TURBOMOLE Users Forum
TURBOMOLE Modules => Statpt => Topic started by: patrikrydberg on November 28, 2011, 10:32:53 AM
-
What's the real key-word for updating the hessian at every X steps during geometry optimizations using statpt?
in different version of the manuals it has different names.
I've found the following
hessfreq
hssfreq
hssfrq
If I wan't the update to be equal to a full hessian calculation with aoforce, do I need to change any other keyword as well?
-
Hi,
only hssfreq is appears in the manual. define writes just that and statpt only understands that.
Where did you find the other syntax???
Cheers,
Arnim
-
in manuals for older versions there are other spellings.
-
and still using "hssfreq 1" leads to a normal powell hessian update, not a full frequency calculation at each step.
There is no mentioning of this keyword in statpt.out, does that indicate there is something wrong?
-
Hi,
actually, it is a bit confusing. statpt interprets the hssfreq keyword and sets calchess in optinfo. From optinfo jobex decides, if an aoforce calculation is done or not.
It can be check direktly from the standard output stream, if an aoforce calculation is done or not. A message from 'force' should appear.
For hssfreq 1 it should look like this:
convgrep will be taken out of the TURBODIR directory
ridft ended normally
OPTIMIZATION CYCLE 1
rdgrad ended normally
force ended normally
statpt ended normally
ridft ended normally
OPTIMIZATION CYCLE 2
rdgrad ended normally
force ended normally
statpt ended normally
ridft ended normally
OPTIMIZATION CYCLE 3
rdgrad ended normally
force ended normally
statpt ended normally
ridft ended normally
...
Best regards,
Arnim
-
I don't get that message.
Found that the hessian from my last aoforce calculation has not been saved to a file, but only saved as a projected hessian in the control file. Could this possibly be the problem?
-
recomputed the hessian and saved it to file, still no success
I'm using version 6.3.1 with B3-LYP, no RI approximation.
Could there possibly be some keyword conflict? i.e. hssfreq does not work if I use something else at the same time?
So far my output from the ongoing optimization is
dscf ended normally
grad ended normally
grad ended normally
statpt ended normally
dscf ended normally
grad ended normally
grad ended normally
grad ended normally
grad ended normally
0 -------------------------------------FORTRAN server ends
could the extra "grad" lines indicate something?
-
You probably are running it with PARA_ARCH=SMP. Since grad is parallelized only with MPI, all task will signal "grad ended normally", while the SMP parallelized codes dscf and aoforce only write it out once. This shouldn't be a problem. On a test system from me the aoforce call are working properly with your settings. Could you send your example to turbomole@cosmologic.de. Maybe it is easier to see what is happening then.
Arnim
-
yes I'm running with PARA_ARCH=SMP.
Since my current run might be working, I'll await the results. It seems to be running much fewer iterations / hour.
I'll post the results of adding
$hessian file=hessian
to the initial frequency calcuation here once the calculation converges.
-
$hessian file=hessian solved my problem