# The free loop space of spheres

Let $$n>1$$. The homology of the free loop space $$\Lambda S^n$$ of the sphere $$S^n$$ contains two torsion if $$n$$ is even. Thus the fibration $$\Omega S^n\rightarrow \Lambda S^n\rightarrow S^n$$ is not trivial if $$n$$ is even (Here $$\Omega S^n$$ denotes the based loop space.).

The odd dimensional spheres do not have torsion in the homology of the free loop space and one can compute that $$H_*(\Lambda S^{2k+1};\mathbb Z)\cong H_*(\Omega S^{2k+1}\times S^{2k+1};\mathbb Z)$$. Hence the homology does not obstruct the existence of a trivialization of the free loop fibration. Indeed $$\Lambda S^3$$ and $$\Omega S^3\times S^3$$ are homeomorphic, which can be proven using the group structure on $$S^3$$.

Is the free loop space fibration always trivial for odd dimensional spheres ? My guess would be that this is not the case, with a possible exception of $$S^7$$.

• It definitely splits for $S^7$ since it's an H-space: mathoverflow.net/a/207856/36146 – Najib Idrissi May 31 at 12:38
• Very nice question! One observation is that you can't use cup products in cohomology to detect non-splitting, by a result of Menichi. Also it seems that $LS^n$ and $\Omega S^n\times S^n$ are rationally homotopy equivalent for $n$ odd (by looking at the minimal models). This doesn't quite do it though. – Mark Grant May 31 at 15:19
• I was wondering if the fact that the Whitehead square $\[\iota_n,\iota_n]\in \pi_{2n-1}(S^n)$ is non-trivial if $n\neq 1,3,7$ can be used to detect non-splitting. It certainly gives non-splitting of the analogous fibration where maps from $S^1$ are replaced by maps from $S^p$. – Mark Grant May 31 at 15:23
• Your desired equivalence exists after p-localizing for p odd, because odd-dimensional spheres are p-local H-spaces. – skd May 31 at 16:37
• Here is the link to the paper that was already mentioned: jstor.org/stable/43741886?seq=1#metadata_info_tab_contents – Jens Reinhold May 31 at 19:25

I am grateful to Tobias Barthel, who sent me the following paper of J. Aguadé:

"On the space of free loops of an odd sphere". Pub. Mat. UAB No 25, June 1981.

1. $$\Lambda S^{2n+1}\simeq S^{2n+1}\times \Omega S^{2n+1}$$;
3. $$n=0,1,3$$.
That 3 implies 2 is due to the fact that these spheres are $$H$$-spaces, which was already noted. 2 implies 1 is trivial. Hence the only thing Aguadé shows is that 1 implies 3.
For this it is assumed that there is a map $$f$$ inducing a homotopy equivalence. From this there is an induced map $$h:S^1\times S^{2n+1}\times \Omega S^{2n+1}\rightarrow S^{2n+1}$$. Aguadé applies the Hopf construction to this map to get a map $$\tilde g:S^{2n+1}*(S^1\times S^{2n})\rightarrow S^{2n+2}.$$ The space on the left is a wedge of spheres $$S^{2n+3}\vee S^{4n+2}\vee S^{4n+3}$$, thus there is an induced map $$g:S^{4n+3}\rightarrow S^{2n+2}$$. Aguadé shows that this map has Hopf invariant one, hence the result follows.