You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
MP2 and CC support for NWChem input files is broken. NWChem moved those calculations into the tensor contraction engine, so the syntax is
TCE
CCSD
end
task TCE energy
rather than
task CCSD energy
Not absolutely necessary, but it is a capability that would be nice to have on release. I built a new ai.nwchem that had it in, but it failed at another point, and I will try again.
Confirmed with NWChem 7.2.3 (water, 6-31G*): ECCE's task ccsd energy stops with "IO error ... End of File", while the TCE form (tce / ccsd / end, task tce energy) runs (CCSD total energy -76.207801 Eh). This will not make 9.0.0, which is being tested tonight; it is the first item after the release and will ship in 9.0.1 soon after: the TCE block in ai.nwchem for MP2 and the CC methods, the energy parser for TCE output, and real runs of each method.
Correction to my comment above: the "IO error ... End of File" line in NWChem 7.2.3 only means a missing CCSD restart file; the job then converges from an MP2 guess and exits normally, so task ccsd energy itself still works for closed shells. The real problem was in ECCE: it could not generate any MP2 or coupled-cluster input for NWChem (a routine aborted for every method except SCF/DFT). Fixed on main for 9.0.1 (ed8f00e): RMP2, UMP2, RCCSD and RCCSD(T) through NWChem's direct modules as before, and new open-shell UCCSD, UCCSD(T), ROCCSD and ROCCSD(T) through TCE (tce block, task tce …), with their energies read. Checked with real NWChem runs on water and the OH radical; TCE and the ccsd module agree to 2e-8 Eh on water. Verify warns about combinations NWChem cannot run (closed-shell-only modules with an open-shell system).
Activity
ohlincha commentedon Oct 12, 2026
Confirmed with NWChem 7.2.3 (water, 6-31G*): ECCE's
task ccsd energystops with "IO error ... End of File", while the TCE form (tce/ccsd/end,task tce energy) runs (CCSD total energy -76.207801 Eh). This will not make 9.0.0, which is being tested tonight; it is the first item after the release and will ship in 9.0.1 soon after: the TCE block in ai.nwchem for MP2 and the CC methods, the energy parser for TCE output, and real runs of each method.NWChem: MP/CC decks generate again; open-shell CC through TCE; TCE en…
ohlincha commentedon Oct 13, 2026
Correction to my comment above: the "IO error ... End of File" line in NWChem 7.2.3 only means a missing CCSD restart file; the job then converges from an MP2 guess and exits normally, so
task ccsd energyitself still works for closed shells. The real problem was in ECCE: it could not generate any MP2 or coupled-cluster input for NWChem (a routine aborted for every method except SCF/DFT). Fixed on main for 9.0.1 (ed8f00e): RMP2, UMP2, RCCSD and RCCSD(T) through NWChem's direct modules as before, and new open-shell UCCSD, UCCSD(T), ROCCSD and ROCCSD(T) through TCE (tceblock,task tce …), with their energies read. Checked with real NWChem runs on water and the OH radical; TCE and the ccsd module agree to 2e-8 Eh on water. Verify warns about combinations NWChem cannot run (closed-shell-only modules with an open-shell system).ohlincha commentedon Oct 13, 2026
Fixed in 9.0.1 (ed8f00e): MP2 and coupled-cluster inputs are generated again; open-shell CC runs through TCE, and TCE energies are read.