[Wien] NMR calculation

[Robert Laskowski]

Hi,

x_nmr runs in stages. First, it executes 7 lapw1 runs, in nmr_* directories 
(nmr_q0, nmr_pqx ....), can you see it? After that it will start nmr program, 
that calculates induced current. Both lapw1 and nmr are executed in parallel, 
in your case it is only k-point parallelization. Files nmr.out.* contain  the 
progress of the nmr progran in current mode.  There should be file for each k-
point chunk. Is there anything inside? After that nmr program start in 
integration mode, this can also be in parallel, but only over MPI and with 
nmr_integ: flag in .machines file. If the vectors are done, and you have correct
nmr_* directories, and nmr program crashed for come reason do:

x nmr -p -noinit

this will execute nmr in mode current, and integration. There is no restart 
implemented beyond that.

NMR calculations are more expensive then an ordinary SCF, this is because the 
basis is extended, and the full spectrum of vectors and eigenvalues are 
computed. The lapack call in WIEN2k eigensolver (lapw1)  is optimized to solve 
the eigenvalue problem in a specified energy window. Maybe calling routine made 
specifically for full spectrum in lapw1 (for nmr execution)  would give some 
speedup. 

regards

Robert

On 18 March 2015 PM 1:36:58 Ary Ferreira wrote:
> Please,
> 
> I'm trying to compute the NMR chemical shifts of a compound with the
> version 14.2. I'm doing it in parallel with two nodes (machineA and
> machineB), each one with 16 cores. So, I prepared a .machines file like
> this:
> 
> 1:machineA:1
> ... plus 15 identical lines
> 1:machineB:1
> ... plus 15 identical lines
> granularity:1
> extrafine:1
> 
> In such conditions, the SCF converges in about 1:30 hours. To run it, I set
> the structural information with the makestruct script and used the
> init_lapw script with the option -b. Then, I run WIEN2k with:
> 
> run_lapw -p -fc 1
> 
> After that, I used the following command to setup the NMR-LOs:
> 
> x_nmr_lapw -mode in1 -focus Li
> 
> And finally:
> 
> x_nmr_lapw -p
> 
> However, the NMR calculation had not been after 4 hours. My first question
> is: can I check the parallel execution of nmr as can be done for lapw1 and
> lapw2 with their respective testpara*_lapw scripts?
> 
> The second question is: is it possible to restart this NMR calculation?
> 
> Thank you very much!
> 
> Ary Ferreira

-- 
==================================================
Dr. Robert Laskowski

Senior Scientist, Materials Science & Engineering Department
Institute of High Performance Computing, A*STAR
1 Fusionopolis Way, #16-16, Connexis, Singapore 138632
Tel(Off): +65. 64191493     Fax: +65. 64632536
=================================================