ChemCompute keeps a record of every job's outcome. Across about 1.5 million GAMESS, Psi4
and PySCF jobs, roughly one in six ended with an error. A handful of causes account
for most of them:
| Cause | Share of failed jobs | Fix |
|---|---|---|
| Charge and multiplicity don't fit the molecule | 26% | Usually a missing hydrogen, or a charge set on a neutral structure |
| Ran out of time | 14% | Longer time limit, more cores, or continue from where it stopped |
| SCF did not converge | 13% | Check the structure; turn on SCF Damping |
| Basis set missing for an element | 5% | Mostly iodine with 6-31G*; pick a basis set that covers it |
| Input errors | 5% | Uploaded GAMESS files, and jobs submitted with no molecule |
| Geometry search did not converge | 5% | Continue with Do More Calculations |
| Out of memory | 2% | More cores (about 2 GB each) |
| Atoms too close | 1% | Remove duplicate atoms |
| Cancelled | 3% | Usually by the user, or by a resubmit |
| Many other messages | 24% | Read the Summary; see the full list |
The first four can all be caught before you submit: check the structure in the 3D
panel, count the electrons, look for a basis-set warning under the basis set choices, and
read the run-time estimate.
Jobs from uploaded GAMESS input files fail about four times in ten, against about once
in seven for the guided form. That's mostly from a few formatting rules; see
Using your own input file.
About one NAMD job in seven ends with an error. Four in five of those fail while the
system is being built, usually because of a ligand or a molecule that couldn't be
parameterized. See NAMD and TINKER errors. TINKER jobs
rarely fail.
Figures are from ChemCompute's job records up to October 2026.