A solution to Banach's isometric conjecture

Abstract: Banach asked in 1932 whether a real Banach space $X$ whose $n$-dimensional subspaces, for some fixed $1<n<\dim X$, are all isometric must be a Hilbert space.Gromov proved the conjecture for even $n$, and subsequent work settled several odd-dimensional cases. We prove the conjecture for every odd $n$, including all previously unresolved cases. Together with Gromov's even-dimensional result, this completes Banach's isometric conjecture in the real case. The proof combines bundle topology with Brouwer degree theory.

• The paper proves Banach’s isometric conjecture for every odd-dimensional subspace parameter in real Banach spaces, completing the real case together with Gromov’s even-dimensional result.
• The authors reduce the problem to a hyperplane theorem and use principal-bundle topology, finite homotopy groups, positive-degree sphere maps, and Lipschitz global sections to construct exact families of section isometries.
• A Brouwer degree argument shows that powers of the model norm are homogeneous polynomials, and unique factorization forces its square to be a positive-definite quadratic form, proving the relevant convex bodies are ellipsoids and the spaces are Hilbert.

The problem and its history

In 1932 Banach asked whether a real or complex Banach space $X$ in which, for some fixed $2 \le n < \dim X$, all $n$-dimensional linear subspaces are linearly isometric must be a Hilbert space. The affirmative answer is known as Banach's isometric conjecture. The problem reduces to codimension one: if the conjecture holds whenever $\dim X = n+1$, then the parallelogram identity holds on every pair of vectors of $X$, and $X$ is Hilbert by the Jordan–von Neumann theorem.

The historical record is as follows. Auerbach, Mazur and Ulam settled the real case $n=2$ [AMU1935]. Dvoretzky's theorem implies the conjecture for every $n \ge 2$ when $X$ is infinite-dimensional and real, and Milman extended this to complex spaces [Dvoretzky1959, Milman1971]. Gromov proved the conjecture for every even $n$ over both fields, and for odd $2 \le n < \dim X$0 when $2 \le n < \dim X$1 (real) or $2 \le n < \dim X$2 (complex) [Gromov1967]. Bor, Hernández-Lamoneda, Jiménez-Desantiago and Montejano handled odd $2 \le n < \dim X$3, $2 \le n < \dim X$4, with a possible exception at $2 \le n < \dim X$5 [BorEtAl2021]; Ivanov, Mamaev and Nordskova settled $2 \le n < \dim X$6, $2 \le n < \dim X$7 [IMN2023]; Bracho and Montejano treated the complex case $2 \le n < \dim X$8 [BrachoMontejano2021].

The paper under discussion closes the remaining gap: it proves the conjecture for every odd $2 \le n < \dim X$9 in the real case, thereby completing the conjecture together with Gromov's even-dimensional theorem. The main results are:

• Banach space form: if a real Banach space has all $n$0-dimensional subspaces isometric for some $n$1, then $n$2 is a Hilbert space.
• Convex-geometric form: an origin-symmetric convex body $n$3 whose $n$4-dimensional central sections are all linearly equivalent ($n$5) is an ellipsoid.
• Hyperplane theorem (the technical core): for odd $n$6, an origin-symmetric convex body $n$7 whose central hyperplane sections are all linearly equivalent is an ellipsoid.

Relation to prior bundle-theoretic approaches

The proof shares its starting point with Gromov and with Ivanov–Mamaev–Nordskova. Fixing one model section $n$8, the exact linear maps from $n$9 onto the sections $\dim X = n+1$0 form a principal $\dim X = n+1$1-bundle over $\dim X = n+1$2. Earlier global arguments used such bundles to constrain the symmetry group $\dim X = n+1$3 itself; in particular, Bor et al. combined structure-group reduction results with representation theory to force $\dim X = n+1$4 to be an affine body of revolution, a strategy that requires detailed classification of compact group actions.

This paper takes a different route: it neither classifies $\dim X = n+1$5 nor produces additional symmetries. Instead, after reducing the structure group to the identity component $\dim X = n+1$6, the bundle class lies in $\dim X = n+1$7, which is finite because $\dim X = n+1$8 is odd. Pullback by a self-map $\dim X = n+1$9 of suitable positive degree annihilates this class, yielding a global Lipschitz family $X$0 of exact section isometries. The ambient body then supplies the link to algebra via Brouwer degree theory.

Finite-moment coordinates and the local isometry bundle

The construction begins by normalizing the model section so that its covariance operator satisfies $X$1. Two lemmas make finite moment data detect symmetries exactly:

• Finite detection: since $X$2 is compact (a closed subgroup of $X$3 once moments are normalized), Noetherianity of the polynomial ring of matrix entries shows that finitely many normalized moment tensors $X$4 have stabilizer exactly $X$5. The argument uses Stone–Weierstrass to pass from equality of all polynomial integrals to equality of measures, hence of supports.
• Orbit coordinates: the orbit map $X$6 is a diffeomorphism onto a compact embedded submanifold, with locally Lipschitz inverse.

These coordinates convert pointwise equivalences into regularity. For each $X$7, applying a Euclidean isometry $X$8 and whitening by $X$9 produces bodies orthogonally equivalent to $X$0; Lipschitz dependence of radial functions (via standard gauge/radial estimates) propagates through moments and square roots, so the orbit map yields Lipschitz local representatives $X$1 with $X$2. Composing gives local exact maps $X$3 satisfying $X$4 and $X$5, and the transition functions $X$6 are globally Lipschitz on overlaps. Thus $X$7 is a principal $X$8-bundle with a Lipschitz atlas.

Globalization: killing the obstruction by degree

Reducing to $X$9 uses that $n=2$0 is finite, so $n=2$1 is a finite covering; simple connectivity of $n=2$2 lets one component define a Lipschitz principal $n=2$3-subbundle. Since $n=2$4 is positive and even, Serre's finiteness theorem makes $n=2$5 finite, so the classifying class $n=2$6 has finite order $n=2$7. Precomposition by a degree-$n=2$8 self-map $n=2$9 of $n \ge 2$0 multiplies the class by $n \ge 2$1, trivializing the pullback.

Topological triviality alone is insufficient—the degree argument requires Lipschitz regularity. The paper proves a general upgrade: a topologically trivial principal bundle with Lipschitz atlas over a compact smooth manifold admits a global Lipschitz section. The proof approximates a continuous section inside the associated vector bundle $n \ge 2$2 using a partition of unity, then retracts equivariantly onto the orbit subbundle via a Lipschitz tubular-neighborhood retraction onto $n \ge 2$3. The outcome is the key structural result: there exist $n \ge 2$4, a smooth degree-$n \ge 2$5 map $n \ge 2$6, and a Lipschitz family $n \ge 2$7 with

$n \ge 2$8

Exactness is preserved throughout: no approximation leaves the isometry bundle.

The degree argument and polynomial rigidity

Fixing $n \ge 2$9 with $X$0, the map $X$1 lands in $X$2, and since $X$3, an explicit geodesic homotopy identifies $X$4 as homotopic to $X$5, so $X$6. The radial homeomorphism $X$7 sending $X$8 to $X$9 is orientation preserving (isotopic to a dilation), so $n$0 has degree $n$1 and $n$2 has degree $n$3 into the oriented boundary $n$4.

Homogeneous extension of $n$5 to the ball yields a Lipschitz cone map $n$6 factoring as $n$7, whence $n$8 for $n$9 and zero outside $2 \le n < \dim X$00. Applying the signed degree formula (valid because $2 \le n < \dim X$01 is Lipschitz and maps the boundary sphere into the null set $2 \le n < \dim X$02) with test function $2 \le n < \dim X$03 gives

$2 \le n < \dim X$04

Since $2 \le n < \dim X$05, the integrand is a homogeneous polynomial in $2 \le n < \dim X$06 of degree $2 \le n < \dim X$07 with integrable coefficients, so the left side is a homogeneous polynomial. Consequently $2 \le n < \dim X$08 is a homogeneous polynomial for every $2 \le n < \dim X$09—a strong rigidity statement from which only $2 \le n < \dim X$10 are needed.

Setting $2 \le n < \dim X$11, the identities for $2 \le n < \dim X$12 give polynomials $2 \le n < \dim X$13 and $2 \le n < \dim X$14 satisfying $2 \le n < \dim X$15. Unique factorization in $2 \le n < \dim X$16 forces $2 \le n < \dim X$17: for each irreducible $2 \le n < \dim X$18, coprimality of $2 \le n < \dim X$19 and $2 \le n < \dim X$20 makes valuations satisfy $2 \le n < \dim X$21, $2 \le n < \dim X$22. Comparing homogeneous degrees shows the quotient $2 \le n < \dim X$23 is quadratic, and $2 \le n < \dim X$24 for $2 \le n < \dim X$25. Hence $2 \le n < \dim X$26 is a positive-definite quadratic form and $2 \le n < \dim X$27 is an ellipsoid.

Completion of the proofs

With $2 \le n < \dim X$28 an ellipsoid, every hyperplane section of $2 \le n < \dim X$29 is an ellipsoid. Since $2 \le n < \dim X$30, any two vectors lie in some $2 \le n < \dim X$31-dimensional subspace whose unit ball is such a section, so the parallelogram identity holds throughout $2 \le n < \dim X$32 and $2 \le n < \dim X$33 is an ellipsoid by Jordan–von Neumann. Lifting to arbitrary Banach spaces: for odd $2 \le n < \dim X$34, any $2 \le n < \dim X$35-dimensional subspace $2 \le n < \dim X$36 has all hyperplane sections isometric (they are $2 \le n < \dim X$37-dimensional subspaces of $2 \le n < \dim X$38), so the hyperplane theorem makes $2 \le n < \dim X$39 an ellipsoid and the parallelogram identity holds on $2 \le n < \dim X$40; arbitrariness of $2 \le n < \dim X$41 gives it on $2 \le n < \dim X$42. Even $2 \le n < \dim X$43 follows from Gromov. The convex-geometric formulation follows by viewing $2 \le n < \dim X$44 as the ambient normed space.

The paper notes precisely where parity enters: oddness of $2 \le n < \dim X$45 makes $2 \le n < \dim X$46 even, ensuring finiteness of the relevant homotopy group; and the first two polynomial identities are consecutive powers of $2 \le n < \dim X$47, enabling the divisibility argument. Notably, the degree method actually yields polynomiality of $2 \le n < \dim X$48 for all $2 \le n < \dim X$49, though only two cases are used.

Limitations and open questions

The result is confined to the real case; the complex conjecture remains open for odd $2 \le n < \dim X$50 beyond the cases covered by Gromov ($2 \le n < \dim X$51) and Bracho–Montejano ($2 \le n < \dim X$52). The proof also relies essentially on the codimension-one reduction, so the method does not directly address situations where only higher-codimension subspace information is available. Whether the degree-theoretic mechanism here can be adapted to the complex field—where the analogous homotopy groups behave differently—is a question the paper does not address. The authors also disclose that generative AI tools contributed to the approach to the globalization theorem and to initial drafts of parts of Sections 2–3, subsequently checked and rewritten by the authors, who take full responsibility for the content.

Conclusion

The paper completes Banach's isometric conjecture in the real case by proving it for all odd $2 \le n < \dim X$53, resolving the last open finite-dimensional cases left after Gromov's work. Its contribution is methodological as much as definitive: rather than classifying symmetry groups of sections, it kills the topological obstruction to a global family of exact section isometries by pullback along a positive-degree map, then extracts algebraic rigidity of the model norm from a signed-degree identity. The combination of principal-bundle topology, Lipschitz selection, and Brouwer degree theory yields a comparatively short path to a result that had resisted earlier approaches requiring detailed Lie-theoretic input.