CCSDT calculation of F2 at twice equilibrium distance cannot converge


Forum Vet
I have tried one example from GAMESS using NWCHEM, i.e., the energy of F2 at twice equilibrium.

It seems NWCHEM6.8 CCSDT calculation of F2 at twice equilibrium cannot converge, although I have not wait for the error signal to appear.

The precompiled GAMESS binary coupled cluster calculation with singles, doubles and active space triples has been successful on macOS, although with
Note: The following floating-point exceptions are signalling: IEEE_UNDERFLOW_FLAG IEEE_DENORMAL

But
ddikick.x: exited gracefully.

The results using the precompiled GAMESS binary on macOS got the same results as those in the authors' article indexed in the example.


Very Best Regards!

Forum Vet
GAMESS 2018 R2 source code, which requires compilation, has implemented ROHF active-space triples, and complementary CC(t,3) for CCSDt, both of which may reproduce the correlations in ground state biradicals and bond breaking situations.
Professor Piecuch's group from Michigan State University contributes to these new coupled cluster methods, evaluated and assembled into GAMESS by Professor Gordon's group.

I will try the compilation.

Forum Vet
After changing something in the input, NWCHEM6.8 made the calculation converge at the 63th step, where the total energy obtained was 0.00527 hartrees from that by GAMESS according to the original setting with additional data provided when the basis sets are the same. If the explicit basis set is truly identical with that in NWCHEM, the total energy difference can be 0.00115 hartrees.


I will try the compilation on Ubuntu and macOS.

Forum Vet
It is a pleasure to use this Monday released GAMESS2018R3 binary for macOS to try the method for CH, and have repeatedly got the identical energies of CCSDt, cc(t,3), ect., with official GAMESS results, and one note:
...
Note: The following floating-point exceptions are signalling: IEEE_DENORMAL
ddikick.x: exited gracefully.

Actually the precompiled binary has made the calculation successful.

The cpu needs to be updated.
The energies calculated by the benchmarkable CCSDT of NWCHEM6.8 for CH, repeated on Ubuntu and MAC OS 10.13.6 and almost identical, using the same geometry, same symmetry, and same kind of basis set are only from 0.00333 to 0.00175 hartrees different from all those obtained from GAMESS according to the original active-space setting.

Forum Vet
When the basis set is increased to aug-cc-pvtz, the total energy of CCSDT of CH is lower than the result of the same method using CBS/mix basis set extrapolation in an article by Drs. in EMSL at that stage, and all thsoe obtained by GAMESS are from 0.0106 to 0.00886 hartrees from that, but both will cause negligible spectroscopic constants differences from the calculated and the experimental data indexed, according to the article, perhaps the geometry provided by the GAMESS example is better.

The CCSDT total energy with NWCHEM6.8 in the case of aug-cc-pvtz is further 0.00005 hatrees lower using the mp2 optimized CH geometry by GAMESS 2018 R3, perhaps because of a larger basis set, which of cause will also be anticipated to produce a very good agreement.
The active space coupled cluster total energy of CH3 with aug-cc-pvtz using GAMESS might be much less hardware demanding, thus much faster, than that obtained by CCSDT using NECHEM to match experimental physical quanties very well.

Congratulations for the achievement of efforts from authors and editors of both programs in GAMESS and NWCHEM.


Forum >> NWChem's corner >> Running NWChem