Landau Damping

In 1946, Lev Landau published a result so counterintuitive that parts of the plasma physics community spent nearly two decades refusing to believe it: a wave in a perfectly collisionless plasma — no friction, no collisions, no resistivity, nothing in the equations that looks like dissipation — still dies. Not through any dissipative term, but through a conspiracy between the wave and the handful of particles that happen to surf it at almost exactly its phase velocity. The prediction sat unverified until 1964, when Malmberg and Wharton finally measured it on a plasma column in San Diego and found Landau's exponent, cleanly, on a bench. You arrive equipped: the Plasma Frequency and Dielectric Response article built $\omega_p$ , the Debye length, and the fluid dielectric function $\varepsilon(\omega)=1-\omega_p^2/\omega^2$ — a function with no imaginary part and therefore no damping, ever. This article is about what that fluid description structurally cannot see: keep track of which particles move at which velocities and a new term appears, proportional to the slope of the velocity distribution at the wave's phase velocity. The mathematics that extracts it — Landau's analytic continuation of an initial-value problem, with an integration contour that must be dented below a pole — is one of the most celebrated calculations in theoretical physics, and we do it honestly here. Mastering it buys you the working currency of kinetic plasma theory: why fusion machines drive current with waves, why beams of electrons make plasmas sing, and what "collisionless damping" actually means — a question subtle enough that its rigorous nonlinear answer earned a Fields Medal in 2010.

Key takeaways

A Langmuir wave in a collisionless plasma decays because particles moving near its phase velocity $v_\phi=\omega/k$ exchange energy with it resonantly — no collisions required.
The linearized Vlasov–Poisson system gives a dielectric function with a singular velocity integral; treating the problem as an initial-value (Laplace) problem forces the integration contour to pass below the pole at $v=v_\phi$ — the Landau contour.
The pole's residue makes $\varepsilon$ complex: $\gamma\propto f_0'(v_\phi)$ . A Maxwellian has negative slope everywhere, so every Langmuir wave damps; a beam-generated positive slope reverses the sign into growth (bump-on-tail instability).
The damping has a cliff: for $k\lambda_D=0.1$ the rate is $\gamma/\omega_p\sim10^{-20}$ (utterly negligible); by $k\lambda_D=0.5$ it is $\gamma/\omega_p\approx0.15$ and the wave dies in about one period.
Landau damping is not dissipation: linear theory conserves entropy, the field energy hides in fine phase-space structure, and the plasma echo experiment recovers it — the damping is reversible phase mixing.
KEY IDEA: the fate of a plasma wave is written in the slope of the velocity distribution at its phase velocity — read off by the residue of the causally-deformed Landau contour.

The core insight

The pole at $v=\omega/k$ in the linearized Vlasov dispersion integral must be handled with the causal (Laplace) contour; its imaginary contribution — the Landau residue $i\pi f_0'(\omega/k)$ — encodes resonant wave–particle energy exchange and damps (or grows) the wave at a rate proportional to the distribution's slope at the phase velocity.

Where this lives today

Landau damping is a working tool wherever waves and plasmas share a room. In tokamaks it is the point: lower-hybrid current drive launches waves whose phase velocity is deliberately parked on the fast tail of the electron distribution ( $v_\phi\approx3$ – $4\,v_{th}$ ), so that Landau absorption transfers wave momentum to superthermal electrons and drives the toroidal current that keeps the machine running — gigajoules of plasma current sustained by exactly the resonance integral derived above. In inertial-confinement fusion the same physics appears as a menace: stimulated Raman scattering pumps laser energy into Langmuir waves, and how much of it returns as preheating hot electrons depends on where those waves sit on the damping cliff. In space physics, spacecraft from Voyager to Parker Solar Probe measure Langmuir waves and suprathermal tails whose coexistence is regulated by Landau resonance. And the mathematics has a life of its own: Mouhot and Villani's 2011 proof that Landau damping survives in the full nonlinear Vlasov equation — for small enough, smooth enough perturbations — resolved a sixty-year controversy about whether the linear result was an artifact, and contributed to Villani's Fields Medal. Lynden-Bell's "violent relaxation" applies the identical mechanism to gravity: galaxies, too, Landau-damp.

Origins · 1 minLev Landau and how this was discovered

Lev Landau

1908–1968 · the 1946 calculation that taught physics how to damp without dissipating

Landau wrote "On the electron plasma oscillations" in 1946, freshly returned to work after his 1938–39 imprisonment during the purges (Kapitza's letter to Molotov secured his release). The paper is a masterclass in taking a mathematical subtlety physically seriously: where Vlasov had computed the natural-looking principal value and found undamped modes, Landau insisted the problem was an initial-value problem, ran the Laplace transform honestly, and found the contour — and the damping — that now bear his name. He was famously scornful about it ("Vlasov's dispersion equation... is wrong"), and famously right, though experimental confirmation (Malmberg–Wharton, 1964–66) and the echo demonstration of reversibility (1968) both came only after his 1962 car accident had ended his working life. The damping band here is distinct from the Landau–Lifshitz radiation-reaction band met in the E9 radiation chapter — same Landau, different immortality. His Course of Theoretical Physics, written with Lifshitz, remains the standard against which theoretical physics prose is measured.

History

1938

— Vlasov formulates the collisionless kinetic equation for plasmas; his principal-value treatment of the wave problem finds no damping

1946

— Landau solves the initial-value problem via Laplace transform: the causal contour yields collisionless damping

1949

— Bohm and Gross derive the thermal dispersion correction and the physical wave-particle resonance picture

1964

— Malmberg and Wharton measure spatial Landau damping of Langmuir waves on a cesium plasma column, confirming the exponential rate

1968

— Malmberg, Wharton, Gould and O'Neil observe plasma wave echoes: the damping is reversible phase mixing, not dissipation

2011

— Mouhot and Villani prove nonlinear Landau damping, closing the mathematical case

The physical picture

Picture the wave as a set of moving potential troughs sliding through the plasma at $v_\phi=\omega/k$ , and the electrons as surfers spread over every speed. A surfer moving slightly slower than the wave sees a trough slowly overtake him, gets caught on its front face, and is pushed forward: he gains energy, and the wave pays. A surfer slightly faster runs up the back of a trough and is held back: he loses energy, and the wave collects. Individually these exchanges are symmetric. Statistically they are not — the outcome depends on whether slow surfers outnumber fast ones in the narrow resonant band around $v_\phi$ , and that head count is exactly the slope $f_0'(v_\phi)$ .

Figure

The surfer census — resonant exchange is decided by the slope of the distribution at the phase velocity

On a Maxwellian the slope is negative at every velocity: slower particles always outnumber faster ones in the resonant band, the census always favors the gainers, and every Langmuir wave pays a tax to its surfers. That is Landau damping. Run the logic backwards and you get something just as important: shoot a beam through the plasma so that $f_0$ has a rising stretch, park a wave's phase velocity on that stretch, and the census flips — the wave collects energy and grows exponentially. Damping and instability are one mechanism read in two directions. The second face of the phenomenon is subtler. Where does the wave's energy "go" if nothing dissipates? Into structure. The perturbed distribution carries a factor $e^{-ikvt}$ : each velocity class oscillates at its own pace, so the velocity-space ripples wind up finer and finer as time passes — like cream stirred into coffee, streaked but never truly mixed. The electric field is a velocity integral of this ever-finer corrugation, and integrals over rapidly oscillating functions fade. The field dies; the information survives, hidden in phase-space filaments. This is why Landau damping is reversible in principle — and, astonishingly, in practice: the plasma echo experiment "unstirs" the coffee. How strong is the tax? That depends on where the phase velocity sits on the distribution, which the wavelength controls through $k\lambda_D$ . The damping has a genuine cliff:

Figure

The damping cliff — twenty orders of magnitude between a wavelength of 60 Debye lengths and one of 13

At $k\lambda_D=0.1$ the phase velocity sits $10$ thermal speeds out on the tail, where essentially no surfers live: the wave e-folds only after $6\times10^{18}$ oscillation periods — for practical purposes undamped. At $k\lambda_D=0.5$ the phase velocity has come down to $2.6$ thermal speeds, the resonant band is populated, and the wave dies in $1.4$ periods. Langmuir waves therefore exist as recognizable modes only in the long-wavelength window, and anything with $k\lambda_D\gtrsim0.5$ is less a wave than a brief shudder. This cliff — twenty orders of magnitude across a factor of five in wavelength — is the sharpest regime boundary in this track.

The math: the initial-value problem and its pole

The fluid theory of the Plasma Frequency and Dielectric Response article treated the electrons as a single cold stream and got $\varepsilon(\omega)=1-\omega_p^2/\omega^2$ : real for real $\omega$ , hence zero damping, forever. The kinetic upgrade replaces the fluid density by the full phase-space distribution $f(x,v,t)$ , evolved collisionlessly by the Vlasov equation and coupled to Poisson's equation. For one-dimensional electrostatic oscillations of electrons (charge $-e$ ) against a fixed neutralizing ion background $n_0$ :

\frac{\partial f}{\partial t}+v\,\frac{\partial f}{\partial x}-\frac{e}{m}E\,\frac{\partial f}{\partial v}=0, \qquad \frac{\partial E}{\partial x}=-\frac{e}{\varepsilon_0}\int\!\big(f-n_0\hat f_0\big)\,\mathrm{d}v ,

where: $f(x,v,t)$ is the electron phase-space density, $E$ the self-consistent electric field, and $\hat f_0$ the equilibrium velocity distribution normalized to $\int\hat f_0\,\mathrm{d}v=1$ . Read physically: the left equation says phase-space density is carried unchanged along particle orbits (Liouville for a collisionless gas); the right says charge separation sources the field that bends those very orbits. The nonlinearity — the field acting back on $f$ — is what we now linearize away, and it is precisely the term whose eventual revenge (particle trapping) sets the boundary of everything derived below.

Derivation AFrom the Vlasov equation to the causal dispersion relation

The goal is the kinetic dielectric function $\varepsilon(k,\omega)$ whose zeros are the plasma's electrostatic modes, and — just as important — the rule for evaluating it, because the defining integral is singular at $v=\omega/k$ . The strategy is Landau's: linearize, transform, and insist throughout that we are solving an initial-value problem (perturb at $t=0$ , watch what happens after), because that physical insistence is what resolves the mathematical ambiguity at the pole. The thing to watch is the moment the Laplace transform's convergence condition turns into a statement about which side of the pole the velocity contour must pass.

Step 1. Linearize about the uniform equilibrium. Write $f=n_0\hat f_0(v)+f_1(x,v,t)$ with $|f_1|\ll n_0\hat f_0$ and $E$ first-order (the equilibrium has no field). Dropping the second-order product $E\,\partial_v f_1$ ,

\frac{\partial f_1}{\partial t}+v\,\frac{\partial f_1}{\partial x}=\frac{e}{m}\,n_0\,\hat f_0'(v)\,E .

Why linearization is honest here: wave fields in the linear regime displace electrons by far less than a wavelength per period, so orbit distortion is a small correction — until the resonant particles, for whom the wave is stationary, accumulate displacement without bound. Linear theory is a finite-time theory for them; we return to this at the trapping boundary.

Step 2. Transform. Fourier in $x$ ( $f_1\propto e^{ikx}$ ) and Laplace in $t$ — this is the crux, and we keep its bookkeeping explicit. The Laplace transform $\tilde f_1(p)=\int_0^\infty f_1 e^{-pt}\mathrm{d}t$ converges only for $\operatorname{Re}p$ large enough, i.e. writing $p=-i\omega$ , only for $\operatorname{Im}\omega$ sufficiently large and positive. Every formula below is therefore born in the upper half of the complex $\omega$ -plane, where the response to an initial kick is still growing transients, and must be analytically continued downward to reach the physical, decaying answers. Algebraically the transform turns derivatives into multipliers exactly as Fourier does (initial-condition terms drive the response but do not alter the dispersion function):

(-i\omega+ikv)\,f_1=\frac{e}{m}\,n_0\hat f_0'(v)\,E \quad\Longrightarrow\quad f_1=\frac{e\,n_0}{m}\,\frac{\hat f_0'(v)\,E}{ik\,(v-\omega/k)} .

Why we refuse to just "set $\omega$ real": for real $\omega$ the denominator vanishes at $v=\omega/k$ and the theory is silent about what the integral through the singularity means. For $\operatorname{Im}\omega>0$ the pole sits safely off the real axis and every integral is unambiguous. The physics (causality: response comes after the kick) has quietly chosen a side.

Step 3. Close the loop with Poisson. The perturbed density is $n_1=\int f_1\,\mathrm{d}v$ , and $ikE=-(e/\varepsilon_0)n_1$ . Substituting $f_1$ and cancelling the common factor $E$ :

1-\frac{e^2 n_0}{\varepsilon_0 m k^2}\int\frac{\hat f_0'(v)}{v-\omega/k}\,\mathrm{d}v=0 \qquad\Longrightarrow\qquad \varepsilon(k,\omega)=1-\frac{\omega_p^2}{k^2}\int_C\frac{\hat f_0'(v)}{v-\omega/k}\,\mathrm{d}v=0 ,

where: $\omega_p^2=n_0e^2/\varepsilon_0 m$ and $C$ is, for now, simply the real axis — legitimate because the pole $\omega/k$ still floats above it. This is the kinetic dielectric function: the fluid $-\omega_p^2/\omega^2$ has become a weighted census of the whole distribution, with the weight blowing up at the resonant velocity.

Step 4. Continue analytically — the Landau contour. The dispersion relation's roots have $\operatorname{Im}\omega<0$ (damped waves), so we must evaluate $\varepsilon$ where the transform integral no longer converges; the function survives by analytic continuation, and continuation demands that $\varepsilon(k,\omega)$ change smoothly as $\omega$ moves. As $\operatorname{Im}\omega\to0^+$ and then negative, the pole $v=\omega/k$ descends onto and then through the real $v$ -axis. The only way the integral can vary smoothly is for the contour to get out of the way: $C$ must dent downward, passing below the pole, and stay below it when the pole crosses into the lower half-plane. That deformed path is the Landau contour.

Step 5. Extract the pole's contribution. For $\omega$ real (the marginal case: contour indented by a semicircle under the pole), the Plemelj decomposition splits the integral into a principal value plus half a residue:

\int_C\frac{\hat f_0'(v)}{v-\omega/k}\,\mathrm{d}v =\mathrm{P}\!\!\int\frac{\hat f_0'(v)}{v-\omega/k}\,\mathrm{d}v\;+\;i\pi\,\hat f_0'\!\left(\frac{\omega}{k}\right) .

The principal value is the reactive, fluid-like part; the $i\pi\hat f_0'(v_\phi)$ term is new, imaginary, and entirely kinetic — the resonant particles' line in the ledger. Everything called Landau damping lives in this term. Verified: in the cold limit $\hat f_0\to\delta(v)$ the integral gives $\int\delta'(v)/(v-\omega/k)\,\mathrm{d}v=k^2/\omega^2$ and $\varepsilon\to1-\omega_p^2/\omega^2$ — the fluid dielectric of the Plasma Frequency article, recovered exactly. (The AllQFT base states this dispersion relation with a plus sign, which fails this very check; the sign above is the one consistent with its own intermediate steps and with Krall–Trivelpiece.)

The contour deformation is the entire intellectual content of the calculation, so it deserves its own picture. Vlasov himself, in 1938, treated the same integral as a principal value — a symmetric, agnostic choice that knows nothing about time's arrow — and concluded that plasma waves neither damp nor grow. Landau's 1946 paper is, at bottom, one observation: perturb at $t=0$ and ask what happens later is not symmetric in time, the Laplace transform encodes that asymmetry as "start in the upper half $\omega$ -plane," and analyticity then dictates the contour with no further freedom:

Figure

The Landau contour — causality starts the pole upstairs; analytic continuation drags the contour under it

With the evaluation rule fixed, extracting the actual frequencies is bookkeeping: split $\varepsilon$ into real and imaginary parts, find where the real part vanishes (the oscillation frequency), and let the small imaginary part shift the root off the real axis (the damping rate).

Go deeper · Derivation B · 2 minBohm-Gross frequency and the Maxwellian damping rate

Derivation BBohm-Gross frequency and the Maxwellian damping rate

The goal is the two numbers an experimenter measures for a Langmuir wave of given $k$ : its real frequency $\omega_r$ and its damping rate $\gamma$ , for a Maxwellian equilibrium. The strategy is standard perturbation theory on the dispersion relation — expand the principal-value integral in moments for the fast-wave regime $v_\phi\gg v_{th}$ , then evaluate the residue term at the resulting phase velocity. The step to watch is the last: the exponential in $\gamma$ is exquisitely sensitive to $v_\phi$ , so the Bohm–Gross correction to $\omega_r$ cannot be dropped there even though it is small — it is the origin of the famous $-3/2$ in the exponent.

Step 1. Real part by moments. Integrate the principal value by parts, $\mathrm{P}\!\int\hat f_0'/(v-v_\phi)\,\mathrm{d}v=\mathrm{P}\!\int\hat f_0/(v-v_\phi)^2\,\mathrm{d}v$ , and expand for $v\ll v_\phi$ : $(v-v_\phi)^{-2}=v_\phi^{-2}(1+2v/v_\phi+3v^2/v_\phi^2+\cdots)$ . Odd moments of the Maxwellian vanish and $\langle v^2\rangle=v_{th}^2=k_BT_e/m$ :

\operatorname{Re}\varepsilon=1-\frac{\omega_p^2}{\omega^2}\left(1+\frac{3k^2v_{th}^2}{\omega^2}\right)=0 \qquad\Longrightarrow\qquad \omega_r^2\simeq\omega_p^2\left(1+3k^2\lambda_D^2\right),

where $\lambda_D=v_{th}/\omega_p$ . This is the Bohm–Gross dispersion relation: plasma oscillations acquire a thermal stiffness, propagating with group velocity $v_g\simeq3k\lambda_D^2\,\omega_p\,(\omega_p/\omega_r)$ instead of standing still like the cold mode.

Step 2. Imaginary part from the residue. With the Plemelj term, $\operatorname{Im}\varepsilon=-\pi(\omega_p^2/k^2)\hat f_0'(v_\phi)$ . For a root $\omega=\omega_r+i\gamma$ with $|\gamma|\ll\omega_r$ , Taylor-expand $\varepsilon(\omega_r+i\gamma)\approx\varepsilon(\omega_r)+i\gamma\,\partial_\omega\operatorname{Re}\varepsilon$ and demand it vanish:

\gamma=-\frac{\operatorname{Im}\varepsilon(\omega_r)}{\partial_\omega\operatorname{Re}\varepsilon(\omega_r)} =\frac{\pi}{2}\frac{\omega_p^3}{k^2}\,\hat f_0'\!\left(\frac{\omega_r}{k}\right)\!,

using $\partial_\omega\operatorname{Re}\varepsilon\simeq2\omega_p^2/\omega_r^3\simeq2/\omega_p$ to leading order. The sign statement is now a theorem: $\gamma$ has the sign of the slope at the phase velocity. Negative slope, damping; positive slope, growth; zero slope, marginal.

Step 3. Insert the Maxwellian. With $\hat f_0(v)=(2\pi v_{th}^2)^{-1/2}e^{-v^2/2v_{th}^2}$ , the slope is $\hat f_0'(v_\phi)=-(v_\phi/v_{th}^2)\hat f_0(v_\phi)$ , and $v_\phi^2/2v_{th}^2=(1+3k^2\lambda_D^2)/2k^2\lambda_D^2=1/(2k^2\lambda_D^2)+3/2$ :

\boxed{\ \gamma=-\sqrt{\frac{\pi}{8}}\ \frac{\omega_p}{(k\lambda_D)^3}\ \exp\!\left[-\frac{1}{2k^2\lambda_D^2}-\frac{3}{2}\right]\ }

Why the $-3/2$ matters: it is the Bohm–Gross correction evaluated inside the exponential — dropping it overestimates the damping by a factor $e^{3/2}\approx4.5$ at every $k$ . A "small" correction to $\omega_r$ is a factor-of-several correction to $\gamma$ ; exponential sensitivity is the rule whenever a rate reads a distribution's tail.

Step 4. Read the formula. The prefactor grows as $(k\lambda_D)^{-3}$ but the exponential crushes everything at small $k\lambda_D$ : numerically $\gamma/\omega_p=2.7\times10^{-20}$ at $k\lambda_D=0.1$ , $6.5\times10^{-5}$ at $0.2$ , $2.0\times10^{-2}$ at $0.3$ , and $0.15$ at $0.5$ — the cliff plotted in the physical picture. Beyond $k\lambda_D\approx0.5$ the expansion $|\gamma|\ll\omega_r$ that produced the formula is failing precisely because damping is becoming order unity; there the exact root of $\varepsilon(k,\omega)=0$ , written with the plasma dispersion function $Z(\zeta)$ of Fried and Conte, must be found numerically, and the "wave" is dead within a period anyway. Verified: the formula and all four quoted numbers were recomputed directly from the boxed expression; the rate matches Chen eq. 7-129 and Nicholson Ch. 6, and the $k\lambda_D=0.5$ value agrees with the exact kinetic root to within the accuracy the weak-damping expansion promises.

The dispersion machinery says that the wave damps; it is silent on who got the energy. The census argument from the physical picture can be made quantitative with nothing more than momentum-conservation bookkeeping on the resonant particles, and doing so closes the logical loop — the same slope $\hat f_0'(v_\phi)$ appears from a completely independent direction.

Go deeper · Derivation C · 2 minThe surfer theorem — energy bookkeeping at the resonance

Derivation CThe surfer theorem — energy bookkeeping at the resonance

The goal is to derive, not assert, the statement that the wave's energy loss rate is proportional to $-\hat f_0'(v_\phi)$ , by following the resonant particles themselves. The strategy is to work in the wave frame, where the wave is a stationary corrugated potential and resonant dynamics is transparent, and to track the kinetic-energy change of particles on each side of $v_\phi$ over a time short compared with the trapping period. The honesty to watch: individual particles both gain and lose; only the census is decisive, and it must emerge from the algebra.

Step 1. Sit in the wave frame. The wave is a stationary potential $\Phi(x')=\Phi_0\cos kx'$ ; a particle with lab velocity $v$ has wave-frame velocity $u=v-v_\phi$ . Particles with $|u|$ small compared with the trapping speed $u_T=\sqrt{2e\Phi_0/m}$ are trapped, oscillating in the troughs with bounce frequency $\omega_B=k\,u_T/\sqrt{2}=\sqrt{ek^2\Phi_0/m}$ ; faster ones pass over the crests, modulated but free.

Step 2. Exchange in the wave frame conserves speed, reverses direction. A trapped particle that entered moving forward ( $u>0$ , lab-fast) is turned around by the potential after half a bounce: its wave-frame velocity maps $u\to-u$ . Back in the lab frame its kinetic energy changes by

\Delta W=\tfrac{m}{2}\big[(v_\phi-u)^2-(v_\phi+u)^2\big]=-2m\,v_\phi\,u .

A lab-fast particle ( $u>0$ ) loses $2mv_\phi u$ ; a lab-slow one ( $u<0$ ) gains the same. The exchange is antisymmetric in $u$ — exactly the individual-surfer symmetry of the picture.

Step 3. Take the census. Over a half-bounce time, particles within the resonant band $|u|\lesssim u_T$ have swapped sides. The net energy given to particles is the antisymmetric exchange weighted by the population difference across $v_\phi$ . Expanding $\hat f_0(v_\phi+u)\simeq\hat f_0(v_\phi)+u\,\hat f_0'(v_\phi)$ , the symmetric term cancels between the two signs of $u$ and:

\frac{\mathrm{d}W_{\text{particles}}}{\mathrm{d}t}\ \propto\ -\,v_\phi\!\int_{-u_T}^{u_T}\!u\cdot u\,\hat f_0'(v_\phi)\,\mathrm{d}u\ \propto\ -\,\hat f_0'(v_\phi)\, ,

positive for a falling distribution. Particles gain, so the wave — whose energy is the only other account — loses: $\gamma<0$ when $\hat f_0'(v_\phi)<0$ , with growth for an inverted slope, in sign-by-sign agreement with the residue calculation.

Step 4. Read off the boundary of linear theory. The argument used "half a bounce" as its clock: linear Landau damping is valid only for times short compared with $\omega_B^{-1}\propto\Phi_0^{-1/2}$ . For larger amplitudes or longer times the trapped particles complete full orbits, the resonant region of $\hat f_0$ flattens into a plateau ( $\hat f_0'(v_\phi)\to0$ ), and the damping switches itself off — O'Neil's nonlinear saturation, with the wave amplitude oscillating at $\omega_B$ as energy sloshes between field and trapped particles. Verified: the sign and the $\hat f_0'(v_\phi)$ proportionality reproduce the kinetic-theory result; the trapping-time criterion $\gamma\gg\omega_B$ for linear validity matches O'Neil (1965).

Limiting cases

case	result	meaning
Cold plasma, $\hat f_0\to\delta(v)$	$\varepsilon=1-\omega_p^2/\omega^2$ , $\gamma=0$	The fluid dielectric of the Plasma Frequency article is the zero-temperature shadow of the kinetic one; no spread, no resonant particles, no damping.
$k\lambda_D\ll1$	$\gamma\sim-\omega_p(k\lambda_D)^{-3}e^{-1/2k^2\lambda_D^2}\to0$	Phase velocity far out on the tail: essentially undamped Langmuir waves — the regime where the wave concept is clean.
$k\lambda_D\gtrsim0.5$	$\|\gamma\|\sim\omega_r$	Phase velocity near thermal: the mode is obliterated within a period; beyond this the weak-damping expansion itself fails.
$\hat f_0'(v_\phi)=0$ (plateau)	$\gamma=0$	Gainers and losers balance exactly — the fixed point that quasilinear flattening drives the system toward.
$\hat f_0'(v_\phi)>0$ (beam bump)	$\gamma>0$ : growth	Inverse Landau damping — the bump-on-tail instability; the census now favors the wave.
$T_e\gg T_i$ (ion-acoustic branch)	weakly damped sound	$v_s=\sqrt{k_BT_e/m_i}$ then sits many ion-thermal speeds above the ion resonance; for $T_e\sim T_i$ it sits on top of it and ion Landau damping kills the mode.

Common trapLandau damping is not dissipation — and the principal value is not a shortcut

Two frauds, one per clause. First: nothing is dissipated. Linear Landau damping conserves the plasma's entropy exactly; the field's energy is transferred reversibly to resonant particles and to ever-finer corrugations of $f_1$ in velocity space, where it hides rather than heats. Call it damping of the field, not of the plasma — the echo experiment (below) retrieves it, which no genuinely dissipative process would permit. Second: you cannot get the physics by declaring the singular integral a principal value and moving on. The principal value is the time-symmetric choice — it is exactly Vlasov's 1938 calculation, and it predicts undamped waves. The $i\pi$ residue is not a mathematical garnish; it is where causality enters the theory, and dropping it is not an approximation but a different (and wrong) physical problem. If your dispersion relation came out real, you didn't simplify — you erased the arrow of time.

Worked examples

1 · The cliff, on a bench-top plasma (confidence builder)

A laboratory plasma column has $n_e=10^{18}\,\mathrm{m^{-3}}$ and $T_e=10\,\mathrm{eV}$ . Where is the Langmuir-wave window?

Step 1. The scales. $\omega_p=\sqrt{n_ee^2/\varepsilon_0m}=5.64\times10^{10}\,\mathrm{s^{-1}}$ ( $f_p=8.98\,\mathrm{GHz}$ ), $v_{th}=\sqrt{k_BT_e/m}=1.33\times10^{6}\,\mathrm{m/s}$ , so $\lambda_D=v_{th}/\omega_p=23.5\,\mathrm{\mu m}$ .

Step 2. A long wave: $k\lambda_D=0.1$ , i.e. wavelength $\lambda=2\pi\lambda_D/0.1=1.48\,\mathrm{mm}$ . The formula gives $\gamma=2.7\times10^{-20}\,\omega_p=1.5\times10^{-9}\,\mathrm{s^{-1}}$ — an e-folding time of $6.6\times10^{8}$ seconds, twenty-one years. The phase velocity is $v_\phi=10.2\,v_{th}$ : no surfers live there. For any practical purpose the wave is undamped, and collisions (neglected throughout) will in reality kill it first.

Step 3. A short wave: $k\lambda_D=0.5$ , wavelength $0.30\,\mathrm{mm}$ . Now $\omega_r=1.32\,\omega_p$ and $\gamma=0.15\,\omega_p=8.5\times10^{9}\,\mathrm{s^{-1}}$ : the e-folding time is $0.12\,\mathrm{ns}$ , about $1.4$ oscillation periods. Same plasma, factor of five in wavelength, nineteen orders of magnitude in damping rate. The usable Langmuir spectrum of this plasma is confined to wavelengths above roughly a third of a millimetre.

2 · Designing the Malmberg–Wharton measurement (exam level)

You want to measure Landau damping spatially: launch a wave from a probe into a quiescent column and watch its amplitude decay along the axis, à la Malmberg and Wharton (1964–66). Column: $n_e=5\times10^{14}\,\mathrm{m^{-3}}$ , $T_e=9.5\,\mathrm{eV}$ . Where should you operate so the decay length is a convenient few centimetres?

Step 1. Scales again. $\omega_p=1.26\times10^{9}\,\mathrm{s^{-1}}$ ( $f_p=201\,\mathrm{MHz}$ ), $v_{th}=1.29\times10^{6}\,\mathrm{m/s}$ , $\lambda_D=1.03\,\mathrm{mm}$ .

Step 2. Convert temporal to spatial damping. A steadily driven wave decays in space with $k_i=\gamma/v_g$ , where the Bohm–Gross group velocity is $v_g=\mathrm{d}\omega/\mathrm{d}k=3kv_{th}^2/\omega_r$ . The decay length is $L=v_g/|\gamma|$ .

Step 3. Scan the operating point. At $k\lambda_D=0.25$ ( $\lambda=25.8\,\mathrm{mm}$ ): $\gamma/\omega_r=2.8\times10^{-3}$ , $v_g=8.9\times10^{5}\,\mathrm{m/s}$ , $L=23\,\mathrm{cm}$ — decay too gentle for a half-metre column. At $k\lambda_D=0.35$ ( $\lambda=18.4\,\mathrm{mm}$ ): $\gamma/\omega_r=4.7\times10^{-2}$ and $L=1.7\,\mathrm{cm}$ — dead too fast to trace cleanly. The sweet spot is near $k\lambda_D\approx0.30$ , wavelength $\lambda\approx21\,\mathrm{mm}$ , drive frequency $f\approx f_p\sqrt{1+3(0.3)^2}\approx227\,\mathrm{MHz}$ , giving $L\approx4\,\mathrm{cm}$ — several clean e-foldings along the column, with wavelength and decay length both comfortably measurable. This exponential-in- $1/k^2\lambda_D^2$ tunability is precisely how the original experiment could sweep the damping across its measurable window and confirm Landau's exponent, eighteen years after the prediction.

Going further

References. Jackson, Classical Electrodynamics, 3rd ed., §7.6 (waves in dispersive media) and plasma sections · Zangwill, Modern Electrodynamics, Ch. 22 (plasma physics) · Landau, J. Phys. USSR 10, 25 (1946) — original paper · Krall & Trivelpiece, Principles of Plasma Physics, Ch. 8 · Nicholson, Introduction to Plasma Theory, Ch. 6 · Chen, Introduction to Plasma Physics and Controlled Fusion, Ch. 7.

Provenance. A verifier should re-derive the Landau dispersion relation by linearizing Vlasov-Poisson and confirm the boxed dispersion integral with the causal contour (Jackson §7; Krall-Trivelpiece Ch. 8). Check the Bohm-Gross relation $\omega_r^2=\omega_p^2(1+3k^2\lambda_D^2)$ and the Maxwellian damping formula $\gamma=-\sqrt{\pi/8}\,\omega_p (k\lambda_D)^{-3}\exp[-1/(2k^2\lambda_D^2)-3/2]$ against Chen (Introduction to Plasma Physics, eq. 7-129) and Nicholson Ch. 6. Verify the Plemelj formula and the plasma dispersion function $Z(\zeta)$ (Fried & Conte). No fabricated numerical constants are used; all symbolic results are standard textbook values. Lane addendum: the base's dispersion-relation card carried a sign inconsistent with its own derivation steps (its cold-plasma limit gives $1+\omega_p^2/\omega^2$ ); corrected here and verified against the cold limit and Krall–Trivelpiece. All damping-rate numbers, the worked examples, and the cliff figure were recomputed from the boxed formula.

Prerequisites. Vlasov-Poisson equation · Plasma oscillations and the plasma frequency (see Plasma Frequency and Dielectric Response, this chapter) · Fourier and Laplace transforms · Contour integration and residues · Maxwell-Boltzmann distribution.

Leads to. Bump-on-tail and two-stream instabilities · Quasilinear theory · Plasma wave echoes · Synchrotron Radiation.

Why did Vlasov's principal-value calculation give no damping when the same integral, evaluated on the Landau contour, does?

The principal value is symmetric under time reversal: it weighs approach from above and below the pole equally, which is the mathematical statement of "no arrow of time." Landau's contour breaks that symmetry deliberately, because the physical question — perturb now, observe later — is not time-symmetric. The Laplace transform encodes the asymmetry by defining everything first in the upper half $\omega$ -plane, and analytic continuation then adds the residue $i\pi\hat f_0'(v_\phi)$ that the principal value discards. Vlasov solved a legitimate mathematical problem; it just wasn't the one nature poses.

Where exactly does the wave's energy go if the plasma has no collisions and gains no entropy?

Two ledgers. Part of it becomes ordered kinetic energy of the resonant particles — the surfers genuinely speed up, which is real, recoverable energy. The rest hides in the phase-space corrugation of $f_1\propto e^{-ikvt}$ : as velocity classes dephase, the field (a velocity integral of $f_1$ ) decays even though $f_1$ itself never shrinks. Nothing is thermalized in linear theory. Collisions, however rare, eventually smooth the fine filaments — that is the moment the hidden energy becomes heat and entropy actually rises. Landau damping delays the dissipative verdict; it does not issue it.

Is the plasma echo really evidence, or just a theoretical curiosity?

It is the decisive experiment. Launch a wave at frequency $\omega_1$ , let it Landau-damp away completely, then launch a second at $\omega_2$ at time $\tau$ later. At time $\approx\omega_2\tau/(\omega_2-\omega_1)$ a third wave appears spontaneously — the echo — built from the interference of the two fossil corrugations in velocity space as they transiently rephase. Malmberg, Wharton, Gould and O'Neil saw it in 1968, exactly as predicted. A dissipated wave cannot echo; a phase-mixed one can. This is as close as plasma physics gets to Loschmidt's reversibility demon on a lab bench.

The damping rate formula has the distribution's slope at one single velocity. Does the rest of the distribution matter at all?

It matters through the real part: the whole distribution shapes $\mathrm{P}\!\int\hat f_0'/(v-v_\phi)\mathrm{d}v$ and hence $\omega_r$ , which in turn sets where $v_\phi$ lands — so the tail is read at a location the body chose. But the imaginary part is genuinely local: only particles within the narrow resonant band exchange net energy with the wave, which is why a tiny beam-induced bump, invisible in the density or pressure, can flip a huge plasma from absorbing to unstable. Kinetic theory's whole point is that moments are not enough.

How does Landau damping switch off nonlinearly, and what does the saturated state look like?

Linear theory treats resonant orbits as unperturbed straight lines, valid only for times short compared with the bounce period of particles trapped in the wave's troughs. Once trapping completes ( $t\gtrsim\omega_B^{-1}$ , $\omega_B=\sqrt{ek^2\Phi_0/m}$ ), the trapped population phase-mixes within the trough, flattening $\hat f_0$ across the resonant band into a plateau. With $\hat f_0'(v_\phi)\to0$ the damping self-extinguishes, and the amplitude rings at $\omega_B$ around a finite level (O'Neil 1965). The criterion: linear Landau damping requires $|\gamma|\gg\omega_B$ , i.e. small enough amplitude.

Why must the electron temperature greatly exceed the ion temperature for ion-acoustic waves to exist?

The ion-acoustic speed is $v_s\simeq\sqrt{k_BT_e/m_i}$ , set by electron pressure and ion inertia. Compare it with the ion thermal speed $v_{th,i}=\sqrt{k_BT_i/m_i}$ : their ratio is $\sqrt{T_e/T_i}$ . If $T_e\sim T_i$ the wave's phase velocity sits right on the ion thermal bulk — maximal ion Landau damping, and the mode is overdamped into nonexistence. For $T_e\gtrsim10\,T_i$ the phase velocity sits $\gtrsim3$ ion-thermal speeds out on the ion tail, few ions surf, and the plasma carries sound. Same cliff, different species.

Is there a version of this where the wave grows instead of damps, and how is it stopped?

Yes — run the census backwards. Any distribution with $\hat f_0'(v_\phi)>0$ at some accessible phase velocity (an electron beam through a plasma is the classic source) makes $\gamma>0$ for waves resonant on the rising slope: the bump-on-tail instability, the kinetic sibling of the two-stream instability. Growth saturates by the same physics that saturates damping: the growing wave traps and scatters the resonant particles, quasilinear diffusion flattens the bump into a plateau, and the free energy ends up as a broadened wave spectrum plus a heated tail. The marginal state, in both directions, is the plateau $\hat f_0'(v_\phi)=0$ .

Landau's calculation continues epsilon analytically. Are there formulations that avoid complex analysis entirely?

Yes — van Kampen (1955) constructed the exact real-frequency eigenmodes of the linearized Vlasov equation, which are singular continuum modes; Landau damping then emerges as the destructive interference (phase mixing) of a continuum superposition rather than as a complex root, and the two pictures are mathematically equivalent. The Laplace route is shorter for computing rates; the van Kampen route makes the reversibility and the echo transparent. Case completed the equivalence proof in 1959 — worth one sentence here and a lifetime elsewhere.

Landau Damping

Where this lives today

The physical picture

The math: the initial-value problem and its pole

Limiting cases

Worked examples

1 · The cliff, on a bench-top plasma (confidence builder)

2 · Designing the Malmberg–Wharton measurement (exam level)

Going further

Where this leads