Increase natmmx without reducing other parameters

Posted on : April 14, 2017 (Fri) 21:03:26

by 池田敦俊

Dear AkaiKKR users,

When I increase natmmx in specx.f, I get the same error as written in this thread:
http://kkr.issp.u-tokyo.ac.jp/bbs/thread.php?id=154
I already got rid of the error by reducing the value of some other parameters.

My question is, if I buy and install (enough) larger physical memory on my computer, can I increase natmmx without reducing other parameters?

Sincerely,
Ikeda

[Re:05] Increase natmmx without reducing other parameters

Posted on : April 18, 2017 (Tue) 17:41:41

by 池田敦俊

Dear KKR Administrators,

-mcmodel=medium option removed the compile error, and now I can increase the natmmx without changing the other parameters.
I didn't know this option since I'm new to compiling programs by myself.
Thank you for the advice.

I understand that calculation of superlattices with large bzqlty takes time, so anyway I will try with bzqlty=0 first.

Sincerely,
Ikeda

[Re:04] Increase natmmx without reducing other parameters

Posted on : April 17, 2017 (Mon) 20:22:17

by Administrator

Dear Ikeda,

For natom=40, it is enough to use bzqlty=1 for self-consistent iteration (we normally start with bzqlty=0, to get initial rough convergence and then switch to 1). Only after having gotten a good convergence, you may increase bzqlty, say 3, to further calculate some physical quantities accurately. This is important only for memory usage but for CPU time.

Concerning the memory size, it is necessary for gfortran or old versions of Intel compiler to use
-mcmodel=medium
as compiler option to get 64bit addressing (probably you know that). Also uncomment the common statement in specx.f (now is commented out) some times solves the problems.

Best regards,

KKR Administrators

[Re:03] Increase natmmx without reducing other parameters

Posted on : April 17, 2017 (Mon) 17:32:26

by 池田敦俊

Dear KKR Administrators,

I just realized that when calculating, the process is using only 3.5% of the memory.
CPU is almost 100% used.
The operating system is 64 bit, so this may be a problem of the compiler.
I hope this information is useful.

Sincerely,
Ikeda

[Re:02] Increase natmmx without reducing other parameters

Posted on : April 17, 2017 (Mon) 16:29:45

by 池田敦俊

Dear KKR Administrators,

Thank you for the reply.
40 atoms per unit cell using 2 GB memory sound unbelievable to me.

Here are my parameters in the source code.
-------------------------------------------
parameter
& (natmmx=40, ncmpmx=4, msizmx=408, mxlmx=4, nk1x=72, nk3x=0,
& msex=201, ngmx=15, nrpmx=650, ngpmx=650, npmx=350, msr=400)
c
c--- the following part is used for all cases.
parameter
& (lengx=msex, ndmx=natmmx*(natmmx-1)+1, mxlfac=5-mxlmx,
& lastmx=2900d0/3d0**mxlfac, ntypmx=ncmpmx, nkmx=nk1x+nk3x)
-------------------------------------------
Since I want to calculate a superlattice, I increased natmmx and msizmx and decreased unnecessary ncmpmx.
I got rid of the compile error by reducing nk1x and nk3x a lot.
(With this nkmx, only a small bzqlty is allowed.)

My computer has a physical memory of 49446628 kB and swap of 16773116 kB.
I compiled the source code with gfortran just by commenting and uncommenting out some parts of makefile.
The operating system is scientific lunux 7.3.

I would be very grateful if you could provide me with any advice.

Sincerely,
Ikeda

[Re:01] Increase natmmx without reducing other parameters

Posted on : April 14, 2017 (Fri) 22:25:09

by Administrator

Dear Ikeda,

certainly yes. However, it might be helpful if you let us know your parameter setting at specx.f. For example, on my small laptop (MacBook Air 2GB memory), I can run a system with 40 atoms per unit cell (compiled with Intel compiler). Of course it may depend on the system setting (swap and so on).

Best regards,

KKR Administrators