Linux: Gaussian 16

| Feature | Gaussian 16 | ORCA 5/6 | Q-Chem 6 | |------------------------|----------------------|-----------------------|-----------------------| | Cost | Paid (site license) | Free for academic | Paid (lower than G16) | | OpenMP scaling | Good (≤32 cores) | Excellent | Good | | GPU support | None | Yes (NVIDIA) | Yes (limited) | | TD-DFT | Stable | Very fast | Fast | | ONIOM | Native | Not native (custom) | No | | Ease of input | Cryptic keywords | Simpler | Moderate | | Documentation | Dense, dated | Modern, well-written | Good |

sudo mkdir -p /opt/gaussian/g16
sudo mkdir -p /scratch/gaussian  # High-speed scratch space

Gaussian 16 Rev C.01+ supports NVIDIA GPUs for DFT (B3LYP, PBE0, M06-2X) and RI-MP2. On Linux, the speedup is dramatic (3-5x for hybrid functionals).

Summary

  • Use checkpoint files (%chk) to save wavefunction and restart large calculations. Use formatted checkpoint (.fchk) when you need properties or visualization support.
  • For large correlated calculations, include options to reduce disk churn (density fitting where available), or use local approximations if acceptable. Action: build a standard header template for your lab with %mem, %nprocshared, %chk, and GAUSS_SCRDIR that matches cluster allocations.

    • Quick checklist to get started (actionable)

    If you want, I can:

    Source your bash configuration:

    source ~/.bashrc

    Run a quick test job:

    cd /opt/g16/tests
g16 < test000.com > test000.log

    If you see Normal termination of Gaussian 16, congratulations — Gaussian is working.