05.09.08 · symplectic / kam

Infinite-dimensional Poisson manifolds and Hamiltonian evolution equations

shipped3 tiersLean: none

Anchor (Master): Gardner 1971 *The Korteweg-de Vries equation as a Hamiltonian system* (J. Math. Phys. 12, originator); Zakharov-Faddeev 1971 *Korteweg-de Vries equation: a completely integrable Hamiltonian system* (Funct. Anal. Appl. 5); Magri 1978 *A simple model of the integrable Hamiltonian equation* (J. Math. Phys. 19, bi-Hamiltonian pair); Faddeev-Takhtajan 1987 *Hamiltonian Methods in the Theory of Solitons* Part I; Olver 1993 §7.1-7.2

Intuition Beginner

In ordinary mechanics a system is a few moving particles, and its state is a short list of numbers: positions and momenta. Hamiltonian mechanics packages the law of motion into a single energy function and a bracket that turns energy into motion. A wave on water is different. Its state is not a few numbers but a whole shape — the height of the surface at every point along the channel. The state is a function, and the space of all such shapes is endlessly many-dimensional.

The surprising discovery of the 1970s is that wave equations like the one Korteweg and de Vries wrote for shallow water are themselves Hamiltonian systems, run on this enormous space of shapes. There is an energy, now an integral over the whole channel rather than a sum over particles, and there is a bracket that turns that energy into the motion of the wave. The machinery of classical mechanics survives the jump to infinitely many degrees of freedom.

What changes is the bracket. With finitely many particles the bracket is a fixed antisymmetric table of numbers. With a wave it becomes an operator that differentiates and integrates the shape. Choosing that operator is choosing the geometry of the infinite-dimensional state space, and a good choice makes a hard nonlinear wave equation read like a tidy energy flow.

Visual Beginner

Picture the surface of a long narrow channel seen from the side. At each point $x$ along the channel the water has a height $u$ , so the state of the whole system is the entire profile — the curve traced by the surface, not any single number.

To the whole profile the energy assigns one number $H$ : add up a contribution from every point and you get the total energy of that shape. The bracket then acts like a steering rule. It takes the way the energy changes if you nudge the surface at each point — a new profile called the energy gradient — and feeds it through an operator that differentiates along the channel. The output is the rate at which the surface moves. Energy in, motion out: the same loop as a swinging pendulum, but now the swinging object is an entire wave.

Worked example Beginner

Take the simplest energy on profiles: the total squared height, written $H$ , obtained by adding up half of $u$ times $u$ at every point along the channel. We want to see how the steering rule produces motion.

Step 1. Find the energy gradient. If you raise the surface by a tiny bump at one point, the energy changes in proportion to the height there. So the gradient of $H$ is just the profile $u$ itself.

Step 2. Apply the steering operator. The simplest operator is "take the slope along the channel," which sends a profile to its rate of change in $x$ . Applied to the gradient $u$ , it gives the slope of $u$ , written $u_{x}$ .

Step 3. Read off the motion. The rule says the surface moves at the rate given by the output: $u_{t} = u_{x}$ . This is a wave that slides rightward without changing shape, at unit speed.

Step 4. Sanity check with numbers. Suppose at one instant the profile is a single bump centered at $x = 0$ . The slope $u_{x}$ is positive just left of the peak and negative just right of it. Adding a little of $u_{x}$ shifts the whole bump to the right. A moment later the same bump sits at a slightly larger $x$ .

What this tells us: an energy plus a steering operator produces a definite motion of the entire profile. The energy chose what is being conserved; the operator chose how the conserved quantity drives the flow. Replacing this toy energy with a more interesting one, keeping the same operator, is exactly how the shallow-water wave equation arises.

Check your understanding Beginner

Exercise (easy, multiple choice).

In the infinite-dimensional picture, what plays the role of a single "point" of the phase space?

A. A single number, the height of the wave at the origin
B. A pair of numbers, position and momentum
C. An entire profile $u (x)$ — the height of the surface at every point
D. The total energy of the system

Hint

The state has to record everything you would need to predict the wave's future.

Answer

C. A point of the field phase space is a whole function $u (x)$ , the entire shape of the surface. Feedback-correct: this is what makes the space infinite-dimensional — you need the height at every point, not a finite list. Feedback-wrong: a single number (A) or a pair (B) describes a particle, not a field; the energy (D) is a number assigned to a state, not the state itself.

Formal definition Intermediate+

Work on the algebra $A$ of differential functions of one field $u$ over the line: smooth functions $f (x, u, u_{x}, u_{xx}, \dots)$ of $x$ , $u = u (x)$ , and finitely many $x$ -derivatives $u_{j} = D^{j} u$ , where $D = \partial_{x} + \sum_{j \geq 0} u_{j + 1} \partial_{u_{j}}$ is the total derivative. A functional is an equivalence class $$ F[u] = \int f,dx, \qquad f \in \mathcal{A}, $$ where two densities define the same functional when they differ by a total derivative $D g$ (assuming fields decaying at infinity, or periodic, so boundary terms vanish). The space of functionals is $F = A / D A$ .

The variational (or functional) derivative of $F = \int f d x$ is the Euler operator applied to the density: $$ \frac{\delta F}{\delta u} = E(f) := \sum_{j \geq 0} (-D)^j, \frac{\partial f}{\partial u_j}. $$ The operator $E$ annihilates total derivatives, $E (D g) = 0$ , so $δ F / δ u$ depends only on the functional, not on the chosen density. The symbols $D$ , $E$ , $δ / δ u$ , and the formal antiderivative $D^{- 1}$ used below are recorded in _meta/NOTATION.md.

A Hamiltonian operator is a linear differential (or pseudo-differential) operator $D$ on $A$ that is skew-adjoint, $D^{*} = - D$ with respect to the pairing $⟨ P, Q ⟩ = \int P Q d x$ , and whose associated bracket $$ {F, G} = \int \frac{\delta F}{\delta u}, \mathcal{D}, \frac{\delta G}{\delta u},dx $$ satisfies the Jacobi identity ${{F, G}, H} + (cyclic) = 0$ for all functionals $F, G, H$ . Skew-adjointness gives antisymmetry of the bracket; the Jacobi identity is an additional condition on $D$ , expressed below as the vanishing of a functional tri-vector. The pair $(F, {\cdot, \cdot})$ is an infinite-dimensional Poisson manifold in the algebraic sense ^{[Olver 1993]}.

The Hamiltonian evolution equation generated by a Hamiltonian functional $H = \int H d x$ is $$ u_t = \mathcal{D}, \frac{\delta H}{\delta u}. $$ A functional $F$ is conserved along this flow when $d F / d t = {F, H} = 0$ . Two functionals are in involution when ${F, G} = 0$ .

Counterexamples to common slips

Skew-adjointness is necessary but not sufficient. For first-order constant-coefficient $D$ it forces Jacobi, but a general skew-adjoint operator with $u$ -dependent coefficients can violate the Jacobi identity; the tri-vector condition must be checked separately.
The variational derivative is not the partial derivative $\partial f / \partial u$ . The Euler operator includes the alternating sum of total derivatives of $\partial f / \partial u_{j}$ ; dropping the higher terms gives the wrong gradient whenever the density depends on $u_{x}$ or higher.
A constant Hamiltonian density is not invisible. Two densities differing by a total derivative give the same functional, but $δ / δ u$ already quotients this out — one must not also discard genuine $u$ -dependence that merely looks like a derivative.
$D^{- 1}$ is a formal antiderivative. In the second KdV structure $\partial_{x}^{- 1}$ acts only on expressions that are total $x$ -derivatives along the hierarchy; it is not a bounded inverse and applying it to an arbitrary density is not a closed operation.

Key theorem with proof Intermediate+

Theorem (the constant-coefficient operator $D = D = \partial_{x}$ is Hamiltonian; KdV is Hamiltonian for it). The operator $D = \partial_{x}$ is skew-adjoint and its bracket ${F, G} = \int (δ F / δ u) \partial_{x} (δ G / δ u) d x$ satisfies the Jacobi identity, so $(F, {\cdot, \cdot})$ is a Poisson manifold. The Korteweg-de Vries equation $$ u_t = u_{xxx} + 6 u u_x $$ is the Hamiltonian evolution equation $u_{t} = \partial_{x} (δ H / δ u)$ for $$ H = \int \Big( u^3 - \tfrac12 u_x^2 \Big),dx. $$

Proof. Skew-adjointness. For densities $P, Q$ decaying at infinity, integration by parts gives $\int P (\partial_{x} Q) d x = - \int (\partial_{x} P) Q d x$ , the boundary term vanishing. Hence $\partial_{x}^{*} = - \partial_{x}$ , and the bracket is antisymmetric: ${F, G} = \int (δ F / δ u) \partial_{x} (δ G / δ u) d x = - \int \partial_{x} (δ F / δ u) (δ G / δ u) d x = - {G, F}$ .

Jacobi identity. Write $ξ_{F} = δ F / δ u$ . For a constant-coefficient operator the Jacobiator reduces to a single computation. The functional bivector of $D = \partial_{x}$ is $Θ = \frac{1}{2} \int θ \land θ_{x} d x$ , where $θ$ is the basic vertical one-form (the uni-vector dual to $δ / δ u$ ) and $θ_{x} = D θ$ . The Jacobi identity is equivalent to the vanishing of the prolonged bivector $pr v_{D θ} (Θ)$ . Since the coefficient of $D = \partial_{x}$ does not depend on $u$ , the evolutionary field $v_{D θ} = v_{θ_{x}}$ annihilates the constant coefficient of $Θ$ , so $pr v_{D θ} (Θ) = \frac{1}{2} \int θ_{x} \land θ_{x} d x + \frac{1}{2} \int θ \land θ_{xx} d x$ . The first term vanishes by antisymmetry of the wedge, and the second is a total derivative of $\frac{1}{2} θ \land θ_{x}$ minus $\frac{1}{2} \int θ_{x} \land θ_{x} d x$ , both of which vanish in $F$ . Hence the Jacobiator is zero and ${\cdot, \cdot}$ is a Poisson bracket.

KdV in Hamiltonian form. Compute $δ H / δ u$ for $H = \int (u^{3} - \frac{1}{2} u_{x}^{2}) d x$ . With density $H = u^{3} - \frac{1}{2} u_{x}^{2}$ , $$ \frac{\delta H}{\delta u} = \frac{\partial \mathcal{H}}{\partial u} - D,\frac{\partial \mathcal{H}}{\partial u_x} = 3u^2 - D(-u_x) = 3u^2 + u_{xx}. $$ Applying $D = \partial_{x}$ , $$ u_t = \partial_x\big(3u^2 + u_{xx}\big) = 6 u u_x + u_{xxx}, $$ which is the KdV equation. $□$

Bridge. The pair $(D, H) = (\partial_{x}, \int (u^{3} - \frac{1}{2} u_{x}^{2}) d x)$ exhibits KdV as a Hamiltonian flow on a space of functions: it builds toward the bi-Hamiltonian theory of Magri, where a second compatible Hamiltonian operator $D_{2}$ generates the same equation from a different Hamiltonian and the recursion operator $R = D_{2} D_{1}^{- 1}$ produces the whole hierarchy; this is exactly the infinite-dimensional shape of the finite Poisson bracket ${F, G} = π^{ij} \partial_{i} F \partial_{j} G$ , with the antisymmetric matrix $π^{ij}$ replaced by the skew operator $D$ and the gradient $\partial_{i} F$ by the variational derivative $δ F / δ u$ ; it is dual to the cotangent-algebroid bracket on a Poisson manifold $[03.04.19]$ , the finite-dimensional shadow whose anchor $π^{♯}$ becomes the Hamiltonian operator $D$ ; and it appears again in the involution ${H_{m}, H_{n}} = 0$ of the conserved functionals, which is the central insight turning a single nonlinear wave equation into an integrable hierarchy with infinitely many conservation laws.

Exercises Intermediate+

Exercise 5 (medium, symbolic).

Write the linear wave equation $u_{tt} = u_{xx}$ in first-order Hamiltonian form. Introduce a momentum field $v = u_{t}$ , take the state $(u, v)$ , the energy $H = \frac{1}{2} \int (v^{2} + u_{x}^{2}) d x$ , and the operator $D = (0 - 1 10)$ .

Hint

Compute $δ H / δ u$ and $δ H / δ v$ , then apply the constant skew matrix $D$ .

Answer

$δ H / δ v = v$ and $δ H / δ u = - u_{xx}$ (from $\frac{1}{2} u_{x}^{2}$ via $- D u_{x} = - u_{xx}$ ). Then $$ \begin{pmatrix} u_t \ v_t\end{pmatrix} = \mathcal{D}\begin{pmatrix} \delta H/\delta u \ \delta H/\delta v\end{pmatrix} = \begin{pmatrix} v \ u_{xx}\end{pmatrix}. $$ So $u_{t} = v$ and $v_{t} = u_{xx}$ , i.e. $u_{tt} = v_{t} = u_{xx}$ , the wave equation. $D$ is the constant symplectic matrix and is Hamiltonian. Rubric: full credit for both variational derivatives and recovering the second-order equation.

Exercise 6 (medium, short-answer).

State the second (Magri) Hamiltonian operator of KdV and the Hamiltonian for which it reproduces the KdV equation, given the first structure uses $H_{2} = \int (u^{3} - \frac{1}{2} u_{x}^{2}) d x$ .

Hint

The second operator is third-order: $D_{2} = \partial_{x}^{3} + \frac{2}{3} (u \partial_{x} + \partial_{x} u)$ ; it pairs with a lower conserved functional.

Answer

The second structure is $D_{2} = \partial_{x}^{3} + \frac{2}{3} (u \partial_{x} + \partial_{x} u) = \partial_{x}^{3} + \frac{4}{3} u \partial_{x} + \frac{2}{3} u_{x}$ , and KdV is $u_{t} = D_{2} (δ H_{1} / δ u)$ with the momentum Hamiltonian $H_{1} = \frac{1}{2} \int u^{2} d x$ , whose variational derivative is $u$ . Indeed $D_{2} u = u_{xxx} + \frac{4}{3} u u_{x} + \frac{2}{3} u u_{x} = u_{xxx} + 2 u u_{x}$ — up to the standard normalisation/scaling of $u$ this is the KdV flow; with the conventional coefficient choice $D_{2} = \partial_{x}^{3} + 2 (u \partial_{x} + \partial_{x} u)$ one gets exactly $u_{xxx} + 6 u u_{x}$ . The two structures generate the same equation from different Hamiltonians: $u_{t} = D_{1} δ H_{2} = D_{2} δ H_{1}$ . Rubric: full credit for the third-order $D_{2}$ , the pairing with $H_{1} = \frac{1}{2} \int u^{2}$ , and the bi-Hamiltonian statement.

Exercise 8 (hard, short-answer).

The nonlinear Schrödinger equation $i ψ_{t} = - ψ_{xx} + 2∣ ψ ∣^{2} ψ$ is Hamiltonian. Identify the Poisson structure (the bracket on the complex field $ψ, \overset{ˉ}{ψ}$ ) and the Hamiltonian functional, and explain how the factor $i$ encodes the skew operator.

Hint

Treat $ψ$ and $\overset{ˉ}{ψ}$ as conjugate fields; the operator is $D = (0 i - i 0)$ and $H = \int (∣ ψ_{x} ∣^{2} + ∣ ψ ∣^{4}) d x$ .

Answer

Take the state $(ψ, \overset{ˉ}{ψ})$ with the canonical bracket ${ψ (x), \overset{ˉ}{ψ} (y)} = - i δ (x - y)$ , i.e. the constant skew operator $D = (0 i - i 0)$ , and Hamiltonian $H = \int (∣ ψ_{x} ∣^{2} + ∣ ψ ∣^{4}) d x$ . The variational derivative $δ H / δ \overset{ˉ}{ψ} = - ψ_{xx} + 2∣ ψ ∣^{2} ψ$ , so $ψ_{t} = - i δ H / δ \overset{ˉ}{ψ} = - i (- ψ_{xx} + 2∣ ψ ∣^{2} ψ)$ , which is $i ψ_{t} = - ψ_{xx} + 2∣ ψ ∣^{2} ψ$ . The factor $i$ is exactly the off-diagonal skew structure pairing $ψ$ with its conjugate momentum $i \overset{ˉ}{ψ}$ ; NLS is the field analogue of a complex harmonic oscillator $i \overset{z}{˙} = \partial H / \partial \overset{z}{ˉ}$ . Rubric: full credit for the canonical bracket, the Hamiltonian, the variational derivative, and the role of $i$ as the symplectic operator.

Exercise 9 (hard, short-answer).

Explain why, for a constant-coefficient skew operator $D$ , skew-adjointness alone implies the Jacobi identity, but for a $u$ -dependent $D$ it does not. Reference the tri-vector / prolongation condition.

Hint

The Jacobiator measures the failure of $pr v_{D θ} (Θ)$ to vanish; the prolongation acts on the coefficients of $D$ .

Answer

The Jacobi identity for a Hamiltonian operator is equivalent to the vanishing of the functional tri-vector $pr v_{D θ} (Θ)$ , where $Θ = \frac{1}{2} \int θ \land D θ d x$ is the bivector and the prolonged evolutionary field $v_{D θ}$ differentiates the coefficients of $D$ along the field. When $D$ has constant coefficients, $v_{D θ}$ annihilates those coefficients, and the surviving terms are total derivatives or wedges of equal forms, which vanish in $F$ ; so skew-adjointness suffices. When $D$ depends on $u$ (as in the second KdV structure), the prolongation hits the coefficients and produces a generally nonzero tri-vector; skew-adjointness only secures antisymmetry of the bracket, not Jacobi, and the tri-vector identity becomes a genuine constraint (satisfied for $D_{2}$ , which is why the pair is bi-Hamiltonian). Rubric: full credit for the tri-vector characterisation, the constant-coefficient annihilation argument, and the $u$ -dependent failure mode.

Lean formalization Intermediate+

Mathlib contains neither the algebra of differential functions modulo total derivatives, the Euler operator, nor skew-adjoint Hamiltonian operators on a field space, so the unit is formalisation-free; the obstacles are catalogued in Mathlib gap analysis. The aspirational target statement has the schematic form:

-- Aspirational, not currently realisable in Mathlib.
theorem kdv_is_hamiltonian
    (D : HamiltonianOperator)      -- D = ∂ₓ, skew-adjoint, Jacobi verified
    (H : Functional)               -- H = ∫ (u^3 - ½ uₓ²) dx
    (hD : D = totalDeriv)
    (hH : H = ∫ (u^3 - (1/2) * uₓ^2)) :
    evolutionEquation D H = (fun u => uₓₓₓ + 6 * u * uₓ) :=
  sorry

theorem partial_x_hamiltonian :
    (totalDeriv : Operator).SkewAdjoint ∧ JacobiIdentity (bracketOf totalDeriv) :=
  sorry

The statements need Functional (the quotient $A / D A$ ), the variational derivative varDeriv realising the Euler operator $E$ , HamiltonianOperator carrying SkewAdjoint and JacobiIdentity predicates, and bracketOf assembling ${F, G} = \int (δ F / δ u) D (δ G / δ u) d x$ . The closest existing Mathlib infrastructure is Derivation and the integration API, but the variational bicomplex and the Schouten bracket on functional multivectors — which express the Jacobi identity — are absent. The human reviewer is the correctness gate for the bracket and involution computations.

Advanced results Master

The Gardner-Zakharov-Faddeev bracket. Gardner 1971 (J. Math. Phys. 12 ^{[Gardner 1971]}) recognised KdV as a Hamiltonian system on the space of functions $u (x)$ with the first structure $D_{1} = \partial_{x}$ and Hamiltonian $H_{2} = \int (u^{3} - \frac{1}{2} u_{x}^{2}) d x$ , and Zakharov-Faddeev 1971 (Funct. Anal. Appl. 5 ^{[Zakharov-Faddeev 1971]}) exhibited the action-angle variables that make the system completely integrable: the scattering data of the associated Schrödinger operator are canonically conjugate, and the infinitely many conserved functionals $H_{n}$ are the Hamiltonians of commuting flows. The bracket ${u (x), u (y)} = \partial_{x} δ (x - y)$ is the Lie-Poisson bracket of the abelian Lie algebra of functions under addition, twisted by the cocycle $\int u \partial_{x} v d x$ ; this is the field-theoretic shadow of the cotangent-algebroid bracket of a Poisson manifold.

The bi-Hamiltonian pair. Magri 1978 (J. Math. Phys. 19 ^{[Magri 1978]}) found a second, compatible Hamiltonian operator $$ \mathcal{D}_2 = \partial_x^3 + 2(u,\partial_x + \partial_x, u), $$ skew-adjoint, with its own Jacobi identity, such that every pencil $D_{2} + λ D_{1}$ is Hamiltonian. KdV is doubly Hamiltonian, $u_{t} = D_{1} δ H_{2} = D_{2} δ H_{1}$ , and the recursion operator $R = D_{2} D_{1}^{- 1} = \partial_{x}^{2} + 4 u + 2 u_{x} \partial_{x}^{- 1}$ steps along the hierarchy $H_{n + 1}$ from $H_{n}$ via $D_{1} δ H_{n + 1} = D_{2} δ H_{n}$ . The compatibility of the pencil is equivalent to the heredity (vanishing Nijenhuis torsion) of $R$ , linking this structure to the master-symmetry construction $[05.09.11]$ .

Involution and the hierarchy. The conserved functionals $H_{n}$ generated by the recursion operator satisfy ${H_{m}, H_{n}}_{D_{1}} = 0$ for all $m, n$ : they are in involution with respect to either bracket, which is the infinite-dimensional Liouville-integrability statement. The proof is the Lenard-Magri recursion: ${H_{m}, H_{n}}_{1} = \int δ H_{m} D_{1} δ H_{n} = \int δ H_{m} D_{2} δ H_{n - 1} = - \int D_{2} δ H_{m} δ H_{n - 1} = - \int D_{1} δ H_{m + 1} δ H_{n - 1} = {H_{n - 1}, H_{m + 1}}_{1}$ , and iterating shifts indices until the bracket meets itself and vanishes by antisymmetry. The involution of the $H_{n}$ is the analytic content of integrability, replacing the dimension count of the finite Liouville-Arnold theorem $[05.02.03]$ .

Casimirs of the field bracket. For the first structure $D_{1} = \partial_{x}$ , a functional $C$ is a Casimir when $\partial_{x} (δ C / δ u) = 0$ , i.e. $δ C / δ u$ is constant; the mass $M = \int u d x$ is the generating Casimir, central to the bracket. For the second structure $D_{2}$ the kernel is spanned by different functionals, and the shift between the two Casimir families along the pencil is what drives the Lenard recursion. This is the infinite-dimensional instance of the degenerate-Poisson Casimir theory $[05.09.12]$ : the symplectic leaves are the level sets of the conserved mass, and the bracket is degenerate precisely along the Casimir directions.

The classical r-matrix. Faddeev-Takhtajan 1987 (Hamiltonian Methods in the Theory of Solitons ^{[Faddeev-Takhtajan 1987]}) recast the field Poisson brackets of soliton equations through the monodromy matrix $T (λ)$ of the auxiliary linear problem, with ${T (λ) \otimes T (μ)} = [r (λ - μ), T (λ) \otimes T (μ)]$ for a classical $r$ -matrix solving the classical Yang-Baxter equation. This packages the involution of the conserved functionals — the traces of powers of $T$ — into a single algebraic identity, and is the classical limit of the quantum inverse scattering method. The field Poisson manifold, the bi-Hamiltonian pair, and the $r$ -matrix are three presentations of the same Poisson structure on the phase space of fields.

Synthesis. The Hamiltonian formulation of evolution equations transports the entire apparatus of finite-dimensional mechanics to a space of functions: the gradient becomes the variational derivative $δ F / δ u = E (f)$ , the antisymmetric matrix becomes a skew-adjoint operator $D$ , and the Jacobi identity becomes the vanishing of a functional tri-vector. For KdV this single move produces four interlocking structures. The Gardner-Zakharov-Faddeev bracket $D_{1} = \partial_{x}$ realises the equation as an energy flow; the compatible second operator $D_{2}$ makes it bi-Hamiltonian and supplies the recursion operator $R = D_{2} D_{1}^{- 1}$ ; the recursion generates an involutive tower ${H_{m}, H_{n}} = 0$ that is the infinite-dimensional Liouville integrability; and the classical $r$ -matrix compresses that involution into one Yang-Baxter identity on the monodromy. The Casimirs of the two operators interleave along the Poisson pencil, driving the Lenard-Magri recursion that ties the four faces together. One skew operator on a space of functions, correctly chosen, manufactures the whole integrable structure, and each presentation is a different projection of the same field Poisson geometry.

Full proof set Master

Proposition (the Euler operator annihilates total derivatives). For any $g \in A$ , $E (D g) = 0$ , so the variational derivative is well-defined on functionals.

Proof. It suffices to check on monomials and use linearity. The total derivative satisfies $\partial_{u_{j}} (D g) = D (\partial_{u_{j}} g) + \partial_{u_{j - 1}} g$ (with the convention $\partial_{u_{- 1}} = 0$ ), since $D = \partial_{x} + \sum_{k} u_{k + 1} \partial_{u_{k}}$ and only the term $u_{j} \partial_{u_{j - 1}}$ contributes the extra summand. Then $$ E(Dg) = \sum_{j\geq 0}(-D)^j,\partial_{u_j}(Dg) = \sum_{j\geq 0}(-D)^j\big(D,\partial_{u_j}g + \partial_{u_{j-1}}g\big). $$ The first family of terms is $\sum_{j} (- D)^{j} D \partial_{u_{j}} g = - \sum_{j} (- D)^{j + 1} \partial_{u_{j}} g$ , and the second is $\sum_{j \geq 1} (- D)^{j} \partial_{u_{j - 1}} g = \sum_{k \geq 0} (- D)^{k + 1} \partial_{u_{k}} g$ . These two sums are negatives of one another and cancel termwise, so $E (D g) = 0$ . $□$

Proposition (skew-adjointness $\Leftrightarrow$ antisymmetry of the bracket). A differential operator $D$ satisfies $D^{*} = - D$ under the pairing $⟨ P, Q ⟩ = \int P Q d x$ if and only if the bracket ${F, G} = \int (δ F / δ u) D (δ G / δ u) d x$ is antisymmetric for all functionals.

Proof. Suppose $D^{*} = - D$ . Writing $ξ = δ F / δ u$ , $η = δ G / δ u$ , $$ {F,G} = \int \xi,\mathcal{D}\eta,dx = \int (\mathcal{D}^\ast\xi),\eta,dx = -\int (\mathcal{D}\xi),\eta,dx = -\int \eta,\mathcal{D}\xi,dx = -{G,F}. $$ Conversely, antisymmetry for all $F, G$ forces $\int ξ (D + D^{*}) η d x = 0$ for all variational derivatives $ξ, η$ ; since these span a large enough subspace and the symmetric part $\frac{1}{2} (D + D^{*})$ is a differential operator detected by such pairings, $D + D^{*} = 0$ . $□$

Proposition (Lenard-Magri involution). Let $D_{1}, D_{2}$ be a compatible Hamiltonian pair and ${H_{n}}$ a Lenard sequence with $D_{1} δ H_{n + 1} = D_{2} δ H_{n}$ for all $n$ . Then ${H_{m}, H_{n}}_{1} = 0$ for all $m, n$ , where ${F, G}_{1} = \int δ F D_{1} δ G d x$ .

Proof. Write $ξ_{k} = δ H_{k}$ . Skew-adjointness of $D_{1}, D_{2}$ gives $\int ξ_{m} D_{i} ξ_{n} d x = - \int ξ_{n} D_{i} ξ_{m} d x$ . Using the Lenard relation $D_{1} ξ_{k + 1} = D_{2} ξ_{k}$ , $$ {H_m, H_n}1 = \int \xi_m,\mathcal{D}1\xi_n,dx = \int \xi_m,\mathcal{D}2\xi{n-1},dx = -\int (\mathcal{D}2\xi_m),\xi{n-1},dx = -\int (\mathcal{D}1\xi{m+1}),\xi{n-1},dx = {H{n-1}, H_{m+1}}1. $$ Thus the pairing satisfies ${H_m, H_n}1 = {H{n-1}, H{m+1}}_1 $, a s hi f tt ha tp r eser v es$ m + n $w hi l e m o v in g t h e in d i ces t o w a r d e a c h o t h er . I t er a t in g$ |m - n| $s t e p sr e a c h es ab r a c k e t o f a f u n c t i o na l w i t hi t se l f,$ {H_k, H_k}1 = 0 $b y an t i sy mm e t r y, w h e n$ m + n $i se v e n; w h e n$ m + n $i so dd i t r e a c h es$ {H_k, H{k+1}}1 = {H_k, H{k+1}}_1 $f or ce d e q u a l t o i t so w nn e g a t i v e t h r o ug h o n e m or es hi f t, a g ain v ani s hin g . H e n ce a l l t h e$ H_n $a r e inin v o l u t i o n .$ \square$

Connections Master

Cotangent algebroid of a Poisson manifold 03.04.19. The finite-dimensional shadow: a Poisson bracket ${F, G} = π^{ij} \partial_{i} F \partial_{j} G$ is built from the bivector $π$ with anchor $π^{♯} : T^{*} M \to T M$ . The infinite-dimensional bracket replaces $π^{ij}$ by the skew operator $D$ and the gradient $\partial_{i} F$ by the variational derivative $δ F / δ u$ ; the Jacobi identity, which for $π$ is the vanishing of the Schouten bracket $[π, π] = 0$ , becomes the vanishing of the functional tri-vector. The cotangent-Lie-algebroid bracket of one-forms is the model whose field version is the bracket ${u (x), u (y)} = \partial_{x} δ (x - y)$ .
Master symmetries and the Fuchssteiner construction 05.09.11. The recursion operator $R = D_{2} D_{1}^{- 1}$ assembled from the bi-Hamiltonian pair is exactly the recursion operator of the master-symmetry construction; the compatibility of the Poisson pencil is equivalent to the heredity (vanishing Nijenhuis torsion) of $R$ , and the conserved Hamiltonians $H_{n} =$ integrals of the densities generated by $R$ are the involutive family produced there. The master symmetry is the conformal generator rescaling the two operators $D_{1}, D_{2}$ with unit relative weight.
Casimir functions of degenerate Poisson structures 05.09.12. The field bracket $D_{1} = \partial_{x}$ is degenerate; its Casimir is the mass $M = \int u d x$ , whose level sets are the symplectic leaves of the field phase space. The interleaving of the Casimir families of $D_{1}$ and $D_{2}$ along the Poisson pencil is the mechanism behind the Lenard-Magri recursion, the infinite-dimensional instance of distinguished functions of a degenerate structure.
Integrable system 05.02.03. The involution ${H_{m}, H_{n}} = 0$ of the conserved functionals is the infinite-dimensional analogue of the commuting integrals of a Liouville-Arnold system. The action-angle variables of Zakharov-Faddeev — the scattering data of the auxiliary Schrödinger operator — are the field-theoretic action-angle coordinates, completing the analogy with the finite-dimensional theory.

Historical & philosophical context Master

The Hamiltonian reading of nonlinear waves began with Gardner 1971 ^{[Gardner 1971]} (J. Math. Phys. 12), the fourth in the Korteweg-de Vries series, which exhibited KdV as a Hamiltonian system on the space of functions $u (x)$ with bracket generated by $\partial_{x}$ and the cubic-energy Hamiltonian. In the same year Zakharov and Faddeev 1971 ^{[Zakharov-Faddeev 1971]} (Funct. Anal. Appl. 5) identified the scattering data as action-angle variables and proved complete integrability, fixing the infinite-dimensional analogue of the Liouville-Arnold theorem for KdV.

Gelfand and Dikii 1975 ^{[Gelfand-Dikii 1975]} (Russ. Math. Surveys 30) developed the formal variational calculus — the Euler operator, the algebra of differential functions, the quotient by total derivatives — that makes the functional Poisson bracket rigorous as algebra rather than analysis. Magri 1978 ^{[Magri 1978]} (J. Math. Phys. 19) discovered the second compatible Hamiltonian operator, recasting integrability as the existence of a compatible Poisson pencil and producing the recursion operator as its ratio. Olver 1977 and the canonical exposition Olver 1993 ^{[Olver 1993]} (Applications of Lie Groups to Differential Equations, 2nd ed., §7.1-§7.2) gave the prolongation test for the Jacobi identity of a Hamiltonian operator and embedded the whole theory in the geometry of jet spaces.

Faddeev and Takhtajan 1987 ^{[Faddeev-Takhtajan 1987]} (Hamiltonian Methods in the Theory of Solitons) organised the field Poisson brackets through the classical $r$ -matrix and the monodromy of the auxiliary linear problem, the classical limit of the quantum inverse scattering method developed in Leningrad. The construction is now standard across the soliton literature: KdV, the nonlinear Schrödinger equation, the sine-Gordon equation, and the Toda lattice each carry an explicit Hamiltonian operator, a compatible second structure, and an $r$ -matrix governing the involution of their conserved functionals.

Bibliography Master

@article{Gardner1971,
  author = {Gardner, Clifford S.},
  title = {Korteweg--de {V}ries equation and generalizations. {IV}. {T}he {K}orteweg--de {V}ries equation as a {H}amiltonian system},
  journal = {Journal of Mathematical Physics},
  volume = {12},
  year = {1971},
  pages = {1548--1551},
}

@article{ZakharovFaddeev1971,
  author = {Zakharov, Vladimir E. and Faddeev, Ludvig D.},
  title = {Korteweg--de {V}ries equation: a completely integrable {H}amiltonian system},
  journal = {Functional Analysis and its Applications},
  volume = {5},
  year = {1971},
  pages = {280--287},
}

@article{GelfandDikii1975,
  author = {Gelfand, Israel M. and Dikii, Leonid A.},
  title = {Asymptotic behaviour of the resolvent of {S}turm--{L}iouville equations and the algebra of the {K}orteweg--de {V}ries equations},
  journal = {Russian Mathematical Surveys},
  volume = {30},
  year = {1975},
  pages = {77--113},
}

@article{Magri1978,
  author = {Magri, Franco},
  title = {A simple model of the integrable {H}amiltonian equation},
  journal = {Journal of Mathematical Physics},
  volume = {19},
  year = {1978},
  pages = {1156--1162},
}

@article{Olver1977,
  author = {Olver, Peter J.},
  title = {Evolution equations possessing infinitely many symmetries},
  journal = {Journal of Mathematical Physics},
  volume = {18},
  year = {1977},
  pages = {1212--1215},
}

@book{Olver1993,
  author = {Olver, Peter J.},
  title = {Applications of {L}ie Groups to Differential Equations},
  publisher = {Springer},
  series = {Graduate Texts in Mathematics},
  volume = {107},
  edition = {2nd},
  year = {1993},
}

@book{FaddeevTakhtajan1987,
  author = {Faddeev, Ludvig D. and Takhtajan, Leon A.},
  title = {Hamiltonian Methods in the Theory of Solitons},
  publisher = {Springer},
  series = {Springer Series in Soviet Mathematics},
  year = {1987},
}

@book{DrazinJohnson1989,
  author = {Drazin, Philip G. and Johnson, Robin S.},
  title = {Solitons: an Introduction},
  publisher = {Cambridge University Press},
  year = {1989},
}

@book{Dickey2003,
  author = {Dickey, L. A.},
  title = {Soliton Equations and {H}amiltonian Systems},
  publisher = {World Scientific},
  edition = {2nd},
  year = {2003},
}

Prerequisites

05.02.02
05.09.11
05.09.12
03.04.19
05.02.01

Tier anchors

beginner: Olver *Applications of Lie Groups to Differential Equations* §7.1 informal opening; Drazin-Johnson *Solitons: an Introduction* Ch. 2-3 on KdV as an energy flow
intermediate: Olver *Applications of Lie Groups to Differential Equations* 2nd ed. §7.1-7.2; Dickey *Soliton Equations and Hamiltonian Systems* Ch. 1
master: Gardner 1971 *The Korteweg-de Vries equation as a Hamiltonian system* (J. Math. Phys. 12, originator); Zakharov-Faddeev 1971 *Korteweg-de Vries equation: a completely integrable Hamiltonian system* (Funct. Anal. Appl. 5); Magri 1978 *A simple model of the integrable Hamiltonian equation* (J. Math. Phys. 19, bi-Hamiltonian pair); Faddeev-Takhtajan 1987 *Hamiltonian Methods in the Theory of Solitons* Part I; Olver 1993 §7.1-7.2

References

Gardner 1971 — Korteweg-de Vries equation and generalizations. IV. The Korteweg-de Vries equation as a Hamiltonian system · Journal of Mathematical Physics 12, originator of KdV as a Hamiltonian system on the space of functions
Zakharov-Faddeev 1971 — Korteweg-de Vries equation: a completely integrable Hamiltonian system · Functional Analysis and its Applications 5, action-angle variables and complete integrability
Magri 1978 — A simple model of the integrable Hamiltonian equation · Journal of Mathematical Physics 19, the compatible bi-Hamiltonian pair for KdV
Faddeev-Takhtajan 1987 — Hamiltonian Methods in the Theory of Solitons · Springer, Part I, field-theoretic Poisson brackets and the classical r-matrix
Olver 1993 — Applications of Lie Groups to Differential Equations, 2nd ed. · Springer GTM 107, §7.1-§7.2 Hamiltonian operators and the functional Jacobi condition
Gelfand-Dikii 1975 — Asymptotic behaviour of the resolvent of Sturm-Liouville equations and the algebra of the Korteweg-de Vries equations · Russian Mathematical Surveys 30, formal variational calculus and the Euler operator

Estimated time

beginner: 18m
intermediate: 45m
master: 85m