38
$\begingroup$

This is a follow-up to another question.

A good interpretation of having an $n$-dimensional hole in a space $X$ is that some image of the sphere $\mathbb{S}^n$ in this space given by a mapping $f: \mathbb{S}^n \rightarrow X$ cannot shrink down to a point. The matter of "shrinking to a point" is best expressed by being $f$ homotopic to some constant map. Next, the homotopy groups $\pi_n$ can be defined as the homotopy classes of base-point preserving maps from $\mathbb{S}^n$ to $X$. In this way it might be argued that the homotopy groups $\pi_n$ should best capture the holes in $X$.

But this is not so. One has the most satisfying result that for $i \geq 1$ the homology $H_i (\mathbb{S}^n) $ is nontrivial iff $n = i$. But the higher homotopy groups of spheres are very complicated.

Why does this complication occur? Why are homology groups far better for capturing the holes than the homotopy groups, which are intuitively better suited, but are not actually so? In the case of $1$-dimensional holes, the homology $H_1$ and $\pi_1$ captures the holes equally well; but of course in this case the former is the abelianization of the latter.

$\endgroup$
14
  • 12
    $\begingroup$ I don't like the use of the term "hole", but in any case, one could argue the opposite: The complexity of homotopy groups is a reflection of the complicated ways in which holes can be manifested. For example, a connected space with simple nonabelian fundamental group will have trivial first homology, but I would claim that it has a nontrivial one dimensional hole. $\endgroup$
    – S. Carnahan
    May 15, 2010 at 1:32
  • 14
    $\begingroup$ Think about the Hopf map from $S^3$ to $S^2$. This shows that a 3-dimensional sphere can wind around, in a non-trivial way, a 2-dimensional "hole" (using your terminology). So one reason that homotopy groups are complicated is that higher dimensional spheres can wind around lower dimensional holes. $\endgroup$
    – Emerton
    May 15, 2010 at 6:21
  • 2
    $\begingroup$ If "hole" is not the acceptable terminology, then please suggest a better one. I thought it best reflects the intuitive picture. $\endgroup$
    – Akela
    May 15, 2010 at 6:25
  • 18
    $\begingroup$ I think a better terminology would be "cycle." Then the answer is that homology is better at detecting cycles than homotopy is because cycles are defined to be what homology detects. (-:O $\endgroup$ May 15, 2010 at 6:40
  • 4
    $\begingroup$ To elaborate on the Hopf map idea, the reason higher-dimensional things can "wrap around" lower-dimensional one (the only reason, over the rationals) is that preimages of subspaces of the lower-dimensional objects can be linked. $\endgroup$
    – Dev Sinha
    May 15, 2010 at 16:35

3 Answers 3

35
$\begingroup$

Homology also has complicated and unintuitive aspects if one goes beyond nice spaces like CW complexes. A surprising example of this is the subspace of Euclidean 3-space consisting of the union of a countable number of 2-spheres with a single point in common and the diameters of the spheres approaching zero. (This is the 2-dimensional analog of the familiar "Hawaiian earring" space.) Then the amazing fact is that the n-th homology group of this space with rational coefficients is nonzero, and even uncountable, for each n > 1. This was shown by Barratt and Milnor in a 1962 paper in the AMS Proceedings.

The result holds also with integer coefficients, with homology classes that are in the image of the Hurewicz homomorphism. So one could say that this strange behavior comes from homotopy groups but just happens to persist in homology. I would guess that there are also examples where the homology is not in the image of the Hurewicz homomorphism, so it does not come directly from homotopy groups.

$\endgroup$
1
  • 14
    $\begingroup$ But this is an anomaly of singular homology. The Cech cohomology of this Hawaiian pearl is what you'd expect, right? $\endgroup$ May 23, 2010 at 17:59
15
$\begingroup$

For another example that homology and homotopy capture holes differently: note that the curve $C$ in the bitorus below is homologous to zero but not homotopic to zero.

Also, a remarkable difference between them is that while an homology chain can be subdivided into smaller chains, a map from a sphere cannot be subdivided into smaller maps from a sphere. Historically, higher homotopy groups were only pursued after Hopf's fundamental example of a map $S^3 \to S^2$ not homotopic to zero, which was already commented above (I recommend the reading of H. Samelson "$\pi_3(S^2)$, H. Hopf, W. K. Clifford, F. Klein", In: History of Topology, p.575-578, Elsevier, Amsterdam, 1999.)

$\endgroup$
1
  • 2
    $\begingroup$ Much more generally, the Kan--Thurston theorem shows that there are finite cell complexes with fairly 'small' homotopy groups (zero in dimension >1) but arbitrarily complicated homology groups. $\endgroup$
    – HJRW
    Jul 3, 2017 at 8:35
12
$\begingroup$

I am no specialist, but I believe that it is not that the homology and homotopy capture better or worse the holes. I believe that they capture them differently.

Obviously, it is far easier to compute homology groups, but on the other hand, they give much less information in a lot of cases. I like Whitehead Theorem ( http://en.wikipedia.org/wiki/Whitehead_theorem ) as an example of how powerfull can homotopy groups. I don't know a counterpart for homology groups.

The other example I like, is the two dimensional torus. On the one hand, the two dimensional homology is non trivial while the second homotopy group is trivial which could indicate that homology is capturing a hole that homotopy does not capture. However, homotopy has already captured the holes of the torus (since the first homotopy group is non trivial). I think this example shows how it should be difficult to define the "dimension" of a hole.

So, I believe that homotopy groups should be considered all together and they will escentially capture all the holes.

$\endgroup$
2
  • 1
    $\begingroup$ One could think of the torus a different way: you can wrap a two-dimensional cycle around it, namely iself; this gives the fundamental class in $H^2$. But this two-dimensional cycle is not a 2-sphere, it is a 2-torus, and so it can't be detected by $\pi_2$. $\endgroup$
    – Emerton
    May 15, 2010 at 14:19
  • 8
    $\begingroup$ Whitehead's theorem does have a homology counterpart (for simply connected spaces, say); see Hatcher's AT, Cor. 4.33. E.g. by Whitehead plus Hurewicz, a simply connected homology $n$-sphere is a homotopy $n$-sphere. $\endgroup$
    – Tim Perutz
    May 15, 2010 at 15:56

Your Answer

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

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