Dynamical Friction

A cannonball fired through air slows because it slams into molecules; a massive body gliding through a sea of stars slows for a subtler reason — it never touches anything. As it moves, its gravity deflects the stars it passes, bending their paths so that they crowd into a trailing overdensity, a gravitational wake strung out behind it. That wake is a lump of extra mass sitting permanently downstream, and its gravity pulls backward, opposing the very motion that created it. The body is, quite literally, dragged down by the gravity of the density enhancement it raises in its own path. This is dynamical friction: a purely gravitational, collisionless drag arising from the collective response of a many-body system, not from any physical collision. Subrahmanyan Chandrasekhar put the effect on a firm footing in 1943 by summing the momentum kicks from independent two-body encounters between the body of mass $M$ and a background of field stars of mass $m$ , modelled as an infinite homogeneous sea with an isotropic Maxwellian velocity distribution. The result — proportional to $M^2$ , to the background density $\rho$ , and modulated by the fraction of stars slower than the body — carries a Coulomb logarithm $\ln\Lambda$ identical in form to the one in plasma physics and the two-body relaxation time. Crude as its assumptions sound, the formula governs how globular clusters and satellite galaxies spiral inward, how massive black holes sink to galactic centres after mergers, and how heavy stars segregate to cluster cores. It is the discreteness effect that the smooth collisionless Boltzmann equation, by construction, cannot see.

Key takeaways

Dynamical friction is the gravitational drag on a massive body of mass $M$ moving through a background of lighter stars: the body's gravity focuses the stars into a trailing wake whose pull decelerates it.
Chandrasekhar's formula gives the deceleration $\mathrm{d}\mathbf{v}_M/\mathrm{d}t=-4\pi G^2 M\rho\,\ln\Lambda\,[\mathrm{erf}(X)-2Xe^{-X^2}/\sqrt{\pi}]\,\mathbf{v}_M/v_M^3$ with $X=v_M/(\sqrt{2}\sigma)$ .
The drag scales as $M^2$ and, for fast bodies, as $v_M^{-2}$ ; in the isotropic case only stars slower than the body contribute, so friction pushes the system toward energy equipartition.
The Coulomb logarithm $\ln\Lambda=\ln(b_{\max}/b_{\min})$ encodes the range of impact parameters; galactic values run $\ln\Lambda\sim3$ – $30$ .
The drag is strongest not for the fastest bodies but for those moving at roughly the dispersion speed, $v_M\approx1.4\sigma$ , and falls away in both the slow and fast limits.
Key idea: a moving mass polarises the background into a trailing wake whose gravity pulls it back; the force scales as $M^2\rho/v^2$ for fast bodies and vanishes for the slowest, so friction drives massive bodies to sink and systems toward equilibrium.

The core insight

A moving mass polarises the stellar background into a trailing gravitational wake whose pull opposes the motion; the drag scales as $M^2\rho/v^2$ for fast bodies, depends only on stars slower than the body, and so drives massive objects to sink toward the centre and the system toward energy equipartition.

Where this lives today

Dynamical friction drives some of the most active questions in galaxy formation and gravitational-wave astronomy. In the hierarchical assembly of galaxies, friction drags accreted dwarfs and dark subhaloes toward the centres of their hosts, so semi-analytic models encode a Chandrasekhar-based sinking time — with an effective $\ln\Lambda$ calibrated against N-body simulations — to decide which satellites merge and which survive. In the gravitational-wave era it governs the earliest stage of forming a massive black-hole binary: after two galaxies merge, friction against the stars brings their central black holes together efficiently down to about a parsec, at which point the supply of stars to scatter runs out and the pair stalls — the "final parsec problem" that stellar hardening, gas torques, or a third black hole must solve before the pair can radiate gravitational waves for LISA and pulsar-timing arrays. Closer to home, friction sets the radial mass segregation of globular clusters, where heavy stars and stellar-mass black holes sink to the core, and it explains the deficit of low-mass, tightly bound clusters in the inner Galaxy that should long since have spiralled in and dissolved. It even shapes debates about the Fornax dwarf's clusters, whose failure to sink is used to constrain the central dark-matter profile.

The physical picture

Imagine $M$ ploughing to the right through a static sea of stars. Each star it approaches is pulled toward its track; having swung past, the star's deflected orbit carries it behind the body, where it lingers. Gravitational focusing repeats this for every star along the path, and the accumulated deflections build a permanent overdensity trailing the body — the wake. Because the wake sits behind $M$ , its gravity has a component pointing backward along the motion. Friction is nothing more than $M$ feeling the pull of the mass concentration its own passage produces. Nothing collides; the drag is entirely the collective gravity of rearranged orbits.

Figure

A massive body M ploughs rightward through a sea of field stars; gravitational focusing piles up a trailing overdensity — the wake — whose backward pull is the friction force, scaling as F ∝ M²ρ/v² for a fast body

Two features make dynamical friction different from ordinary drag. First, only stars moving slower than the body get caught in the wake; faster stars overtake it and stream away, contributing nothing in the isotropic case. This is why friction drives a system toward equipartition — a heavy body keeps losing energy until it moves no faster than a typical field star. Second, the drag is not monotonic in speed. A very fast body sees an effectively cold, static background and its drag falls as $v_M^{-2}$ , like a charge Rutherford-scattering off a plasma; a very slow body feels a gentle linear restoring drag, like Stokes friction in a warm fluid. In between, the drag is maximal, peaking when the body moves at about $1.4$ times the one-dimensional dispersion. The next figure plots the exact velocity dependence Chandrasekhar's formula predicts.

Figure

The dimensionless velocity factor G(X) = erf(X) − 2X·e^(−X²)/√π climbs from 0 to 1 as X = v_M/(√2 σ) grows, reaching 0.428 at X = 1; because the deceleration goes as G(X)/v² the drag peaks near v_M ≈ 1.4 σ and then falls off as 1/v² for fast bodies

The $M^2$ scaling is the last decisive feature. The wake's mass grows in proportion to $M$ , because a deeper potential well gathers more stars, and the force that wake exerts back grows with $M$ again — so friction is savagely more effective for heavy bodies. A black hole or a satellite galaxy spirals in over a few billion years, while an individual star, for which $M=m$ , feels essentially nothing.

The core equations

Chandrasekhar's master formula gives the deceleration of a body of mass $M$ and velocity $\mathbf{v}_M$ moving through an infinite homogeneous background of density $\rho$ and isotropic Maxwellian dispersion $\sigma$ . The drag points antiparallel to the motion, with the entire velocity dependence packed into the bracket.

\frac{\mathrm{d}\mathbf{v}_M}{\mathrm{d}t}=-\,4\pi G^2 M\,\rho\,\ln\Lambda\left[\mathrm{erf}(X)-\frac{2X}{\sqrt{\pi}}e^{-X^2}\right]\frac{\mathbf{v}_M}{v_M^3}\qquad(1)

where: $M$ — mass of the perturbing body; $\mathbf{v}_M,\,v_M$ — its velocity and speed; $\rho$ — mass density of the field stars; $\ln\Lambda$ — Coulomb logarithm; $X=v_M/(\sqrt{2}\sigma)$ — speed in units of the dispersion; $\sigma$ — one-dimensional velocity dispersion; $G$ — Newton's constant. Reading it: the deceleration grows with the background density and with the body's mass, but the bracket suppresses it for both very slow bodies (bracket $\to0$ ) and very fast ones (bracket $\to1$ while $v_M^{-2}$ shrinks it). Multiplying the deceleration by $M$ gives the drag force, which exposes the characteristic $M^2$ dependence that separates black holes from stars.

\mathbf{F}_{\rm df}=-\,4\pi G^2 M^2\,\rho\,\ln\Lambda\left[\mathrm{erf}(X)-\frac{2X}{\sqrt{\pi}}e^{-X^2}\right]\frac{\mathbf{v}_M}{v_M^3}\qquad(2)

where: $\mathbf{F}_{\rm df}$ — dynamical friction force on the body; all other symbols as above. Reading it: one factor of $M$ comes from the wake's mass, the other from the force that wake exerts, so friction is negligible on a single star and dominant on a massive clump. The bracket is a dimensionless function of $X$ alone; naming it $G(X)$ makes the velocity structure transparent and is the quantity plotted above.

G(X)=\mathrm{erf}(X)-\frac{2X}{\sqrt{\pi}}e^{-X^2},\qquad X=\frac{v_M}{\sqrt{2}\,\sigma}\qquad(3)

where: $\mathrm{erf}$ — the error function; $G(X)$ — the fraction of the background mass moving slower than $v_M$ , in the isotropic Maxwellian case. Reading it: $G$ rises from $0$ at $X=0$ to $1$ as $X\to\infty$ , equals $0.428$ at $X=1$ , and — divided by $v_M^2$ — makes the drag peak near $v_M\approx1.4\sigma$ . The logarithm collects a divergent integral over impact parameters $b$ into a single number set by the largest and smallest relevant encounter distances.

\ln\Lambda=\ln\!\left(\frac{b_{\max}}{b_{\min}}\right),\qquad b_{\min}\simeq\frac{G(M+m)}{v_{\rm typ}^2}\qquad(4)

where: $b_{\max}$ — largest impact parameter (system size or the scale over which the background stays uniform); $b_{\min}$ — impact parameter for a large-angle deflection, or the body's own size if extended; $v_{\rm typ}\sim\sqrt{v_M^2+\sigma^2}$ — typical relative speed; $m$ — field-star mass. Reading it: only the ratio of scales matters, so $\ln\Lambda$ is a slowly varying number of order a few to a few tens, not a precise constant. For a satellite on a circular orbit in an isothermal galaxy, integrating the angular-momentum loss gives the time to spiral to the centre — the practical output of the whole theory.

t_{\rm df}\simeq\frac{1.17}{\ln\Lambda}\,\frac{r_i^2\,v_c}{G M}\qquad(5)

where: $t_{\rm df}$ — sinking time; $r_i$ — initial orbital radius; $v_c$ — circular (and roughly constant) velocity of the isothermal halo. Reading it: the sinking time falls as $1/M$ , so a satellite ten times more massive sinks ten times faster; light bodies effectively never arrive. Finally, the background that all of this integrates over is the isotropic Maxwellian distribution function, the single assumption from which Chandrasekhar's bracket follows.

f(\mathbf{v}_a)=\frac{\rho}{(2\pi\sigma^2)^{3/2}}\,e^{-v_a^2/2\sigma^2},\qquad \rho=\int f(\mathbf{v}_a)\,\mathrm{d}^3v_a\qquad(6)

where: $f$ — field-star velocity distribution; $\mathbf{v}_a,\,v_a$ — a field star's velocity and speed; $\rho$ — total mass density. Reading it: the Gaussian in velocity is what turns the beam calculation into the $\mathrm{erf}(X)$ form; change the distribution and the bracket changes with it.

Derivation AChandrasekhar's formula from summing two-body encounters

The goal is to build the master formula from the ground up: treat the background as an infinite homogeneous sea, compute the momentum a single hyperbolic encounter removes from the body along its direction of motion, sum over all impact parameters to get the drag from a monochromatic beam of field stars, and finally average that beam result over the isotropic Maxwellian distribution. The transverse kicks cancel by symmetry, the impact-parameter integral supplies the Coulomb logarithm, and the velocity average supplies the $\mathrm{erf}(X)$ bracket, so that three familiar pieces assemble into equation (1).

Step 1. Count encounters. Field stars of number density $n$ stream past the body at relative speed $v$ ; the rate of encounters with impact parameter between $b$ and $b+\mathrm{d}b$ is $n\,v\,2\pi b\,\mathrm{d}b$ , the flux through an annulus of area $2\pi b\,\mathrm{d}b$ . Why: friction is the cumulative effect of a continuous rain of independent gravitational scatterings, so we must weight each impact parameter by how often it occurs.

Step 2. Take one encounter. A hyperbolic two-body scattering deflects the relative velocity through an angle $\psi$ with $\tan(\psi/2)=G(M+m)/(bv^2)$ ; the body's velocity change antiparallel to its motion, after averaging over the encounter, is

\delta v_\parallel=\frac{2mv}{M+m}\sin^2\!\frac{\psi}{2}=\frac{2mv}{M+m}\left[1+\frac{b^2v^4}{G^2(M+m)^2}\right]^{-1}.

Why: only the component of the kick opposite the motion survives the sum — the perpendicular kicks average to zero over the azimuth — so this is the piece that becomes drag.

Step 3. Integrate over impact parameter. Summing $\delta v_\parallel$ over the encounter rate gives an integral $\int b\,\mathrm{d}b/(1+b^2v^4/G^2(M+m)^2)$ that grows logarithmically:

\int_0^{b_{\max}}\frac{2\pi b\,\mathrm{d}b}{1+b^2v^4/G^2(M+m)^2}=\pi\frac{G^2(M+m)^2}{v^4}\ln\!\left(1+\Lambda^2\right)\approx2\pi\frac{G^2(M+m)^2}{v^4}\ln\Lambda.

Why: the kick falls only as $b^{-2}$ at large $b$ , so distant gentle encounters matter collectively as much as close ones — the divergence, cut off at $b_{\max}$ , is the origin of the Coulomb logarithm, with $\Lambda=b_{\max}v^2/[G(M+m)]$ .

Step 4. Assemble the beam result. Putting the rate and the integrated kick together, a monochromatic beam of field stars of velocity $\mathbf{v}_a$ decelerates the body as

\frac{\mathrm{d}\mathbf{v}_M}{\mathrm{d}t}=-4\pi G^2(M+m)\,m\,n\,\ln\Lambda\,\frac{\mathbf{v}_M-\mathbf{v}_a}{|\mathbf{v}_M-\mathbf{v}_a|^3}.

With $m\ll M$ we replace $(M+m)\,m\,n\to M\rho$ , where $\rho=mn$ . Why: this is Chandrasekhar's drag from a single stream; the real background is a superposition of such streams, so the final step is to add them up.

Step 5. Average over the Maxwellian. Integrating the beam result against $f(\mathbf{v}_a)$ , the angular part behaves like a shell theorem in velocity space: contributions from stars faster than the body cancel, leaving only those with $v_a<v_M$ ,

\int f(\mathbf{v}_a)\frac{\mathbf{v}_M-\mathbf{v}_a}{|\mathbf{v}_M-\mathbf{v}_a|^3}\,\mathrm{d}^3v_a=4\pi\!\left[\int_0^{v_M}v_a^2 f\,\mathrm{d}v_a\right]\frac{\mathbf{v}_M}{v_M^3}.

Why: this is precisely where the "only slower stars matter" rule comes from — it is the gravitational analogue of Newton's shell theorem applied to the distribution of velocities.

Step 6. Do the Gaussian integral. For the Maxwellian, $4\pi\int_0^{v_M}v_a^2 f\,\mathrm{d}v_a=\rho\,G(X)=\rho\,[\mathrm{erf}(X)-2Xe^{-X^2}/\sqrt{\pi}]$ , and substituting reproduces equation (1) exactly. Why: the error function is just the enclosed velocity-space mass of a Gaussian, so the bracket is literally the fraction of stars slower than the body — the shell theorem made quantitative.

Go deeper · Derivation B · 1 minSinking time of a satellite in an isothermal halo

Derivation BSinking time of a satellite in an isothermal halo

The goal is to turn the instantaneous drag into a total lifetime: how long a satellite of mass $M$ on a near-circular orbit takes to spiral from radius $r_i$ to the centre of a singular isothermal halo. We fix the background with $\rho(r)=v_c^2/(4\pi G r^2)$ and $\sigma=v_c/\sqrt{2}$ , evaluate the friction torque, and integrate the angular-momentum loss, arriving at the standard result $t_{\rm df}=1.17\,r_i^2 v_c/(\ln\Lambda\,GM)$ used to argue that massive satellites sink within a Hubble time while light ones survive indefinitely.

Step 1. Fix the background. In a singular isothermal sphere the circular velocity $v_c$ is independent of radius, so a body on a circular orbit has $v_M=v_c$ and, with $\sigma=v_c/\sqrt{2}$ , exactly $X=1$ . The bracket is then $G(1)=\mathrm{erf}(1)-2e^{-1}/\sqrt{\pi}=0.428$ . Why: the isothermal sphere is the one profile where $X$ is constant along the whole inspiral, making the integral tractable in closed form.

Step 2. Evaluate the drag. Using $\rho=v_c^2/(4\pi G r^2)$ in equation (2) at $X=1$ ,

F_{\rm df}=-0.428\,\frac{4\pi\ln\Lambda\,G^2 M^2\rho}{v_c^2}=-0.428\,\frac{\ln\Lambda\,G M^2}{r^2}.

Why: the explicit $r^{-2}$ density conveniently cancels the $v_c^2$ , leaving a drag that depends on radius only through $r^{-2}$ — a clean torque to integrate.

Step 3. Write the angular-momentum equation. The orbital angular momentum is $L=M v_c r$ , and the friction torque removes it at rate $r F_{\rm df}$ :

\frac{\mathrm{d}}{\mathrm{d}t}\big(M v_c r\big)=r F_{\rm df}=-0.428\,\frac{\ln\Lambda\,G M^2}{r}.

Since $v_c$ is constant, $M v_c\,\mathrm{d}r/\mathrm{d}t=-0.428\,\ln\Lambda\,GM^2/r$ . Why: on the assumption that the orbit stays nearly circular, tracking angular momentum is equivalent to tracking the radius, and the torque drives a slow inward drift.

Step 4. Integrate to the centre. Separating and integrating $r\,\mathrm{d}r$ from $r_i$ to $0$ ,

\int_{r_i}^{0}r\,\mathrm{d}r=-0.428\,\frac{\ln\Lambda\,GM}{v_c}\,t_{\rm df}\;\Rightarrow\; t_{\rm df}=\frac{r_i^2 v_c}{2\times0.428\,\ln\Lambda\,GM}=\frac{1.17}{\ln\Lambda}\frac{r_i^2 v_c}{GM}.

Why: the coefficient $1/(2\times0.428)=1.17$ is the standard Binney and Tremaine value, and the $r_i^2/M$ dependence is the practical punchline — sinking time grows with the square of the starting radius and falls inversely with mass.

Limiting cases

The bracket $G(X)$ and the mass scaling control every regime of interest; the table collects the forms one actually uses (isotropic Maxwellian background, $X=v_M/\sqrt{2}\sigma$ ).

Case	Result	Meaning
Fast body $v_M\gg\sigma$	$\dfrac{\mathrm{d}\mathbf{v}_M}{\mathrm{d}t}\to-\dfrac{4\pi G^2 M\rho\ln\Lambda}{v_M^2}\hat{\mathbf{v}}_M$	$G\to1$ ; drag $\propto v_M^{-2}$ , the Rutherford form for a cold static background.
Slow body $v_M\ll\sigma$	$\dfrac{\mathrm{d}\mathbf{v}_M}{\mathrm{d}t}\to-\dfrac{4}{3}\sqrt{2\pi}\,\dfrac{G^2 M\rho\ln\Lambda}{\sigma^3}\mathbf{v}_M$	$G\approx4X^3/3\sqrt{\pi}$ ; linear Stokes-like drag driving equipartition.
Peak drag	$v_M\approx1.4\sigma$ $(X\approx0.97)$	$G(X)/v_M^2$ is maximal; friction is strongest at roughly the dispersion speed, not for the fastest bodies.
Mass scaling	$\dfrac{\mathrm{d}v_M}{\mathrm{d}t}\propto M,\quad t_{\rm df}\propto M^{-1}$	Heavier bodies decelerate faster and sink sooner; individual stars barely respond.
Extended perturber	$b_{\min}\sim\max\!\big(GM/v_M^2,\,r_{\rm half}\big)$	A finite size raises $b_{\min}$ and lowers $\ln\Lambda$ relative to a point mass.

Common trapTreating ln Λ as a universal constant and the velocity factor as always active

Two traps trip people up. First, $\ln\Lambda$ is not a fixed number: it is $\ln(b_{\max}/b_{\min})$ and must be re-estimated for each system from its actual largest and smallest impact parameters — a point mass, an extended cluster, and a satellite embedded in a live halo all take different values, and getting it wrong rescales every sinking time. Second, the bracket $\mathrm{erf}(X)-2Xe^{-X^2}/\sqrt{\pi}$ is derived for an infinite, homogeneous, isotropic Maxwellian background; do not apply it blindly to anisotropic, rotating, or strongly inhomogeneous systems. In a rotating disc, for instance, the wake is deflected and can even push a body outward or sideways, and the "only slower stars matter" rule no longer holds as stated.

Worked examples

1 · A satellite galaxy sinking in the Milky Way halo (fully worked)

A satellite of mass $M=2\times10^{10}\,M_\odot$ is on a circular orbit at $r_i=50\ \mathrm{kpc}$ in an isothermal Milky Way halo with $v_c=220\ \mathrm{km\,s^{-1}}$ . Estimate the Coulomb logarithm and the time to sink to the centre. Does it arrive within a Hubble time?

Step 1. Estimate $\ln\Lambda$ . Take $b_{\max}=r_i=50\ \mathrm{kpc}$ and $b_{\min}=GM/v_c^2$ . With $G=4.301\times10^{-6}\ \mathrm{kpc\,(km/s)^2}\,M_\odot^{-1}$ , $b_{\min}=(4.301\times10^{-6})(2\times10^{10})/220^2=86020/48400=\mathbf{1.78\ kpc}$ . Then $\Lambda=50/1.78=28.1$ and $\ln\Lambda=\mathbf{3.34}$ .

Step 2. Form the mass-and-radius factor. $\dfrac{r_i^2 v_c}{GM}=\dfrac{(50)^2(220)}{(4.301\times10^{-6})(2\times10^{10})}=\dfrac{5.5\times10^{5}}{8.60\times10^{4}}=6.39\ \mathrm{kpc/(km/s)}$ . Converting with $1\ \mathrm{kpc/(km/s)}=0.978\ \mathrm{Gyr}$ gives $6.25\ \mathrm{Gyr}$ .

Step 3. Apply the sinking-time formula. $t_{\rm df}=\dfrac{1.17}{\ln\Lambda}\dfrac{r_i^2 v_c}{GM}=\dfrac{1.17}{3.34}\times6.25\ \mathrm{Gyr}=0.350\times6.25\ \mathrm{Gyr}=\mathbf{2.2\ Gyr}$ .

Step 4. Interpret. The satellite spirals to the centre in about $2.2$ billion years, comfortably inside the $13.8\ \mathrm{Gyr}$ age of the Universe. A $2\times10^{10}\,M_\odot$ satellite at $50\ \mathrm{kpc}$ sinks in $\sim2\ \mathrm{Gyr}$ — which is exactly why massive satellites like the progenitors of accreted dwarfs are dragged into the inner galaxy and disrupted, seeding stellar haloes.

2 · Why globular clusters survive and stars feel nothing

Compare the sinking time of a $10^7\,M_\odot$ globular cluster on a circular orbit at $r_i=8\ \mathrm{kpc}$ (same $v_c=220\ \mathrm{km\,s^{-1}}$ halo) with that of the satellite above, and then of a single $1\,M_\odot$ star at the same radius.

Step 1. Coulomb logarithm for the cluster. $b_{\min}=GM/v_c^2=(4.301\times10^{-6})(10^7)/48400=8.9\times10^{-4}\ \mathrm{kpc}=\mathbf{0.89\ pc}$ , and with $b_{\max}=8\ \mathrm{kpc}$ , $\Lambda=8/8.9\times10^{-4}=9000$ , so $\ln\Lambda=\mathbf{9.1}$ — larger than the satellite's because a compact cluster admits much closer encounters.

Step 2. Sinking time of the cluster. $\dfrac{r_i^2 v_c}{GM}=\dfrac{(8)^2(220)}{(4.301\times10^{-6})(10^7)}=\dfrac{14080}{43.0}=327\ \mathrm{kpc/(km/s)}=320\ \mathrm{Gyr}$ , so $t_{\rm df}=\dfrac{1.17}{9.1}\times320\ \mathrm{Gyr}=\mathbf{41\ Gyr}$ .

Step 3. Compare. $41\ \mathrm{Gyr}$ is three times the age of the Universe, so a $10^7\,M_\odot$ cluster at $8\ \mathrm{kpc}$ does not sink — unlike the far more massive satellite of Example 1. The $t_{\rm df}\propto1/M$ scaling does the work: dropping from $2\times10^{10}$ to $10^7\,M_\odot$ lengthens the time roughly two thousandfold.

Step 4. Push to a single star. For $M=1\,M_\odot$ the same scaling multiplies $t_{\rm df}$ by another $10^7$ , giving $\sim4\times10^{8}\ \mathrm{Gyr}$ . An individual star is, for all practical purposes, frictionless. This is the observational face of the $M^2$ law: friction reshapes galaxies through their massive substructures while leaving ordinary stellar orbits untouched.

Physics of the portrait

Dynamical friction is inseparable from the man who derived it. Chandrasekhar wrote it down in 1943 as one strand of a vast programme to understand how the graininess of real star systems — the fact that they are made of a finite number of point masses, not a smooth fluid — drives their slow evolution.

Origins · 1 minSubrahmanyan Chandrasekhar and how this was discovered

Subrahmanyan Chandrasekhar

1910–1995 · stellar dynamics and the theory of gravitational encounters

Subrahmanyan Chandrasekhar was among the most systematic theorists of twentieth-century astrophysics, a scholar who would exhaust an entire subject and then move on. Born in Lahore and educated in Madras and Cambridge, he is best remembered for the Chandrasekhar limit — the maximum mass of a white dwarf, found on the boat to England in 1930 — for which he shared the 1983 Nobel Prize. But around 1943 his attention turned to stellar dynamics, and in the classic Astrophysical Journal paper (ApJ 97, 255) he derived the dynamical friction formula by summing independent two-body gravitational encounters. Working with John von Neumann, he also treated the statistics of the fluctuating gravitational field produced by a random distribution of stars — the noise whose slow rectification is friction. The result was initially controversial: many doubted that a smooth stellar system could exhibit friction at all, since there are no collisions. Chandrasekhar's insight was that the body polarises the background into a wake and feels its own reflected gravity, an idea later shown to be equivalent to the linear-response and fluctuation-dissipation treatments. His rigour — starting from an idealised infinite medium yet extracting an answer that survives in messy real galaxies — made the formula the durable foundation of galactic dynamics it remains today.

History

1943

— Chandrasekhar derives the dynamical friction formula from two-body encounters

1975

— Tremaine, Ostriker and Spitzer apply friction to sinking globular clusters and galactic nuclei

1976

— White models galaxy merging in clusters driven by dynamical friction

1987

— Binney and Tremaine's Galactic Dynamics standardises the modern treatment

2005

— Cosmological N-body simulations calibrate the effective Coulomb logarithm for subhalo decay

Going further

Texts. Binney & Tremaine, Galactic Dynamics, 2nd ed., §8.1 (eqs. 8.3–8.7 and the sinking time 8.24) — the definitive treatment · Chandrasekhar, S. (1943), "Dynamical Friction I", ApJ 97, 255 — the original derivation · Carroll & Ostlie, An Introduction to Modern Astrophysics, 2nd ed., stellar-dynamics chapters · Mo, van den Bosch & White, Galaxy Formation and Evolution, §12 (satellite dynamics and merging) · Dodelson, Modern Cosmology (structure-formation context).

Provenance. Equation (1) and the bracket $G(X)$ follow Binney & Tremaine eqs. 8.6–8.7; the sinking time (5) is their eq. 8.24. A verifier should recompute: (i) the impact-parameter integral giving $\ln\Lambda$ ; (ii) the velocity-space average yielding $\mathrm{erf}(X)-2Xe^{-X^2}/\sqrt{\pi}$ ; (iii) the value $G(1)=\mathrm{erf}(1)-2e^{-1}/\sqrt{\pi}=0.428$ and the coefficient $1/(2\times0.428)=1.17$ ; (iv) the slow-limit coefficient $\tfrac{4}{3}\sqrt{2\pi}=3.34$ (note: this corrects a mis-simplified factor in the migrated source); (v) the drag maximum at $X\approx0.97$ , i.e. $v_M\approx1.4\sigma$ . Worked-example figures use $G=4.301\times10^{-6}\ \mathrm{kpc\,(km/s)^2}M_\odot^{-1}$ and $1\ \mathrm{kpc/(km/s)}=0.978\ \mathrm{Gyr}$ ; the range $\ln\Lambda\sim3$ – $30$ is standard (BT §8.1).

Prerequisites. Gravitational Two-Body Problem · Two-Body Relaxation Time · Distribution Function and Phase Space · Gravitational N-Body Problem. Leads to. Virial Theorem in Stellar Systems · Jeans Equations · Collisionless Boltzmann Equation.

Ask the tutor: Want to see how the wake would deform in a rotating disc, or how the sinking time changes if you swap the isothermal halo for an NFW profile? Ask and we will work it through.

Is dynamical friction really friction if nothing ever collides?

It is friction in the sense that matters — a velocity-dependent force that removes kinetic energy and momentum from the body — but its mechanism is purely gravitational. The body deflects the stars it passes, and their rearranged orbits pile into an overdensity behind it whose gravity pulls the body back. No star ever touches the body, and no energy is lost to heat in the ordinary sense; instead the energy is transferred to the field stars, gently heating the background. The word "friction" is a strong analogy, not a claim about collisions: it is a collective, collisionless drag arising from the discreteness of the many-body gravitational field.

Why does only the wake behind the body matter, and not the stars ahead of it?

The stars ahead do exert a forward pull, but by symmetry it is exactly balanced in a static, uniform medium — until the body's motion breaks that symmetry. As the body moves, gravitational focusing sends the deflected stars streaming to its rear, where they accumulate faster than they can disperse. The forward and backward pulls no longer cancel: there is a permanent excess of mass behind, and its net backward pull is the drag. Equivalently, in velocity space, the stars slower than the body are the ones it can "catch" and drag along, and their reaction decelerates it; faster stars overtake and escape, contributing nothing on average. The asymmetry is entirely a consequence of the body's own motion.

Why does the force scale as M squared rather than linearly like ordinary drag?

Two factors of the mass enter. The first sets the size of the wake: a heavier body has a deeper potential well, so it gathers more field stars into its overdensity, and the wake mass grows in proportion to $M$ . The second sets the force that wake exerts back: gravity is proportional to the mass being pulled, so the backward force on the body grows with $M$ again. Multiplying the two gives $F_{\rm df}\propto M^2$ . Ordinary hydrodynamic drag scales with cross-sectional area and is roughly linear in size because it does not have this self-amplifying feedback — the object does not gravitationally build up the very medium that then pulls on it.

What value should I use for the Coulomb logarithm in practice?

Estimate it, do not memorise it. Write $\ln\Lambda=\ln(b_{\max}/b_{\min})$ : take $b_{\max}$ as the size of the system or the scale over which the background density is roughly uniform (often the orbital radius), and $b_{\min}$ as the larger of $GM/v^2$ (the deflection scale) and the physical radius of the perturber if it is extended. For a point-like satellite in a galaxy this typically gives $\ln\Lambda\sim3$ – $10$ ; for a compact object admitting very close encounters it can reach $\sim20$ – $30$ . A common refinement for a satellite in a live halo is $\ln\Lambda\approx\ln(1+M_{\rm host}/M_{\rm sat})$ . Because the result depends only logarithmically on the cutoffs, a factor-of-two uncertainty in either scale changes the sinking time by only tens of percent — but ignoring the variation entirely can be badly wrong.

Does dynamical friction ever speed a body up or heat the background?

Yes to both, and they are the same statement. The energy the sinking body loses does not vanish — it is deposited into the field stars, which are stirred to slightly higher velocities as they absorb the wake's momentum. Over many crossing times this is precisely the heating that drives a cluster toward equipartition and, in extreme cases, core collapse. As for speeding up: in a non-Maxwellian or rotating background the "only slower stars matter" rule breaks, and the wake can be deflected so its pull acquires a component along or across the motion. In a rotating disc, dynamical friction can even transfer angular momentum outward and, in some geometries, push a bar or satellite outward rather than in.

Why does a formula derived for an infinite uniform background work in real galaxies?

Because the drag is set almost entirely by local quantities and a logarithm. The dominant contributions come from encounters at impact parameters comparable to the local density scale, so what matters is the $\rho$ , $\sigma$ , and $v_M$ where the body currently sits — not the global structure. As long as these vary slowly over a wake length, the local homogeneous approximation is good, and the only genuinely non-local ingredient, $\ln\Lambda$ , is adjusted to the system's real scales. N-body simulations bear this out: with a sensibly chosen $\ln\Lambda$ , Chandrasekhar's formula reproduces measured orbital-decay rates to within tens of percent, which is why it remains the workhorse of semi-analytic galaxy modelling.