# CCSD(T) triples crash

8 months 4 weeks ago - 8 months 4 weeks ago #647 by negoty
CCSD(T) triples crash was created by negoty
Dear fellow users and/or Mihaly,

I have been trying to calculate CCSD(T) for an iron(II) open shell complex. It proceeded fine until after the CCSD iterations converged but crashes as soon as MRCC starts with the triples. Unfortunately I have no indication what might have gone wrong. Also the closed shell calculation for a similar complex (and same basis set) did not show any problems. I would appreciate any help.

Input:
# CCSD(T)
calc=ccsd(t)
mem=30000MB
scfmaxit=100
scfext=15
#symm=off
scfiguess=ao

occ=17,6,7,14/14,5,7,14
core=corr

basis=atomtype
Fe:aug-cc-pwCVTZ
O:aug-cc-pCVDZ
H:aug-cc-pVDZ

mult=5
charge=2

unit=angs
geom=xyz
19

Fe       0.000000000      0.000000000      0.000000000
O        0.000000000      0.000000000      2.132470000
H       -0.633127000      0.454504000      2.712386000
H        0.633127000     -0.454504000      2.712386000
O        0.792100000      1.961200000      0.000000000
H        1.035508000      2.485723000      0.781097000
H        1.035508000      2.485723000     -0.781097000
O       -1.994703000      0.776644000      0.000000000
H       -2.240397000      1.716560000      0.000000000
H       -2.823416000      0.269539000      0.000000000
O        0.000000000      0.000000000     -2.132470000
H        0.633127000     -0.454504000     -2.712386000
H       -0.633127000      0.454504000     -2.712386000
O       -0.792100000     -1.961200000      0.000000000
H       -1.035508000     -2.485723000     -0.781097000
H       -1.035508000     -2.485723000      0.781097000
O        1.994703000     -0.776644000      0.000000000
H        2.240397000     -1.716560000      0.000000000
H        2.823416000     -0.269539000      0.000000000

Output snippet:
 Starting CCSD iterations...

Residual norm     Correlation energy [au]
-------------------------------------------------------------
Iteration  1       0.140337245242        -2.374586200128
Iteration  2       0.038603017564        -2.415046033203
Iteration  3       0.023668053366        -2.424282322957
Iteration  4       0.006670109361        -2.423934062142
Iteration  5       0.004151952029        -2.423624339568
Iteration  6       0.001602094607        -2.423832362904
Iteration  7       0.000653346502        -2.423868138936
Iteration  8       0.000227946660        -2.423853663306
Iteration  9       0.000093990545        -2.423856281824
Iteration 10       0.000051560742        -2.423858074921
Iteration 11       0.000034027268        -2.423857390054
Iteration 12       0.000029867000        -2.423857465312
Iteration 13       0.000028601675        -2.423857531877
Iteration 14       0.000027947157        -2.423857114354
Iteration 15       0.000026686483        -2.423856633713
Iteration 16       0.000020955653        -2.423857579666
Iteration 17       0.000009710641        -2.423857695840

CCSD iteration has converged in 17 steps.

Total CCSD energy [au]:     -1720.807613612760
CPU time [min]:  2830.799                   Wall time [min]:   631.293

Calculation of (T) correction...

Fatal error in uccsd.
Program will stop.

************************ 2018-10-23 10:04:17 *************************
Error at the termination of mrcc.
**********************************************************************

All the best, Benedikt
Last edit: 8 months 4 weeks ago by negoty.

8 months 3 weeks ago #648
Replied by nagypeter on topic CCSD(T) triples crash
Dear Benedikt,

sorry for the late reply, the tests took a long time during last week.
In this case the program ran out of memory. There is usually an indication in such cases, sorry for that.
You can find an updated version of the uccsd.f code in the download area that requires much less memory, so you can run this example with the 30 GB set in your input file.

If you are interested in even larger systems, you might want to use the density fitting version [calc=DF-CCSD(T)], for which the integral transformation is much faster. The CCSD(T) part will take the same time both with and without DF.

All the best,
Peter

8 months 2 weeks ago #649 by negoty
Replied by negoty on topic CCSD(T) triples crash
Dear Peter,

thanks for your reply. So far I have been using the binaries so I'll try and compile MRCC with the updated uccsd source code. I'll keep DF-CCSD(T) in mind although the integral transformation does not bother me at this time, I keep stumbling over other issues but I think I'll put that in a separate thread.