Runtest problem

Top Page > Browsing

Runtest problem

Date: 2007/01/22 16:46
Name: Zsolt: Dear Dr Ozaki,

Recently, as I was trying to compile the latest version of openmx on sr11k,
I found and interesting feature. When I run the test calculations usin the
-runtest option, i got the following results:

1 input_example/Benzene.dat Elapsed time(s)= 86.14 diff Utot= 0.082752177659 diff Force= 0.000000006381
2 input_example/C60.dat Elapsed time(s)= 294.23 diff Utot= 0.825157960276 diff Force= 0.000049728716
3 input_example/Cdia.dat Elapsed time(s)= 37.90 diff Utot= 0.016872857246 diff Force= 0.000000000257
4 input_example/CO.dat Elapsed time(s)= 116.96 diff Utot= 0.017376391063 diff Force= 0.000024878317
5 input_example/Cr2.dat Elapsed time(s)= 86.86 diff Utot= 0.059579005969 diff Force= 0.000000000001
6 input_example/Crys-MnO.dat Elapsed time(s)= 378.32 diff Utot=16.479818474526 diff Force= 0.403381185682
7 input_example/GaAs.dat Elapsed time(s)= 247.37 diff Utot= 0.000000000211 diff Force= 0.000000000529
8 input_example/Glycine.dat Elapsed time(s)= 138.50 diff Utot= 0.026971308078 diff Force= 0.000121072116
9 input_example/Graphite4.dat Elapsed time(s)= 64.60 diff Utot= 0.066758721522 diff Force= 0.000000000498
10 input_example/H2O.dat Elapsed time(s)= 55.51 diff Utot= 0.000000000000 diff Force= 0.000000000001
11 input_example/H2O-EF.dat Elapsed time(s)= 59.06 diff Utot= 0.000000000003 diff Force= 0.000000000069
12 input_example/HYb.dat Elapsed time(s)= 212.22 diff Utot= 0.000000000005 diff Force= 0.000000000001
13 input_example/Methane.dat Elapsed time(s)= 115.88 diff Utot= 0.016269215899 diff Force= 0.000000001426
14 input_example/Mol_MnO.dat Elapsed time(s)= 260.29 diff Utot= 7.805308637538 diff Force= 0.009363967135

Total elapsed time (s) 2153.84500

First, I thought that the fftw2 or fftw3 library makes the trouble. But using either of them, the same difference in Utot occured.
Finally, I figured out, that I was using the pao and vps data of 2006. So I restored them to the data packed with openmx3.0 (2004) and
I got this result:

1 input_example/Benzene.dat Elapsed time(s)= 88.19 diff Utot= 0.000000000000 diff Force= 0.000000000002
2 input_example/C60.dat Elapsed time(s)= 332.34 diff Utot= 0.000000000009 diff Force= 0.000000000003
3 input_example/Cdia.dat Elapsed time(s)= 45.25 diff Utot= 0.000000000006 diff Force= 0.000000000137
4 input_example/CO.dat Elapsed time(s)= 153.95 diff Utot= 0.000009673729 diff Force= 0.000000000084
5 input_example/Cr2.dat Elapsed time(s)= 99.68 diff Utot= 0.059579005969 diff Force= 0.000000000001
6 input_example/Crys-MnO.dat Elapsed time(s)= 336.63 diff Utot= 0.000000000085 diff Force= 0.000000000015
7 input_example/GaAs.dat Elapsed time(s)= 254.58 diff Utot= 0.000000000211 diff Force= 0.000000000529
8 input_example/Glycine.dat Elapsed time(s)= 145.42 diff Utot= 0.000000000000 diff Force= 0.000000000001
9 input_example/Graphite4.dat Elapsed time(s)= 68.10 diff Utot= 0.000000000000 diff Force= 0.000000000000
10 input_example/H2O.dat Elapsed time(s)= 62.85 diff Utot= 0.000000000000 diff Force= 0.000000000001
11 input_example/H2O-EF.dat Elapsed time(s)= 65.40 diff Utot= 0.000000000003 diff Force= 0.000000000069
12 input_example/HYb.dat Elapsed time(s)= 230.72 diff Utot= 0.000000000005 diff Force= 0.000000000001
13 input_example/Methane.dat Elapsed time(s)= 119.30 diff Utot= 0.000000000026 diff Force= 0.000000000646
14 input_example/Mol_MnO.dat Elapsed time(s)= 272.95 diff Utot= 0.000000000174 diff Force= 0.000000000007

Total elapsed time (s) 2275.36100

These later results are consistent with the reference data, however the calculations using the updated (2006)
pseudo potentials (VPS) and pseudo atomic orbitals (PAO) results in significant and in some cases huge differences in dUtot as compared to data from 2004.
(The reason for the still existing but slight difference in the case of CO and Cr2 is still unknown for me).

So, do you think it is Ok?! If so, then which data are more relaible?
Thank you.

yours,
Zsolt

Page: [1]

Re: Runtest problem ( No.1 )

Date: 2007/01/19 01:16
Name: Guillaume Lucas <guillaume.lucas@psi.ch>

Hi Zsolt

Are you sure you use the good 2006 DFT_DATA?

In my case every thing worked great for the runtest...except that I encountered the same issues concerning Cr2 and CO.

It seems to be related to the new routine 'Total_Energy.c' of OpenMX Ver. 3.1. provided by Dr. Ozaki (see "Bug fixed in OpenMX3.1, No.2" thread) which corrects a bug in non-collinear calculations.

I found no significative differences with the reference calculations without patching openMX 3.1 but using the code with this new routine I found similar runtests as you (with DFT_DATA provided with the code):

1 input_example/Crys-MnO.dat Elapsed time(s)= 148.68 diff Utot= 0.000000000142 diff Force= 0.000000000008
2 input_example/H2O-EF.dat Elapsed time(s)= 19.78 diff Utot= 0.000000000006 diff Force= 0.000000000084
3 input_example/Glycine.dat Elapsed time(s)= 18.97 diff Utot= 0.000000000000 diff Force= 0.000000000000
4 input_example/H2O.dat Elapsed time(s)= 12.53 diff Utot= 0.000000000000 diff Force= 0.000000000001
5 input_example/Mol_MnO.dat Elapsed time(s)= 122.77 diff Utot= 0.000000000054 diff Force= 0.000000000108
6 input_example/Cr2.dat Elapsed time(s)= 31.09 diff Utot= 0.059579005969 diff Force= 0.000000000001
7 input_example/Cdia.dat Elapsed time(s)= 11.53 diff Utot= 0.000000000004 diff Force= 0.000000000151
8 input_example/C60.dat Elapsed time(s)= 118.63 diff Utot= 0.000000000007 diff Force= 0.000000000006
9 input_example/Benzene.dat Elapsed time(s)= 13.95 diff Utot= 0.000000000000 diff Force= 0.000000000002
10 input_example/Graphite4.dat Elapsed time(s)= 10.78 diff Utot= 0.000000000004 diff Force= 0.000000000001
11 input_example/HYb.dat Elapsed time(s)= 87.49 diff Utot= 0.000000000001 diff Force= 0.000000000000
12 input_example/Methane.dat Elapsed time(s)= 10.19 diff Utot= 0.000000000001 diff Force= 0.000000001317
13 input_example/GaAs.dat Elapsed time(s)= 148.53 diff Utot= 0.000000000999 diff Force= 0.000000000469
14 input_example/CO.dat Elapsed time(s)= 78.57 diff Utot= 0.000009673730 diff Force= 0.000000000322

Hence, as the Cr2 test is a non-collinear calculation, it is clearly due to the used of this corrected routine. In the case of CO, I just don't know why there is this (slight) difference.

Sincerely yours

Guillaume Lucas

Re: Runtest problem ( No.2 )

Date: 2007/01/22 18:22
Name: Zsolt

Hi Guillaume,

Thanks for the suggestion.

I've double checked if I use the corrected 'Total_Energy.c' (and also DFT.c) routine. Yes, I do!
So I just tried the same test with openmx3.0.
With database 2004 (zipped together with openmx3.0) I got perfect run.
Even Cr2 and CO are OK:
.
.
.
4 input_example/CO.dat Elapsed time(s)= 137.30 diff Utot= 0.000000000000 diff Force= 0.000000000032
5 input_example/Cr2.dat Elapsed time(s)= 121.08 diff Utot= 0.000000000000 diff Force= 0.000000000000
.
.
.

Nevertheless, with the database 2006 using openmx3.0, this is what came up:

1 input_example/Benzene.dat Elapsed time(s)= 43.00 diff Utot= 0.082752177659 diff Force= 0.000000006381
2 input_example/C60.dat Elapsed time(s)= 281.57 diff Utot= 0.825157960272 diff Force= 0.000049728715
3 input_example/Cdia.dat Elapsed time(s)= 54.31 diff Utot= 0.016872854206 diff Force= 0.000000002269
4 input_example/CO.dat Elapsed time(s)= 137.34 diff Utot= 0.017373858881 diff Force= 0.000102855310
5 input_example/Cr2.dat Elapsed time(s)= 120.68 diff Utot= 0.000000000000 diff Force= 0.000000000001
6 input_example/Crys-MnO.dat Elapsed time(s)= 510.79 diff Utot=16.479818474275 diff Force= 0.403381140554
7 input_example/GaAs.dat Elapsed time(s)= 348.61 diff Utot= 0.000000000213 diff Force= 0.000000000184
8 input_example/Glycine.dat Elapsed time(s)= 87.76 diff Utot= 0.026971308079 diff Force= 0.000121072116
9 input_example/Graphite4.dat Elapsed time(s)= 57.68 diff Utot= 0.066758726607 diff Force= 0.000000000914
10 input_example/H2O-EF.dat Elapsed time(s)= 62.06 diff Utot= 0.000000000001 diff Force= 0.000000000192
11 input_example/H2O.dat Elapsed time(s)= 57.68 diff Utot= 0.000000000000 diff Force= 0.000000000000
12 input_example/HYb.dat Elapsed time(s)= 234.16 diff Utot= 0.000000000003 diff Force= 0.000000000005
13 input_example/Methane.dat Elapsed time(s)= 58.94 diff Utot= 0.016269215898 diff Force= 0.000000000008
14 input_example/Mol_MnO.dat Elapsed time(s)= 256.89 diff Utot= 7.805308713329 diff Force= 0.009363957910

Total elapased time (s) 2311.45800

So the problem still exists. The reason why I suspect that the data for relativistic pseudopotentials (VPS) and pseudo-atomic orbitals (PAO)
are to be blamed is because if you open the same files ( for instance Mn7.0) of 2004 and 2006 to compare,
you can see that they were generated using different conditions.
However, this should not result in such a huge difference in Etot.

Any other suggestions, ideas?
Could anyone report, if this problem exists on other platform and/or compilation?

yours,
Zsolt

Page: [1]