Chern's Affine Conjecture

Chern's Affine Conjecture

Definitions and mathematical statement

An affine structure on an $n$-manifold is an atlas whose transition maps are restrictions of affine transformations of ${\mathbb{R}}^n$. Equivalently, its tangent bundle has a connection $\nabla$ with zero curvature and zero torsion. A closed manifold is compact without boundary. Its Euler characteristic is \[ \chi(M)=\sum_{i=0}^n(-1)^i\dim_{{\mathbb{Q}}} H_i(M;{\mathbb{Q}}). \] Chern's usual conjecture, in positive dimension, is that every closed affine manifold satisfies $\chi(M)=0$. Completeness of the affine connection is not assumed; this distinguishes the question from compact complete affine manifolds.

The positive-dimensional convention needs to be explicit: if the catalogue's unqualified phrase every closed affine manifold includes dimension zero, a single point has a flat torsion-free tangent connection and Euler characteristic one. That degenerate literal extension is not the standard conjecture.

Formulation and concept references: [S1].

Short English statement

Must a closed positive-dimensional manifold admitting affine coordinate changes have Euler characteristic zero, even without completeness?

Research attempt

Scope and status audit (27 September 2026).

The manifold is smooth, closed, and of positive dimension. It is enough to treat connected manifolds, since Euler characteristic adds over components. Passing to a finite cover of degree $q$ multiplies Euler characteristic by $q$ and pulls back the affine structure. Thus an orientation double cover reduces the problem to the oriented case. No developing map is assumed injective or surjective, and no completeness hypothesis is inserted.

The supplied 2026 source [S1, Remark 2.4] explicitly states the conjecture. Klingler's published theorem [E1] proves it under the additional hypothesis of a parallel volume form. That theorem is acknowledged, not rederived here. Status searches must distinguish a proof claim from an accepted result: Ge's arXiv record [E2] is explicitly withdrawn and reports errors, including an interchange of limits. Its unchanged abstract cannot be used as a theorem. Cocos's arXiv record [E3] contains a general proof claim; the present note has not audited that argument and does not use it. These checks justify retaining the catalogue's unresolved/disputed-claim status, not announcing a settlement from a search result.

The main route pursued here repairs the tempting but invalid instruction "put the flat curvature into the Euler Pfaffian." A compatible metric connection is constructed exactly, its residual curvature is computed, and a sufficient analytic vanishing criterion is derived. The construction also supplies a local counterexample to pointwise Euler-density vanishing. Alongside it, elementary holonomy subcases and the two-dimensional obstruction are checked to delimit where a counterexample could occur.

Elementary reductions which do not need completeness.

For an oriented closed manifold of odd dimension $n$, rational Poincare duality gives $b_i=b_{n-i}$. In the Euler sum, the terms with indices $i$ and $n-i$ have opposite signs, and there is no middle index. Hence $\chi(M)=0$. Applying this on an orientation double cover proves the same for nonorientable odd-dimensional closed manifolds. The remaining issue is therefore even dimension.

Lemma 498.1 (invariant line subcase). If the linear holonomy of a closed affine manifold preserves a one-dimensional subspace, then its Euler characteristic is zero.

Proof. Parallel transport of the invariant subspace defines a rank-one subbundle $L\subset TM$: path dependence is precisely the holonomy action, which preserves the chosen line. Pass, if necessary, to the double cover which orients $L$. An oriented real line bundle is trivial: choose a bundle metric and take the positive unit vector in each fiber. This is a smooth nowhere-zero section of $L$, hence of the tangent bundle of the cover. The Poincare--Hopf theorem gives Euler characteristic zero there. Dividing by the finite covering degree proves the assertion for $M$. $\square$

The parallel transport need not fix a vector, and may rescale it; the proof intentionally uses a line subbundle rather than incorrectly claiming that a parallel vector field exists. More generally, a finite holonomy orbit of lines becomes an invariant line after passing to the finite cover corresponding to its stabilizer. The same argument applies.

If the closure of linear holonomy is compact, average a positive inner product over this compact group and extend it by parallel transport. Holonomy invariance makes the resulting Riemannian metric globally well-defined and parallel. The flat connection is now a metric connection, so its Euler form vanishes, giving $\chi(M)=0$ by Chern--Gauss--Bonnet. Here the averaging is over a compact group with finite Haar measure; there is no analogous averaging over an arbitrary noncompact linear holonomy group. The compact case is one place where the tempting curvature argument does become valid after its missing hypothesis is verified.

Surface check using the precise flat-bundle inequality.

For a closed oriented surface $\Sigma_g$ with $g\ge1$, Milnor's Theorem 1 [E4] implies that a flat oriented real two-plane bundle $E$ satisfies \[ |\langle e(E),[\Sigma_g]\rangle|<g. \] For $E=T\Sigma_g$ its Euler number is $2-2g$. If $g\ge2$, then $|2-2g|=2g-2\ge g$, a contradiction. If $g=0$, a flat bundle on the simply connected sphere has a global parallel frame by path-independent parallel transport, whereas the tangent bundle has Euler number two. Thus $g=1$ is the only oriented possibility and has Euler characteristic zero. The orientation cover handles every nonorientable closed surface.

This is an application of a cited theorem, not an independent proof of Milnor's inequality. The indexed primary article supplied its exact strict bound and corollary; direct retrieval from its archive encountered an access barrier, recorded in [E4]. The inequality is enough for the surface conclusion even for a flat tangent connection with torsion. In higher dimensions that extra strength cannot simply be assumed.

The metric correction and its surviving curvature.

Let $\nabla$ be a flat connection on $TM$, and choose any smooth positive definite bundle metric $g$. The following identities use flatness but not yet torsion-freeness. Define an endomorphism-valued one-form \[ B=g^{-1}\nabla g,\qquad A=\tfrac12 B, \qquad D=\nabla+A. \] The notation means that $g(B_XY,Z)=(\nabla_Xg)(Y,Z)$, so each $B_X$ is self-adjoint with respect to $g$. This definition is intrinsic; the matrix formulas below are computed in a local $\nabla$-parallel frame.

Lemma 498.2 (exact correction identity). The connection $D$ is compatible with $g$, and its curvature is \begin{equation}\label{wsref-0-72-61-34-39-38-3a-63-75-72-76-61-74-75-72-65} R^D=-A\wedge A=-\tfrac14B\wedge B, \qquad R^D(X,Y)=-[A_X,A_Y]. \end{equation}

Proof. Since $\nabla_Xg=gB_X$ in matrix notation and $B_X$ is $g$-self-adjoint, \[ (D_Xg)(Y,Z)=(\nabla_Xg)(Y,Z) -g(A_XY,Z)-g(Y,A_XZ)=0. \] In a parallel frame $B=g^{-1}\mathrm dg$. Differentiating the inverse matrix gives $\mathrm d(g^{-1})=-g^{-1}(\mathrm dg)g^{-1}$, hence \[ \mathrm dB=-B\wedge B, \qquad \mathrm dA=-2A\wedge A. \] The original connection matrix is zero in this frame, and therefore $R^D=\mathrm dA+A\wedge A=-A\wedge A$. Because both sides transform tensorially, the identity is global. Evaluating the wedge product on two vectors gives the displayed commutator formula.$\square$

When $n=2r$, Chern--Weil theory applied to the metric connection $D$ represents $e(TM)$ by its Euler form \begin{equation}\label{wsref-0-72-61-34-39-38-3a-65-75-6c-65-72} e(D)=\frac{1}{(2\pi)^r}\operatorname{Pf}(\Omega_D), \qquad \chi(M)=\int_M e(D), \end{equation} where $\Omega_D$ is the skew-symmetric curvature matrix in an oriented $g$-orthonormal frame, with a consistent curvature sign convention. The Pfaffian is not being applied to an arbitrary matrix in an arbitrary parallel frame. Formula (ra498:curvature) shows exactly why flatness of $\nabla$ does not imply flatness of the compatible connection: the endomorphisms measuring the metric variation need not commute. The Euler-form construction and the nonmetric-connection distinction are discussed in [E5, Appendix C].

A local stress test, even with a Hessian metric.

On a disk about the origin in $\mathbb R^2$, use the ordinary flat torsion-free connection and the potential \[ f(x,y)=\tfrac12(x^2+y^2)+\tfrac16x^3-\tfrac12xy^2. \] Its Hessian is \[ g=\begin{pmatrix}1+x&-y\\-y&1-x\end{pmatrix} =I+xS+yT,\quad S=\begin{pmatrix}1&0\\0&-1\end{pmatrix},\quad T=\begin{pmatrix}0&-1\\-1&0\end{pmatrix}. \] The eigenvalues are $1\pm\sqrt{x^2+y^2}$, so $g$ is positive definite on the unit disk. At the origin the coordinate frame is orthonormal and \[ A_{\partial_x}=S/2,\qquad A_{\partial_y}=T/2, \qquad R^D(\partial_x,\partial_y) =-\tfrac14[S,T] =\begin{pmatrix}0&1/2\\-1/2&0\end{pmatrix}. \] Thus the degree-two Euler density at the origin is nonzero (its magnitude is $1/(4\pi)$ in the coordinate area form). In this example the third derivatives of $f$ are totally symmetric, so $A_XY=A_YX$ for coordinate vectors. Hence $D$ is also torsion-free and is the Levi--Civita connection of $g$. The nonzero density is not an artifact of allowing torsion in the corrected connection.

This is not a counterexample to the conjecture: the disk is not closed. It is a counterexample to the stronger local claim that any metric Euler density must vanish on an affine chart. By inserting a cutoff of a small multiple of $xS+yT$ in a coordinate disk on a flat two-torus, one obtains a globally positive metric with the same nonzero commutator near the disk center. The torus still has Euler characteristic zero. Its Euler integral therefore cancels across the manifold. A successful general argument must control a global integral or cohomology class, not merely use the existence of locally flat coordinates.

A quantitative sufficient criterion and its scaling obstruction.

The exact formula yields an analytic route with a clearly measurable target. Let $|A|_g$ denote the Hilbert--Schmidt norm on $T^*M\otimes\operatorname{End}(TM)$, using $g$ on all factors. At each point choose a $g$-orthonormal basis. For basis vectors $e_i,e_j$, \[ |R^D(e_i,e_j)|\le2|A_{e_i}|\,|A_{e_j}|\le2|A|_g^2. \] The coefficient of a Pfaffian is a finite sum, depending only on $r$, of products of $r$ curvature components. Consequently a dimensional constant $C_r<\infty$ satisfies \begin{equation}\label{wsref-0-72-61-34-39-38-3a-65-6e-65-72-67-79} |\chi(M)|\le\int_M|e(D)| \le C_r\int_M|A|_g^{2r}\,\mathrm d\operatorname{vol}_g. \end{equation} This estimate follows just from the Pfaffian expansion and the commutator bound; it requires no choice of a global frame.

Corollary 498.3 (conditional analytic vanishing). If there are smooth metrics $g_j$ for which \[ \int_M\left|\tfrac12g_j^{-1}\nabla g_j\right|_{g_j}^{2r} \,\mathrm d\operatorname{vol}_{g_j}\longrightarrow0, \] then $\chi(M)=0$. In fact a single metric making the right side of (ra498:energy) smaller than one is enough, because $\chi(M)$ is an integer.

Proof. Apply (ra498:energy) to each metric. The Euler integral is the same topological integer for every metric, so its absolute value tends to zero. The last assertion is the same inequality plus integrality. $\square$

A naive scaling attempt fails exactly at the critical exponent. If $\widehat g=c^2g$ for a positive constant $c$, then $\widehat A=A$ as an endomorphism-valued one-form. Its endomorphism norm is unchanged, its covector norm is multiplied by $c^{-1}$, and volume by $c^{2r}$. Therefore \[ \int_M|\widehat A|_{\widehat g}^{2r} \,\mathrm d\operatorname{vol}_{\widehat g} =\int_M|A|_g^{2r}\,\mathrm d\operatorname{vol}_g. \] Making a metric uniformly large cannot achieve the hypothesis. Nor does a partition of unity of local parallel metrics control it: differentiating the weights produces nonzero $\nabla g$ in the overlaps. No estimate making that critical energy arbitrarily small has been obtained here. The criterion is sufficient and may be stronger than what the conjecture needs.

Why the missing ingredient must use more than flatness.

Milnor's Theorem 2 [E4] supplies flat oriented real two-plane bundles with nonzero Euler number: over a genus-two surface the Euler number one satisfies $|1|<2$. Such a bundle is not that surface's tangent bundle, whose Euler number is $-2$. It nevertheless disproves the general assertion that every flat oriented bundle has zero real Euler class. Thus the metric correction cannot automatically have an exact Euler form solely because its starting bundle connection was flat.

For a tangent connection the extra torsion-free condition is an integrability statement. In a local parallel frame $E_1,\ldots,E_n$, \[ T^\nabla(E_i,E_j)=\nabla_{E_i}E_j-\nabla_{E_j}E_i-[E_i,E_j] =-[E_i,E_j]. \] If torsion vanishes, the parallel frame fields commute and furnish local affine coordinates. None of the algebra in Lemma 498.2 or the energy bound uses this commutation. It is exactly the extra structure still available to exploit, rather than a condition already spent in a nonexistent curvature cancellation. Conversely, even using a local Hessian metric does not force pointwise cancellation, as the explicit disk shows.

The remaining analytical target in this approach is therefore a global construction using the affine-coordinate compatibility to produce either the small-energy metrics of Corollary 498.3, or a global primitive for $e(D)$, or a direct vanishing of its integral. No such construction is provided. Local primitives on charts do not automatically glue: the overlap corrections carry the Euler class one is trying to eliminate. One cannot use Chern--Simons transgression from $\nabla$ as if the Euler Pfaffian were an invariant polynomial for arbitrary general-linear connections; that would repeat the initial error in a different notation.

Verification and outcome.

The positive-dimensional exclusion prevents the point from being called a counterexample. Finite covers preserve affine structures and have the stated Euler scaling. The invariant-line proof does not assume compact holonomy, and the compact-holonomy proof does not average over an infinite measure. The metric identities were checked in parallel frames and then interpreted tensorially; orthonormal frames are used only where the Euler Pfaffian requires them. An exact rational matrix check in the local research record verifies the Hessian, commutator, and nonzero curvature at the origin. It is not a numerical experiment on the space of closed affine manifolds.

Outcome: UNRESOLVED. The work derives the metric-correction identity and its critical-energy criterion, proves elementary holonomy and parity subcases, and applies the known surface inequality. It also gives an explicit local obstruction to a frequently tempting proof shortcut. The first unproved step is global Euler-integral control in even dimensions beyond these subcases, using torsion-free affine compatibility without assuming completeness or a parallel volume form. No closed affine counterexample and no general vanishing proof have been produced.

Sources

• S1 Emmanuel Gnandi, Michel Nguiffo Boyom, and Stephane Puechmorel, Canonical foliations of statistical manifolds with statistical models, Remark 2.4. https://link.springer.com/article/10.1007/s41884-026-00193-8. Formulation source.
• E1 Bruno Klingler, Chern's conjecture for special affine manifolds, Annals of Mathematics 186 $2017$, no. 1, 69--95. https://annals.math.princeton.edu/2017/186-1/p02. Publisher theorem summary inspected 27 September 2026; the extra parallel-volume hypothesis is essential to the cited statement.
• E2 Jianquan Ge, Proof of Chern's conjecture on affine manifolds, arXiv:2002.03105v2, withdrawn 27 July 2020. https://arxiv.org/abs/2002.03105v2. Withdrawal and author error notice inspected 27 September 2026; not used as a theorem.
• E3 Mihail Cocos, Proof of Chern conjecture for flat affine manifolds, arXiv:1504.04852v12, 6 December 2025. https://arxiv.org/abs/1504.04852v12. Record and abstract inspected 27 September 2026; general proof claim not independently audited or used here.
• E4 John Milnor, On the existence of a connection with curvature zero, Commentarii Mathematici Helvetici 32 $1958$, 215--223, Theorems 1 and 2 and the surface corollary on the first page. https://doi.org/10.1007/BF02564579; archival scan https://www.e-periodica.ch/cntmng?pid=com-001:1957:32::15. Indexed primary-paper theorem text inspected 27 September 2026; direct archive retrieval encountered an access barrier.
• E5 John W. Milnor and James D. Stasheff, Characteristic Classes, Princeton University Press, 1974, Appendix C, Euler form and the distinction between orthogonal and general linear connections. https://webhomes.maths.ed.ac.uk/~v1ranick/papers/milnstas.pdf. Indexed Appendix C passage inspected 27 September 2026; the full scan exceeded the browser retrieval size limit.