Efficient open-source GPU implementation for multi-agent autochemotactic 2D modeling

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.