Colouring signed analogues of Kneser, Schrijver, and Borsuk graphs

Let \mathcal{C}_i be the collection of connected components of C_i. Assume X \in \mathcal{C}_i. Since C_i is symmetric, -X\in \mathcal{C}_i. If X\cap (-X) \neq \emptyset (and hence X=-X), then we are done. Thus we may assume for X\cap (-X) = \emptyset for each element X of \mathcal{C}_i.

Let \mathcal{A}=\bigcup_{1\leq i \leq d} \mathcal{C}_i. Then \mathcal{A} satisfies all the conditions of Theorem 9.

Thus there are distinct sets X_1, X_2, \ldots X_{d+1} in \mathcal{A} and a point x such that

By the pigeonhole principle, two of these sets are in the same \mathcal{C}_i, leading to a contradiction.

Let d=n-k. We first embed \pm[n] into \mathbb{R}^{d+1} with the assistance of the odd moment curve. More precisely, let

for each i\in \pm[n] and let V=\{v_i:i\in \pm[n]\}. By the definition v_{-i}=-v_i for all i. Let V^+=\{v_{i}: i\in [n]\}. By a property of the moment curve, no hyperplane that passes through the origin intersects V^{+} in more than d point (see Lemma 1.6.4 of [M03]).

We now claim that the mapping i\mapsto w_i=v_i/|v_i| is the desired embedding of \pm[n] in the sphere S^{d}.

Let a=(a_1,a_2,\ldots,a_{d+1})\in S^d. The hyperplane h_a=\{x\in \mathbb{R}^{d+1}:x\cdot a=0\} passing through the origin and perpendicular to a partitions \mathbb{R}^{d+1} into three regions, namely h_a, h_a^+=\{x\in \mathbb{R}^{d+1}:x\cdot a>0\}, and h_a^-=\{x\in \mathbb{R}^{d+1}: x\cdot a<0\}. The open hemisphere centered at a is H_a=S^d\cap h_a^+. We shall find an alternating k-set X\in \mathcal{A}(n,k) whose image \{w_i:i\in X\} is contained in H_a, equivalently \{v_i:i\in X\} \subset h_a^+.

To do so, we first continuously move the vector a\in S^{d} to increase the number of points of V contained in the hyperplane h_a while no points of V get swept through by h_a, i.e., each v_i stays in h_a\cup h_a^+ or h_a\cup h_a^- that it originally belonged to.

Since no d+1 points of V^+=\{v_{i}: i\in [n]\} is on h_a, and noting that 0\in h_a, we can do this by gradually increasing (one at a time) the intersection h_a\cap V^+ while fixing the subspace generated by the vectors already in h_a\cap V, until we reach the vector a'=(a'_1, a'_2, \ldots, a'_{d+1}) such that |h_{a'}\cap V^+|=d. Furthermore, observe that v_i\in h_{a'} if and only if v_{-i} \in h_{a'}, thus |h_{a'}\cap V|=2d at the end of this process.

Thus, |V\setminus h_{a'}|=2k, and, since V is antipodally symmetric about the origin, we must have |V\cap h_{a'}^+|=|V\cap h_{a'}^-|=k. The process of obtaining h_{a'} from h_{a} guarantees that V\cap h_{a'}^+\subseteq V\cap h_a^+ and V\cap h_{a'}^-\subseteq V\cap h_a^-. Hence, to complete the proof, it suffices to show that V\cap h_{a'}^+ is the image of an alternating k-set.

Let p(x)=a'_1x+a'_2x^3+\cdots+a'_{d+1}x^{2d+1}. By the choice of a', p(x) has 2d+1 simple roots: 0 and d pairs of antipodal elements of \pm[n]. Observe that v_i\in h_{a'}^+ if and only if (-1)^ip(i)>0. Hence X=\{i\in\pm[k+d]:v_i\in h_{a'}^+\}=\{i\in\pm[k+d]:(-1)^ip(i)>0\}.

To complete the proof it is enough to prove that:

Claim. X\in\mathcal{A}(n,k), that is, X is an alternating k-set.

Proof of Claim. First of all, since v_i and v_{-i} are on the opposite sides of h_{a'}, X does not contain an antipodal pair of indices. Hence X \in \binom{[n]}{\pm k}.

To see that X is alternating, suppose, to the contrary, that i_l and i_{l+1} are two indices in X of the same sign with adjacent absolute values, that is, there is no j\in X with |i_l|<|j|<|i_{l+1}|. This implies that, all the integers in (i_l,i_{l+1}) are (simple) roots of p(x).

If i_l,i_{l+1} are of the same parity, then p(i_l) and p(i_{l+1}) are of the same sign. So, the number of roots of p(x) on (i_l,i_{l+1}) is even, contradicting the fact that the number of integral points in (i_l,i_{l+1}) is odd. Similarly, if i_l and i_{l+1} are of opposite parities, then p(i_l) and p(i_{l+1}) are of opposite signs and similarly we get a contradiction. Therefore, X is alternating.

The lower bound follows from Proposition 12, Theorem 6, and Theorem 8. Hence it is enough to give an (n-k+1)-colouring for \widehat{\mathit{KS}}(n,k)^- (hence also for \widehat{\mathit{SS}}(n,k)^-). To that end, we observe that \mathcal{B}_i^+(n,k) for i=1,2,\ldots, n-k+1 covers all vertices of \widehat{\mathit{KS}}(n,k) because the first nonzero element of each vertex is positive.

The first claim is clear once we recall from the proof of Theorem 6 that \mathit{KS}(n,k)^- contains S(2n,k) as a subgraph.

To see the second part, first notice that \mathcal{B}_i^+(n,k), i=1,2,\ldots, n-k+2 is an (n-k+2)-colouring of \mathit{SS}(n,k)^-, establishing the upper bound.

The lower bound is already implicitly proved along the way of proving Theorem 8, so we give a sketch of it. Using Theorem 11, we embed \pm[n] in S^{n-k} in such a way that i and -i are antipodal for each i\in [n], and any open hemisphere contains an alternating k-set.

Suppose there is an (n-k+1)-colouring c of \mathit{SS}(n,k)^-, let

for i\in [n-k+1].

Since each hemisphere contains an alternating k-set, A_1,A_2,\ldots,A_{n-k+1} gives an open cover of S^{n-k}. By the Borsuk-Ulam theorem (cf. Theorem 17), there is an A_i that contains a pair of antipodal points of S^{n-k}. However, this gives a pair of disjoint alternating k-sets in the colour class c^{-1}(i). A contradiction.

Note that S(n-1,1)=K_{n-1}. We need to show that for any switching equivalent copy of \widehat{\mathit{SS}}(n,2), its negative subgraph has clique number at least n-1.

Let B be the bipartite graph with parts \{1, 2, \ldots, n\} and \{-1, -2, \ldots, -n\}, where \{i, -j\} is an edge if i< j. The vertices of \widehat{\mathit{SS}}(n,2) are the edges of B, where two vertices (that is, the edges of B) are connected by a negative edge if they form a matching. Thus the clique number of the subgraph induced by the negative edges in \widehat{\mathit{SS}}(n,2) is the maximum size of a matching in B. To switch a vertex \{i,-j\} means to replace this edge by \{j,-i\} in B.

So we need to prove that for any subset S of E(B), replacing each edge \{i,-j\} \in S with edge \{j,-i\}, the resulting bipartite graph B' has a matching of size n-1. By Kőnig’s theorem, it suffices to show that the minimum size of a vertex cover of B' is n-1.

Note that for any i, d_B(i)=n-i and d_B(-i) = i-1. So d_B(i)+ d_B(-i)=n-1. Replacing edge \{i,-j\} with edge \{-i, j\} does not change the sum of the degrees of i and -i. So d_{B'}(i)+ d_{B'}(-i)=n-1 for i \in [n].

Let C be a cover of B'. If for some i, C \cap \{i,-i\} = \emptyset, then all neighbours of i and -i in B' must be in C. Thus |C|\geq n-1. Otherwise, C \cap \{i,-i\} \ne \emptyset for i=1,2,\ldots, n, and hence |C|\geq n. This completes the proof of Theorem 16.

Assume that antipodally symmetric sets F_1, F_2,\ldots, F_d is a closed cover of S^d. Further, suppose for each i\in[d] there is an open neighbourhood U_i\supset F_i that does not connect any pair of antipodal points.

By compactness of the sphere, there is an \varepsilon_i>0 such that F_i\subset F_{i,\varepsilon_i}\subset U_i where F_{i, \varepsilon_i} is the \varepsilon_i-neighbourhood of F_i. Being a subset of U_i, F_{i,\varepsilon_i} does not connect a pair of antipodal points either. For every \varepsilon_0>0, let \varepsilon=\min_{i\in\{0,1,2,\ldots,d\}}\varepsilon_i. We claim that F_i is a balanced set in \mathit{BS}(d, \varepsilon).

If x, y\in F_i belong to different connected components of F_{i,\varepsilon_i}, then \operatorname{dist}(x,y)\geq 2\varepsilon_i>\varepsilon. In that case x and y cannot be joined by a positive edge in \mathit{BS}(d, \varepsilon). On the other hand, if x, y\in F_i belong to the same component of F_{i,\varepsilon_i}, since -x is not in the same component as x, we have \operatorname{dist}(-x,y)\geq 2\varepsilon_i>\varepsilon. This shows that vertices in the same component cannot be joined by a negative edge.

We now claim that F_{i} is a balanced set of \mathit{BS}(d, \varepsilon). Being symmetric, if there is a negative cycle in F_{i}, there is a negative cycle, say C, with exactly one negative edge. In this cycle, the sequence of positive edges implies that all its vertices are in the same component of F_{i,\varepsilon_i}. But the two ends of the only negative edge must be on two different components. The collection of F_{i}, 1\leq i \leq d, then gives a balanced d-colouring of \mathit{BS}(d, \varepsilon), contradicting (a).

A symmetric open cover A_1, \ldots, A_{d} of S^d, corresponds to a natural open cover U_1, \ldots, U_{d} of \mathbb{R}\mathrm{P}^d through the quotient map q: S^d\to \mathbb{R}\mathrm{P}^d that identifies the antipodal points. Then, each of the U_i’s is non-contractible if and only if the corresponding A_i connects a pair of antipodal points of S^d, (see[Hatcher2001], Example 1.43 for more details).