Decide what deserves the next calculation or experiment.
Explore large composition and process spaces with surrogate models, multi-fidelity evidence, high-fidelity simulation and a closed loop between computation and measurement.
Start with the bottleneck that consumes expensive evidence.
Materials programmes are constrained by the cost of trustworthy evidence. Candidate spaces may be enormous, while density-functional calculations, specialist simulations and laboratory experiments remain expensive. The central question is therefore not how to generate more candidates, but how to choose the next evaluation that can most improve the decision.
Turn a broad search space into an evidence-driven sequence of candidate evaluations, with uncertainty visible at every stage and high-fidelity resources reserved for the points that matter most.
Turn a design space into a sequence of evidence-buying decisions.
The visual model shows the relationship between observed points, the surrogate surface and the next proposed evaluation, keeping uncertainty in the decision rather than only in a report.
From objective to evidence and next action.
- 01Define the target property profile, decision variables, allowable composition/process ranges and hard constraints.
- 02Register existing compositions, measurements, simulation outputs and dataset lineage so previous evidence becomes part of the programme.
- 03Establish an evidence ladder: inexpensive approximations, surrogate predictions, high-fidelity calculations and measured experiments remain distinct.
- 04Fit Gaussian-process or other quantitative surrogate models where the data and model assumptions support them.
- 05Use uncertainty and objective value to prioritize the next candidate instead of ranking only by predicted mean performance.
- 06Call trusted simulation or DFT adapters when a high-fidelity evaluation is justified and the required runtime is provisioned.
- 07Ingest computational or laboratory results together with artifact hashes, source references and evidence type.
- 08Refit the model, update the candidate ranking and repeat until the programme reaches its decision or stopping criteria.
The programme combines modelling, execution and evidence.
Design-space definition
Bounded variables, objectives, constraints and candidate records.
Surrogate modelling
Gaussian-process models with predictive mean, variance and intervals.
Multi-fidelity learning
Combine lower-cost evidence with calibrated high-fidelity observations.
Scientific execution
Connect trusted simulation, DFT, HPC and result-ingestion paths.
Uncertainty-aware selection
Use uncertainty and objective value to decide where more evidence is worth buying.
Reproducibility
Bind datasets, protocols, software state, artifacts and decisions into reproducible evidence.
What connects into the programme.
- Historical measurements and experimental datasets
- Candidate compositions and process variables
- Low- and high-fidelity simulation outputs
- Property targets and engineering constraints
- Optional DFT/HPC environments and laboratory result feeds
What the team gets back.
- Ranked candidates with supporting quantitative evidence
- Surrogate models and uncertainty estimates
- High-fidelity execution records and artifacts
- Experiment priorities and next-evaluation decisions
- Reproducibility packs and benchmark-ready evidence
Why the operating loop matters.
- Screen a broader design space before committing expensive resources.
- Use high-fidelity compute where it can materially change candidate selection.
- Make uncertainty visible instead of presenting point estimates as certainty.
- Preserve failed and negative evaluations so the programme does not relearn the same lesson.
- Create a repeatable path from computational hypothesis to measured evidence.