Firefly and PC GAMESS-related discussion club

Learn how to ask questions correctly  
We are NATO-free zone


Alex Granovsky

Dear Dawid,

I'm sorry for delay with my reply. Indeed, the use of DLCs for
geometry optimization (with or without constrains, this does not
matter) is the recommended way with Firefly. It is good you have
already figured this out.

As to torsion angle variations you observed, I'm sure that
there are no variations in frozen torsion at the geometries
along the energy optimization path itself. Please double check this.
At the same time, at distorted intermediate geometries used to
compute numerical gradients, violation of constrains is indeed
possible and is even required.

On a separate note, I have a comment concerning your use of XP mode
with Firefly. Upon examination of your previously posted input and
output (from system-blows-up.tar.gz archive) it seems to me you do
not use XP mode optimally. Indeed, the job was running in parallel
using 12 processes on a single host wn0361. There is nothing wrong
with this so far. However, np variable in $smp group was set to 24
and you requested plain XP mode.

As you can find in the output:

     Switching to XP mode with    12 groups of processes.
     Maximum size of group is      1 process(es).

This means that there were 12 XP groups and hence 12 XP master
processes. As requested, each XP master used 24 threads during
summation of XMCQDPT2 series, i.e., 288 threads in aggregate.
At the same time, there were only 56 cores available (again, this
number can be found in output). Use of 256 threads simultaneously
caused severe resource oversubscription resulting in serious
performance degradation. It would be much better to run this job
in parallel using e.g., 14 processes and requesting plain XP mode
with np=4 in $smp group. This would lead to 14*4 = 56 threads in
aggregate hence avoiding oversubscription and resulting in better
overall performance.

Hope this helps.

Kind regards,
Alex Granovsky

On Wed Mar 22 '17 11:15am, Dawid wrote
>Dear All,

>I have figured out how to do it.
>I attach my input file, so that others who struggle with this
>have a working example.

>By the way, I noticed that while the constraints generally work,
>the constrained dihedral somehow varies during optimization
>by less than 0.5 degrees. Is that related to the way the Z-matrix
>was automatically generated?

>Best wishes,
>Dawid Grabarek

>On Mon Mar 13 '17 11:49am, Dawid wrote
>>Dear Alex,

>>Thank you for helping me out with this.

>>Nevertheless, after all I have decided to do my constrained
>>optimization somehow different.
>>I'd like to provide the coordinates in XYZ format and constrain
>>one dihedral angle (torsion). My system blows up and I get
>>warnings on the singularity (if I understand it correctly) of the
>>G matrix.
>>Could you have one more look at my input and output, please?

>>Best wishes,
>>Dawid Grabarek
>>On Sat Mar 11 '17 8:30pm, Alex Granovsky wrote
>>>Dear Dawid,

>>>there are two errors in your original input file, the first is an
>>>extra line in the $DATA group and the second is too long input line
>>>in the $ZMT group (current limit is 80 symbols per line).

>>>I have attached fixed input file for your reference.

>>>Kind regards,
>>>Alex Granovsky
>>>On Thu Mar 9 '17 10:31pm, Dawid wrote
>>>>Dear Firefly Users,

>>>>I encounter issues with z-matrix definition in Firefly. I attach
>>>>my input and output files. Could you explain why I get this error,

>>>>Best wishes,

[ Previous ] [ Next ] [ Index ]           Tue Mar 28 '17 9:34pm
[ Reply ] [ Edit ] [ Delete ]           This message read 382 times