Particle-in-Cell Methods
Particle-in-cell couples moving particles to a grid-based field, a workhorse of plasma simulation with distinctive parallel-computing demands.
Particles and a grid together
The particle-in-cell (PIC) method models a plasma as many charged macro-particles moving through space, coupled to electromagnetic fields represented on a grid. Each step: deposit particle charge and current onto the grid, solve the field equations on the grid, interpolate the fields back to particle positions, and push the particles forward. It captures kinetic effects that fluid models miss.
The compute profile
- Particle push: highly data-parallel, arithmetic-heavy
- Charge deposition (scatter): particles write to grid cells, causing memory contention
- Field solve: a grid-based, often communication-bound step
Parallelization
PIC codes combine domain decomposition of the grid with distribution of particles. The challenge is that particles move between subdomains as the simulation evolves, so they must be migrated between processes and the workload drifts, demanding periodic load balancing. Deposition, where many particles update shared cells, needs care to avoid races on both CPU and GPU.
On GPUs
PIC maps well to GPUs because the particle push is massively parallel, but the scatter step suffers from atomic-update contention and irregular memory access. Sorting particles by cell improves locality and coalescing, and is a standard optimization that turns a memory-bound deposition into a far faster one.
Role in fusion design
PIC and related kinetic methods simulate plasma behavior that informs magnetic-confinement design. Kronos studies its machines, the Hyperion breeder and the burner generators, in simulation using such tools; these remain design and simulation efforts, with no hardware net-gain claim before first-of-a-kind operation.