP
20.0
TEMPERATURE
300.0
MAXCPUTIME
85000
SUBTRACT COMVEL ROTVEL
25
TRAJECTORY XYZ SAMPLE
-10
ISOLATED MOLECULE
CENTER MOLECULE
STORE
1000
&END
&SYSTEM
ANGSTROM
SYMMETRY
8
CELL ABSOLUTE
13.0 13.0 8.0 0.0 0.0 0.0
CUTOFF
70.0
&END
&DFT
FUNCTIONAL BLYP
GC-CUTOFF
5.D-6
&END
&ATOMS
ISOTOPES
14.0067
12.0112
2.0000
1.0080
*N_MT_BLYP.psp KLEINMAN-BYLANDER
LMAX=P
4
1.480434744000 -1.261526338000 0.012486679000
-1.428836888000 -1.263268937000 0.020030520000
-1.482423266000 1.260607946000 0.012029901000
1.430161665000 1.261259358000 0.004480994000
*C_MT_BLYP_1.psp KLEINMAN-BYLANDER
LMAX=P
20
2.764321879000 1.617573818000 -0.000125061000
3.855614684000 0.719804456000 -0.000057419000
3.883340408000 -0.678014026000 0.004378686000
2.816852951000 -1.603192445000 0.010125751000
2.919542448000 -3.046779447000 0.014515219000
1.637225743000 -3.547145051000 0.019451505000
0.739381956000 -2.412298909000 0.018118048000
-0.677939805000 -2.411241253000 0.021787063000
-1.565841060000 -3.552172948000 0.027767194000
-2.851571375000 -3.063067365000 0.029521295000
-2.763080194000 -1.618905630000 0.024634604000
-3.853925457000 -0.720052370000 0.024562307000
-3.883570717000 0.678208069000 0.020129751000
-2.818027537000 1.604764172000 0.014380689000
-2.917826697000 3.049092800000 0.009981340000
-1.634343765000 3.546387751000 0.005051935000
-0.738698022000 2.409860388000 0.006396510000
0.678687181000 2.408624576000 0.002727905000
1.566148163000 3.550017372000 -0.003252572000
2.852089024000 3.061544408000 -0.005009259000
*H_MT_BLYP.psp KLEINMAN-BYLANDER
LMAX=S
12
4.829896620000 1.209449328000 -0.004160981000
4.874865030000 -1.131837359000 0.003273118000
3.847285348000 -3.609138791000 0.013924943000
1.339951403000 -4.589537686000 0.023583475000
-1.257219301000 -4.591087509000 0.030318008000
-3.775889801000 -3.630740840000 0.033747037000
-4.828126595000 -1.209793447000 0.028665970000
-4.875902354000 1.130204086000 0.021243303000
-3.843471723000 3.614955577000 0.010554882000
-1.334243713000 4.588096584000 0.000914849000
1.257025926000 4.588704910000 -0.005801358000
3.776400021000 3.629099090000 -0.009234599000
*H_MT_BLYP.psp KLEINMAN-BYLANDER
LMAX=S
2
1.216317908000 -0.004972921000 0.009118666000
-1.218997172000 0.003787130000 0.015400566000
&END
More information about the CPMD-list
mailing list