Summit

Overview

Summit provides tools for optimising chemical processes, with reactions as its initial focus. Its workflow uses optimisation algorithms to explore conditions such as temperature and stoichiometry against one or more objectives, including yield or enantiomeric excess. It is a toolkit for constructing and evaluating optimisation workflows, rather than a single predictive model or a documented laboratory-control system.

The README describes two main components: eight implemented optimisation strategies and reaction benchmarks built from mechanistic or data-driven simulations. These benchmarks provide a setting for evaluating strategies before considering their use with physical experiments. The workflow connects an optimisation domain, a strategy and an experiment through an iterative runner. Inputs include the domain of reaction conditions, objective definitions and a selected benchmark; outputs include evaluated objective values and a Pareto plot for inspecting competing objectives.

The supplied quick-start example uses SnarBenchmark, MultitoSingleObjective, SOBO and Runner. It combines the benchmark's two objectives through an expression and configures a closed-loop run with a maximum of 50 iterations. This illustrates how objective transformation, strategy selection and simulated experiments fit together, but it does not establish performance on a user's reaction system.

The example also contains a documentation inconsistency: its prose calls the strategy Nelder-Mead, while its code instantiates SOBO. Users should resolve that discrepancy against the linked documentation before adapting the example. The source excerpts do not document direct hardware integration or demonstrate real-laboratory outcomes.

Key Features

• Eight optimisation strategies are described as implemented for searching reaction conditions through iterative optimisation.
• Mechanistic and data-driven reaction benchmarks support simulation-based evaluation of optimisation strategies.
• `MultitoSingleObjective` combines multiple objectives into a single expression for use with a single-objective strategy.
• `Runner` connects a strategy and an experiment to execute a closed-loop workflow with a configurable iteration limit.
• `SnarBenchmark` provides the reaction simulation used in the quick-start example, which includes Pareto plotting of its two objectives.

Use Cases

• Suggested evaluation: compare optimisation strategies on simulated reaction benchmarks before selecting an approach for a reaction-development study.
• Suggested evaluation: explore how different objective combinations affect the trade-off between the SnAr benchmark's `sty` and `e_factor` objectives.
• Suggested evaluation: use the documented tutorial to prototype an iterative reaction-condition search before assessing suitability for a physical experimental workflow.

How to Use

1. Read the project README and documentation to identify the benchmark, optimisation domain and objectives relevant to your evaluation.
2. Check the supplied package configuration's declared Python constraint, ^3.8, <3.11, when planning an environment. Treat this as a dependency declaration, not evidence of tested compatibility on your system. Follow the README's installation guidance.
3. Work through the tutorial to understand how strategies interact with simulated experiments. Begin with the documented benchmark rather than assuming a connection to laboratory equipment.
4. Inspect the quick-start components: SnarBenchmark, MultitoSingleObjective, SOBO and Runner. Resolve the mismatch between the example's Nelder-Mead description and its SOBO implementation before choosing a strategy.
5. Define the objective transformation and iteration budget for your intended evaluation, then inspect the resulting objective trade-offs using the example's Pareto-plot workflow. Record simulation findings separately from any later physical-experiment results.