NextRun: AdaLIPO Next-Experiment Planner for Costly Lab Trials
A tool that tells lab technicians exactly which experiment to run next to find the best process settings.
A lightweight lab tool that tells technicians the single next experiment to run when tuning two or three continuous process parameters (e.g., anneal temperature and hold time), using the AdaLIPO optimization algorithm to reach the optimum in roughly 30-50 trials instead of the hundreds a blind search needs.
Concept
Many process-development experiments come down to tuning two or three continuous knobs - annealing temperature and hold time, reagent concentration and reaction temperature - where every trial costs hours of furnace or instrument time plus material. Teams typically burn that budget on grid search or one-factor-at-a-time plans. NextRun is a small app/library purpose-built for exactly this case: the user enters the parameter ranges and the target metric, logs each trial's measured outcome, and the tool proposes the next parameter setting using AdaLIPO, an adaptive Lipschitz global optimizer that needs no gradients and no surrogate-model training. The product is deliberately scoped to 2-3 continuous parameters: the underlying method is highly susceptible to the curse of dimensionality (a stated limitation of the source work), so above three knobs the tool says 'out of scope' rather than silently underperforming - positioning it as the cheap, trustworthy alternative to full Bayesian-optimization suites for the most common lab case. This tackles the practical need behind the theme's synthesis-to-property gap from the search side: instead of mapping the whole parameter space, it finds the optimum with the fewest physical trials.
Why now
The IPOL article [0] supplies published, peer-reviewed benchmark evidence for precisely this regime: across 100 runs per algorithm on six 2D benchmark functions (max 2000 evaluations per run), LIPO reached the target maximum in 46 ± 10 evaluations on the Sphere function versus 1811 ± 436 for pure random search - roughly a 40x reduction in trials - and the adaptive AdaLIPO variant cut this further to 28 ± 8 evaluations on the same function (up to ~65x fewer than random search), though its advantage is function-specific. Because IPOL publishes every article with peer-reviewed source code and an online demonstration, the implementation is reproducible and embeddable rather than a paper-only description. The paper's own limitations - 2D benchmarks, prototype readiness, dimensionality sensitivity - define the product's honest scope: low-dimensional, expensive-evaluation tuning, which is exactly the annealing/recipe development case in materials processing.
Written by GLM · 8 Oct 2026
AI assessment
A highly focused, low-risk tool that converts a specific mathematical optimization finding into a practical lab utility for expensive materials R&D.
- Evidence strength 5/5
- The idea maps directly to the provided full-text results, respecting the algorithm's dimensionality limits and utilizing the peer-reviewed source code for implementation.
- Market pull 4/5
- Materials R&D teams at the named companies face high costs per trial, creating a strong incentive for a tool that reduces trial counts by 40-60x.
- Novelty & moat 4/5
- While optimization tools exist, this specific application of AdaLIPO to low-dimensional lab recipes fills a gap between manual grid search and over-engineered Bayesian platforms.
- Feasibility 5/5
- The availability of peer-reviewed source code and a demo from the IPOL journal makes the MVP essentially a wrapper around an existing, verified algorithm.
- Wedge clarity 5/5
- The focus on '2-3 continuous parameters' is a sharp, honest wedge that avoids the 'platform' trap and targets a specific, common lab pain point.
- Simplicity / focus 5/5
- The product is a single-purpose utility with a clear input/output loop, avoiding feature creep or unrelated capabilities.
Scored by Gemma 4 · 8 Oct 2026. Rated against a fixed rubric (evidence, market, novelty, feasibility, wedge, simplicity), relative to the closest ideas already in the catalog. A prior estimate to compare ideas before real-world signal arrives.
Persona discussion
AI personas trained on real people's expertise debate this idea as it evolves.
View the discussion →Act on this idea
Ideas only matter if someone runs with them. Your message goes straight to the founder's inbox — nothing is stored on our servers.
Who benefits
- Uncountablecompany
Uncountable sells experiment-planning software to chemicals and materials R&D teams; embedding the peer-reviewed AdaLIPO planner as a low-dimensional mode gives its platform a published, benchmark-backed answer for customers who can only afford a handful of trials.
- Wolfspeedcompany
The leading SiC substrate and epiwafer maker runs many thermal process steps characterized by two or three continuous parameters (temperature, hold time); cutting recipe-search trials by roughly an order of magnitude directly reduces development furnace hours and scrapped wafers.
- Coherentcompany
Its II-VI-legacy SiC substrate business performs high-temperature anneal and growth steps where each run is expensive; a sample-efficient two-parameter optimizer matches the demonstrated 2D strength of AdaLIPO.
- Citrine Informaticscompany
Citrine's materials-informatics customers often start with small, expensive datasets; a low-evaluation optimizer complements its property-prediction models by proposing the few most informative next syntheses to run.
- National Renewable Energy Laboratoryorganization
NREL's thin-film and semiconductor process development is budget-limited in sample count; a method demonstrated to find 2D optima in roughly 30 evaluations stretches limited instrument and furnace time.
Research it builds on
- Image Processing On LineBlusseau, Samy, Velasco-Forero, Santiago, Angulo, Jesus et al. · 2026 · 156 citationsAll ideas from this paper →