Free Circle Actions and Positive Ricci Curvature on Manifolds with the Cohomology Ring of S^2\times S^5

For (1), see, e.g., [Br72a, Corollary VI.2.5]. Part (2) follows from the fact that, by definition, two principal circle bundles are isomorphic if and only if the corresponding circle actions are equivalent.

This is a direct consequence of the long exact sequence of homotopy groups of the fibration M\to N (see, e.g., [DK01, Corollary 6.44]), in combination with the fact that \pi_i\bigl(S^1\bigr)=0 for all i\geq 2.

This is well known, see, e.g., [MS74, Example 15.6] for the Pontryagin classes of \mathbb{C} P^3 and [Re22, Lemma 4.1.4] for the statement on connected sums.

Set H_1=H^2(N_1) and H_2=H^2(N_2) and let a\in H^2(N)\cong\mathbb{Z} be a generator. Then, by Lemma 2.6, we can identify H^2(N_i\# N)\cong H_i\oplus \mathbb{Z} a and the trilinear form \mu_{N_i\# N} splits according to this direct sum. A diffeomorphism N_1\# N\to N_2\# N induces an isomorphism \phi\colon H_1\oplus \mathbb{Z} a\to H_2\oplus \mathbb{Z} a such that \phi^*\mu_{N_2\# N}=\mu_{N_1\# N}. In particular, H_1 and H_2 have the same rank. Further, since b_3(N_i\# N)=b_3(N_i)+b_3(N), we have b_3(N_1)=b_3(N_2).

Now assume that \mu_N is non-trivial. Let (u_1,\dots,u_k) be a basis of H_1 and (v_1,\dots,v_k) a basis of H_2. We write

with \mu_{ij},\mu_j,\lambda_i,\lambda\in\mathbb{Z}.

In the following, we will identify a trilinear form \mu\colon H\times H\times H\to\mathbb{Z} with a bilinear map H\times H\to H^*=\mathrm{Hom}(H,\mathbb{Z}) via (x,y)\mapsto (z\mapsto\mu(x,y,z)) and we will write xy\in H^* for the image of (x,y) under this map. For example, we have a^2=\mu_N(a,a,a)a^*\in (\mathbb{Z} a)^*, where a^*(a)=1. We set \alpha=\mu_N(a,a,a), which, by assumption, is non-zero.

With this notation, we have u_i a=0, and hence

Since v_j v_\ell\in H_2^*, it follows that either \lambda_i=0 for all i or \lambda=0. We distinguish these two cases:

Case 1: \lambda_i=0 for all i. In this case, \phi(u_i)\in H_2 for all i. Hence, \phi restricts to an isomorphism \phi\colon H_1\to H_2, which, by assumption, preserves \mu and p_1. Thus, the systems of invariants \bigl(H_1,\mu_{N_1},p_1(N_1),w_2(N_1),\tfrac{b_3(N_1)}{2}\bigr) and \bigl(H_2,\mu_{N_2},p_1(N_2),w_2(N_2),\tfrac{b_3(N_2)}{2}\bigr) are equivalent via \phi and so N_1 and N_2 are diffeomorphic.

Case 2: \lambda=0. Since (\phi(u_1),\dots,\phi(u_k),\phi(a)) is a basis of H_2\oplus\mathbb{Z} a and \phi(a) has no a-component, we can choose the basis (u_1,\dots,u_k) such that \lambda_1=1 and \lambda_i=0 for all i\in\{2,\dots,k\}. In particular, (\phi(u_2),\dots,\phi(u_k),\phi(a)) is a basis of H_2. Hence, there are \nu_2,\dots,\nu_k, {\nu\in\mathbb{Z}} with

and so

Thus, if we replace u_1 by u_1-\nu_2 u_2 -\dots-\nu_k u_k, which still results in a basis (u_1,\dots,u_k) of H_1, we have

Hence,

so \nu=0 and \phi(u_1)=a. Since \mu_{N_2\# N} splits according to the direct sum decomposition H_2\oplus\mathbb{Z} a, the decomposition H_1\oplus \mathbb{Z} a=\mathbb{Z} u_1\oplus \mathrm{span}(u_2,\dots,u_k)\oplus \mathbb{Z} a is therefore a splitting of \mu_{N_1\# N}.

By defining the basis v_1,\dots,v_k so that \phi(a)=v_1 and \phi(u_i)=v_i for i=2,\dots,k (recall that (\phi(u_2),\dots,\phi(u_k),\phi(a)) is a basis of H_2), we therefore obtain that the isomorphism

splits according to this decomposition, where we have set H_1'=\mathrm{span}(u_2,\dots,u_k) and H_2'=\mathrm{span}(v_2,\dots,v_k). If N_i' denotes the closed, simply-connected 6-manifold with torsion-free homology with invariants \bigl(H_i',\mu_{N_i}|_{H_i'},p_1(N_i)|_{H_i'},w_2(N_i)|_{H_i'},\tfrac{b_3(N_i)}{2}\bigr), we therefore obtain that there is a diffeomorphism

that preserves this connected sum decomposition. In particular, N_1\cong N_1'\# N is diffeomorphic to N_2\cong N_2'\# N.

Finally, assume that N=S^2\times S^4. We set K_i=\ker(\mu_{N_i},p_1(N_i)). Then, by Lemma 2.6,

By Lemma 2.8, there exists a complement H_i' of K_i inside H_i. We choose the complement H_i' so that w_2(N_i)\in H_i'\mod 2 whenever w_2(N_i)\not\in K_i\mod 2. Hence,

is an isomorphism that maps K_1\oplus\mathbb{Z} a to K_2\oplus\mathbb{Z} a since \phi preserves \mu_{N_i\# N} and p_1(N_i\# N). Since \phi also preserves w_2(N_i), we have w_2(N_1)\in H_1'\mod 2 if and only if w_2(N_2)\in H_2'\mod 2.

If \mathrm{pr}_{H_2'} denotes the projection to H_2' along this splitting, the composition

is therefore an isomorphism that preserves w_2(N_i) if w_2(N_1)\in H_1'\mod 2. We claim that it also preserves \mu_{N_i}|_{H_i'} and p_1(N_i)|_{H_i'}. Indeed, if we write \phi(x)=x'+x_K with x'\in H_2' and x_K\in K_2\oplus\mathbb{Z} a for x\in H_1', we have for x,y,z\in H_1',

and

Hence, H_1' and H_2' are isomorphic via an isomorphism that preserves \mu_{N_i}|_{H_i'} and p_1(N_i)|_{H_i'}. Since both the trilinear and linear form are trivial on K_i and w_2(N_1)\in K_1 if and only if w_2(N_2)\in K_2, this isomorphism extends to an isomorphism between H_1=H_1'\oplus K_1 and {H_2=H_2'\oplus K_2} that preserves \mu_{N_i}, p_1(N_i) and w_2(N_i). It follows that the systems of invariants \bigl(H_1,\mu_{N_1},p_1(N_1),w_2(N_1),\tfrac{b_3(N_1)}{2}\bigr) and \bigl(H_2,\mu_{N_2},p_1(N_2),w_2(N_2),\tfrac{b_3(N_2)}{2}\bigr) are equivalent and hence, by Theorem 2.5, N_1 and N_2 are diffeomorphic.

It follows from the multilinearity of the form \mu_N and the linearity of the form p_1(N) that K=\ker(\mu_N,p_1(N)) is a subspace of H^2(N). Further, by the elementary divisor theorem, there exists a basis (u_1,\dots,u_k) of H^2(N) and \lambda_1,\dots,\lambda_\ell\in \mathbb{N} such that (\lambda_1 u_1,\dots,\lambda_\ell u_\ell) is a basis of K. By definition, since \lambda_i u_i\in K, we have u_i\in K and so \lambda_i=1. Hence, K is a direct summand in H^2(N).

Part (1) is precisely the content of Proposition 3.4. The diffeomorphism classification then follows from the fact that, by [KS98, Theorem 1], the invariants s_1, s_2, \tilde{s}_3 uniquely determine the oriented diffeomorphism type. Similarly, for the homeomorphism classification, we use that, by [KS98, Theorem 1], the invariants 28s_1, s_2, \tilde{s}_3 uniquely determine the oriented homeomorphism type. Since s_1 takes values in \bigl(\frac{1}{28}\mathbb{Z}\bigr)/\mathbb{Z}, we have 28s_1=0 and hence it suffices to consider the invariants s_2 and \tilde{s}_3.

It is shown in [Wa22, Section 5] that there are at most 672 oriented diffeomorphism types (resp. 24 oriented homeomorphism types) of spin cohomology S^2\times S^5s. Further, explicit examples are constructed where s_2 can take all possible values in \bigl(\frac{1}{12}\mathbb{Z}\bigr)/\mathbb{Z}, and \tilde{s}_3 can take all possible values in \bigl(\frac{1}{2}\mathbb{Z}\bigr)/\mathbb{Z}, while s_1 takes values in \bigl(\frac{1}{28}\mathbb{Z}\bigr)/\mathbb{Z}.

Finally, we can realise all values of \bigl(\frac{1}{28}\mathbb{Z}\bigr)/\mathbb{Z} for s_1 since, if M is an arbitrary spin cohomology S^2\times S^5, we have

by (3) of Example 3.3. Thus, these examples cover all possible values of \bigl(\frac{1}{28}\mathbb{Z}\bigr)/\mathbb{Z}, \bigl(\frac{1}{12}\mathbb{Z}\bigr)/\mathbb{Z} and \bigl(\frac{1}{2}\mathbb{Z}\bigr)/\mathbb{Z} for s_1, s_2 and \tilde{s}_3, respectively. In particular, since there exist at most 28\cdot 12\cdot 2=672 oriented diffeomorphism types of spin cohomology S^2\times S^5s, these examples cover all oriented diffeomorphism types.

We calculate \pi_4 for each of the 24 oriented homeomorphism types of spin cohomology S^2\times S^5s.

We first consider the case \tilde{s}_3=\frac{1}{2}. Tables ? and ? show that all 12 oriented homeomorphism types can be realised by manifolds of the form N_{(-1,-1,\alpha)}(e). More precisely, we extract the following from Tables ? and ?, respectively.

This shows that all oriented homeomorphism types M with (s_2(M),\tilde{s}_3(M))=\bigl(\frac{j}{12},\frac{1}{2}\bigr) are of the form N_{(-1,-1,\alpha)}(e) with \alpha odd. By Lemmas 2.4 and 4.14, we have

whenever \alpha is odd. Hence, \pi_4(M)=0 by Lemma 2.4.

Now assume that M is a spin cohomology S^2\times S^5 with \tilde{s}_3(M)=0. It is shown in [Kr98, p. 471] that either \pi_4(M)=0 or \pi_4(M)\cong\mathbb{Z}/2. Assume \pi_4(M)=0. Let M' be the spin cohomology S^2\times S^5 with invariants

By Theorem 3.5, M and M' are orientation-preserving homotopy equivalent, since, by the first part of the proof, \pi_4(M')=0=\pi_4(M).

By [EM16, Corollary 4.3], the invariants 2s_2 and s_3 are oriented homotopy invariants. Hence, also \tilde{s}_3=s_3-4s_2 is an oriented homotopy invariant. It follows that \tilde{s}_3(M)=\tilde{s}_3(M'), which is a contradiction. Hence, \pi_4(M)\cong\mathbb{Z}/2.

Alternatively, if s_2(M) is of the form \frac{j}{6}, we can also use a similar proof as in the case \tilde{s}_3=\frac{1}{2} by using the manifolds in Table ? in combination with Lemmas 2.4 and 4.14.

The fact that the given s-invariants are the only possible s-invariants of a spin cohomology S^2\times S^5 with a free circle action is the content of Lemma  4.11. These values are also realised by Proposition 4.12, except if

In this case, we now construct a closed, simply-connected non-spin 6-manifold N with torsion-free homology such that M\cong N(e) for some e\in H^2(N).

Set B=1, C=0, D=1 and v=0, and let A\in\mathbb{Z} with A\equiv 2\mod4 and u\in\mathbb{Z} be odd. Then, by Lemma 4.4, there is a unique closed, simply-connected non-spin 6-manifold N realising these values. Let e\in H^2(N) as in Lemma 4.4. Then, by Lemma 4.10, the manifold N(e) is a spin cohomology S^2\times S^5 with invariants

Since \frac{A}{2} is odd, 6s_2(N(e)) is an odd integer, and any odd integer \mod6 can be realised by choosing A appropriately. Since u is odd, 7s_1(N(e)) is an integer, and by choosing u appropriately, any integer \mod7 can be realised.

Moreover, we have p_1(N)(e)=48u+A. Hence, since for each fixed triple \bigl(\tfrac{i}{7},\frac{1+2j}{6},0\bigr) of invariants we can choose u arbitrarily large, there are an infinite number of values realising these values. This proves Theorem 4.1 in this case.

It remains to prove part (3) of Theorem 4.1 when the orbit space is of the form N_A or N_{(\alpha,0)}. Suppose first that M is a spin cohomology S^2\times S^5 that admits a free circle action with orbit space of the form N_A. By Lemmas 4.6 and 4.10, the s-invariants of N_A(e) are polynomials in (\lambda,\mu,\alpha_1,\alpha_2,\alpha_3) with rational coefficients. Let T be the least common multiple of the denominators of these coefficients. Then, by item (1) of Lemma 4.13, there exist infinitely many integer tuples \bigl(\lambda',\mu',\alpha_1',\alpha_2',\alpha_3'\bigr) with

in the same congruence class \mod T. It follows that there are infinitely many tuples A'=\bigl(\lambda',\mu',\alpha_1',\alpha_2',\alpha_3'\bigr) defining manifolds N_{A'} and classes e'\in H^2(N_{A'}) such that N_{A'}(e') is a spin cohomology S^2\times S^5 with the same s-invariants as M, hence N_{A'}(e')\cong M. By [Re24b, Proposition 4.12], there are only finitely many triples (\alpha_1,\alpha_2,\alpha_3) defining diffeomorphic manifolds N_A. Hence, there are infinitely many pairwise non-diffeomorphic manifolds N_{A'} and classes e'\in H^2(N_{A'}) with N_{A'}(e')\cong M. By Proposition 2.2, this proves part (3) in this case.

Finally, let M be a spin cohomology S^2\times S^5 that admits a free circle action with orbit space of the form N_{(\alpha,0)}. Here the argument goes along the same lines as in the previous case, where we use item (2) of Lemma 4.13 to construct infinitely many values of (\lambda,\mu,\alpha), and we use [Re24b, Proposition 4.12] to deduce that N_{(\alpha,0)} and N_{(\alpha',0)} are not diffeomorphic when \alpha\neq\pm\alpha'.

As seen in the proof of Theorem 4.1, when M is a spin cohomology S^2\times S^5 with s-invariants as in Theorem 4.2, then M admits infinitely many pairwise non-equivalent free circle actions such that each of these actions has quotient space of the form N_A or N_{(\alpha,0)}. Since manifolds of this form admit a Riemannian metric of positive Ricci curvature by [Re23, Theorem B] and [GW09, Theorem 2.7.3], the claim follows from Theorem 2.3.

This can be shown by verifying that the corresponding system of invariants satisfies (?). Alternatively, we can give an explicit geometric construction as follows.

We consider the following bipartite graph:

We now replace each labelled vertex with label \alpha_i with the linear D^3-bundle over S^4 with first Pontryagin class 4\alpha_i[S^4]^*, and we replace the unlabelled vertex with the trivial D^4-bundle S^3\times D^4 over S^3. Plumbing these bundles according to this graph results in a 7-manifold whose boundary N_A satisfies the conditions as claimed, see [Re24b, Lemma 4.6 and Proposition 4.12].

The proof is a tedious but straightforward calculation using Lemmas 4.4 and 4.5. The calculation for the value of \det(N_A,e)=AC-B^2 can be simplified as follows: Since AC-B^2 is the determinant of the bilinear form (x,y)\mapsto \mu_{N_A}(x,y,e) on H^2(N_A), we can also compute it from the basis (e_1,e_2) instead of (e,f), i.e.,

As in Lemma 4.5, this can be shown by verifying that the corresponding system of invariants satisfies (?). We give the following alternative construction.

We define N_{\alpha,\beta} as the total space of the linear S^2-bundle over \mathbb{C} P^2 with first Pontryagin class (4\alpha+\beta)\bigl[\mathbb{C} P^2\bigr]^* and second Stiefel–Whitney class \beta x\mod 2, where x\in H^2\bigl(\mathbb{C} P^2\bigr)\cong \mathbb{Z} is a generator. The claim then follows from [Re23, Corollary 5.8].

The proof is again a straightforward calculation. Here we have AC-B^2=\mu^2\alpha-\lambda^2. Since (3) of Lemma 4.4 is equivalent to \mu being even and \lambda being odd, we have \mu^2\alpha-\lambda^2\equiv -1\mod4, hence we cannot have \mu^2\alpha-\lambda^2=1.

Items (1) and (3) are the content of [Xu25, Lemma 3]. For item (2), it is shown in the proof of [Xu25, Lemma 3] that

Here, k=4A+24u, l=4D+24v and M_e is the matrix

When \det(N,e)=\det(M_e)=1, the signature \sigma(M_e) equals \sigma(M_e)=\mathrm{sgn}(A) since A and C have the same sign and so

Further,

A tedious but straightforward calculation using that A is even, B,C are odd and AC-B^2=1 then finishes the proof.

First, suppose that N is spin with AC-B^2=-1. Since B is odd, we can write it as B=2B'+1 and hence

Since C is odd, it follows that A is divisible by 8. Hence, if \tilde{s}_3(N(e))=0, we obtain

for j=\frac{A}{8}\bigl(2C^2-2BD+D^2\bigr), and if \tilde{s}_3(N(e))=\frac{1}{2}, we obtain

for j=3(D+1)u+2\frac{A}{8}\bigl(2C^2-2BD+D^2\bigr).

Next, suppose that N is spin with AC-B^2=1. If \tilde{s}_3(N(e))=0, we have that D is odd and we obtain

We now show that the term in the bracket is divisible by 4 and hence s_2(N(e))=\frac{j}{3}. Indeed, since B, C, D are odd, we have C^2\equiv 1\mod 4, BD\equiv\pm 1\mod 4 and D^2\equiv 1\mod 4. Hence,

Moreover, since C is odd, we have C^2\equiv 1\mod 8 and therefore \frac{1+C^2}{2}\equiv 1\mod 4. Finally, since AC=B^2+1\equiv 2\mod 8, so \frac{A}{2}C\equiv 1\mod 4, we have that \frac{A}{2}\equiv C\mod 4. It follows that

If \tilde{s}_3(N(e))=\frac{1}{2}, we have that D is even and again by using that \frac{1+C^2}{2}\equiv 1\mod 4 we obtain

for

Finally, suppose that N is non-spin. Then B is odd while A and C are even. In particular, B^2\equiv 1\mod 8 and all of A^2, AC, C^2 are divisible by 4. It follows that

Since B is odd, D(B+D) is even, so -AD(B+D)\equiv 0\mod 4. Hence, s_2(N(e)) is of the form \frac{j}{6}.

To analyse s_1(N(e)) when N is non-spin, we write s_1(N(e))=\frac{k}{28} and we obtain

Since we have AC-B^2=-1, it follows as in the first case that AC is divisible by 8. Moreover, since B is odd, D(B+D) is even as above. Hence, every term in this expression is even and therefore k is even.

To calculate the s-invariants of the total spaces N_A(e) and N_{(\alpha,0)}(e), one can combine Lemma 4.10 with Lemmas 4.6 and 4.8. However, the resulting formulae appear to be highly complicated (see, e.g., [EZ14, Propositions 5.2 and 7.11] for some special cases). On the other hand, a simple computer search shows that all the values as claimed are realised. The results of such a search are given in Tables ?–? below.

(1) First assume that \lambda,\mu,(\lambda+\mu)\neq 0. We set A=\lambda\alpha_1, B=\mu\alpha_2 and C=(\lambda+\mu)\alpha_3. Then (?) is equivalent to

which in turn is equivalent to

For n\in\mathbb{Z}, we define A_n=A+n(A-C), B_n=B+n(2C+(n+1)(A-C)) and C_n=C+n(A-C). Then

Hence, for n=m\lambda\mu(\lambda+\mu)T with m\in\mathbb{Z}, we obtain for

a sequence of solutions of (?) in the same congruence class of (\lambda,\mu,\alpha_1,\alpha_2,\alpha_3)\mod T.

Since we assumed that \lambda,\mu,(\lambda+\mu)\neq 0, this sequence is constant in m only when A-C=0 and C=0, i.e., A=C=0 and therefore \alpha_1=\alpha_3=0. In this case the left-hand side of (?) does not depend on \alpha_2, so we can freely choose its value. In particular we can assign infinitely many values to \alpha_2 while leaving it unchanged \mod T.

Finally, assume that one of \lambda, \mu, (\lambda+\mu), say (\lambda+\mu), vanishes. In this case the left-hand side of (?) does not depend on \alpha_3, and as before we can freely choose its value.

(2) First assume \mu\neq0. Then, for m\in\mathbb{Z} we set

A calculation shows that \bigl(\lambda^{(m)},\mu^{(m)},\alpha^{(m)}\bigr) is a sequence of solutions of (?) in the same congruence class of (\lambda,\mu,\alpha)\mod T. Since \mu\neq 0, this sequence is non-constant.

Finally, if \mu=0, the left-hand side of (?) does not depend on \alpha, so we can freely choose its value while leaving it unchanged \mod T.

The proof is a tedious but straightforward calculation using Lemmas 4.6, 4.8 and 4.10. The values of (A,B,C,D,u,v) are given by (0,-1,-1,-\alpha-1,0,0) (with (s,t)=(0,-1)) in the first case and by (0,1,0,\alpha,0,0) (with (s,t)=(1,0)) in the second case. To identify the total spaces with S^2\times S^5 and S^5\mathrel{\widetilde{\times} } S^2\cong\mathrm{SU}(3), we use Example 3.3.

Suppose that M satisfies the hypotheses of Theorem 4.1. Then, it was shown in the proof of Theorem 4.1 that there are infinitely many principal circle bundles \pi_i\colon M\to N_i, i\in\mathbb{N}, where N_i is a family of closed, simply-connected 6-manifolds that are pairwise non-diffeomorphic. By Proposition 5.2, there are infinitely many principal circle bundles

with \ell\geq 1 and m\geq 0 even, and

with \ell\geq 2 and m\geq 1 odd.

By an iterated application of Proposition 2.7, for fixed \ell and m, the families of manifolds

are pairwise non-diffeomorphic. Hence, the circle actions on the total spaces are all pairwise non-equivalent by Proposition 2.2.

Finally, all the base spaces admit Riemannian metrics of positive Ricci curvature, and hence, by Theorem 2.3, the total spaces admit invariant Riemannian metrics of positive Ricci curvature. Indeed, the manifolds N_i in the proof of Theorem 4.1 are of the form N_A, which admit core metrics as defined in [Bu19], see [Re23, Theorem B]. Further, also the manifolds \mathbb{C} P^3, S^2\times S^4 and S^3\times S^3 admit core metrics by [Bu19, Theorem C], [Re23, Theorem C] and [Bu20, Theorem B], respectively. By [Bu19, Theorem B], connected sums of manifolds with core metrics admit Riemannian metrics of positive Ricci curvature. Hence, all the base spaces admit Riemannian metrics of positive Ricci curvature.

This is essentially the content of [Xu25, Lemma 11]. For convenience, we give the proof below.

Let e_\pi\in H^2(N) be the Euler class of the principal circle bundle \pi and let a_i\in H^2\bigl(\mathbb{C} P^3\bigr) denote a generator for the i-th \mathbb{C} P^3-summand in N_{\ell,m}=N\#_{(\ell-1)}\mathbb{C} P^3\#_{\frac{m}{2}}\bigl(S^3\times S^3\bigr). We set

Let \pi_{\ell,m}\colon M_{\ell,m}\to N_{\ell,m} be the unique principal circle bundle with Euler class e_{\ell,m}. Then, by [GR25, Theorem A], we have a diffeomorphism

Here, \widetilde{\Sigma}_e B for a manifold B and a class e\in H^2(B) denotes the suspension operation of [Du26] and [GR25], that is, \widetilde{\Sigma}_e B is the result of surgery on the total space of the principal circle bundle over B with Euler class e along a fibre using the non-standard framing.

Finally, by [GR25, Theorem B], we have diffeomorphisms

which proves the first case. The proof of the second case is analogous.

By Lemma 4.4, N is simply-connected, has vanishing third Betti number and torsion-free cohomology. Hence, the same holds for N_1 and N_2. It follows that N_1 and N_2 satisfy b_2(N_1)=b_2(N_2)=1, since otherwise it would follow from Theorem 2.5 that one of N_1 and N_2 is the 6-sphere S^6.

Let e=\lambda_1 x_1+\lambda_2 x_2\in H^2(N_1\# N_2)\cong H^2(N) be the Euler class corresponding to the principal circle bundle M\to N. Since the number AC-B^2 in Lemma  4.4 is the determinant of the bilinear form (y,z)\mapsto \mu_N(y,z,e) in the basis (e,f), we can equally compute it by considering the basis (x_1,x_2) instead of (e,f). Here we obtain from (3) of Lemma 2.6

Hence, all of \lambda_1, \lambda_2, \mu_{N_1}(x_1,x_1,x_1), \mu_{N_2}(x_2,x_2,x_2) are equal to \pm1.

Now suppose that one of N_i, say N_1, is non-spin. Then, by (3) of Lemma 4.4, w_2(N_1)=x_1\mod 2, so we have

for all odd a_1\in\mathbb{Z} by (?). In particular, we have

so 24p_1(N_1)(x_1)\equiv 0\mod 48, thus p_1(N_1)(x_1) is even. It follows that \mu_{N_1}(x_1,x_1,x_1)\equiv p_1(N_1)(x_1)\mod 48 is even as well, which is a contradiction. Hence, both N_1 and N_2 are spin.=-1

Since \mu_{N_1}(x_1,x_1,x_1)=\pm1, it then follows that both N_i have the cohomology ring of \mathbb{C} P^3 and hence are homotopy \mathbb{C} P^3s (see, e.g., [Re24b, p. 41] and (2) of Lemma 2.6).

Finally, since the restriction of e to each N_i is primitive as \lambda_i=\pm1, it follows from [GR25, Theorems A and B] that M\cong M_1\# M_2\# \bigl(S^2\times S^5\bigr), where M_i is the total space of the principal circle bundle over N_i with Euler class \lambda_i x_i. Each M_i is a homotopy 7-sphere by [MY66, Lemma 6], so M is of the form \bigl(S^2\times S^5\bigr)\#\Sigma with \Sigma a homotopy 7-sphere. Hence, M is homeomorphic to S^2\times S^5.