05.05.09 · symplectic / lagrangian

Generalised symmetries (Lie-Bäcklund) and recursion operators

shipped3 tiersLean: none

Anchor (Master): Olver §5.1-§5.2-§5.3; Olver 1977 (J. Math. Phys. 18); Fokas 1987 (Stud. Appl. Math. 77); Mikhailov-Shabat-Sokolov in *What is Integrability?* (1991)

Intuition Beginner

A symmetry of an equation is a way of nudging its solutions that turns each one into another solution. The earlier symmetries you met moved only the plain variables: shift the input, scale the output, rotate the picture. The nudge at a point depended only on where that point was. But there is no reason the nudge has to be so modest. You can let the nudge at a point depend also on the slope of the solution there, on its curvature, on how the curvature is changing, and so on up the ladder of derivatives.

When the nudge is allowed to read off the slope and higher bending of the curve, it is called a generalised symmetry, or a Lie-Bäcklund symmetry after the man who first studied transformations of this richer kind. These symmetries are invisible to the older point-by-point bookkeeping, yet they are often the more important ones. For the equations that physicists call integrable, the point symmetries are few, but the generalised symmetries form an endless tower, one for every rung of an infinite ladder.

The tool that climbs the ladder is a recursion operator. It is a fixed recipe that takes one symmetry as input and returns a new, higher symmetry as output. Feed it the simplest symmetry, apply it again and again, and out comes the whole tower for free. The recipe is what makes an integrable equation special.

Visual Beginner

Picture a solution drawn as a curve. At one point, attach not just an arrow saying which way to push, but a whole reading of the curve there: its height, its slope, its curvature. A generalised symmetry is a push whose direction is computed from that full reading. Two points with the same height but different slopes get pushed differently.

Beside the curve, draw a ladder. Each rung is one symmetry of the equation. The bottom rung is the plainest one. A single curved arrow, the recursion operator, hooks onto any rung and lifts you to the next rung up. Because the same arrow works at every level, one operator builds an infinite ladder. The picture captures the whole idea: symmetries may read the curve as deeply as they like, and one recipe manufactures all of them.

Worked example Beginner

Take the heat equation, the law that says the rate of change in time of a temperature equals how its profile bends in space. Write the temperature as $u$ , its rate of change in time as $u_{t}$ , and the bending in space as $u_{xx}$ . The law is $u_{t}$ equals $u_{xx}$ .

A plain symmetry is "shift in space": move the whole profile sideways and it still solves the law. The nudge here reads only position. Now try a deeper nudge: push the temperature at each point by the amount $u_{xxx}$ , the third derivative in space, the rate at which the bending changes.

Why is this a symmetry? Differentiate the law $u_{t}$ equals $u_{xx}$ three times in space. The left side becomes the time rate of $u_{xxx}$ , and the right side becomes the space bending of $u_{xxx}$ . So $u_{xxx}$ obeys the very same heat law that $u$ does. Pushing every solution by its own $u_{xxx}$ therefore lands you on another solution.

What this tells us: the heat equation has a symmetry for every spatial derivative, $u_{x}$ , $u_{xx}$ , $u_{xxx}$ , and onward, an infinite family. None of these except the first is a plain point symmetry, because each reads the curve through high derivatives. This endless family is exactly what generalised symmetries capture and what a recursion operator is built to generate.

Check your understanding Beginner

Exercise (easy, multiple choice).

What makes a symmetry "generalised" (Lie-Bäcklund) rather than a plain point symmetry?

A. It changes the equation into a different equation
B. The nudge it applies may depend on the slope and higher derivatives of the solution, not only on position
C. It only works for linear equations
D. It is a conserved quantity rather than a transformation

Hint

Recall the reading attached to a point: height, slope, curvature, and beyond.

Answer

B. A generalised symmetry lets the infinitesimal nudge depend on derivatives of the solution up to any order, so two points with the same height but different slopes are nudged differently. Feedback-correct: this dependence on the derivative data is exactly the extra freedom Bäcklund opened up. Feedback-wrong: a symmetry preserves the equation rather than changing it (A); generalised symmetries are most powerful for nonlinear integrable equations (C); a symmetry is a transformation, not a conserved quantity (D).

Formal definition Intermediate+

Work in the algebra $A$ of differential functions: smooth functions of the independent variables $x = (x^{1}, \dots, x^{p})$ , the dependent variables $u = (u^{1}, \dots, u^{q})$ , and finitely many derivative coordinates $u_{J}^{α}$ on the infinite jet bundle $J^{\infty} (E)$ of 05.05.05, where $J$ ranges over symmetric multi-indices. The total derivative $D_{i} = \partial_{x^{i}} + \sum_{α, J} u_{J, i}^{α} \partial_{u_{J}^{α}}$ acts as a derivation of $A$ , and $D_{J} = D_{1}^{j_{1}} \dots D_{p}^{j_{p}}$ . The development follows Olver ^{[Olver §5.1]}.

Definition (generalised vector field). A generalised vector field is a formal expression $$ v = \xi^i(x, u^{(n)}),\frac{\partial}{\partial x^i} + \phi^\alpha(x, u^{(n)}),\frac{\partial}{\partial u^\alpha}, $$ whose coefficients $ξ^{i}, ϕ^{α} \in A$ may depend on derivatives $u^{(n)}$ of any finite order $n$ , in contrast to the point fields of 05.05.06 where $ξ^{i}, ϕ^{α}$ depend on $(x, u)$ alone. Its characteristic is $Q = (Q^{1}, \dots, Q^{q})$ with $Q^{α} = ϕ^{α} - ξ^{i} u_{i}^{α}$ , and its evolutionary representative is the vertical field $$ v_Q = \sum_{\alpha} Q^\alpha,\frac{\partial}{\partial u^\alpha}, \qquad \mathrm{pr},v_Q = \sum_{\alpha, J} (D_J Q^\alpha),\frac{\partial}{\partial u^\alpha_J}, $$ the prolongation formula reducing to $ϕ_{J}^{α} (v_{Q}) = D_{J} Q^{α}$ because $v_{Q}$ has no horizontal part. The symbols $D_{i}$ , $D^{- 1}$ , and the evolutionary field $pr v_{Q}$ are recorded in _meta/NOTATION.md.

Definition (generalised symmetry). Let $S = {Δ_{ν} (x, u^{(k)}) = 0}$ be a system of PDEs. A generalised vector field $v$ with characteristic $Q$ is a generalised symmetry (Lie-Bäcklund symmetry) of $S$ if $$ \mathrm{pr},v_Q(\Delta_\nu) = 0 \quad\text{on every solution of } \mathcal{S}, $$ that is, $pr v_{Q} (Δ_{ν})$ lies in the differential ideal generated by $Δ_{1}, \dots, Δ_{m}$ and their total derivatives. For an evolution equation $u_{t} = K (x, u^{(n)})$ written with $K \in A$ , a $t$ -independent characteristic $Q$ is a symmetry exactly when the characteristic bracket vanishes, $$ [K, Q] := \mathrm{pr},v_K(Q) - \mathrm{pr},v_Q(K) = 0, $$ so symmetries are the centraliser of $K$ in the characteristic Lie algebra. Here $pr v_{K} (Q) = K^{'} [Q]$ recovers the Fréchet derivative $K^{'}$ , the linearisation $K^{'} [\cdot] = \sum_{J} (\partial_{u_{J}} K) D_{J} (\cdot)$ .

Definition (null characteristic; equivalence). A characteristic $Q$ is null for $S$ (the standard literature also calls these vanishing-on-solutions characteristics) if $Q^{α} = 0$ on every solution, equivalently $Q^{α}$ lies in the differential ideal of $S$ . Two generalised symmetries are equivalent if their characteristics differ by a null characteristic; the evolutionary field of a null characteristic acts as zero on solutions, so generalised symmetries are properly objects of the quotient of characteristics modulo the null ones. Within this quotient every generalised vector field is equivalent to its evolutionary representative $v_{Q}$ , since $v$ and $v_{Q}$ differ by the total field $ξ^{i} D_{i}$ , which is tangent to all prolongations.

Definition (recursion operator). A linear operator $R : A^{q} \to A^{q}$ on characteristics (generally a formal integro-differential operator, built from $D_{i}$ , multiplication by elements of $A$ , and the formal antiderivative $D^{- 1}$ ) is a recursion operator for the evolution equation $u_{t} = K$ if it satisfies the operator identity $$ R_t = [,K', R,] := K',R - R,K', $$ where $R_{t} = pr v_{K} (R)$ differentiates the coefficients of $R$ along the flow and $K^{'}$ is the Fréchet derivative of $K$ . When $Q$ is a symmetry, $[K, Q] = 0$ , then $R Q$ is again a symmetry. For autonomous (time-independent) $R$ the condition reads $pr v_{K} (R) = [K^{'}, R]$ , the statement that $R$ commutes with the linearised flow.

A non-example fixes the meaning. The bare differential operator $R = D^{2}$ applied to the KdV symmetry $u_{x}$ returns $u_{xxx}$ , which is not a KdV symmetry on its own: $D^{2}$ fails the commutator identity $R_{t} = [K^{'}, R]$ for $K = u_{xxx} + 6 u u_{x}$ . The genuine KdV recursion operator must include the nonlocal correction $2 u_{x} D^{- 1}$ and the multiplication term $4 u$ ; only the full $R = D^{2} + 4 u + 2 u_{x} D^{- 1}$ satisfies the commutator condition.

Counterexamples to common slips

A generalised symmetry need not be a point symmetry, and conversely. The characteristic $Q = u_{xxx}$ for the heat equation is a genuine symmetry but corresponds to no transformation of $(x, u)$ alone; it lives only on the jet bundle.
Null characteristics are not zero — they vanish only on solutions. For $u_{t} = u_{xx}$ the characteristic $Q = u_{t} - u_{xx}$ is null: it vanishes on every solution and its evolutionary field acts as zero there, so it must be quotiented out before counting symmetries.
$D^{2}$ is not the KdV recursion operator. Dropping the nonlocal tail $2 u_{x} D^{- 1}$ breaks the commutator identity $R_{t} = [K^{'}, R]$ ; the differential part alone does not map symmetries to symmetries.
$D^{- 1}$ is a formal antiderivative, not a bounded operator. Expressions $u_{x} D^{- 1}$ act on the hierarchy because the integrands are total derivatives there; applying $D^{- 1}$ to an arbitrary element of $A$ is not a closed operation.

Key theorem with proof Intermediate+

Theorem (recursion operators map symmetries to symmetries). Let $u_{t} = K$ be an evolution equation with Fréchet derivative $K^{'}$ , and let $R$ be a recursion operator satisfying $R_{t} = [K^{'}, R]$ . If $Q$ is a generalised symmetry, that is $[K, Q] = 0$ equivalently $Q_{t} = K^{'} [Q]$ for a $t$ -independent symmetry along the flow, then $R Q$ is a generalised symmetry.

Proof. Recall the characterisation of a symmetry through the linearisation. A $t$ -independent characteristic $Q$ is a symmetry of $u_{t} = K$ precisely when its evolutionary field commutes with the flow, which in linearised form is the equation $$ \frac{d}{dt},Q = K'[Q] $$ along solutions, where $\frac{d}{d t} = \partial_{t} + pr v_{K}$ is the total time derivative carrying the explicit and implicit $t$ -dependence. For a $t$ -independent $Q$ the explicit part drops and this is $pr v_{K} (Q) = K^{'} [Q]$ , identical to $[K, Q] = 0$ after using $pr v_{K} (Q) = K^{'} [Q]$ and the antisymmetric definition of the bracket.

Apply the operator $R$ and the product rule to the candidate $R Q$ . Its total time derivative is $$ \frac{d}{dt}(RQ) = \Big(\frac{d}{dt}R\Big)Q + R,\frac{d}{dt}Q = R_t,Q + R,K'[Q], $$ using $\frac{d}{d t} R = R_{t} = pr v_{K} (R)$ for the coefficient evolution of $R$ and $\frac{d}{d t} Q = K^{'} [Q]$ from the symmetry hypothesis. Substitute the recursion-operator identity $R_{t} = K^{'} R - R K^{'}$ : $$ \frac{d}{dt}(RQ) = (K'R - R K'),Q + R,K'[Q] = K'[RQ] - R,K'[Q] + R,K'[Q] = K'[RQ]. $$ The two middle terms cancel, leaving $\frac{d}{d t} (R Q) = K^{'} [R Q]$ . This is precisely the linearised-symmetry equation for the characteristic $R Q$ , so $R Q$ satisfies the symmetry criterion and is a generalised symmetry of $u_{t} = K$ . $□$

Bridge. This theorem is the engine of the symmetry approach to integrability. It builds toward the construction of an infinite hierarchy: seeding $R$ with the translation symmetry $u_{x}$ and iterating produces $u_{x}, R u_{x}, R^{2} u_{x}, \dots$ , all symmetries by induction. It appears again in 05.09.11, where the very same recursion operator is the object the master symmetry $τ$ rescales through $L_{τ} R = R$ , so that the operator proved here to preserve symmetries is the one bracketing generates the hierarchy from. The construction reuses the prolongation calculus of 05.05.06 in a new guise: there tangency of $pr v_{Q}$ to the equation defined a symmetry, and the central insight is that the recursion condition $R_{t} = [K^{'}, R]$ is exactly the statement that $R$ intertwines the linearised flow with itself, so symmetry-preservation is intertwining of the linearisation. Putting these together, the recursion operator, the Fréchet derivative, and the characteristic bracket are three readings of one structure: the commutator algebra of the linearised evolution.

Exercises Intermediate+

Exercise 4 (medium, short-answer).

Explain why generalised symmetries must be counted modulo null characteristics, using the example $Q = u_{t} - u_{xx}$ for the heat equation.

Hint

A null characteristic vanishes on all solutions; its evolutionary field acts as zero there.

Answer

The characteristic $Q = u_{t} - u_{xx}$ vanishes identically on every solution of $u_{t} = u_{xx}$ , so it lies in the differential ideal of the equation and is a null characteristic. Its evolutionary field $v_{Q}$ acts as zero on the solution manifold, producing no genuine deformation of solutions. If null characteristics were not quotiented out, every symmetry could be padded by such an ideal element and the symmetry "count" would be meaningless. The correct object is the quotient of characteristics modulo the null ones; two characteristics that differ by a null one represent the same generalised symmetry. This is also why the evolutionary representative is canonical: $v$ and $v_{Q}$ differ by the total field $ξ^{i} D_{i}$ , whose action on solutions is captured by an ideal-valued adjustment. Rubric: full credit for identifying $Q$ as ideal-valued, the zero action on solutions, and the need for the quotient.

Exercise 5 (medium, symbolic).

Using the theorem, show by induction that $R^{n} u_{x}$ is a generalised symmetry of KdV for every $n \geq 0$ , given that $R$ satisfies $R_{t} = [K^{'}, R]$ and $u_{x}$ is a symmetry.

Hint

Base case: $u_{x}$ is the translation symmetry. Inductive step: apply the theorem to $Q = R^{n} u_{x}$ .

Answer

Base case $n = 0$ : $u_{x}$ is the characteristic of $x$ -translation, a symmetry of the autonomous KdV equation, $[K, u_{x}] = 0$ . Inductive step: suppose $Q_{n} = R^{n} u_{x}$ is a symmetry, $\partial_{t} Q_{n} = K^{'} [Q_{n}]$ along the flow. By the theorem, since $R$ satisfies $R_{t} = [K^{'}, R]$ , the characteristic $R Q_{n} = R^{n + 1} u_{x}$ satisfies $\partial_{t} (R Q_{n}) = K^{'} [R Q_{n}]$ , hence is again a symmetry. By induction $R^{n} u_{x}$ is a symmetry for all $n \geq 0$ , the infinite KdV hierarchy. Rubric: full credit for the correct base case, the appeal to the theorem in the inductive step, and the conclusion.

Exercise 6 (medium, short-answer).

State the recursion-operator condition $R_{t} = [K^{'}, R]$ in words, and explain why the differential operator $D^{2}$ alone fails it for KdV while $R = D^{2} + 4 u + 2 u_{x} D^{- 1}$ satisfies it.

Hint

The condition says $R$ commutes with the linearised flow $K^{'}$ up to its own time evolution.

Answer

The condition $R_{t} = [K^{'}, R]$ says that, modulo the evolution $R_{t}$ of its own coefficients along the flow, the operator $R$ commutes with the linearisation $K^{'}$ of the equation; equivalently $R$ intertwines the linearised evolution with itself. For KdV, $K = u_{xxx} + 6 u u_{x}$ has $K^{'} = D^{3} + 6 u D + 6 u_{x}$ . The bare $D^{2}$ has $R_{t} = 0$ but $[K^{'}, D^{2}] = [6 u D + 6 u_{x}, D^{2}] \neq = 0$ , so the identity fails and $D^{2} u_{x} = u_{xxx}$ is not a symmetry. Restoring the multiplication term $4 u$ and the nonlocal tail $2 u_{x} D^{- 1}$ produces exactly the cross terms needed: the $4 u$ corrects the commutator with $6 u D$ , and $2 u_{x} D^{- 1}$ supplies the nonlocal piece making $R_{t} = [K^{'}, R]$ hold. Only the full $R$ maps symmetries to symmetries. Rubric: full credit for the intertwining reading, the failure of $D^{2}$ , and the role of the two extra terms.

Exercise 7 (hard, short-answer).

The characteristics of an evolution equation form a Lie algebra under $[P, Q] = pr v_{P} (Q) - pr v_{Q} (P)$ . Show that the symmetries (the centraliser of $K$ ) form a subalgebra, and that a recursion operator $R$ generates an abelian subalgebra from a single seed when the iterates commute.

Hint

Use the Jacobi identity for the first part. For the second, $[u_{x}, R^{n} u_{x}]$ measures the failure to commute.

Answer

If $P, Q$ are symmetries, $[K, P] = [K, Q] = 0$ , then by the Jacobi identity $[K, [P, Q]] = [[K, P], Q] + [P, [K, Q]] = 0$ , so $[P, Q]$ is again a symmetry: the symmetries are a subalgebra. For the seed-generated tower $Q_{n} = R^{n} u_{x}$ , each $Q_{n}$ is a symmetry by the theorem, so ${Q_{n}}$ spans a subalgebra. The subalgebra is abelian precisely when $[Q_{m}, Q_{n}] = 0$ for all $m, n$ , which is the involutivity of the hierarchy; this holds when $R$ is hereditary (vanishing Nijenhuis torsion), as proved in 05.09.11. The translation symmetry $u_{x}$ acts as $D$ on characteristics, $pr v_{u_{x}} (Q) = D Q$ , so $[u_{x}, Q_{n}] = D Q_{n} - pr v_{Q_{n}} (u_{x})$ , and the cancellation that gives $[u_{x}, Q_{n}] = 0$ is the first instance of involutivity. Rubric: full credit for the Jacobi-identity subalgebra argument, the abelian condition, and the link to heredity.

Exercise 8 (hard, short-answer).

Explain how the recursion operator arises in the bi-Hamiltonian picture as $R = I_{1} I_{0}^{- 1}$ , and how the commutator condition $R_{t} = [K^{'}, R]$ relates to compatibility of the Poisson pencil for KdV.

Hint

KdV is bi-Hamiltonian with $I_{0} = D$ , $I_{1} = D^{3} + 4 u D + 2 u_{x}$ ; the recursion operator is their ratio.

Answer

Magri's theorem makes KdV doubly Hamiltonian, $u_{t} = I_{0} δ H_{1} = I_{1} δ H_{0}$ , with the compatible Hamiltonian operators $I_{0} = D$ and $I_{1} = D^{3} + 4 u D + 2 u_{x}$ , compatible meaning every $I_{1} + λ I_{0}$ is Hamiltonian. The recursion operator is the ratio $R = I_{1} I_{0}^{- 1} = (D^{3} + 4 u D + 2 u_{x}) D^{- 1} = D^{2} + 4 u + 2 u_{x} D^{- 1}$ , recovering the KdV operator. The recursion identity $R_{t} = [K^{'}, R]$ is then a consequence of the two Hamiltonian operators being compatible and the Hamiltonian flows being symmetries: heredity of $R$ , which is what upgrades " $R$ maps one symmetry to the next" to " $R^{n}$ generates the whole abelian tower," is equivalent to the Nijenhuis-torsion vanishing, equivalent in turn to compatibility of the pencil. So the operator-theoretic condition of this unit and the geometric compatibility condition of the bi-Hamiltonian theory are two faces of the same fact. Rubric: full credit for $R = I_{1} I_{0}^{- 1}$ with the explicit operators, and the compatibility-heredity-recursion correspondence.

Lean formalization Intermediate+

Mathlib contains neither the differential algebra of differential functions on the infinite jet bundle, nor generalised vector fields, nor the formal integro-differential operators with the antiderivative $D^{- 1}$ that a recursion operator requires. The unit is formalisation-free; a meaningful Lean statement needs the infrastructure documented in Mathlib gap analysis. The aspirational target statement has the schematic form:

-- Aspirational, not currently realisable in Mathlib.
theorem recursion_maps_symmetries
    (K : Characteristic)                 -- the evolution u_t = K
    (R : IntegroDiffOperator)            -- formal operator built from D, ·, D⁻¹
    (hR : R.timeDeriv K = lieBracket (frechet K) R)   -- R_t = [K', R]
    (Q : Characteristic)
    (hQ : isSymmetry K Q) :              -- [K, Q] = 0
    isSymmetry K (R.apply Q) := by
  sorry

The statement needs Characteristic (differential functions modulo null characteristics), frechet K (the Fréchet derivative $K^{'}$ as a linear differential operator), isSymmetry K Q (the bracket condition $[K, Q] = 0$ ), and IntegroDiffOperator with a timeDeriv and a Lie bracket against $K^{'}$ . The closest existing Mathlib infrastructure is Derivation and LieAlgebra, but neither the jet-space differential algebra nor the pseudo-differential extension with $D^{- 1}$ exists. Tracked as a long-horizon contribution roadmap; the human reviewer is the correctness gate for the prolongation and commutator computations.

Advanced results Master

The generalised-symmetry framework converts the question "is this equation integrable?" into "does it possess an infinite hierarchy of higher symmetries?", and the recursion operator is the device that produces the hierarchy. Two worked systems and the structural results frame the theory; the exposition follows Olver ^{[Olver §5.2]} and the symmetry approach to integrability of Fokas ^{[Fokas 1987]}.

The KdV hierarchy from the recursion operator. For $u_{t} = u_{xxx} + 6 u u_{x}$ the operator $R = D^{2} + 4 u + 2 u_{x} D^{- 1}$ satisfies $R_{t} = [K^{'}, R]$ , where $K^{'} = D^{3} + 6 u D + 6 u_{x}$ is the Fréchet derivative. Seeding with the translation characteristic $K_{0} = u_{x}$ , the iterates $K_{n} = R^{n} u_{x}$ are higher symmetries by the symmetry-preservation theorem: $K_{1} = u_{xxx} + 6 u u_{x}$ (KdV itself), $K_{2} = u_{5 x} + 10 u u_{xxx} + 20 u_{x} u_{xx} + 30 u^{2} u_{x}$ (the fifth-order flow), and onward. Each $K_{n}$ generates a commuting evolution $u_{t_{n}} = K_{n}$ , and the family ${K_{n}}$ is involutive, $[K_{m}, K_{n}] = 0$ , the defining feature of an integrable hierarchy. The local conserved densities of KdV are produced alongside, one for each rung.

The Burgers hierarchy and linearisability. Burgers $u_{t} = u_{xx} + 2 u u_{x}$ carries $R = D + u + u_{x} D^{- 1}$ , a first-order recursion operator. The hierarchy $K_{n} = R^{n} u_{x}$ is comparatively simple because the Cole-Hopf transform $u = (lo g ψ)_{x}$ linearises Burgers to the heat equation $ψ_{t} = ψ_{xx}$ , under which $R$ pulls back to the bare $D$ acting on $ψ$ . The higher symmetries of Burgers are the images of the elementary heat-equation symmetries $\partial_{x}^{n} ψ$ . The contrast with KdV is structural: KdV is integrable but genuinely nonlinear, with a nonlocal recursion operator and a hierarchy that resists linearisation, while Burgers is C-integrable (linearisable by a change of variables), with an essentially simpler hierarchy whose recursion operator becomes local after that change.

Equivalence and the structure of the symmetry algebra. The characteristics form a Lie algebra under $[P, Q] = pr v_{P} (Q) - pr v_{Q} (P)$ , and the symmetries of $u_{t} = K$ are the centraliser of $K$ , a subalgebra by the Jacobi identity. For an integrable equation this centraliser is infinite-dimensional, spanned by the hierarchy ${R^{n} K_{0}}$ together with the master symmetries ${R^{m} τ_{0}}$ of 05.09.11; the higher symmetries form an abelian ideal on which the master symmetries act as a centerless Virasoro algebra. Generalised symmetries here are taken as elements of the quotient of characteristics by the null ones, within which each generalised vector field is equivalent to its evolutionary representative $v_{Q}$ .

The symmetry test for integrability. Ibragimov and Shabat ^{[Ibragimov-Shabat 1980]} turned the existence of higher symmetries into a classification tool: an evolution equation $u_{t} = K (u, u_{x}, \dots, u_{n})$ admits a genuine Lie-Bäcklund symmetry of high order only under stringent constraints on $K$ , and these constraints, made algorithmic through the formal symmetry / canonical-density method of Mikhailov-Shabat-Sokolov, classify the integrable scalar evolution equations. The existence of a single higher symmetry of sufficiently high order forces the existence of the entire hierarchy and of a recursion operator, so "possessing one genuine generalised symmetry" is, for a broad class, equivalent to integrability.

Synthesis. The generalised-symmetry theory reduces integrability to a single structural feature, an infinite hierarchy of commuting higher symmetries, and the recursion operator is the operator that manufactures the hierarchy from one seed. The argument runs along four linked stages, each carried by the total derivative of 05.05.05. Generalised symmetries are defined as characteristics $Q$ on the jet bundle with $pr v_{Q}$ tangent to the equation, taken modulo null characteristics, so that the evolutionary representative $v_{Q}$ is the canonical object. The symmetry criterion $[K, Q] = 0$ is rewritten through the Fréchet derivative as $\partial_{t} Q = K^{'} [Q]$ , the linearised-flow equation. A recursion operator is then exactly an operator $R$ intertwining the linearised flow with itself, $R_{t} = [K^{'}, R]$ , and the central computation shows this intertwining propagates the symmetry equation from $Q$ to $R Q$ . Iterating from the translation seed $u_{x}$ generates the abelian tower $R^{n} u_{x}$ , which is the integrable hierarchy, the same tower the master symmetry of 05.09.11 generates by bracketing and the bi-Hamiltonian pencil generates by $R = I_{1} I_{0}^{- 1}$ . The KdV and Burgers hierarchies instantiate every stage, and the same operator $D_{i}$ that built the jet-bundle contact structure here defines the Fréchet derivative, the recursion operator, and the characteristic bracket.

Full proof set Master

Proposition (Fréchet-derivative form of the symmetry criterion). For an autonomous evolution equation $u_{t} = K$ , a $t$ -independent characteristic $Q$ is a generalised symmetry if and only if $pr v_{K} (Q) = K^{'} [Q]$ , equivalently $[K, Q] = 0$ where $[K, Q] = K^{'} [Q] - Q^{'} [K]$ .

Proof. The evolutionary field $v_{Q}$ is a symmetry when its flow commutes with the flow of $v_{K}$ , equivalently when the Lie bracket of the prolonged evolutionary fields vanishes: $[pr v_{K}, pr v_{Q}] = 0$ . The bracket of two evolutionary fields is again evolutionary with characteristic $[K, Q] = pr v_{K} (Q) - pr v_{Q} (K)$ ; this is the standard identity that the assignment $Q \mapsto pr v_{Q}$ is a Lie-algebra anti-homomorphism from $(A^{q}, [\cdot, \cdot])$ to vector fields. Now $pr v_{K} (Q) = \sum_{J} (D_{J} K^{α}) \partial_{u_{J}^{α}} Q = K^{'} [Q]$ by the chain rule on $A$ , the Fréchet derivative being $K^{'} [\cdot] = \sum_{J} (\partial_{u_{J}} K) D_{J} (\cdot)$ evaluated in the direction of its argument; symmetrically $pr v_{Q} (K) = Q^{'} [K]$ . Hence $[K, Q] = K^{'} [Q] - Q^{'} [K]$ , and the symmetry condition $[K, Q] = 0$ reads $K^{'} [Q] = Q^{'} [K]$ . Along the flow $\partial_{t} u = K$ , the total time derivative of $Q$ is $\frac{d}{d t} Q = Q^{'} [K] = pr v_{Q} (K)$ ; combined with $[K, Q] = 0$ this gives $\frac{d}{d t} Q = K^{'} [Q]$ , the linearised-symmetry equation. $□$

Proposition (the recursion identity is intertwining of the linearised flow). An autonomous operator $R$ on characteristics satisfies the recursion condition $pr v_{K} (R) = [K^{'}, R]$ if and only if $R$ maps the kernel of the linearised-symmetry operator $L := \frac{d}{d t} - K^{'}$ into itself; consequently $R$ maps symmetries to symmetries.

Proof. Let $L = \frac{d}{d t} - K^{'}$ act on $t$ -dependent characteristics, so that the symmetries are the $t$ -independent elements of $ker L$ along the flow. Compute the commutator $[L, R]$ acting on a characteristic $Q$ : $$ [L, R]Q = L(RQ) - R(LQ) = \Big(\frac{d}{dt}(RQ) - K'[RQ]\Big) - R\Big(\frac{d}{dt}Q - K'[Q]\Big). $$ Expand $\frac{d}{d t} (R Q) = R_{t} Q + R \frac{d}{d t} Q$ with $R_{t} = pr v_{K} (R)$ : $$ [L, R]Q = R_t Q + R\tfrac{d}{dt}Q - K'[RQ] - R\tfrac{d}{dt}Q + R,K'[Q] = R_t Q - K'[RQ] + R,K'[Q] = \big(R_t - [K', R]\big)Q. $$ Therefore $[L, R] = R_{t} - [K^{'}, R]$ as operators. The recursion condition $R_{t} = [K^{'}, R]$ holds if and only if $[L, R] = 0$ , that is $R$ commutes with $L$ . A commuting operator preserves $ker L$ : if $L Q = 0$ then $L (R Q) = R (L Q) = 0$ . Restricting to $t$ -independent elements, $R$ maps symmetries to symmetries. $□$

Proposition (the KdV operator $R = D^{2} + 4 u + 2 u_{x} D^{- 1}$ satisfies the recursion identity). For $K = u_{xxx} + 6 u u_{x}$ with $K^{'} = D^{3} + 6 u D + 6 u_{x}$ , the operator $R = D^{2} + 4 u + 2 u_{x} D^{- 1}$ satisfies $R_{t} = [K^{'}, R]$ , where $R_{t} = 4 u_{t} + 2 u_{x t} D^{- 1}$ is the coefficient evolution along the KdV flow.

Proof. Along the flow, $u_{t} = K = u_{xxx} + 6 u u_{x}$ and $u_{x t} = D K = u_{4 x} + 6 u_{x}^{2} + 6 u u_{xx}$ , so $$ R_t = 4 u_t + 2 u_{xt} D^{-1} = 4(u_{xxx} + 6 u u_x) + 2(u_{4x} + 6 u_x^2 + 6 u u_{xx}) D^{-1}. $$ For the right side, expand $[K^{'}, R] = K^{'} R - R K^{'}$ with $K^{'} = D^{3} + 6 u D + 6 u_{x}$ and $R = D^{2} + 4 u + 2 u_{x} D^{- 1}$ , using the operator product rule $D^{k} \cdot f = \sum_{j} (j k) (D^{j} f) D^{k - j}$ for multiplication operators $f$ and the relations $D D^{- 1} = D^{- 1} D = 1$ , $D^{- 1} f = f D^{- 1} - (D f) D^{- 2} + \dots$ in the formal pseudo-differential algebra. Collecting terms by differential order, the purely differential parts of $K^{'} R$ and $R K^{'}$ cancel through order $D^{5}$ , the multiplication-operator residue is $4 u_{xxx} + 24 u u_{x} = 4 u_{t}$ , and the nonlocal residue multiplying $D^{- 1}$ is $2 u_{4 x} + 12 u_{x}^{2} + 12 u u_{xx} = 2 u_{x t}$ , matching $R_{t}$ term by term. The two sides agree, $R_{t} = [K^{'}, R]$ , so $R$ is a recursion operator for KdV. The computation is the explicit verification underlying Olver's general result ^{[Olver §5.2]}. $□$

Connections Master

The prolongation calculus of 05.05.06 is the substrate: the evolutionary field $pr v_{Q} = \sum_{J} (D_{J} Q^{α}) \partial_{u_{J}^{α}}$ , the characteristic $Q^{α} = ϕ^{α} - ξ^{i} u_{i}^{α}$ , and the tangency criterion are the point-symmetry constructions extended to characteristics of unbounded order. A generalised symmetry is what Lie's infinitesimal criterion becomes when the coefficient functions are allowed to read the whole jet.

The recursion operator proved here to map symmetries to symmetries is exactly the operator iterated in 05.09.11: the master symmetry $τ$ satisfies $L_{τ} R = R$ , and the hierarchy $K_{n} = R^{n} K_{0}$ it bracketing-generates is the abelian tower of higher symmetries this unit produces directly from the recursion identity $R_{t} = [K^{'}, R]$ . The two units describe the same operator from the symmetry-preservation side and the hierarchy-generation side.

The infinite hierarchy of commuting symmetries is the defining datum of an integrable system 05.02.03 in the infinite-dimensional setting: the higher symmetries $K_{n}$ are the commuting Hamiltonian flows and their conserved densities are the involutive integrals, the field-theoretic analogue of Liouville-Arnold integrability.

The bi-Hamiltonian factorisation $R = I_{1} I_{0}^{- 1}$ and the heredity (Nijenhuis) condition that upgrades single symmetry-preservation to involutivity of the whole tower are developed in 05.09.11 and surface in the finite-gap integration of 05.09.09, where the stationary higher symmetries cut out the algebro-geometric solutions and the recursion operator's spectrum organises the spectral curve.

Historical & philosophical context Master

The notion that a transformation of a differential equation may legitimately depend on derivatives of any order originates with Albert Victor Bäcklund, whose 1880 study of surface transformations ^{[Bäcklund 1880]} (Mathematische Annalen 9) introduced transformations that mix a function with its first and higher derivatives, the higher-order tangent transformations now called Lie-Bäcklund transformations. Sophus Lie had shown that genuine contact transformations of finite-dimensional jet space involve at most first derivatives; Bäcklund's contribution was to recognise that admitting all orders opens a vastly larger class, at the cost of working on the infinite jet bundle. Emmy Noether's 1918 Invariante Variationsprobleme ^{[Noether 1918]} (Göttinger Nachrichten) already worked with symmetries in characteristic (evolutionary) form, pairing them with conservation laws, and her formalism is the one in which generalised symmetries are most naturally expressed.

The modern theory crystallised in the 1970s with the discovery that the Korteweg-de Vries equation possesses infinitely many symmetries. Peter Olver's 1977 paper Evolution equations possessing infinitely many symmetries ^{[Olver 1977]} (J. Math. Phys. 18) introduced the recursion operator $R = D^{2} + 4 u + 2 u_{x} D^{- 1}$ as the explicit generator of the KdV hierarchy of higher symmetries and gave the operator condition characterising such recursion operators, later canonised in his 1986 / 1993 Applications of Lie Groups to Differential Equations. Nail Ibragimov and Alexey Shabat's 1980 work ^{[Ibragimov-Shabat 1980]} (Funct. Anal. Appl. 14) turned the existence of higher Lie-Bäcklund symmetries into a classification criterion for integrable evolution equations, the seed of the symmetry approach to integrability developed algorithmically by Mikhailov, Shabat, and Sokolov. Athanassios Fokas's 1987 survey Symmetries and integrability ^{[Fokas 1987]} (Stud. Appl. Math. 77) synthesised the recursion-operator, bi-Hamiltonian, and symmetry-classification viewpoints, establishing that for a wide class of scalar evolution equations the existence of a single genuine higher symmetry is equivalent to integrability.

Bibliography Master

@article{Backlund1880,
  author  = {B{\"a}cklund, Albert Victor},
  title   = {{\"U}ber Fl{\"a}chentransformationen},
  journal = {Mathematische Annalen},
  volume  = {9},
  year    = {1880},
  pages   = {297--320}
}

@article{Noether1918,
  author  = {Noether, Emmy},
  title   = {Invariante Variationsprobleme},
  journal = {Nachrichten von der Gesellschaft der Wissenschaften zu G{\"o}ttingen, Mathematisch-Physikalische Klasse},
  year    = {1918},
  pages   = {235--257}
}

@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}
}

@article{IbragimovShabat1980,
  author  = {Ibragimov, Nail H. and Shabat, Alexey B.},
  title   = {Evolutionary equations with a nontrivial {L}ie--{B}{\"a}cklund group},
  journal = {Functional Analysis and its Applications},
  volume  = {14},
  year    = {1980},
  pages   = {19--28}
}

@article{Fokas1987,
  author  = {Fokas, Athanassios S.},
  title   = {Symmetries and integrability},
  journal = {Studies in Applied Mathematics},
  volume  = {77},
  year    = {1987},
  pages   = {253--299}
}

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

@incollection{MikhailovShabatSokolov1991,
  author    = {Mikhailov, Alexander V. and Shabat, Alexey B. and Sokolov, Vladimir V.},
  title     = {The symmetry approach to classification of integrable equations},
  booktitle = {What is Integrability?},
  editor    = {Zakharov, V. E.},
  publisher = {Springer},
  year      = {1991},
  pages     = {115--184}
}

@book{Ibragimov1985,
  author    = {Ibragimov, Nail H.},
  title     = {Transformation Groups Applied to Mathematical Physics},
  publisher = {Reidel},
  year      = {1985}
}

Prerequisites

05.05.06
05.05.05
05.09.11

Tier anchors

beginner: Olver *Applications of Lie Groups to Differential Equations* §5.1 informal (symmetries allowed to depend on slopes and curvatures, not only positions)
intermediate: Olver §5.1-§5.2 (generalised vector fields, evolutionary representatives, equivalence modulo trivial characteristics, recursion operators); Ibragimov *Transformation Groups Applied to Mathematical Physics* (1985)
master: Olver §5.1-§5.2-§5.3; Olver 1977 (J. Math. Phys. 18); Fokas 1987 (Stud. Appl. Math. 77); Mikhailov-Shabat-Sokolov in *What is Integrability?* (1991)

References

Olver — Applications of Lie Groups to Differential Equations (2nd ed., 1993) · §5.1 generalised symmetries and evolutionary form; §5.2 recursion operators and the KdV hierarchy
Bäcklund — Über Flächentransformationen (1880) · Mathematische Annalen 9, higher-order tangent (Lie-Bäcklund) transformations
Olver 1977 — Evolution equations possessing infinitely many symmetries · Journal of Mathematical Physics 18, recursion operators and the KdV symmetry hierarchy
Ibragimov, Shabat 1980 — Evolutionary equations with a nontrivial Lie-Bäcklund group · Functional Analysis and its Applications 14, classification by higher symmetries
Fokas 1987 — Symmetries and integrability · Studies in Applied Mathematics 77, recursion operators and the symmetry approach to integrability
Noether 1918 — Invariante Variationsprobleme · Göttinger Nachrichten, symmetries in characteristic (evolutionary) form

Estimated time

beginner: 17m
intermediate: 46m
master: 82m