TCE runs fail for ROHF MP2


Just Got Here
Hi,

I'm wanting to do an MP2 energy calculation on an ROHF starting point, and the TCE module seems to consistently fails with various errors, no matter how I vary the memory, algorithm and IO settings. Any help would be much appreciated. I'm running nwchem 6.1.1. I've appended my input file and also my output error.

Cheers,
Andrew



INPUT:



start
memory stack 1000 mb heap 500 mb global 2000 mb
basis SPHERICAL
* library STO-3G
end
geometry noautoz noautosym
Mn 0.0 0.0 0.0
N 1.963319 -0.0 0.0
C 2.769515 1.117594 -0.0
N 2.388384 2.388384 -0.0
C 4.165968 0.704005 0.0
C 5.36092 1.425703 0.0
H 5.351885 2.511198 -0.0
C 6.553234 0.704372 0.0
H 7.501597 1.234817 0.0
C 2.769515 -1.117594 0.0
C 4.165968 -0.704005 0.0
C 5.36092 -1.425703 0.0
H 5.351885 -2.511198 0.0
C 6.553234 -0.704372 0.0
H 7.501597 -1.234817 0.0
N -0.0 1.963319 -0.0
C -1.117594 2.769515 -0.0
N -2.388384 2.388384 -0.0
C -0.704005 4.165968 -0.0
C -1.425703 5.36092 -0.0
H -2.511198 5.351885 -0.0
C -0.704372 6.553234 -0.0
H -1.234817 7.501597 -0.0
C 1.117594 2.769515 -0.0
C 0.704005 4.165968 -0.0
C 1.425703 5.36092 -0.0
H 2.511198 5.351885 -0.0
C 0.704372 6.553234 -0.0
H 1.234817 7.501597 -0.0
N -1.963319 0.0 -0.0
C -2.769515 -1.117594 0.0
N -2.388384 -2.388384 0.0
C -4.165968 -0.704005 -0.0
C -5.36092 -1.425703 0.0
H -5.351885 -2.511198 0.0
C -6.553234 -0.704372 0.0
H -7.501597 -1.234817 0.0
C -2.769515 1.117594 -0.0
C -4.165968 0.704005 -0.0
C -5.36092 1.425703 -0.0
H -5.351885 2.511198 -0.0
C -6.553234 0.704372 -0.0
H -7.501597 1.234817 -0.0
N -0.0 -1.963319 0.0
C 1.117594 -2.769515 0.0
N 2.388384 -2.388384 0.0
C 0.704005 -4.165968 0.0
C 1.425703 -5.36092 0.0
H 2.511198 -5.351885 0.0
C 0.704372 -6.553234 0.0
H 1.234817 -7.501597 0.0
C -1.117594 -2.769515 0.0
C -0.704005 -4.165968 0.0
C -1.425703 -5.36092 0.0
H -2.511198 -5.351885 0.0
C -0.704372 -6.553234 0.0
H -1.234817 -7.501597 0.0
end

charge 0
scf
ROHF
doublet
vectors input HCORE
tol2e 1e-9
sym off
adapt off
maxiter 200
end

task scf energy

tce
scf
mp2
freeze atomic
io eaf
tilesize 5
attilesize 10
end
task tce energy



OUTPUT:



Global files accessible by all nodes assumed

Parallel file system coherency ......... OK

Integral file          = ./test2.aoints.0
Record size in doubles = 65536 No. of integs per rec = 43688
Max. records in memory = 0 Max. records in file = ********
No. of bits per label = 8 No. of bits per value = 64


#quartets = 8.066D+06 #integrals = 3.107D+07 #direct =  0.0% #cached =100.0%


Fock matrix recomputed
1-e file size = 68450
1-e file name = ./test2.f1
Cpu & wall time / sec 124.6 128.7

tce_ao2e: fast2e=1
half-transformed integrals in memory

2-e (intermediate) file size =      -496817356
2-e (intermediate) file name = ./test2.v2i
------------------------------------------------------------------------
put_block: eaf problem 5
------------------------------------------------------------------------
------------------------------------------------------------------------
current input line :
85: task tce energy
------------------------------------------------------------------------
------------------------------------------------------------------------
An error occured while trying to read or write to disk space
------------------------------------------------------------------------
For more information see the NWChem manual at http://nwchemgit.github.io/index.php/NWChem_Documentation


For further details see manual section:                                                                                                                                                                                                                                                                
0:0:put_block: eaf problem:: 5

Clicked A Few Times
Hi Andrew
the TCE implementation of the mp2 approach is a by-product of the coupled cluster iterations and therefore is rather inefficient in large scale calculations. For example, all possible 2-electron integrals are formed when this keyword is involed.

For large MP2 calculations you should use the MP2 module of nwchem (please see the documentation).
You have several options there, including: semi-direct, fully-direct, and resolution-of-identity based MP2 formulations.

Best,
Karol

Just Got Here
Hi Karol,

I'm aware of that. And the built-in MP2 procedures would be my preferred option, and I have used them in the past. However the built-in MP2 procedures require a RHF or UHF starting point. I have a ROHF starting point, so I can't use the built-in MP2 procedure. (I've already done UHF MP2 calculations on this molecule, but the system has a lot of spin-contamination so UHF results are too unreliable, and I therefore want ROHF MP2 results) A new feature of enabling MP2 on ROHF starting points would be nice for future releases. Anyway, that's why I'm using TCE, because it's currently nwchem's only documented way of doing ROHF MP2 calcs.

I don't particularly mind the inefficiency of the TCE's MP2 implementation. I just want it to work. The documentation says that the TCE should be able to do MP2 calculations from an ROHF starting point but no matter what IO / 2emet / memory settings I use, I am unable to get it to run without an error. Even the settings I supplied above, which I normally find to be helpful for problem cases, are not working.

Thanks,
Andrew

Gets Around
Don't use EAF. USE IO=GA and tilesize=30 or so. Same for attilesize, although you don't need it given that you are not using 2eorb or 2emet>1. For ROHF, you should use 2eorb and 2emet=13.

Negative filesize indicates you did a 32b integer build but need to index more than 2^31 elements in a global array. You should rebuild with 64b integer support enabled. It should be the default for NWCHEM_TARGET=LINUX64 so I don't know how you got where you are.

I don't know why you force the HCORE guess. That's not nearly as good as the atomic density guess that is the default.


Forum >> NWChem's corner >> Running NWChem