The Virial Theorem

Every modern N-body code ships with a virial diagnostic. The ratio $2T/|V|$ is computed at each snapshot, and its settling to unity is the standard operational test that a simulated halo, star cluster, or merger remnant has finished relaxing — equation (5) running live inside gravity solvers from GADGET to modern GPU codes. Simulators also read it in reverse: a halo with $2T/|V|$ far from 1 is flagged as unrelaxed and its "virial mass" quoted with suspicion. On the observational side, cluster masses estimated from velocity dispersions via the virial theorem are now routinely cross-checked against weak gravitational lensing, which weighs the same clusters with entirely independent physics. The Euclid mission's survey data — with cluster catalogs growing through 2026 — makes this comparison at scale, and the concordance of dispersion-based virial masses with lensing masses is one of the quiet, load-bearing consistency checks of the dark-matter paradigm. Meanwhile Chandrasekhar's stellar-structure virial still anchors the theory of white dwarfs: the balance between degenerate-electron kinetic energy and gravitational binding, run through the relativistic version of the theorem, is what sets the Chandrasekhar mass limit that every Type Ia supernova respects.

Picture a comet on a closed, eccentric orbit. Near perihelion it whips past the Sun — kinetic energy spikes, potential energy dives; near aphelion it crawls, and the ledger reverses. Instant by instant, $T$ and $V$ swing wildly and tell you nothing universal. But watch their running time averages: within a few periods each settles onto a fixed value, and the two values are locked in the ratio $2\langle T\rangle = -\langle V\rangle$ — the same lock for every bound orbit in a $1/r$ potential, whatever its eccentricity. The virial theorem is the statement that this lock is not a Kepler coincidence but a consequence of boundedness alone.

Why should averages be constrained when instants are not? Think of $G=\sum_i\mathbf{p}_i\cdot\mathbf{r}_i$ as a bookkeeping scalar. Its growth rate collects $2T$ from the kinetic side and $\sum_i\mathbf{F}_i\cdot\mathbf{r}_i$ from the force side. If the books are never allowed to run off to infinity — bounded positions, bounded momenta — then whatever one term contributes on average, the other must on average cancel. That forced cancellation is the theorem. And it shows exactly where the hypothesis bites: on an escape trajectory $G$ grows without bound, its average growth rate stays finite forever, and no relation between $\langle T\rangle$ and $\langle V\rangle$ survives.

The exponent in $2\langle T\rangle = n\langle V\rangle$ encodes how steeply the potential climbs with distance. A stiff confining potential (large $n$) demands a large share of kinetic energy to keep the motion trapped; an attractive $1/r$ potential ($n=-1$) lets the system bind with negative total energy, storing twice as much (negative) potential energy as kinetic. And there is a deeper reason this particular $G$ works, unpacked below: $G$ generates dilations, so the theorem is scaling analysis promoted to an exact statement about time averages.

Everything begins with a single scalar built from the phase-space coordinates. Sum the dot product of each particle's momentum with its position and you have a function $G(q,p)$ on phase space — no dynamics yet, just bookkeeping:

where: $\mathbf{p}_i$ — the momentum of particle $i$; $\mathbf{r}_i$ — its position vector; $G$ — a scalar phase-space function with the dimensions of action (momentum times length). For $\mathbf{p}_i = m_i\dot{\mathbf{r}}_i$ it equals $\tfrac12\,\mathrm{d}I/\mathrm{d}t$ with $I=\sum_i m_i r_i^2$ the moment of inertia about the origin — $G$ measures whether the system is currently swelling ($G>0$) or contracting ($G<0$). Differentiate along the motion and the product rule splits $\dot G$ into two terms with sharply different physical meanings. One collects the momenta acting on the velocities; the other collects the forces acting on the positions:

where: $\mathbf{F}_i=\dot{\mathbf{p}}_i$ — the net force on particle $i$ (Newton's second law); $T=\tfrac12\sum_i m_i\dot{\mathbf{r}}_i^2$ — the total kinetic energy, entering doubled because $\mathbf{p}_i\cdot\dot{\mathbf{r}}_i = m_i\dot{\mathbf{r}}_i^2$ is twice each particle's kinetic energy. The combination $\sum_i\mathbf{F}_i\cdot\mathbf{r}_i$ is Clausius's virial of the forces (up to the conventional factor $-\tfrac12$). The same statement reads more elegantly in the language of this chapter. $G$ is a phase-space function, so its evolution is a Poisson bracket with the Hamiltonian — the universal law from the article on time evolution of phase-space functions. For $H = \sum_i p_i^2/2m_i + V(q)$ a two-line computation gives

where: $\{G,H\}$ — the Poisson bracket of the generator with the Hamiltonian; $\mathbf{r}_i\cdot\partial V/\partial\mathbf{r}_i$ — the radial stretch of the potential, equal to $-\mathbf{F}_i\cdot\mathbf{r}_i$ for conservative forces, so (3) and (2) agree. Read it structurally: $G$ generates the dilation $\mathbf{r}\to\lambda\mathbf{r}$, $\mathbf{p}\to\mathbf{p}/\lambda$, and $\{G,H\}$ measures how $H$ responds to that rescaling — the kinetic term (which scales as $\lambda^{-2}$, hence the $2T$) pitted against the potential term (which contributes its radial stretch). Unlike the conserved quantities of the conservation-laws article, $G$ does not commute with $H$; the theorem instead exploits the weaker, subtler fact that its bracket averages to zero. Averages need a definition before they can be manipulated. The relevant one is the infinite-time average along the actual trajectory — for periodic motion, one period already gives the exact answer:

where: $f$ — any dynamical quantity evaluated along the motion; $\tau$ — the averaging window. For motion of period $T_p$, the integral over each period repeats, so $\langle f\rangle$ equals the one-period average exactly — no limit required. Now apply (4) to $\dot G$ itself. A time average of a total derivative telescopes: the integral of $\dot G$ is just $G$ at the endpoints, and boundedness pins the numerator while the denominator grows without limit,

where: $G(\tau)-G(0)$ — the net change of the generator over the window, bounded by $2G_{\max}$ whenever positions and momenta stay finite. This is the only place boundedness enters, and it is the step that fails for unbound motion: on a hyperbolic orbit $\dot G \to 2E > 0$ and the average never dies.

Averaging equation (2) term by term and feeding in (5) yields the theorem in its most general dress — a statement about forces, valid whether or not they come from a potential:

where: $\langle T\rangle$ — the time-averaged total kinetic energy; the right side is twice Clausius's virial $-\tfrac12\langle\sum_i\mathbf{F}_i\cdot\mathbf{r}_i\rangle$. Note what is not assumed: no periodicity, no integrability, no two-body structure — chaotic triple systems and $10^{3}$-galaxy clusters qualify just as well as a Kepler ellipse. When the forces derive from a potential that is a homogeneous function of the coordinates — $V(\lambda\mathbf{r}_1,\dots) = \lambda^n V(\mathbf{r}_1,\dots)$, which covers every power law $V\propto r^n$ — Euler's theorem converts the virial into a multiple of $V$ itself, and the theorem becomes an energy-partition rule:

where: $n$ — the degree of homogeneity; $\langle V\rangle$ — the time-averaged potential energy. One number fixes the entire average energy budget: $n=2$ splits it evenly, $n=-1$ binds it, and the crossover through $n=0$ marks potentials (like $\log r$) whose average kinetic energy is fixed independently of orbit size. Combining (7) with energy conservation $E = \langle T\rangle + \langle V\rangle$ eliminates either average in favour of the conserved total — the form you reach for in astrophysical estimates:

where: $\langle E\rangle = E$ — the conserved total energy. For gravity ($n=-1$): $E = -\langle T\rangle = \tfrac12\langle V\rangle$, so a bound system's kinetic energy is minus its total energy — lose energy and the system heats up, the negative heat capacity that drives stellar evolution.

Which form of the theorem answers which question

Nothing in derivation A referred to a single particle or a single orbit, and this is where the Hamiltonian-chapter version of the theorem earns its keep. For a statistically stationary many-body system — a relaxed star cluster, a gas in equilibrium — the time average along one long trajectory can be traded for an ensemble average over phase space. In the canonical ensemble the exchange becomes a theorem of its own: integrating by parts against the Boltzmann weight $e^{-H/k_BT}$ gives the generalized equipartition theorem,

where: $x_k$ — any phase-space coordinate, position or momentum; $\delta_{kl}$ — the Kronecker delta; $T_{\text{temp}}$ — the absolute temperature. Choosing $x_k = x_l = p$ gives $\langle p^2/m\rangle = k_BT_{\text{temp}}$ — each momentum degree of freedom carries $\tfrac12 k_BT_{\text{temp}}$, the ensemble shadow of the $2T$ in (2); choosing $x_k = x_l = q$ recovers the potential-side virial. The virial theorem and equipartition are one bookkeeping read in two ensembles. The gravitational case carries the most dramatic consequence. For a self-gravitating system in virial equilibrium, equation (8) with $n=-1$ says $E = -\langle T\rangle$: the heat capacity is negative. Radiate energy away ($E$ decreases) and $\langle T\rangle$ must rise — the system contracts and heats up. A star that loses heat gets hotter, which is why gravitational contraction ignites nuclear burning; a globular cluster whose core loses stars to evaporation contracts and heats toward the gravothermal catastrophe. No laboratory gas behaves this way, and the difference is traceable to a single sign: $n=-1$ instead of walls. The theorem also crosses into quantum mechanics essentially intact. For a stationary state (an energy eigenstate), Ehrenfest's theorem applied to the operator $\hat G = \tfrac12\sum_i(\hat{\mathbf{r}}_i\cdot\hat{\mathbf{p}}_i + \hat{\mathbf{p}}_i\cdot\hat{\mathbf{r}}_i)$ gives $2\langle\hat T\rangle = \langle\mathbf{r}\cdot\nabla V\rangle$ — expectation values in place of time averages, with stationarity of the state playing the role that boundedness played classically. For hydrogen it yields $\langle\hat T\rangle = -E_n$ exactly, a stringent test applied routinely to variational wavefunctions.

For a homogeneous potential $V\propto r^{n}$ the theorem reads $2\langle T\rangle=n\langle V\rangle$, and the physically important exponents deserve a table of their own:

1 · Checking the lock on a harmonic oscillator (confidence builder)

A block of mass $m = 2.0\ \mathrm{kg}$ oscillates on a spring with angular frequency $\omega = 3.0\ \mathrm{rad/s}$ and amplitude $A = 0.5\ \mathrm{m}$. Compute $\langle T\rangle$ and $\langle V\rangle$ over one period explicitly and verify the virial prediction.

Step 1. Write the motion and the two energies. With $x(t)=A\cos\omega t$: $T(t) = \tfrac12 m\omega^2A^2\sin^2\omega t$ and $V(t) = \tfrac12 m\omega^2A^2\cos^2\omega t$. Their sum is constant, $E = \tfrac12 m\omega^2 A^2 = \tfrac12(2.0)(9.0)(0.25) = 2.25\ \mathrm{J}$.

Step 2. Average over one period. Since $\langle\sin^2\omega t\rangle = \langle\cos^2\omega t\rangle = \tfrac12$ over a full period, $\langle T\rangle = \tfrac14 m\omega^2A^2 = \tfrac14(2.0)(9.0)(0.25) = \boxed{1.125\ \mathrm{J}}$ and $\langle V\rangle = \boxed{1.125\ \mathrm{J}}$ — equal, as the theorem demands.

Step 3. Check against the virial prediction. The spring potential $V=\tfrac12 kx^2$ is homogeneous of degree $n=2$, so equation (7) says $2\langle T\rangle = 2\langle V\rangle$, i.e. $\langle T\rangle = \langle V\rangle$, and equation (8) says $E = 2\langle T\rangle = 2(1.125) = 2.25\ \mathrm{J}$. Both match the explicit averages.

Step 4. Note what would fail instantaneously. At $t=0$ the split is $T=0$, $V=2.25\ \mathrm{J}$ — nowhere near equality. The theorem constrains the averages, never the snapshot.

2 · Weighing a galaxy cluster from its velocity dispersion (exam level)

The most famous application of the theorem turns two observables — how fast the members move and how big the system is — into a total mass, including every gram that does not shine.

A rich cluster has line-of-sight velocity dispersion $\sigma \approx 1000\ \mathrm{km/s}$ and characteristic radius $R \approx 3\ \mathrm{Mpc}$. Estimate its virial mass and compare with the luminous mass, $\sim 2\times10^{13}\ M_\odot$ in stars plus $\sim 1.4\times10^{14}\ M_\odot$ in hot gas.

Step 1. Set up the gravitational virial. For a self-gravitating system in equilibrium, $2\langle T\rangle + \langle V\rangle = 0$ with $\langle T\rangle = \tfrac12 M\langle v^2\rangle$ and $\langle V\rangle = -\alpha\,GM^2/R$, where $\alpha$ is a structure constant of order unity encoding the mass profile. Solving for the mass: $M = \langle v^2\rangle R/(\alpha G)$.

Step 2. Get the order of magnitude first, ignoring all factors of order one: $M \sim \sigma^2 R/G$. With $\sigma^2 = (1.0\times10^{6}\ \mathrm{m/s})^2 = 1.0\times10^{12}\ \mathrm{m^2/s^2}$ and $R = 3\ \mathrm{Mpc} = 9.26\times10^{22}\ \mathrm{m}$:

Dividing by $M_\odot = 1.99\times10^{30}\ \mathrm{kg}$ gives $\boxed{M \sim 7\times10^{14}\ M_\odot}$.

Step 3. Restore the honest factors. The line-of-sight dispersion sees one of three velocity components, so for isotropic orbits $\langle v^2\rangle = 3\sigma^2$, pushing the estimate up to $\sim 2\times10^{15}\ M_\odot$; the profile constant $\alpha$ pulls it back down by a factor of a few. The standard estimator $M \approx 5\sigma^2 R_e/G$ (with $R_e$ a projected effective radius) packages both corrections. Honest statement: $M_{\mathrm{vir}} \approx 10^{15}\ M_\odot$ within a factor of 2, the uncertainty dominated by profile and anisotropy assumptions, not by the theorem.

Step 4. Confront the luminous mass. Stars plus hot gas total $\approx 1.6\times10^{14}\ M_\odot$ — about 16 percent of the virial mass. Some $84$ percent of the gravitating matter is dark. This is Zwicky's 1933 Coma argument with modern numbers, and its logic has never been overturned — only confirmed, by X-ray hydrostatic masses and by gravitational lensing.

Step 5. Check the stationarity hypothesis. The crossing time is $R/\sigma \approx 9.26\times10^{22}/10^{6}\ \mathrm{s} \approx 3\ \mathrm{Gyr}$ — comfortably shorter than the $\sim10\ \mathrm{Gyr}$ age of a rich cluster, so the central regions have had time to relax and the virial average is meaningful. For a merging cluster this box fails, and virial masses are quoted with correspondingly fat error bars.

References. Goldstein, Poole & Safko, Classical Mechanics, 3rd ed., §3.4 · Landau & Lifshitz, Mechanics, 3rd ed., §10 · Morin, Introduction to Classical Mechanics, Ch. 5 · Taylor, Classical Mechanics, §13 · Binney & Tremaine, Galactic Dynamics, 2nd ed., §4.8 (tensor virial and astrophysical applications)

Provenance. Derivations follow the standard treatments in Goldstein, Poole & Safko Classical Mechanics 3rd ed. §3.4 (general theorem, power-law case) and Landau & Lifshitz Mechanics §10 (homogeneous potentials, energy partition). A verifier should: (1) reconfirm $\dot G=2T+\sum\mathbf F\cdot\mathbf r$ by the product rule; (2) check Euler's theorem step $\mathbf r\cdot\nabla V=nV$ for $V\propto r^n$; (3) verify the harmonic ($\langle T\rangle=\langle V\rangle$) and Kepler ($E=-T$) special cases; (4) confirm the ideal-gas derivation via $\nabla\cdot\mathbf r=3$ and the divergence theorem giving $2\langle T\rangle=3PV$. The dark-matter (Zwicky 1933) and ideal-gas (Clausius 1870) historical attributions are textbook-standard.

Builds on: The Hamiltonian Function · Hamiltonian as Total Energy · Generalized Momenta · Phase Space. Leads to: Hamiltonian Flow and the Phase Fluid · Phase Portraits and Trajectories · The Harmonic Oscillator in Phase Space. See also the companion article Virial Theorem for Central Forces for the orbit-by-orbit story, and The Kepler Problem for the $n=-1$ dynamics itself.