Efficient open-source GPU implementation for multi-agent autochemotactic 2D modeling
Description
We present a high-performance open-source GPU implementation for simulating large ensembles of autochemotactic active Brownian particles in two dimensions. The code couples continuous overdamped dynamics of self-propelled particle with a discrete chemoattractant field. This software enables simulations of autochemotactic channel formation, collective search strategies, multi-target foraging dynamics, and first-passage problems in both free exploration and goal-directed navigation scenarios, where effects of rotational and translational chemotactic responses, chemosecretion, resetting and target reinforcement are important. Our implementation addresses these challenges through GPU-accelerated algorithms: two-dimensional parallelization of chemoattractant deposition, multiplicative renormalization for efficient field decay, spatial hashing for particle interactions, and asynchronous multi-stream execution for ensemble statistics. We benchmark on NVIDIA A100 hardware and characterize algorithmic scaling with number of particles and chemoattractant deposition cutoff radius. We also compare GPU execution times with these on multi-core Intel Xeon and AMD EPYC CPU systems. We release the implementation as open-source software with examples and documentation, addressing community calls for unified computational tools in motile active matter.