MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Recall that Sperner's lemma is essentially a combinatorial version of the topological statement "A map from $S^n$ to $S^n$ with degree one cannot be nullhomotopic."

My question is, does there exist an analogous combinatorial lemma corresponding to the statement "A map from $S^3$ to $S^2$ with Hopf invariant one cannot be nullhomotopic"?

share|cite|improve this question
I don't know enough algebraic geometry to answer your question, but you may be looking for Tucker's lemma's_lemma which is a combinatorial version of the Borsuk-Ulam theorem. – Tony Huynh Jan 25 '14 at 3:13
This comment is algebraic rather than "combinatorial" so I don't give it as an answer. But Loday and I in Topology 26 (1987) 311-334, defined a tensor product $G \otimes H$ of groups which act on each other and on themselves by conjugation, and showed that $\pi_3SK(G,1)$ is the kernel of the "commutator map" $\kappa: G \otimes G \to G$. When you take $G=\mathbb Z$, you get the Hopf map is $1 \otimes 1$. – Ronnie Brown Jan 25 '14 at 10:29
@TonyHuynh The Borsuk-Ulam theorem is still inherently about $\pi_n(S^n)$, so Tucker's lemma is arguably just a different combinatorial version of the same topological statement as Sperner's lemma (although its precise relationship with Sperner's lemma is less than clear -- see this question). – Jim Belk Jan 25 '14 at 18:56
Lovely question! – Gil Kalai Jan 30 '14 at 20:42
up vote 6 down vote accepted

Sperner's Lemma. Let me start by talking through the implications for topology of Sperner's Lemma. The point is a map $S^1 \rightarrow S^1$ is null-homotopic exactly if it extends to a map $D^1 \rightarrow S^1$. Then the 3 coloring of Sperner's Lemma gives a map from the boundary of the triangulated triangle, to the boundary of a triangle (with the vertices corresponding to the three colors). The condition on the coloring amounts to requiring the map to be degree 1.

Since a map from a simplicial disc to a triangle maps only to its boundary if and only if no 2-simplex maps to the entire triangle, the conclusion follows.

Null-homotopic maps. Now, as Vidit Nanda has noted, a map from $S^3 \rightarrow S^2$ is similarly null-homotopic if and only if it extends to a map $D^4 \rightarrow S^2$. Thus, to show a simplicial map $f$ from the boundary of a triangulated 4-simplex $\Delta$ to the boundary of a tetrahedron is not null-homotopic, it suffices to show that some 3-simplex $\sigma$ in $\Delta$ is "rainbow" -- that is, that the 4 vertices of $\sigma$ map to the 4 vertices of the tetrahedron. (Rainbow being a good word if you think of the 4 vertices as being named Red, Blue, Green, and Yellow.)

The Hopf invariant. At this point, I can answer Jim's question: Yes, there exists a combinatorial analogue of Sperner's Lemma showing that a map with Hopf invariant 1 is not null-homotopic. The definition of simplicial cohomology is purely combinatorial, and the above allows us to translate "null-homotopic" into combinatorics. There surely exists a (possibly complicated) encoding of "Hopf invariant 1" into statements about colors.

Towards a concrete answer According to P Hilton's "An introduction to homotopy theory" Chapter 6, we can define the Hopf invariant for $f:S^3 -> S^2$ as follows: take a point $x$ of $S^2$. The inverse image of $x$ under $f$ is a (homology 1-) cycle $C$, and so there is a homology 2-chain $D$ which has $C$ as its boundary. Since $f$ collapses the boundary of $D$ (and leaves the rest alone), we get that $f(D)$ is a homology 2-cycle in $S^2$. If $f(D)$ is the fundamental cycle, then we say $f$ has Hopf invariant 1.

Thus, a map has Hopf invariant 1 if we can find $D$ and $C$ as above such that the induced map from $D/C \rightarrow S^2$ has degree 1.

Hilton's formulation of the Hopf invariant however does seem to be somewhat more tractable to turning into a combinatorial condition than the standard one from e.g. Wikipedia -- Hilton's is easier to work with simplicially! Incidentally, Hilton seems to say that his formulation is closer than the standard to Hopf's original statements.

Let me try to translate Hopf-invariant-one into a coloring condition:

A theorem: Let $f$ be a 4-coloring of the vertices of a triangulated 5-simplex $\Delta$, such that there is a subcomplex $D \subseteq \partial \Delta$ which is a triangulated polygon (i.e., 2-disc) satisfying the following coloring properties:
i) The boundary of $D$ is monochromatic (say, Red),
ii) No interior vertices of $D$ are colored Red, and
iii) The subcomplex of $D$ induced by the interior vertices of $D$ satisfies the Sperner condition on Green, Blue, and Yellow -- that is, there are boundary vertices $v_1, v_2, v_3$ colored Green, Blue, and Yellow, such that the outside path from $v_1$ to $v_2$ has all vertices Green or Blue, etc.
Then $\Delta$ has a rainbow (Red-Green-Blue-Yellow) simplex.

An illustration of the conditions on $D$ follows:

Example of $D$

Proof: If there is a rainbow simplex on $\partial \Delta$, then the result is trivial. Otherwise, $f$ induces a map $\partial \Delta \rightarrow S^2$, and the conditions above directly imply that $D/\partial D$ (with the induced coloring) satisfies the conditions of Sperner's Lemma for the boundary of the 2-simplex. We get that the map $f \vert_D$ induces a degree 1 map from $D / \partial D \cong S^2 \rightarrow S^2$, and thus that $f \vert_{\partial \Delta}$ has Hopf invariant 1. Thus, $f \vert_{\partial \Delta}$ is not null-homotopic, and the theorem follows from the discussion above of null-homotopic maps. $\square$

Loose ends The above theorem is a start towards what @JimBelk asked for, but there are several things that are unsatisfactory about it:

  1. It is not clear to me that the homology chain $D$ in the definition of the Hopf invariant need be a disc. This would be the main obstacle to proving from the above theorem that a map with Hopf invariant 1 is not null-homotopic.
    (If I understand Hopf's map correctly though, one can indeed take the homology chain to be a disc here. Since this is the only example in these dimensions, requiring $D$ to be a disc might be within the boundaries of good taste.)
  2. It would be nice to have a direct combinatorial proof of the above theorem. The main obstacle is seeing how the interior of the simplicial $D^4$ yields an interior of $D / \partial D$. (If you can understand some 3-dimensional interior of $D / \partial D$, then I guess it'll follow from Sperner.)
share|cite|improve this answer
Russ, I'm having a hard time trying to decipher the asymmetry in your combinatoria condition involving the simplicial 2-disk. Would you not ask for four such disks, one with each color along the bounding circle? – Vidit Nanda Feb 3 '14 at 18:25
If I'm understanding Hilton's formulation, I think it should suffice to take one such disc, corresponding with $f^{-1}$ of a single point. Of course, if $\Delta$ is sufficiently finely subdivided, it shouldn't matter what color you pick. – Russ Woodroofe Feb 3 '14 at 19:25
The relevant page of Hilton's text can be reached (at least for now) by Google Books at… – Russ Woodroofe Feb 3 '14 at 21:16

It seems really hard to impose combinatorial Sperner-like conditions which would guarantee the nontriviality of the Hopf invariant. But if you allow things to get slightly more algebraic by constructing a local system (or a cellular sheaf), then the following statement is a reasonable candidate.

Let $K$ be a triangulated $D^4$, i.e., any polyhedral subdivision of the $4$-simplex (with the triangulated $S^3$ in its boundary being denoted $\partial K$), let $L$ be the $3$-simplex on four vertices (which we might label R, G, B, Y). Finally, let $\rho:K \to L$ be any simplicial map so that the following conditions are satisfied:

  1. For $\sigma \in L$ of dimension $2$, (i.e., RGB, RGY, RBY, GBY) the corresponding fiber $\rho^{-1}(\sigma)$ restricted to $\partial K$ has the rational homology of a circle.
  2. For any face relation $\tau < \sigma$ in $L$ where $\dim \tau = 1$ and $\dim \sigma = 2$, the inclusion $\rho^{-1}(\sigma) \hookrightarrow \rho^{-1}(\tau)$ (again restricted to $\partial K$) induces an isomorphism on rational homology,

Then, there exists a simplex of $K$ which is sent by $\rho$ to the unique $3$-simplex of $L$.

From your point of view, the question of central interest might be "what combinatorial restrictions on $\rho$ guarantee 1. and 2. above?". I have no ideas about this yet.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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