I was trying to generate SQS for AB alloy with concentration (0.6-0.4 with 20 atom supercell) and (0.375-0.625 with 32 atom supercell) alloy for both FCC and BCC lattice. For 0.6-0.4 concentration and two body correlation up to first NN, mcsqs didn’t generate any sqs (in bestsqs.out file) even after running for 3days both in FCC and BCC supercell. For 0.375-0.625 concentration and two body correlation up to first NN, mcsqs generates the bestsqs but when I included the two body correlation up to 2nd NN same thing happens as the 0.4-0.6 concentration.
So, is it normal behavior and should it take more times for these concentration or is there any problem with my installation or bug?
This is unusual: mcsqs usually outputs at least a poor sqs quickly.
Maybe the code gets stuck trying to enumerate supercells.
How big is the mcsqs.out file? (# of lines)
As a test, you could also specify your own very short sqscell.out file and the -rc option and see if it speeds up.
Could you send me a .tar file with all your input files for debugging purposes?
I am extremely sorry for my late reply. I tried to specify a 32 atom supercell both for BCC and FCC at 0.375-0.625 concentration with two body correlation up to 2nd NN (mcsqs generates the bestsqs when the two body correlation was up to the 1st NN) but got nothing after running for 3 days. Don’t know whether I am doing any mistakes. I attached all the files. Please check.
I tried a few but all seemed to work so far.
Can you give me one specific subdirectory where the problem occurs?
Attached directory is for BCC at 0.375-0.625 concentration with pair correlation upto 2nd NN. Even after specifying the supercell of 32 atom it didn’t give any result.
I just noticed in your submit.sh that you are using the command
mcsqs -rc<sqscell.out -n=32 >sqs32low.outIt should be
mcsqs -rc -n=32 >sqs32low.outCould that be the problem?
It is working on my end.
Also, when you send a file make sure the extension reflects the type. (you used .tar.gz but it was not gzipped. It was a .tar )
Sorry for the extension. I tried according to your suggestion
mcsqs -rc -n=32 >sqs32low.outBut still no output.
try
echo | mcsqs -rcalso try it on the command line
(not in a submission script)
finally, try
ps | grep mcsqs
to find the process id [pid] and then
gdb [pid]and tell me in which routine it is stuck.
The program was running for almost 22:30 hrs. The attached ‘debug.txt’ file is the output of this command
gdb -p [pid]
After the gdb debug, I killed the job. After killing the job from gdb, mcsqs wrote the output files ‘bestsqs.out’, ‘bestcorr.out’ etc. though the correlations are still not optimized. I attached the whole directory with output.
From looking at your submission script I think the issue is that your system is configured to write the files on some temporary space on the compute node that is later copied to your working directory with the command
cp -Rf * $pwdSo mcsqs is working - it’s just that you are looking at the wrong directory. You need to log in to the right node and look at the scratch space. In your stript, before mcsqs add
cat $PBS_NODEFILE
echo $SCRATCH_JOBTo see where the files are.
It’s normal that not all correlations are optimized. It is sometimes not possible with the given cell size.
Yes, it temporarily writes on the computer node and on completion of the job it writes the output files back on the work directory.
My original query is that at some concentration like 0.375-0.625 correlations aren’t getting optimized. Actually, at these concentrations, even two body correlations up to 2NN aren’t getting optimized for 32 atom BCC supercell (and even after specifying a ‘sqscell’ via ‘-rc’ option). In earlier posts you mentioned that at your end some of these are working. I am attaching one directory for BCC 32 atom supercell at 0.375-0.625 concentration. Could please check whether this is getting optimized for two body correlations up to 2NN and how much time its taking? For me, after running this for 3 days it’s ‘bestcorr’ looks like this,
2 0.866025 0.062500 0.062500 0.000000
2 1.000000 0.041667 0.062500 -0.020833
Objective_function= -0.979167
Thank you very much Professor for your continuous help and support.
As you go away from compositions that are simple fractions, it gets more difficult to find exact SQS.
So I would say this is normal. Note that a discrepancy of 0.02 is pretty small!
You can try larger sqs and that would help a lot.
One way to check if the answer is optimal is to see if multiple instances of mcsqs converge to the same quality of sqs.
(Make sure they are started with a different random seed (-sd) or started at different times (if using seed clock, the default).)
Thank you…I will also try larger supercells.