The Delta-f Method
The delta-f method evolves only the perturbation away from a known background distribution, sharply reducing particle noise in kinetic simulations.
Splitting the distribution
Full-f particle methods sample the entire distribution function, so statistical noise contaminates even the large, well-known background. The delta-f method writes the distribution as a known analytic background f0 plus a perturbation delta-f, and lets the computational particles carry only delta-f as a weight. Since delta-f is typically much smaller than f0, the same number of particles yields far less noise in the quantity of interest.
Each particle carries a time-evolving weight whose equation of motion follows from the kinetic equation applied to the split. The background is treated analytically, and only its deviation is sampled statistically.
Where it applies
Delta-f is most effective when the perturbation stays small relative to the background, as in linear and weakly nonlinear microturbulence and instability growth. It is the standard approach in many gyrokinetic codes studying transport driven by small-amplitude fluctuations.
Limits and extensions
When the perturbation grows large, or when the background itself must evolve, the noise advantage erodes and particle weights can spread, degrading accuracy. Adaptive and full-f hybrids periodically reset or renormalize the background to keep the split efficient. Careful weight control is needed to avoid a slow growth of sampling error over long runs.
- Distribution split f = f0 + delta-f
- Particles carry weights evolving delta-f, not full f
- Noise reduced roughly by the ratio delta-f/f0
- Weight growth limits validity in strongly nonlinear regimes
By focusing computational effort on the physically interesting perturbation, the delta-f method makes first-principles turbulence and transport studies tractable at resolutions that full-f sampling could not reach economically.