Vector Resonant Relaxation

• Vector resonant relaxation is a gravitational mechanism where coherent torques rapidly change the directions of orbital angular momentum in near-Keplerian stellar systems while keeping orbital energies nearly fixed.
• It employs orbit-averaged Hamiltonians and multipolar expansions to model both ballistic and diffusive changes in orbital-plane orientations, elucidating fast reorientation dynamics.
• This mechanism explains astrophysical phenomena such as the isotropization of inner S-clusters and warped disc structures in young stellar populations, informing dark cluster constraints.

Vector resonant relaxation is the collective, long-range process by which the directions of orbital angular momentum vectors in a nearly-Keplerian stellar system evolve through coherent gravitational torques, while orbital energies and, on the VRR timescale, angular-momentum magnitudes remain nearly fixed. In nuclear star clusters around a supermassive black hole, it is the fastest gravitational mechanism shaping the geometry of stellar orbits, acting on orbital-plane orientations much faster than scalar resonant relaxation changes eccentricities and much faster than non-resonant two-body relaxation changes orbital energies (Fouvry et al., 2018, Máthé et al., 2022).

1. Physical setting and dynamical variables

VRR operates in systems where a central massive black hole dominates the potential, so stars move on nearly Keplerian ellipses and the total potential is almost stationary. In this regime it is natural to use orbital elements or Delaunay variables such as $(M,\omega,\Omega,\Lambda,J,z)$, with

$$\Lambda = \sqrt{G M_\bullet a}, \qquad J = \Lambda \sqrt{1-e^2}, \qquad z = J \cos I,$$

where $a$ is the semimajor axis, $e$ the eccentricity, and $I$ the inclination. The Keplerian frequency is

$\nu_{\rm Kep}(a) = \sqrt{\frac{G M_\bullet}{a^3}}.$

Because the orbital period is much shorter than secular timescales, one averages over the fast Kepler motion and treats each orbit as an orbit-averaged “wire” or annulus; on these timescales the semimajor axis is adiabatically conserved, and for VRR the principal slow variable is the unit angular-momentum vector $\hat{\mathbf L}$ (Tep et al., 2021).

A complementary formulation labels each orbit by the conserved parameters

$K=(m,a,e),$

together with the time-dependent orbital-plane normal $\hat{\boldsymbol L}(t)$. On timescales longer than orbital motion and apsidal precession, but shorter than scalar resonant relaxation and non-resonant relaxation, stars are therefore represented by annuli extending from pericentre to apocentre. VRR is the dynamics of a set of long-range coupled unit vectors $\hat{\boldsymbol L}_i(t)$, and it changes inclination and node while leaving $$\Lambda = \sqrt{G M_\bullet a}, \qquad J = \Lambda \sqrt{1-e^2}, \qquad z = J \cos I,$$0, $$\Lambda = \sqrt{G M_\bullet a}, \qquad J = \Lambda \sqrt{1-e^2}, \qquad z = J \cos I,$$1, and $$\Lambda = \sqrt{G M_\bullet a}, \qquad J = \Lambda \sqrt{1-e^2}, \qquad z = J \cos I,$$2 essentially fixed (Martínez et al., 2020).

2. Orbit-averaged Hamiltonian and torque structure

The orbit-averaged gravitational interaction between annuli admits a multipolar expansion. In pairwise form, the VRR Hamiltonian may be written as

$$\Lambda = \sqrt{G M_\bullet a}, \qquad J = \Lambda \sqrt{1-e^2}, \qquad z = J \cos I,$$3

where $$\Lambda = \sqrt{G M_\bullet a}, \qquad J = \Lambda \sqrt{1-e^2}, \qquad z = J \cos I,$$4 are Legendre polynomials and the coupling coefficients $$\Lambda = \sqrt{G M_\bullet a}, \qquad J = \Lambda \sqrt{1-e^2}, \qquad z = J \cos I,$$5 encode the dependence on masses, semimajor axes, eccentricities, and radial overlap. Only even $$\Lambda = \sqrt{G M_\bullet a}, \qquad J = \Lambda \sqrt{1-e^2}, \qquad z = J \cos I,$$6 contribute, and the evolution is dominated by torques among stars with radially overlapping orbits (Kocsis et al., 2014).

In mean-field notation, the same dynamics is expressed through spherical harmonics on the orientation sphere,

$$\Lambda = \sqrt{G M_\bullet a}, \qquad J = \Lambda \sqrt{1-e^2}, \qquad z = J \cos I,$$7

with

$$\Lambda = \sqrt{G M_\bullet a}, \qquad J = \Lambda \sqrt{1-e^2}, \qquad z = J \cos I,$$8

The multipole coefficients $$\Lambda = \sqrt{G M_\bullet a}, \qquad J = \Lambda \sqrt{1-e^2}, \qquad z = J \cos I,$$9 arise from orbit-averaging the Newtonian interaction over the annuli, and stronger couplings occur for similar semimajor axes and for more radial orbits, which have smaller $a$0 and are easier to reorient (Martínez et al., 2020).

After averaging over orbital phase and often over apsidal precession, the annulus surface density becomes

$a$1

so the dynamical variables reduce to orbit normals $a$2. The equations of motion take the precessional form

$a$3

which preserves $a$4 and therefore the eccentricity on the VRR timescale (Kocsis et al., 2014).

3. Timescale hierarchy, coherence, and stochastic evolution

In the Galactic Centre, the relevant ordering for an S-star such as S2 is

$a$5

with characteristic values $a$6 yr for the orbital period, $a$7 yr for in-plane precession, $a$8, $a$9, and $e$0. This hierarchy captures the central fact that orbital-plane reorientation is much faster than eccentricity relaxation and enormously faster than classical two-body relaxation (Tep et al., 2021).

A standard estimate for the VRR timescale is

$e$1

or equivalently $e$2. The $e$3 enters because the residual torque is produced by Poisson fluctuations of many annuli, so the net torque amplitude scales with the root-mean-square mass fluctuation rather than with the total mass (Bar-Or et al., 2018, Kocsis et al., 2014).

The stochastic description separates a ballistic regime from a diffusive regime. For the background multipole coefficients, the two-point correlation is well approximated by a temporal Gaussian,

$e$4

with $e$5 and a coherence time $e$6. At $e$7, orientation changes grow ballistically; at $e$8, they become diffusive (Martínez et al., 2020).

Numerically, VRR is well described as a random walk of orbit normals on the sphere, with angular step size ranging from $e$9-$I$0 radian. The autocorrelation of the spherical multipole moments decays exponentially, and its e-folding time provides an operational definition of the VRR timescale (Kocsis et al., 2014).

4. Relation to scalar resonant relaxation, non-resonant relaxation, and general kinetic theory

The standard distinction is between vector resonant relaxation, which changes orbital orientation, and scalar resonant relaxation, which changes the magnitude of angular momentum and therefore eccentricity. The residual torques associated with scalar RR are randomized by the in-plane orbital precession, whereas the residual torques associated with vector RR persist on longer timescales, as they are randomized by the changes of the orbital orientations themselves. This is why VRR is typically faster than SRR but can only affect the direction of the angular momentum vector (Bar-Or et al., 2018).

In a nearly-Keplerian nucleus, relativistic apsidal precession and mass precession strongly affect SRR. Schwarzschild precession can destroy apsidal resonances and produce the “Schwarzschild barrier,” suppressing scalar resonant relaxation at very high eccentricity. By contrast, the in-plane character of Schwarzschild precession implies that it suppresses changes in $I$1 much more directly than changes in $I$2; VRR therefore remains comparatively robust except very close to the black hole. Loss-cone feeding is primarily controlled by diffusion in $I$3, namely scalar RR plus two-body relaxation, not by VRR alone (Bar-Or et al., 2018).

A more general kinetic perspective is that the separation into scalar and vector channels is exact only in restricted limits. In the first-principles Ring kinetic theory of Keplerian stellar systems, resonant relaxation is driven by both apsidal and nodal resonances, so in general geometries the distinction between scalar-RR and vector-RR disappears. The same two-Ring correlation function and wake function underlie changes in $I$4, $I$5, $I$6, and $I$7; scalar and vector RR are then different projections of one secular collisional process (Sridhar et al., 2015).

5. Astrophysical manifestations and equilibrium structures

In the Galactic Centre, VRR has been invoked to explain both the isotropization of the inner S-cluster and the persistence of anisotropic outer populations. The inner S-stars have ages comparable to or larger than the VRR time, while the outer young stars still display disc-like alignment. For the young clockwise disc, the normals to the mean orbital planes in the inner and outer third differ by $I$8–$I$9, and this warp arises naturally through vector resonant relaxation between the disc and the surrounding old stellar cluster (Kocsis et al., 2010, Martínez et al., 2020).

VRR also enters directly in constraints on the unresolved dark cluster around Sgr A*. In modelling the S-cluster, the old background cluster is commonly assumed to be spherically symmetric in orientations and thermal in eccentricity, an assumption justified by the short VRR timescale of old stars. At the same time, systems with a large mass spectrum can depart from this simple picture: heavy components such as intermediate-mass black holes can follow strongly anisotropic structures, aligned within the same disc, so VRR is both a mechanism of isotropization for light old populations and a mechanism of anisotropic ordering for massive components (Tep et al., 2021).

This mass dependence is central to the equilibrium theory. Maximum-entropy calculations show that the most likely outcome of VRR in galactic nuclei is that the orbital planes of the most massive stars spontaneously self-align within a narrow disc, and the heaviest stellar objects are found to live within a thin equatorial disk. Microcanonical explorations of the $\nu_{\rm Kep}(a) = \sqrt{\frac{G M_\bullet}{a^3}}.$0 landscape likewise show anisotropic mass segregation: the distribution of more massive objects is more flattened than that for lighter objects, suggesting that stellar black holes reside in disc-like structures within nuclear star clusters for a wide range of initial conditions (Magnan et al., 2021, Máthé et al., 2022).

VRR is not restricted to SMBH-dominated nuclei. Direct $\nu_{\rm Kep}(a) = \sqrt{\frac{G M_\bullet}{a^3}}.$1-body simulations show that it operates efficiently in globular clusters with $\nu_{\rm Kep}(a) = \sqrt{\frac{G M_\bullet}{a^3}}.$2, where the distribution of orbital planes relaxes much more rapidly than the distribution of the magnitude of angular momentum and the radial action. This implies an internal statistical equilibrium in orientation space while the full cluster remains out of equilibrium, a state described as quenched disorder (Meiron et al., 2018).

A distinct application concerns binaries near a massive black hole. The Lidov–Kozai mechanism is only effective when a binary is highly inclined relative to its orbit around the black hole, but VRR can bring a low-inclination binary into an “active” Lidov–Kozai regime. Monte Carlo calculations find that the merger fraction enhancement due to LK-VRR dynamics is up to a factor of $\nu_{\rm Kep}(a) = \sqrt{\frac{G M_\bullet}{a^3}}.$3 for $\nu_{\rm Kep}(a) = \sqrt{\frac{G M_\bullet}{a^3}}.$4, decreases sharply with increasing $\nu_{\rm Kep}(a) = \sqrt{\frac{G M_\bullet}{a^3}}.$5, and for the Galactic Center remains $\nu_{\rm Kep}(a) = \sqrt{\frac{G M_\bullet}{a^3}}.$6 times lower than the LIGO/VIRGO limit (Hamers et al., 2018).

6. Numerical methods, statistical theories, and open problems

The secular and stochastic character of VRR has motivated several complementary calculational frameworks. The symplectic integrator N-ring follows the dynamical evolution of a cluster of $\nu_{\rm Kep}(a) = \sqrt{\frac{G M_\bullet}{a^3}}.$7 stars through vector resonant relaxation by averaging pairwise interactions over the orbital period and periapsis-precession timescale; it was used to evolve clusters containing over $\nu_{\rm Kep}(a) = \sqrt{\frac{G M_\bullet}{a^3}}.$8 stars for tens of relaxation times and established the random-walk description of orbit normals (Kocsis et al., 2014). For young stellar discs, a piecewise Markovian prescription calibrated to simulations gives an efficient description of neighbour separation and disc dilution in orientation space (Martínez et al., 2020).

At the level of equilibrium statistical mechanics, one line of work maps orbit-averaged annuli to rigid, disk-shaped tops on a bounded canonical phase space. In that formulation the effective Hamiltonian consists of a kinetic precession–nutation term plus a long-range interaction

$\nu_{\rm Kep}(a) = \sqrt{\frac{G M_\bullet}{a^3}}.$9

and the resulting thermal equilibria display ordered and disordered phases of orbital planes. A related maximum-entropy method determines the long-term distribution of orientations for multi-population nuclei, though in its axisymmetric implementation it excludes spontaneous symmetry breaking such as warped disks (Roupas, 2019, Magnan et al., 2021).

Recent work has pushed the dynamical theory beyond bare stochastic closures. One-loop implementations of the Martin–Siggia–Rose formalism for VRR quantitatively predict the two-point two-time correlation function, the renormalised three-point interaction vertex, and the three-point three-time correlation function, using VRR as a long-range, non-linear, and correlated testbed for closure theory (Flores et al., 26 Sep 2025).

Several issues remain open. A complete treatment should couple VRR to SRR and non-resonant relaxation in multiple populations, because the same dark cluster properties that shape eccentricity relaxation also shape orientation relaxation. Background isotropy, test-particle limits, and axisymmetry assumptions simplify current calculations but exclude anisotropic equilibria, self-gravitating disc responses, and feedback between massive and light components. More generally, the broad Ring kinetic theory indicates that apsidal and nodal resonances are parts of one secular collisional problem, so a full theory of VRR must ultimately be embedded in a unified kinetic description of resonant relaxation rather than treated as an isolated channel (Tep et al., 2021, Sridhar et al., 2015).