Sphere bundles and bundles over spheres are everywhere and are excellent things to get one's hands dirty with.

(1a) But when can we have a bundle $S^n \to S^m \to B?$ It seems like requiring the total space of a sphere bundle to be a sphere is pretty restrictive.

(1b) Does the answer to (1a) depend on a choice of category (PL, TOP, etc)?

There's an $S^1-$ bundle over $CP^1$ with total space $S^3$, but that's the only example I can find.

(2) Do people know examples other than the one above?

  • 3
    $\begingroup$ How about the Hopf bundles? Those are the only bundles where the fibre, total space and base are all spheres -- a result, I believe, due to Frank Adams? Experts here will correct me if I'm wrong. $\endgroup$ Nov 3, 2010 at 2:21
  • $\begingroup$ José's comment came at the same time as my answer. I believe him that they're the only examples. $\endgroup$ Nov 3, 2010 at 2:29
  • 1
    $\begingroup$ Here we're allowing the base to be arbitrary. $\endgroup$
    – Romeo
    Nov 3, 2010 at 2:46
  • $\begingroup$ The answer to (1b) is that the category does matter. Milnor found seven smooth 7-manifolds, all different in DIFF but all PL homeomorphic to $S^7$, which are $S^3$-bundles over $S^4$. [Jose: is your quotation of Adams precisely correct? Isn't there an infinite sequence of $S^3$-bundles over $S^4$ with Euler class $\pm 1$ but varying $p_1$, and a subsequence for which the total space is $S^7$ even in DIFF?] $\endgroup$
    – Tim Perutz
    Nov 3, 2010 at 4:28
  • $\begingroup$ Oops. I didn't read closely enough. Of course, if the base need not be a sphere either, there are the more general Hopf fibrations that Neil describes very well in his answer. $\endgroup$ Nov 3, 2010 at 12:28

4 Answers 4


More generally, there are bundles $S^0\to S^n\to \mathbb{R}P^n$ and $S^1\to S^{2n+1}\to\mathbb{C}P^n$ and $S^3\to S^{4n+3}\to\mathbb{H}P^n$. There is also an "octonionic projective plane" $\mathbb{O}P^2$ but you have to construct it in a nonobvious way as the quotient $F_4/Spin(9)$ of exceptional Lie groups; more obvious constructions do not work because $\mathbb{O}$ is not associative. This gives a bundle $S^7\to S^{23}\to\mathbb{O}P^2$. I think it is known that there is no bundle $S^7\to S^{8k+7}\to B$ for $k>2$ but I do not know a proof of that. In general, if we have $S^n\to S^m\to B$ with $n>0$ and $m>1$ then the homotopy long exact sequence of the fibration shows that $B$ is simply connected, so the Serre spectral sequence $H^{\ast}(B;H^\ast(S^n))\to H^\ast(S^m)$ has untwisted coefficients and we find that $H^\ast(B)=\mathbb{Z}[x]/x^k$ where $|x|=n+1$ and $m=(n+1)k-1$. Adams's Hopf Invariant One theorem (applied to the attaching map for the $2(n+1)$-cell in $B$) now implies that $n\in\{1,3,7\}$. If $n=1$ then the element $x$ gives a map $B\to K(\mathbb{Z},2)=\mathbb{C}P^\infty$ and we deduce that $B$ is homotopy equivalent to $\mathbb{C}P^{k-1}$. I don't immediately see how to deal with the cases $n=3$ or $n=7$ but I suspect that all examples are at least homotopy equivalent to examples that have been mentioned already. I don't know about homeomorphism or diffeomorphism.

  • 4
    $\begingroup$ One can construct $OP^2$ by attaching a 16-cell to $S^8$ via the Hopf bundle map $S^{15}\to S^8$, and this Hopf bundle can be constructed just using the octonions. (Details can be found in Example 4.47 of my algebraic topology book.) So Lie groups aren't needed to construct $OP^2$. If one had a bundle $S^7 \to S^{23} \to OP^2$, couldn't one attach a 24-cell to $OP^2$ via the map $S^{23} \to OP^2$ to construct an $OP^3$ (which doesn't exist)? $\endgroup$ Nov 3, 2010 at 9:21
  • 2
    $\begingroup$ There are exotic $\mathbb{HP}^n$s X such that X is 1-connected and has the same integral cohomology ring as $\mathbb{HP}^n$, but not the same homotopy type. There is also up to homotopy a fibration $S^3\to S^{4n-3}\to X$. This can be found in an article by McGibbon: jstor.org/stable/1998715?seq=3 I don't know if these X can be chosen as manifolds and the fibrations as actual bundles. Probably at least the first part should be accesible to the methods of surgery since we have here a simply connected Poincare duality space. $\endgroup$ Nov 3, 2010 at 9:27
  • $\begingroup$ Allen is right. I remember reading somewhere (probably the article on Octonions by Baez, that although $\mathbb O \mathbb P^n$ exists for $n=1$ and $n=2$ only, the associated Hopf fibration $S^7 \to S^{8n+7} \to \mathbb O \mathbb P^n$ exists only for $n=1$. $\endgroup$ Nov 3, 2010 at 12:31
  • $\begingroup$ @Niel: very nice! $\endgroup$
    – Romeo
    Nov 4, 2010 at 0:43

This question was studied in a paper by Browder who proves the following.

Thereom. Consider any Serre fibration of $F\to S^n\to B$ where $F$, $B$ are connected polyhedra. Then $F$ is homotopy equivalent to $S^1$, $S^3$, or $S^7$. If $F=S^1$, then $B$ is homotopy equivalent to $CP^k$ with $2k+1=n$. If $F=S^7$, then $B$ is homotopy equivalent to $S^8$.

One expects that if $F= S^3$, then $B$ is homotopic to a quaternionic projective space of a suitable dimension but Browder mentions (if I am reading it correctly) that this is not true, and there are other examples.

Caution: there might be something wrong with the above theorem because it rules out existence of $S^7$ Hopf bundles over the octonian projective plane. Is there such a bundle? Hatcher's comment above says there isn't, am I reading it right? Could someone clarify the situation? I care because Browder's theorem was used in some geometric problems involving boundary at infinity of nonnegatively curved manifolds.

  • $\begingroup$ Sorry to return to this old answer, but I wanted to record a response to your caution. Browder only says in passing that $B$ should be $S^8$, saying that $H^*(B)$ is $Z[y_8]/y_8^l$ and then saying that $l=2$, citing this 3 page note of Adem: ncbi.nlm.nih.gov/pmc/articles/PMC1063836 ... but the note of Adem clearly indicates not that $l=2$ but rather that $l\le 3$. I think that clears up the situation? $\endgroup$ Apr 28, 2015 at 2:51
  • $\begingroup$ @GabrielC.Drummond-Cole: thank you for checking into Adem's note. The question remains (in my mind) whether $l=3$ is realized. Perhaps Browder knew it is not, which would explain his "$l=2$". Perhaps Hatcher's argument in the comments above justifies this (I have not thought about this recently). $\endgroup$ Apr 28, 2015 at 12:12

The Hopf fibrations give 4 such examples:

1) $S^3$ is an $S^1$ bundle over $\mathbb C \mathbb P^1 \cong S^2$, as you mentioned.

2) $S^7$ is an $S^3$ bundle over $\mathbb H \mathbb P^1 \cong S^4$, the quaternionic projective line.

3) $S^{15}$ is an $S^7$ bundle over $\mathbb O \mathbb P^1 \cong S^8$, the octonionic projective line.

4) $S^0$ is an $S^1$ bundle over $\mathbb R \mathbb P^1 \cong S^1$.

I don't know if there are any others.

  • 2
    $\begingroup$ These are certainly the only examples where all three spaces are spheres. But certainly there are more: all complex projective spaces have circle bundles whose total spaces are spheres. $\endgroup$ Nov 3, 2010 at 2:33
  • $\begingroup$ What's the map $S^9 \to CP^4$ giving a circle bundle? $\endgroup$
    – Romeo
    Nov 3, 2010 at 2:36
  • $\begingroup$ @Spiro: thanks for adding the other 2 classics! $\endgroup$
    – Romeo
    Nov 3, 2010 at 2:37
  • $\begingroup$ $CP^4 = (C^5-0)/C^*$, and since $C^5 \simeq R^10$, we can get an isomorphism $(C^5-0)/C^* \simeq S^9/U(1)$. So take $S^9 \subset C^5$ and then quotient by the natural diagonal action of the complex numbers of unit length. $\endgroup$
    – David Roberts
    Nov 3, 2010 at 2:54
  • $\begingroup$ that should be $R^{10}$ not $R^10$. $C^*$ is the nonzero complex numbers. $\endgroup$
    – David Roberts
    Nov 3, 2010 at 2:54

Here's a sketch. Take a look at the Gysin sequence. This will tell you that $H^i(B)$ and $H^{i+n+1}(B)$ are isomorphic except in two dimensions, which gives you (if $n$ is odd) a restriction on the possible cohomology rings for $B$: truncated polynomial algebras or (polynomial algebra tensor an exterior algebra). The latter cannot happen because such a space would be an infinite CW-complex and cannot be the base space of a fiber bundle with total space $S^m$ (something just homotopy equivalent to $S^m$ would be possible, though.) Hopf invariant 1 will then tell you that there's also a restriction on $n$. I'm guessing you'll end up with just the mentioned bundles over $RP^n$, $CP^n$, $HP^n$ and $OP^n$ (the latter for $n\leq 2$) and possibly some twisted versions of these.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.