Violent Relaxation

A galaxy is a paradox waiting to be resolved. Its stars are so sparse that no two have ever collided, and the time random star-star encounters would take to shuffle their orbits into equilibrium — the two-body relaxation time — is millions of times the age of the Universe. By that reckoning a galaxy should still remember every detail of how it formed, a frozen archive of chaotic infall. Yet ellipticals and dark-matter halos are the opposite: smooth, virialized, and so regular that a single formula describes their light. Something relaxes them, fast, and it is not collisions. Donald Lynden-Bell named and solved this in 1967. During gravitational collapse or a major merger the collective potential $\Phi(\mathbf r,t)$ — the smooth well every star shares — does not sit still: it plunges, overshoots, and rings on a dynamical timescale. Because a star's specific energy obeys $\mathrm dE/\mathrm dt=\partial\Phi/\partial t$ , a moving well pumps energy into and out of every orbit at once, and in a few crossing times the ordered infall is scrambled into a broad, smooth energy distribution. This is violent relaxation: the mean field itself, not discrete encounters, drives the system to equilibrium, independently of how many stars there are. Lynden-Bell went further and asked what equilibrium the mixing should reach. Maximizing an entropy that respects Liouville's theorem, he found a distribution formally identical to Fermi–Dirac statistics — an isothermal-like state whose dispersion is independent of stellar mass, so violent relaxation makes no mass segregation. Real systems never quite get there: the stirring shuts off as the potential steadies, so relaxation stays incomplete, leaving the near-universal de Vaucouleurs and NFW profiles we observe.

Key takeaways

Violent relaxation is the rapid redistribution of stellar energies driven by a strongly time-varying collective potential during collapse, merging, or virialization of a self-gravitating system.
Because $\Phi(\mathbf r,t)$ changes on a dynamical timescale, individual star energies $E=\tfrac12v^2+\Phi$ are not conserved even though the system is collisionless: $\mathrm dE/\mathrm dt=\partial\Phi/\partial t$ , and the mean field does the shuffling.
It acts on a few crossing times $t_{\mathrm{cross}}\sim(G\bar\rho)^{-1/2}$ , independent of particle number $N$ — vastly faster than two-body relaxation $t_{\mathrm{relax}}\sim[N/(8\ln N)]\,t_{\mathrm{cross}}$ , which for a galaxy exceeds the Hubble time a million-fold.
Lynden-Bell (1967) predicted a Fermi–Dirac-like, mass-independent equilibrium; incomplete relaxation and phase-mixing yield the observed de Vaucouleurs $R^{1/4}$ / Sérsic and NFW profiles instead.
It explains how ellipticals and dark-matter halos reach smooth, virialized, memory-erased states out of chaotic cold collapse — and why merger remnants all look generic.
Key idea: in a rapidly changing collective potential a star's energy is not conserved ( $\mathrm dE/\mathrm dt=\partial\Phi/\partial t\neq0$ ), so the mean field itself relaxes the system in a few crossing times — no star-star collisions required.

The core insight

In a rapidly changing collective potential, individual stellar energies are not conserved ( $\mathrm dE/\mathrm dt=\partial\Phi/\partial t\neq0$ ), so the mean field itself drives the system to equilibrium in a few crossing times — no star-star collisions required.

Where this lives today

Violent relaxation is the connective tissue of modern structure formation, tested most sharply in the supercomputer. Cosmological N-body simulations — from Aquarius and Millennium to today's IllustrisTNG zoom-ins — assemble dark-matter halos by collapse and merger, and the halos relax into the near-universal NFW (or Einasto) profile within a few free-fall times, exactly the incomplete-Lynden-Bell outcome. The mass–concentration relation those profiles obey, calibrated on these runs, is now a workhorse of lensing and cluster cosmology. Observationally, integral-field spectrographs (MUSE, SAMI, MaNGA) map the smooth, pressure-supported velocity fields of ellipticals and confirm both the de Vaucouleurs surface brightness and the absence of mass segregation that violent relaxation predicts; the mechanism also acts in real time as galaxies plunge into forming clusters and phase-mix into the intracluster light. The open frontier is the incompleteness itself: why does relaxation stop where it does, and can a predictive theory — beyond Lynden-Bell's idealized maximum — explain why halos pick Einasto/NFW rather than any other profile? That question sits at the live interface of statistical mechanics, dynamics, and cosmology.

The physical picture

Picture a marble rolling in a bowl whose walls are being violently reshaped while it rolls. Even with no friction and no other marbles to bump, the marble gains or loses energy simply because the floor moves under it. Every star in a collapsing galaxy feels exactly this: the deepening, overshooting, ringing collective well pumps energy in and out of its orbit until the well steadies. That collective pumping — not encounters — is violent relaxation, and it is why orbital energy is the one thing not conserved even in a perfectly collisionless system. The first figure makes the mechanism a single picture: the shared potential $\Phi(r,t)$ drops and rings, and because the floor is moving, $\mathrm dE/\mathrm dt=\partial\Phi/\partial t$ carries each star to a new energy.

Figure

The mean-field well Φ(r,t) deepens and rings during collapse: because the floor moves under each star, dE/dt = ∂Φ/∂t ≠ 0 and orbital energies are pumped up and down — collective relaxation with no star-star collisions

The process erases memory. Because the energy a star gains depends on the phase of its orbit relative to the potential's swings, cold, ordered infall is scrambled into a broad, smooth energy distribution in a handful of crossing times — which is why merger remnants all look like generic ellipticals regardless of what fell together. The second figure lays the clocks side by side: violent relaxation finishes in a few crossing times while collisional relaxation would take a million Hubble times.

Figure

The clocks of relaxation for an N ≈ 10¹¹ galaxy: violent relaxation finishes in a few crossing times t_cross ≈ 5×10⁷ yr, while two-body relaxation needs t_relax ≈ N/(8 ln N) t_cross ≈ 2×10¹⁶ yr — over a million times the Hubble age 1.4×10¹⁰ yr, so collisions never act

The core equations

The engine of the whole subject is a one-line kinematic fact: in a collisionless system the only thing that can change an orbit's energy is the explicit time-dependence of the shared potential.

\frac{\mathrm dE}{\mathrm dt}=\frac{\mathrm d}{\mathrm dt}\!\left(\tfrac12 v^2+\Phi\right)=\frac{\partial\Phi}{\partial t} \qquad (1)

where: $E$ — specific (per unit mass) orbital energy of a star; $v$ — its speed; $\Phi(\mathbf r,t)$ — the mean-field potential, generally time-dependent. A static well ( $\partial_t\Phi=0$ ) conserves every orbit's energy; a swinging well does not — that is the entire mechanism. Violent relaxation is the evolution of $f$ under the coupled collisionless Boltzmann and Poisson equations, with the potential sourced by $f$ itself. The nonlinearity of this closed loop lets small coherent perturbations grow into the large potential swings that do the relaxing.

\frac{\partial f}{\partial t}+\mathbf v\!\cdot\!\nabla f-\nabla\Phi\!\cdot\!\frac{\partial f}{\partial\mathbf v}=0,\qquad \nabla^2\Phi=4\pi G\!\int f\,\mathrm d^3v \qquad (2)

where: $f(\mathbf r,\mathbf v,t)$ — phase-space distribution function; $\Phi$ — self-consistent potential; $G$ — Newton's constant. Along every orbit the fine-grained $f$ is constant ( $\mathrm df/\mathrm dt=0$ , Liouville) — the density is merely stirred and folded, never created or destroyed. The clock of relaxation is the crossing time — the time for a star to traverse the system once — which, through the virial estimate, is just the free-fall time and depends only on the mean density.

t_{\mathrm{cross}}\sim\frac{R}{\sigma}\sim\frac{1}{\sqrt{G\bar\rho}} \qquad (3)

where: $R$ — characteristic radius; $\sigma$ — velocity dispersion; $\bar\rho$ — mean mass density. Crucially there is no $N$ here: violent relaxation completes in a few $t_{\mathrm{cross}}$ whether the system has $10^6$ or $10^{12}$ stars. Contrast collisional two-body relaxation, whose timescale carries an explicit factor of $N$ and is therefore astronomically longer for any real galaxy — the fact that makes a collective mechanism mandatory.

t_{\mathrm{relax}}\approx\frac{N}{8\ln N}\,t_{\mathrm{cross}} \qquad (4)

where: $N$ — number of stars; $\ln N\approx\ln\Lambda$ — the Coulomb logarithm. For $N\sim10^{11}$ the prefactor is $\sim5\times10^{8}$ , pushing $t_{\mathrm{relax}}$ to $\sim10^{16}\,$ yr — so encounters simply cannot relax a galaxy in a Hubble time. Lynden-Bell asked what equilibrium the mixing in (2) reaches. Maximizing a phase-space mixing entropy at fixed mass, energy, and maximum fine-grained density gives a distribution formally identical to Fermi–Dirac statistics.

\bar f(E)=\frac{\rho_1}{e^{\beta(E-\mu)}+1}\;\xrightarrow[\text{dilute}]{}\;\rho_1\,e^{-\beta(E-\mu)} \qquad (5)

where: $\rho_1$ — maximum fine-grained phase density set by the initial state; $\beta$ — inverse "temperature" multiplier; $\mu$ — chemical-potential-like multiplier. Because $E$ enters as a specific energy with no stellar-mass weighting, reading (5) predicts the headline result: no mass segregation, and a mass-independent isothermal sphere in the dilute limit. The end-state is dynamical, not thermal, equilibrium — the system settles onto the virial relation tying its kinetic and potential energies.

2T+W=0,\qquad T=\tfrac12\!\int\rho\,\sigma^2\,\mathrm d^3r,\quad W=-\tfrac12\!\int\rho\,\Phi\,\mathrm d^3r \qquad (6)

where: $T$ — total kinetic energy; $W$ — total gravitational potential energy; $\rho$ — mass density. Reading (6): once virialized, $E=T+W=W/2=-T$ , the identity behind the cold-collapse example. Incomplete relaxation plus phase-mixing yields not the pure isothermal sphere of (5) — of infinite mass anyway — but a regular, finite family of profiles, summarized by the Sérsic law with the de Vaucouleurs $n=4$ case for ellipticals.

I(R)=I_e\exp\!\left\{-b_n\!\left[\left(\tfrac{R}{R_e}\right)^{1/n}-1\right]\right\} \qquad (7)

where: $I(R)$ — projected surface brightness; $R_e$ — effective (half-light) radius; $n$ — Sérsic index ( $n=4$ is the $R^{1/4}$ law). Reading (7): the universality of ellipticals, and analogously NFW halos, is the fingerprint of violent relaxation that stopped short of completion.

Derivation AHow a moving potential pumps orbital energy

The goal is to prove the central claim of the whole subject: that in a smooth collisionless system the specific energy of a star can change only through the explicit time-dependence of the shared potential, so that $\mathrm dE/\mathrm dt=\partial\Phi/\partial t$ . We start from the definition of orbital energy, differentiate along the trajectory, and insert the equation of motion, watching the two "motion" terms cancel and leave the explicit clock behind — the result that makes violent relaxation possible and also shows exactly when it switches off.

Step 1. Write the specific energy of one moving star, $E(t)=\tfrac12|\mathbf v(t)|^2+\Phi(\mathbf r(t),t)$ , where $\mathbf r$ and $\mathbf v$ evolve and $\Phi$ may depend explicitly on $t$ . Why: energy is not a fixed label on an orbit here — we must track it along the path because $\Phi$ is a moving landscape.

Step 2. Differentiate along the trajectory with the chain rule, separating the two ways $\Phi$ changes: $\mathrm dE/\mathrm dt=\mathbf v\!\cdot\!\dot{\mathbf v}+\nabla\Phi\!\cdot\!\dot{\mathbf r}+\partial\Phi/\partial t$ . Why: $\Phi$ changes both because the star moves to a new place (the convective term $\nabla\Phi\!\cdot\!\dot{\mathbf r}$ ) and because the well shifts in time ( $\partial\Phi/\partial t$ ) — only the second is genuine relaxation.

Step 3. Insert Newton's law in the mean field, $\dot{\mathbf v}=-\nabla\Phi$ and $\dot{\mathbf r}=\mathbf v$ , so $\mathbf v\!\cdot\!\dot{\mathbf v}=-\mathbf v\!\cdot\!\nabla\Phi$ . Why: the star responds to the smooth collective gradient, not to individual neighbours — the collisionless assumption made concrete.

Step 4. Add the kinetic and convective terms: $-\mathbf v\!\cdot\!\nabla\Phi+\nabla\Phi\!\cdot\!\mathbf v=0$ , an exact cancellation, leaving $\mathrm dE/\mathrm dt=\partial\Phi/\partial t$ . Why: a star sliding through a frozen landscape trades kinetic for potential energy exactly and keeps $E$ fixed; only a landscape that moves under it can change $E$ .

Step 5. Read off the regimes. If $\partial\Phi/\partial t=0$ energy is conserved and there is no violent relaxation — only phase-mixing at fixed $E$ . If $\Phi$ swings on the crossing time, energies are pumped up and down until the swinging stops. Why: the process is self-limiting — as the system virializes and the potential steadies, the engine that relaxed it shuts off, guaranteeing incomplete relaxation.

Go deeper · Derivation B · 1 minWhy a galaxy is collisionless over a Hubble time

Derivation BWhy a galaxy is collisionless over a Hubble time

The goal is to show quantitatively that discrete star-star encounters cannot relax a galaxy within the age of the Universe, which is what forces a collective mechanism in the first place. We estimate the mean-square velocity deflection a star accumulates in one passage, ask how many passages are needed for the deflections to add up to order the star's own speed, and convert that count into a time — a galaxy's star count then reveals a relaxation time orders of magnitude beyond the Hubble time.

Step 1. Sum the small-angle deflections from all two-body encounters in one crossing. The standard impulse calculation gives $(\Delta v/v)^2_{\rm one\ crossing}\approx 8\ln\Lambda/N$ , with $\ln\Lambda\approx\ln N$ the Coulomb logarithm cutting off very close and very distant encounters. Why: each encounter is feeble because gravity is long-range and the field nearly smooth, so it takes an enormous number to matter — the $1/N$ is the signature of granularity.

Step 2. Relaxation means the accumulated deflections reach order unity, $\sum(\Delta v/v)^2\sim1$ , requiring $n_{\mathrm{relax}}\approx N/(8\ln N)$ crossings. Why: this defines the relaxation time — when a star has "forgotten" its original velocity through the graininess of the potential rather than its smooth part.

Step 3. Convert crossings to time, $t_{\mathrm{relax}}=n_{\mathrm{relax}}\,t_{\mathrm{cross}}\approx[N/(8\ln N)]\,t_{\mathrm{cross}}$ . Why: the crossing time is the natural tick of any self-gravitating system, so every dynamical timescale is cleanest as a multiple of it.

Step 4. Evaluate for a galaxy: with $N=10^{11}$ , $\ln N\approx25$ , so $N/(8\ln N)\approx5\times10^{8}$ . Taking $t_{\mathrm{cross}}\sim5\times10^{7}\,$ yr gives $t_{\mathrm{relax}}\sim2\times10^{16}\,$ yr — over a million times the Hubble age $1.4\times10^{10}\,$ yr. Why: this is the crux of the subject — collisional relaxation is off by $10^6$ , so whatever smooths galaxies must be collective and $N$ -independent: violent relaxation.

Go deeper · Derivation C · 1 minThe Lynden-Bell distribution from entropy maximization

Derivation CThe Lynden-Bell distribution from entropy maximization

The goal is to predict the equilibrium a violently relaxing system heads toward by treating phase space statistically, while respecting the one constraint ordinary gas thermodynamics lacks: Liouville's theorem forbids the fine-grained phase density from ever exceeding its initial value. Coarse-graining phase space and counting the ways to distribute the incompressible phase fluid produces a mixing entropy formally identical to fermion counting; maximizing it at fixed mass and energy yields a Fermi–Dirac-like distribution that carries no dependence on individual stellar mass.

Step 1. Impose the Liouville constraint: the fine-grained $f$ takes only the values present initially (idealize as $0$ or a maximum $\rho_1$ ), and coarse-graining can dilute a cell but never raise it above $\rho_1$ . Why: this "phase-space exclusion" is the exact analogue of the Pauli principle — you cannot pack phase fluid denser than it started, so occupied and empty elements behave like fermions and holes.

Step 2. Write the mixing entropy counting arrangements of occupied versus empty elements, $S=-\int[\bar f\ln\bar f+(\rho_1-\bar f)\ln(\rho_1-\bar f)]\,\mathrm d^6w/\rho_1$ , with $\mathrm d^6w=\mathrm d^3r\,\mathrm d^3v$ . Why: this is the log of the number of microstates for a given coarse-grained $\bar f$ , its form fixed entirely by the maximum-occupancy constraint of Step 1.

Step 3. Maximize $S$ subject to fixed mass $M=\int\bar f\,\mathrm d^6w$ and energy $E_{\rm tot}=\int\bar f\,E\,\mathrm d^6w$ by extremizing $S-\alpha M-\beta E_{\rm tot}$ with multipliers $\alpha=-\beta\mu$ and $\beta$ . Why: mass and energy are the genuinely conserved quantities, so they alone set the shape of the most-probable coarse-grained state.

Step 4. Setting the variation to zero gives $\ln[(\rho_1-\bar f)/\bar f]=\beta E+\alpha$ , i.e. $\bar f(E)=\rho_1/[e^{\beta(E-\mu)}+1]$ , the Fermi–Dirac form. Why: the "+1" is the fingerprint of the exclusion constraint; drop it and you recover the Maxwell–Boltzmann exponential of ordinary gas statistics.

Step 5. Read off the consequences. In the dilute limit $\bar f\ll\rho_1$ the distribution becomes $\bar f\propto e^{-\beta E}$ , an isothermal sphere; and because $E$ is specific energy, $\beta$ multiplies energy per unit mass, so heavy and light stars share the same $\beta$ and relax identically. Why: this mass-independence — no mass segregation — is Lynden-Bell's sharpest prediction and the cleanest discriminator from two-body relaxation, which instead drives equipartition and sinks massive stars.

Limiting cases

The regime of violent relaxation is set by comparing the potential's variability timescale to the dynamical and collisional clocks. The table collects the corners.

Case	Result	Meaning
Static potential (relaxation off)	$\partial\Phi/\partial t=0\Rightarrow\mathrm dE/\mathrm dt=0$	every orbit conserves energy; only phase-mixing (at fixed $E$ ) and glacial two-body relaxation remain
Violent-relaxation regime	$t_{\mathrm{vr}}\sim t_{\mathrm{cross}}\sim(G\bar\rho)^{-1/2}$	during collapse or merging the well swings on the crossing time; energies redistribute in a few $t_{\mathrm{cross}}$ , independent of $N$
Collisional (two-body) regime	$t_{\mathrm{relax}}\approx\frac{N}{8\ln N}\,t_{\mathrm{cross}}\gg t_{\mathrm{cross}}$	for $N\sim10^{11}$ , $t_{\mathrm{relax}}\gg t_{\mathrm{Hubble}}$ ; encounters are irrelevant, so violent relaxation is essential
Complete Lynden-Bell limit	$\bar f\propto[e^{\beta(E-\mu)}+1]^{-1}\to e^{-\beta E}$	idealized full mixing gives a mass-independent isothermal-like state — never reached exactly, because stirring shuts off

Common trapViolent relaxation is just fast collisional thermalization

It is not, and the differences are physical. (i) It is collisionless: it conserves the fine-grained phase density exactly (Liouville), so it cannot raise fine-grained entropy — only the coarse-grained mixing entropy grows. (ii) Its clock is the crossing time $(G\bar\rho)^{-1/2}$ , not the two-body time $\propto N$ — essentially independent of particle number. (iii) It produces no mass segregation, because the kick $\partial_t\Phi$ acts per unit mass; two-body relaxation instead drives equipartition and sinks heavy stars. (iv) It is always incomplete: the potential fluctuations damp as the system virializes, so the stirring dies before the full Lynden-Bell distribution (of infinite mass) is reached. The end-state is a regular but non-isothermal profile that keeps partial memory of the collapse.

Worked examples

1 · Two clocks for an elliptical galaxy (fully worked)

An elliptical has half-light radius $R\approx10\,$ kpc, line-of-sight dispersion $\sigma\approx200\,$ km/s, and $N\approx10^{11}$ stars. How long does violent relaxation take, and how long would two-body relaxation take? Compare both to the Hubble time.

Step 1. Crossing time. With $R=10\,$ kpc $=3.09\times10^{20}\,$ m and $\sigma=2.0\times10^5\,$ m/s, $t_{\mathrm{cross}}\sim R/\sigma=1.54\times10^{15}\,$ s. Dividing by $3.156\times10^{7}\,$ s/yr gives t_cross ≈ 4.9×10⁷ yr ≈ 49 Myr. This is the free-fall clock $(G\bar\rho)^{-1/2}$ in disguise — the virial mass $M\sim\sigma^2R/G\approx9\times10^{10}\,M_\odot$ confirms an $\sim10^{11}$ -star galaxy.

Step 2. Violent-relaxation time. The mean field takes only a few crossing times to swing, overshoot, and settle, so t_vr ≈ (a few)×t_cross ≈ 2×10⁸ yr — comfortably shorter than the $\sim10^{10}\,$ yr since the galaxy formed, so it has relaxed many times over.

Step 3. Two-body relaxation time. With $\ln(10^{11})=25.3$ , $8\ln N=203$ , so $N/(8\ln N)=4.9\times10^{8}$ . Then $t_{\mathrm{relax}}=4.9\times10^{8}\times4.9\times10^{7}\,$ yr $=$ 2.4×10¹⁶ yr.

Step 4. Compare. With $t_H\approx1.4\times10^{10}\,$ yr, $t_{\mathrm{relax}}/t_H\approx\mathbf{1.7\times10^{6}}$ : collisional relaxation is over a million Hubble times away, while violent relaxation finished in the first few hundred Myr. The galaxy is smooth and virialized today because of the collective mechanism; not a single star-star encounter has contributed.

2 · Cold collapse virializes at half the initial radius (fully worked)

A cold (zero-velocity) uniform sphere of mass $M$ , radius $R_i$ collapses and violently relaxes. Using energy conservation and the virial theorem, find its final virial radius and speed. Evaluate for $M=10^{11}\,M_\odot$ , $R_i=20\,$ kpc.

Step 1. Initial energy. A uniform sphere has $W_i=-\tfrac{3}{5}GM^2/R_i$ , and starts cold, so $T_i=0$ and $E_i=W_i=-\tfrac{3}{5}GM^2/R_i$ . Total energy is conserved through the collapse because violent relaxation only redistributes energy among stars, never changing the system's total.

Step 2. Final state. Once relaxed, $2T_f+W_f=0$ , so $T_f=-\tfrac12 W_f$ and $E_f=T_f+W_f=\tfrac12 W_f$ . Setting $E_f=E_i$ gives $\tfrac12 W_f=W_i$ , i.e. $W_f=2W_i$ .

Step 3. Solve for the radius. Since $W\propto-1/R$ for a uniform sphere, $W_f=2W_i$ gives $1/R_f=2/R_i$ , i.e. R_f = R_i/2. Cold collapse always virializes at half the initial radius — a clean, memory-erasing result independent of $M$ .

Step 4. Characteristic speed. From $T_f=-\tfrac12 W_f=\tfrac35 GM^2/R_i$ and $T_f=\tfrac12 M\langle v^2\rangle$ , $\langle v^2\rangle=\tfrac65 GM/R_i$ . For $M=10^{11}\,M_\odot$ , $R_i=20\,$ kpc ( $GM=1.33\times10^{31}\,$ m³/s², $R_i=6.17\times10^{20}\,$ m), $\langle v^2\rangle=2.6\times10^{10}\,$ m²/s², so v_rms ≈ 160 km/s and a one-dimensional $\sigma\approx93\,$ km/s. These land squarely in the observed range for a modest elliptical — from nothing but energy conservation and the virial theorem, and independent of $M$ .

Physics of the portrait

Violent relaxation is a rare case where a single 1967 paper both named a phenomenon and set out its complete statistical mechanics in one stroke.

Origins · 1 minDonald Lynden-Bell and how this was discovered

Donald Lynden-Bell

1935–2018 · statistical mechanics of stellar systems

Donald Lynden-Bell was a British astrophysicist at Cambridge whose career reads as a catalogue of the field's foundational ideas. In 1962, with Olin Eggen and Allan Sandage, he argued from stellar orbits that the Milky Way's halo formed in a rapid monolithic collapse — the "ELS" picture that framed galaxy formation for a generation. Turning to the dynamics of that collapse, in 1967 he published "Statistical mechanics of violent relaxation in stellar systems" (MNRAS 136, 101), coining the term, identifying the mean-field energy exchange $\mathrm dE/\mathrm dt=\partial_t\Phi$ as its engine, and deriving a Fermi–Dirac-like equilibrium from a phase-space exclusion principle rooted in Liouville's theorem. The next year, with Roger Wood, he explained the "gravothermal catastrophe": self-gravitating systems have negative specific heat and can never reach a true entropy maximum — precisely why violent relaxation must be incomplete. In 1969 he proposed that the dead engines of quasars are supermassive black holes in ordinary galactic nuclei, now the consensus view, and he shared the inaugural Kavli Prize in Astrophysics in 2008. The through-line of his work is a physicist's instinct for importing deep ideas — statistical mechanics, thermodynamic instability, phase-space geometry — into gravitational dynamics.

History

1943

— Chandrasekhar quantifies two-body relaxation and dynamical friction in stellar systems

1962

— Eggen, Lynden-Bell and Sandage tie the Galaxy's shape to a rapid collapse

1967

— Lynden-Bell derives the statistical mechanics of violent relaxation

1982

— van Albada's cold-collapse simulations reproduce the de Vaucouleurs profile

1996

— Navarro, Frenk and White find the universal NFW dark-matter halo profile

Going further

Texts. Binney & Tremaine, Galactic Dynamics, 2nd ed., §4.10.2 (violent relaxation) and §1.2 (relaxation times) · Lynden-Bell, D. 1967, MNRAS 136, 101, "Statistical mechanics of violent relaxation in stellar systems" · Carroll & Ostlie, An Introduction to Modern Astrophysics, 2nd ed., Ch. 24–25 · Mo, van den Bosch & White, Galaxy Formation and Evolution, Ch. 5–6 (collisionless dynamics, virialization) · Dodelson, Modern Cosmology, halo formation and virialization. Key results to keep in mind: the chain-rule cancellation yields $\mathrm dE/\mathrm dt=\partial_t\Phi$ ; for $N=10^{11}$ with $t_{\mathrm{cross}}\approx4.9\times10^{7}\,$ yr, the two-body prefactor $N/(8\ln N)\approx5\times10^{8}$ gives $t_{\mathrm{relax}}\approx2.4\times10^{16}\,$ yr; entropy maximization produces the Fermi–Dirac-like distribution; and cold collapse of a uniform sphere virializes at $R_f=R_i/2$ with $v_{\rm rms}\approx160\,$ km/s.

Prerequisites. Collisionless Boltzmann Equation · Distribution Function and Phase Space · Two-Body Relaxation Time · Virial Theorem in Stellar Systems. Leads to. Jeans Theorem · Galactic Potential Models · Orbits in Spherical Potentials.

Ask the tutor: Ask me to walk you through why $\mathrm dE/\mathrm dt=\partial\Phi/\partial t$ shuts relaxation off the moment a galaxy virializes, or to work the cold-collapse "half the radius" result step by step.

Why is a galaxy called "collisionless" when it clearly has gravity acting?

"Collisionless" does not mean gravity is absent — it means the granular, star-by-star part of gravity is negligible over the age of the Universe. Each star responds to the smooth collective potential of all the others, not to the sharp tug of any neighbour. The measure is the two-body relaxation time $t_{\mathrm{relax}}\approx[N/(8\ln N)]\,t_{\mathrm{cross}}$ , which for $N\sim10^{11}$ is $\sim10^{16}\,$ yr — a million Hubble times. Over any realistic timescale the discrete encounters average out and the potential acts as if perfectly smooth, so the mean field dominates — exactly the setting in which violent relaxation operates.

If energy is conserved for the system, how can individual star energies change?

These are two bookkeeping levels. The total energy of the isolated system is conserved, but that total is continually reshuffled among the stars: when the collective potential deepens or rings, $\mathrm dE/\mathrm dt=\partial\Phi/\partial t$ moves specific energy into some orbits and out of others. One star climbs to a wider orbit while another sinks, and the sum stays fixed. Violent relaxation is precisely this redistribution — a broadening of the distribution of individual energies at constant total energy — turning cold, narrow-in-energy infall into a broad, smooth, relaxed energy spectrum.

What is the difference between violent relaxation and phase-mixing?

Both smooth a system, but differ. Violent relaxation changes individual orbital energies through a time-varying potential ( $\partial_t\Phi\neq0$ ), acting during the active collapse or merger. Phase-mixing acts even in a static potential: orbits at slightly different energies have slightly different periods, so a localized clump shears out and winds up in phase space, filling the region smoothly at fixed energy. They cooperate — violent relaxation broadens the energy distribution while the potential swings, phase-mixing keeps smoothing afterward. Neither raises the fine-grained entropy; both raise the coarse-grained (mixing) entropy.

Why does Lynden-Bell's distribution look like Fermi–Dirac statistics?

Because of a constraint ordinary gases lack. Liouville's theorem conserves the fine-grained phase density along every orbit, so it can never exceed its initial maximum $\rho_1$ — you cannot compress phase fluid beyond its starting density. Counting the ways to arrange incompressible phase elements under this maximum-occupancy rule is formally identical to counting fermions, which also cannot double-occupy a state. Maximizing the resulting entropy gives $\bar f=\rho_1/[e^{\beta(E-\mu)}+1]$ , the Fermi–Dirac form, with the " $+1$ " as the fingerprint of exclusion. In the dilute limit the constraint stops biting and the distribution relaxes to the Maxwell–Boltzmann exponential of an isothermal sphere.

Why is violent relaxation always incomplete, and why does that matter?

Because it switches itself off. The engine is a fluctuating potential, but as the system virializes those fluctuations damp — the well steadies, $\partial_t\Phi\to0$ , and the stirring dies before the idealized Lynden-Bell equilibrium is reached. There is a second reason: that full distribution is an isothermal sphere with infinite mass, so no finite system could realize it. The consequence matters: real galaxies and halos retain partial memory of their initial conditions. Instead of one universal state you get a limited family of regular but non-isothermal profiles — de Vaucouleurs and Sérsic for ellipticals, NFW or Einasto for halos — whose details encode formation history.

How is violent relaxation tested if we cannot watch a galaxy form?

Three ways. First, numerical experiments: "cold collapse" simulations start a near-static sphere, watch it collapse, overshoot, and ring, and confirm it settles into a relaxed $R^{1/4}$ -like profile in a few free-fall times — exactly the predicted timescale and end-state. Second, cosmological N-body simulations produce halos that relax into universal NFW/Einasto profiles, matching the incomplete-relaxation prediction. Third, observations: integral-field spectroscopy of ellipticals confirms their smooth, pressure-supported kinematics, de Vaucouleurs profiles, and — critically — the absence of mass segregation that distinguishes violent from two-body relaxation.