betaFeedback

WorksMathematics

Constant Chern holomorphic sectional curvature on two-step solvmanifolds

Abstract

Let M=Γ\GM=Γ\backslash G be a compact solvmanifold with a left-invariant Hermitian structure whose Lie algebra is two-step solvable. If the Chern holomorphic sectional curvature is constant, then the derived algebra is JJ-invariant and the Chern connection is flat. At the Lie-algebra level, we also prove that constant Chern holomorphic sectional curvature is impossible for a unimodular two-step solvable Hermitian Lie algebra of mixed Freibert-Swann type.

Introduction

Let (Mn,J,g)(M^n,J,g) be a Hermitian manifold with Chern curvature tensor RR. The Chern holomorphic sectional curvature is

H(X)=RXX‾XX‾∣X∣4,0≠X∈T1,0M.H(X)=\frac{R_{X\overline{X}X\overline{X}}}{\lvert X\rvert^4},\qquad0\ne X\in T^{1,0}M.

For Kähler metrics, HH determines RR. For a general Hermitian metric, it determines only the part of RR symmetric separately in its holomorphic and antiholomorphic indices. The classical rigidity conjecture asserted that on a compact Hermitian manifold, H≡cH\equiv c forces the metric to be Kähler if c≠0c\ne0 and Chern flat if c=0c=0. This is true for complex surfaces [2, 3, 24, 1]. In dimension three, Ma–Nie treated normal balanced metrics in the nonpositive range, and Chen–Li removed the normality assumption [21, 7]. Qin–Tian settled c≠0c\ne0 without additional assumptions, while Chen–Zheng settled c=0c=0 [22, 11]. The zero-curvature assertion, however, fails in higher dimensions: Qin–Tian [22] constructed compact balanced metrics with H≡0H\equiv0 but R≢0R\not\equiv0 in every complex dimension n≥7n\ge7. They also proved Kähler rigidity for c≤0c\le0 in Fujiki’s class CC and for c≠0c\ne0 on compact balanced fourfolds. The conjectured rigidity also holds for locally conformally Kähler metrics [6, 16] and Bismut-torsion-parallel metrics [8, 27, 28, 29]. Related results for Chern–Kähler-like, Bismut–Kähler-like, and certain pluriclosed metrics appear in [25, 23, 5]. Chen–Zheng also proved that for Hermitian connections with parallel torsion, nonzero constant holomorphic sectional curvature forces the torsion to vanish [9]. In a related direction, Tang [26] recently resolved the Yang–Zheng conjecture on constant real bisectional curvature

2020 Mathematics Subject Classification. 53C55.

Key words and phrases. Chern connection, holomorphic sectional curvature, Hermitian Lie algebra, solvmanifold, two-step solvable Lie algebra. [30, 32]. For a survey see [31]; related work on Chern curvature of twistor spaces and orthogonal complex structures on flat tori appears in [12, 18].

We consider left-invariant Hermitian metrics on compact solvmanifolds M=Γ\GM = \Gamma\backslash G. Let g=Lie⁡(G)\mathfrak{g} = \operatorname{Lie}(G) and g′=[g,g]\mathfrak{g}' = [\mathfrak{g},\mathfrak{g}]. The existence of the lattice implies that g\mathfrak{g} is unimodular. The role of Lie groups in flatness questions is already apparent in the theorems of Boothby and Hano [4, 15]. Li–Zheng proved rigidity for complex nilmanifolds and studied almost-abelian and related solvable groups [20, 19]. More directly relevant here is the theorem of Huang–Zheng [17]: a compact Lie-Hermitian solvmanifold with constant Chern holomorphic sectional curvature is Chern flat when g′\mathfrak{g}' is JJ-invariant. Even when g′\mathfrak{g}' is abelian, this invariance need not hold. We prove that for a unimodular two-step solvable Hermitian Lie algebra it follows from the constancy of HH itself. Huang–Zheng’s theorem then yields Chern flatness on compact quotients.

Set

h=g′∩Jg′,V=g′∩h⊥,W=(g′+Jg′)⊥.\mathfrak{h} = \mathfrak{g}' \cap J\mathfrak{g}', \qquad V = \mathfrak{g}' \cap\mathfrak{h}^{\perp}, \qquad W = (\mathfrak{g}' + J\mathfrak{g}')^{\perp}.

The Freibert–Swann decomposition [14, 13] is

g=h⊕(V⊕JV)⊕W,g′=h⊕V.(1)\mathfrak{g} = \mathfrak{h} \oplus(V \oplus JV) \oplus W, \qquad\mathfrak{g}' = \mathfrak{h} \oplus V. \tag*{(1)}

These summands are mutually orthogonal, except possibly VV and JVJV. The cases h=0\mathfrak{h} = 0, V=0V = 0 and W=0W = 0 are called pure types I, II and III, respectively; the type is mixed if all three spaces are nonzero. Throughout, two-step solvable means that g′\mathfrak{g}' is abelian, including g′=0\mathfrak{g}' = 0.

Our main result is the following.

Theorem 1.1. Let M=Γ\GM = \Gamma\backslash G be a compact Lie-Hermitian solvmanifold with two-step solvable Lie algebra. If its left-invariant Hermitian metric has constant Chern holomorphic sectional curvature, then

Jg′=g′,H≡0,R=0.J\mathfrak{g}' = \mathfrak{g}', \qquad H \equiv0, \qquad R = 0.

It remains to exclude V≠0V \ne0. The mixed case admits the following Lie-algebraic statement, without a lattice assumption.

Theorem 1.2. Let (g,J,g)(\mathfrak{g}, J, g) be a unimodular two-step solvable Hermitian Lie algebra. If h\mathfrak{h}, VV and WW in (1) are all nonzero, then its Chern holomorphic sectional curvature cannot be constant.

The proof separates the three cases with V≠0V \ne0. In pure type I, unimodularity yields a nilpotent commutative associative algebra, which the curvature condition excludes in the nonabelian case. In pure type III, the corresponding quotient algebra satisfies A2=AA^2 = A; a curvature trace identity gives a contradiction. In the mixed case, the same trace argument yields nilpotence, and further curvature identities eliminate VV.

Preliminaries

Write g1,0\mathfrak{g}^{1,0} for the ii-eigenspace of JJ in g⊗C\mathfrak{g} \otimes\mathbb{C}, and use a bar for complex conjugation. For matrices, M∗=M‾TM^* = \overline{M}^{T} and M−T=(MT)−1M^{-T} = (M^T)^{-1}. We follow the structure-constant conventions of [10, 17]. If e=(e1,…,en)e=(e_1,\ldots,e_n) is a frame of g1,0\mathfrak{g}^{1,0} with dual coframe φ=(φ1,…,φn)\varphi=(\varphi_1,\ldots,\varphi_n), then

[ei,ej]=∑kCijkek,[ei,ej‾]=∑k(Dkji‾ek−Dkijek‾).(2)[e_i,e_j]=\sum_k C^k_{ij}e_k,\qquad[e_i,\overline{e_j}]=\sum_k\left(\overline{D^i_{kj}}e_k-D^j_{ki}\overline{e_k}\right). \tag*{(2)}

For a unitary frame, the Chern curvature is [17 Lemma 2]

Rijˉkℓˉ=∑s(DkisDℓjs‾−DℓisDsjk‾−DsijDℓsk‾−DsjiDksℓ‾).(3)R_{i\bar{j}k\bar{\ell}}=\sum_s\left(D^s_{ki}\overline{D^s_{\ell j}}-D^s_{\ell i}\overline{D^k_{sj}}-D^j_{si}\overline{D^k_{\ell s}}-D^i_{sj}\overline{D^\ell_{ks}}\right). \tag*{(3)}

Only the part of RR symmetric in its two holomorphic and its two antiholomorphic entries enters HH. Put

R^ijˉkℓˉ=14(Rijˉkℓˉ+Rkjˉiℓˉ+Riℓˉkjˉ+Rkℓˉijˉ).(4)\widehat{R}_{i\bar{j}k\bar{\ell}}=\frac{1}{4}\left(R_{i\bar{j}k\bar{\ell}}+R_{k\bar{j}i\bar{\ell}}+R_{i\bar{\ell}k\bar{j}}+R_{k\bar{\ell}i\bar{j}}\right). \tag*{(4)}

Then H≡cH\equiv c is equivalent to [17 Lemma 1]

R^ijˉkℓˉ=c2(gijˉgkℓˉ+giℓˉgkjˉ).(5)\widehat{R}_{i\bar{j}k\bar{\ell}}=\frac{c}{2}\left(g_{i\bar{j}}g_{k\bar{\ell}}+g_{i\bar{\ell}}g_{k\bar{j}}\right). \tag*{(5)}

Lemma 2.1. Let (g,J,g)(\mathfrak{g},J,g) be a Hermitian Lie algebra. If 0≠x∈g0\ne x\in\mathfrak{g} satisfies [x,Jx]=0[x,Jx]=0, then H(x−iJx)≥0H(x-iJx)\ge0.

Proof. Take a unitary frame with e1=(x−iJx)/(2∣x∣)e_1=(x-iJx)/(\sqrt{2}|x|). Since [e1,e1‾]=0[e_1,\overline{e_1}]=0, (2) gives D11s=0D^s_{1 1}=0 for every ss. Formula (3) yields

H(e1)=R11ˉ11ˉ=∑s∣D11s∣2≥0.H(e_1)=R_{1\bar{1}1\bar{1}}=\sum_s\left|D^s_{11}\right|^2\ge0.

The Jacobi identity also gives the following relations [10 (5)]:

∑r(CikrDjrℓ+DjirDrkℓ−DjkrDriℓ)=0(6)\sum_r\left(C^r_{ik}D^\ell_{jr}+D^r_{ji}D^\ell_{rk}-D^r_{jk}D^\ell_{ri}\right)=0 \tag*{(6)}

and

∑r(CikrDrℓj‾−CrkjDrℓi‾+CrijDrℓk‾−DriℓDjrk‾+DrkℓDjri‾)=0.(7)\sum_r\left(C^r_{ik}\overline{D^j_{r\ell}}-C^j_{rk}\overline{D^i_{r\ell}}+C^j_{ri}\overline{D^k_{r\ell}}-D^\ell_{ri}\overline{D^k_{jr}}+D^\ell_{rk}\overline{D^i_{jr}}\right)=0. \tag*{(7)}

The unimodularity condition, which holds whenever GG has a compact quotient, is

∑s(Csis+Dsis)=0for every i.(8)\sum_s\left(C^s_{si}+D^s_{si}\right)=0\qquad\text{for every }i. \tag*{(8)}

For a possibly nonunitary frame, let

H=(gijˉ),Dk=(Diki)i,j.\mathcal{H}=(g_{i\bar{j}}),\qquad D_k=(D^i_{ik})_{i,j}.

Denote by AkA_k and BkB_k the matrices of ∇ek\nabla_{e_k} and ∇‾eˉk\overline{\nabla}_{\bar{e}_k} on g1,0\mathfrak{g}^{1,0}. The connection formula of [10 Lemma 1] gives

Ak=H−TDkTHT,Bk=−Dk,(9)A_k=\mathcal{H}^{-T}D_k^T\mathcal{H}^T,\qquad B_k=-D_k, \tag*{(9)}

and therefore

Rkℓˉ=AkBℓ−BℓAk−∑pDpℓk‾Ap+∑pDpkℓBp.(10)R_{k\bar{\ell}}=A_kB_\ell-B_\ell A_k-\sum_p\overline{D^k_{p\ell}}A_p+\sum_pD^\ell_{pk}B_p. \tag*{(10)}

The transposes in (9) matter when the frame is not unitary.

For a finite-dimensional real commutative associative algebra AA, write Lx(y)=x∘yL_x(y)=x\circ y and set A1=AA^1=A, Am+1=A∘AmA^{m+1}=A\circ A^m.

Lemma 2.2. Let AA be a finite-dimensional real commutative associative algebra.

  1. If tr⁡Lx=0\operatorname{tr}L_x=0 for all x∈Ax\in A, then AA is nilpotent.

  2. If AA is not nilpotent, there exists y∈Ay\in A with Ly2=Ly≠0L_y^2=L_y\ne0. In particular, this holds if A2=A≠0A^2=A\ne0.

Proof. Suppose that some LxL_x is not nilpotent. Write its minimal polynomial as mx(t)=tkq(t)m_x(t)=t^kq(t) with q(0)≠0q(0)\ne0. As tmax⁡{k,1}t^{\max\{k,1\}} and qq are coprime, there exists a real polynomial pp with p≡0(modtmax⁡{k,1})p\equiv0\pmod{t^{\max\{k,1\}}} and p≡1(modq)p\equiv1\pmod q. Thus p(0)=0p(0)=0 and p(Lx)p(L_x) is a nonzero idempotent. Associativity implies p(Lx)=Lyp(L_x)=L_y for some y∈Ay\in A, so

tr⁡Ly=tr⁡p(Lx)=rank⁡p(Lx)>0.\operatorname{tr}L_y=\operatorname{tr}p(L_x)=\operatorname{rank}p(L_x)>0.

If, instead, every LxL_x is nilpotent, the commuting family {Lx:x∈A}\{L_x:x\in A\} can be simultaneously strictly upper triangularized over C\mathbb{C}. Hence, with d=dim⁡Ad=\dim A, any product of dd such operators vanishes and Ad+1=0A^{d+1}=0. The trace hypothesis in (1) excludes the nonzero idempotent above, proving nilpotence, and nonnilpotence gives (2). Finally, A2=A≠0A^2=A\ne0 implies Am=AA^m=A for every m≥1m\ge1. □

Pure type I

Suppose that g′∩Jg′=0\mathfrak{g}'\cap J\mathfrak{g}'=0, and put A=g′A=\mathfrak{g}'.

Proposition 3.1. Let g\mathfrak{g} be a unimodular two-step solvable Lie algebra with complex structure JJ and g′∩Jg′=0\mathfrak{g}'\cap J\mathfrak{g}'=0. Define

x∘y=[Jx,y],x,y∈A=g′.(11)x\circ y=[Jx,y],\qquad x,y\in A=\mathfrak{g}'. \tag*{(11)}

Then (A,∘)(A,\circ) is a nilpotent commutative associative algebra.

Proof. The integrability of JJ reads

[X,Y]−[JX,JY]+J[JX,Y]+J[X,JY]=0.[X,Y]-[JX,JY]+J[JX,Y]+J[X,JY]=0.

The first two terms lie in AA, while the others lie in JAJA. Since A∩JA=0A\cap JA=0,

[JX,JY]=[X,Y],[JX,Y]=−[X,JY].(12)[JX,JY]=[X,Y],\qquad[JX,Y]=-[X,JY]. \tag*{(12)}

In particular, JJ is abelian and ∘\circ is commutative. Because AA is abelian, Jacobi gives [Lx,Ly]z=[[Jx,Jy],z]=0[L_x,L_y]z=[[Jx,Jy],z]=0 for x,y,z∈Ax,y,z\in A. Therefore

(x∘y)∘z=z∘(x∘y)=LzLxy=LxLzy=x∘(y∘z).(x\circ y)\circ z=z\circ(x\circ y)=L_zL_xy=L_xL_zy=x\circ(y\circ z).

Choose a JJ-invariant complement WW with g=A⊕JA⊕W\mathfrak{g}=A\oplus JA\oplus W. The map ad⁡Jx\operatorname{ad}_{Jx} restricts to LxL_x on AA, vanishes on JAJA, and maps WW into AA. Thus tr⁡(ad⁡Jx)=tr⁡Lx=0\operatorname{tr}(\operatorname{ad}_{Jx})=\operatorname{tr}L_x=0 by unimodularity. Lemma 2.2 proves that AA is nilpotent. □

Theorem 3.2. A nonabelian unimodular two-step solvable Hermitian Lie algebra of pure type I cannot have constant Chern holomorphic sectional curvature.

Proof. Let A=g′A=\mathfrak{g}' and use the orthogonal JJ-invariant splitting g=(A⊕JA)⊕W\mathfrak{g}=(A\oplus JA)\oplus W. By Proposition 3.1, the algebra AA is nonzero and nilpotent. If W=0W=0, then (12) gives A=g′=[JA,A]=A2A=\mathfrak{g}'=[JA,A]=A^2, a contradiction. Thus W≠0W\ne0. Choose a unitary frame adapted to (A⊕JA)1,0⊕W1,0(A\oplus JA)^{1,0}\oplus W^{1,0}, with a,b,…a,b,\ldots denoting W1,0W^{1,0}-indices. All brackets lie in ACA_{\mathbb{C}}, so

Daqp=0for all p,q and a∈W1,0.(13)D^{p}_{aq}=0 \qquad\text{for all }p,q\text{ and }a\in W^{1,0}. \tag*{(13)}

Also C=0C=0, since JJ is abelian. If H≡cH\equiv c, formula (3) gives

c=Raaˉaaˉ=−∑r∣Draa∣2≤0.(14)c=R_{a\bar{a}a\bar{a}}=-\sum_r\lvert D^a_{ra}\rvert^2\le0. \tag*{(14)}

Suppose c=0c=0. Then

Draa=0for all r,a.(15)D^a_{ra}=0 \qquad\text{for all }r,a. \tag*{(15)}

With C=0C=0, equation (7) becomes

∑r(−DriℓDjrk‾+DrkℓDjri‾)=0.\sum_r\left(-D^\ell_{ri}\overline{D^k_{jr}}+D^\ell_{rk}\overline{D^i_{jr}}\right)=0.

Take k=jk=j and sum over jj. The first term vanishes by (8), leaving

∑r,jDrjℓDjri‾=0.(16)\sum_{r,j}D^\ell_{rj}\overline{D^i_{jr}}=0. \tag*{(16)}

In particular,

∑r,kDrkaDkra‾=0.(17)\sum_{r,k}D^a_{rk}\overline{D^a_{kr}}=0. \tag*{(17)}

Since R^=0\widehat{R}=0 by (5), expanding 4R^aaˉkkˉ4\widehat{R}_{a\bar{a}k\bar{k}} with (3), (13), and (15) yields

0=∑r(∣Dkar∣2−∣Drak∣2−∣Drka∣2−2Re⁡(DrkaDkra‾)).(18)0=\sum_r\left(\lvert D^r_{ka}\rvert^2-\lvert D^k_{ra}\rvert^2-\lvert D^a_{rk}\rvert^2-2\operatorname{Re}(D^a_{rk}\overline{D^a_{kr}})\right). \tag*{(18)}

Summation over kk cancels the first two terms, and (17) removes the last one. Hence Drka=0D^a_{rk}=0 for all a,r,ka,r,k. As C=0C=0, we have [ea,ej]=0[e_a,e_j]=0; moreover [ea,ej‾][e_a,\overline{e_j}] has type (0,1)(0,1) by (2), and must vanish because it lies in ACA_{\mathbb{C}} and AC∩g0,1=0A_{\mathbb{C}}\cap\mathfrak{g}^{0,1}=0. Thus WW is central, giving A=g′=[JA,A]=A2A=\mathfrak{g}'=[JA,A]=A^2, again a contradiction. Finally, if c<0c<0, take 0≠x0\ne x in the last nonzero power of AA. Then x∘A=0x\circ A=0, in particular [Jx,x]=0[Jx,x]=0, which contradicts Lemma 2.1. □

Pure type III

Assume now that g′+Jg′=g\mathfrak{g}'+J\mathfrak{g}'=\mathfrak{g}. With h=g′∩Jg′\mathfrak{h}=\mathfrak{g}'\cap J\mathfrak{g}', decomposition (1) reduces to

g=h⊕(V⊕JV),g′=h⊕V.(19)\mathfrak{g}=\mathfrak{h}\oplus(V\oplus JV),\qquad\mathfrak{g}'=\mathfrak{h}\oplus V. \tag*{(19)}

Write r=dim⁡Chr=\dim_{\mathbb{C}}\mathfrak{h} and q=dim⁡RVq=\dim_{\mathbb{R}}V. Since a nonzero solvable Lie algebra cannot equal its derived algebra, q>0q>0.

Proposition 4.1. The space h\mathfrak{h} is a JJ-invariant abelian ideal. The quotient g‾=g/h\overline{\mathfrak{g}}=\mathfrak{g}/\mathfrak{h} has an abelian complex structure and satisfies

g‾=V⊕JV,g‾′=V.\overline{\mathfrak{g}}=V\oplus JV,\qquad\overline{\mathfrak{g}}'=V.

Moreover, VV admits a commutative associative product ∘\circ with V∘V=VV\circ V=V.

Proof. The space h\mathfrak{h} is JJ-invariant and abelian by definition. For u∈hu\in\mathfrak{h} and v∈g′v\in\mathfrak{g}', the integrability of JJ, together with u,Ju,v∈g′u,Ju,v\in\mathfrak{g}', gives

[Jv,Ju]=J[Jv,u].[Jv,Ju]=J[Jv,u].

Both sides lie in g′\mathfrak{g}', so [Jv,u]∈h[Jv,u]\in\mathfrak{h}. Since [g′,h]=0[\mathfrak{g}',\mathfrak{h}]=0 and g=g′+Jg′\mathfrak{g}=\mathfrak{g}'+J\mathfrak{g}', we have [g,h]⊂h[\mathfrak{g},\mathfrak{h}]\subset\mathfrak{h}. Thus JJ descends to g‾\overline{\mathfrak{g}}, where (19) gives g‾=V⊕JV\overline{\mathfrak{g}}=V\oplus JV and g‾′=g′/h=V\overline{\mathfrak{g}}'=\mathfrak{g}'/\mathfrak{h}=V. Here VV and JVJV denote their images in g‾\overline{\mathfrak{g}}. Their intersection is zero, and the argument of Proposition 3.1 shows that the quotient complex structure is abelian.

Choose a real basis xr+1,…,xnx_{r+1},\ldots,x_n of VV and set

eα=12(xα−iJxα),r+1≤α≤n.e_\alpha=\frac{1}{2}(x_\alpha-iJx_\alpha),\qquad r+1\leq\alpha\leq n.

Together with a unitary basis e1,…,ere_1,\ldots,e_r of h1,0\mathfrak{h}^{1,0}, these vectors form an admissible frame in the sense of [10]. The middle block need not be unitary. The structure relations of [10], Lemma 4 include

Cpqα=0,Di=0,Dαβα=0,Ciαj=−Djαi‾,Dγαβ=Dγβα∈iR.(20)C^\alpha_{pq}=0,\qquad D_i=0,\qquad D^\alpha_{\alpha\beta}=0,\qquad C^j_{i\alpha}=-\overline{D^i_{j\alpha}},\qquad D^\beta_{\gamma\alpha}=D^\alpha_{\gamma\beta}\in i\mathbb{R}. \tag*{(20)}

Write

Dγαβ=iaαβγ,aαβγ=aβαγ∈R,(21)D^\beta_{\gamma\alpha}=ia^\gamma_{\alpha\beta},\qquad a^\gamma_{\alpha\beta}=a^\gamma_{\beta\alpha}\in\mathbb{R}, \tag*{(21)}

and define

xα∘xβ=∑γaαβγxγ.(22)x_\alpha\circ x_\beta=\sum_\gamma a^\gamma_{\alpha\beta}x_\gamma. \tag*{(22)}

The product is commutative. Equation (6), restricted to middle indices, gives [Lα,Lβ]=0[L_\alpha,L_\beta]=0, and hence associativity. Finally, g‾′=V\overline{\mathfrak{g}}'=V is generated by [JV,V][JV,V], because VV and JVJV are both abelian. As

[Jxα,xβ]≡2∑γaαβγxγ(modh),[Jx_\alpha,x_\beta]\equiv2\sum_\gamma a^\gamma_{\alpha\beta}x_\gamma\pmod{\mathfrak{h}},

we conclude that V=V∘VV=V\circ V. □

Lemma 4.2. Let (g,J,g)(\mathfrak{g},J,\mathfrak{g}) be a nonabelian unimodular two-step solvable Hermitian Lie algebra of pure type III. Then h≠0\mathfrak{h}\ne0, and in the admissible frame above,

Di=0,Riy‾=Rx‾i=0(i≤r)D_i=0,\qquad R_{i\overline{y}}=R_{\overline{x}i}=0\quad(i\leq r)

for all x,y∈g1,0x,y\in\mathfrak{g}^{1,0}.

Proof. If r=0r=0, there are only middle indices. Thus C=0C=0, and (8) and (21) give

0=∑γDγαγ=itr⁡Lα.0=\sum_{\gamma}D^{\gamma}_{\gamma\alpha}=i\operatorname{tr}L_{\alpha}.

Hence tr⁡Lx=0\operatorname{tr}L_x=0 for all x∈Ax\in A, contrary to Lemma 2.2 since A2=A≠0A^2=A\ne0. Thus r>0r>0. For i≤ri\le r, (20) gives Di=0D_i=0 and therefore Ai=Bi=0A_i=B_i=0. In (10), all terms containing AiA_i or DpiyD^y_{pi} vanish. Since h\mathfrak h is an ideal, DpyiD^i_{py} can be nonzero only for an h\mathfrak h-index pp, for which Ap=0A_p=0. Thus Riyˉ=0R_{i\bar y}=0, and similarly Rxiˉ=0R_{x\bar i}=0.

Theorem 4.3. A nonabelian unimodular two-step solvable Hermitian Lie algebra of pure type III cannot have constant Chern holomorphic sectional curvature.

Proof. Suppose H≡cH\equiv c. By Lemma 4.2, Riiˉiiˉ=0R_{i\bar i i\bar i}=0 for i≤ri\le r, so c=0c=0 and R^=0\widehat R=0. The same lemma and (4) give

Rαβˉijˉ=0,1≤i,j≤r.(23)R_{\alpha\bar\beta i\bar j}=0,\qquad1\le i,j\le r. \tag*{(23)}

The metric matrix in the admissible frame is H=(Ir00G)\mathcal H=\begin{pmatrix}I_r&0\\0&G\end{pmatrix} with G=G∗>0G=G^*>0. For a middle index α\alpha, write

Dα=(PαFα0iLα),(Pα)ij=Diαj,(Fα)iγ=Diαγ.(24)D_\alpha=\begin{pmatrix}P_\alpha&F_\alpha\\0&iL_\alpha\end{pmatrix},\qquad(P_\alpha)_{ij}=D^j_{i\alpha},\qquad(F_\alpha)_{i\gamma}=D^\gamma_{i\alpha}. \tag*{(24)}

Equation (9) becomes

Aα=(PαT0G−TFαTiG−TLαGT),Bβ=(−Pβ‾−Fβ‾0iLβ).(25)A_\alpha=\begin{pmatrix}P_\alpha^T&0\\G^{-T}F_\alpha^T&iG^{-T}L_\alpha G^T\end{pmatrix},\qquad B_\beta=\begin{pmatrix}-\overline{P_\beta}&-\overline{F_\beta}\\0&iL_\beta\end{pmatrix}. \tag*{(25)}

Put Uα=FαTU_\alpha=F_\alpha^T and pα=tr⁡Pαp_\alpha=\operatorname{tr}P_\alpha. Taking the trace of the h1,0\mathfrak h^{1,0}-block of (10) and using (23), we obtain

0=tr⁡(Uβ∗G−TUα)+i∑γaαβγ(pγ−pγ‾).(26)0=\operatorname{tr}(U_\beta^*G^{-T}U_\alpha)+i\sum_\gamma a^\gamma_{\alpha\beta}(p_\gamma-\overline{p_\gamma}). \tag*{(26)}

The commutator of the PP-blocks has zero trace, and unimodularity with (20)–(21) gives

(pγ−pγ‾)+itr⁡Lγ=0,i(pγ−pγ‾)=tr⁡Lγ.(27)(p_\gamma-\overline{p_\gamma})+i\operatorname{tr}L_\gamma=0,\qquad i(p_\gamma-\overline{p_\gamma})=\operatorname{tr}L_\gamma. \tag*{(27)}

Since Lxα∘xβ=LαLβL_{x_\alpha\circ x_\beta}=L_\alpha L_\beta, it follows that

tr⁡(LαLβ)+tr⁡(Uβ∗G−TUα)=0.(28)\operatorname{tr}(L_\alpha L_\beta)+\operatorname{tr}(U_\beta^*G^{-T}U_\alpha)=0. \tag*{(28)}

For x=∑αtαxα∈Ax=\sum_\alpha t_\alpha x_\alpha\in A with tα∈Rt_\alpha\in\mathbb R, set Lx=∑αtαLαL_x=\sum_\alpha t_\alpha L_\alpha and Ux=∑αtαUαU_x=\sum_\alpha t_\alpha U_\alpha. Contracting (28) with tαtβt_\alpha t_\beta yields

tr⁡(Lx2)+tr⁡(Ux∗G−TUx)=0.(29)\operatorname{tr}(L_x^2)+\operatorname{tr}(U_x^*G^{-T}U_x)=0. \tag*{(29)}

Since G−TG^{-T} is positive definite,

tr⁡(Lx2)≤0for all x∈A.(30)\operatorname{tr}(L_x^2)\le0\qquad\text{for all }x\in A. \tag*{(30)}

But A2=A≠0A^{2}=A\ne0. Lemma 2.2 supplies y∈Ay\in A with Ly2=Ly≠0L_{y}^{2}=L_{y}\ne0, and then tr⁡(Ly2)=tr⁡Ly=rank⁡Ly>0\operatorname{tr}(L_{y}^{2})=\operatorname{tr}L_{y}=\operatorname{rank}L_{y}>0, contradicting (30). □

The mixed type

To prove Theorem 1.2, let g\mathfrak{g} be unimodular and assume

h≠0,V≠0,W≠0.\mathfrak{h}\ne0,\qquad V\ne0,\qquad W\ne0.

Put r=dim⁡Ch1,0r=\dim_{\mathbb{C}}\mathfrak{h}^{1,0}, q=dim⁡RVq=\dim_{\mathbb{R}}V, and m=dim⁡CW1,0m=\dim_{\mathbb{C}}W^{1,0}. Choose an admissible frame as in [10 Lemma 4]: e1,…,ere_{1},\ldots,e_{r} form a unitary basis of h1,0\mathfrak{h}^{1,0}, the middle vectors satisfy

V=span⁡R{eα+eˉα:r+1≤α≤r+q},V=\operatorname{span}_{\mathbb{R}}\{e_{\alpha}+\bar{e}_{\alpha}:r+1\le\alpha\le r+q\},

and the remaining eae_{a} form a unitary basis of W1,0W^{1,0}. Throughout this section, i,ji,j denote h\mathfrak{h}-indices; α,β,γ\alpha,\beta,\gamma denote middle indices; and a,ba,b denote WW-indices. The metric matrix has the form

H=diag⁡(Ir,G,Im),G=G∗>0.(31)\mathcal{H}=\operatorname{diag}(I_{r},G,I_{m}),\qquad G=G^{*}>0. \tag*{(31)}

For unrestricted indices p,qp,q, the structure relations in [10 Lemma 4] give

Cpqα=Cpqα=0,Daqp=0,Dpji=Dpjα=0,Dαβi=0.(32)C_{pq}^{\alpha}=C_{pq}^{\alpha}=0,\qquad D_{aq}^{p}=0,\qquad D_{pj}^{i}=D_{pj}^{\alpha}=0,\qquad D_{\alpha\beta}^{i}=0. \tag*{(32)}

and

Cjαi=−Diαj‾,Dγαβ=Dγβα∈iR,Dαyx=−Dαxy‾,(33)C_{j\alpha}^{i}=-\overline{D_{i\alpha}^{j}},\qquad D_{\gamma\alpha}^{\beta}=D_{\gamma\beta}^{\alpha}\in i\mathbb{R},\qquad D_{\alpha y}^{x}=-\overline{D_{\alpha x}^{y}}, \tag*{(33)}

where x,yx,y may range over the full frame. In particular,

Di=(00Qi00Yi000),Dα=(PαFαQα0iLαYα000).(34)\mathcal{D}_{i}= \begin{pmatrix} 0&0&Q_{i}\\ 0&0&Y_{i}\\ 0&0&0 \end{pmatrix}, \qquad \mathcal{D}_{\alpha}= \begin{pmatrix} P_{\alpha}&F_{\alpha}&Q_{\alpha}\\ 0&iL_{\alpha}&Y_{\alpha}\\ 0&0&0 \end{pmatrix}. \tag*{(34)}

Here Pα,Fα,Qα,Yα,LαP_{\alpha},F_{\alpha},Q_{\alpha},Y_{\alpha},L_{\alpha} have sizes r×rr\times r, r×qr\times q, r×mr\times m, q×mq\times m, and q×qq\times q, respectively.

Lemma 5.1. If H≡cH\equiv c in the mixed case, then c=0c=0. Moreover, in an admissible frame,

Dpab=0for all W-indices a,b and every p,(35)D_{pa}^{b}=0\qquad\text{for all }W\text{-indices }a,b\text{ and every }p, \tag*{(35)}

and the matrices Za=(Dija)i,j≤rZ_{a}=(D_{ij}^{a})_{i,j\le r} are skew-symmetric.

Proof. For a unit vector ei∈h1,0e_{i}\in\mathfrak{h}^{1,0}, (34) and (10) give

Riiˉiiˉ=∑a∣Diia∣2≥0.R_{i\bar{i}i\bar{i}}=\sum_{a}\lvert D_{ii}^{a}\rvert^{2}\ge0.

For a unit vector ea∈W1,0e_{a}\in W^{1,0}, the vanishing of the WW-rows in (32) gives Apea=0A_{p}e_{a}=0. With da=Daead_{a}=\mathcal{D}_{a}e_{a}, we obtain

Raaˉaaˉ=−daTHTda‾≤0.R_{a\bar{a}a\bar{a}}=-d_{a}^{T}\mathcal{H}^{T}\overline{d_{a}}\le0.

Both quantities equal cc, so c=0c=0 and Dpaa=0D_{pa}^{a}=0 for every pp. This holds for any unit vector of W1,0W^{1,0}. Polarizing DpZZ=0D_{pZ}^{Z}=0 with Z=∑azaeaZ=\sum_{a}z_{a}e_{a} gives ∑a,bz‾azbDp‾ba=0\sum_{a,b}\overline{z}_a z_bD^a_{\overline{p}b}=0 and hence (35). Replacing eie_i by an arbitrary u∈h1,0u\in\mathfrak{h}^{1,0} in the first identity yields Duua=uTZau=0D^a_{uu}=u^T Z_a u=0. Therefore ZaT=−ZaZ_a^T=-Z_a.

Write Dγαβ=iaαβγD^\beta_{\gamma\alpha}=ia^\gamma_{\alpha\beta} with aαβγ=aβαγ∈Ra^\gamma_{\alpha\beta}=a^\gamma_{\beta\alpha}\in\mathbb{R}. On A=VA=V define

xα∘xβ=∑γaαβγxγ,xα=eα+e‾α.(36)x_\alpha\circ x_\beta=\sum_\gamma a^\gamma_{\alpha\beta}x_\gamma,\qquad x_\alpha=e_\alpha+\overline{e}_\alpha. \tag*{(36)}

By (6), [Lα,Lβ]=0[L_\alpha,L_\beta]=0; as in Proposition 3.1, this makes AA a commutative associative algebra. In particular,

Lx∘y=LxLy,x,y∈A.(37)L_{x\circ y}=L_xL_y,\qquad x,y\in A. \tag*{(37)}

For x=∑αxαxα∈Ax=\sum_\alpha x^\alpha x_\alpha\in A with real coefficients, extend the matrices in (34) linearly to Px,Fx,Qx,Yx,LxP_x,F_x,Q_x,Y_x,L_x, and put Ux=FxTU_x=F_x^T and pα=tr⁡Pαp_\alpha=\operatorname{tr}P_\alpha. In curvature expressions, xx denotes the corresponding (1,0)(1,0)-vector ∑αxαeα\sum_\alpha x^\alpha e_\alpha. For each WW-index aa, introduce the q×rq\times r matrix KaK_a and the r×mr\times m matrix Ξx\Xi_x by

(Ka)αi=Dαia,(Ξx)ia=∑αxα(GKa)αi.(38)(K_a)_{\alpha i}=D^a_{\alpha i},\qquad(\Xi_x)_{ia}=\sum_\alpha x^\alpha(GK_a)_{\alpha i}. \tag*{(38)}

Thus QxQ_x and Ξx\Xi_x are both r×mr\times m. Unimodularity and (32)–(33) give

(pα−p‾α)+itr⁡Lα=0,i(pα−p‾α)=tr⁡Lα.(39)(p_\alpha-\overline{p}_\alpha)+i\operatorname{tr}L_\alpha=0,\qquad i(p_\alpha-\overline{p}_\alpha)=\operatorname{tr}L_\alpha. \tag*{(39)}

Lemma 5.2. For every real x∈Ax\in A,

4∑i=1rRxx‾ii‾=tr⁡(Lx2)+tr⁡(Ux∗G−TUx)+∥Qx+Ξx∥HS2.(40)4\sum_{i=1}^{r}R_{x\overline{x}i\overline{i}}=\operatorname{tr}(L_x^2)+\operatorname{tr}(U_x^*G^{-T}U_x)+\lVert Q_x+\Xi_x\rVert_{\mathrm{HS}}^2. \tag*{(40)}

Here ∥⋅∥HS\lVert\cdot\rVert_{\mathrm{HS}} denotes the Hilbert–Schmidt norm.

Proof. The connection matrices in the middle direction xx are

Ax=(PxT00G−TFxTiG−TLxGT0QxTYxTGT0),Bx=(−P‾x−F‾x−Q‾x0iLx−Y‾x000).A_x=\begin{pmatrix} P_x^T & 0 & 0\\ G^{-T}F_x^T & iG^{-T}L_xG^T & 0\\ Q_x^T & Y_x^TG^T & 0 \end{pmatrix},\qquad B_x=\begin{pmatrix} -\overline{P}_x & -\overline{F}_x & -\overline{Q}_x\\ 0 & iL_x & -\overline{Y}_x\\ 0 & 0 & 0 \end{pmatrix}.

For i≤ri\le r they are

Ai=(000000QiTYiTGT0),Bi=(00−Q‾i00−Y‾i000).A_i=\begin{pmatrix} 0 & 0 & 0\\ 0 & 0 & 0\\ Q_i^T & Y_i^TG^T & 0 \end{pmatrix},\qquad B_i=\begin{pmatrix} 0 & 0 & -\overline{Q}_i\\ 0 & 0 & -\overline{Y}_i\\ 0 & 0 & 0 \end{pmatrix}.

Taking the trace over the unitary basis of h1,0\mathfrak{h}^{1,0} in (10), the h\mathfrak{h}-block of the first commutator contributes

tr⁡(−PxTP‾x+P‾xPxT+F‾xG−TFxT+Q‾xQxT)=tr⁡(Ux∗G−TUx)+∥Qx∥HS2.\operatorname{tr}\left(-P_x^T\overline{P}_x+\overline{P}_xP_x^T+\overline{F}_xG^{-T}F_x^T+\overline{Q}_xQ_x^T\right)=\operatorname{tr}\left(U_x^*G^{-T}U_x\right)+\lVert Q_x\rVert_{\mathrm{HS}}^2.

By (39) and (37), the bracket-coefficient contribution is

i∑γ(x∘x)γ(pγ−p‾γ)=tr⁡Lx∘x=tr⁡(Lx2).i\sum_\gamma(x\circ x)^\gamma(p_\gamma-\overline{p}_\gamma)=\operatorname{tr}L_{x\circ x}=\operatorname{tr}(L_x^2).

The four terms in the symmetrization are therefore

∑iRxxˉiˉiˉ=tr⁡(Lx2)+tr⁡(Ux∗G−TUx)+∥Qx∥HS2,(41)\sum_i R_{x\bar{x}\bar{i}\bar{i}}=\operatorname{tr}(L_x^2)+\operatorname{tr}(U_x^*G^{-T}U_x)+\lVert Q_x\rVert_{\mathrm{HS}}^2, \tag*{(41)}
∑iRixˉiˉxˉ=tr⁡(Qx∗Ξ‾x),(42)\sum_i R_{i\bar{x}\bar{i}\bar{x}}=\operatorname{tr}(Q_x^*\overline{\Xi}_x), \tag*{(42)}
∑iRxiˉiˉx=tr⁡(Ξ‾x∗Qx),(43)\sum_i R_{x\bar{i}\bar{i}x}=\operatorname{tr}(\overline{\Xi}_x^*Q_x), \tag*{(43)}
∑iRiiˉxxˉ=∥Ξx∥HS2.(44)\sum_i R_{i\bar{i}x\bar{x}}=\lVert\Xi_x\rVert_{\mathrm{HS}}^2. \tag*{(44)}

Indeed, the mixed contractions are ∑i,a(Qx)ia(Ξ‾x)ia\sum_{i,a}(Q_x)_{ia}(\overline{\Xi}_x)_{ia} and its conjugate, while the last one is ∑i,a∣(Ξ‾x)ia∣2\sum_{i,a}|(\overline{\Xi}_x)_{ia}|^2. Adding the four identities proves (40). □

Lemma 5.3. The algebra A=(V,∘)A=(V,\circ) is nilpotent. Moreover, for every real x∈Ax\in A,

Fx=0,Qx+Ξ‾x=0.(45)F_x=0,\qquad Q_x+\overline{\Xi}_x=0. \tag*{(45)}

Proof. Lemma 5.1 gives c=0c=0, hence R^=0\widehat{R}=0. Since G−TG^{-T} is positive definite, (40) implies tr⁡(Lx2)≤0\operatorname{tr}(L_x^2)\leq0 for all x∈Ax\in A. If AA were not nilpotent, Lemma 2.2 would give y∈Ay\in A with Ly2=Ly≠0L_y^2=L_y\neq0 and hence tr⁡(Ly2)=rank⁡Ly>0\operatorname{tr}(L_y^2)=\operatorname{rank}L_y>0, a contradiction. Thus AA is nilpotent and tr⁡(Lx2)=0\operatorname{tr}(L_x^2)=0 for every xx. The other two terms in (40) are nonnegative, so both vanish, proving (45). □

Let Ann⁡(A)={z∈A:z∘A=0}\operatorname{Ann}(A)=\{z\in A:z\circ A=0\}. For every WW-index aa, put

Sa=(Dαβa)α,β,Yx=(S1x,…,Smx).(46)S_a=(D^a_{\alpha\beta})_{\alpha,\beta},\qquad Y_x=(S_1x,\ldots,S_mx). \tag*{(46)}

We use ∥v∥G−12=v∗G−1v\lVert v\rVert_{G^{-1}}^2=v^*G^{-1}v for v∈Cqv\in\mathbb{C}^q and ∥v∥G2=vTGv=v∗Gv\lVert v\rVert_G^2=v^TGv=v^*Gv for real vv. In particular, ∥v∥G>0\lVert v\rVert_G>0 for every nonzero real vv.

Lemma 5.4. For u∈A2u\in A^2 and z∈Ann⁡(A)z\in\operatorname{Ann}(A),

uTGz=0.(47)u^TGz=0. \tag*{(47)}

Consequently A2=0A^2=0.

Proof. By (45), the h\mathfrak{h}-block decouples when all four curvature arguments are middle vectors. The middle block of (10) is

−G−TLxTGTLx+LxG−TLxTGT+Y‾xYxTGT−G−T(Lx2)TGT−Lx2.-G^{-T}L_x^TG^TL_x+L_xG^{-T}L_x^TG^T+\overline{Y}_xY_x^TG^T-G^{-T}(L_x^2)^TG^T-L_x^2.

Contracting with the real middle vector xx gives

Rxxˉxˉxˉ=∥LxTGx∥G−12+∑a∣xTGSax∣2−∥x∘x∥G2−2Re⁡(xTGLx2x).(48)R_{x\bar{x}\bar{x}\bar{x}}=\lVert L_x^TGx\rVert_{G^{-1}}^2+\sum_a\lvert x^TGS_ax\rvert^2-\lVert x\circ x\rVert_G^2-2\operatorname{Re}(x^TGL_x^2x). \tag*{(48)}

For z∈Ann⁡(A)z\in\operatorname{Ann}(A), Lz=0L_z=0, and (48) with x=zx=z gives

zTGSaz=0for every a.(49)z^TGS_az=0\qquad\text{for every }a. \tag*{(49)}

Fix y∈Ay \in A and replace xx by y+tzy+tz, t∈Rt \in\mathbb{R}. Then Lx=LyL_x=L_y, x∘x=y∘yx\circ x=y\circ y, and Ly2z=0L_y^2z=0. By (49), the coefficient of t2t^2 in (48) is

∥LyTGz∥G−12+∑a∣zTGSay+yTGSaz∣2.\left\|L_y^T Gz\right\|_{G^{-1}}^2+\sum_a\left|z^TGS_ay+y^TGS_az\right|^2.

Since Rxxˉxˉx=0R_{x\bar{x}\bar{x}x}=0, this coefficient vanishes. Positive definiteness gives LyTGz=0L_y^TGz=0 for every yy, and therefore (y∘u)TGz=0(y\circ u)^TGz=0 for all y,u∈Ay,u\in A. This proves (47). If A2≠0A^2\ne0 and N≥3N\ge3 is the nilpotency index, then 0≠AN−1⊂A2∩Ann⁡(A)0\ne A^{N-1}\subset A^2\cap\operatorname{Ann}(A). Taking 0≠z∈AN−10\ne z\in A^{N-1} and u=zu=z in (47) gives zTGz=0z^TGz=0, a contradiction. □

Fix a WW-index aa and set

Ea=(Dpka)p,k≤r+q=(ZaQ(a)KaSa),H0=diag⁡(Ir,G).(50)E_a=\left(D^a_{pk}\right)_{p,k\le r+q}= \begin{pmatrix} Z_a & Q^{(a)}\\ K_a & S_a \end{pmatrix}, \qquad\mathcal{H}_0=\operatorname{diag}(I_r,G). \tag*{(50)}

where Za=(Dija)Z_a=\left(D^a_{ij}\right) and Q(a)=(Diαa)Q^{(a)}=\left(D^a_{i\alpha}\right). Since A2=0A^2=0, all of AA belongs to Ann⁡(A)\operatorname{Ann}(A), and (48) implies xTGSax=0x^TGS_ax=0 for every real x∈Ax\in A. Thus

(GSa)T=−GSa.(51)(GS_a)^T=-GS_a. \tag*{(51)}

Also ZaT=−ZaZ_a^T=-Z_a by Lemma 5.1, while (45) and (38) give Q(a)=−(GKa)TQ^{(a)}=-(GK_a)^T. Consequently,

(H0Ea)T=−H0Ea.(52)(\mathcal{H}_0E_a)^T=-\mathcal{H}_0E_a. \tag*{(52)}

Lemma 5.5. Suppose H≡0H\equiv0, Dpab=0D^b_{pa}=0 for all WW-indices a,ba,b and every pp, and (H0Ea)T=−H0Ea(\mathcal{H}_0E_a)^T=-\mathcal{H}_0E_a. Then Ea=0E_a=0 for every aa.

Proof. We can make the non-WW block unitary without changing the WW-frame, since the curvature contraction below is tensorial. Choose TT satisfying

T∈GL⁡(r+q,C),TTH0T‾=I,T\in\operatorname{GL}(r+q,\mathbb{C}),\qquad T^T\mathcal{H}_0\overline{T}=I,

and put ep′=∑kTkpeke'_p=\sum_kT_{kp}e_k for p≤r+qp\le r+q. The matrices with upper index aa transform as

Ea′=T‾−1EaT=TTH0EaT,E'_a=\overline{T}^{-1}E_aT=T^T\mathcal{H}_0E_aT,

which is skew-symmetric by (52). This change of frame need not preserve the real middle space, but it preserves the curvature trace and the vanishing of the WW-rows.

In the resulting unitary frame, sum the four terms of 4R^aaˉkkˉ=04\widehat{R}_{a\bar{a}k\bar{k}}=0 over k≤r+qk\le r+q. By (3),

0=∑k,p≤r+q(∣Dka′p∣2−∣Dpa′k∣2)−∥Ea′∥HS2−2Re⁡tr⁡(Ea′Ea′‾).0=\sum_{k,p\le r+q}\left(\left|D'^p_{ka}\right|^2-\left|D'^k_{pa}\right|^2\right)-\left\|E'_a\right\|_{\mathrm{HS}}^2-2\operatorname{Re}\operatorname{tr}\left(E'_a\overline{E'_a}\right).

The double sums cancel. Skew-symmetry gives

tr⁡(Ea′Ea′‾)=∑p,kEa,pk′Ea,kp′‾=−∥Ea′∥HS2.\operatorname{tr}\left(E'_a\overline{E'_a}\right)=\sum_{p,k}E'_{a,pk}\overline{E'_{a,kp}}=-\left\|E'_a\right\|_{\mathrm{HS}}^2.

It follows that ∥Ea′∥HS2=0\left\|E'_a\right\|_{\mathrm{HS}}^2=0, and hence Ea=0E_a=0. □

Proof of Theorem 1.2. By Lemma 5.3, AA is nilpotent. Lemma 5.4 gives A2=0A^2=0, and Lemma 5.5 then implies Ea=0E_a=0 for all aa. Equations (34), (45), and (50) reduce the matrices to

Di=0,Dα=(Pα00000000),Da=(∗∗0000000).\mathcal{D}_i=0,\qquad\mathcal{D}_{\alpha}=\begin{pmatrix} P_{\alpha} & 0 & 0 \\ 0 & 0 & 0 \\ 0 & 0 & 0 \end{pmatrix},\qquad\mathcal{D}_a=\begin{pmatrix} * & * & 0 \\ 0 & 0 & 0 \\ 0 & 0 & 0 \end{pmatrix}.

Thus Dαkj=0D^j_{\alpha k}=0 for every middle first lower index α\alpha and all j,kj,k. Also Cjkα=0C^\alpha_{jk}=0 by (32). The bracket formula (2) therefore shows that no Lie bracket has a real VV-component, contrary to g′=h⊕V\mathfrak{g}'=\mathfrak{h}\oplus V with V≠0V\ne0. □

Proof of Theorem 1.1. The abelian case is Chern flat. Otherwise, if h=0\mathfrak{h}=0 we apply Theorem 3.2; if W=0W=0, Theorem 4.3; and if h\mathfrak{h}, VV, and WW are all nonzero, Theorem 1.2, since a compact quotient implies unimodularity. Hence V=0V=0, or Jg′=g′J\mathfrak{g}'=\mathfrak{g}'. The result of Huang–Zheng [17] gives H≡0H\equiv0 and R=0R=0. □

The nonexistence arguments use the lattice only through unimodularity, and therefore also give the following Lie-algebraic conclusion.

Corollary 5.6. Let (g,J,g)(\mathfrak{g},J,g) be a unimodular two-step solvable Hermitian Lie algebra with constant Chern holomorphic sectional curvature. Then Jg′=g′J\mathfrak{g}'=\mathfrak{g}'.

Theorem 1.1 concerns the original invariant metric. It makes no assertion about non-invariant metrics on the underlying complex solvmanifold, since averaging does not preserve constant holomorphic sectional curvature.

Acknowledgments. We thank Professor Fangyang Zheng for his guidance and helpful discussions.

Generative AI disclosure. The authors used OpenAI ChatGPT as an assisting tool in the development of the paper. The authors take full responsibility for the contents of the paper.

Declaration on competing interests. All authors declare that there are no competing interests for this article.

References

School of Mathematical Sciences, Chongqing Normal University, Chongqing 401331, China

Email address: [email protected]

School of Mathematical Sciences, Chongqing Normal University, Chongqing 401331, China

Email address: [email protected]

Email address: [email protected]

References

  1. [1]V. Apostolov, J. Davidov, and O. Muškarov, Compact self-dual Hermitian surfaces, Trans. Amer. Math. Soc. 348 (1996), no. 8, 3051–3063. doi:10.1090/S0002-9947-96-01585-1.DOI
  2. [2]A. Balas, Compact Hermitian manifolds of constant holomorphic sectional curvature, Math. Z. 189 (1985), no. 2, 193–210. doi:10.1007/BF01175044.DOI
  3. [3]A. Balas and P. Gauduchon, Any Hermitian metric of constant non-positive (Hermitian) holomorphic sectional curvature on a compact complex surface is Kähler, Math. Z. 190 (1985), no. 1, 39–43. doi:10.1007/BF01159161.DOI
  4. [4]W. M. Boothby, Hermitian manifolds with zero curvature, Michigan Math. J. 5 (1958), no. 2, 229–233. doi:10.1307/mmj/1028998068.DOI
  5. [5]K. Broder and K. Tang, On Hermitian manifolds with vanishing curvature, Math. Z. 309 (2025), no. 1, Paper No. 12, 14 pp. doi:10.1007/s00209-024-03651-0.DOI
  6. [6]H. Chen, L. Chen, and X. Nie, Chern–Ricci curvatures, holomorphic sectional curvature and Hermitian metrics, Sci. China Math. 64 (2021), no. 4, 763–780. doi:10.1007/s11425-019-9566-y.
  7. [7]S. Chen and J. Li, Compact balanced threefolds and LCK manifolds with constant holomorphic sectional curvature, preprint (2026), arXiv:2608.05598.arxiv.org/abs/2608.05598
  8. [8]S. Chen and F. Zheng, Bismut torsion parallel metrics with constant holomorphic sectional curvature, to appear in Chinese Ann. Math. Ser. B; arXiv:2405.09110v2.arxiv.org/abs/2405.09110
  9. [9]S. Chen and F. Zheng, Hermitian connections with parallel torsion and constant holomorphic sectional curvature, preprint (2026), arXiv:2609.33374.arxiv.org/abs/2609.33374
  10. [10]S. Chen and F. Zheng, Streets–Tian conjecture holds for 2-step solvmanifolds, J. Geom. Phys. 209 (2025), Paper No. 105390, 13 pp. doi:10.1016/j.geomphys.2024.105390.DOI
  11. [11]S. Chen and F. Zheng, Hermitian threefolds with constant holomorphic sectional curvature, preprint (2026), arXiv:2609.39282.arxiv.org/abs/2609.39282
  12. [12]J. Davidov, G. Grantcharov, and O. Muškarov, Curvature properties of the Chern connection of twistor spaces, Rocky Mountain J. Math. 39 (2009), no. 1, 27–48. doi:10.1216/RMJ-2009-39-1-27.DOI
  13. [13]M. Freibert and A. Swann, Compatibility of balanced and SKT metrics on two-step solvable Lie groups, Transform. Groups 30 (2025), no. 1, 235–265. doi:10.1007/s00021-023-09796-2.
  14. [14]M. Freibert and A. Swann, Two-step solvable SKT shears, Math. Z. 299 (2021), nos. 3–4, 1703–1739. doi:10.1007/s00209-021-02753-3.DOI
  15. [15]J.-I. Hano, On Kaehlerian homogeneous spaces of unimodular Lie groups, Amer. J. Math. 79 (1957), no. 4, 885–900. doi:10.2307/2372440.DOI
  16. [16]Z. Huang and X. Wan, Compact locally conformal Kähler manifolds with constant Chern holomorphic sectional curvature, preprint (2026), arXiv:2606.24425.arxiv.org/abs/2606.24425
  17. [17]X. Huang and F. Zheng, On solvmanifolds with the Chern–Ricci commutator and constant holomorphic sectional curvature, Internat. J. Math. 36 (2025), no. 6, Paper No. 2550006. doi:10.1142/S0129167X25500065.DOI
  18. [18]G. Khan, B. Yang, and F. Zheng, The set of all orthogonal complex structures on the flat 6-tori, Adv. Math. 319 (2017), 451–471. doi:10.1016/j.aim.2017.08.037.DOI
  19. [19]Y. Li and F. Zheng, A note on almost abelian groups with constant holomorphic sectional curvature, Proc. Amer. Math. Soc. 153 (2025), no. 12, 5369–5383. doi:10.1090/proc/17390.DOI
  20. [20]Y. Li and F. Zheng, Complex nilmanifolds with constant holomorphic sectional curvature, Proc. Amer. Math. Soc. 150 (2022), no. 1, 319–326. doi:10.1090/proc/15724.DOI
  21. [21]F. Ma and X. Nie, A remark on compact balanced threefolds with constant holomorphic sectional curvature, Kodai Math. J. 47 (2024), no. 1, 90–98. doi:10.2996/kmj47106.
  22. [22]X. Qin and Y. Tian, Chern holomorphic sectional curvature: canonical ampleness, rigidity, and flatness counterexamples, preprint (2026), arXiv:2609.25762v5.arxiv.org/abs/2609.25762
  23. [23]P. P. Rao and F. Zheng, Pluriclosed manifolds with constant holomorphic sectional curvature, Acta Math. Sin. (Engl. Ser.) 38 (2022), no. 6, 1094–1104. doi:10.1007/s10114-022-1046-1.DOI
  24. [24]T. Sato and K. Sekigawa, Hermitian surfaces of constant holomorphic sectional curvature, Math. Z. 205 (1990), no. 4, 659–668. doi:10.1007/BF02571270.DOI
  25. [25]K. Tang, Holomorphic sectional curvature and Kähler-like metric (in Chinese), Sci. Sin. Math. 51 (2021), no. 12, 2013–2024. doi:10.1360/SSM-2020-0038.DOI
  26. [26]K. Tang, Hermitian manifolds with constant holomorphic sectional curvature or real bisectional curvature, preprint (2026), arXiv:2609.14392.arxiv.org/abs/2609.14392
  27. [27]H. Wang, Bismut-torsion-parallel Hermitian manifolds with constant Chern holomorphic sectional curvature, preprint (2026), arXiv:2608.13386v2.arxiv.org/abs/2608.13386
  28. [28]Q. Wang and F. Zheng, Balanced Bismut torsion-parallel fourfolds with constant holomorphic sectional curvature, J. Geom. Anal. 36 (2026), Paper No. 377. doi:10.1007/s12220-026-02620-9.DOI
  29. [29]Q. Wang and F. Zheng, Balanced Bismut torsion-parallel manifold with constant holomorphic sectional curvature, preprint (2026), arXiv:2609.26089.arxiv.org/abs/2609.26089
  30. [30]X. Yang and F. Zheng, On real bisectional curvature for Hermitian manifolds, Trans. Amer. Math. Soc. 371 (2019), no. 4, 2703–2718. doi:10.1090/tran/7445.
  31. [31]F. Zheng, Constant holomorphic sectional curvature conjecture and Fino-Vezzoni conjecture, Surv. Differ. Geom. 28 (2023), no. 1, 289–314. doi:10.4310/sdg.260101003933.DOI
  32. [32]W. Zhou and F. Zheng, Hermitian threefolds with vanishing real bisectional curvature (in Chinese), Sci. Sin. Math. 52 (2022), no. 7, 757–764; English version: arXiv:2103.04296. doi:10.1360/SCM-2021-0109.DOI

Paper details

vlnc:W_aq4k5zvs
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

OpenAI ChatGPT

View statement
Looking for related papers…
Contents