Part V

Discovery Through Simulation and Optimization

Differentiable programming, scientific simulation, and intelligent experiment design

Overview

Part V explores how simulation and optimization accelerate the discovery loop. Differentiable programming makes entire simulation pipelines trainable end-to-end, while world models learn to predict system dynamics from data. The chapters cover optimization techniques for navigating vast design spaces, principled experiment design that maximizes information gain per trial, and experiment registries that ensure reproducibility across discovery campaigns.

Chapters

Chapter 42 Differentiable Programming Making simulations and computational pipelines end-to-end differentiable for gradient-based discovery. Chapter 43 Scientific Simulation AI-accelerated numerical simulation, surrogate models, and learned solvers for scientific computing. Chapter 44 World Models Learning predictive models of system dynamics for planning, counterfactual reasoning, and exploration. Chapter 45 Optimization for Discovery Black-box, multi-objective, and combinatorial optimization methods for navigating design spaces. Chapter 46 Experiment Design Active learning, Bayesian experimental design, and adaptive strategies that maximize information per trial. Chapter 47 Experiment Registries Tracking experiments, results, and metadata to ensure reproducibility across discovery campaigns.