Let M=Ω/Γ be a closed properly convex real projective 4-manifold. We prove the Chern-Hopf-Thurston conjecture for this class of manifolds, i.e. χ(M)≥0. The argument transfers Akagawa's double-form identity on the Cheng–Yau cone over Ω to a Bochner identity on M, from which we deduce that Hq(M;ΛpE∗)=0 whenever q<p(n+1−p)/(n+1), in arbitrary dimension n, for some flat bundle E naturally attached to M. In dimension four, this implies that the cohomology with coefficients in Λ2E∗ concentrates in degree two.
Introduction
The Chern–Hopf–Thurston conjecture predicts that every closed aspherical manifold of dimension 2m satisfies
(−1)mχ(M)≥0.(1)
The conjecture has been established under some geometric, complex-analytic, and topological assumptions. For closed real hyperbolic manifolds, the predicted sign follows from the Gauss–Bonnet theorem, and more generally it holds for closed locally symmetric manifolds of noncompact type [15]. Gromov proved the strict inequality for closed Kähler hyperbolic manifolds [11], while Cao–Xavier and Jost–Zuo established the non-strict inequality for compact Kähler manifolds with nonpositive sectional curvature [3, 13]. In dimension four, the conjecture holds for closed Riemannian manifolds with nonpositive sectional curvature [6], closed aspherical complex surfaces [12], closed nonpositively curved piecewise Euclidean cubical manifolds [8], and closed generalized Haken manifolds [7]. For closed aspherical manifolds whose fundamental groups contain an infinite amenable normal subgroup, Cheeger–Gromov proved the stronger conclusion χ(M)=0[4]. More recently, Deng–Wang established the conjecture for closed aspherical complex projective manifolds whose fundamental groups admit a linear representation with finite kernel over a field of arbitrary characteristic [9].
In this paper we establish the conjecture for closed properly convex real projective 4-manifolds. A properly convex projective n-manifold is a quotient M=Ω/Γ, where Ω⊂RPn is an open convex set bounded in an affine chart, and Γ acts freely, properly discontinuously, and cocompactly by projective transformations. The domain Ω is contractible, so M is aspherical. For a general domain, the Cheng–Yau construction supplies a canonical Hessian metric on its cone [5]. We use this metric to prove a vanishing result for flat cohomology with coefficients in exterior powers of the holonomy representation. For oriented four-manifolds, this yields the following:
Theorem A. Let M be a closed properly convex real projective 4-manifold. Then χ(M)≥0. More precisely, if M is oriented, let E→M be the flat rank-five bundle defined by the cone-preserving unimodular lift of its holonomy. Then
10χ(M)=dimRH2(M;Λ2E∗).(2)
The holonomy lift and its flat bundle are constructed in § 2. No strict convexity, boundary regularity, or irreducibility hypothesis is imposed, and the proof makes no assumption on the sectional curvature of a metric on M. The identity (2) explains the sign by realizing a positive multiple of the Euler characteristic as the dimension of a cohomology group.
The result also completes the determination of the possible Euler characteristics in the oriented four-dimensional case. Kobayashi, using an earlier Pontryagin-form identity of Avez, proved the rational vanishing of the Pontryagin classes of real projective manifolds [14 Theorem 4.16]. In dimension four, this forces the signature to vanish and the Euler characteristic to be even. These restrictions alone leave its sign undetermined. Riolo–Seppi–Slavich construct closed oriented properly convex projective four-manifolds with every positive even Euler characteristic [16 Theorem 1.3]. Together with the simplex quotients, for which χ=0, Theorem A shows that the possible values are precisely the even non-negative integers.
Cohomology with flat coefficients. The vanishing theorem underlying Theorem A holds in arbitrary dimension. Let M be oriented of dimension n, put N=n+1, and let E be the standard flat rank-N bundle defined by its unimodular holonomy lift. For 0≤p≤N, set
Fp=ΛpE∗.
The flat connection defines an exterior differential dp on Ω∙(M;Fp), and its cohomology is denoted by H∙(M;Fp).
Theorem B. Let M be a closed oriented properly convex real projective n-manifold, n≥2, and let E be its standard unimodular flat bundle. For 0≤p≤n+1 and 0≤q≤n,
Hq(M;ΛpE∗)=0ifq<n+1p(n+1−p).(3)
The theorem follows from a Bochner identity on the compact manifold M. The Cheng–Yau metric on the cone determines a Riemannian metric k on M and a metric on Fp. Let dp∗ be the resulting formal adjoint and let Δp,q=dpdp∗+dp∗dp. We prove the quantitative estimate
∥dpα∥2+∥dp∗α∥2≥(Np(N−p)−q)∥α∥2,α∈Ωq(M;Fp).(4)
More precisely, we construct first-order operators Ap,q, Bp,q on bundles over M such that
Δp,q=Ap,q∗Ap,q+Bp,q∗Bp,q+(p−q−Np2)Id.(5)
The bundles, metrics, and operators are defined in § 4.3. Taking the inner product with α gives (4). The flat connection need not preserve the fiber metric, but the corresponding elliptic Hodge theory still represents flat cohomology by harmonic forms; we include the standard argument in Lemma 3.6.
Theorem A is a consequence of (3). For N=5, and p=2,3 we have
5p(5−p)=56>1.
Theorem B therefore eliminates degrees zero and one cohomology with coefficients in F2 and F3. A parallel volume form on E identifies F2∗ with F3. Poincaré duality then eliminates degrees three and four for F2. Its cohomology is thus concentrated in degree two, and the Euler characteristic with rank-ten flat coefficients yields (2). In higher even dimensions, the vanishing range and its dual can leave other degrees besides the middle one, so the theorem does not by itself establish the sign conjecture there.
Organization of the paper.§ 2 constructs the unimodular holonomy bundle and the radial product. § 3 defines the double-form operators and derives their identity in the cone over Ω following [1]. § 4 transfers it to the aforementioned Bochner identity on M and proves Theorem B. § 5 gives the topological argument for Theorem A. Appendix § 6 gives a direct proof in coordinates of the identities in [1].
AI declaration. Chat GPT Sol 6.1 was used for literature review, proofreading and editing. All mathematical ideas are original of the author.
Acknowledgment. The author acknowledges support from the Italian Ministry of University and Research (MUR) through the FIS3 project “Geometry and Topology of higher Teichmüller spaces”, grant no. FIS-2024-05142.
The Cheng–Yau cone and the holonomy bundle
We work in dimension n≥2 and write N=n+1. Unless explicitly stated otherwise, M=Ω/Γ is oriented. We begin by fixing the lift of the holonomy.
The unimodular lift
Choose an affine chart in which Ω is bounded and convex. After identifying this chart with {x0=1}⊂RN, let
C={t(1,z):t>0,z∈Ω}⊂V=RN.
This is an open convex cone. Its radial projection identifies C/R>0 with Ω.
Lemma 1.1.The holonomy of an oriented properly convex projective manifold has a unique lift Γ→SL(V) that preserves C.
Proof. By [2 Section 2.1], the holonomy has a unique lift to
SL±(V)={A∈GL(V):∣detA∣=1}.
whose image preserves the cone C. We check that this lift takes values in SL(V). The radial projection C→Ω has naturally oriented fibers: each fiber is a positive ray, oriented by increasing radial parameter. Every cone-preserving linear map preserves these fiber orientations. Moreover, the holonomy acts on Ω by orientation-preserving maps, because M=Ω/Γ is oriented. The lifted maps therefore preserve the induced orientation of C, so their determinants are positive. Uniqueness follows from the uniqueness of the cone-preserving lift to SL±(V). □
We use the same symbol Γ for the lifted subgroup of SL(V). Define
E=(Ω×V)/Γ,γ⋅(z,v)=(γz,γv).(6)
Local constant sections on Ω define its flat connection. The induced bundle Fp=ΛpE∗ has monodromy Λp(γ−T). Since Γ⊂SL(V), any fixed volume form on V defines a parallel volume form on E.
The canonical potential
Let D denote the standard flat affine connection on C, and choose a constant volume form on V. We use the following normalization of the Cheng–Yau canonical potential.
Theorem 1.2 (Cheng–Yau). There is a unique smooth strictly convex functionϕonCsatisfying
detD2ϕ=e2ϕ(7)
such that the Hessian metricg=Ddϕis complete.
This is the canonical potential theorem of Cheng–Yau [5]; see also [10 Theorem 1.3 and Corollary 1.1] for a precise formulation and the uniqueness statement. Completeness is used here to characterize the potential; the compact Bochner estimate below uses only local identities. The determinant in (7) is computed in affine coordinates of unit volume for the chosen constant volume form.
Lemma 1.3.IfA∈GL(V)preservesC, then
ϕ(Ax)=ϕ(x)−log∣detA∣.(8)
In particular,
ϕ(tx)=ϕ(x)−Nlogt(t>0),ϕ(γx)=ϕ(x)(γ∈Γ).(9)
The transformationsAand the radial dilations preserveg.
Proof. Set ψ(x)=ϕ(Ax)+log∣detA∣. The chain rule gives
D2ψ(x)=AT(D2ϕ)(Ax)A.
It follows that
detD2ψ(x)=∣detA∣2e2ϕ(Ax)=e2ψ(x).
The function ψ is strictly convex, and its Hessian metric is A∗g, which is complete because A is a diffeomorphism of C. Uniqueness in Theorem 1.2 gives ψ=ϕ, which proves (8). Taking A=tId gives the first identity in (9), and taking A=γ∈Γ gives the second. Applying Dd to (8) proves A∗g=g. □
The radial product
Let H be the vector field, defined by Hx=x in the affine identification TxC=V. Set
Σ=ϕ−1(0),k=g∣TΣ.
Every radial ray meets Σ once. Since ϕ is Γ-invariant, the radial projection induces a diffeomorphism Σ/Γ≅M. We use k also for the quotient metric.
Proposition 1.4.The map
Φ:R×Σ⟶C,Φ(r,x)=erx,
is aΓ-equivariant diffeomorphism, and
Φ∗g=Ndr2+k.(10)
In particular,X=C/Γidentifies withR×Mwith this product metric. In these coordinates,
H=∂r,ϕ=−Nr,∣dr∣g2=N−1,dvolg=Ndrdvolk.(11)
Proof. In affine coordinates H=∑ixi∂i. Differentiating ϕ(tx)=ϕ(x)−Nlogt in t at t=1 gives
i∑xiϕi=−N.
Differentiating this equation in xj then gives
ϕj+i∑xiϕij=0.
Since gij=ϕij, this is the coordinate form of
g(H,Y)=−dϕ(Y),gradgϕ=−H.(12)
In particular, g(H,H)=−dϕ(H)=N, and H is orthogonal to every level set of ϕ. The inverse of Φ is
y⟼(−Nϕ(y),eϕ(y)/Ny).
This proves that Φ is a diffeomorphism. To compute its metric in coordinates, write a local parametrization of Σ as σ(z) and put σa=∂aσ. Then
dΦ(∂r)=erσ,dΦ(∂a)=erσa.
Radial dilation preserves g by Lemma 1.3. Equivalently, for fixed ambient vectors v,w, gerx(v,w)=e−2rgx(v,w). Therefore the product metric coefficients are
gerσ(erσ,erσ)=N,
gerσ(erσ,erσa)=0,
gerσ(erσa,erσb)=kab.
These identities prove (10). Moreover, r=−ϕ/N, and the displayed inverse shows that Γ acts trivially on r and by its original action on Σ. Thus the product is equivariant and descends to X=R×M. Finally, the inverse metric and determinant of the block matrix diag(N,kab) give ∣dr∣2=N−1 and dvolg=Ndrdvolk, while dΦ(∂r)=H gives the remaining radial formulas. □
Double forms and Akagawa’s identity
The cone has two structures that will be used separately: its flat affine connection and its Hessian metric. The flat connection defines the two differentials below, while the metric defines their adjoints.
Double forms, exterior operators, and inner products
Let (Y,D) be a flat affine manifold of dimension N. A double form of bidegree (p,q) is a smooth section of
ΛpT∗Y⊗ΛqT∗Y.
We denote the space of these sections by Ap,q(Y). In affine coordinates x1,…,xN, write
Here and below repeated indices are summed from 1 to N unless otherwise indicated. The components are alternating separately in the indices on the two sides of the semicolon; no symmetry between the two groups is imposed.
Exterior multiplication takes place independently in the two factors. In particular,
(ξ⊗η)∧(ξ′⊗η′)=(ξ∧ξ′)⊗(η∧η′)
for decomposable double forms. There is no sign from passing a form in the first factor past a form in the second factor. This is the convention of [1 Definition 2.1.2].
For a covector ξ, let ε(ξ) and εˉ(ξ) denote wedge multiplication in the first and second factors. For a vector Z, let ι(Z) and ιˉ(Z) denote contraction in the respective factors. Our contraction convention is
Thus contraction inserts the vector in the first slot of the chosen factor.
Write ei=ε(dxi), eˉi=εˉ(dxi), ii=ι(∂i), and iˉi=ιˉ(∂i). In each factor the usual exterior-algebra relations hold:
iiej+ejii=δijId,iˉieˉj+eˉjiˉi=δijId.
Two wedge operators, or two contraction operators, anticommute within the same factor. All operators acting on different factors commute. Put Di=D∂i. In an affine coordinate frame, Di differentiates the components of the entire double form ordinarily. Define
∂=i∑eiDi,∂ˉ=i∑eˉiDi.(13)
For example,
(∂ω)i0…ip;J=r=0∑p(−1)r∂irωi0…ir…ip;J.
and ∂ˉ has the identical formula in the second group of indices. These definitions are independent of the affine chart. Since affine derivatives commute and the two exterior factors commute, they give
∂2=∂ˉ2=0,∂∂ˉ=∂ˉ∂.(14)
The last equality has no minus sign. In particular, the two factors are not being combined into a single exterior algebra with the usual total-degree sign convention.
For fixed p, the first exterior factor is a flat coefficient bundle. More precisely,
To check this description, choose any local parallel frame sA of the first exterior factor and write ω=∑AsA⊗ωA. Then
∂ˉω=A∑sA⊗dωA.
The differential on the right requires only the connection on the coefficient bundle; the ωA are ordinary differential forms.
Suppose now that g=Ddϕ is positive definite, and write gij=ϕij and (gij)=(gij)−1. The metric on T∗Y is g−1. We use its determinant metric on exterior powers and the tensor-product metric on double forms. Explicitly,
For example, on a double (1,1)-form this is gikgjℓωi;jηk;ℓ. The L2 inner product is defined with the Riemannian density:
(ω,η)Y=∫Y⟨ω,η⟩gdet(gij)dx1⋯dxN.
The notation Acp,q(Y) means compact support.
Let δ and δˉ be the formal adjoints of ∂ and ∂ˉ for these inner products. They are determined by integration by parts on compactly supported smooth test forms. Define the local differential operators
□=∂δ+δ∂,□ˉ=∂ˉδˉ+δˉ∂ˉ.
The Hessian identities
Regard g as the double (1,1)-form gijdxi⊗dxj, and let L be wedge multiplication by this double form. Its pointwise adjoint is
Λ=gijiiiˉj.
Indeed, ei∗=giaia and eˉj∗=gjbiˉb, so L∗=gijgiagjbiaiˉb=gabiaiˉb. In components,
(Λω)I;J=gabωaI;bJ.
Thus Λ traces one covariant index from each factor. For a (1,1)-form, Λω=gijωi;j.
The first Koszul form and its metric dual are
α=21dlogdet(gij),Xα=α♯g.
An affine coordinate change multiplies det(gij) by a positive constant, so its logarithmic differential is well defined. The local Hessian Kähler identities are
Λ∂+∂Λ=−δˉ+ˉ(Xα),Λ∂ˉ+∂ˉΛ=−δ+(Xα).(16)
These are [1 Theorem 2.3.6]. In Appendix § 6, we prove (16) directly in coordinates. For the Cheng–Yau potential, its Monge–Ampère equation gives 21dlogdet(D2ϕ)=dϕ. The radial identity (12) then gives
α=dϕ,Xα=−H.(17)
The following calculation explains how the Euler field introduces the exterior degrees.
Lemma 2.1.On double forms of bidegree (p,q) on C,
∂(H)+(H)∂=DH+pId,∂ˉˉ(H)+ˉ(H)∂ˉ=DH+qId.(18)
Proof. Since DiH=∂i, differentiation of contraction gives
The second equality uses ei(H)+(H)ei=xiId. On an elementary exterior monomial of first degree p, the operator ∑ieii counts its p indices: a term vanishes unless dxi occurs, and contraction followed by reinsertion gives that monomial for each such index. Thus ∑ieii=pId. The identical calculation in the second factor gives ∑ieˉiˉi=qId and proves the second identity. These formulas also appear in [1 Corollary 3.3.3]. □
The Lie derivative in the H-direction of the entire covariant tensor is
(LHω)I;J=Ha∂aωI;J+(p+q)ωI;J,
because ∂iHa=δia contributes once for every covariant slot. Hence LHω=DHω+(p+q)ω. In particular, dilation invariance means DHω=−(p+q)ω.
Proposition 2.2 (Akagawa’s cone identity).On double forms of bidegree (p,q) on the Cheng–Yau cone,
□ˉ=□+(p−q)Id.(19)
Proof. This is the local identity established in [1 proof of Main Theorem 3, p. 86]. Put I=(Xα) and Iˉ=ˉ(Xα). Solving (16) for the adjoints gives
δ=−Λ∂ˉ−∂ˉΛ+I,δˉ=−Λ∂−∂Λ+Iˉ.
Substitution into the two Laplacians yields
□=−∂Λ∂ˉ−∂∂ˉΛ−Λ∂ˉ∂−∂ˉΛ∂+∂I+I∂,
□=−∂ˉΛ∂−∂ˉ∂Λ−Λ∂∂ˉ−∂Λ∂ˉ+∂ˉIˉ+Iˉ∂ˉ.
The four terms involving Λ agree between the two lines: the only interchanges needed are ∂ˉ∂ˉ=∂ˉ∂ˉ, with Λ left in its original position. Their difference is therefore
□−□=−(∂I+I∂)+(∂ˉIˉ+Iˉ∂ˉ).
By (17), I=−ι(H) and Iˉ=−ιˉ(H). Applying Lemma 2.1 gives
□−□=(DH+p)−(DH+q)=(p−q)Id.
This proves the asserted identity. □
Proposition 2.3. The four operators ∂,∂ˉ,δ,δˉ descend to X=C/Γ, and (19) holds on X.
Proof. Each element of Γ is affine for D and preserves g. Its pullback therefore commutes with both differentials and preserves the pointwise tensor metrics and the Riemannian density. It also commutes with their formal adjoints: the defining integration-by-parts identity on a coordinate neighborhood is unchanged by this pullback. □
The next section rewrites (19) as an identity of operators on M.
A Bochner identity
The aim of this section is to express the two sides of Akagawa’s cone identity as operators on bundles over M. In particular, we define the operators in the Bochner identity (5) and compute the action of □ and □ on dilation invariant forms.
The coefficient bundle and its metric
The identification TC=C×V identifies its flat cotangent bundle with C×V∗. The action on the fiber is γ−T. Its pth exterior power therefore descends to a flat bundle Fp on X=C/Γ. Radial projection gives a flat bundle isomorphism
Fp≅πM∗Fp,πM:R×M⟶M.(20)
To describe it, write a point of the cone as erσ(y), where σ(y)∈Σ. A covector with affine components λi at σ(y) is identified with the covector having the same components λi at erσ(y). This is flat parallel transport in the ambient space V. It is Γ-equivariant, so the description gives the stated isomorphism on the quotient. In particular, the connection in the radial direction is ordinary differentiation.
Under (20), ∂ˉ is the exterior covariant derivative of πM∗Fp, as explained in (15). In a local flat coefficient frame sA, it is characterized by
∂ˉ(A∑sA⊗ωA)=A∑sA⊗dωA.
where ωA are ordinary differential forms on R×M.
The inverse Cheng–Yau metric g−1 induces a fiber metric on Fp=ΛpT∗X. Under the identification (20), the fiber of Fp at (0,y) is identified with (Fp)y. We denote by Hp,y the inner product transported to (Fp)y through this identification. This defines a fiber metric Hp on Fp.
Lemma 3.1. Under (20), the fiber metric on Fp over (r,y) is
e2prHp(y).(21)
Proof. Fix t>0 and differentiate ϕ(tx)=ϕ(x)−Nlogt in the affine variables x. The first and second derivatives give
tϕi(tx)=ϕi(x),t2ϕij(tx)=ϕij(x).
Consequently, for the matrices of the metric and its inverse,
gij(erx)=e−2rgij(x),gij(erx)=e2rgij(x).(22)
For covectors λ,μ we then have
⟨λ,μ⟩erx=i,j∑gij(erx)λiμj=e2r⟨λ,μ⟩x.
The exterior metric is the determinant metric:
⟨λ1∧⋯∧λp,μ1∧⋯∧μp⟩=det(⟨λa,μb⟩)a,b=1p.
Each entry of this p×p matrix acquires the factor e2r, and its determinant acquires e2pr. Decomposable exterior tensors span the fiber, so bilinearity proves (21). □
We regard double (p,q)-forms on X as ordinary q-forms on R×M with coefficients in πM∗Fp:
ΛpT∗X⊗ΛqT∗X≅πM∗Fp⊗ΛqT∗(R×M).
Their pointwise metric is therefore the tensor product of the metric in Lemma 3.1 with the metric on ordinary q-forms induced by the product Riemannian metric Ndr2+k. In particular, for s∈(Fp)y and α∈ΛqT(r,y)∗(R×M),
∣s⊗α∣g2=e2pr∣s∣Hp2∣α∣Ndr2+k2.
On Ωq(M;Fp), we use the pointwise metric Hp⊗Λqk−1 and the volume density dvolk. Let dp be the flat exterior derivative, let dp∗ be its formal adjoint, and put
The differential and its formal adjoint on the product
Every Fp-valued q-form on X has a unique decomposition
ω=a(r)+dr∧b(r),
where a(r)∈Ωq(M;Fp) and b(r)∈Ωq−1(M;Fp). The form is horizontal if b=0. Wedge multiplication by dr in this decomposition acts in the second exterior factor. By definition of the metric (see (11)), together with dvolg=Ndrdvolk, give
To compute the adjoint, take a compactly supported test form c(r)+dr∧d(r) with c(r)∈Ωq−1(M,Fp) and d(r)∈Ωq−2(M,Fp). Its derivative paired with ω=a+dr∧b is
N∫Re2pr((a,dpc)M+N1(b,∂rc−dpd)M)dr.
The terms involving dp become (dp∗a,c)M and (−dp∗b,d)M by integration on M. Compact support in r gives
For a horizontal form f(r)a(y), the formulas reduce to
∂ˉ(fa)=fdpa+f′dr∧a,δˉ(fa)=fdp∗a.(27)
We will need the Laplacian on such forms. Applying (26) to the two components of ∂ˉ(fa) gives
δˉ∂ˉ(fa)=fdp∗dpa−N1(f′′+2pf′)a−f′dr∧dp∗a,
∂ˉδˉ(fa)=fdpdp∗a+f′dr∧dp∗a.
The radial exterior terms cancel, so
□(fa)=fΔp,qa−N1(f′′+2pf′)a.(28)
This holds for arbitrary smooth functions f.
A Bochner identity on the compact quotient
We now define all the bundles and operators needed for the Bochner identity (5). The level hypersurface Σ gives a section σ of E: at the point represented by x∈Σ, its value is x∈V. The equivariance of Σ makes this section well defined on M. Let DE be the flat connection on E. Since DYEσ is the derivative of the immersion of Σ, it identifies TM with a subbundle of E. Moreover, dϕ(σ)=−N and dϕ(DYEσ)=0, so
E=Rσ⊕DEσ(TM).
In this subsection, a tangent vector Y regarded as an element of E means DYEσ. The cone metric on the slice defines a fiber metric G on E:
G(σ,σ)=N,G(σ,Y)=0,G(Y,Z)=k(Y,Z).(29)
The metric on ΛpE∗ induced by G is precisely the metric Hp defined in Section 3.1. Define ζ∈E∗ by ζ(σ)=1 and ζ(TM)=0. Then
ζ=G(σ,⋅)/N,ζ♯G=σ/N,∣ζ∣2=N−1.
For any local frame ea of TM, let θa∈E∗ annihilate σ and be dual to ea on TM. When working on the cone, we use the same symbols ζ,θa for their extensions by flat parallel transport along the rays, so their affine components are kept fixed. In a coordinate frame ea=∂ya, write σa=∂yaσ. The product parametrization Φ(r,y)=erσ(y) satisfies
dΦ=erσdr+erb∑σbdyb.
Applying the defining duality relations gives
Φ∗ζ=er(ζ(σ)dr+b∑ζ(σb)dyb)=erdr,
Φ∗θa=er(θa(σ)dr+b∑θa(σb)dyb)=erdya.
Thus, suppressing pullback by Φ, we write
ζ=erdr,θa=erdya.
In particular, θa1∧⋯∧θaq=eqrdya1∧⋯∧dyaq.
For 0≤p,q≤N, set
Bp,q=ΛpE∗⊗ΛqE∗,
with the tensor metric induced by G and the base measure dvolk. The inclusion of covectors annihilating σ gives an isometric bundle map
Jq:ΛpE∗⊗ΛqT∗M⟶Bp,q.(30)
It is the identity on the first factor. In a coordinate coframe,
Jq(λ⊗dya1∧⋯∧dyaq)=λ⊗θa1∧⋯∧θaq.
Note that
ΛqE∗=Jq(ΛqT∗M)⊕ζ∧Jq−1(Λq−1T∗M).(31)
Here the notation Jq in the single exterior factor denotes the same inclusion without the first-factor identity. Thus Jq∗ discards the terms containing a second-factor ζ and identifies the horizontal part with ordinary forms. In particular, Jq∗Jq=Id.
Let Dtot be the connection on Bp,q induced by DE on both factors, in other words
DXtot(λ⊗η)=(DXλ)⊗η+λ⊗DXη,
with the induced exterior-power connections on λ,η. In a local flat frame of E, this is ordinary differentiation of all components.
Define
Pp,qu=a=1∑nε(θa)Deatotu−(p+q)ε(ζ)u,(32)
Pp,q:Γ(Bp,q)⟶Γ(Bp+1,q).
Let Pp,q∗:Γ(Bp+1,q)⟶Γ(Bp,q) be its formal adjoint for the metrics induced by G and the volume dvolk, i.e.
(Pp,qu,v)M=(u,Pp,q∗v)M(33)
for all smooth sections u,v.
For the adjoint calculation below, we will also use the following identity
Lemma 3.3.The adjoint of exterior multiplication by ζ is the operator
ε(ζ)∗=ι(σ/N).
Proof. At each point, choose a G-orthonormal basis v0=σ/N,v1,…,vn, with dual basis ω0,…,ωn. Then ζ=ω0/N. On the orthonormal basis of exterior monomials, exterior multiplication by ω0 and contraction by v0 have transposed matrices. Hence
ε(ζ)∗=N1ε(ω0)∗=N1ι(v0)=ι(σ/N).
The identity extends to double forms since both operators act as the identity on the second exterior factor. □
We set exterior powers and corresponding operators to zero outside degrees 0,…,N. Define
Ap,q=Pp,qJq,Bp,q=Pp−1,q∗Jq.(34)
Their domains and targets are
Ap,q:Ωq(M;Fp)⟶Γ(Bp+1,q),
Bp,q:Ωq(M;Fp)⟶Γ(Bp−1,q).
Proposition 3.4. On Ωq(M;Fp) one has the differential-operator identity
Conversely, dilation invariance forces precisely this scaling of the affine components. Its inverse, denoted by R0, evaluates the full tensor at r=0.
We next compute the operator induced by ∂ on dilation invariant double forms. Put m=p+q. In local coordinates ya on M, we have
D∂r(Up,qu)=−me−mru,D∂a(Up,qu)=e−mrDatotu.
Using
∂=ε(dr)D∂r+a∑ε(dya)D∂a
we compute
∂(Up,qu)=e−(m+1)r(a∑ε(ϑa)Datotu−mε(ζ)u).
The expression in parentheses is Pp,qu by definition. Therefore,
∂(Up,qu)=Up+1,q(Pp,qu).(38)
To identify its adjoint operator, choose local frames of Bp−1,q and Bp,q over an open subset of M, and extend their elements invariantly to X using Up−1,q and Up,q, respectively. These extensions form local frames of the corresponding double-form bundles on X. In these invariant frames, the operator ∂ acts like
∂=Pp−1,q+ε(ζ)∂r.
Indeed, its action on the frame elements is given by ∂U=UP, while its action on their scalar coefficients follows from the Leibniz rule. The formal adjoint can now be computed in these same frames. We first make a remark about the metric used to compute the adjoint in this frame. Under flat affine transport, the metric on covectors scales by e2r. Hence the induced metric on double forms of bidegree (p,q) scales by e2(p+q)r. On the other hand, each invariant extension has the coefficient factor e−(p+q)r. Consequently, for u,v∈Γ∞(M,Bp,q),
Thus the invariant extension is a pointwise isometry, and the Gram matrices of the invariant frames are independent of r. Moreover,
dvolg=Ndrdvolk.
Integration by parts on compactly supported test forms therefore shows that the radial derivative of the scalar coefficients in these frames has formal adjoint
∂r∗=−∂r.
Furthermore,
ε(ζ)∗=ι(σ/N).
Consequently,
δ=Pp−1,q∗−ι(σ/N)∂r.
Here ∂r differentiates the scalar coefficients relative to the invariant frames. The derivatives of the frame elements themselves are already included in Pp−1,q. For u∈Γ∞(M,Bp,q), the coefficients of Up,qu relative to these frames are independent of r. It follows that
δ(Up,qu)=Up−1,q(Pp−1,q∗u).
Together with the corresponding formula for ∂, this gives
□(Up,qu)=Up,q(Lp,qu),(39)
where
Lp,q=Pp,q∗Pp,q+Pp−1,qPp−1,q∗.
We next compute □ on the extension of Jqη, where η∈Ωq(M;Fp). Recall that in a flat coefficient frame, if
η=I,A∑ηI;A(y)dxI⊗dyA
then the inclusion Jq replaces dyA by θA at the slice. Since θA=eqrdyA, in flat coordinates we have
Finally, we apply Jq∗ on both sides and, since Jq∗Jq=Id and the adjoints of Ap,q, Bp,q satisfy
Ap,q∗=Jq∗Pp,q∗andBp,q∗=Jq∗Pp−1,q,
we have the desired
Δp,q=Ap,q∗Ap,q+Bp,q∗Bp,q+(p−q−Np2)Id.
The formula for Qp,q(a) follows by taking the inner product (Δp,qa,a).
Computation of cohomology groups
The Bochner identity implies the following:
Proposition 3.5.For everya∈Ωq(M;Fp),
Qp,q(a)≥(Np(N−p)−q)∥a∥M2.(42)
Proof. Both squared norms on the right-hand side of (36) are nonnegative. Removing them gives (42).
The flat connections need not preserve the positive fiber metrics used in this estimate. We recall the passage from the corresponding Hodge Laplacian to flat cohomology.
Lemma 3.6.LetFbe a flat real vector bundle over a closed Riemannian manifold, equipped with any positive fiber metric. LetdFbe its flat exterior derivative anddF∗its formal adjoint. Every class inHq(M;F)has a unique representative in
kerΔF,q,ΔF,q=dFdF∗+dF∗dF.
Proof. Flatness gives dF2=0. The principal symbol of dF is exterior multiplication by a covector, tensored with IdF. Its adjoint symbol is contraction by the metric dual covector. The relation ε(ξ)ι(ξ♯)+ι(ξ♯)ε(ξ)=∣ξ∣2Id shows that the principal symbol for ΔF,q is ∣ξ∣2Id. Thus this operator is elliptic. It is formally self-adjoint and nonnegative, since
(ΔF,qa,a)=∥dFa∥2+∥dF∗a∥2.
On a closed manifold, self-adjoint elliptic spectral theory gives its finite-dimensional smooth kernel, its orthogonal projection P onto ker(ΔF,q), and a Green operator G, satisfying
Id=P+ΔFG=P+GΔF.
These operators preserve smooth forms. From dF2=(dF∗)2=0 one obtains, by expanding the products, dFΔF=ΔFdF and the corresponding identity for dF∗. The same commutation holds with G. Therefore, if dFa=0, then dFGa=GdFa=0, so
a=Pa+ΔFGa=Pa+dFdF∗Ga.
The harmonic form Pa represents the same cohomology class. If a harmonic form is exact, say a=dFb, then ∥a∥2=(dF∗a,b)=0. This proves uniqueness as well. □
Proof of Theorem B. Apply Lemma 3.6 to Fp with the metrics in Proposition 3.5. A harmonic representative a has Qp,q(a)=0. If q<p(N−p)/N, the coefficient in (42) is positive, and hence a=0. Therefore Hq(M;Fp)=0. □
The Euler characteristic in dimension four
We now turn the analytic estimate into the topological result in Theorem A. We recall explicitly a well-known fact about cohomology with flat coefficients.
Lemma 4.1. Let M be a closed oriented n-manifold and let F be a flat real vector bundle. Integration of the fiberwise evaluation pairing induces a nondegenerate pairing
Hq(M;F)×Hn−q(M;F∗)⟶R.(43)
Moreover,
χ(M;F)=rank(F)χ(M).(44)
Proof of Theorem A. Assume first that M is oriented. For n=4, the standard bundle E has rank N=5. Taking p=2 and p=3 in Theorem B gives
H0(M;F2)=H1(M;F2)=0,H0(M;F3)=H1(M;F3)=0,(45)
because p(5−p)/5=6/5 in both cases.
Let ν be a parallel nonzero volume form on E. Contraction with ν is a flat bundle isomorphism
Λ2E⟶Λ3E∗,v∧w⟼ι(w)ι(v)ν.
Consequently F2∗≅F3. Applying Lemma 4.1 and (45), we obtain
H4(M;F2)≅H0(M;F3)∗=0,H3(M;F2)≅H1(M;F3)∗=0.
The only possibly nonzero cohomology group of F2 is therefore H2(M;F2). Since rank(F2)=(25)=10, (44) gives
10χ(M)=χ(M;F2)=dimRH2(M;F2)≥0.
This proves both conclusions for oriented M.
If M is nonorientable, its orientation double cover M is again a closed properly convex projective four-manifold. The oriented case applies to M. Euler characteristic multiplies by the degree of a finite cover, so 2χ(M)=χ(M)≥0. □
The spectrum of Euler characteristics.
Example 4.2. The inequality is attained by convex projective four-tori. Take the positive cone C=R>05. Its projectivization is an open simplex. The determinant-one positive diagonal group is
A={diag(eu1,…,eu5):u1+⋯+u5=0}≅R4.
In logarithmic coordinates, A acts by translations on R4. A lattice Γ<A therefore acts freely and cocompactly on the simplex, with quotient T4. Its Euler characteristic is zero.
Corollary 4.3. The Euler characteristics of closed oriented properly convex real projective four-manifolds are exactly the non-negative even integers.
Proof. The splitting E=Rσ⊕DσE(TM) in § 4.3 gives a smooth bundle isomorphism TM⊕R≅E. Since E is flat, Chern–Weil theory and stability give p1(TM)=0. This is the present case of Kobayashi’s theorem [14 Theorem 4.16], based on Avez’s identity; see [16 Section 2.2]. The Hirzebruch signature theorem gives σ(M)=0. Writing b2=b2++b2−, this implies b2+=b2−, and Poincaré duality gives
χ(M)=2−2b1+b2=2−2b1+2b2+∈2Z.
It is nonnegative by Theorem A. Conversely, zero is realized in Example 4.2, and every positive even integer is realized by [16 Theorem 1.3]. □
Appendix A. Coordinate proof of Akagawa’s identities
We prove (16) for a positive Hessian metric g=Ddϕ. The argument is local and uses only the first derivatives of g. In particular, neither the canonical Monge–Ampère equation, nor completeness is assumed.
Metric derivatives and the Levi–Civita connection
Choose affine coordinates x1,…,xN and put Di=D∂i. Thus Didxj=0, and the covariant components of g and its cubic derivative are
gij=ϕij,Cijk=21ϕijk.
The tensor C is totally symmetric. We reserve ∇g for the Levi–Civita connection, to distinguish it from D. Since every first derivative of g is a third derivative of ϕ, its Christoffel symbols simplify to
For the density μ=det(gij), Jacobi’s determinant formula gives
∂iμ=21μgab∂igab=αiμ.
Fix a point x0. A constant linear change of affine coordinates makes gij(x0)=δij. These remain affine coordinates; they are not Riemannian normal coordinates, and the derivatives of g must still be retained. At this point,
Consequently ∇igω=Diω−Siω. At x0, its expression is
Si=j,k∑Cijk(ejik+eˉjiˉk).
The matrix Γi is self-adjoint for the relevant metric: lowering its upper index gives gkℓΓijℓ=Cijk, which is symmetric in j,k. Its induced action Si on exterior tensors is therefore self-adjoint. Since ∇g preserves their metric,
Take smooth test double forms supported in the affine chart. The integral of the ordinary derivative ∂i(μ⟨ω,η⟩g) is zero. Using ∂iμ=αiμ and (A.5), we get
Solving for the first term gives the exact local formal adjoint
Di∗=−Di−αi+2Si.(51)
Pointwise wedge multiplication has adjoints
ei∗=gijij,eˉi∗=gijˉiˉj.
It follows that, on input of bidegree (p,q),
δˉ=i,j∑Di∗(gijˉiˉj).
Here gijˉiˉj first maps to bidegree (p,q−1); the Di∗ that follows it acts on that lower bidegree. Expanding the derivative and then setting gij(x0)=δij gives
There are two cubic contributions in ∑iSiiˉi. The one acting only in the second factor vanishes:
T:=i,j,k∑Cijkeˉjiˉkiˉi=0.
Indeed, exchanging the dummy indices i,k leaves Cijk unchanged and replaces iˉkiˉi by iˉiiˉk=−iˉkiˉi, so T=−T. The mixed contribution remains. We obtain
For this second formula, Si acts after contraction on bidegree (p−1,q), and its pure first-factor contribution vanishes by the same symmetry argument.
Commuting contraction past the exterior derivatives
We now compute Λ∂+∂Λ at x0. The coefficients gij in Λ must again be differentiated before evaluating there. Using (A.4),
In the first line, operators in different factors commute, and Dk commutes with contractions in the affine frame. The exterior relation ιiek+ekιi=δikId gives the first term in the second line. Total symmetry of C and a renaming of indices rewrite its cubic term as
−2i,j,k∑Cijkejιkιˉi.
Comparison with (A.7) therefore gives
Λ∂+∂Λ=−δˉ+i∑αiιˉi=−δˉ+ιˉ(Xα).
At x0, Xα=∑iαi∂i, which explains the last expression. The mixed cubic terms agree exactly and cancel; only the trace α remains.
where cross-factor contractions commute and the final comparison uses (A.8). This proves both identities in (16). Since x0 was arbitrary and the operators are intrinsic, the identities hold on the entire Hessian manifold.
References
References
[1][Aka19] S. Akagawa, Vanishing theorems of $L^2$-cohomology groups on Hessian manifolds, J. Math. Soc. Japan 71 (2019), no. 1, 65–89. doi:10.2969/jmsj/73797739.arxiv.org/abs/1702.06678
[2][BDL18] S. A. Ballas, J. Danciger, and G.-S. Lee, Convex projective structures on nonhyperbolic three-manifolds, Geom. Topol. 22 (2018), no. 3, 1593–1646. doi:10.2140/gt.2018.22.1593.DOI
[3][CX01] J. Cao and F. Xavier, Kähler parabolicity and the Euler number of compact manifolds of non-positive sectional curvature, Math. Ann. 319 (2001), no. 3, 483–491. doi:10.1007/PL00004444.DOI
[4][CG86] J. Cheeger and M. Gromov, $L^2$-cohomology and group cohomology, Topology 25 (1986), no. 2, 189–215.
[5][CY82] S.-Y. Cheng and S.-T. Yau, The real Monge–Ampère equation and affine flat structures, in Proceedings of the 1980 Beijing Symposium on Differential Geometry and Differential Equations, Science Press, Beijing, 1982, 339–370.
[6]S.-S. Chern, On curvature and characteristic classes of a Riemann manifold, Abh. Math. Sem. Univ. Hamburg 20 (1955), 117–126. doi:10.1007/BF02960745.DOI
[7]M. W. Davis and A. L. Edmonds, Euler characteristics of generalized Haken manifolds, Algebr. Geom. Topol. 14 (2014), no. 6, 3701–3716. doi:10.2140/agt.2014.14.3701.DOI
[8]M. W. Davis and B. Okun, Vanishing theorems and conjectures for the ℓ²-homology of right-angled Coxeter groups, Geom. Topol. 5 (2001), 7–74. doi:10.2140/gt.2001.5.7.DOI
[9]Y. Deng and B. Wang, Linear Chern–Hopf–Thurston conjecture, arXiv:2405.12012 (2024).arxiv.org/abs/2405.12012
[10]D. J. F. Fox, A Schwarz lemma for Kähler affine metrics and the canonical potential of a proper convex cone, Ann. Mat. Pura Appl. (4) 194 (2015), no. 1, 1–42. doi:10.1007/s10231-013-0362-6.
[11]M. Gromov, Kähler hyperbolicity and L²-Hodge theory, J. Differential Geom. 33 (1991), no. 1, 263–292. doi:10.4310/jdg/1214446039.DOI
[12]F. E. A. Johnson and D. Kotschick, On the signature and Euler characteristic of certain four-manifolds, Math. Proc. Cambridge Philos. Soc. 114 (1993), no. 3, 431–437. doi:10.1017/S0305004100071711.DOI
[13]J. Jost and K. Zuo, Vanishing theorems for L²-cohomology on infinite coverings of compact Kähler manifolds and applications in algebraic geometry, Comm. Anal. Geom. 8 (2000), no. 1, 1–30. doi:10.4310/CAG.2000.v8.n1.a1.DOI
[14]S. Kobayashi, Projectively invariant distances for affine and projective structures, in Differential Geometry (Warsaw, 1979), Banach Center Publications 12, PWN, Warsaw, 1984, 127–152.
[16]S. Riolo, A. Seppi, and L. Slavich, Convex projective manifolds, symmetric spaces and geometric decompositions, arXiv:2607.05662.arxiv.org/abs/2607.05662