Fokker-Planck Collision Operator Overview

• The Fokker-Planck collision operator is a drift-diffusion formulation that models binary collisions while conserving mass, momentum, and energy.
• It is central to kinetic theory and plasma physics, underpinning numerical schemes and machine learning surrogates with entropy dissipation and equilibrium relaxation properties.
• Multi-species extensions and advanced discretization methods ensure scalable, accurate plasma simulations applicable to turbulence, cosmology, and runaway electron dynamics.

The Fokker-Planck collision operator is a central construct in kinetic theory, statistical physics, and plasma physics, modeling the evolution of velocity distribution functions under the influence of binary collisions. Specializations such as the Fokker-Planck-Landau (FPL) and Dougherty (Lenard–Bernstein) operators are of particular relevance in plasma modeling, where their properties of conservation, entropy dissipation, and consistent relaxation to equilibrium underpin both theoretical investigations and practical simulation schemes. The operator appears in diverse contexts, from turbulence-resolving gyrokinetic codes to cosmological Boltzmann hierarchies and operator-learning neural surrogates.

1. Fundamental Structure and Formulations

The Fokker-Planck collision operator arises as the small-angle limit of the Boltzmann integral, representing collisional dynamics as a second-order differential—or equivalently, drift-diffusion—operator in velocity space. For a single-species distribution $f(v)$, the Landau (Coulomb) form is

$$C[f](v) = \frac{\partial}{\partial v_i} \left[ \int d^3v' \, A_{ij}(v,v')\, [f(v')\,\partial_{v_j} f(v) - f(v)\,\partial_{v'_j} f(v')] \right]$$

with the symmetric kernel

$$A_{ij}(v, v') = \Gamma \frac{ |v - v'|^2 \delta_{ij} - (v_i - v'_i)(v_j - v'_j) }{ |v - v'|^3 }$$

and $\Gamma$ encapsulating the interaction strength and Coulomb logarithm (Miller et al., 2020). Alternative formulations introduce either Rosenbluth potentials or recast the operator into drift-diffusion form: $$C[f](v) = -\partial_{v_i}[A_i(v) f(v)] + \frac12 \partial_{v_i} \partial_{v_j}[D_{ij}(v) f(v)]$$ where $A_i$ and $D_{ij}$ are velocity-dependent friction and diffusion tensors, constructed to reproduce exact momentum and heat-exchange rates in binary scattering (Ali-Haïmoud, 2018).

For practical and model-closure purposes—especially in computational plasma physics—simplifications are made, leading to the "Dougherty" (Lenard–Bernstein) operator of the advective-diffusive type: $$C[f] = \nu \, \nabla_v \cdot \left[ (\mathbf v - \mathbf u) f + t^2 \nabla_v f \right]$$ where $t^2$ and $\mathbf u$ are set by the instantaneous moments of $f$ (Francisquez et al., 2020, Hakim et al., 2019). This model preserves the principal invariants and the entropy structure (H-theorem), with particular rules for cross-species coupling (see below).

2. Conservation Laws and H-Theorem

The Fokker-Planck collision operator is constructed to annihilate the collision invariants—particle number, total momentum, and kinetic energy: $$\int C[f]\,d^3v = 0, \quad \int v\,C[f]\,d^3v = 0, \quad \int v^2\,C[f]\,d^3v = 0$$ both in the continuous analytic and, under appropriate discretization strategies, the numerical sense (Miller et al., 2020, Francisquez et al., 2020, Hakim et al., 2019, Wang et al., 2024). The operator also guarantees, via its symmetric structure, monotonic dissipation of entropy (the H-theorem): $$\frac{d}{dt} H[f] = \int C[f]\,( \ln f + 1 )\,d^3v \leq 0$$ with equality if and only if $f$ is a Maxwellian matching the moments. These properties persist under multi-species generalizations, with coupled forms ensuring exact conservation of global momentum and energy and global entropy dissipation (Francisquez et al., 2021, Filbet et al., 2022, Habbershaw et al., 2024).

3. Multi-Species and Cross-Species Coupling

In practical plasmas, fully nonlinear, multi-species collision operators are required. The generalization introduces coupled terms: $$C_s[f] = \sum_r \nu_{sr} \nabla_v \cdot \left[ (v-u_{sr}) f_s + t_{sr}^2 \nabla_v f_s \right]$$ where the cross-species primitive moments $u_{sr}, t_{sr}^2$ are obtained by imposing conservation symmetry and matching the physical relaxation rates of velocity and temperature derived from the underlying Boltzmann operator (Francisquez et al., 2021, Habbershaw et al., 2024, Filbet et al., 2022). The precise choice of these moments can be tuned to best approximate the full Coulomb relaxation rate in cases of mass and temperature disparity.

Recent models admit additional parameters—such as mixture-weighting coefficients $α_{i,j}, β_{i,j}$—that, under certain sum rules, recover the exact momentum and heat-exchange rates of the full Boltzmann operator (Habbershaw et al., 2024). These models maintain entropy monotonicity, positive-definiteness, and correct relaxation to a Maxwellian equilibrium with common velocity and temperature for all species.

4. Numerical Discretization: Methods and Conservation

Conservative, high-order discretization is essential for accurate long-time integration. Established discontinuous Galerkin (DG) schemes rely on "weak-equality" for reconstructing primitive moments, multiple integrations by parts to enforce conservation, and recovery-based polynomial expansions for evaluating derivatives (Francisquez et al., 2020, Hakim et al., 2019).

For the axisymmetric 0D-2V Fokker-Planck-Rosenbluth equation, spectral schemes based on Legendre polynomials (angular coordinate) and "King" expansions (speed coordinate) achieve exponential convergence. Post-step manifold-projections correct residual mass, momentum, and energy errors, enforcing exact conservation down to machine precision (Wang et al., 2024). Implicit-in-time integrators (e.g., backward-Euler, Crank–Nicolson, JFNK) allow for large stable timesteps even under collision-frequency stiffness.

Operator-splitting and spectral approaches using Hermite expansions have also been developed for full Fokker-Planck-Landau equations, with quadratic collision terms handled at low Hermite indices and diffusive surrogates for tail modes (Li et al., 2020). These strategies facilitate efficient and accurate simulation across a wide range of Knudsen and collisionality regimes.

5. Machine Learning and Operator Surrogates

The high computational cost of directly evaluating the dense, nonlinear Fokker-Planck-Landau operator in modern gyrokinetic codes has motivated the development of machine-learned surrogates. In XGC, an encoder–decoder neural network—combining VGG-style convolutional encoders and ReNet recurrent bottlenecks—is trained to map input velocity-space histograms to the change $\Delta f$ prescribed by the FPL collision operator (Miller et al., 2020). The neural network is penalized for violations of mass, momentum, and energy conservation

$$L_{\text{total}} = L_2 + \lambda_n L_n + \lambda_m L_m + \lambda_E L_E$$

enabling the network to capture both the effect of collisions and to preserve the fundamental invariants with accuracy down to $10^{-4}$ for relative moment violations.

Inference via such surrogates offers $O(n)$ scaling in the number of plasma species, in sharp contrast to the $O(n^2)$ scaling of explicit linear-algebra solvers, enabling feasible simulation for multispecies turbulence problems. Machine-learned surrogates have been further enhanced by augmented Lagrangian penalization and alternative architectures (e.g., U-Nets, physics-informed graph convnets), and are currently being integrated into production 5D gyrokinetic workflows (Miller et al., 2020, Lee et al., 2022).

Operator learning with PINNs/encoder–decoders has also found application for the Fokker-Planck-Landau equation outside of XGC, for instance, in the "opPINN" framework, which demonstrates mesh-free, convergent approximations with theoretical guarantees for the surrogate operator error (Lee et al., 2022).

6. Applications and Physical Contexts

Fokker-Planck collision operators govern transport and relaxation in a wide variety of kinetic problems:

• Turbulent transport in fusion devices: The gyro-averaged and finite Larmor radius (FLR) versions of the operator, coupled with gyrokinetic solvers, capture collisional thermalization, neoclassical transport, and impurity effect in magnetically confined plasmas (Miller et al., 2020, Bostan et al., 2012).
• Cosmological Boltzmann codes: In cosmic microwave background and large-scale-structure analyses, Boltzmann-Fokker-Planck formulations accurately model DM–baryon momentum and heat exchange, essential for parameter estimation of dark-matter interactions (Ali-Haïmoud, 2018).
• Runaway electron production: Hybrid Fokker-Planck plus Boltzmann operators are required to handle both diffusive (small-angle) and discrete, non-diffusive (large-angle/hard) collisions in weakly-ionized plasmas, with direct implications for runaway generation during tokamak startup (Lee et al., 3 Sep 2025).
• Quantum and degenerate electron systems: Quantum Fokker-Planck models preserve Pauli blocking, exact conservation, and entropy principles for degenerate electrons colliding elastically with ions and among themselves, with DG discretizations delivering unconditional stability and accuracy (Le, 2021).

7. Outlook and Ongoing Developments

Current developments target further improvements in:

• Scalable operator learning: Multi-species, higher-dimensional, and non-local operators via deep neural surrogates, embedding constraint penalties, and, where possible, operator-theoretic spectral structure (Miller et al., 2020, Lee et al., 2022).
• Discrete conservation and entropy enforcement: Advanced spectral and DG schemes, manifold projection methods, and post-processing stages guarantee exact conservation even for complex coupled systems (Wang et al., 2024, Francisquez et al., 2020).
• Extensive parameter matching: Flexible multi-parameter models calibrated to Boltzmann–Coulomb rates for arbitrary mixtures and mass ratios (Habbershaw et al., 2024, Francisquez et al., 2021).
• Hybrid kinetic–fluid models: Sharp asymptotic limits (e.g., adiabatic electron treatment in multispecies Fokker-Planck) supporting hybrid simulation strategies (Filbet et al., 2022).

These methodological advances position the Fokker-Planck collision operator—and its modern numerical and data-driven realizations—as essential tools for first-principles, conservation-law-respecting kinetic simulation in plasma science and beyond.