Paramus — Catalyst Candidate Screen

Overview

Featurize metal/ligand candidates, predict activity-relevant descriptors, and rank by proxies for activity/selectivity. The value is the screening loop + catalysis judgment, not one descriptor.

Guidance vs execution: descriptors and energies come from tested Paramus QM/ML tools. Do not estimate binding energies in ad-hoc code.

When to Use

• Ranking a set of catalyst candidates for a target transformation
• Prioritizing metal/ligand combinations before experiment Do not use for: full microkinetic modeling (hand off to the kinetics engine).

Workflow

Candidate catalysts + target reaction / key intermediate
  ↓ 1. Build & validate structures   → metal-ligand / surface models
  ↓ 2. Electronic / geometric descr. → Paramus descriptor tools
  ↓ 3. Binding / activation energies  → Paramus QM engine or ML surrogate
  ↓ 4. Rank by activity/selectivity proxy → scaling-relation / descriptor model
  ↓ 5. Report                         → ranked candidates + descriptors + provenance

Procedure

1. Discover tools via search/get_schema ("adsorption energy", "electronic descriptor", "QM single point").
2. Build and validate each catalyst model; report failures.
3. Compute the activity-relevant descriptors/energies with provenance.
4. Rank using the appropriate proxy (e.g. binding-energy scaling relation); state the rule.
5. Report ranked candidates with descriptors, proxy value, and confidence.

Domain Judgment

• Descriptor-based ranking is a prioritization, not a rate — never report turnover numbers.
• Beware scaling-relation limits; flag candidates far from the model's domain.
• Selectivity often trades against activity — present both, don't collapse to one score blindly.

Fallbacks

• QM engine unavailable → rank on ML/descriptor proxies; clearly mark reduced fidelity.
• Endpoint unreachable → stop and report; no local energy estimates.