The metric Blaschke theorem

Source and verification

This article explains family 344 of the October 6, 2026 OpenAI collection. Research conclusions are attributed to its manuscripts. The review covers statements, introductions, and selected proof mechanisms; it is not an independent verification of the complete proofs.

Overview

Suppose every geodesic on a closed Riemannian manifold minimizes for exactly the same distance, equal to the diameter. Family 344 reports that this forces the metric to be a standard compact rank-one symmetric metric, up to scale. Its proof compares the volume with the appropriate model in both directions. The lower bound comes from Jacobi fields and a weighted convexity identity; the difficult quaternionic upper bound comes from the geometry and integral topology of the space of closed geodesics. (OpenAI, 2026)

🏷️ Injectivity radius and the common cut distance

Let $(M,g)$ be a connected closed smooth Riemannian manifold of positive dimension. For $p\in M$, the exponential map sends a tangent vector $v$ to the point reached at time one by the geodesic with initial velocity $v$. The injectivity radius ${\mathrm{inj}}_p$ is the supremum of tangent-ball radii on which this map is a diffeomorphism onto its image.

The global injectivity radius and diameter are

$$\begin{array}{rl}\mathrm{inj}(M,g)& =\underset{p}{\inf }{\mathrm{inj}}_p,\\ \mathrm{diam}(M,g)& =\underset{p,q}{\sup }{d}_g(p,q).\end{array}$$

A Blaschke manifold satisfies equality of these two numbers. Denote the common positive value by $D$.

Every unit-speed geodesic segment of length $t<D$ then minimizes uniquely. Indeed, a minimizing path to its endpoint exists, has length at most $t$, and has an initial vector in the injectivity ball. Injectivity identifies that vector with the one defining the given segment. Continuity gives minimality at $t=D$. No segment longer than $D$ can minimize, because that would produce a distance larger than the diameter.

The cut locus of $p$ consists of endpoints where minimizing geodesics from $p$ cease to minimize upon extension. Here all cut times equal $D$. Classical Blaschke structure, imported by the manuscript, says that all unit-speed geodesics are simply closed with common prime period $2D$. The cut map from the unit tangent sphere at $p$ has great-sphere fibers of a fixed dimension $k$.

These are stronger conclusions than merely saying that all geodesics are closed. Conversely, a common closed-geodesic period by itself is not the hypothesis of this post.

Metric classification reported by family 344

If $\mathrm{inj}(M,g)=\mathrm{diam}(M,g)$, the manuscript proves that $(M,g)$ is isometric, after constant rescaling, to a round sphere, standard real, complex, or quaternionic projective space, or the Cayley plane. (OpenAI, 2026)

Real, complex, and quaternionic projective spaces parametrize lines over their respective scalar systems and carry standard homogeneous metrics. The Cayley plane is the exceptional compact rank-one symmetric space of real dimension sixteen. Their detailed algebraic constructions are unnecessary for the comparison argument below.

🏷️ Models and the volume reduction

Classical structure associates to $M$ a model $M_0$ through its cut-locus geometry and integral topology. This initially identifies a model type, not an isometry.

The circle, sphere-type, and real-projective-type metric cases are classical inputs. In the remaining cases, normalize $D=\pi /2$ and the prime period $T=\pi$. Write $n=\dim M$, $d=n-1$, and $m=d-k$.

Model	Real dimension $n$	Cut-fiber dimension $k$
Complex projective space of dimension $r\ge 2$	$2r$	$1$
Quaternionic projective space of dimension $r\ge 2$	$4r$	$3$
Cayley plane	$16$	$7$

The projective lines are already sphere cases. In each remaining type, $m,k>0$ and $M$ is simply connected.

On the model, the normal radial curvature operator has eigenvalues one and four with multiplicities $m$ and $k$. Solving the scalar Jacobi equations gives its polar volume density

$$w(t)=(\sin t{)}^m{\left|\frac{\sin 2t}{2}\right|}^k.$$

Consequently the model volume at the chosen scale is

$$V_0=|S^d|{\int }_0^{D}w(t)dt.$$

The manuscript’s analytic theorem is $V\ge V_0$, with equality precisely for the model metric. Thus an independently established upper bound $V\le V_0$ completes the classification. No sectional-curvature bound or curvature eigenspace decomposition is assumed for $M$.

🏷️ Jacobi fields as a symplectic family

A normal Jacobi field $J$ along a unit-speed geodesic $\gamma$ describes the first-order separation of nearby geodesics. It solves

$$J^{\prime \prime }+R(J,\dot{\gamma })\dot{\gamma }=0,$$

where primes are covariant derivatives. The initial pair $(J(0),J^{\prime }(0))$ determines the solution, so the space $\mathcal{J}$ of normal fields has dimension $2d$.

For two such fields, the Wronskian

$$\omega (I,J)=⟨I,J^{\prime }⟩-⟨I^{\prime },J⟩$$

is constant along the geodesic. Differentiation cancels the first-derivative terms, and symmetry of the curvature operator cancels the remaining terms. This makes $\mathcal{J}$ a symplectic vector space.

The fields vanishing at time $t$ form a $d$-dimensional Lagrangian subspace $L_t$: the Wronskian vanishes on it. Its motion has the positive inner product

$$G_t(I,J)=⟨I^{\prime }(t),J^{\prime }(t)⟩.$$

The cut geometry determines intersections of these subspaces. Within one period, $L_x\cap L_y$ is zero except at separation $D$, where its dimension is $k$. Moreover the entire geodesic flow has period $T$, so differentiation of that flow identity makes the Jacobi fields and their derivatives $T$-periodic. This does not assert trivial parallel transport around the closed geodesic.

Using orthonormal frames of $(L_x,G_x)$ and $(L_y,G_y)$, let $f(x,y)$ be the absolute determinant of their Wronskian pairing. It is the absolute normal Jacobi determinant. For $0<y-x<D$, it is exactly the polar volume density; no additional factor $(y-x{)}^d$ belongs in that polar formula.

🏷️ A fixed chart and two determinant factors

Use initial data $(q,p)=(J(0),J^{\prime }(0))$ as coordinates on $\mathcal{J}$. Away from times $0,D,T$, the subspaces can be written

$$\begin{array}{c}{L}_t=\{(q,A(t)q)\},\\ A(t)=A(t{)}^{\mathsf{T}},A^{\prime }(t)>0.\end{array}$$

Symmetry expresses the Lagrangian condition; positivity expresses the positive velocity form $G_t$. Normalizing the graph frames gives

$$f(x,y)=\frac{|\det (A(y)-A(x))|}{\sqrt{\det A^{\prime }(x)\det A^{\prime }(y)}}.$$

Choose the first $k$ coordinate directions from the common vanishing fields at times zero and $D$. In that one fixed basis write

$$\begin{array}{c}A=\left(\begin{array}{cc}{V}_b& C\\ C^{\mathsf{T}}& H\end{array}\right),\\ S=V_b^{\prime }-C^{\prime }(H^{\prime }{)}^{-1}(C^{\prime }{)}^{\mathsf{T}}.\end{array}$$

The lower block has size $m$, and $S$ has size $k$. Positive definiteness gives $H^{\prime }>0$, $S>0$, and $\det A^{\prime }=\det H^{\prime }\det S$. The pole at $D$ occurs only in the first block; $H$ extends smoothly through that time. These facts are derived from the cut multiplicity, without a parallel curvature splitting.

For endpoint differences $\Delta A=A(y)-A(x)$, define the second Schur complement

$$\begin{array}{rl}Q(x,y)& =\Delta V_b\\ & -(\Delta C)(\Delta H{)}^{-1}(\Delta C{)}^{\mathsf{T}}.\end{array}$$

Block elimination gives $\det \Delta A=\det \Delta H\det Q$. Thus the density separates into a regular factor from $H$ and a cut factor from $Q$, normalized respectively by the endpoint determinants of $H^{\prime }$ and $S$.

The important differential estimates are

$${\partial }_{y}Q\ge S(y),-{\partial }_{x}Q\ge S(x).$$

They follow by completing a positive matrix square after differentiating the Schur complement. Invertibility of $\Delta H$ suffices; its sign can be either positive or negative.

The matrix $Q$ vanishes on the diagonal and at cut-separated endpoint pairs. Fold each endpoint into the half-period: write $x=u+iD$ and $y=v+jD$, with $0<u<v<D$ and $i,j\in \{0,1\}$. Integrating from either zero gives

$$\begin{array}{rl}Q(x,y)& \ge {\int }_u^{v}S(t+iD)dt,\\ Q(x,y)& \ge {\int }_u^{v}S(t+jD)dt.\end{array}$$

Both hold even when the actual order is $x=u+D>v=y$. Each integration keeps its moving endpoint within one chart interval, so it does not cross the pole.

Taking determinant roots converts these matrix bounds to scalar ones. Put ${\delta }_r(B)=\det (B{)}^{1/r}$ for positive definite $r$-by-$r$ matrices. Its concavity and degree-one homogeneity give

$${\delta }_r\left({\int }_a^{b}B(t)dt\right)\ge {\int }_a^b{\delta }_r(B(t))dt.$$

The resulting scalar profiles are $h=(\det H^{\prime }{)}^{1/m}$ and $s_i(t)=(\det S(t+iD){)}^{1/k}$. Pole expansions show that their ratios to ${\csc }^{2}t$ and $4{\csc }^2(2t)$, respectively, stay between positive constants. This endpoint control is needed for the logarithmic integrals that follow.

🏷️ Weighted cancellation and the equality defect

The scalar reference profile on an interval of length $b$ is

$$p_0(t)=\frac{{\lambda }^2}{{\sin }^2(\lambda t)},\lambda =\pi /b.$$

For a positive profile $p$, put $g=\log (p/p_0)$ and assume $g$ bounded. On each interval $(x,y)$ let ${\nu }_{xy}$ be the probability measure with density proportional to $p_0$.

The manuscript proves a weighted identity: if $W(b-z)=W(z)$ and $W(z)/\sin (\lambda z)$ is bounded, the weighted integral of

$$\log \frac{{\int }_x^{y}p}{{\int }_x^y{p}_0}-\frac{g(x)+g(y)}{2}$$

equals the weighted integral of

$$\log {\int }_x^y{e}^{g}d{\nu }_{xy}-{\int }_x^{y}gd{\nu }_{xy}.$$

The second expression is nonnegative by Jensen’s inequality. The equality of integrals comes from cancellation of the interval averages against the endpoint terms: reflection symmetry gives the same integrated coefficient for $g(t)$ in both. The bounded ratio hypotheses ensure finite integrals despite the poles of $p_0$.

Apply this identity on the full period for the regular determinant factor. For the cut factor, the four endpoint choices have weights, with $z=v-u$,

$${\left(\frac{\sin 2z}{2}\right)}^k\left(\begin{array}{cc}{\sin }^{m}z& {\cos }^{m}z\\ {\cos }^{m}z& {\sin }^{m}z\end{array}\right).$$

Every row and column has the same sum, symmetric under $z\mapsto D-z$. Summing the four comparisons therefore supplies exactly the symmetry needed on the half-period. This is why both endpoint orders had to be treated in the Schur estimates.

The combined result is a nonnegative weighted integral of $\log (f/w)$. Put $r=f/w$ off the diagonal and cut lines. The identity

$$f-w=w\log r+w(r-1-\log r)$$

retains a strict-convexity remainder. Its last factor is nonnegative and vanishes only at $r=1$. Integrating over the endpoint square $\mathcal{R}=(0,T{)}^2$ gives

$${\int }_{\mathcal{R}}f\ge {\int }_{\mathcal{R}}w(|y-x|).$$

Equality forces $f=w$ almost everywhere, hence everywhere by continuity. The proof thereby keeps a usable equality statement without classifying equality separately in every matrix estimate.

🏷️ From individual geodesics to metric rigidity

Let $U=SM$ be the unit tangent bundle, with Liouville measure $d\mu$ given by base volume times ordinary unit-sphere area. It is invariant under the geodesic flow and has total mass $|S^d|V$.

Polar integration from each base point over times $0<t<D$ gives $V$. Over a whole period it gives $2V$, using reversal in the second half. Translation along the flow therefore yields

$$\begin{array}{rl}{\int }_U{\int }_0^T{\int }_0^T{f}_{\xi }& =2T{V}^2,\\ {\int }_U{\int }_0^T{\int }_0^{T}w& =2TV{V}_0.\end{array}$$

The geodesic comparison proves $V\ge V_0$ after division by $2TV>0$.

If $V=V_0$, every geodesic defect vanishes: the defects are continuous, nonnegative, and integrated against a measure of full support. Their equality cases give the model radial density in every direction at every point.

A Riemannian manifold is harmonic when its volume density in normal coordinates is radial. Here the Cartesian density is the polar density divided by $t^d$, so equality proves harmonicity. The manuscript then invokes Szabó’s classification of compact simply connected harmonic manifolds. The model type fixes the symmetric space, and the common diameter fixes its scale.

🏷️ The quaternionic upper bound and its normalization

The preceding equality argument still requires an upper volume bound. For complex-projective type, the manuscript verifies the homotopy-type hypothesis of Yang’s period-normalized volume theorem. For the quaternionic and Cayley planes, smooth incidence classification supplies the homeomorphism hypothesis of a classical equal-volume theorem. Higher quaternionic dimensions require a separate argument, also valid for the quaternionic plane.

Suppose $n=4r$, $r\ge 2$. Because every geodesic is simple with prime period $T$, time translation acts freely on $U=SM$. The quotient

$$B=U/S^1$$

is the manifold of oriented geodesics with the starting point forgotten. Its dimension is $8r-2$.

The contact form ${\lambda }_w(\zeta )=⟨w,d\pi (\zeta )⟩$ measures velocity in the geodesic direction. Dividing by $T$ gives a circle-bundle connection with integral one along each orbit. Its curvature descends to a symplectic form on $B$, representing an integral Euler class $e$.

Integrating first along the circle gives the contact volume identity

$$\begin{array}{c}N:={\int }_B{e}^{4r-1},\\ N=\frac{(4r-1)!|S^{4r-1}|}{T^{4r}}V.\end{array}$$

The factorial comes from the exterior power of $d\lambda$, which pairs each horizontal normal direction with its vertical counterpart. For the equally scaled standard model,

$$N_0=\frac{1}{2r+1}\left(\genfrac{}{}{0ex}{}{4r}{2r}\right).$$

Thus $V\le V_0$ is equivalent to $N\le N_0$. The rest of the manuscript computes enough integral topology to bound this number. The following dependencies are essential; we record the characteristic-class computations as inputs rather than developing algebra.

The cut-incidence manifold consists of pairs $(p,q)$ at distance $D$. Its actual normal four-plane bundle identifies the great three-sphere of directions from $p$ to $q$. At a fixed $p$, the four-planes associated with distinct endpoints intersect only at zero. This rank restriction controls the tangent bundle’s Pontryagin classes, which are topological invariants of real vector bundles. Initially their formula contains one odd integer coefficient. The signature theorem—relating characteristic classes to the signature of the middle-dimensional intersection pairing—forces its absolute value to be one. Small ranks are computed exactly; a uniform estimate handles all larger ranks.

Next, the integral Gysin sequences for the unit-tangent sphere bundle and the geodesic circle bundle determine the lattice of cohomology classes on $B$. In particular, the pairing

$$\begin{array}{c}(a,b)⟼{\int }_{B}eab,\\ a,b\in H^{4r-2}(B;\mathbb{Z})\end{array}$$

has determinant of absolute value $r+1$ in an integral basis. Rational dimensions alone would lose this normalization.

The first negative bundle of free-loop energy provides the remaining link. A negative direction is a variation decreasing the energy to second order. The manuscript identifies a descended rank-three negative bundle $E\to B$ and an attaching map from its sphere bundle to $M$. The actual bundle identification transfers the incidence characteristic class. The attaching map pulls back the degree-four generator of $M$, and integration over its sphere fibers gives precisely $e$, with coefficient one after orientation. Primitivity is established over every prime field, so a hidden integral multiple cannot alter the final volume scale.

Vanishing of the $(r+1)$st power of that pulled-back class then imposes relations on $B$. Combining these relations with the middle determinant gives the sharp bound

$$0<N\le \frac{1}{2r+1}\left(\genfrac{}{}{0ex}{}{4r}{2r}\right).$$

The manuscript’s final finite coefficient calculation uses a nonzero quarter-integral parameter and termwise bounds whose sum equals the displayed model number. This is a theorem about all ranks, not a check of finitely many dimensions. With the contact identity it supplies $V\le V_0$, and the analytic equality case finishes the metric classification. (OpenAI, 2026)

🏷️ Key steps and proof outlines

Differentiating the Schur complement retains a positive remainder

Use the fixed graph chart from above and put $R=(\Delta C)(\Delta H{)}^{-1}$. Direct differentiation gives

$${\partial }_{y}Q=V_b^{\prime }(y)-C^{\prime }(y)R^T-R{C}^{\prime }(y{)}^T+R{H}^{\prime }(y)R^T.$$

Completing the square yields the exact identity

$${\partial }_{y}Q=S(y)+[R-C^{\prime }(y)H^{\prime }(y{)}^{-1}]H^{\prime }(y)[R-C^{\prime }(y)H^{\prime }(y{)}^{-1}{]}^T.$$

The last term is positive semidefinite because $H^{\prime }(y)>0$. The same formula at $x$ gives $-{\partial }_{x}Q\ge S(x)$. Notice that positivity of $\Delta H$ was not used: its inverse exists even when the actual endpoint order is reversed. (OpenAI, 2026)

At a cut-separated pair, the nullity of $\Delta A$ is exactly $k$. Block congruence identifies this nullity with that of the $k\times k$ matrix $Q$, so $Q=0$. Near the diagonal, Taylor expansion gives $Q(x,y)=(y-x)S(x)+O((y-x{)}^2)$. These two zero sets are the anchors for the folded endpoint integrations. Each endpoint moves within one half-period, avoiding the chart pole; the positive matrix estimate therefore survives even for the folded pair whose actual first endpoint is later than its second.

Scalar averaging uses both endpoint orders

Determinant-root concavity converts the integrated matrix inequalities into interval averages of positive scalar profiles. Their ratios to the singular reference profiles stay bounded above and below, making their logarithms integrable with the chosen weights.

For the cut block, there are four folded endpoint choices. Their weight matrix has equal row sums and equal column sums. Summing all four therefore gives the same coefficient to the left and right endpoint logarithms. Reflection symmetry then cancels these terms against the integrated interval averages. This cancellation is what permits Jensen’s inequality to be used with its correct weights; an unweighted average or a single endpoint ordering would leave unmatched terms.

The final scalar defect carries all the rigidity

After the weighted logarithmic comparison, write $r=f/w$. The identity

$$f-w=w\log r+w(r-1-\log r)$$

separates the established logarithmic contribution from a pointwise nonnegative remainder. The function $r-1-\log r$ vanishes only at $r=1$. Thus equality of the integrated volumes forces equality of the radial determinant, without separately classifying every earlier matrix equality case.

Flow invariance converts the two-endpoint integral into $2T{V}^2$ for the actual metric and $2TV{V}_0$ for the model. If the independent upper-volume argument gives $V\le V_0$, all these nonnegative defects vanish. Full support and continuity upgrade vanishing almost everywhere to every geodesic. The model polar determinant then gives radial Cartesian density after division by $t^{n-1}$, supplying the harmonicity hypothesis for the final classification theorem. The topological upper-bound argument remains a separate input, including its integral normalization; the local Jacobi comparison alone proves only the lower bound and its equality consequence.

🏷️ Connections

Entropy rigidity also passes from an equality statement to a rigid metric through regularity and dynamical structure. Here the decisive equality is in volume, and the retained defect yields harmonic radial density.

The nearby Lagrangian family provides background on symplectic and Lagrangian geometry. The Lagrangian spaces used here are finite-dimensional spaces of Jacobi fields, encoding conjugate times rather than embedded submanifolds of a cotangent bundle.

🏷️ Manuscripts and verification scope

• The metric Blaschke theorem — pinned manuscript and companion source.

No family scope page was present at lean/docs/344.md in the pinned snapshot. This does not establish that all related formalization code is absent. No Lean build, axiom audit, or manuscript-to-formal-statement equivalence check was performed for these notes.

OpenAI collection provides prerequisites and the reading sequence; coverage register records completed and pending families.

🐻  OpenAI 2026. The metric Blaschke theorem.