This question is a follow-up to a previous question answered by Neil Strickland:

Map from simplex to itself that preserves sub-simplices

Let $B$ denote the closed unit ball in $\mathbb{R}^2$ and let $\Delta_{n-1}$ denote the $(n-1)$-simplex. I have a continuous function $f(x_1,\dots,x_n):B^n \rightarrow \Delta_{n-1}$ defined for all subsets $\lbrace x_1,\dots,x_n\rbrace \subset B$ of size $n$ that satisfy $x_i \neq x_j$ for all pairs $i,j$ (in other words, the function is only defined if all of the $n$ arguments are distinct). This function has the property that, if $\sigma$ denotes a permutation, then $f(\sigma(x_1,\dots,x_n)) = \sigma(f(x_1,\dots,x_n))$. In other words, permuting the arguments of the function merely permutes the output. My question is: are there non-trivial sufficient conditions on $f$ under which the point $(1/n , \dots, 1/n)$ lies in the image of this map? (or, even better, is this always the case?)

Here's one property of the map $f$ that I can add regarding the requirement that arguments be distinct: if $\lbrace \mathbf{x}_k \rbrace$ is a sequence of $n$-tuples (with distinct entries) in $B$ that converges to an $n$-tuple $\bar{\mathbf{x}}$ with (possibly) non-distinct entries, then the limit of $f(\mathbf{x}_k)$ exists if and only if, for each pair of entries $x_i^k$ and $x_j^k$ in the $n$-tuple, the unit direction vector from $x_i^k$ to $x_j^k$ (i.e. $\frac{x_i^k - x_j^k}{||x_i^k - x_j^k||}$) has a limit.

  • 1
    $\begingroup$ I don't think this is true as stated — or rather, as stated modulo making $B$ the closed unit ball in ${\bf R}^n$, rather than ${\bf R}^2$. Removing from $B$ all points where two coordinates coincide leaves $n!$ connected components, one for each ordering of the $n$ coordinates. These are permuted simply transitively by coordinate permutations. So I can fix a component $c_0$, and a point $x_0 \in \Delta_{n-1}$ with distinct coordinates, and map each component $\sigma(c_0)$ to $\sigma(x_0)$. This function satisfies your hypotheses but not the desired conclusion. $\endgroup$ – Noam D. Elkies Jul 13 '11 at 4:06
  • 2
    $\begingroup$ Sorry, I think I didn't explain this properly (or I've misunderstood your explanation); what I meant was that $f$ is a map that takes an $n$-tuple of points in $B$ and spits out a point in $\Delta_{n-1}$, and that $f$ is defined if each of the $n$ arguments in the $n$-tuple is distinct. So for example, if $n=2$, then the $2$-tuple $( (1/2,1/2),(1/3,1/3)$ is allowed, but $((0,1),(0,1))$ is not. This doesn't seem to be the same as "Removing from $B$ all points where two coordinates coincide". Did I miss something? Thanks! $\endgroup$ – Jennifer Gao Jul 13 '11 at 6:01
  • 1
    $\begingroup$ Sorry, I indeed misread your question: you're going from $(2-{\rm ball})^n$ to $\Delta$, not from $(n-{\rm ball})$ to $\Delta$ as I thought. $\endgroup$ – Noam D. Elkies Jul 13 '11 at 15:30
  • $\begingroup$ I just now added one more condition on the map that may be helpful. Also, thanks to Neil and gowers for the helpful comments. $\endgroup$ – Jennifer Gao Jul 13 '11 at 21:38
  • $\begingroup$ If there is a counterxample then there is one with your "extra property". (Remove aneighborhood of your hyperlines and inductively extend the function so that it is constant along the radials.) $\endgroup$ – Anton Petrunin Jul 14 '11 at 7:30

This turns out to be a remarkably interesting question. I don't have a complete answer but here is a start.

First, for any space $X$ I'll write $F_n(X)\subset X^n$ for the space distinct $n$-tuples. The question asks whether there exist $\Sigma_n$-equivariant maps $f:F_n(B^2)\to\Delta_{n-1}$ with the fixed point $b=(1/n,\dotsc,1/n)$ not in the image. If there is such a map then we can push it away from $b$ to the boundary of $\Delta_{n-1}$, which is homeomorphic to $S^{n-2}$. On the other side, there is an obvious embedding $i:B\to\mathbb{R}^2$ and one can choose an embedding $j:\mathbb{R}^2\to B$ such that $ij$ and $ji$ are isotopic to the respective identity maps; using this we see that $F_n(B^2)$ is equivariantly homotopy equivalent to the space $X=F_n(\mathbb{R}^2)$. This space is well-known: the following paper is one entry point to the literature:

   author={Cohen, F. R.},
   title={On configuration spaces, their homology, and Lie algebras},
   journal={J. Pure Appl. Algebra},
   review={\MR{1344842 (96d:55005)}},

In particular:

  1. $\pi_1(X)$ is the pure braid group $Br_n$ on $n$ strings. Moreover, the higher homotopy groups are trivial, so $X$ is the classifying space $BBr_n$.

  2. $X$ has an equivariant deformation retract $X_0$ that is a finite simplicial complex of dimension $n-1$. This means that $H^k(X)=0$ for $k>n-1$.

  3. The cohomology of $X$ is completely known, together with the action of $\Sigma_n$. In particular, the top group $H^{n-1}(X)$ is the module known as $\text{Lie}(n)$ (or maybe the dual of that?). As a $\mathbb{Z}[\Sigma_{n-1}]$-module this is free of rank one, but the $\Sigma_n$-action is harder to describe. The standard description also implies that all Steenrod operations in $H^*(X_0;\mathbb{Z}/p)$ are trivial.

If we can show that there is no $\Sigma_n$-equivariant map from $X_0$ to $S^{n-2}$ then we will be done.

In the case $n=2$ we just have $X_0=S^1$ and $S^{n-2}=S^0$ with $\Sigma_2$ acting antipodally on both sides: it is clear that there is no equivariant map, as required.

In the case $n=3$ we have $S^{n-2}=S^1=K(\mathbb{Z},1)$, so the nonequivariant mapping set is $[X_0,S^1]=H^1(X_0)$, and one can check that this is just $\mathbb{Z}^3$ with the action given by permuting the coordinates and multiplying by the signature. The only fixed point for this action is zero, so any map $X_0\to S^1$ that is equivariant-up-to-homotopy is nonequivariantly homotopic to a constant map. Here the action of $\Sigma_3$ on $S^1$ is generated by a reflection and a rotation through $2\pi/3$, so there are no fixed points. This means that constant maps $X_0\to S^1$, although equivariant-up-to-homotopy, cannot be equivariant on the nose. I suspect that there are no equivariant maps, and it should be possible to prove this by equivariant obstruction theory (ie working up the skeleta of $X_0$) but I do not see the details at the moment.

For $n>3$ we still have an evident map $[X_0,S^{n-2}]\to H^{n-2}(X_0)$, but it need not be bijective. We can compare $S^{n-2}$ with the fibre of the map $Sq^2:K(\mathbb{Z},n-2)\to K(\mathbb{Z}/2,n)$, recalling that $Sq^2$ acts trivially on $H^*(X_0)$, which should give an explicit description of $[X_0,S^{n-2}]$. With a bit of representation theory we should be able to calculate the group of equivariant-up-to-homotopy maps $X_0\to S^{n-2}$. We would then need some equivariant obstruction theory to improve this to understand whether there are any equivariant maps. Because $\Sigma_n$ acts freely on $X_0$ and $S^{n-2}$ is nonequivariantly $(n-3)$-connected and $X_0$ is $(n-1)$-dimensional, this obstruction theory will only involve the last two or three skeleta of $X_0$, so it should hopefully be tractable.


Let me rephrase the question and give a complete answer:

What are conditions for the existence of a $\Sigma_n$-equivariant map $$f: F_n(B)\to \Delta_{n-1}\backslash(\tfrac{1}{n},\dots,\tfrac{1}{n})\; ?$$ Here $B\subset \mathbb{R}^2$ denotes the closed unit ball, $F_n(B)$ is the ordered configuration space of $n$ points on $B$ with $\Sigma_n$ acting by permutation and $\Sigma_n$ acts on $\Delta_{n-1}:=\{(x_1,\dots,x_n)\in\mathbb{R}^n: \sum_ix_i=1, 0\leq x_i\leq 1\}$ by permuting the coordinates.

As Neil Strickland pointed out, using an equivariant homeomorphism, we can replace $\Delta_{n-1}\backslash(\tfrac{1}{n},\dots,\tfrac{1}{n})$ by $\partial\Delta_{n-1}\cong_{\Sigma_n} S^{n-2}.$ This sphere $S^{n-2}$ can be viewed as $S(W_n)$, where $W_n:=\{(x_1,\dots,x_n)\in\mathbb{R}^n: \sum_ix_i=0\}$ and $\Sigma_n$ is again acting by permuting the coordinates. Using equivariant homotopy equivalences, we can replace $F_n(B)$ by $F(\mathbb{R}^2)$ and further use an equivariant model of dimension $n-1$, which Neil Strickland calls $X_0$ and I will call $\mathcal{F}_n(\mathbb{R}^2)$. So an equivalent question is:

What are conditions for the existence of a $\Sigma_n$-equivariant map $$f: \mathcal{F}_n(\mathbb{R}^2)\to S(W_n)\; ?$$

Fortunately, there is a complete answer to that questions; it is given by Blagojević and Ziegler: arxiv, springer:

Such a map exists if and only if $n\geq 2$ is a not a prime power

The authors use equivariant obstruction theory and an explicit model $\mathcal{F_n(\mathbb{R}^2)}$, which is called $\mathcal{F}(d,n)$ in their notation. Just plug in $d=2$ and the result is the main theorem of section 4 in the paper on the arxiv.

From the nonexistence of the map we conclude that the barycenter $(\frac{1}{n},\dots,\frac{1}{n})$ is hit: If $n$ is a prime power, then your continous equivariant function $B^n\to\Delta_{n-1}$ will always hit the barycenter and if $n$ is not a prime power, there exist equivariant functions that miss the barycenter.

  • 2
    $\begingroup$ Fascinating! From skimming the paper, it appears that their argument is entirely obstruction-theoretic and doesn't exhibit an explicit map when $n$ is not a prime power. I would love to see such an explicit construction. $\endgroup$ – Eric Wofsey Sep 3 '13 at 13:38
  • 1
    $\begingroup$ @Eric yes, that would be interesting indeed! $\endgroup$ – Moritz Firsching Sep 4 '13 at 7:40

This isn't an answer to your question but rather a comment that is too long for a comment. I once tried to solve what was then an open problem called the Knaster hypothesis. It's a beautiful and highly plausible statement: that if $f$ is a continuous function defined on the sphere $S^n$ and $x_1,\dots,x_{n+1}$ are points in $S^n$, then there is an orthogonal map $T\in SO(n+1)$ such that $f(Tx_1)=\dots=f(Tx_{n+1})$. One reason for being interested in the conjecture was that, as observed by Milman, it easily implies Dvoretzky's theorem with an essentially optimal bound in all parameters.

When $n=1$ we have two points in the circle and the result can be proved by an easy application of the intermediate value theorem. When $n=2$ the result is true but nothing like so easy. In that case we can think about the map from $SO(3)$ to $\mathbb{R}^3$ that takes $T$ to the triple $(f(Tx_1),f(Tx_2),f(Tx_3))$. This map is continuous, and if $f$ is a counterexample to the Knaster hypothesis then the image of this map does not intersect the line $x=y=z$. So one can hope to try to prove the result by contradiction, showing that the image somehow "surrounds" the line $x=y=z$ enough to be forced to contain it (just as a map from the closed disc to itself that preserves the boundary must map something to the centre of the disc). It is this that I am reminded of by your question, though I don't claim that there is a close connection.

If I remember correctly a proof along those lines can be given when $n=2$, but the conjecture in general turned out to be false. This interesting (but slightly disappointing since the hypothesis would have been such a nice theorem) result was proved by Kashin and Szarek. Their function $f$ was just the $\ell_\infty$ norm. The set of points was slightly more complicated but not too bad.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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