Optimize a multi-component formulation to hit property targets by defining the component/ratio design space under mixture constraints, screening candidates, and optimizing the blend against multiple objectives — orchestrating Paramus mixture-DOE, property-prediction, and optimization tools.When to use: a user needs to find component ratios that meet several property targets at once.Multi-tool WORKFLOW with mixture-design judgment.
Paramus — Formulation Property Optimization
Overview
Find component ratios that meet several property targets at once. The value is the mixture-aware design→predict→optimize pipeline honoring the "components sum to 100%" constraint — not one optimizer call.
Guidance vs execution: mixture designs, predictions, and optimization come from tested Paramus tools. Do not hand-roll a mixture optimizer in ad-hoc code.
When to Use
- Balancing multiple property targets across component ratios
- "What blend hits target viscosity AND stability AND cost?" Do not use for: single-component property lookup or pure stability assessment (use the stability screen).
Workflow
Components + ratio ranges + property targets + constraints
↓ 1. Define mixture design space → components sum to 100%, bounds
↓ 2. Generate mixture design → Paramus mixture-DOE tool
↓ 3. Predict properties → per-candidate property estimates
↓ 4. Multi-objective optimize → trade-off surface / Pareto set
↓ 5. Rank & recommend → candidate blends + provenance
Procedure
- Discover tools via
search/get_schema("mixture design", "formulation", "multi-objective"). - Define the mixture space with the sum-to-one constraint and component bounds.
- Generate a mixture-appropriate design (simplex/constrained), not a naive factorial.
- Predict each target property for candidates.
- Optimize across objectives; expose the trade-off/Pareto set, not a single hidden pick.
- Rank and recommend blends with provenance.
Domain Judgment
- Use mixture designs — components are not independent factors; standard DOE misleads.
- Surface trade-offs explicitly; multi-objective problems rarely have one dominant winner.
- Include practical constraints (cost, regulatory limits) as real bounds, not afterthoughts.
Fallbacks
- Mixture/optimization tool unavailable → deliver a manual constrained design + target table, clearly scoped.
- Endpoint unreachable → stop and report; do not fabricate optimized ratios.
Tools this skill may use
Candidate deterministic tools an agent is likely to route to when running this skill. The skill decides which to call at runtime.
-
Brain HPC Bofire Doe
Data ScienceGenerate experimental design -
Full Factorial
Data ScienceGenerate full factorial design with customizable levels for each factor -
Box Behnken
Data ScienceGenerate a Box-Behnken experimental design for response surface methodology. This design places experimental runs at the midpoints of edges of the design space (no corner points), making it more efficient than full factorial designs while still allowing estimation of quadratic effects. Ideal for optimizing reaction conditions, formulations, and process parameters when extreme combinations should be avoided. -
Latin Hypercube
Data ScienceGenerate Latin Hypercube Sampling design for space-filling experiments -
Brain HPC Bofire Campaign
Data ScienceMulti-cycle Bayesian optimization campaign manager. Track experiments across multiple optimization cycles with actions: create, add_cycle, get_history, get_summary, list.
Browse the full deterministic layer in the tool browser.
Used in these use cases
Customer scenarios that orchestrate this skill end-to-end.