Fig. 1.

(A) Optimizing patchy particle interactions. The optimization parameters (shown on the Top) are the patch locations and the interaction matrix of patch strengths. These parameters are used to run a forward simulation, and a loss function is computed. We then take the gradient of the loss function with respect to the optimization parameters and update the parameters accordingly. The loss functions vary for different optimization targets. (B) Gradient of loss function respect to parameters for optimization. The pseudo-code demonstrates how the gradient is computed based on the parameters for optimization. (C) Extrapolation to more performant MD engines. We test optimal parameters in HOOMD-blue, showing both that optimal parameters are valid across different MD engines and enabling rapid testing for longer simulations with more particles.