Kinetic Field Theory: Phase-Space Dynamics

• Kinetic Field Theory is a statistical field theory that models classical particle trajectories in phase space to derive macroscopic observables via functional differentiation.
• It employs generating functionals and perturbative expansions, offering an analytic alternative to direct numerical simulations in dense fluids and cosmological structure formation.
• Applications include predicting non-linear matter power spectra and neutrino clustering, showcasing its versatility across astrophysical and condensed matter systems.

Kinetic Field Theory (KFT) is a non-equilibrium statistical field theory for ensembles of classical particles in phase space. Its defining move is to keep the microscopic Hamiltonian trajectories of particles as the primary dynamical objects, encode their statistics in a generating functional, and obtain macroscopic observables—density fields, response functions, power spectra, and higher cumulants—by functional differentiation. In this sense KFT is simultaneously a reformulation of classical kinetic theory and an analytic particle-based alternative to fluid perturbation theory and direct $N$-body simulation. The framework was first developed for Newtonian many-particle systems and dense fluids, where it reproduces the continuity and Jeans equations in the free theory and yields the BBGKY hierarchy when interactions are included, and was later specialized to cosmological structure formation, where it provides exact free-streaming results, systematic interaction expansions, and non-perturbative closures for the non-linear matter power spectrum (Das et al., 2011, Viermann et al., 2014, Bartelmann et al., 2016).

1. Genealogy and scope

The earliest field-theoretic formulations in this lineage recast Newtonian kinetic theory as an MSR-type functional integral over particle trajectories, with the number density $\rho$ and a response density $B$ identified as the core collective variables (Das et al., 2011). In that setting, the formalism was designed to support a self-consistent perturbation expansion in an effective interaction potential and to study dense-fluid phenomena such as ergodic–nonergodic transitions. A later formulation made the connection to classical kinetic theory explicit by showing that the free theory reproduces the continuity and Jeans equations of a collisionless gas and that a perturbative treatment of two-particle interactions yields the BBGKY hierarchy, with a truncation criterion tied to perturbative order (Viermann et al., 2014). Within the dense-fluid program, the low-frequency dynamics near the ergodic–nonergodic transition was found to be the same for Smoluchowski and Newtonian dynamics, despite their term-by-term density expansions being fundamentally different (Das et al., 2013).

Cosmological KFT emerged by applying the same basic field-theoretic machinery to cold dark matter viewed as an ensemble of classical particles in an expanding background. In this version, the emphasis shifted from dense-fluid response functions to the non-linear evolution of large-scale structure, with particles propagated in comoving coordinates and initial conditions drawn from Gaussian random fields fixed by the linear matter power spectrum (Bartelmann et al., 2016, Fabis et al., 2017). The cosmological theory was then extended in several directions: exact treatment of Gaussian phase-space correlations, fluctuation–dissipation relations for free-streaming ensembles, systematic comparison to Eulerian SPT, higher-order interaction expansions, modified-gravity applications, and non-CDM components such as relic neutrinos (Dombrowski et al., 2017, Kozlikin et al., 2020, Pixius et al., 2022, Heisenberg et al., 2019, Holm et al., 2023).

A common misconception is to identify KFT with a homogeneous cosmological formalism only. More recent work removes statistical homogeneity and isotropy altogether and develops KFT for compact, spatially inhomogeneous systems. In that setting, first-order perturbation theory and an iterated mean-field approximation were tested in solvable toy models, with the mean-field theory maintaining positivity and capturing collapse dynamics, and first-order perturbation theory reproducing critical phenomena in a self-gravitating sheet model (Bartelmann et al., 2 Sep 2025).

2. Microscopic and field-theoretic construction

The basic variables of KFT are particle phase-space coordinates $x_j=(q_j,p_j)$, or collectively $\boldsymbol{x}$. Dynamics is Hamiltonian, and the generating functional is built by enforcing the classical equations of motion inside a path integral. In a generic formulation one writes

$$Z = \int\mathcal{D}[\varphi]\int\mathcal{D}[\varphi_i]\, P(\varphi|\varphi_i)P(\varphi_i),$$

where $P(\varphi_i)$ is the initial distribution and $P(\varphi|\varphi_i)$ becomes a functional delta for deterministic classical trajectories (Heisenberg et al., 2019). In the MSR formulation for classical particles, the free generating functional takes the form

$$Z_0[\boldsymbol J,\boldsymbol K] = \int d\Gamma\, \exp\left( i\int_0^\infty dt\, \langle\boldsymbol J,\bar{\boldsymbol x}(t)\rangle \right),$$

with $d\Gamma$ the initial phase-space measure and $\rho$0 the free trajectory propagated by a retarded Green’s function (Bartelmann et al., 2016).

This structure has two immediate consequences. First, the path integral collapses to an ensemble average over initial conditions once the retarded Green’s function is known (Viermann et al., 2014). Second, interactions can be isolated in an operator acting on the free theory,

$\rho$1

so perturbation theory is organized in powers of the interaction operator rather than in powers of the density contrast (Bartelmann et al., 2016, Pixius et al., 2022).

Macroscopic observables are represented by composite operators. For the density field in Fourier space,

$\rho$2

and the corresponding operator is

$\rho$3

Acting with $\rho$4 such operators on the generating functional produces the $\rho$5-point density cumulants (Heisenberg et al., 2019). In the dense-fluid and Newtonian kinetic-theory formulations, the second core collective field is the response density $\rho$6, whose correlators encode how the system responds to perturbations and which enters the interaction action bilinearly with $\rho$7 (Das et al., 2011, Dombrowski et al., 2017).

A decisive conceptual difference from Eulerian fluid approaches is that KFT works directly in full phase space. There is no assumption of a single-valued velocity field, no moment truncation at the level of the equations of motion, and no shell-crossing pathology in the microscopic description. In cosmological language, “Hamiltonian trajectories cannot cross in phase space,” so KFT remains well-defined where SPT and LPT become problematic (Heisenberg et al., 2019, Pixius et al., 2022).

3. Free evolution, initial correlations, and resummation

The free theory in KFT is not trivial. Because the initial ensemble is correlated, exact free streaming already generates non-linear density cumulants. This point became central once the free generating functional was factorized while retaining the full hierarchy of initial momentum correlations. The factorized form shows that each particle pair contributes a universal factor $\rho$8, interpretable as a nonlinearly evolved density-fluctuation power spectrum for that pair, together with delta-function terms enforcing conservation constraints (Bartelmann et al., 2016). The same work shows that the complete hierarchy of initial momentum correlations is responsible for a large part of the characteristic non-linear deformation and mode transport in the density-fluctuation power spectrum, even before interactions are included (Bartelmann et al., 2016).

The exact treatment of Gaussian phase-space correlations was then extended beyond pure momentum correlations. By introducing a diagrammatic language inspired by the Mayer cluster expansion, explicit expressions for phase-space density cumulants of arbitrary $\rho$9-point order were obtained in the free theory, fully capturing the non-linear coupling of free-streaming kinematics induced by initial density-density, density-momentum, and momentum-momentum correlations (Fabis et al., 2017). This result supplies the exact free baseline on which interacting perturbation theory is built.

A second important development was the comparison to Eulerian SPT in the non-interacting regime. That analysis showed that KFT contains a complete resummation of SPT in free-streaming kinematics and that the exact free-streaming solution of KFT cannot be recovered in any finite order of SPT (Kozlikin et al., 2020). The statement is precise: expanding the exact KFT density power spectrum in powers of the initial power spectrum reproduces the SPT loop series order by order, but the full KFT result contains exponentials and damping factors that encode infinitely many SPT diagrams (Kozlikin et al., 2020).

The free theory also exhibits a kinematic balance between diffusion and structure accumulation. In the Born approximation, the non-linear power spectrum can be decomposed into a diffusive one-particle factor, a two-particle accumulation term arising from initial momentum correlations, and an interaction factor. The analysis of the time derivative of the power spectrum shows that diffusion and accumulation are delicately balanced because of the Gaussian form of the initial conditions, and that the response to arbitrary gradient forces is related to the evolution of diffusion through kinematic fluctuation–dissipation relations rooted in a time-reversal symmetry of the generating functional (Dombrowski et al., 2017).

4. Interactions, perturbation theory, and non-perturbative closures

In the cosmological formulation, pair interactions are encoded by

$B$0

with $B$1 the Fourier-space interaction potential. Expanding $B$2 yields a diagrammatic perturbation theory analogous to Feynman diagrams, but based on phase-space trajectories and response insertions rather than fluid vertices (Bartelmann et al., 2016, Pixius et al., 2022). A defining feature of KFT perturbation theory is that it expands in interactions, not in $B$3, and therefore does not use $B$4 as the organizing parameter (Pixius et al., 2022).

A complementary line of work keeps the average over initial conditions until the very last step. This reorganization greatly clarifies the perturbative treatment of interacting $B$5-body systems and allows a direct physical interpretation of intermediate results. Specialized to cosmological structure formation, it reproduces the linear growth of the cosmic density fluctuation power spectrum on all scales from microscopic, Newtonian particle dynamics alone, provided one keeps only linear initial correlations and sums sufficiently high interaction order (Heisenberg et al., 2022).

At the practical level, KFT has developed both perturbative and non-perturbative closures. The most widely used non-perturbative result is the mean-field expression for the non-linear matter power spectrum,

$B$6

with $B$7. Here $B$8 describes damping due to trajectory dispersion, $B$9 encodes the mean interaction, and the remaining integral builds up correlations from the initial momentum correlator $x_j=(q_j,p_j)$0 (Heisenberg et al., 2019). This equation is closed, analytic, non-perturbative, and parameter-free once the initial conditions and the background/interaction model are specified (Heisenberg et al., 2019).

Perturbative improvements beyond first order have also been worked out explicitly for inertial Zel’dovich trajectories. A more rigorous split between inertial motion and interactions leads to a modified Poisson equation and a Yukawa-like regulated interaction potential in Fourier space, suppressing spurious large-scale forces while recovering Newtonian behavior on small scales (Pixius et al., 2022). With this improved interaction treatment, second-order corrections to the late-time dark matter power spectrum systematically improve agreement with simulations on intermediate scales (Pixius et al., 2022).

5. Cosmological predictions and applications

The baseline cosmological application of KFT is the non-linear matter power spectrum in $x_j=(q_j,p_j)$1CDM. In that setting, the mean-field non-perturbative power-spectrum expression has been shown to agree very well with $x_j=(q_j,p_j)$2-body simulations up to $x_j=(q_j,p_j)$3 at $x_j=(q_j,p_j)$4, which established the framework as a viable analytic alternative to direct numerical evolution on mildly and moderately non-linear scales (Heisenberg et al., 2019). A separate perturbative analysis with improved interaction operators found that, at $x_j=(q_j,p_j)$5, first-order KFT agrees with simulations at the $x_j=(q_j,p_j)$6 level up to $x_j=(q_j,p_j)$7, while including interactions up to second order extends that range to $x_j=(q_j,p_j)$8; at $x_j=(q_j,p_j)$9, second order reaches $\boldsymbol{x}$0 (Pixius et al., 2022).

KFT has also been used as a framework for modified gravity. In generalized Proca vector–tensor theories, the formal structure of the KFT power-spectrum equation is unchanged; only the background expansion $\boldsymbol{x}$1, the effective gravitational constant $\boldsymbol{x}$2, the propagators $\boldsymbol{x}$3, and the time variable $\boldsymbol{x}$4 are modified (Heisenberg et al., 2019). For the parameter sets studied there, the non-linear matter power spectrum is enhanced relative to GR+$\boldsymbol{x}$5CDM, with the enhancement peaking around $\boldsymbol{x}$6; for one non-minimally coupled model with $\boldsymbol{x}$7, $\boldsymbol{x}$8, and $\boldsymbol{x}$9, the enhancement reaches up to $$Z = \int\mathcal{D}[\varphi]\int\mathcal{D}[\varphi_i]\, P(\varphi|\varphi_i)P(\varphi_i),$$0 at that scale at $$Z = \int\mathcal{D}[\varphi]\int\mathcal{D}[\varphi_i]\, P(\varphi|\varphi_i)P(\varphi_i),$$1 (Heisenberg et al., 2019).

A distinct application is the local clustering of relic neutrinos. In the $$Z = \int\mathcal{D}[\varphi]\int\mathcal{D}[\varphi_i]\, P(\varphi|\varphi_i)P(\varphi_i),$$2-one-body approximation, neutrinos are treated as test particles moving in an external gravitational potential sourced by CDM and baryons, so the full $$Z = \int\mathcal{D}[\varphi]\int\mathcal{D}[\varphi_i]\, P(\varphi|\varphi_i)P(\varphi_i),$$3-particle KFT reduces to a one-particle theory with a perturbative interaction expansion (Holm et al., 2023). The first-order density correction can be written as a Born-type integral of the halo density along free-streaming trajectories, and for neutrino masses consistent with cosmological bounds the resulting local overdensity agrees with state-of-the-art calculations at the percent level for $$Z = \int\mathcal{D}[\varphi]\int\mathcal{D}[\varphi_i]\, P(\varphi|\varphi_i)P(\varphi_i),$$4 (Holm et al., 2023). In the spherically symmetric case considered there, the numerical evaluation of the KFT integral for a given mass point takes $$Z = \int\mathcal{D}[\varphi]\int\mathcal{D}[\varphi_i]\, P(\varphi|\varphi_i)P(\varphi_i),$$5, compared with $$Z = \int\mathcal{D}[\varphi]\int\mathcal{D}[\varphi_i]\, P(\varphi|\varphi_i)P(\varphi_i),$$6 for full $$Z = \int\mathcal{D}[\varphi]\int\mathcal{D}[\varphi_i]\, P(\varphi|\varphi_i)P(\varphi_i),$$7-one-body backtracking (Holm et al., 2023).

These applications illustrate a general pattern: KFT preserves a universal microscopic formalism while changing only the background Hamiltonian, interaction kernel, and initial ensemble. This suggests a broad portability across cosmological sectors and matter components, provided the relevant Hamiltonian and initial correlations can be specified.

6. Conceptual status, limitations, and current directions

KFT occupies a distinctive place among analytic approaches to many-particle dynamics. It is not a fluid closure, because it never replaces the microscopic ensemble by a single-valued velocity field; it is not merely a reformulation of $$Z = \int\mathcal{D}[\varphi]\int\mathcal{D}[\varphi_i]\, P(\varphi|\varphi_i)P(\varphi_i),$$8-body simulation, because the ensemble is handled analytically through a generating functional; and in cosmology it is not simply another $$Z = \int\mathcal{D}[\varphi]\int\mathcal{D}[\varphi_i]\, P(\varphi|\varphi_i)P(\varphi_i),$$9-expansion, because perturbation theory is organized in the interaction operator while the free kinematics and initial-correlation hierarchy are kept intact (Pixius et al., 2022, Kozlikin et al., 2020).

At the same time, the framework has clear limitations. Mean-field cosmological KFT neglects higher interaction correlations and does not capture screening mechanisms such as Vainshtein screening on deeply non-linear scales; proceeding to such regimes requires going beyond the mean-field approximation and retaining more of the non-linear structure of $P(\varphi_i)$0 (Heisenberg et al., 2019). In the neutrino application, first-order Born trajectories underestimate clustering for larger neutrino masses, where bound orbits and stronger curvature of trajectories become important (Holm et al., 2023). In higher-order cosmological perturbation theory, one-loop expansions of the $P(\varphi_i)$1 kernels and practical replacements of exact damping factors are acknowledged approximations whose breakdown on small scales motivates non-perturbative evaluation or higher-loop treatments (Pixius et al., 2022).

A second misconception is that KFT is intrinsically tied to translation-invariant thermodynamic limits. Recent work on compact systems argues otherwise by formulating KFT without assumptions of statistical homogeneity and isotropy. In a solvable toy model with short-ranged interactions, first-order perturbation theory was compared to an iterated mean-field approximation scheme, and the mean-field theory was found to maintain positivity and capture collapse dynamics, allowing analytic estimates of blow-up times; in a self-gravitating sheet model, first-order perturbation theory reproduced critical phenomena (Bartelmann et al., 2 Sep 2025). A plausible implication is that future KFT developments may connect cosmological large-scale structure, compact self-gravitating systems, and classical kinetic theory more tightly than earlier homogeneous formulations allowed.

The current research trajectory therefore has three intertwined aims: to sharpen the microscopic formalism, to improve interaction treatments beyond first order and beyond mean field, and to broaden the class of admissible systems from homogeneous cosmologies to finite inhomogeneous ensembles. Across these variants, the invariant core remains the same: KFT is a field theory of classical phase-space trajectories whose observables are generated, not postulated.