Page 1 of 1

runqmc stops without any error.

Posted: Sat Jan 07, 2017 4:55 pm
by Sharma_Omprakash
Dear users,

The runqmc stops before finishing the calculations without any error, instead it says "Job is finished" in OUT file. Please do needful in this regard.

CASINO VERSION="v2.13.602 [Mike Towler] (12 December 2016)"
OS : Ububtu 16.04
gfortran version 5.4.0
FFLAGS_opt = -Ofast -fprotect-parens -fpeel-loops -fcray-pointer

Re: runqmc stops without any error.

Posted: Sat Jan 07, 2017 5:36 pm
by Katharina Doblhoff
Please give more detais.
What is the last output it gives and what is the input? How do you know that it stops before the end?
Thanks

Re: runqmc stops without any error.

Posted: Sat Jan 07, 2017 5:51 pm
by Sharma_Omprakash
Hi Katharina

Here I have attached the input and out put file for electron-hole bilayer.
Thank
Rajesh

Re: runqmc stops without any error.

Posted: Sun Jan 08, 2017 7:09 am
by Sharma_Omprakash
How do you know that it stops before the end?
I have set the opt_cycles = 100 , but before it finish all cycles stopped at 5th cycle as you can see in the OUT file. And This happened many times.

Re: runqmc stops without any error.

Posted: Sun Jan 08, 2017 11:38 am
by Neil Drummond
Dear Rajesh,

I am running your example on four cores with the same random seed as you... I've got to 69 cycles so far without problems (see attached out file). If you diff the out file with yours you will see that the first VMC calculation is exactly the same. After that differences creep in, because optimisation is a very numerically sensitive procedure.

I am also running your example with debugging flags, and have reached 40 cycles without encountering problems.

Is it possible that some other issue on your computer caused the job to halt?

Best wishes,

Neil.

Re: runqmc stops without any error.

Posted: Sun Jan 08, 2017 11:58 am
by Neil Drummond
Dear Rajesh,

Note that you are using the same two-body terms between pairs of electrons and between electrons and holes. Since the electrons and holes have the same mass, it would be best to define a custom spin dependence with the pair groupings (1-1,2-2) and (1-2). Alternatively you could set the spin-dependence of U to 1, which gives the grouping (1-1) (1-2) (2-2), which is appropriate for the case where the electron and hole masses are different.

Best wishes,

Neil.

Re: runqmc stops without any error.

Posted: Sun Jan 08, 2017 5:08 pm
by Sharma_Omprakash
Dear Sir,
Thank you for your reply.
Is it possible that some other issue on your computer caused the job to halt?
I have been looking for it, I feel like this issue may be related to Ubuntu 16.04/gfortran5.4.0 since I have tried to run the same jobs on the other PC with the same configurations as mentioned in first post, but job halt on other PC too. Job sometimes complete but most of the time it stops, particularly when I see the job in the morning I found it stopped without any errors. I am also running some jobs on the workstation with ubuntu14.04, gfortran4.8.4 and casino version 1.13.473(1 march 2015) but on this workstation job never stopped like this.
it would be best to define a custom spin dependence with the pair groupings (1-1,2-2) and (1-2).
How can I define custom spin dependence?

Regards
Rajesh

Re: runqmc stops without any error.

Posted: Sun Jan 08, 2017 8:36 pm
by Neil Drummond
Dear Rajesh,

I've attached modified input files for the example you sent. The main changes I made are: (i) I defined a custom spin-dependence (see the block at the end of input) and used it in the U term; (ii) I added a plane-wave two-body p term to the Jastrow; (iii) I removed the QCUSP term from the Jastrow; (iv) I used varmin_linjas rather than madmin to optimise the wave function.

Hope this helps. Best wishes,

Neil.

P.S. I ran under Ubuntu 16.10, using gfortran 6.2.0.