Scientists develop a faster way to optimize experiments with mixed variables

Researchers generalized a mathematical technique for optimizing expensive experiments when some variables are continuous and others are discrete — making it possible to use gradient-based methods in fully mixed search spaces. This means labs can run fewer iterations to find good settings when testing physical systems with both tunable and categorical parameters.

Bayesian optimization has been a workhorse in materials science, chemistry, and drug discovery because it cuts the number of expensive experiments needed to find good parameters. But it broke down when problems mixed continuous variables (like temperature) with discrete ones (like which chemical to use) — forcing labs back to slower grid search or random sampling. This generalizes an existing method to handle that mixed case efficiently, which matters most in autonomous lab settings where every iteration costs time and money. The practical effect is narrower than it sounds: this solves a real bottleneck for a specific category of optimization problem, but only for labs that already know how to implement Bayesian optimization.