This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Strong cohomological rigidity of Hirzebruch surface bundles in Bott towers

Hiroaki Ishida Department of Mathematics and Computer Science, Graduate School of Science and Engineering, Kagoshima University [email protected]
Abstract.

We show the strong cohomological rigidity of Hirzebruch surface bundles over Bott manifolds. As a corollary, we have that the strong cohomological rigidity conjecture is true for Bott manifolds of dimension 88.

Key words and phrases:
Bott manifold, Bott tower, cohomological rigidity, toric manifolds
2020 Mathematics Subject Classification:
Primary 57S12; Secondary 57R19, 57S25, 14M25
This work is supported by JSPS KAKENHI Grant Number JP20K03592

1. Introduction

A Bott tower of height nn is an iterated P1\mathbb{C}P^{1}-bundle

(1.1) Bn{B_{n}}Bn1{B_{n-1}}{\cdots}B1{B_{1}}B0={a point}{B_{0}=\{\text{a point}\}}πn\scriptstyle{\pi_{n}}πn1\scriptstyle{\pi_{n-1}}π2\scriptstyle{\pi_{2}}π1\scriptstyle{\pi_{1}}

where each fibration πj:BjBj1\pi_{j}\thinspace\colon B_{j}\to B_{j-1} is the projectivization of a Whitney sum of two complex line bundles over Bj1B_{j-1}. Each BjB_{j} is called a Bott manifold. The notion of the Bott tower is introduced in [GK1994]. By definition, BjB_{j} is a closed manifold of dimension 2j2j and B1B_{1} is nothing but the projective line P1\mathbb{C}P^{1}. Bott manifolds of dimension 44 are known as Hirzebruch surfaces and introduced in [Hirzebruch1951].

By composing the first kk fibrations πn,,πnk+1\pi_{n},\dots,\pi_{n-k+1}, we have a fiber bundle BnBnkB_{n}\to B_{n-k} whose fibers are Bott manifold of dimension 2k2k. Moreover the cohomology H(Bn)H^{*}(B_{n}) of BnB_{n} has a structure of H(Bnk)H^{*}(B_{n-k})-algebra. In this paper, we clarify the relation between the topology of Hirzebruch surface bundles over a Bott manifold BnB_{n} and the structure of the cohomology as H(Bn)H^{*}(B_{n})-algebras. The main theorem of this paper is the following.

Theorem 1.1.

Let

Bn+2{B_{n+2}}Bn+1{B_{n+1}}Bn{B_{n}}{\cdots}B1{B_{1}}B0={a point}{B_{0}=\{\text{a point}\}}πn+2\scriptstyle{\pi_{n+2}}πn+1\scriptstyle{\pi_{n+1}}πn\scriptstyle{\pi_{n}}π2\scriptstyle{\pi_{2}}π1\scriptstyle{\pi_{1}}

and

Bn+2{B_{n+2}^{\prime}}Bn+1{B_{n+1}^{\prime}}Bn{B_{n}}{\cdots}B1{B_{1}}B0={a point}{B_{0}=\{\text{a point}\}}πn+2\scriptstyle{\pi_{n+2}^{\prime}}πn+1\scriptstyle{\pi_{n+1}^{\prime}}πn\scriptstyle{\pi_{n}}π2\scriptstyle{\pi_{2}}π1\scriptstyle{\pi_{1}}

be Bott towers of height n+2n+2. Let φ:H(Bn+2)H(Bn+2)\varphi\thinspace\colon H^{*}(B_{n+2})\to H^{*}(B_{n+2}^{\prime}) be an isomorphism as H(Bn)H^{*}(B_{n})-algebras. Then, there exists a bundle isomorphism f:Bn+2Bn+2f\thinspace\colon B_{n+2}^{\prime}\to B_{n+2} over BnB_{n} such that f=φf^{*}=\varphi.

This study is motivated by the strong cohomological rigidity conjecure posed in [Choi2015].

Conjecture 1.2 (Strong cohomological rigidity conjecture for Bott manifolds).

Any graded ring isomorphism between the cohomology rings of two Bott manifolds is induced by a diffeomorphism.

So far, no counterexamples are known to Conjecture 1.2 and some partial affirmative results are known. Conjecture 1.2 is known to be true for Bott manifolds of dimension up to 66 (see [Choi2015, Theorem A]). Conjecture 1.2 is also true if we restrict Bott manifolds to \mathbb{Q}-trivial (see [CM2012, Corollary 5.1]) or /2\mathbb{Z}/2\mathbb{Z}-trivial ([CMM2015, Theorem 1.3]) ones. A Bott manifold BnB_{n} is said to be \mathbb{Q}-trivial (respectively, /2\mathbb{Z}/2\mathbb{Z}-trivial) if H(Bn)H((P1)n)H^{*}(B_{n})\otimes\mathbb{Q}\cong H^{*}((\mathbb{C}P^{1})^{n})\otimes\mathbb{Q} (respectively, H(Bn)/2H((P1)n)/2H^{*}(B_{n})\otimes\mathbb{Z}/2\mathbb{Z}\cong H^{*}((\mathbb{C}P^{1})^{n})\otimes\mathbb{Z}/2\mathbb{Z}). In [Choi2015, Problem 3.5], it is pointed out that Theorem 1.1 implies that Conjecture 1.2 is also true for Bott manifolds of dimension 88.

This paper is organized as follows. In Section 2, we recall some preliminary facts about Bott manifolds that we need. In Section 3, we study the topology of Hirzebruch surfaces. Especially, we show an S1S^{1}-equivariant version of the cohomological rigidity of a Hirzebruch surface. Sections 4, 5, 6 are devoted to prove Theorem 1.1. In Section 4, we study the automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra. In Section 5, we show that any automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra is induced by a bundle automorphism. In Section 6, we show that isomorphism classes of Hirzebruch surface bundles over a Bott manifold BnB_{n} are distinguished by their cohomologies as H(Bn)H^{*}(B_{n})-algebras. These results obtained in Sections 5 and 6 imply Theorem 1.1 immediately.

Throughout this paper, all cohomologies are taken with \mathbb{Z}-coefficient.

Acknowledgement. The author is grateful to the anonymous referee for the invaluable comments and useful suggestions on improving the text.

2. Preliminaries

In this section we recall elementary facts about Bott manifolds for later use.

Lemma 2.1 ([CMS2010, Lemma 2.1]).

Let BB be a smooth manifold. Let LL be a complex line bundle over BB and VV a complex vector bundle over BB. Then, the projectivizations of VV and LVL\otimes V are isomorphic as bundles over BB.

Thanks to Lemma 2.1 we may assume that one of line bundles over BjB_{j} is the product line bundle ¯\underline{\mathbb{C}} in (1.1) without loss of generality. Throughout this paper, we assume that each fibration BjBj1B_{j}\to B_{j-1} is a projectivization of P(¯ξj)Bj1P(\underline{\mathbb{C}}\oplus\xi_{j})\to B_{j-1}, where ¯\underline{\mathbb{C}} is a product line bundle and ξj\xi_{j} is a complex line bundle over Bj1B_{j-1}.

Let BB be a smooth manifold and VV a complex vector bundle over BB. The tautological line bundle γ\gamma of P(V)P(V) is a line bundle over P(V)P(V) given by

γ={(,v)P(V)×Vv},\gamma=\{(\ell,v)\in P(V)\times V\mid\ell\ni v\},

here we think of P(V)P(V) as a set of all lines passing 0 in VV. The first Chern classes of tautological bundles play a role to describe the cohomology ring of Bott manifolds. Lemma 2.2 and Theorem 2.3 below are well known facts. For leader’s convenience, we give brief proofs.

Lemma 2.2.

Let BB be a smooth manifold. Let ξ\xi be a complex line bundle over BB. Then H(P(¯ξ))H^{*}(P(\underline{\mathbb{C}}\oplus\xi)) is isomorphic to H(B)[X]/(X2c1(ξ)X)H^{*}(B)[X]/(X^{2}-c_{1}(\xi)X) as an H(B)H^{*}(B)-algebra, where degX=2\deg X=2.

Proof.

Let γ\gamma be the tautological line bundle of P(¯ξ)P(\underline{\mathbb{C}}\oplus\xi). Let xBx\in B and (¯ξ)x(\underline{\mathbb{C}}\oplus\xi)_{x} the fiber of ¯ξ\underline{\mathbb{C}}\oplus\xi at xx. Then, the restriction of γ\gamma to P((¯ξ)x)P((\underline{\mathbb{C}}\oplus\xi)_{x}) is nothing but the tautological line bundle over the projective line P((¯ξ)x)P((\underline{\mathbb{C}}\oplus\xi)_{x}). Since H(P((¯ξ)x))H^{*}(P((\underline{\mathbb{C}}\oplus\xi)_{x})) is generated by 11 and the first Chern class of the tautological line bundle as a \mathbb{Z}-module, Leray-Hirsch Theorem (see [Hatcher2002] for example) yields that H(P(¯ξ))H^{*}(P(\underline{\mathbb{C}}\oplus\xi)) is generated by 11 and c1(γ)c_{1}(\gamma) as an H(B)H^{*}(B)-module.

Let π:P(¯ξ)B\pi\thinspace\colon P(\underline{\mathbb{C}}\oplus\xi)\to B be the projection. We show that c1(γ)(c1(γ)+πc1(ξ))=0c_{1}(\gamma)(-c_{1}(\gamma)+\pi^{*}c_{1}(\xi))=0. Let γ\gamma^{\perp} be the orthogonal complement of γ\gamma in ¯ξ\underline{\mathbb{C}}\oplus\xi for some hermitian metric. Since γγ=π(¯ξ)\gamma\oplus\gamma^{\perp}=\pi^{*}(\underline{\mathbb{C}}\oplus\xi), by comparing Chern classes we have that c1(γ)=c1(γ)+πc1(ξ)c_{1}(\gamma^{\perp})=-c_{1}(\gamma)+\pi^{*}c_{1}(\xi) and c1(γ)c1(γ)=0c_{1}(\gamma)c_{1}(\gamma^{\perp})=0. Thus we have c1(γ)(c1(γ)+πc1(ξ))=0c_{1}(\gamma)(-c_{1}(\gamma)+\pi^{*}c_{1}(\xi))=0.

The computation above shows that the homomorphism H(B)[X]/(X2c1(ξ)X)H(P(¯ξ))H^{*}(B)[X]/(X^{2}-c_{1}(\xi)X)\to H^{*}(P(\underline{\mathbb{C}}\oplus\xi)) given by Xc1(γ)X\mapsto c_{1}(\gamma) is an isomorphism as H(B)H^{*}(B)-algebras. ∎

Let

B:Bn{B_{\bullet}\thinspace\colon B_{n}}Bn1{B_{n-1}}{\cdots}B1{B_{1}}B0={a point}{B_{0}=\{\text{a point}\}}πn\scriptstyle{\pi_{n}}πn1\scriptstyle{\pi_{n-1}}π2\scriptstyle{\pi_{2}}π1\scriptstyle{\pi_{1}}

be a Bott tower of height nn. Let γj\gamma_{j} be the tautological line bundle over Bj=P(¯ξj)B_{j}=P(\underline{\mathbb{C}}\oplus\xi_{j}). For integers j,kj,k with 1jkn1\leq j\leq k\leq n, we define

xj(k):=(πj+1πk)(c1(γj))H2(Bk),αj(k):=(πjπk)(c1(ξj))H2(Bk).\begin{split}x_{j}^{(k)}&:=(\pi_{j+1}\circ\dots\circ\pi_{k})^{*}(c_{1}(\gamma_{j}))\in H^{2}(B_{k}),\\ \alpha_{j}^{(k)}&:=(\pi_{j}\circ\dots\circ\pi_{k})^{*}(c_{1}(\xi_{j}))\in H^{2}(B_{k}).\end{split}

By applying Lemma 2.2 eventually, we have the following.

Theorem 2.3.

Let BkB_{k} be the Bott manifold as above. The followings hold:

  1. (1)

    αj(k)\alpha_{j}^{(k)} is a linear combination of x(k)x_{\ell}^{(k)}, =1,,j1\ell=1,\dots,j-1.

  2. (2)

    H(Bk)=[x1(k),,xk(k)]/((xj(k))2αj(k)xj(k)j=1,,k)H^{*}(B_{k})=\mathbb{Z}[x_{1}^{(k)},\dots,x_{k}^{(k)}]/((x_{j}^{(k)})^{2}-\alpha_{j}^{(k)}x_{j}^{(k)}\mid j=1,\dots,k).

Put xj:=xj(n)x_{j}:=x_{j}^{(n)}. It follows from Theorem 2.3 that H(Bn)H^{*}(B_{n}) is generated by degree 22 elements x1,,xnx_{1},\dots,x_{n} and they form a basis of H2(Bn)H^{2}(B_{n}). We call them the standard generators of H(Bn)H^{*}(B_{n}) with respect to ξ1,,ξn\xi_{1},\dots,\xi_{n}.

Suppose that

αj(k)==1j1ajx(k),aj.\alpha_{j}^{(k)}=\sum_{\ell=1}^{j-1}a_{j}^{\ell}x_{\ell}^{(k)},\quad a_{j}^{\ell}\in\mathbb{Z}.

Let γj(k)\gamma_{j}^{(k)} be the line bundle over BkB_{k} given by

γj(k):=(πj+1πk)γj.\gamma_{j}^{(k)}:=(\pi_{j+1}\circ\dots\circ\pi_{k})^{*}\gamma_{j}.

Since c1(γj(k))=xj(k)c_{1}(\gamma_{j}^{(k)})=x_{j}^{(k)} and isomorphism classes of line bundles are distinguished by their first Chern classes, we have that the line bundle ξj\xi_{j} over Bj1B_{j-1} is isomorphic to (γ1(j1))aj1(γj1(j1))ajj1(\gamma_{1}^{(j-1)})^{\otimes a_{j}^{1}}\otimes\dots\otimes(\gamma_{j-1}^{(j-1)})^{\otimes a_{j}^{j-1}}. For simplicity, we assume that ξj=(γ1(j1))aj1(γj1(j1))ajj1\xi_{j}=(\gamma_{1}^{(j-1)})^{\otimes a_{j}^{1}}\otimes\dots\otimes(\gamma_{j-1}^{(j-1)})^{\otimes a_{j}^{j-1}}. Let ρj:(S1)kS1\rho_{j}\thinspace\colon(S^{1})^{k}\to S^{1} be the homomorphism given by ρj(t)==1j1taj\rho_{j}(t)=\prod_{\ell=1}^{j-1}t_{\ell}^{a_{j}^{\ell}} for t=(t1,,tk)(S1)kt=(t_{1},\dots,t_{k})\in(S^{1})^{k}. We use the same symbol even if kk is different. Let MkM_{k} be the quotient manifold (S3)k/(S1)k(S^{3})^{k}/(S^{1})^{k} by the action of (S1)k(S^{1})^{k} on (S3)k(S^{3})^{k} given by

(t1,,tk)((z1,w1),,(zk,wk))=((t1z1,t1ρ1(t)1w1),,(tkz1,tkρk(t)1wk)).\begin{split}&(t_{1},\dots,t_{k})\cdot((z_{1},w_{1}),\dots,(z_{k},w_{k}))\\ &=((t_{1}z_{1},t_{1}\rho_{1}(t)^{-1}w_{1}),\dots,(t_{k}z_{1},t_{k}\rho_{k}(t)^{-1}w_{k})).\end{split}

Let pk:MkMk1p_{k}\thinspace\colon M_{k}\to M_{k-1} be the map induced by the projection (S3)k(S3)k1(S^{3})^{k}\to(S^{3})^{k-1}. Then pk:MkMk1p_{k}\thinspace\colon M_{k}\to M_{k-1} is a P1\mathbb{C}P^{1}-bundle. Therefore we have the iterated P1\mathbb{C}P^{1}-bundle

M:Mn{M_{\bullet}\thinspace\colon M_{n}}Mn1{M_{n-1}}{\cdots}M1{M_{1}}M0={a point}.{M_{0}=\{\text{a point}\}.}pn\scriptstyle{p_{n}}pn1\scriptstyle{p_{n-1}}p2\scriptstyle{p_{2}}p1\scriptstyle{p_{1}}

Let LkL_{k} be the quotient ((S3)k×)/(S1)k((S^{3})^{k}\times\mathbb{C})/(S^{1})^{k} by the action of (S1)k(S^{1})^{k} on (S3)k×(S^{3})^{k}\times\mathbb{C} given by

(t1,,tk)((z1,w1),,(zk,wk),v)=((t1z1,t1ρ1(t)1w1),,(tkz1,tkρk(t)1wk),tk1v).\begin{split}&(t_{1},\dots,t_{k})\cdot((z_{1},w_{1}),\dots,(z_{k},w_{k}),v)\\ &=((t_{1}z_{1},t_{1}\rho_{1}(t)^{-1}w_{1}),\dots,(t_{k}z_{1},t_{k}\rho_{k}(t)^{-1}w_{k}),t_{k}^{-1}v).\end{split}

For a homomorphism ρ:(S1)kS1\rho\thinspace\colon(S^{1})^{k}\to S^{1}, let LρL_{\rho} be the ((S3)k×)/(S1)k((S^{3})^{k}\times\mathbb{C})/(S^{1})^{k} by the action of (S1)k(S^{1})^{k} on (S3)k×(S^{3})^{k}\times\mathbb{C} given by

(t1,,tk)((z1,w1),,(zk,wk),v)=((t1z1,t1ρ1(t)1w1),,(tkz1,tkρk(t)1wk),ρ(t)1v).\begin{split}&(t_{1},\dots,t_{k})\cdot((z_{1},w_{1}),\dots,(z_{k},w_{k}),v)\\ &=((t_{1}z_{1},t_{1}\rho_{1}(t)^{-1}w_{1}),\dots,(t_{k}z_{1},t_{k}\rho_{k}(t)^{-1}w_{k}),\rho(t)^{-1}v).\end{split}

LkL_{k} and LρL_{\rho} are line bundles over MkM_{k}.

Proposition 2.4 (See also [CR2005, Section 3]).

For all kk there exists a diffeomorphism fk:MkBkf_{k}\thinspace\colon M_{k}\to B_{k} such that πkfk=fk1pk\pi_{k}\circ f_{k}=f_{k-1}\circ p_{k}. Moreover, there exists a line bundle isomorphism f~k:Lkγk\widetilde{f}_{k}\thinspace\colon L_{k}\to\gamma_{k} which induces fkf_{k}.

Proof.

Induction on kk. Consider the case when k=1k=1. Since ξ1\xi_{1} is a line bundle over a point, we have that ξ1\xi_{1} is the product line bundle and a1(0)=0a_{1}^{(0)}=0. Therefore ρ1(t)=1\rho_{1}(t)=1 and hence B1B_{1} and M1M_{1} are nothing but P1\mathbb{C}P^{1}.

The tautological line bundle γ\gamma of P1\mathbb{C}P^{1} is

γ={([z,w],(u,v))P1×2(u,v)[z,w]}.\gamma=\{([z,w],(u,v))\in\mathbb{C}P^{1}\times\mathbb{C}^{2}\mid(u,v)\in[z,w]\}.

The smooth map S3×S3×2S^{3}\times\mathbb{C}\to S^{3}\times\mathbb{C}^{2} given by ((z,w),λ)((z,w),(λz,λw))((z,w),\lambda)\mapsto((z,w),(\lambda z,\lambda w)) induces the isomorphism f~1\widetilde{f}_{1} of line bundles between L1M1=B1L_{1}\to M_{1}=B_{1} and γB1\gamma\to B_{1}.

Suppose that the proposition holds for k1k-1. For =1,,k1\ell=1,\dots,k-1, let LL_{\ell}^{\prime} be the pull-back line bundle (pk1p+1)L(p_{k-1}\circ\dots\circ p_{\ell+1})^{*}L_{\ell}. By the induction hypothesis, LL_{\ell}^{\prime} is isomorphic to γ(k1)\gamma_{\ell}^{(k-1)}. Since ξk\xi_{k} is isomorphic to (γ1(k1))ak1(γk1(k1))akk1(\gamma_{1}^{(k-1)})^{\otimes a_{k}^{1}}\otimes\dots\otimes(\gamma_{k-1}^{(k-1)})^{\otimes a_{k}^{k-1}}, we have that

fk1(ξk)(L1)ak1(Lk1)akk1.f_{k-1}^{*}(\xi_{k})\cong(L_{1}^{\prime})^{\otimes a_{k}^{1}}\otimes\dots\otimes(L_{k-1}^{\prime})^{\otimes a_{k}^{k-1}}.

On the other hand, (L1)ak1(Lk1)akk1(L_{1}^{\prime})^{\otimes a_{k}^{1}}\otimes\dots\otimes(L_{k-1}^{\prime})^{\otimes a_{k}^{k-1}} is isomorphic to LρkL_{\rho_{k}} as line bundles over Mk1M_{k-1}. Therefore there exists a line bundle isomorphism gk:Lρkξkg_{k}\thinspace\colon L_{\rho_{k}}\to\xi_{k} that induces fk1f_{k-1}. The vector bundle isomorphism between ¯Lρk\underline{\mathbb{C}}\oplus L_{\rho_{k}} and ¯ξk\underline{\mathbb{C}}\oplus\xi_{k} induces the diffeomorphism fk:MkBkf_{k}\thinspace\colon M_{k}\to B_{k} such that πkfk=fk1pk\pi_{k}\circ f_{k}=f_{k-1}\circ p_{k}. The smooth map (S3)k×(S3)k×2(S^{3})^{k}\times\mathbb{C}\to(S^{3})^{k}\times\mathbb{C}^{2} given by

((z1,w1),,(zk,wk),λ)((z1,w1),,(zk,wk),(λzk,λwk))((z_{1},w_{1}),\dots,(z_{k},w_{k}),\lambda)\mapsto((z_{1},w_{1}),\dots,(z_{k},w_{k}),(\lambda z_{k},\lambda w_{k}))

induces the isomorphism between LkL_{k} and the tautological line bundle of P(¯Lρk)=MkP(\underline{\mathbb{C}}\oplus L_{\rho_{k}})=M_{k}. Remark that fk(γk)f_{k}^{*}(\gamma_{k}) is isomorphic to the tautological line bundle of MkM_{k}. By composing, we have that there exists a line bundle isomorphism f~k:Lkγk\widetilde{f}_{k}\thinspace\colon L_{k}\to\gamma_{k} which induces fkf_{k}.

The proposition is proved. ∎

Theorem 2.5 ([Ishida2012, Theorem 3.1]).

Rank 22 decomposable vector bundles over a Bott manifold are distinguished by their total Chern classes.

Let

B:Bn{B_{\bullet}^{\prime}\thinspace\colon B_{n}^{\prime}}Bn1{B_{n-1}^{\prime}}{\cdots}B1{B_{1}^{\prime}}B0={a point}{B_{0}^{\prime}=\{\text{a point}\}}πn\scriptstyle{\pi_{n}^{\prime}}πn1\scriptstyle{\pi_{n-1}^{\prime}}π2\scriptstyle{\pi_{2}^{\prime}}π1\scriptstyle{\pi_{1}^{\prime}}

be another Bott tower of height nn such that πj:BjBj1\pi_{j}^{\prime}\thinspace\colon B_{j}^{\prime}\to B_{j-1}^{\prime} is a projectivization P(¯ξ)Bj1P(\underline{\mathbb{C}}\oplus\xi^{\prime})\to B_{j-1}^{\prime}, where ξj\xi_{j}^{\prime} is a complex line bundle over Bj1B_{j-1}^{\prime}. Let x1,,xnx_{1}^{\prime},\dots,x_{n}^{\prime} be the standard generators of H(Bn)H^{*}(B_{n}^{\prime}) with respect to ξ1,,ξn\xi_{1}^{\prime},\dots,\xi_{n}^{\prime}.

Theorem 2.6 ([Ishida2012, Theorem 1.1]).

Let φ:H(Bn)H(Bn)\varphi\thinspace\colon H^{*}(B_{n})\to H^{*}(B_{n}^{\prime}) be an isomorphism as graded algebras. Assume that the representation matrix of φ\varphi with respect to x1,,xnx_{1},\dots,x_{n} and x1,,xnx_{1}^{\prime},\dots,x_{n}^{\prime} is an upper triangular matrix. Then, there exists a diffeomorphism fk:BkBkf_{k}\thinspace\colon B_{k}^{\prime}\to B_{k} such that πkfk=fk1πk\pi_{k}\circ f_{k}=f_{k-1}\circ\pi_{k}^{\prime} for all kk.

3. The Hirzebruch surfaces

Let aa\in\mathbb{Z}. The Hirzebruch surface Σa\Sigma_{a} is the total space of the projectivization of the rank 22 vector bundle over P1\mathbb{C}P^{1}:

Σa=P(γa)P1.\Sigma_{a}=P(\mathbb{C}\oplus\gamma^{\otimes a})\to\mathbb{C}P^{1}.

In particular, Σa\Sigma_{a} is a Bott manifold of dimension 44. By Theorem 2.3, the cohomology algebra H(Σa)H^{*}(\Sigma_{a}) is of the form

H(Σa)=[x1,x2]/(x12,x2(x2ax1)).H^{*}(\Sigma_{a})=\mathbb{Z}[x_{1},x_{2}]/(x_{1}^{2},x_{2}(x_{2}-ax_{1})).

The strong cohomological rigidity for Hirzebruch surfaces is already shown in [CM2012]. In this section, we review it briefly and study the certain diffeomorphism of Σa\Sigma_{a}.

First we shall review the classification of the diffeomorphism types of Hirzebruch surfaces given in [Hirzebruch1951] briefly. Let a,ba,b\in\mathbb{Z}. By direct computation the isomorphism type of the cohomology algebra H(Σa)H^{*}(\Sigma_{a}) is determined by the parity of aa. In case when aa is even, say a=2ka=2k, then we have that

Σ2k=P(¯γ2k)P(γ(k)γk)(by Lemma 2.1)P(¯¯)(by Theorem 2.5)=Σ0=P1×P1.\begin{split}\Sigma_{2k}&=P(\underline{\mathbb{C}}\oplus\gamma^{\otimes 2k})\\ &\cong P(\gamma^{\otimes(-k)}\otimes\gamma^{\otimes k})\quad(\text{by Lemma \ref{lemm:tensor}})\\ &\cong P(\underline{\mathbb{C}}\oplus\underline{\mathbb{C}})\quad(\text{by Theorem \ref{theo:decomposable}})\\ &=\Sigma_{0}=\mathbb{C}P^{1}\times\mathbb{C}P^{1}.\end{split}

In case when aa is odd, say a=2k+1a=2k+1, then we have that

Σ2k+1=P(¯γ2k+1)P(γ(k)γk+1)(by Lemma 2.1)P(¯γ)(by Theorem 2.5)=Σ1.\begin{split}\Sigma_{2k+1}&=P(\underline{\mathbb{C}}\oplus\gamma^{\otimes 2k+1})\\ &\cong P(\gamma^{\otimes(-k)}\otimes\gamma^{\otimes k+1})\quad(\text{by Lemma \ref{lemm:tensor}})\\ &\cong P(\underline{\mathbb{C}}\oplus\gamma)\quad(\text{by Theorem \ref{theo:decomposable}})\\ &=\Sigma_{1}.\end{split}

Thus the diffeomorphism types of Hirzebruch surfaces are determined by the parity of aa, and hence they are determined by the isomorphism types of the cohomologies.

We shall see the group of automorphisms of H(Σa)H^{*}(\Sigma_{a}). Since x1x_{1} and x2x_{2} form a basis of H2(Σa)H^{2}(\Sigma_{a}) and H(Σa)H^{*}(\Sigma_{a}) is generated by x1x_{1} and x2x_{2}, the automorphism of H(Σa)H^{*}(\Sigma_{a}) is determined by the images of x1x_{1} and x2x_{2}. By direct computations one can see that there are 88 automorphisms of H(Σa)H^{*}(\Sigma_{a}). Among them, there are 44 diffeomorphisms whose representation matrices with respect to x1,x2x_{1},x_{2} are the following upper triangular matrices

(1001),(1001),(1a01),(1a01).\begin{pmatrix}1&0\\ 0&1\end{pmatrix},\begin{pmatrix}-1&0\\ 0&-1\end{pmatrix},\begin{pmatrix}1&a\\ 0&-1\end{pmatrix},\begin{pmatrix}-1&-a\\ 0&1\end{pmatrix}.

Remaining 44 automorphisms have different forms by the parity of aa:

  • In case when aa is even, representation matrices are

    ±(a2a2411a2),±(a2a24+11a2).\pm\begin{pmatrix}\frac{a}{2}&\frac{a^{2}}{4}-1\\ -1&-\frac{a}{2}\end{pmatrix},\pm\begin{pmatrix}\frac{a}{2}&\frac{a^{2}}{4}+1\\ -1&-\frac{a}{2}\end{pmatrix}.
  • In case when aa is odd, representation matrices are

    ±(aa2122a),±(aa2+122a).\pm\begin{pmatrix}a&\frac{a^{2}-1}{2}\\ -2&-a\end{pmatrix},\pm\begin{pmatrix}a&\frac{a^{2}+1}{2}\\ -2&-a\end{pmatrix}.

To show the strong cohomological rigidity for Hirzebruch surfaces, we need the following: for any automorphism φ\varphi of H(Σ0)H^{*}(\Sigma_{0}) (respectively, H(Σ1)H^{*}(\Sigma_{1})), there exists a diffeomorphism ff of Σ0\Sigma_{0} (respectively, Σ1\Sigma_{1}) such that f=φf^{*}=\varphi.

For P1\ell\in\mathbb{C}P^{1}, we denote by P1\ell^{\perp}\in\mathbb{C}P^{1} the orthogonal complement of \ell in 2\mathbb{C}^{2}. Let φ:H(Σ0)H(Σ0)\varphi\thinspace\colon H^{*}(\Sigma_{0})\to H^{*}(\Sigma_{0}) be an isomorphism. The representation matrix of φ\varphi with respect to x1x_{1}, x2x_{2} is a signed permutation matrix and vice versa. A signed permutation matrix of size 22 is the identity matrix or a multiplication of several (0110)\begin{pmatrix}0&1\\ 1&0\end{pmatrix} and (1001)\begin{pmatrix}-1&0\\ 0&1\end{pmatrix}. Therefore φ\varphi is induced by the identity map or a composition of diffeomorphisms (1,2)(2,1)(\ell_{1},\ell_{2})\mapsto(\ell_{2},\ell_{1}) and (1,2)(1,2)(\ell_{1},\ell_{2})\mapsto(\ell_{1}^{\perp},\ell_{2}) for (1,2)Σ0=P1×P1(\ell_{1},\ell_{2})\in\Sigma_{0}=\mathbb{C}P^{1}\times\mathbb{C}P^{1}.

Let a=±1a=\pm 1. We will construct a diffeomorphism of Σa\Sigma_{a} explicitly for all automorphism of H(Σa)H^{*}(\Sigma_{a}). For a moment, we use column vector notation to represent elements in 2\mathbb{C}^{2}. By Proposition 2.4, Σa\Sigma_{a} is diffeomorphic to the quotient (S3)2/(S1)2(S^{3})^{2}/(S^{1})^{2} by the action of (S1)2(S^{1})^{2} on (S3)2(S^{3})^{2} given by

(t1t2)((z1w1),(z2w2))=((t1z1t1w1),(t2z2t1at2w2)).\begin{pmatrix}t_{1}\\ t_{2}\end{pmatrix}\cdot(\begin{pmatrix}z_{1}\\ w_{1}\end{pmatrix},\begin{pmatrix}z_{2}\\ w_{2}\end{pmatrix})=(\begin{pmatrix}t_{1}z_{1}\\ t_{1}w_{1}\end{pmatrix},\begin{pmatrix}t_{2}z_{2}\\ t_{1}^{-a}t_{2}w_{2}\end{pmatrix}).

We identify Σa\Sigma_{a} with the quotient (S3)2/(S1)2(S^{3})^{2}/(S^{1})^{2} via the diffeomorphism. We denote by [(z1w1),(z2w2)]Σa\begin{bmatrix}\begin{pmatrix}z_{1}\\ w_{1}\end{pmatrix},\begin{pmatrix}z_{2}\\ w_{2}\end{pmatrix}\end{bmatrix}\in\Sigma_{a} the equivalence class of ((z1w1),(z2w2))(S3)2(\begin{pmatrix}z_{1}\\ w_{1}\end{pmatrix},\begin{pmatrix}z_{2}\\ w_{2}\end{pmatrix})\in(S^{3})^{2}. For a complex number zz and kk\in\mathbb{Z}, we define

z[k]:={zkif k>0,1if k=0,z¯kif k<0.z^{[k]}:=\begin{cases}z^{k}&\text{if $k>0$},\\ 1&\text{if $k=0$},\\ \overline{z}^{-k}&\text{if $k<0$}.\end{cases}

Let f:ΣaΣaf\thinspace\colon\Sigma_{a}\to\Sigma_{a} be the map given by

[(z1w1),(z2w2)][(|z2|4+|w2|4)1/2(z1z2[2a]w1¯w2[2a]w1z2[2a]+z1¯w2[2a]),(z2¯w2)].\begin{bmatrix}\begin{pmatrix}z_{1}\\ w_{1}\end{pmatrix},\begin{pmatrix}z_{2}\\ w_{2}\end{pmatrix}\end{bmatrix}\mapsto\begin{bmatrix}(|z_{2}|^{4}+|w_{2}|^{4})^{-1/2}\begin{pmatrix}z_{1}z_{2}^{[-2a]}-\overline{w_{1}}w_{2}^{[-2a]}\\ w_{1}z_{2}^{[-2a]}+\overline{z_{1}}w_{2}^{[-2a]}\end{pmatrix},\begin{pmatrix}\overline{z_{2}}\\ w_{2}\end{pmatrix}\end{bmatrix}.

By direct computation we have that ff is well-defined. Moreover, ff admits a smooth inverse, because

(|z2|4+|w2|4)1/2(z1z2[2a]w1¯w2[2a]w1z2[2a]+z1¯w2[2a])(|z_{2}|^{4}+|w_{2}|^{4})^{-1/2}\begin{pmatrix}z_{1}z_{2}^{[-2a]}-\overline{w_{1}}w_{2}^{[-2a]}\\ w_{1}z_{2}^{[-2a]}+\overline{z_{1}}w_{2}^{[-2a]}\end{pmatrix}

is the first column vector of the special unitary matrix

(z1w1¯w1z1¯)(|z2|4+|w2|4)1/2(z2[2a]w2[2a]w2[2a]z2[2a]).\begin{pmatrix}z_{1}&-\overline{w_{1}}\\ w_{1}&\overline{z_{1}}\end{pmatrix}(|z_{2}|^{4}+|w_{2}|^{4})^{-1/2}\begin{pmatrix}z_{2}^{[-2a]}&-w_{2}^{[2a]}\\ w_{2}^{[-2a]}&z_{2}^{[2a]}\end{pmatrix}.

We claim that f(x1)=2ax2+x1f^{*}(x_{1})=-2ax_{2}+x_{1}. To see this, we consider the pull-back LL of the tautological line bundle of P1\mathbb{C}P^{1} by π2:ΣaP1\pi_{2}\thinspace\colon\Sigma_{a}\to\mathbb{C}P^{1}, that is,

L={([(z1w1),(z2w2)],(uv))Σa×2λ such that (uv)=(λz1λw1)}.L=\{(\begin{bmatrix}\begin{pmatrix}z_{1}\\ w_{1}\end{pmatrix},\begin{pmatrix}z_{2}\\ w_{2}\end{pmatrix}\end{bmatrix},\begin{pmatrix}u\\ v\end{pmatrix})\in\Sigma_{a}\times\mathbb{C}^{2}\mid{}^{\exists}\lambda\in\mathbb{C}\text{ such that }\begin{pmatrix}u\\ v\end{pmatrix}=\begin{pmatrix}\lambda z_{1}\\ \lambda w_{1}\end{pmatrix}\}.

Then the pull-back f(L)f^{*}(L) is

{([(z1w1),(z2w2)],(uv))Σa×2λ such that (uv)=(λ(z1z2[2a]w1¯w2[2a])λ(w1z2[2a]+z1¯w2[2a]))}.\{(\begin{bmatrix}\begin{pmatrix}z_{1}\\ w_{1}\end{pmatrix},\begin{pmatrix}z_{2}\\ w_{2}\end{pmatrix}\end{bmatrix},\begin{pmatrix}u\\ v\end{pmatrix})\in\Sigma_{a}\times\mathbb{C}^{2}\mid{}^{\exists}\lambda\in\mathbb{C}\text{ such that }\begin{pmatrix}u\\ v\end{pmatrix}=\begin{pmatrix}\lambda(z_{1}z_{2}^{[-2a]}-\overline{w_{1}}w_{2}^{[-2a]})\\ \lambda(w_{1}z_{2}^{[-2a]}+\overline{z_{1}}w_{2}^{[-2a]})\end{pmatrix}\}.

On the other hand, f(L)f^{*}(L) is isomorphic to L1L2(2a)L_{1}^{\prime}\otimes L_{2}^{\otimes(-2a)}, where L1L_{1}^{\prime} and L2L_{2} are defined as in the proof of Proposition 2.4. In fact, the map L1L2(2a)f(L)L_{1}^{\prime}\otimes L_{2}^{\otimes(-2a)}\to f^{*}(L) given by

[(z1w1),(z2w2),λ]([(z1w1),(z2w2)],(λ(z1z2[2a]w1¯w2[2a])λ(w1z2[2a]+z1¯w2[2a])))\begin{bmatrix}\begin{pmatrix}z_{1}\\ w_{1}\end{pmatrix},\begin{pmatrix}z_{2}\\ w_{2}\end{pmatrix},\lambda\end{bmatrix}\mapsto(\begin{bmatrix}\begin{pmatrix}z_{1}\\ w_{1}\end{pmatrix},\begin{pmatrix}z_{2}\\ w_{2}\end{pmatrix}\end{bmatrix},\begin{pmatrix}\lambda(z_{1}z_{2}^{[-2a]}-\overline{w_{1}}w_{2}^{[-2a]})\\ \lambda(w_{1}z_{2}^{[-2a]}+\overline{z_{1}}w_{2}^{[-2a]})\end{pmatrix})

is a line bundle isomorphism. By definition, c1(L)=x1c_{1}(L)=x_{1}. The argument above shows that c1(f(L))=2ax2+x1c_{1}(f^{*}(L))=-2ax_{2}+x_{1}. Therefore f(x1)=2ax2+x1f^{*}(x_{1})=-2ax_{2}+x_{1}. By the same argument we have that f(x2)=x2f^{*}(x_{2})=-x_{2}. Let g1:ΣaΣag_{1}\thinspace\colon\Sigma_{a}\to\Sigma_{a} be the diffeomorphism given by

[(z1w1),(z2w2)][(z1w1),(w2¯z2¯)].\begin{bmatrix}\begin{pmatrix}z_{1}\\ w_{1}\end{pmatrix},\begin{pmatrix}z_{2}\\ w_{2}\end{pmatrix}\end{bmatrix}\mapsto\begin{bmatrix}\begin{pmatrix}z_{1}\\ w_{1}\end{pmatrix},\begin{pmatrix}-\overline{w_{2}}\\ \overline{z_{2}}\end{pmatrix}\end{bmatrix}.

Then the representation matrix of g1g_{1}^{*} is (1a01)\begin{pmatrix}1&a\\ 0&-1\end{pmatrix}. Let g2:ΣaΣag_{2}\thinspace\colon\Sigma_{a}\to\Sigma_{a} be the diffeomorphism given by

[(z1w1),(z2w2)][(w1¯z1¯),(w2z2)].\begin{bmatrix}\begin{pmatrix}z_{1}\\ w_{1}\end{pmatrix},\begin{pmatrix}z_{2}\\ w_{2}\end{pmatrix}\end{bmatrix}\mapsto\begin{bmatrix}\begin{pmatrix}-\overline{w_{1}}\\ \overline{z_{1}}\end{pmatrix},\begin{pmatrix}w_{2}\\ z_{2}\end{pmatrix}\end{bmatrix}.

Then the representation matrix of g2g_{2}^{*} is (1a01)\begin{pmatrix}-1&-a\\ 0&1\end{pmatrix}. These computations show that the diffeomorphism φ\varphi is induced by one of the identity map or a composition of several ff, g1g_{1} and g2g_{2}.

Finally, we remark that ff, g1g_{1} and g2g_{2} are equivariant with respect to the following S1S^{1}-action on Σa\Sigma_{a}. We define the S1S^{1}-action on Σa\Sigma_{a} via

t[(z1w1),(z2w2)][(z1t2w1),(z2taw2)]t\cdot\begin{bmatrix}\begin{pmatrix}z_{1}\\ w_{1}\end{pmatrix},\begin{pmatrix}z_{2}\\ w_{2}\end{pmatrix}\end{bmatrix}\mapsto\begin{bmatrix}\begin{pmatrix}z_{1}\\ t^{2}w_{1}\end{pmatrix},\begin{pmatrix}z_{2}\\ t^{-a}w_{2}\end{pmatrix}\end{bmatrix}

for tS1t\in S^{1} and [(z1w1),(z2w2)]Σa\begin{bmatrix}\begin{pmatrix}z_{1}\\ w_{1}\end{pmatrix},\begin{pmatrix}z_{2}\\ w_{2}\end{pmatrix}\end{bmatrix}\in\Sigma_{a}. We shall state this as a lemma for later use.

Lemma 3.1.

Let a=±1a=\pm 1. For any automorphism φ:H(Σa)H(Σa)\varphi\thinspace\colon H^{*}(\Sigma_{a})\to H^{*}(\Sigma_{a}), there exists an S1S^{1}-equivariant diffeomorphism f:ΣaΣaf\thinspace\colon\Sigma_{a}\to\Sigma_{a} such that f=φf^{*}=\varphi.

Remark 3.2.

Let a=±1a=\pm 1. It is known that Σa\Sigma_{a} is diffeomorphic to the connected sum P2#P2¯\mathbb{C}P^{2}\#\overline{\mathbb{C}P^{2}}. Using certain involutions on P2\mathbb{C}P^{2} and focusing on the connected sum, one can construct diffeomorphisms that induce all automorphisms of the cohomology of Σa\Sigma_{a}, see [CM2012, Proof of Lemma 5.4]. For Lemma 3.1, we focus on the quotient construction (Proposition 2.4) and construct the equivariant diffeomorphisms.

4. Hirzebruch surface bundles and algebra automorphisms

Let

Bn+2{B_{n+2}}Bn+1{B_{n+1}}Bn{B_{n}}{\cdots}B1{B_{1}}B0={a point}.{B_{0}=\{\text{a point}\}.}πn+2\scriptstyle{\pi_{n+2}}πn+1\scriptstyle{\pi_{n+1}}πn\scriptstyle{\pi_{n}}π2\scriptstyle{\pi_{2}}π1\scriptstyle{\pi_{1}}

be a Bott tower of height n+2n+2 such that πj:BjBj1\pi_{j}\thinspace\colon B_{j}\to B_{j-1} is a projectivization P(¯ξj)Bj1P(\underline{\mathbb{C}}\oplus\xi_{j})\to B_{j-1}, where ξj\xi_{j} is a complex line bundle over Bj1B_{j-1}. We think of H(Bn)H^{*}(B_{n}) and H(Bn+1)H^{*}(B_{n+1}) as subalgebras of H(Bn+2)H^{*}(B_{n+2}) via the injections πn+1\pi_{n+1}^{*} and πn+2\pi_{n+2}^{*}.

By applying Lemma 2.2 twice, we have that H(Bn+2)H^{*}(B_{n+2}) is freely generated by xn+2=c1(γn+2)x_{n+2}=c_{1}(\gamma_{n+2}) and xn+1=c1(γn+1)x_{n+1}=c_{1}(\gamma_{n+1}) as an H(Bn)H^{*}(B_{n})-algebra, where γn+1\gamma_{n+1} and γn+2\gamma_{n+2} are tautological line bundles of Bn+1=P(¯ξn+1)B_{n+1}=P(\underline{\mathbb{C}}\oplus\xi_{n+1}) and Bn+2=P(¯ξn+2)B_{n+2}=P(\underline{\mathbb{C}}\oplus\xi_{n+2}), respectively. Suppose that c1(ξn+2)=axn+1+yc_{1}(\xi_{n+2})=ax_{n+1}+y for aa\in\mathbb{Z} and yH2(Bn)y\in H^{2}(B_{n}). Then the composed bundle Bn+2BnB_{n+2}\to B_{n} is a fiber bundle with fiber Σa\Sigma_{a}. Moreover, there is a natural isomorphism between H(Bn+2)/H>(Bn)H^{*}(B_{n+2})/H^{>}(B_{n}) and H(Σa)H^{*}(\Sigma_{a}). Let x1¯\overline{x_{1}} and x2¯\overline{x_{2}} be the image of xn+1x_{n+1} and xn+2x_{n+2} by the projection H2(Bn+2)H2(Σa)H2(Bn+2)/H2(Bn)H^{2}(B_{n+2})\to H^{2}(\Sigma_{a})\cong H^{2}(B_{n+2})/H^{2}(B_{n}), respectively. We study the condition of aa, c1(ξn+1)c_{1}(\xi_{n+1}) and yy for an automorphism φ\varphi of H(Σa)H^{*}(\Sigma_{a}) to extend to an algebra automorphism φ~\widetilde{\varphi} of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra. Remark that φ\varphi and φ~\widetilde{\varphi} are determined by their restrictions to H2(Σa)H^{2}(\Sigma_{a}) and H2(Bn+2)H^{2}(B_{n+2}), respectively. In the sequel, we suppose that u1,u2H2(Bn)u_{1},u_{2}\in H^{2}(B_{n}) and φ~:H2(Bn+2)H2(Bn+2)\widetilde{\varphi}\thinspace\colon H^{2}(B_{n+2})\to H^{2}(B_{n+2}) is a homomorphism which preserves elements in H2(Bn)H^{2}(B_{n}).

  • (0-i)

    Suppose that the representation matrix of φ\varphi with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} is (1001)\begin{pmatrix}1&0\\ 0&1\end{pmatrix}. Suppose that φ~(xn+1)=xn+1+u1\widetilde{\varphi}(x_{n+1})=x_{n+1}+u_{1} and φ~(xn+2)=xn+2+u2\widetilde{\varphi}(x_{n+2})=x_{n+2}+u_{2}. Then

    φ~(xn+1)φ~(xn+1c1(ξn+1))=2xn+1u1+u1(u1c1(ξn+1))\begin{split}&\widetilde{\varphi}(x_{n+1})\widetilde{\varphi}(x_{n+1}-c_{1}(\xi_{n+1}))\\ &=2x_{n+1}u_{1}+u_{1}(u_{1}-c_{1}(\xi_{n+1}))\end{split}

    and

    φ~(xn+2)φ~(xn+2axn+1y)=xn+2(2u2au1)axn+1u2+u2(u2au1y).\begin{split}&\widetilde{\varphi}(x_{n+2})\widetilde{\varphi}(x_{n+2}-ax_{n+1}-y)\\ &=x_{n+2}(2u_{2}-au_{1})-ax_{n+1}u_{2}+u_{2}(u_{2}-au_{1}-y).\end{split}

    Therefore φ~\widetilde{\varphi} becomes an automorphism of H(Bn)H^{*}(B_{n}) if and only if u1=u2=0u_{1}=u_{2}=0 because H(Bn+2)H^{*}(B_{n+2}) is freely generated by 1,xn+1,xn+2,xn+1xn+21,x_{n+1},x_{n+2},x_{n+1}x_{n+2} as an H(Bn)H^{*}(B_{n})-module. These computations show that φ\varphi always uniquely extends to an automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra.

  • (0-ii)

    Suppose that the representation matrix of φ\varphi with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} is (1a01)\begin{pmatrix}1&a\\ 0&-1\end{pmatrix}. Suppose that φ~(xn+1)=xn+1+u1\widetilde{\varphi}(x_{n+1})=x_{n+1}+u_{1} and φ~(xn+2)=xn+2+axn+1+u2\widetilde{\varphi}(x_{n+2})=-x_{n+2}+ax_{n+1}+u_{2}. Then

    φ~(xn+1)φ~(xn+1c1(ξn+1))=2xn+1u1+u1(u1c1(ξn+1))\begin{split}&\widetilde{\varphi}(x_{n+1})\widetilde{\varphi}(x_{n+1}-c_{1}(\xi_{n+1}))\\ &=2x_{n+1}u_{1}+u_{1}(u_{1}-c_{1}(\xi_{n+1}))\end{split}

    and

    φ~(xn+2)φ~(xn+2axn+1y)=xn+2(au12u2+2y)+axn+1(u2au1y)+u2(u2au1y).\begin{split}&\widetilde{\varphi}(x_{n+2})\widetilde{\varphi}(x_{n+2}-ax_{n+1}-y)\\ &=x_{n+2}(au_{1}-2u_{2}+2y)+ax_{n+1}(u_{2}-au_{1}-y)+u_{2}(u_{2}-au_{1}-y).\end{split}

    Therefore φ~\widetilde{\varphi} becomes an automorphism of H(Bn+2)H^{*}(B_{n+2}) if and only if u1=0u_{1}=0 and u2=yu_{2}=y.

    These computations show that φ\varphi always uniquely extends to an automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra.

For other 66 automorphisms of H(Σa)H^{*}(\Sigma_{a}), we need to separate cases by the value of aa.

  1. (1)

    Suppose that a=0a=0.

    1. (1-i)

      Suppose that the representation matrix of φ\varphi with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} is (1001)\begin{pmatrix}-1&0\\ 0&-1\end{pmatrix}. Suppose that φ~(xn+1)=xn+1+u1\widetilde{\varphi}(x_{n+1})=-x_{n+1}+u_{1} and φ~(xn+2)=xn+2+u2\widetilde{\varphi}(x_{n+2})=-x_{n+2}+u_{2}. Then

      φ~(xn+1)φ~(xn+1c1(ξn+1))=xn+1(2c1(ξn+1)2u1)+u1(u1c1(ξn+1))\begin{split}&\widetilde{\varphi}(x_{n+1})\widetilde{\varphi}(x_{n+1}-c_{1}(\xi_{n+1}))\\ &=x_{n+1}(2c_{1}(\xi_{n+1})-2u_{1})+u_{1}(u_{1}-c_{1}(\xi_{n+1}))\end{split}

      and

      φ~(xn+2)φ~(xn+2y)=xn+2(2y2u2)+u2(u2y).\begin{split}&\widetilde{\varphi}(x_{n+2})\widetilde{\varphi}(x_{n+2}-y)\\ &=x_{n+2}(2y-2u_{2})+u_{2}(u_{2}-y).\end{split}

      Therefore φ~\widetilde{\varphi} becomes an automorphism of H(Bn+2)H^{*}(B_{n+2}) if and only if u1=c1(ξn+1)u_{1}=c_{1}(\xi_{n+1}) and u2=yu_{2}=y.

      These computations show that, φ\varphi always uniquely extends to an automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra.

    2. (1-ii)

      Suppose that the representation matrix of φ\varphi with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} is (1001)\begin{pmatrix}-1&0\\ 0&1\end{pmatrix}. Then φ\varphi always uniquely extends to an automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra because (0-ii), (1-i) and (1001)=(1001)(1001)\begin{pmatrix}-1&0\\ 0&1\end{pmatrix}=\begin{pmatrix}-1&0\\ 0&-1\end{pmatrix}\begin{pmatrix}1&0\\ 0&-1\end{pmatrix}.

    3. (1-iii)

      Suppose that the representation matrix of φ\varphi with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} is (0110)\begin{pmatrix}0&-1\\ -1&0\end{pmatrix}. Suppose that φ~(xn+1)=xn+2+u1\widetilde{\varphi}(x_{n+1})=-x_{n+2}+u_{1} and φ~(xn+2)=xn+1+u2\widetilde{\varphi}(x_{n+2})=-x_{n+1}+u_{2}. Then

      φ~(xn+1)φ~(xn+1c1(ξn+1))=xn+2(y2u1+c1(ξn+1))+u1(u1c1(ξn+1))\begin{split}&\widetilde{\varphi}(x_{n+1})\widetilde{\varphi}(x_{n+1}-c_{1}(\xi_{n+1}))\\ &=x_{n+2}(y-2u_{1}+c_{1}(\xi_{n+1}))+u_{1}(u_{1}-c_{1}(\xi_{n+1}))\end{split}

      and

      φ~(xn+2)φ~(xn+2y)=xn+1(c1(ξn+1)2u2+y)+u2(u2y).\begin{split}&\widetilde{\varphi}(x_{n+2})\widetilde{\varphi}(x_{n+2}-y)\\ &=x_{n+1}(c_{1}(\xi_{n+1})-2u_{2}+y)+u_{2}(u_{2}-y).\end{split}

      Therefore φ~\widetilde{\varphi} becomes an automorphism of H(Bn+2)H^{*}(B_{n+2}) if and only if u1=u2=(c1(ξn+1)+y)/2u_{1}=u_{2}=(c_{1}(\xi_{n+1})+y)/2 and c1(ξn+1)2=y2c_{1}(\xi_{n+1})^{2}=y^{2}.

      These computations show that, φ\varphi uniquely extends to an automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra if and only if c1(ξn+1)2=y2c_{1}(\xi_{n+1})^{2}=y^{2} and c1(ξn+1)±yc_{1}(\xi_{n+1})\pm y is even.

    4. (1-iv)

      Suppose that the representation matrix of φ\varphi with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} is (0110)\begin{pmatrix}0&1\\ 1&0\end{pmatrix}. Then φ\varphi uniquely extends to an automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra if and only if c1(ξn+1)2=y2c_{1}(\xi_{n+1})^{2}=y^{2} and c1(ξn+1)±yc_{1}(\xi_{n+1})\pm y is even because (1-i), (1-iii) and

      (0110)=(0110)(1001).\begin{pmatrix}0&1\\ 1&0\end{pmatrix}=\begin{pmatrix}0&-1\\ -1&0\end{pmatrix}\begin{pmatrix}-1&0\\ 0&-1\end{pmatrix}.
    5. (1-v)

      Suppose that the representation matrix of φ\varphi with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} is ±(0110)\pm\begin{pmatrix}0&1\\ -1&0\end{pmatrix}. Then φ\varphi uniquely extends to an automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra if and only if c1(ξn+1)2=y2c_{1}(\xi_{n+1})^{2}=y^{2} and c1(ξn+1)±yc_{1}(\xi_{n+1})\pm y is even because (0-ii), (1-iii), (1-iv) and

      ±(0110)=±(0110)(1001).\pm\begin{pmatrix}0&1\\ -1&0\end{pmatrix}=\pm\begin{pmatrix}0&-1\\ -1&0\end{pmatrix}\begin{pmatrix}1&0\\ 0&-1\end{pmatrix}.
  2. (2)

    Suppose that aa is nonzero even.

    1. (2-i)

      Suppose that the representation matrix of φ\varphi with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} is (1001)\begin{pmatrix}-1&0\\ 0&-1\end{pmatrix}. Suppose that φ~(xn+1)=xn+1+u1\widetilde{\varphi}(x_{n+1})=-x_{n+1}+u_{1} and φ~(xn+2)=xn+2+u2\widetilde{\varphi}(x_{n+2})=-x_{n+2}+u_{2}. Then

      φ~(xn+1)φ~(xn+1c1(ξn+1))=xn+1(2c1(ξn+1)2u1)+u1(u1c1(ξn+1)\begin{split}&\widetilde{\varphi}(x_{n+1})\widetilde{\varphi}(x_{n+1}-c_{1}(\xi_{n+1}))\\ &=x_{n+1}(2c_{1}(\xi_{n+1})-2u_{1})+u_{1}(u_{1}-c_{1}(\xi_{n+1})\end{split}

      and

      φ~(xn+2)φ~(xn+2axn+1y)=xn+2(2y2u2+au1)+axn+1u2+u2(u2au1y).\begin{split}&\widetilde{\varphi}(x_{n+2})\widetilde{\varphi}(x_{n+2}-ax_{n+1}-y)\\ &=x_{n+2}(2y-2u_{2}+au_{1})+ax_{n+1}u_{2}+u_{2}(u_{2}-au_{1}-y).\end{split}

      Therefore φ~\widetilde{\varphi} becomes an automorphism of H(Bn+2)H^{*}(B_{n+2}) if and only if u1=c1(ξn+1)u_{1}=c_{1}(\xi_{n+1}), u2=0u_{2}=0 and y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}).

      These computations show that, φ\varphi uniquely extends to an automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra if and only if y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}).

    2. (2-ii)

      Suppose that the representation matrix of φ\varphi with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} is (1a01)\begin{pmatrix}-1&-a\\ 0&1\end{pmatrix}. Then φ\varphi uniquely extends to an automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra if and only if y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}) because (0-ii), (2-i) and

      (1a01)=(1001)(1a01).\begin{pmatrix}-1&-a\\ 0&1\end{pmatrix}=\begin{pmatrix}-1&0\\ 0&-1\end{pmatrix}\begin{pmatrix}1&a\\ 0&-1\end{pmatrix}.
    3. (2-iii)

      Suppose that the representation matrix of φ\varphi with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} is (a2a2411a2)\begin{pmatrix}\frac{a}{2}&\frac{a^{2}}{4}-1\\ -1&-\frac{a}{2}\end{pmatrix}. Suppose that φ~(xn+1)=xn+2+a2xn+1+u1\widetilde{\varphi}(x_{n+1})=-x_{n+2}+\frac{a}{2}x_{n+1}+u_{1} and φ~(xn+2)=a2xn+2+(a241)xn+1+u2\widetilde{\varphi}(x_{n+2})=-\frac{a}{2}x_{n+2}+(\frac{a^{2}}{4}-1)x_{n+1}+u_{2}. Then

      φ~(xn+1)φ~(xn+1c1(ξn+1))=xn+2(y2u1+c1(ξn+1))+axn+1(u1+a24c1(ξn+1))+u1(u1c1(ξn+1))\begin{split}&\widetilde{\varphi}(x_{n+1})\widetilde{\varphi}(x_{n+1}-c_{1}(\xi_{n+1}))\\ &=x_{n+2}(y-2u_{1}+c_{1}(\xi_{n+1}))\\ &\quad+ax_{n+1}(u_{1}+\frac{a-2}{4}c_{1}(\xi_{n+1}))+u_{1}(u_{1}-c_{1}(\xi_{n+1}))\end{split}

      and

      φ~(xn+2)φ~(xn+2axn+1y)=a4xn+2(2au1+(2a)y)+xn+1((1a416)c1(ξn+1)2u2+(a241)(au1y))+u2(u2au1y).\begin{split}&\widetilde{\varphi}(x_{n+2})\widetilde{\varphi}(x_{n+2}-ax_{n+1}-y)\\ &=\frac{a}{4}x_{n+2}(2au_{1}+(2-a)y)\\ &\quad+x_{n+1}((1-\frac{a^{4}}{16})c_{1}(\xi_{n+1})-2u_{2}+(\frac{a^{2}}{4}-1)(-au_{1}-y))\\ &\quad+u_{2}(u_{2}-au_{1}-y).\end{split}

      Therefore φ~\widetilde{\varphi} becomes an automorphism of H(Bn+2)H^{*}(B_{n+2}) if and only if u1=2a4c1(ξn+1)u_{1}=\frac{2-a}{4}c_{1}(\xi_{n+1}), u2=4a28c1(ξn+1)u_{2}=\frac{4-a^{2}}{8}c_{1}(\xi_{n+1}), y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}) and (4a2)c1(ξn+1)2=0(4-a^{2})c_{1}(\xi_{n+1})^{2}=0.

      These computations show that, φ\varphi uniquely extends to an automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra if and only if 2±a4c1(ξn+1)\frac{2\pm a}{4}c_{1}(\xi_{n+1}) is integral, y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}) and (4a2)c1(ξn+1)2=0(4-a^{2})c_{1}(\xi_{n+1})^{2}=0.

    4. (2-iv)

      Suppose that the representation matrix of φ\varphi with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} is (a2a24+11a2)\begin{pmatrix}-\frac{a}{2}&-\frac{a^{2}}{4}+1\\ 1&\frac{a}{2}\end{pmatrix}. Suppose that φ~(xn+1)=xn+2a2xn+1+u1\widetilde{\varphi}(x_{n+1})=x_{n+2}-\frac{a}{2}x_{n+1}+u_{1} and φ~(xn+2)=a2xn+2+(a24+1)xn+1+u2\widetilde{\varphi}(x_{n+2})=\frac{a}{2}x_{n+2}+(-\frac{a^{2}}{4}+1)x_{n+1}+u_{2}. Then

      φ~(xn+1)φ~(xn+1c1(ξn+1))=xn+2(y+2u1c1(ξn+1))+axn+1(u1+2+a4c1(ξn+1))+u1(u1c1(ξn+1))\begin{split}&\widetilde{\varphi}(x_{n+1})\widetilde{\varphi}(x_{n+1}-c_{1}(\xi_{n+1}))\\ &=x_{n+2}(y+2u_{1}-c_{1}(\xi_{n+1}))\\ &\quad+ax_{n+1}(-u_{1}+\frac{2+a}{4}c_{1}(\xi_{n+1}))+u_{1}(u_{1}-c_{1}(\xi_{n+1}))\end{split}

      and

      φ~(xn+2)φ~(xn+2axn+1y)=a4xn+2(2au1+(2+a)y)+xn+1(2u2(1a24)(au1+y)+(1a416)c1(ξn+1))+u2(u2au1y).\begin{split}&\widetilde{\varphi}(x_{n+2})\widetilde{\varphi}(x_{n+2}-ax_{n+1}-y)\\ &=-\frac{a}{4}x_{n+2}(2au_{1}+(2+a)y)\\ &\quad+x_{n+1}(2u_{2}-(1-\frac{a^{2}}{4})(au_{1}+y)+(1-\frac{a^{4}}{16})c_{1}(\xi_{n+1}))\\ &\quad+u_{2}(u_{2}-au_{1}-y).\end{split}

      Therefore φ~\widetilde{\varphi} becomes an automorphism of H(Bn+2)H^{*}(B_{n+2}) if and only if u1=2+a4c1(ξn+1)u_{1}=\frac{2+a}{4}c_{1}(\xi_{n+1}), u2=4a28c1(ξn+1)u_{2}=-\frac{4-a^{2}}{8}c_{1}(\xi_{n+1}), y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}) and (4a2)c1(ξn+1)2=0(4-a^{2})c_{1}(\xi_{n+1})^{2}=0.

      These computations show that, φ\varphi uniquely extends to an automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra if and only if 2±a4c1(ξn+1)\frac{2\pm a}{4}c_{1}(\xi_{n+1}) is integral, y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}) and (4a2)c1(ξn+1)2=0(4-a^{2})c_{1}(\xi_{n+1})^{2}=0.

    5. (2-v)

      Suppose that the representation matrix of φ\varphi with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} is ±(a2a24+11a2)\pm\begin{pmatrix}\frac{a}{2}&\frac{a^{2}}{4}+1\\ -1&-\frac{a}{2}\end{pmatrix}. Then φ\varphi uniquely extends to an automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra if and only if 2±a4c1(ξn+1)\frac{2\pm a}{4}c_{1}(\xi_{n+1}) is integral, y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}) and (4a2)c1(ξn+1)2=0(4-a^{2})c_{1}(\xi_{n+1})^{2}=0 because (0-ii), (2-iii), (2-iv) and

      ±(a2a24+11a2)=±(a2a2411a2)(1a01).\pm\begin{pmatrix}\frac{a}{2}&\frac{a^{2}}{4}+1\\ -1&-\frac{a}{2}\end{pmatrix}=\pm\begin{pmatrix}\frac{a}{2}&\frac{a^{2}}{4}-1\\ -1&-\frac{a}{2}\end{pmatrix}\begin{pmatrix}1&a\\ 0&-1\end{pmatrix}.
  3. (3)

    Suppose that aa is odd.

    1. (3-i)

      Suppose that the representation matrix of φ\varphi with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} is (1001)\begin{pmatrix}-1&0\\ 0&-1\end{pmatrix}. Suppose that φ~(xn+1)=xn+1+u1\widetilde{\varphi}(x_{n+1})=-x_{n+1}+u_{1} and φ~(xn+2)=xn+2+u2\widetilde{\varphi}(x_{n+2})=-x_{n+2}+u_{2}. Then

      φ~(xn+1)φ~(xn+1c1(ξn+1))=xn+1(2c1(ξn+1)2u1)+u12u1c1(ξn+1)\begin{split}&\widetilde{\varphi}(x_{n+1})\widetilde{\varphi}(x_{n+1}-c_{1}(\xi_{n+1}))\\ &=x_{n+1}(2c_{1}(\xi_{n+1})-2u_{1})+u_{1}^{2}-u_{1}c_{1}(\xi_{n+1})\end{split}

      and

      φ~(xn+2)φ~(xn+2axn+1y)=xn+2(2y2u2+au1)+axn+1u2+u2(u2au1y).\begin{split}&\widetilde{\varphi}(x_{n+2})\widetilde{\varphi}(x_{n+2}-ax_{n+1}-y)\\ &=x_{n+2}(2y-2u_{2}+au_{1})+ax_{n+1}u_{2}+u_{2}(u_{2}-au_{1}-y).\end{split}

      Therefore φ~\widetilde{\varphi} becomes an automorphism of H(Bn+2)H^{*}(B_{n+2}) if and only if u1=c1(ξn+1)u_{1}=c_{1}(\xi_{n+1}), u2=0u_{2}=0 and y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}).

      These computations show that, φ\varphi uniquely extends to an automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra if and only if c1(ξn+1)c_{1}(\xi_{n+1}) is even and y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}).

    2. (3-ii)

      Suppose that the representation matrix of φ\varphi with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} is (1a01)\begin{pmatrix}-1&-a\\ 0&1\end{pmatrix}. Then φ\varphi uniquely extends to an automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra if and only if c1(ξn+1)c_{1}(\xi_{n+1}) is even and y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}) because (0-ii), (3-i) and

      (1a01)=(1001)(1a01).\begin{pmatrix}-1&-a\\ 0&1\end{pmatrix}=\begin{pmatrix}-1&0\\ 0&-1\end{pmatrix}\begin{pmatrix}1&a\\ 0&-1\end{pmatrix}.
    3. (3-iii)

      Suppose that the representation matrix of φ\varphi with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} is (aa2122a)\begin{pmatrix}a&\frac{a^{2}-1}{2}\\ -2&-a\end{pmatrix}. Suppose that φ~(xn+1)=2xn+2+axn+1+u1\widetilde{\varphi}(x_{n+1})=-2x_{n+2}+{a}x_{n+1}+u_{1} and φ~(xn+2)=axn+2+(a212)xn+1+u2\widetilde{\varphi}(x_{n+2})=-{a}x_{n+2}+(\frac{a^{2}-1}{2})x_{n+1}+u_{2}. Then

      φ~(xn+1)φ~(xn+1c1(ξn+1))=2xn+2(2y2u1+c1(ξn+1))+axn+1(2u1+(a1)c1(ξn+1))+u1(u1c1(ξn+1))\begin{split}&\widetilde{\varphi}(x_{n+1})\widetilde{\varphi}(x_{n+1}-c_{1}(\xi_{n+1}))\\ &=2x_{n+2}(2y-2u_{1}+c_{1}(\xi_{n+1}))\\ &\quad+ax_{n+1}(2u_{1}+(a-1)c_{1}(\xi_{n+1}))+u_{1}(u_{1}-c_{1}(\xi_{n+1}))\end{split}

      and

      φ~(xn+2)φ~(xn+2axn+1y)=axn+2(au1+(1a)y)+xn+1(u2+(1a22)(au1+y)+1a44c1(ξn+1))+u2(u2au1y).\begin{split}&\widetilde{\varphi}(x_{n+2})\widetilde{\varphi}(x_{n+2}-ax_{n+1}-y)\\ &=ax_{n+2}(au_{1}+(1-a)y)\\ &\quad+x_{n+1}(-u_{2}+(\frac{1-a^{2}}{2})(au_{1}+y)+\frac{1-a^{4}}{4}c_{1}(\xi_{n+1}))\\ &\quad+u_{2}(u_{2}-au_{1}-y).\end{split}

      Therefore φ~\widetilde{\varphi} becomes an automorphism of H(Bn+2)H^{*}(B_{n+2}) if and only if u1=1a2c1(ξn+1)u_{1}=\frac{1-a}{2}c_{1}(\xi_{n+1}), u2=1a24c1(ξn+1)u_{2}=\frac{1-a^{2}}{4}c_{1}(\xi_{n+1}), y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}) and (1a2)c1(ξn+1)2=0(1-a^{2})c_{1}(\xi_{n+1})^{2}=0.

      These computations show that, φ\varphi uniquely extends to an automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra if and only if c1(ξn+1)c_{1}(\xi_{n+1}) is even, y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}) and (1a2)c1(ξn+1)2=0(1-a^{2})c_{1}(\xi_{n+1})^{2}=0.

    4. (3-iv)

      Suppose that the representation matrix of φ\varphi with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} is (aa2122a)\begin{pmatrix}-a&-\frac{a^{2}-1}{2}\\ 2&a\end{pmatrix}. Suppose that φ~(xn+1)=2xn+2axn+1+u1\widetilde{\varphi}(x_{n+1})=2x_{n+2}-{a}x_{n+1}+u_{1} and φ~(xn+2)=axn+2(a212)xn+1+u2\widetilde{\varphi}(x_{n+2})={a}x_{n+2}-(\frac{a^{2}-1}{2})x_{n+1}+u_{2}. Then

      φ~(xn+1)φ~(xn+1c1(ξn+1))=2xn+2(2y+2u1c1(ξn+1))+axn+1(2u1+(a+1)c1(ξn+1))+u1(u1c1(ξn+1))\begin{split}&\widetilde{\varphi}(x_{n+1})\widetilde{\varphi}(x_{n+1}-c_{1}(\xi_{n+1}))\\ &=2x_{n+2}(2y+2u_{1}-c_{1}(\xi_{n+1}))\\ &\quad+ax_{n+1}(-2u_{1}+(a+1)c_{1}(\xi_{n+1}))+u_{1}(u_{1}-c_{1}(\xi_{n+1}))\end{split}

      and

      φ~(xn+2)φ~(xn+2axn+1y)=axn+2(au1+(a+1)y)+xn+1(u2+a212(au1+y)+1a44c1(ξn+1))+u2(u2au1y).\begin{split}&\widetilde{\varphi}(x_{n+2})\widetilde{\varphi}(x_{n+2}-ax_{n+1}-y)\\ &=-ax_{n+2}(au_{1}+(a+1)y)\\ &\quad+x_{n+1}(u_{2}+\frac{a^{2}-1}{2}(au_{1}+y)+\frac{1-a^{4}}{4}c_{1}(\xi_{n+1}))\\ &\quad+u_{2}(u_{2}-au_{1}-y).\end{split}

      Therefore φ~\widetilde{\varphi} becomes an automorphism of H(Bn+2)H^{*}(B_{n+2}) if and only if u1=1+a2c1(ξn+1)u_{1}=\frac{1+a}{2}c_{1}(\xi_{n+1}), u2=1a24c1(ξn+1)u_{2}=-\frac{1-a^{2}}{4}c_{1}(\xi_{n+1}), y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}) and (1a2)c1(ξn+1)2=0(1-a^{2})c_{1}(\xi_{n+1})^{2}=0.

      These computations show that, φ\varphi uniquely extends to an automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra if and only if c1(ξn+1)c_{1}(\xi_{n+1}) is even, y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}) and (1a2)c1(ξn+1)2=0(1-a^{2})c_{1}(\xi_{n+1})^{2}=0.

    5. (3-v)

      Suppose that the representation matrix of φ\varphi with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} is ±(aa2+122a)\pm\begin{pmatrix}a&\frac{a^{2}+1}{2}\\ -2&-a\end{pmatrix}. Then φ\varphi uniquely extends to an automorphism of H(Bn+2)H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra if and only if c1(ξn+1)c_{1}(\xi_{n+1}) is even, y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}) and (1a2)c1(ξn+1)2=0(1-a^{2})c_{1}(\xi_{n+1})^{2}=0 because (0-ii), (3-iii), (3-iv) and

      ±(aa2+122a)=±(aa2122a)(1a01).\pm\begin{pmatrix}a&\frac{a^{2}+1}{2}\\ -2&-a\end{pmatrix}=\pm\begin{pmatrix}a&\frac{a^{2}-1}{2}\\ -2&-a\end{pmatrix}\begin{pmatrix}1&a\\ 0&-1\end{pmatrix}.

5. Realizing bundle automorphisms

We use the same notations as the previous section. In this section, we show that for any automorphism φ~:H(Bn+2)H(Bn+2)\widetilde{\varphi}\thinspace\colon H^{*}(B_{n+2})\to H^{*}(B_{n+2}) as an H(Bn)H^{*}(B_{n})-algebra there exists a bundle automorphism f~:Bn+2Bn+2\widetilde{f}\thinspace\colon B_{n+2}\to B_{n+2} over BnB_{n} such that f~=φ~\widetilde{f}^{*}=\widetilde{\varphi}. By Theorem 2.6 we know that such f~\widetilde{f} exists if the representation matrix of the descent homomorphism φ:H2(Σa)H2(Σa)\varphi\thinspace\colon H^{2}(\Sigma_{a})\to H^{2}(\Sigma_{a}) is upper triangular. In the sequel of this section we assume that the representation matrix of φ\varphi is not upper triangular. For a cohomology class α\alpha of degree 22, we denote by γkα\gamma_{k}^{\alpha} a line bundle over BkB_{k} such that c1(γkα)=αc_{1}(\gamma^{\alpha}_{k})=\alpha.

  1. (1)

    Suppose that a=0a=0. Then ξn+2\xi_{n+2} is isomorphic to the pull-back of a line bundle over BnB_{n} because c1(ξn+2)=yH2(Bn)c_{1}(\xi_{n+2})=y\in H^{2}(B_{n}). Thus we may assume that ξn+2=πn+1γny\xi_{n+2}=\pi_{n+1}^{*}\gamma_{n}^{y}. Then the P1\mathbb{C}P^{1}-bundle Bn+2Bn+1B_{n+2}\to B_{n+1} is the pull-back of πn+1:P(¯γny)Bn\pi_{n+1}^{\prime}\thinspace\colon P(\underline{\mathbb{C}}\oplus\gamma_{n}^{y})\to B_{n} by the P1\mathbb{C}P^{1}-bundle Bn+1BnB_{n+1}\to B_{n}. Therefore

    Bn+2={(1,2)P(¯ξn+1)×P(¯γny)πn+1(1)=πn+1(2)}.B_{n+2}=\{(\ell_{1},\ell_{2})\in P(\underline{\mathbb{C}}\oplus\xi_{n+1})\times P(\underline{\mathbb{C}}\oplus\gamma_{n}^{y})\mid\pi_{n+1}(\ell_{1})=\pi_{n+1}^{\prime}(\ell_{2})\}.

    By (1-iii), (1-iv) and (1-v) in Section 4, we have that c1(ξn+1)2=y2c_{1}(\xi_{n+1})^{2}=y^{2} and c1(ξn+1)±yc_{1}(\xi_{n+1})\pm y is even. Thus we have

    P(¯γny)P(γn12(c1(ξn+1)y)γn12(c1(ξn+1)+y))P(¯ξn+1)\begin{split}P(\underline{\mathbb{C}}\oplus\gamma^{y}_{n})&\cong P(\gamma_{n}^{\frac{1}{2}(c_{1}(\xi_{n+1})-y)}\oplus\gamma_{n}^{\frac{1}{2}(c_{1}(\xi_{n+1})+y)})\\ &\cong P(\underline{\mathbb{C}}\oplus\xi_{n+1})\end{split}

    by Lemma 2.1 and Theorem 2.5. Through this bundle isomorphism, φ~\widetilde{\varphi} is induced by one of the maps given by (1,2)(2,1)(\ell_{1},\ell_{2})\mapsto(\ell_{2},\ell_{1}), (1,2)(2,1)(\ell_{1},\ell_{2})\mapsto(\ell_{2}^{\perp},\ell_{1}), (1,2)(2,1)(\ell_{1},\ell_{2})\mapsto(\ell_{2},\ell_{1}^{\perp}) or (1,2)(2,1)(\ell_{1},\ell_{2})\mapsto(\ell_{2}^{\perp},\ell_{1}^{\perp}).

  2. (2)

    Suppose that aa is nonzero even. By (2-iii), (2-iv) and (2-v) in Section 4, we have that y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}). Since c1(ξn+2)=axn+1+yc_{1}(\xi_{n+2})=ax_{n+1}+y and xn+1(xn+1c1(ξn+1))=0x_{n+1}(x_{n+1}-c_{1}(\xi_{n+1}))=0, we have

    P(¯ξn+2)P(¯γn+1axn+1+y)P(γn+1a2xn+1γn+1a2(xn+1c1(ξn+1)))P(¯γn+1a2c1(ξn+1))\begin{split}P(\underline{\mathbb{C}}\oplus\xi_{n+2})&\cong P(\underline{\mathbb{C}}\oplus\gamma_{n+1}^{ax_{n+1}+y})\\ &\cong P(\gamma_{n+1}^{-\frac{a}{2}x_{n+1}}\oplus\gamma_{n+1}^{\frac{a}{2}(x_{n+1}-c_{1}(\xi_{n+1}))})\\ &\cong P(\underline{\mathbb{C}}\oplus\gamma_{n+1}^{-\frac{a}{2}c_{1}(\xi_{n+1})})\end{split}

    as bundles over Bn+1B_{n+1} by Lemma 2.1 and Theorem 2.5. Let Bn+2=P(¯γn+1a2c1(ξ1))B_{n+2}^{\prime}=P(\underline{\mathbb{C}}\oplus\gamma_{n+1}^{-\frac{a}{2}c_{1}(\xi_{1})}). Let g~:Bn+2Bn+2\widetilde{g}\thinspace\colon B_{n+2}\to B_{n+2}^{\prime} be the bundle isomorphism. Then the composition (g~1)φ~g~:H(Bn+2)H(Bn+2)(\widetilde{g}^{-1})^{*}\circ\widetilde{\varphi}\circ\widetilde{g}^{*}\thinspace\colon H^{*}(B_{n+2}^{\prime})\to H^{*}(B_{n+2}^{\prime}) is an automorphism as an H(Bn)H^{*}(B_{n})-algebra. This together with (1) yields that (g~1)φ~g~(\widetilde{g}^{-1})^{*}\circ\widetilde{\varphi}\circ\widetilde{g}^{*} is induced by a bundle automorphism f~:Bn+2Bn+2\widetilde{f^{\prime}}\thinspace\colon B_{n+2}^{\prime}\to B_{n+2}^{\prime}. Therefore φ~\widetilde{\varphi} is induced by a bundle automorphism g~1f~g~\widetilde{g}^{-1}\circ\widetilde{f^{\prime}}\circ\widetilde{g}.

  3. (3)

    Suppose that aa is odd and a±1a\neq\pm 1. By (3-iii), (3-iv) and (3-v) in Section 4, we have that c1(ξn+1)c_{1}(\xi_{n+1}) is even and c1(ξn+1)2=0c_{1}(\xi_{n+1})^{2}=0. Thus we have

    Bn+1=P(¯ξn+1)P(¯γnc1(ξn+1))P(γn12c1(ξn+1)γn12c1(ξn+1))P(¯¯)\begin{split}B_{n+1}&=P(\underline{\mathbb{C}}\oplus\xi_{n+1})\\ &\cong P(\underline{\mathbb{C}}\oplus\gamma_{n}^{c_{1}(\xi_{n+1})})\\ &\cong P(\gamma_{n}^{-\frac{1}{2}c_{1}(\xi_{n+1})}\oplus\gamma_{n}^{\frac{1}{2}c_{1}(\xi_{n+1})})\\ &\cong P(\underline{\mathbb{C}}\oplus\underline{\mathbb{C}})\end{split}

    as bundles over BnB_{n} by Lemma 2.1 and Theorem 2.5. Let πn+1:Bn+1Bn\pi_{n+1}^{\prime}\thinspace\colon B_{n+1}^{\prime}\to B_{n} be the product P1\mathbb{C}P^{1}-bundle over BnB_{n}. Let h~:Bn+1Bn+1\widetilde{h}\thinspace\colon B_{n+1}^{\prime}\to B_{n+1} be the bundle isomorphism. Let πn+2:Bn+2Bn+1\pi_{n+2}^{\prime}\thinspace\colon B_{n+2}^{\prime}\to B_{n+1}^{\prime} be the pull-back of πn+2:Bn+2Bn+1\pi_{n+2}\thinspace\colon B_{n+2}\to B_{n+1} by h~\widetilde{h} and g~:Bn+2Bn+2\widetilde{g}\thinspace\colon B_{n+2}^{\prime}\to B_{n+2} the bundle isomorphism induced by h~\widetilde{h}. Then the composition g~φ~(g~1):H(Bn+2)H(Bn+2)\widetilde{g}^{*}\circ\widetilde{\varphi}\circ(\widetilde{g}^{-1})^{*}\thinspace\colon H^{*}(B_{n+2}^{\prime})\to H^{*}(B_{n+2}^{\prime}) is an automorphism as an H(Bn)H^{*}(B_{n})-algebra. Let xn+1,xn+2x_{n+1}^{\prime},x_{n+2}^{\prime} be the first Chern classes of the tautological line bundles of Bn+1B_{n+1}^{\prime}, Bn+2B_{n+2}^{\prime}, respectively. Let xn+1¯\overline{x_{n+1}^{\prime}} and xn+2¯\overline{x_{n+2}^{\prime}} be the image of xn+1x_{n+1}^{\prime} and xn+2x_{n+2}^{\prime} by the projection H2(Bn+2)H2(Bn+2)/H2(Bn)H^{2}(B_{n+2}^{\prime})\to H^{2}(B_{n+2}^{\prime})/H^{2}(B_{n}), respectively. Since the representation matrix of g~:H2(Bn+2)H2(Bn+2)\widetilde{g}^{*}\thinspace\colon H^{2}(B_{n+2})\to H^{2}(B_{n+2}^{\prime}) with respect to x1,,xn,xn+1,xn+2x_{1},\dots,x_{n},x_{n+1},x_{n+2} and x1,,xn,xn+1,xn+2x_{1},\dots,x_{n},x_{n+1}^{\prime},x_{n+2}^{\prime} is upper triangular, we have that the descent homomorphism of g~φ~(g~1)\widetilde{g}^{*}\circ\widetilde{\varphi}\circ(\widetilde{g}^{-1})^{*} with respect to xn+1¯,xn+2¯\overline{x_{n+1}^{\prime}},\overline{x_{n+2}^{\prime}} is not upper triangular. By (3-iii), (3-iv), (3-v) and Bn+1=P(¯¯)B_{n+1}^{\prime}=P(\underline{\mathbb{C}}\oplus\underline{\mathbb{C}}), we have that there exists aa^{\prime}\in\mathbb{Z} such that h~(c1(ξn+2))=axn+1\widetilde{h}^{*}(c_{1}(\xi_{n+2}))=a^{\prime}x_{n+1}^{\prime} . Therefore πn+1πn+2:Bn+2Bn\pi_{n+1}^{\prime}\circ\pi_{n+2}^{\prime}\thinspace\colon B_{n+2}^{\prime}\to B_{n} is a trivial bundle with fiber Σa\Sigma_{a^{\prime}} over BnB_{n}. Since the Hirzebruch surface Σa\Sigma_{a^{\prime}} is strongly cohomological rigid as we saw in Section 3, we have that g~φ~(g~1)\widetilde{g}^{*}\circ\widetilde{\varphi}\circ(\widetilde{g}^{-1})^{*} is induced by a bundle automorphism f~\widetilde{f^{\prime}} of Bn+2BnB_{n+2}^{\prime}\to B_{n}. Thus φ~\widetilde{\varphi} is induced by the bundle automorphism g~f~g~1\widetilde{g}\circ\widetilde{f^{\prime}}\circ\widetilde{g}^{-1}.

  4. (4)

    Suppose that a=±1a=\pm 1. By Lemma 3.1, there exists an S1S^{1}-equivariant diffeomorphism ff of Σa\Sigma_{a} such that f=φf^{*}=\varphi. By (3-iii), (3-iv) and (3-v) in Section 4, we have that c1(ξn+1)c_{1}(\xi_{n+1}) is even and y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}). Let p:ηBnp\thinspace\colon\eta\to B_{n} be a complex line bundle over BnB_{n} with a Hermitian metric such that c1(η)=12c1(ξn+1)c_{1}(\eta)=\frac{1}{2}c_{1}(\xi_{n+1}). Let {hα:p1(Uα)Uα×}\{h_{\alpha}\thinspace\colon p^{-1}(U_{\alpha})\to U_{\alpha}\times\mathbb{C}\} be a local trivialization such that each hαh_{\alpha} preserves the length of vectors. Let {ϕαβ:UαβS1}\{\phi_{\alpha\beta}\thinspace\colon U_{\alpha\beta}\to S^{1}\} be the transition functions of η\eta with respect to the open covering {Uα}\{U_{\alpha}\}. Namely, hαhβ1(x,v)=(x,ϕαβ(x)v)h_{\alpha}\circ h_{\beta}^{-1}(x,v)=(x,\phi_{\alpha\beta}(x)v) for (x,v)Uαβ×(x,v)\in U_{\alpha\beta}\times\mathbb{C}, where Uαβ=UαUβU_{\alpha\beta}=U_{\alpha}\cap U_{\beta}. Since c1(η)=12c1(ξn+1)c_{1}(\eta)=\frac{1}{2}c_{1}(\xi_{n+1}) we may assume that ξn+1=η2\xi_{n+1}=\eta^{\otimes 2} and ξn+2=πn+1η(a)γn+2a\xi_{n+2}=\pi_{n+1}^{*}\eta^{\otimes(-a)}\otimes\gamma_{n+2}^{\otimes a}. Let p2:S(¯ξn+2)Bn+1p_{2}\thinspace\colon S(\underline{\mathbb{C}}\oplus\xi_{n+2})\to B_{n+1} be the unit sphere bundle. Then p2p_{2} is decomposed into the principal S1S^{1}-bundle q2:S(¯ξn+2)Bn+2q_{2}\thinspace\colon S(\underline{\mathbb{C}}\oplus\xi_{n+2})\to B_{n+2} and πn+2:Bn+2Bn+1\pi_{n+2}\thinspace\colon B_{n+2}\to B_{n+1}. Let p1:S(¯ξn+1)Bnp_{1}\thinspace\colon S(\underline{\mathbb{C}}\oplus\xi_{n+1})\to B_{n} be the unit sphere bundle. Then p1p_{1} is decomposed into the principal S1S^{1}-bundle q1:S(¯ξn+1)Bn+1q_{1}\thinspace\colon S(\underline{\mathbb{C}}\oplus\xi_{n+1})\to B_{n+1} and πn+1:Bn+1Bn\pi_{n+1}\thinspace\colon B_{n+1}\to B_{n}.

    S(¯q1ξn+2){S(\underline{\mathbb{C}}\oplus q_{1}^{*}\xi_{n+2})}S(¯ξn+2){S(\underline{\mathbb{C}}\oplus\xi_{n+2})}S(¯ξn+1).{S(\underline{\mathbb{C}}\oplus\xi_{n+1}).}Bn+2{B_{n+2}}Bn+1{B_{n+1}}Bn{B_{n}}q2\scriptstyle{q_{2}}p2\scriptstyle{p_{2}}q1\scriptstyle{q_{1}}p2\scriptstyle{p_{2}}πn+2\scriptstyle{\pi_{n+2}}πn+1\scriptstyle{\pi_{n+1}}

    The pull-back of the S3S^{3}-bundle p2:S(¯ξn+2)Bn+1p_{2}\thinspace\colon S(\underline{\mathbb{C}}\oplus\xi_{n+2})\to B_{n+1} by q1q_{1} is the unit sphere bundle S(¯q1ξn+1)S(\underline{\mathbb{C}}\oplus q_{1}^{*}\xi_{n+1}). The composition S(¯q1ξn+2)BnS(\underline{\mathbb{C}}\oplus q_{1}^{*}\xi_{n+2})\to B_{n} is the fiber product of the unit sphere bundles S(¯η2)S(\underline{\mathbb{C}}\oplus\eta^{\otimes 2}) and S(¯η(a))S(\underline{\mathbb{C}}\oplus\eta^{\otimes(-a)}) over BnB_{n}. Therefore S(¯q1ξn+2)BnS(\underline{\mathbb{C}}\oplus q_{1}^{*}\xi_{n+2})\to B_{n} is an S3×S3S^{3}\times S^{3}-bundle over BnB_{n}. Since ξn+1=η2\xi_{n+1}=\eta^{\otimes 2} and q1ξn+2p2η(a)q_{1}^{*}\xi_{n+2}\cong p_{2}^{*}\eta^{\otimes(-a)}, we have that the S3×S3S^{3}\times S^{3}-bundle S(¯q1ξn+2)BnS(\underline{\mathbb{C}}\oplus q_{1}^{*}\xi_{n+2})\to B_{n} has transition functions {(1,ϕαβ2,1,ϕαβa)}\{(1,\phi_{\alpha\beta}^{2},1,\phi_{\alpha\beta}^{-a})\}. On the other hand, each fiber of Bn+2BnB_{n+2}\to B_{n} is nothing but the quotient of S3×S3S^{3}\times S^{3} by the (S1)2(S^{1})^{2}-action as we saw in Proposition 2.4 and Section 3. Therefore the transition functions of Bn+2BnB_{n+2}\to B_{n} are

    [(z1ϕαβ2(x)w1),(z2ϕαβa(x)w2)]\begin{bmatrix}\begin{pmatrix}z_{1}\\ \phi_{\alpha\beta}^{2}(x)w_{1}\end{pmatrix},\begin{pmatrix}z_{2}\\ \phi_{\alpha\beta}^{-a}(x)w_{2}\end{pmatrix}\end{bmatrix}

    for xUαβx\in U_{\alpha\beta}, [(z1w1),(z2w2)]Σa\begin{bmatrix}\begin{pmatrix}z_{1}\\ w_{1}\end{pmatrix},\begin{pmatrix}z_{2}\\ w_{2}\end{pmatrix}\end{bmatrix}\in\Sigma_{a}. Thus we have that the S1S^{1}-equivariant diffeomorphism ff of Σa\Sigma_{a} commutes with the transition functions. Therefore the diffeomorphism ff of Σa\Sigma_{a} extends to the bundle automorphism f~\widetilde{f} of the Σa\Sigma_{a}-bundle Bn+2BnB_{n+2}\to B_{n}.

6. Algebra isomorphisms of Hirzebruch surface bundles

We use the same notation as Sections 4 and 5. Let ξn+1\xi_{n+1}^{\prime} be a complex line bundle over BnB_{n} and πn+1:Bn+1Bn\pi_{n+1}^{\prime}\thinspace\colon B_{n+1}^{\prime}\to B_{n} be the projectivization P(¯ξn+1)BnP(\underline{\mathbb{C}}\oplus\xi_{n+1}^{\prime})\to B_{n}. Let ξn+2\xi_{n+2}^{\prime} be a complex line bundle over Bn+1B_{n+1}^{\prime} and πn+2:Bn+2Bn+1\pi_{n+2}^{\prime}\thinspace\colon B_{n+2}^{\prime}\to B_{n+1}^{\prime} be the projectivization P(¯ξn+2)Bn+1P(\underline{\mathbb{C}}\oplus\xi_{n+2}^{\prime})\to B_{n+1}^{\prime}. As well as H(Bn+2)H^{*}(B_{n+2}), we think of H(Bn)H^{*}(B_{n}) and H(Bn+1)H^{*}(B^{\prime}_{n+1}) as subalgebras of H(Bn+2)H^{*}(B^{\prime}_{n+2}) via the injections (πn+1)(\pi_{n+1}^{\prime})^{*} and (πn+2)(\pi_{n+2}^{\prime})^{*}. Then H(Bn+2)H^{*}(B_{n+2}^{\prime}) is freely generated by xn+2=c1(γn+2)x_{n+2}^{\prime}=c_{1}(\gamma_{n+2}^{\prime}) and xn+1=c1(γn+1)x_{n+1}^{\prime}=c_{1}(\gamma_{n+1}^{\prime}) as an H(Bn)H^{*}(B_{n})-algebra, where γn+2\gamma_{n+2}^{\prime} and γn+1\gamma_{n+1}^{\prime} are tautological line bundles of Bn+2Bn+1B_{n+2}^{\prime}\to B_{n+1}^{\prime} and Bn+1BnB_{n+1}^{\prime}\to B_{n}, respectively. In this section, we show that if H(Bn+2)H^{*}(B_{n+2}) and H(Bn+2)H^{*}(B_{n+2}^{\prime}) are isomorphic as H(Bn)H^{*}(B_{n})-algebras then Bn+2B_{n+2} and Bn+2B_{n+2}^{\prime} are isomorphic as bundles over BnB_{n}.

Suppose that c1(ξn+2)=axn+1+yc_{1}(\xi_{n+2}^{\prime})=a^{\prime}x_{n+1}^{\prime}+y^{\prime} for aa^{\prime}\in\mathbb{Z} and yH2(Bn)y^{\prime}\in H^{2}(B_{n}). Then the bundle Bn+2BnB_{n+2}^{\prime}\to B_{n} is a fiber bundle with fiber Σa\Sigma_{a^{\prime}}. Let x1¯\overline{x_{1}^{\prime}} and x2¯\overline{x_{2}^{\prime}} be the image of xn+1x_{n+1}^{\prime} and xn+2x_{n+2}^{\prime} by the projection H2(Bn+2)H2(Σa)H2(Bn+2)/H2(Bn)H^{2}(B_{n+2}^{\prime})\to H^{2}(\Sigma_{a^{\prime}})\cong H^{2}(B_{n+2}^{\prime})/H^{2}(B_{n}), respectively. Let ψ~:H2(Bn+2)H2(Bn+2)\widetilde{\psi}\thinspace\colon H^{2}(B_{n+2})\to H^{2}(B_{n+2}^{\prime}) be a homomorphism as \mathbb{Z}-modules. Suppose that

ψ~(x1¯)=ψ11x1¯+ψ21x2¯+v1,ψ~(x2¯)=ψ12x1¯+ψ22x2¯+v2,\begin{split}\widetilde{\psi}(\overline{x_{1}})&=\psi_{11}\overline{x_{1}^{\prime}}+\psi_{21}\overline{x_{2}^{\prime}}+v_{1},\\ \widetilde{\psi}(\overline{x_{2}})&=\psi_{12}\overline{x_{1}^{\prime}}+\psi_{22}\overline{x_{2}^{\prime}}+v_{2},\end{split}

here ψij\psi_{ij}\in\mathbb{Z} for i,j=1,2i,j=1,2 and v1,v2H2(B)v_{1},v_{2}\in H^{2}(B). Assume that ψ~\widetilde{\psi} extends to an isomorphism of H(Bn)H^{*}(B_{n})-algebras. Then ψ~\widetilde{\psi} descends to an isomorphism ψ:H(Σa)H(Σa)\psi\thinspace\colon H^{*}(\Sigma_{a})\to H^{*}(\Sigma_{a^{\prime}}) whose representation matrix with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} and x1¯,x2¯\overline{x_{1}^{\prime}},\overline{x_{2}^{\prime}} is (ψ11ψ12ψ21ψ22)\begin{pmatrix}\psi_{11}&\psi_{12}\\ \psi_{21}&\psi_{22}\end{pmatrix}. In particular, aa and aa^{\prime} have the same parity. If the representation matrix of ψ\psi is upper triangular, then φ~\widetilde{\varphi} is induced by a bundle isomorphism Bn+2Bn+2B_{n+2}\to B_{n+2}^{\prime} by Theorem 2.6. In the sequel of this section, we always assume that the representation matrix of ψ\psi is not upper triangular. Let φ~:H(Bn+2)H(Bn+2)\widetilde{\varphi}\thinspace\colon H^{*}(B_{n+2}^{\prime})\to H^{*}(B_{n+2}^{\prime}) be the automorphism as an H(Bn)H^{*}(B_{n})-algebra given by φ~(xn+1)=xn+1\widetilde{\varphi}(x_{n+1}^{\prime})=x_{n+1}^{\prime} and φ~(xn+2)=xn+2+c1(ξn+2)\widetilde{\varphi}(x_{n+2}^{\prime})=-x_{n+2}^{\prime}+c_{1}(\xi_{n+2}^{\prime}). Then ψ~1φ~ψ~:H(Bn+2)H(Bn+2)\widetilde{\psi}^{-1}\circ\widetilde{\varphi}\circ\widetilde{\psi}\thinspace\colon H^{*}(B_{n+2})\to H^{*}(B_{n+2}) is an automorphism as an H(Bn)H^{*}(B_{n})-algebra. ψ~1φ~ψ~:H(Bn+2)H(Bn+2)\widetilde{\psi}^{-1}\circ\widetilde{\varphi}\circ\widetilde{\psi}\thinspace\colon H^{*}(B_{n+2})\to H^{*}(B_{n+2}) descends to an isomorphism H(Σa)H(Σa)H^{*}(\Sigma_{a})\to H^{*}(\Sigma_{a}) whose representation matrix with respect to x1¯,x2¯\overline{x_{1}},\overline{x_{2}} is (1a01)\begin{pmatrix}-1&-a\\ 0&1\end{pmatrix}.

As before, for a cohomology class α\alpha of degree 22, we denote by γkα\gamma_{k}^{\alpha} a line bundle over BkB_{k} such that c1(γkα)=αc_{1}(\gamma_{k}^{\alpha})=\alpha. We also denote by γn+1α\gamma_{n+1^{\prime}}^{\alpha} and γn+2α\gamma_{n+2^{\prime}}^{\alpha} line bundles over Bn+1B_{n+1}^{\prime} and Bn+2B_{n+2}^{\prime}, respectively.

  1. (1)

    Suppose that aa and aa^{\prime} are 0. Since aa and aa^{\prime} are 0, Bn+2B_{n+2} is isomorphic to the fiber product of Bn+1=P(¯ξn+1)BnB_{n+1}=P(\underline{\mathbb{C}}\oplus\xi_{n+1})\to B_{n} and P(¯γny)BnP(\underline{\mathbb{C}}\oplus\gamma_{n}^{y})\to B_{n} and Bn+2B_{n+2}^{\prime} is the fiber product of Bn+1=P(¯ξn+1)BnB_{n+1}^{\prime}=P(\underline{\mathbb{C}}\oplus\xi_{n+1}^{\prime})\to B_{n} and P(¯γny)BnP(\underline{\mathbb{C}}\oplus\gamma_{n}^{y^{\prime}})\to B_{n}. We will show that P(¯ξn+1)P(¯γny)P(\underline{\mathbb{C}}\oplus\xi_{n+1})\cong P(\underline{\mathbb{C}}\oplus\gamma_{n}^{y^{\prime}}) and P(¯γny)P(¯ξn+1)P(\underline{\mathbb{C}}\oplus\gamma_{n}^{y})\cong P(\underline{\mathbb{C}}\oplus\xi_{n+1}^{\prime}) as bundles over BnB_{n}. Since the primitive square zero elements in H2(Σ0)H^{2}(\Sigma_{0}) are ±x1¯=±x1¯\pm\overline{x_{1}}=\pm\overline{x_{1}^{\prime}} and ±x2¯=±x2¯\pm\overline{x_{2}}=\pm\overline{x_{2}^{\prime}}, and ψ\psi is an isomorphism whose representation matrix is not upper triangular, we have that ψ(x1¯)=s1x2¯\psi(\overline{x_{1}})=s_{1}\overline{x_{2}^{\prime}} and ψ(x2¯)=s2x1¯\psi(\overline{x_{2}})=s_{2}\overline{x_{1}^{\prime}} for some s1,s2=±1s_{1},s_{2}=\pm 1. It follows from ψ~(xn+1)ψ~(xn+1c1(ξn+1))=0\widetilde{\psi}(x_{n+1})\widetilde{\psi}(x_{n+1}-c_{1}(\xi_{n+1}))=0 that

    0=ψ~(xn+1)ψ~(xn+1c1(ξn+1))=xn+2(y+2s1v1s1c1(ξn+1))+v1(v1c1(ξn+1)).\begin{split}0&=\widetilde{\psi}(x_{n+1})\widetilde{\psi}(x_{n+1}-c_{1}(\xi_{n+1}))\\ &=x_{n+2}^{\prime}(y^{\prime}+2s_{1}v_{1}-s_{1}c_{1}(\xi_{n+1}))+v_{1}(v_{1}-c_{1}(\xi_{n+1})).\end{split}

    Thus we have y±c1(ξn+1)y^{\prime}\pm c_{1}(\xi_{n+1}) is even and y2=c1(ξn+1)2y^{\prime 2}=c_{1}(\xi_{n+1})^{2}. Therefore

    P(¯ξn+1)P(¯γnc1(ξn+1))P(γn12(yc1(ξn+1))γn12(c1(ξn+1)+y))P(¯γny)\begin{split}P(\underline{\mathbb{C}}\oplus\xi_{n+1})&\cong P(\underline{\mathbb{C}}\oplus\gamma_{n}^{c_{1}(\xi_{n+1})})\\ &\cong P(\gamma_{n}^{\frac{1}{2}(y^{\prime}-c_{1}(\xi_{n+1}))}\oplus\gamma_{n}^{\frac{1}{2}(c_{1}(\xi_{n+1})+y^{\prime})})\\ &\cong P(\underline{\mathbb{C}}\oplus\gamma_{n}^{y^{\prime}})\end{split}

    by Lemma 2.1 and Theorem 2.5. It follows from ψ~(xn+2)ψ~(xn+2c1(ξn+2))=0\widetilde{\psi}(x_{n+2})\widetilde{\psi}(x_{n+2}-c_{1}(\xi_{n+2}))=0 that

    0=ψ~(xn+2)ψ~(xn+2axn+1y)=xn+1(c1(ξn+1)2s2v2s2y)+v2(v2y).\begin{split}0&=\widetilde{\psi}(x_{n+2})\widetilde{\psi}(x_{n+2}-ax_{n+1}-y)\\ &=x_{n+1}^{\prime}(c_{1}(\xi_{n+1}^{\prime})-2s_{2}v_{2}-s_{2}y)+v_{2}(v_{2}-y).\end{split}

    Thus we have c1(ξn+1)±yc_{1}(\xi_{n+1}^{\prime})\pm y is even and c1(ξn+1)2=y2c_{1}(\xi_{n+1}^{\prime})^{2}=y^{2}. Therefore

    P(¯ξn+1)P(¯γnc1(ξn+1))P(γn12(yc1(ξn+1))γn12(c1(ξn+1)+y))P(¯γny)\begin{split}P(\underline{\mathbb{C}}\oplus\xi_{n+1}^{\prime})&\cong P(\underline{\mathbb{C}}\oplus\gamma_{n}^{c_{1}(\xi_{n+1}^{\prime})})\\ &\cong P(\gamma_{n}^{\frac{1}{2}(y-c_{1}(\xi_{n+1}^{\prime}))}\oplus\gamma_{n}^{\frac{1}{2}(c_{1}(\xi_{n+1}^{\prime})+y)})\\ &\cong P(\underline{\mathbb{C}}\oplus\gamma_{n}^{y})\end{split}

    by Lemma 2.1 and Theorem 2.5. Therefore Bn+2B_{n+2} and Bn+2B_{n+2}^{\prime} are isomorphic as bundles over BnB_{n}.

  2. (2)

    Suppose that aa and aa^{\prime} are even and one of them is nonzero. If aa is nonzero, then it follows from (2-ii) in Section 4 that y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}). Thus by Lemma 2.1 and Theorem 2.5 we have that

    P(¯ξn+2)P(¯γn+1axn+1+y)P(¯γaxn+1a2c1(ξn+1)))P(γn+1a2xn+1γn+1a2(xn+1c1(ξn+1)))P(¯γn+1a2c1(ξn+1))P(¯γn+1y)\begin{split}P(\underline{\mathbb{C}}\oplus\xi_{n+2})&\cong P(\underline{\mathbb{C}}\oplus\gamma_{n+1}^{ax_{n+1}+y})\\ &\cong P(\underline{\mathbb{C}}\oplus\gamma^{ax_{n+1}-\frac{a}{2}c_{1}(\xi_{n+1}))})\\ &\cong P(\gamma_{n+1}^{-\frac{a}{2}x_{n+1}}\oplus\gamma_{n+1}^{\frac{a}{2}(x_{n+1}-c_{1}(\xi_{n+1}))})\\ &\cong P(\underline{\mathbb{C}}\oplus\gamma_{n+1}^{-\frac{a}{2}c_{1}(\xi_{n+1})})\\ &\cong P(\underline{\mathbb{C}}\oplus\gamma_{n+1}^{y})\end{split}

    as bundles over Bn+1B_{n+1}. Therefore we may assume that a=0a=0. Using the same argument, we may assume that a=0a^{\prime}=0. It follows from (1) in this section that Bn+2B_{n+2} and Bn+2B_{n+2}^{\prime} are isomorphic as bundles over BnB_{n}.

  3. (3)

    Suppose that aa and aa^{\prime} are odd. By (3-ii) in Section 4, we have that y=a2c1(ξn+1)y=-\frac{a}{2}c_{1}(\xi_{n+1}). By the same argument we have that y=a2c1(ξn+1)y^{\prime}=-\frac{a^{\prime}}{2}c_{1}(\xi_{n+1}^{\prime}). Since the primitive square zero elements in H2(Σa)H^{2}(\Sigma_{a}) (respectively, H2(Σa)H^{2}(\Sigma_{a^{\prime}})) are ±x1¯\pm\overline{x_{1}} (respectively, ±x1¯\pm\overline{x_{1}^{\prime}}) and ±(2x2¯ax1¯)\pm(2\overline{x_{2}}-a\overline{x_{1}}) (respectively, ±(2x2¯ax1¯)\pm(2\overline{x_{2}^{\prime}}-a^{\prime}\overline{x_{1}^{\prime}})) and the representation matrix of the isomorphism ψ:H2(Σa)H2(Σa)\psi\thinspace\colon H^{2}(\Sigma_{a})\to H^{2}(\Sigma_{a^{\prime}}) is not upper triangular, we have that ψ(x1¯)=s1(2x2¯ax1¯)\psi(\overline{x_{1}})=s_{1}(2\overline{x_{2}^{\prime}}-a^{\prime}\overline{x_{1}^{\prime}}) and ψ(2x2¯ax1¯)=s2x1¯\psi(2\overline{x_{2}}-a\overline{x_{1}})=s_{2}\overline{x_{1}^{\prime}} for s1,s2=±1s_{1},s_{2}=\pm 1. Then ψ(x2¯)=as1x2¯+s2s1aa2x1¯\psi(\overline{x_{2}})=as_{1}\overline{x_{2}^{\prime}}+\frac{s_{2}-s_{1}aa^{\prime}}{2}\overline{x_{1}^{\prime}}. It follows from ψ~(xn+1)ψ~(xn+1c1(ξn+1))=0\widetilde{\psi}(x_{n+1})\widetilde{\psi}(x_{n+1}-c_{1}(\xi_{n+1}))=0 that

    0=ψ~(xn+1)ψ~(xn+1c1(ξn+1))=xn+2(2ac1(ξn+1)+4s1v12s1c1(ξn+1))+s1axn+1(s1ac1(ξn+1)2v1+c1(ξn+1))+v1(v1c1(ξn+1)).\begin{split}0&=\widetilde{\psi}(x_{n+1})\widetilde{\psi}(x_{n+1}-c_{1}(\xi_{n+1}))\\ &=x_{n+2}^{\prime}(-2a^{\prime}c_{1}(\xi_{n+1}^{\prime})+4s_{1}v_{1}-2s_{1}c_{1}(\xi_{n+1}))\\ &\quad+s_{1}a^{\prime}x_{n+1}^{\prime}(s_{1}a^{\prime}c_{1}(\xi_{n+1}^{\prime})-2v_{1}+c_{1}(\xi_{n+1}))\\ &\quad+v_{1}(v_{1}-c_{1}(\xi_{n+1})).\end{split}

    Thus we have v1=12(s1ac1(ξn+1)+c1(ξn+1))v_{1}=\frac{1}{2}(s_{1}a^{\prime}c_{1}(\xi_{n+1}^{\prime})+c_{1}(\xi_{n+1})) and a2c1(ξn+1)2=c1(ξn+1)2a^{\prime 2}c_{1}(\xi_{n+1}^{\prime})^{2}=c_{1}(\xi_{n+1})^{2}. It follows from ψ~(xn+2)ψ~(xn+2c1(ξn+2))=0\widetilde{\psi}(x_{n+2})\widetilde{\psi}(x_{n+2}-c_{1}(\xi_{n+2}))=0 that

    0=ψ~(xn+2)ψ~(xn+2c1(ξn+2))=xn+1(1s1s2aa4c1(ξn+1)+s2v2)+v2(v2s1aa2c1(ξn+1)).\begin{split}0&=\widetilde{\psi}(x_{n+2})\widetilde{\psi}(x_{n+2}-c_{1}(\xi_{n+2}))\\ &=x_{n+1}^{\prime}(\frac{1-s_{1}s_{2}aa^{\prime}}{4}c_{1}(\xi_{n+1}^{\prime})+s_{2}v_{2})\\ &\quad+v_{2}(v_{2}-\frac{s_{1}aa^{\prime}}{2}c_{1}(\xi_{n+1}^{\prime})).\end{split}

    Thus we have v2=s1aas24c1(ξn+1)v_{2}=\frac{s_{1}aa^{\prime}-s_{2}}{4}c_{1}(\xi_{n+1}^{\prime}) and (a2a21)c1(ξn+1)2=0(a^{2}a^{\prime 2}-1)c_{1}(\xi_{n+1}^{\prime})^{2}=0. If a2a21a^{2}a^{\prime 2}\neq 1, then c1(ξn+1)2=0c_{1}(\xi_{n+1}^{\prime})^{2}=0. By considering ψ~1\widetilde{\psi}^{-1} instead of ψ~\widetilde{\psi}, we also have that c1(ξn+1)2=0c_{1}(\xi_{n+1})^{2}=0. By (3-ii) in Section 4, we have that c1(ξn+1)c_{1}(\xi_{n+1}) is even. By the same argument we also have that c1(ξn+1)c_{1}(\xi_{n+1}^{\prime}) is even. Using the same argument as (3) in Section 5, we have that Bn+2B_{n+2} and Bn+2B_{n+2}^{\prime} are trivial bundles over BnB_{n}. Since the parity of aa and aa^{\prime} are the same, we have that Bn+2B_{n+2} and Bn+2B_{n+2}^{\prime} are isomorphic as bundles.

    Suppose that a2a2=1a^{2}a^{\prime 2}=1. Then a=±1a^{\prime}=\pm 1 and we have that c1(ξn+1)2=c1(ξn+1)2c_{1}(\xi_{n+1}^{\prime})^{2}=c_{1}(\xi_{n+1})^{2} and c1(ξn+1)±c1(ξn+1)c_{1}(\xi_{n+1}^{\prime})\pm c_{1}(\xi_{n+1}) is even. Therefore it follows from Lemma 2.1 and Theorem 2.5 that

    P(¯ξn+1)P(¯γnc1(ξn+1))P(γn12(c1(ξn+1)+c1(ξn+1))γn12(c1(ξn+1)+c1(ξn+1)))P(¯γnc1(ξn+1))P(¯ξn+1)\begin{split}P(\underline{\mathbb{C}}\oplus\xi_{n+1})&\cong P(\underline{\mathbb{C}}\oplus\gamma_{n}^{c_{1}(\xi_{n+1})})\\ &\cong P(\gamma_{n}^{\frac{1}{2}(-c_{1}(\xi_{n+1})+c_{1}(\xi_{n+1}^{\prime}))}\oplus\gamma_{n}^{\frac{1}{2}(c_{1}(\xi_{n+1})+c_{1}(\xi_{n+1}^{\prime}))})\\ &\cong P(\underline{\mathbb{C}}\oplus\gamma_{n}^{c_{1}(\xi_{n+1}^{\prime})})\\ &\cong P(\underline{\mathbb{C}}\oplus\xi_{n+1}^{\prime})\end{split}

    as bundles over BnB_{n}. Let h~:Bn+1Bn+1\widetilde{h}\thinspace\colon B_{n+1}\to B_{n+1}^{\prime} be the bundle isomorphism and ξn+2′′\xi_{n+2}^{\prime\prime} the pull-back bundle of ξn+2\xi_{n+2}^{\prime} by h~\widetilde{h}. Let Bn+2′′Bn+1B_{n+2}^{\prime\prime}\to B_{n+1} be the projectivization P(¯ξn+2′′)Bn+1P(\underline{\mathbb{C}}\oplus\xi_{n+2}^{\prime\prime})\to B_{n+1} and xn+2′′x_{n+2}^{\prime\prime} the first Chern class of the tautological line bundle of Bn+2B_{n+2}. Let g~:Bn+2′′Bn+2\widetilde{g}\thinspace\colon B_{n+2}^{\prime\prime}\to B_{n+2}^{\prime} be the isomorphism as bundles over Bn+1B_{n+1}^{\prime} induced by the pull-back as bundles over Bn+1B_{n+1}. We have the commutative diagram

    Bn+2=P(¯ξn+2){B_{n+2}=P(\underline{\mathbb{C}}\oplus\xi_{n+2})}Bn+1=P(¯ξn+1){B_{n+1}=P(\underline{\mathbb{C}}\oplus\xi_{n+1})}Bn+2′′=P(¯ξn+2′′){B_{n+2}^{\prime\prime}=P(\underline{\mathbb{C}}\oplus\xi_{n+2}^{\prime\prime})}Bn,{B_{n},}Bn+1=P(¯ξn+1){B_{n+1}^{\prime}=P(\underline{\mathbb{C}}\oplus\xi_{n+1}^{\prime})}Bn+2=P(¯ξn+2){B_{n+2}^{\prime}=P(\underline{\mathbb{C}}\oplus\xi_{n+2}^{\prime})}h~\scriptstyle{\widetilde{h}}g~\scriptstyle{\widetilde{g}}

    here vertical arrows are isomorphisms as P1\mathbb{C}P^{1}-bundles and right arrows are projections of P1\mathbb{C}P^{1}-bundles. The representation matrix of g~ψ~:H(Bn+2)H(Bn+2′′)\widetilde{g}^{*}\circ\widetilde{\psi}\thinspace\colon H^{*}(B_{n+2})\to H^{*}(B_{n+2}^{\prime\prime}) is not upper triangular because the one of g~\widetilde{g}^{*} is upper triangular but the one of ψ~\widetilde{\psi} is not upper triangular. Thus we have that there exists an odd a′′a^{\prime\prime} such that c1(ξn+2′′)=a′′xn+1a′′2c1(ξn+1)c_{1}(\xi_{n+2}^{\prime\prime})=a^{\prime\prime}x_{n+1}-\frac{a^{\prime\prime}}{2}c_{1}(\xi_{n+1}) by (3-ii) in Section 4. If a′′±1a^{\prime\prime}\neq\pm 1, then a2a′′21a^{2}a^{\prime\prime 2}\neq 1. By the same argument as the case when a2a21a^{2}a^{\prime 2}\neq 1 we have that Bn+2B_{n+2} and Bn+2′′B_{n+2}^{\prime\prime} are trivial bundles over BnB_{n}. If a′′=±1a^{\prime\prime}=\pm 1, then we have that Bn+2B_{n+2} and Bn+2′′B_{n+2}^{\prime\prime} are isomorphic as bundles over Bn+1B_{n+1} because c1(ξn+2′′)=±c1(ξn+2)c_{1}(\xi_{n+2}^{\prime\prime})=\pm c_{1}(\xi_{n+2}). So Bn+2B_{n+2} and Bn+2B_{n+2}^{\prime} are isomorphic as bundles over BnB_{n}.

References