An exact Lagrangian that cannot be untied by Hamiltonian motion

← All digests

Family 340 · Differential geometry
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-26

1. The question

Classical mechanics describes a system by positions and momenta. If positions form a manifold $Q$, its phase space is the cotangent bundle $T^*Q$. The simplest distinguished submanifold consists of all positions with zero momentum: the zero section.

Many other submanifolds can look locally like the zero section from the viewpoint of symplectic geometry. Arnold’s nearby Lagrangian conjecture asks whether every closed exact Lagrangian in this phase space can be obtained from the zero section by a Hamiltonian motion.

The manuscript in family 340 claims a counterexample. Its Lagrangian has exactly the same abstract smooth manifold as the base, yet cannot be moved to the zero section by a Hamiltonian isotopy. The proposed obstruction lives in how the manifold is placed in phase space, beyond its ordinary topology.

The word “nearby” is historical terminology. The unrestricted conjecture concerns every closed exact embedded Lagrangian in a cotangent bundle; it does not assume that the submanifold is a small graph or lies close to the zero section in a specified norm.

2. Background and essential definitions

Positions, momenta, and symplectic area

A point of $T^*Q$ is a pair $(q,p)$, where $q$ is a position and $p$ is a linear functional on tangent vectors at $q$. In local coordinates, the canonical one-form and the paper’s symplectic form are

$$\lambda=\sum_i p_i\,dq_i, \qquad \omega=d\lambda=\sum_i dp_i\wedge dq_i.$$

A symplectic form is a closed, nondegenerate differential two-form. It measures oriented area in position–momentum directions. The sign convention for $\omega$ varies between texts; the manuscript uses the one displayed here.

If $Q$ has dimension $n$, a Lagrangian submanifold $L\subset T^*Q$ has dimension $n$ and satisfies $\omega|_L=0$. It is exact when there is a globally defined function $f:L\to\mathbb R$ such that

$$\lambda|_L=df.$$

Exactness is a global restriction, stronger than the Lagrangian condition. Equivalently, the integral of $\lambda$ around every closed curve in $L$ vanishes. “Closed” here means compact without boundary, and “embedded” excludes self-intersections.

The allowed way to move a Lagrangian

A time-dependent Hamiltonian $H_t$ determines a vector field through

$$\iota_{X_{H_t}}\omega=-dH_t.$$

Its flow is a Hamiltonian isotopy. These motions preserve symplectic structure and form the natural class of deformations in the conjecture. The paper takes the Hamiltonians to have support in one fixed compact subset. For a compact isotopy trace, a cutoff makes this convention harmless.

There are easy positive examples. If $u:Q\to\mathbb R$ is smooth, its differential defines the graph

$$L_u=\{(q,du(q)):q\in Q\}.$$

It is exact Lagrangian and Hamiltonian isotopic to the zero section. The difficulty is that a general embedded exact Lagrangian need not project one-to-one onto $Q$. Its projection can fold, although the full submanifold in phase space remains embedded.

Being diffeomorphic means being the same abstract smooth manifold. Being Hamiltonian isotopic is much stronger: it specifies an allowed ambient motion of the embedding. Confusing these two notions would erase the point of the claimed counterexample.

3. Development history and previous partial results

Early intersection theorems by Hofer and Laudenbach–Sikorav constrained Hamiltonian deformations of the zero section. Generating functions made these constraints accessible through finite-dimensional critical-point theory: intersections become critical points of auxiliary functions. Chaperon, Sikorav, and others developed methods for transporting generating functions under Hamiltonian motion. Hofer, 1985, Laudenbach–Sikorav, 1985.

There are positive low-dimensional isotopy results. Hind’s work treats Lagrangian spheres in $T^*S^2$, and work of Dimitroglou Rizell, Goodman, and Ivrii gives the cotangent-torus theorem. These illustrate real rigidity in specific bases, rather than evidence that can simply be extrapolated to arbitrary dimension. Hind, Dimitroglou Rizell–Goodman–Ivrii.

Meanwhile, the topology of the projection $L\to Q$ became increasingly constrained. Work of Fukaya–Seidel–Smith and Nadler established cohomological and categorical restrictions under additional hypotheses. Abouzaid proved a homotopy-equivalence theorem under a Maslov-class assumption, and Kragh’s work with Abouzaid’s appendix removed that assumption. Abouzaid–Kragh then proved that the projection is a simple homotopy equivalence, a refinement controlling an algebraic obstruction to elementary topological simplification. Abouzaid–Kragh, 2018.

These results concern the projection’s topology, not the full Hamiltonian isotopy class of the embedding. More recent work on twisted generating functions, normal invariants, and Floer homotopy theory imposed still finer necessary conditions. None of those conditions alone identifies the allowed ambient motion.

A particularly direct predecessor is Álvarez-Gavela–Igusa–Sullivan’s work on higher torsion. They constructed Legendrian submanifolds in the one-jet space $J^1Q=T^*Q\times\mathbb R$ with nontrivial tube data even when their projections to the base are homotopic to diffeomorphisms. However, forgetting the extra real coordinate can introduce self-intersections in $T^*Q$. The new manuscript’s essential additional claim is an embedded exact Lagrangian construction that keeps a stable tube obstruction. Version 2 of that predecessor.

4. What the new manuscript claims

For some sufficiently large even integer $N$, set

$$Q=S^9\times S^{N-1}.$$

The manuscript asserts the existence of a closed exact smoothly embedded Lagrangian $L\subset T^*Q$ such that

$$L\cong Q,$$

where $\cong$ denotes diffeomorphism, but no compactly supported Hamiltonian isotopy carries the zero section to $L$.

Both $Q$ and $L$ are connected and simply connected. Thus the obstruction is not an exotic fundamental group or a mismatch of abstract smooth manifolds. The result is an existential high-dimensional example: the theorem does not specify a minimal numerical dimension or assert the construction in every dimension.

If correct, one such example disproves the unrestricted nearby Lagrangian conjecture. It leaves positive theorems for particular bases intact and does not automatically settle variants with additional restrictions.

The construction also distinguishes two roles of a parameter space. Here $S^9$ parametrizes the behavior at infinity of functions within one generating family. It is not a parameter indexing a collection of whole Lagrangians in one fixed cotangent bundle. The paper emphasizes this distinction when comparing its obstruction with vanishing theorems for parametrized Whitehead torsion of families of nearby Lagrangians.

5. Why the advance matters

Importance: 5/5. The nearby Lagrangian conjecture is a central question about rigidity in cotangent bundles. A counterexample satisfying exactness, embeddedness, and equality of smooth type would change the expected relationship between topological classification and Hamiltonian classification.

The result would show that increasingly strong information about the projection can coexist with a remaining symplectic obstruction. Knowing that the Lagrangian has the right homotopy type—or even the right diffeomorphism type—would still not determine how it can move in phase space.

Its proposed mechanism also links generating functions to algebraic K-theory of spaces and smooth families of cobordisms. These tools detect information that simpler invariants discard. The significance of the claim therefore includes both the negative answer and the kind of invariant used to obtain it.

This rating concerns the stated mathematical consequences if the construction works as claimed.

6. The main idea

Encode a Lagrangian by critical points

Introduce auxiliary variables $w\in\mathbb R^r$ and a smooth function $F(q,w)$. Its fiber-critical set is defined by

$$d_wF(q,w)=0.$$

A critical point is sent to the phase-space point

$$(q,w)\longmapsto(q,d_qF(q,w)).$$

With the required regularity of the full critical equation, this gives an exact Lagrangian immersion. The value of $F$ on the critical set supplies its primitive. Regularity involves derivatives in both $q$ and $w$; it does not require every fiber Hessian to be invertible. Embeddedness requires a separate global argument.

Put a nontrivial family at infinity

Far out in the auxiliary variables, the generating functions agree with functions homogeneous of degree two:

$$g_b(rw)=r^2g_b(w),\qquad b\in S^9.$$

On the unit sphere, each negative domain resembles the negative domain of a nondegenerate quadratic form. The paper calls such a domain a smooth tube. Adding positive and negative quadratic variables gives a stabilization operation, allowing tube families of different dimensions to be compared.

Let $\mathcal T$ denote the resulting stable space of tubes. Its elements retain smooth family information. There is also a forgetful map $\gamma:\mathcal T\to BG$ recording only an associated stable spherical fibration; $BG$ is the classifying space for these fibrations.

Using Waldhausen’s tube description and Rognes’s calculations, the paper selects a homotopy class

$$0\ne a\in\pi_9(\mathcal T),\qquad \gamma_*a=0.$$

Here $\pi_9$ records families parametrized by a nine-sphere up to deformation. The family remains nontrivial as smooth tube data even though its coarser spherical-fibration data become trivial.

Separate branches in phase space

The extra sphere factor $S^{N-1}$ provides a variable $v$. A suitable coupling arranges that the critical locus obeys

$$\nabla g_b(w)=R(b)v,$$

where $R(b)>0$ is a chosen scale. Homogeneity parametrizes this locus by the product of spheres, giving the desired abstract smooth type.

The hard geometric issue is whether two critical points can produce the same covector. Equality of the momenta in the sphere directions forces their remaining difference into a specific line. The variation of $R$ over $S^9$ is then used to distinguish them through their base-direction momenta. Near critical points of that variation, the tube family is chosen quadratic, where the relevant critical equation has a unique solution.

Thus the manuscript separately addresses both requirements: a regular critical locus produces an immersion, and the momentum-separation argument upgrades it to an embedding.

Why a Hamiltonian untangling would contradict the tube class

Suppose a Hamiltonian isotopy carried the constructed Lagrangian to the zero section. The paper transports its generating family through finitely many small Hamiltonian steps. Each step adds standard split quadratic variables while preserving the stable tube class at infinity.

At the zero section, the critical locus projects diffeomorphically to the base. Consequently, each fiber would have exactly one nondegenerate critical point. Restricting to a fixed sphere-coordinate slice leaves a family over $S^9$ with the original nontrivial class at infinity.

The manuscript establishes that this situation is impossible. Morse theory relates the negative directions at the sole critical point to the negative domain at infinity. Between a small inner sphere and a large outer sphere, regular level sets form a family of cobordisms. Stabilization and smooth-family classification then force the tube family to be trivial, contradicting the choice of $a$.

The obstruction therefore survives Hamiltonian motion but is incompatible with the simplest possible critical-point structure at the zero section.

7. Scope, limitations, and open questions

The result is a high-dimensional existence statement. It does not identify the lowest dimension permitting a counterexample, classify all exact Lagrangians in the example’s cotangent bundle, or settle the conjecture for every other base.

It also does not claim that $L$ fails to be smoothly isotopic to the zero section under unrestricted ambient deformations. The stated obstruction is to Hamiltonian isotopy. Smooth, symplectic, and Hamiltonian equivalence are different questions and should not be interchanged.

Exactness and embeddedness are crucial. An immersed example with double points, or a nonexact Lagrangian, would not disprove the conjecture under discussion. The manuscript explicitly supplies separate steps for the critical equation, branch separation, and preservation of the obstruction under Hamiltonian transport.

Natural further questions concern which bases retain the conjectured rigidity, which stable tube classes can be realized by embeddings, and whether the construction can be made effective in a smaller dimension. This digest reports the claimed construction and its dependencies without checking the proof or independently assessing novelty.

8. Sources and further reading

The primary predecessors linked above explain intersection theory, positive low-dimensional cases, projection rigidity, and higher tube torsion. Optional background is available under cotangent bundle, Lagrangian submanifold, and Morse theory.

Source manuscripts

• M1. OpenAI, A counterexample to the nearby Lagrangian conjecture (September 23, 2026).

Sources consulted on 26 October 2026 at repository revision 1581d4c. Download the manuscript citations.