Orca Opt

ORCA Geometry Optimization Skill

When to Use

Use this skill when the user wants to:

• Optimize a molecular geometry with ORCA
• Find the minimum energy structure of a molecule
• Relax a molecular cluster or complex

Do NOT use for periodic systems (use VASP or CP2K instead).

Node Parameters (canonical names — these are what the engine actually reads)

⚠️ extra_keywords and extra_blocks are NOT read by the ORCA workflow engine. Earlier versions of this skill recommended them; anything passed via those keys is silently dropped. Use the dedicated fields above. CPCM solvation, SlowConv/SOSCF, and NumFreq currently have no first-class field on the opt/freq/neb_ts/irc nodes — that is a node-def gap, not a usage problem.

MCP Tool Examples — proven Expanse submission flow

Use catgo_workflow (graph-based), NOT catgo_workflow_engine (task-based). The graph-based tool auto-captures the viewer structure on create. The task-based tool's add_task does not, so jobs fail with "No input structure provided". Param keys differ too: graph-based uses method/basis, task-based uses orca_method/orca_basis.

1. Confirm structure is loaded

catgo_view(action: "get_state")

2. Find the Expanse session_id

Session IDs are volatile — they change on every reconnect. Discover the current one:

curl -s http://localhost:8000/api/hpc/connections

Look for the entry with host: login.expanse.sdsc.edu and copy its session_id.

3. Create the workflow (auto-captures viewer structure)

catgo_workflow(action: "create", name: "Benzene optimization")

Returns a workflow with one structure_input node containing the current viewer structure. Note its node ID (e.g., n1777012885-iode).

4. Add the geo_opt node and connect it

Inject extra_blocks: "%output jsongbwfile True jsonpropfile True end" so ORCA emits the JSON files OPI parses on the way back. Without this, OPI parsing falls back to grepping ORCA.out (still works, just less rich).

catgo_workflow(action: "batch", workflow_id: "<wf_id>", operations: [
  {"op": "add_node", "node_type": "geo_opt", "label": "opt",
   "params": {
     "software": "orca",
     "method": "B3LYP",
     "basis": "def2-SVP",
     "charge": 0,
     "multiplicity": 1,
     "extra_blocks": "%output jsongbwfile True jsonpropfile True end"
   }},
  {"op": "connect", "from_id": "<structure_input_id>", "to_id": "opt",
   "from_handle": "structure", "to_handle": "structure"}
])

5. Run with the full HPC run_config

Real opt jobs on non-trivial molecules can run for hours — default to partition: "shared" with a generous walltime. Use debug only for tiny sanity checks (≤ a couple of heavy atoms, single-point or quick test). Read server/templates/orca_generic.sh and pass its contents as default_template.

catgo_workflow(action: "run", workflow_id: "<wf_id>", run_config: {
  "execution_mode": "hpc",
  "default_session_id": "<expanse_session_id>",
  "base_work_dir": "/expanse/lustre/projects/sdp126/jyang25/ORCA/catgo",
  "default_job_params": {
    "nodes": 1, "ntasks": 8, "cpus_per_task": 1,
    "walltime": "04:00:00", "partition": "shared"
  },
  "cluster_configs": {
    "<expanse_session_id>": {
      "account": "sdp126",
      "partition": "shared",
      "module_loads": "module load cpu/0.17.3b\nmodule load gcc/10.2.0/npcyll4\nexport PATH=$HOME/openmpi-4.1.8/bin:$PATH\nexport LD_LIBRARY_PATH=$HOME/openmpi-4.1.8/lib:$LD_LIBRARY_PATH",
      "orca_dir": "/home/jyang25/orca_6_1_1_RRP8",
      "default_template": "<contents of server/templates/orca_generic.sh>",
      "default_job_params": {
        "nodes": 1, "ntasks": 8, "cpus_per_task": 1,
        "walltime": "04:00:00", "partition": "shared"
      }
    }
  }
})

For a quick sanity check (e.g., H2O / methane / single small molecule), override to partition: "debug", walltime: "00:30:00", ntasks: 4. The debug partition caps at 30 min — anything bigger will be rejected after the limit.

The local-scratch template stages I/O to $TMPDIR/orca_$SLURM_JOB_ID and copies results back to the Lustre work_dir. Required on Expanse — Lustre is bad for ORCA's many small temp files.

6. Common parameter variations

For dispersion (non-covalent systems, dimers, π-stacking, H-bonding):

"params": {"method": "B3LYP", "basis": "def2-TZVP", "dispersion": "D3BJ", "charge": 0, "multiplicity": 1}

For D4 (newer Grimme correction, slightly better for metals):

"params": {"method": "B3LYP", "basis": "def2-SVP", "dispersion": "D4", "charge": 0, "multiplicity": 1}

For implicit solvation:

CPCM is currently a node-def gap on opt/freq/neb_ts/irc — there's no first-class solvation/solvent field, and extra_keywords is not read. UV-Vis is the exception (it has dedicated solvation/solvent fields). Until this is fixed, single-point CPCM on a gas-phase optimized geometry, or running on a non-CatGo input file, is the workaround.

For tight convergence (publication quality, pre-freq):

"params": {"method": "B3LYP", "basis": "def2-TZVP", "opt_convergence": "TightOpt", "dispersion": "D3BJ", "charge": 0, "multiplicity": 1}

For open-shell radicals:

"params": {"method": "UB3LYP", "basis": "def2-SVP", "charge": 0, "multiplicity": 2}

7. Monitor

catgo_workflow(action: "status", workflow_id: "<wf_id>")

Or query SLURM directly via the live session:

curl -s "http://localhost:8000/api/hpc/jobs/<job_id>?session_id=<expanse_session_id>"