Stiefel Manifolds and Coloring the Pentagon

James Dovera,1,∗, Murad Özaydına

a
Department of Mathematics, University of Oklahoma, 601 Elm Avenue, Norman, OK,
USA 73019-3103
arXiv:1302.2831v1 [math.GT] 12 Feb 2013

Abstract
We prove Csorba’s conjecture that the Lovász complex Hom(C5 , Kn ) of graph
multimorphisms from the 5–cycle C5 to the complete graph Kn is Z/2Z–
equivariantly homeomorphic to the Stiefel manifold, Vn−1,2 , the space of
(ordered) orthonormal 2–frames in Rn−1 . The equivariant piecewise-linear
topology that we need is developed along the way.
Keywords: Stiefel manifolds, PL topology, group action, graph morphism
complexes
2010 MSC: 57S25, 57Q15, 05E18, 57Q91, 57Q40, 05E45, 57S17

1. Introduction
In his remarkable proof of the Kneser conjecture [14], Lovász gave a lower
bound for the chromatic number of a graph using equivariant algebraic topol-
ogy. This was essentially done via a functor (the edge complex, defined below)
from the category of graphs and graph morphisms to the category G–T OP
(topological spaces equipped with an action of the group G and G–maps, i.e.,
continuous functions commuting with the given actions). In the case of the
edge complex functor, the group G is of order two, and the actions are free.
The edge complex of the complete graph Kn is (equivariantly homeomorphic
to) the sphere S n−2 with the antipodal action, and the punchline is provided
by the Borsuk-Ulam theorem.

∗
Corresponding author
Email addresses: [email protected] (James Dover), [email protected]
(Murad Özaydın)
1
Present address: Department of Mathematical Sciences, Cameron University, Burch
Hall, Room B001, 2800 W. Gore Blvd., Lawton, OK, USA 73505, Phone: 580-581-2488,
Fax: 580-581-2616

Preprint submitted to Elsevier November 20, 2018

 In [6], Csorba observed that, for n = 2, n = 3, and n = 4, the 5–cycle
complex (defined below) of the complete graph Kn , denoted Hom(C5 , Kn ),
is equivariantly homeomorphic to the Stiefel manifold Vn−1,2 of (ordered) or-
thonormal 2-frames in the Euclidean space Rn−1 , and he conjectured that
this was true for all n. The non-equivariant version of Csorba’s conjecture
was proven by C. Schultz in [18], who also proved that Hom(C5 , Kn ) is equiv-
ariantly homotopy equivalent to Vn−1,2 . In this note we give a proof of the
equivariant Csorba conjecture: Hom(C5 , Kn ) is equivariantly homeomorphic
to Vn−1,2 with respect to the actions described below. This provides a rather
mysterious combinatorial model for the Stiefel manifolds Vn−1,2 which are
fundamental topological spaces.
The edge complex and the (5–)cycle complex are special cases of the
Lovász graph multimorphism complex, which we proceed to define precisely
now. A graph morphism f is a function from VΓ to VΛ (where VΓ denotes
the vertex set of the graph Γ) so that the image of an edge of Γ is an edge
of Λ. (In particular, a graph morphism from Γ to the complete graph Kn is
just an admissible coloring of the vertices of Γ.) A multimorphism from Γ
to Λ is a relation φ ⊆ VΓ × VΛ such that:
(i) φ(v) := {w ∈ VΛ |(v, w) ∈ φ} is non-empty for all v ∈ VΓ , and
(ii) any function f ⊆ φ is a graph morphism.
In other words, φ is a multimorphism when there is a function f ⊆ φ and
any function f ⊆ φ is a graph morphism.
The Lovász multimorphism complex Hom(Γ, Λ) [3] is a bifunctor (con-
travariant in the first variable, covariant in the second) assigning a regular
cell complex to the (simple) graphs Γ, Λ. Any regular cell complex is de-
termined by its face poset ([5] p200, [15] Ch 3 §1), and the face poset of
Hom(Γ, Λ) is the set of multimorphisms from Γ to Λ ordered by inclusion.
Geometrically, the cells of the complex Hom(Γ, Λ) are products of simplices;
the cell corresponding to the multimorphism φ:
Y
∆φ(v)
v∈VΓ

where ∆A is the full simplex on the vertex set A. Hom(Γ, Λ) is the union
of all these cells indexed by the multimorphisms. We will identify each cell
with the multimorphism φ indexing it and refer to it as such. The vertices
(0–cells) of Hom(Γ, Λ) are the graph morphisms from Γ to Λ, the vertices of
a given cell φ are obtained by choosing any w from φ(v) for each v ∈ VΓ .

2
 The composition of two multimorphisms is also a multimorphism, hence
Hom(Γ, Λ) is (bi-)functorial for multimorphisms (not just for morphisms, this
extended functoriality was put to very good use in [20]). That is, a multi-
morphism α : Γ0 → Γ induces a (cellular) map Hom(Γ, Λ) → Hom(Γ0 , Λ),
and a multimorphism β : Λ → Λ0 induces a (cellular) map Hom(Γ, Λ) →
Hom(Γ, Λ0 ).
If G and H are groups acting on the graphs Γ and Λ respectively, then
there is an action of G×H on VΓ ×VΛ inducing an action on the relations, i.e.,
subsets of VΓ ×VΛ , which restricts to an action on Hom(Γ, Λ). In particular, a
graph Γ equipped with an action of a group G defines the functor Hom(Γ, −)
from the category of graphs and graph (multi-)morphisms to G–T OP , the
equivariant category of G–spaces and G–maps. Lovász’s proof of the Kneser
conjecture [14] essentially employs the edge complex functor Hom(K2 , −)
with G being the automorphism group of (the edge) K2 , the complete graph
on 2 vertices.
If we denote the vertices of K2 with + and −, then the nontrivial element
of G is the involution on Hom(K2 , Λ) switching φ(+) and φ(−), the subsets
(of VΛ ) assigned to the two vertices of K2 . This action is free because φ(+)
and φ(−) are distinct since they are nonempty and disjoint. In particular,
Hom(K2 , Kn ) is a poset consisting of pairs (A, B) of nonempty disjoint sub-
sets of {1, . . . , n}, ordered by component-wise inclusion, where A = φ(+) and
B = φ(−). Hom(K2 , Kn ) can be geometrically realized as the (n − 2)–sphere
of radius two with respect to the L1 norm (inducing the taxicab metric) in the
hyperplane orthogonal to the diagonal vector (1, 1, . . . , 1) in the Euclidean
space Rn , where φ(+) and φ(−) are the sets of coordinates with positive and
negative values respectively. Switching φ(+) and φ(−) yields the antipodal
action.
Interestingly, long before the definition of Hom(Γ, Λ), the underlying
spaces of Hom(Km , Kn ) figured prominently in two unrelated applications
of equivariant algebraic topology (neither one involving chromatic numbers
or graphs): in Alon’s elegant Necklace Splitting Theorem (with m prime) [1]
and the solution of the prime power case of the Topological Tverberg Prob-
lem ([16] and [22]), which was conjectured by Bárány, Shlossman, and Szűcs
in [2].
The Lovász conjecture (proven by Babson and Kozlov [4], see also [20])
is about the (odd) cycle complexes Hom(Cm , Λ) where Cm is the m–gon.
Hom(Cm , Λ) also has an involution induced by a reflection (of the m–gon).
When m is odd, any such reflection flips an edge and hence induces a free

3
action on Hom(Cm , Λ) as above. The Lovász conjecture reduces to a compu-
tation involving the equivariant cohomology of Hom(Cm , Kn ).
A graph multimorphism φ in Hom(Γ, Λ) is also determined by specifying
−1
φ (w) := {v ∈ VΓ |(v, w) ∈ φ} for each w ∈ VΛ . Clearly, for any φ, each
φ−1 (w) is independent (i.e., no two elements are adjacent in Γ). The in-
dependence complex ind(Γ) of a graph Γ is the simplicial complex with
vertex set VΓ and simplices being independent subsets of VΓ . Equivalently,
ind(Γ) is the flag complex (the largest simplicial complex whose 1–skeleton
is the given graph) of the edge complement of Γ (the graph with the same
vertices as Γ but with the complementary edge set). Any flag complex is the
independence complex of the edge complement of its 1–skeleton.
When Λ = Kn , the only condition on φ−1 (w) is being in ind(Γ), so
Hom(Γ, Kn ) consists of φ such that:
(i) φ(v) is non-empty for all v ∈ VΓ , and
(ii) φ−1 (j) is independent for j = 1, . . . , n.
We can identify Hom(Γ, Kn )>φ with the face poset of the join (over j =
1, . . . , n) of the links of the φ−1 (j) in ind(Γ). Since Hom(Γ, Kn )<φ is the face
poset of the boundary of a product (over the vertices v ∈ VΓ ) of the simplices
on φ(v), Hom(Γ, Kn ) is a (closed) manifold if ind(Γ) is a PL (piecewise-linear)
sphere. In fact, the converse is also true [6]. The complex Hom(Cm , Kn ) is
a manifold only when m = 5 since this is the only time when ind(Cm ) is a
sphere: ind(C3 ) is 3 points, ind(C4 ) is the disjoint union of two edges, ind(C5 )
is a pentagon, and ind(Cm ) with m > 5 has maximal simplices of different
dimensions.
On the topological side, the orthogonal group O2 acts on the Stiefel man-
ifold Vn−1,2 := {(x, y) ∈ S n−2 × S n−2 | x · y = 0} (with the Grassmannian
Grn−1,2 of 2–planes in (n − 1)–space as the quotient). The group O2 is the
semi-direct product of rotations SO2 with a subgroup generated by an arbi-
trary reflection. Two natural involutions on Vn−1,2 are (i) (x, y) 7→ (x, −y)
and (ii) (x, y) 7→ (y, x). Since any two reflections are conjugate via a rotation,
these give equivalent actions. An explicit map Vn−1,2 → Vn−1,2 interchang-
ing the actions (i) and (ii) is (x, y) 7→ √12 (x + y, x − y). Schultz [18] used
the action (i) on the Stiefel manifold. Below, we will use the (equivalent)
action (ii). The corresponding involution on Hom(C5 , Kn ) is induced by any
reflection of the pentagon C5 (they all give equivalent actions).
It is convenient to work with a smaller model: HomI (Γ, Λ), in which
two multimorphisms are considered the same if their values differ only on

4
the independent set I of vertices. Thus, HomI (Γ, Λ) is the subcomplex of
Hom(Γ \ I, Λ) consisting of the cells φ that can be extended to Γ. The
projection from Hom(Γ, Λ) to HomI (Γ, Λ) is a homotopy equivalence [6] since
the fibers are contractible. It turns out that Hom(Γ, Kn ) is homeomorphic
(but not via the aforementioned projection) to HomI (Γ, Kn ) when ind(Γ) is
a PL sphere [18]. For HomI (Γ, Λ) to inherit the G–action (induced from an
action on Γ), the set I needs to be (setwise) G–invariant. When G is the group
of order two generated by the reflection on C5 switching the vertex i with
the vertex 6 − i for i ∈ {1, 2, 3, 4, 5}, there are two G–invariant independent
subsets: {3} and {2, 4}. In [18] Schultz uses I = {2, 4}; we use I = {3}.
We need HomI (C5 , Kn ) and Hom(C5 , Kn ) to be equivariantly homeomorphic
which we get from an equivariant version of a lemma of Schultz [18].
While Hom(C5 , Kn ) (or Hom{3} (C5 , Kn ), which is equivariantly homeo-
morphic to it) is the star of this story, the hero is Hom{3} (P4 , Kn ), where
P4 is the path with four edges from 1 to 5 (the subgraph of C5 missing
the edge {1, 5}). Our story also features P4 \ {3}, which is the disjoint
union of two copies of K2 , namely the edges {1, 2} and {4, 5}. The graph
morphism P4 \ {3} → K2 sending 1 and 5 to + and 2 and 4 to − in-
duces the diagonal embedding of (the sphere) Hom(K2 , Kn ) in Hom(P4 \
{3}, Kn ) = Hom(K2 , Kn ) × Hom(K2 , Kn ). We also have the restriction map
Hom{3} (C5 , Kn ) → Hom{3} (P4 , Kn ) induced from the inclusion of P4 in C5
and the inclusion of Hom{3} (P4 , Kn ) into Hom(P4 \ {3}, Kn ). All these maps
are equivariant with respect to the involutions induced from the aforemen-
tioned i 7→ 6 − i. (The action on Hom(K2 , Kn ) is trivial since it maps
homeomorphically to the diagonal in Hom(P4 \ {3}, Kn ), the fixed point set
of the involution.)
Note that Hom{3} (P4 , Kn ) is not homeomorphic to Hom(P4 , Kn ): We will
show that Hom{3} (P4 , Kn ) is a (2n − 4)–dimensional manifold with boundary
while Hom(P4 , Kn ) is not a manifold, having maximal cells of every dimension
from 2n − 4 to 3n − 6. (Recall that n ≥ 3.)
The face posets of these Lovász complexes above are actually very easy
to describe concretely; all are made up of (ordered) quadruples of nonempty
subsets A, B, C, D of {1, . . . , n} satisfying further conditions. For any cell φ,
A, B, C, D are φ(1), φ(2), φ(5), and φ(4) respectively. In Hom(P4 \ {3}, Kn )
we have A ∩ B = ∅ = C ∩ D. In Hom{3} (P4 , Kn ) we also have that B ∪ D 6=
{1, . . . , n}, and Hom{3} (C5 , Kn ) has the further restriction that A ∩ C = ∅.
The diagonal Hom(K2 , Kn ) has A = C and B = D (in addition to A∩B = ∅).

5
We will denote these cells as
 
A B
φ=
C D

to remind us of their locations in terms of the vertices of the pentagon C5 .

Such an array should not be thought of as a 2 × 2 matrix.
On the topological side we have the involution on S n−2 × S n−2 switching
the two coordinates, with the diagonal S n−2 as the fixed set. An equivariant
regular neighborhood of the diagonal is N := {(x, y) ∈ S n−2 × S n−2 | x · y ≥
0}, which is a manifold with boundary Vn−1,2 = {(x, y) ∈ S n−2 ×S n−2 | x·y =
0}. We prove in Section 5 that Hom{3} (P4 , Kn ) is a manifold with boundary
Hom{3} (C5 , Kn ) and also an equivariant regular neighborhood of the diagonal
in Hom(P4 \ {3}, Kn ) ∼= Hom(K2 , Kn ) × Hom(K2 , Kn ). From the equivariant
PL theory of collapsing, shelling, and regular neighborhoods developed in
Section 4, our main result follows:

Main Theorem. The regular cell complex Hom{3} (P4 , Kn ) is a PL manifold

with boundary Hom{3} (C5 , Kn ) and is equivariantly homeomorphic (with re-
spect to the involution described above) to N := {(x, y) ∈ S n−2 × S n−2 | x ·
y ≥ 0}, where the involution on N interchanges (x, y) with (y, x). The
Stiefel manifold Vn−1,2 = ∂N is therefore equivariantly homeomorphic to
Hom{3} (C5 , Kn ), which is equivariantly homeomorphic to Hom(C5 , Kn ).

The structure of the rest of the paper is as follows: We establish notation

and give basic definitions and prove that Hom(C5 , Kn ) and Hom{3} (C5 , Kn )
are equivariantly homeomorphic in Section 2. The small amount of discrete
Morse theory that we use is in Section 3. Section 4 is devoted to the stat-
ing the equivariant versions of facts we need from piecewise-linear topology
given in [17]. Finally, in Section 5 we specialize to our setup and prove the
equivariant Csorba conjecture.

2. Notation and Basics

All simplicial complexes and posets we consider are finite. An (abstract)
simplicial complex K on a vertex set V is a collection of (finite) subsets
of V such that if σ ⊆ τ ∈ K, then σ ∈ K. The dimension of the simplex σ is
one less than its cardinality, and the n–skeleton of the simplicial complex
K (denoted K n ) is the sub-simplicial complex made up of all simplices of K

6
 0
with dimensionS at most n. However, we will abuse the notation and use K
to also mean {σ | σ ∈ K}. The nonempty simplices σ ∈ K are also called
the faces of K. The geometric realization of K is the topological space
X X
|K| := { tv δv ∈ RV | tv = 1, tv > 0, σ ∈ K \ {∅}}
v∈σ v∈σ

where δv is the standard basis vector of RV corresponding to v ∈ V . The

underlying topological space of a face σ of K is
X X
|σ| := { tv δv ∈ RV | tv = 1, tv ≥ 0}.
v∈σ v∈σ

Note that we are using vertical bars to denote both the cardinality of a finite
set and the geometric realization of a simplex or a simplicial complex; which
one is meant should be clear from the context.
If σ is a simplex in a simplicial complex K, its link is defined lnkK (σ) :=
{τ ∈ K | σ ∩ τ = ∅, σ ∪ τ ∈ K}. If K and L are two simplicial complexes, we
define their join K ∗ L := {σ ∪ τ | σ ∈ K, τ ∈ L}.
A subcomplex L of a simplicial complex K is called full if it satisfies the
property that if σ ∈ K and σ ⊆ L0 , then σ ∈ L.
A regular cell structure on a (compact Hausdorff) topological space
X is a (finite) collection {c} of subspaces (called cells or faces), each home-
omorphic to a (closed) disk of dimension d for some d, such that (1) X is
the disjoint union of the relative interiors of its cells (called the open cells
and denoted by int c), and (2) the boundary of each cell is a union of (lower
dimensional) cells. A topological space with a regular cell structure is a regu-
lar cell complex. For any simplicial complex K, the collection {|σ|}σ∈K\{∅}
gives a regular cell structure on |K|.
For any poset P , its order complex ∆P is the simplicial complex whose
simplices are chains a0 < a1 < . . . < ad , with ai ∈ P . The faces of a
regular cell complex under inclusion form a poset whose order complex is its
barycentric subdivision ([5] p200, [15] Ch 3 §1). Note that this notation is
consistent with ∆A being the full simplex on a vertex set A if the set A is
endowed with any total order. Also note that an abstract simplicial complex
K is itself a poset and includes the empty simplex (unlike its face poset), so
the order complex ∆K is the cone of the barycentric subdivision of K.
A G–simplicial complex, where G is a (finite) group, is a simplicial
complex K equipped with a permutation action of G on the vertex set V

7
so that the induced action on the subsets of V sends simplices to simplices.
Similarly, in a G–cell complex, the group action permutes the cells. For
any cell (or simplex) c, Gc := {g ∈ G | gc = c} is called its stabilizer.
It will be convenient for us to define a G–regular cell complex to be a
topological space with a G–action and a regular cell structure with the group
G permuting the cells so that every closed cell is Gc –homeomorphic to a cone
on its boundary with the apex fixed by Gc . The geometric realization of a
G–simplicial complex K is a G–regular cell complex with cells {|σ|}σ∈K\{∅}
since the stabilizer of a simplex always fixes its centroid. Similarly, if the
cells of a regular cell complex are convex with G acting affinely on each one,
the complex is G–regular. If the stabilizer of each cell (or simplex) fixes the
cell pointwise, the complex is called admissible. If G acts on a poset P
(preserving the order), then ∆P is an admissible G–simplicial complex.
Just as in the non-equivariant case ([5] p200, [15] Ch 3 §1), the face poset
of a G–regular cell complex determines its G–homeomorphism type:

Lemma 2.1. If X is a G-regular cell complex with face poset F, then X is

G–homeomorphic to |∆F |.

The proof below is very similar to the non-equivariant case and is mainly
given for completeness.
Proof. The proof is by induction on the number of cells in X. There is
nothing to prove if X consists
S of a single 0–cell. Now, choose a maximal cell
c and define Y := X \ g∈G int gc with the induced G–cell structure. By
the induction hypothesis, Y ≈G |∆(F \ Gc)|. Also, ∂c ≈Gc |∆F<c |. Then
we take the cone of ∂c with its apex being a point x ∈ int c fixed by Gc .
Extending this coning equivariantly to G∂c gives the homeomorphism from
|∆F | to X.
We are specifically interested in the G–equivariant homeomorphism type
of some Lovász multimorphism complexes Hom(Γ, Kn ), for a graph Γ with a
group of symmetries G. The following equivariant version of Lemma 3.5 in
[18] enables us, in some cases, to work with the smaller and more convenient
G–regular cell complexes HomI (Γ, Kn ), where I is a G–invariant independent
subset of vertices of Γ.

Lemma 2.2. Let Γ be a graph with a G-action, n ≥ 1, and I a G–invariant,

independent subset of the vertex set VΓ . For all v ∈ VΓ , define

8
 Av := {J ∈ ind(Γ) | v ∈ J}, Bv := {J \ I | J ∈ Av }
If there is a G–homeomorphism h : |∆ ind(Γ)| → |∆ ind(Γ \ I)| such that
h(|∆Av |) = |∆Bv | for all v ∈ VΓ , then Hom(Γ, Kn ) is G–homeomorphic to
HomI (Γ, Kn ).

Proof. Following [18], we consider the equivariant poset embedding

n
Y
f : Hom(Γ, Kn ) → ind(Γ) = ind(Γ){1,...,n}
i=1

given by φ 7→ (φ−1 (i))i . Then Hom(Γ, Kn ) ≈G |∆im f |. The poset ind(Γ){1,...,n}

can naturally be identified with those relations φ ⊆ VΓ × {1, . . . , n} that
are multimorphisms from the induced subgraph on the vertices with φ(v)
nonempty to the complete graph Kn . The additional condition that no φ(v)
can be empty yields the following description:
n Y
n 
\ [ Av , i=j
im f =
6 j
ind(Γ), i =
v∈VΓ j=1 i=1

All the Av satisfy the condition that if x ∈ Av and x ≤ y, then y ∈ Av .

Therefore, taking the order complex commutes with unions, and we obtain
that n Yn 
\ [ |∆Av |, i=j
Hom(Γ, Kn ) ≈G
|∆ ind(Γ)|, i 6= j
v∈VΓ j=1 i=1

We use a similar argument for HomI (Γ, Kn ). The image of its embedding in
ind(Γ \ I){1,...,n} has the additional condition that for each vertex in I, there
is some element of {1, . . . , n} that is not related to any of its neighbors in Γ.
We have that, for v ∈ / I, Bv satisfies the same condition as Av above. For
v ∈ I, Bv also satisfies the condition that if x ∈ Bv and y ≤ x, then y ∈ Bv .
Hence,
n Y n 
\ [ |∆Bv |, i=j
HomI (Γ, Kn ) ≈G
|∆ ind(Γ \ I)|, i 6= j
v∈VΓ j=1 i=1

Thus, using the G–homeomorphism from the hypothesis on each coordinate

in the product, we obtain that Hom(Γ, Kn ) ≈G HomI (Γ, Kn ).

9
 {3}
A3
{1,3} {3,5}

{1} {5} {1} {5}

A1
0/ B3 0/
A5
B1
B5
{1,4} {2,5} {1,4} {2,5}
B4 B2
A4
{4} {2,4} A2 {2} {4} {2,4} {2}

Figure 1: |∆ ind(C5 )| and |∆ ind(C5 \ {3})|

In our case, we choose I = {3} as our independent set in C5 . The

existence of a G = {±1}–homeomorphism h : |∆ ind(C5 )| → |∆ ind(C5 \{3})|
satisfying the necessary conditions is easy to see from Figure 1. Thus, we
have Hom(C5 , Kn ) is equivariantly homeomorphic to Hom{3} (C5 , Kn ).

3. Discrete Morse Theory

We will use Robin Forman’s Discrete Morse Theory to find collapsings of
simplicial complexes. More thorough discussions of the subject can be found
in [8] and [12].
Let K be a finite (abstract) simplicial complex. We use the notation σ lτ
if σ < τ and dim σ = dim τ − 1. By a vector, we mean a pair (σ l τ ), where
τ is thought of as the head of the vector and σ the tail. A discrete vector
field on K is defined to be a collection of vectors V = { (σi l τi ) | i ∈ I} such
that each simplex ρ ∈ K belongs to at most one element of V , either as a
head or a tail of a vector.
Given a discrete vector field V on K, we have the notion of a path, which
is a sequence of simplices in K of the form:

σ0 l τ0 m σ1 l τ1 m . . . m σs−1 l τs−1 m σs

where ∀ i : 0 ≤ i < s, (σi l τi ) ∈ V and σi 6= σi+1 . We say a path as above

has length s. By a cycle we mean a path as above with σs = σ0 .
A Morse matching (or a discrete gradient field) is a discrete vector
field V with no cycles. The simplices which are unpaired in V are called
critical.

10
 An equivalent concept to a Morse matching is a height function on K. A
height (or Morse) function is a map h : K → R satisfying ∀ σ ∈ K,

|{ρ l σ | h(ρ) ≥ h(σ)} ∪ {τ m σ | h(τ ) ≤ h(σ)}| ≤ 1

Given a height function h, the corresponding Morse matching is the collection

of pairs (σ l τ ) for which h(σ) ≥ h(τ ). Conversely, it is not difficult to
construct a height function inducing a given Morse matching ([8], [12]). This
height function is clearly not unique. In fact, we may adjust it to be one-to-
one and to take values in N (without changing the Morse matching).
The following is the key lemma from discrete Morse theory we will use in
this paper. A more general version for cell complexes can be found in [12],
Theorem 11.13.
Lemma 3.1. Let K be a finite simplicial complex with a Morse matching
whose critical simplices form a subcomplex L. Then K collapses simplicially
to L.
Proof. Let h : K → N be a one-to-one height function corresponding to the
given Morse matching. Define a new height function h̃ : K → N as follows:
For σ ∈ L, set h̃(σ) = dim (σ). For σ ∈/ L, set h̃(σ) = h(σ) + dim(L). Under
this new height function, all of the simplices in L remain critical, and the
relative heights of all the simplices outside of L are unchanged, preserving
their pairings. Therefore, h̃ corresponds to the same Morse matching as h,
and h̃ is one-to-one on K \ L.
Now, for m ∈ N define

K(m) := {σ ∈ K | ∃ τ ≥ σ such that h̃(τ ) ≤ m}

Note that K(dim(L)) = L. For m ≥ dim(L), either K(m + 1) = K(m) or

K(m + 1) = K(m) ∪ {ρ, τ } where ρ l τ and h̃(τ ) = m + 1 < h̃(ρ). In the
latter case, K(m + 1) collapses to K(m) along the free face ρ. Since K =
K(max{h̃(σ) | σ ∈ K}), K collapses to L via a sequence of these collapsings.

4. Equivariant Neighborhoods
In this section, we discuss the equivariant theory of G–regular neigh-
borhoods with suitable hypotheses, similar to Rourke and Sanderson’s non-
equivariant version [17]. Throughout, we assume G is a finite group, all

11
simplicial complexes are finite, and maps between polyhedra are piecewise-
linear. Also, when there is a G–action on a space X, the product X × I is
assumed to have the G–action g(x, t) = (gx, t).
A simplicial complex K polyhedrally triangulates a space X ⊂ Rn
if the 0–skeleton K 0 is identified with a subset of X so that the canonical
piecewise-linear function from |K| to Rn mapping |σ| of each simplex σ to
the convex hull of its vertices in X is a homeomorphism onto X. Then
X is called a polyhedron, and two polyhedra are equivalent if there is a
piecewise-linear homeomorphism between them. If K is a G–complex, the
ambient vector space Rn into which the polyhedron X is G–embedded is the
underlying space of a (real) linear representation of the group G, and the
piecewise-linear homeomorphism from |K| to X is a G–map, then we have
a polyhedral G–triangulation of a G–polyhedron X. We implicitly
identify each simplex σ of K with the image of its geometric realization in
X. Note that one abstract (G–)simplicial complex may triangulate X in
multiple ways by having different choices for some of the vertices in X, in
which case we will give the simplicial complex different names. We will only
consider polyhedral triangulations.
Let Y ⊂ X be polyhedra and L be a subcomplex of K such that (K, L)
triangulates (N, Y ) where N is a neighborhood of Y in X. Define the sim-
plicial neighborhood of L in K, NK (L) := { σ ∈ K | ∃ τ ∈ K s.t. σ <
τ and τ ∩ L0 6= ∅}. Also, define ṄK (L) := { σ ∈ NK (L) | σ ∩ L0 = ∅}. The
derived subdivision K 0 of K near L is (combinatorially) the simplicial
complex with vertex set K 0 ∪ {vτ | τ ∈ K \ L, τ ∩ L0 6= ∅}, and the simplices
are of the form σ ∪ {vτ1 , . . . vτm } where σ ∈ L or σ ∈ K with σ ∩ L0 = ∅ and
σ < τ1 < . . . < τm . Geometrically, K 0 is realized by selecting the new vertex
vτ in the interior of each simplex τ of K \ L that intersects a simplex of L
and then, in ascending order of dimension, replacing each τ with the cone
(with apex vτ ) on its boundary (which has already been subdivided in the
previous steps).
Now suppose L is full in K (i.e., if a set of vertices in L forms a simplex
in K, then they form a simplex in L) and |ṄK (L)| = ∂|NK (L)| in X. Then
N1 = |NK 0 (L)| is called a regular neighborhood of Y in X. If K and L are
both admissible G–complexes and, when defining K 0 , the set of new vertices
{vτ } is chosen to be G–invariant, we say N1 is a G–regular neighborhood.

Proposition 4.1. If K is a G–simplicial complex, |K| × [0, 1] has an admis-

sible G–triangulation with no new vertices in |K| × (0, 1).

12
Proof. Let F be the face poset of K. Then the order complex ∆(F ×{0, 1}) of
the product poset F ×{0, 1} satisfies |∆(F ×{0, 1})| ≈G |∆F |×|∆{0, 1}| ≈G
|K| × I. Hence ∆(F × {0, 1}) gives the desired G–triangulation.
Let K = {K1 , K2 , . . . , Ks } be a collection of simplicial complexes, each
one triangulating the polyhedron X. For φ ∈ K1 × . . . × Ks , let |φ| :=
T
1≤i≤s |φi |. Define an equivalence relation on K1 × . . . × Ks by φ ∼ ψ ⇔
|φ| = |ψ|. We define a poset CK := {φ ∈ K1 × . . . × Ks | |φ| 6= ∅}/ ∼ with
the partial order [φ] ≤ [ψ] ⇔ |φ| ⊆ |ψ|. We will identify an equivalence class
[φ] with the geometric realization |φ| of its representatives.

Proposition 4.2. Given a collection K = {K1 , . . . , Ks } of simplicial com-

plexes that triangulate a polyhedron X, then CK is a regular cell structure
on X, so that |∆CK | is a triangulation of X and a common subdivision of
K1 , . . . , K s .

Proof. We Tshow first that the openT cells of CK are disjoint. In this discussion,
let |φ| = 1≤i≤s |σTi | and |ψ| = 1≤i≤s T |τi |. Note that if |φ| ∩ |ψ| =
6 ∅, we have
that |φ| ∩ |ψ| = 1≤i≤s (|σi | ∩ |τi |) = 1≤i≤s |σi ∩ τi |. Thus, the intersection
of any two closed cells is a closed cell. If [φ] < [ψ], then, without changing
[ψ] or [φ], we may replace each τi with its minimal face containing |ψ| and
each σi with σi ∩ τi . Then |ψ| ∩ int(|τi |) is always nonempty, and for some
i we have |φ| ⊆ |∂τi |. Therefore we have that |φ| ⊆ ∂|ψ|, implying that the
intersection of any two distinct closed cells must occur on the boundary of
at least one of them, so any two distinct open cells are disjoint.
Each |ψ| is a (nonempty) compact, convexSpolytope, yielding that |ψ| ≈
Dm for some m ≥ 0. Furthermore, ∂|ψ| = [φ]<[ψ] |φ|: If x ∈ ∂|ψ| ⊆ X,
there is a unique σi ∈ Ki such that x ∈ int(|σi |), giving some φ with x ∈ |φ|
and [φ] ≤ [ψ]. However [φ] < [ψ] because x is not in int(|ψ|), S so there is an
i with x not in int(|τi |), i.e., σi 6= τi . Conversely, if x ∈ [φ]<[ψ] |φ|, we have
x ∈ |φ| ⊆ ∂|ψ| as above. This proves that CK is a regular cell structure on
X. Therefore, we have that |∆CK | ≈ X.

Corollary 4.3. If X is a G–polyhedron and K is a nonequivariant trian-

gulation of X, then there is an admissible G–triangulation of X which is a
subdivision of K.

Proof. For each g ∈ G, gK is another triangulation of X because G is acting

linearly on the ambient representation space. Let K = {gK | g ∈ G}. Using

13
the notationTfrom the proof of 4.2, each cell [φ] is given by a map φ : G → K.
Then |φ| = g∈G g|φ(g)| ⊆ X. For h ∈ G, define (hφ)(g) := φ(h−1 g). This
induces anTorder-preserving G–action on CK because, for any φ and any h in
−1
T T
G, |hφ| = g∈G g|φ(h g)| = g∈G hg|φ(g)| = h g∈G g|φ(g)| = h|φ|. Lastly,
CK is a G–regular cell complex because the average of the vertices of any cell
is fixed by the cell’s stabilizer. The result follows by Lemma 2.1.

Lemma 4.4. Let (K, L) be an admissible G–triangulation of (X, Y) with L

full in K and let (K1 , L1 ) be a G–subdivision of (K, L). Then ∃ G–derived
subdivisions K 0 and K10 of K and K1 near L and L1 such that |NK 0 (L)| =
|NK10 (L1 )|.

Proof. We follow the proof of the non-equivariant version, Lemma 3.7 in

[17], and clarify some details with the combinatorial definition of derived
subdivisions. We define a map fL,K = f : X → [0, 1] by setting f (v) = 0 if
v ∈ L0 and f (v) = 1 if v ∈ K 0 \L0 gives a map on the vertices of K which can
be linearly extended on simplices. Choose  small enough so that no vertex
of K1 \ L1 is contained in f −1 [0, ]. Then choose G–derived subdivisions K 0
and K10 of K and K1 near L and L1 respectively with all the new vertices vτ
lying in f −1 (). We can choose these vertices equivariantly because f is G–
invariant and K is admissible. Now we will show that |NK 0 (L)| = f −1 [0, ] =
|NK10 (L1 )|.
The map f takes values of 0 or  on all of the vertices of NK 0 (L) and
NK10 (L1 ), so both of these neighborhoods are contained in f −1 [0, ]. Now let
x be a point in f −1 [0, ] ⊂ |K 0 | = |K10 |; say x is in the interior of the simplex
σ ∪ {vτ1 , . . . , vτk } of K 0 (respectively K10 ) as in the combinatorial definition
of a derived subdivision. Then x = s0 v0 + . . . + sk vk + t1 vτ1 + . . . + tm vτm
where σ = {v0 , . . . , vk } and s0 + . . . + sk + t1 + . . . + tm = 1. Suppose σ
is not in L (resp. L1 ). Then f (vi ) = 1 (resp. f (vi ) > ) for i = 0, . . . , k.
Meanwhile, f (vτj ) =  for j = 1, . . . , m. We have then, in both cases, that
f (x) > (s0 + . . . + sk ) + (t1 + . . . + tm ) = . This is a contradiction, and σ
must be in L (resp. L1 ), yielding that x is in |NK 0 (L)| (resp. |NK10 (L1 )|).

Lemma 4.5. Let v be a vertex of a G–simplicial complex K. If K 0 is a

derived G–subdivision of K near v, then | lnkK (v)| is Gv -homeomorphic to
| lnkK 0 (v)|.
−1
Proof. Assume that the derived vertices are chosen in fv,K () for some  ∈
−1
(0, 1). Then a point in | lnkK 0 (v)| = f () is of the form u + (1 − )v where

14
u ∈ | lnkK (v)|. Mapping this point to u gives the desired homeomorphism.

Theorem 4.6. If N1 and N2 are G–regular neighborhoods of Y in X, then

there exists a G–homeomorphism h : X → X that maps N1 to N2 and is the
identity on Y.
Proof. The proof mirrors that of the non-equivariant version, Theorem 3.8
in [17]. We have Ni = |NKi0 (Li )| for i = 1, 2 as above, where Ki are
G–triangulations of a neighborhood of Y in X. As in [17], K1 and K2
may be subdivided and extended to triangulations of |K1 | ∪ |K2 |. Apply-
ing 4.2, there is a triangulation K0 of |K1 | ∪ |K2 | containing subdivisions
of K1 and K2 and inducing L0 , a common subdivision of L1 and L2 . By
4.3, K0 may be assumed to be a G–triangulation. By 4.4, for i = 1, 2,
we can find G–derived subdivisions K̃i and K0i of Ki and K0 near Li and
L0 respectively such that |NK̃i (Li )| = |NK0i (L0 )|. Therefore we have that
N1 ≈G |NK̃1 (L1 )| = |NK01 (L0 )| ≈G |NK02 (L0 )| = |NK̃2 (L2 )| ≈G N2 where
each G–homeomorphism is induced by a G–isomorphism of G–derived sub-
divisions. Such an isomorphism is given by changing the placement of each
derived vertex vτ within the interior of a simplex τ touching Y and fixing
the placement of every other vertex in the subdivision, so the corresponding
G–homeomorphism is the identity on Y itself. Thus, the same is true of the
composition of these G–homeomorphisms.
Let Y ⊂ X be G–polyhedra. A G–collar on Y in X is a PL G–embedding
c : Y ×I →G X such that c(y, 0) = y and c(Y ×[0, 1)) is an open neighborhood
of Y in X. Suppose that for every a ∈ Y there are (closed, polyhedral)
neighborhoods U and V of a in X and Y respectively with U ∩ Y = V ,
such that for any g ∈ G, gU ∩ U 6= ∅ implies that ga = a and gU = U .
Suppose further that U ≈Ga V × I with v 7→ (v, 0) on V . Then we say that
Y is locally G–collarable in X and we have that GU ≈G GV × I. Local
G–collarability is equivalent to G–collarability.
Theorem 4.7. Suppose Y ⊂ X is locally G–collarable. Then there is a
G–collar on Y in X.
Proof. The proof is essentially the same as that of the non-equivariant ver-
sion, Theorem 2.25 in [17], substituting local G–collars for local collars.
Construct a new G–polyhedron Z := X ∪ Y × [−1, 0] by attaching a G–
collar to Y outside of X, identifying Y ⊂ X with Y × {0}. We will construct

15
 Va x {1}
i

Y x {0}

hi
Y x [-1,0]

Y x {-1}

Va x {-1}
i

Figure 2: Construction of hi

ols
To
a G–homeomorphism between X and Z which carries Y to Y × {−1}. Then

d
an
the preimage of Y × [−1, 0] will be a G–collar on Y in X.
er
For each a ∈ Y , let GVa × I be a local G–collar at a. Using compactness,
rit
W

cover Y with the interiors of finitely many GVa1 , . . . , GVak . Then for each
ee
Fr

i, we will define a G–homeomorphism hi : Z → Z which maps the interior

ith

of Vai × {0} into the interior of Vai × [−1, 0] and is the identity outside
rw

of GVai × [−1, 1]. To do this, we consider a G–triangulation Kai of GVai .

ito
Ed

Taking the product of this triangulation with an interval defines a regular

DF

G–cell structure on GVai × [−1, 1]. We now equivariantly subdivide this cell
ll P

complex by its order complex: We must choose a new vertex vc in the interior
Fi
PD

of each cell c. First, equivariantly select a point yσ in the interior of each

simplex σ ∈ Kai . For the cell |σ| × {1}, choose vc = (yσ , 1). Likewise, for
the cell |σ| × {−1}, choose vc = (yσ , −1). Finally, for the cell |σ| × [−1, 1],
choose vc = (yσ , 0).
Define a different subdivision simply by moving vc to (yσ , − 21 ) when c
intersects the relative interior of GVai × [0, 1] and making no change in the
placement for vc otherwise. Then the G–homeomorphism hi is given by
mapping the first subdivision to the second, since they are both realizations
of the same order complex. Note that hi is the identity except on the relative
interior of GVai × [−1, 1] where hi (y, t) = (y, s) with s < t. Thus, hi maps
all of the relative interior of GVai × {0} into GVai × (−1, 0).
Now define h to be the composition hk ◦ . . . ◦ h1 . Since the interiors of the
local G–collars cover all of Y , h maps all of Y × {0} into Y × (−1, 0). Let K
G–triangulate Y , and thus also h(Y × {0}). Then we consider h(X) ∩ [−1, 0].

16
It has a regular cellular G–structure with a face poset isomorphic to that
of |K| × [−1, 0]. The cells in the former come in three types: simplices
of K triangulating Y × {0} (which correspond to the same in the latter
complex), simplices of K triangulating h(Y × {0}) (which correspond to
simplices of K triangulating Y × {−1}), and the intersection of h(X) with
cells |σ| × [−1, 0] for σ ∈ K (which correspond to the cells |σ| × [−1, 0]).
Therefore, h(X) ∩ [−1, 0] is G–homeomorphic to Y × [−1, 0], fixing Y × {0}.
We extend this homeomorphism by the identity to the rest of X to get the
desired G–homeomorphism h̃ : X → Z carrying Y to Y × {−1}.

Theorem 4.8. If Y ⊂ X is locally G–collarable then a G–regular neighbor-

hood of Y in X is a G–collar.

Proof. We have by 4.7 that Y is G–collarable. Let L = L×0 be an admissible

G–triangulation of Y . Let K be the G–triangulation of the G–collar Y × I as
in 4.1. Now choose a G–derived subdivision K 0 of K near L such that all of
the new vertices have second coordinate 21 . Then |N (L, K 0 )| = Y × [0, 12 ] ≈G
Y × I. The result now follows from 4.6.
We will also make use of the notion of bicollarability. We say Y ⊂ X is
G–bicollarable in X if there exists a G–embedding of Y ×[−1, 1] into X with
(y, 0) 7→ y for all y ∈ Y and Y × (−1, 1) mapping to an open neighborhood
of Y in X.

Theorem 4.9. If N = |NK 0 (L)| is a G–regular neighborhood of Y in X,

then |ṄK 0 (L)| is G–bicollarable in X.
−1
Proof. By 4.6, it suffices to consider the case N = fL,K [0, ] for some  ∈
(0, 1). That is, the derived vertices {vτ } of K 0 were chosen in f −1 (). Let
0 < 1 <  < 2 < 1. Equivariantly, choose alternate derived vertices {vτ1 }
and {vτ2 } in f −1 (1 ) and f −1 (2 ) respectively, giving derived G–subdivisions
K10 and K20 of K near L. Then there are the natural homeomorphisms
hi : |ṄK 0 (L)| →G |ṄKi0 (L)| given by sending each vτ to vτi . We now define a
G–bicollar C : |ṄK 0 (L)| × [−1, 1] →G cl(|NK20 (L)| \ |NK10 (L)|) by setting

|t|h1 (x) + (1 − |t|)x, −1 ≤ t ≤ 0
C(x, t) =
th2 (x) + (1 − t)x, 0≤t≤1

17
 Note that this G–bicollarability can alternatively be expressed as |ṄK 0 (L)|
being G–collarable in both N and in cl(X \ N ).
For the remainder of the section, we wish to consider G–regular neighbor-
hoods within manifolds. For that purpose, we need to define a G–manifold.
We will consider G–polyhedra and G–complexes that are manifolds and that
have particularly well-behaved G–actions.
First, consider an orthogonal representation ρ : G → On (R). We denote
by S(ρ) and D(ρ) the unit sphere and disk respectively in the correspond-
ing representation space. Further, denote by S+ (ρ) the hemisphere with
last coordinate nonnegative and similarly for D+ (ρ). These have unique
piecewise-linear structures coming from their smooth structures [9].
We now inductively define a combinatorial G–sphere. S 0 with a
G–action is a combinatorial 0–dimensional G–sphere. An admissible sim-
plicial G–complex K with |K| (PL) G–homeomorphic to S(ρ) for some
ρ : G → On+1 (R) is an n–dimensional combinatorial G–sphere if for every
v ∈ K 0 , lnkK (v) is an (n − 1)–dimensional combinatorial Gv –sphere, itself
Gv –homeomorphic to S(Rv ⊥ ), where Rv ⊥ is the orthogonal complement in
ρ|Gv of the trivial representation Rv.
Similarly, we may define a combinatorial G–hemisphere by substitut-
ing S+ (ρ) and allowing links of vertices to be n–dimensional G–spheres or
G–hemispheres in the above definition. Finally, a combinatorial G–disk is
simply the cone on a G–sphere or G–hemisphere with a G–fixed point.
An admissible simplicial G–complex K is an n–dimensional combinato-
rial G–manifold if for every v ∈ K 0 , lnkK (v) is an (n−1)–dimensional com-
binatorial Gv –sphere or hemisphere. A G–polyhedron M is an n–dimensional
(PL) G–manifold (with boundary) if its admissible G–triangulations are
n–dimensional combinatorial G–manifolds. (If it is true for one, it is true for
all.) The boundary ∂M is easily shown to be a G–submanifold.
Proposition 4.10. Let M be an n–dimensional G–manifold and M1 be an
n–dimensional G–invariant submanifold with cl(∂M1 ∩int M ) G–bicollarable
in M . Then M1 is a G–manifold.
Proof. Let K be a G–triangulation of M with subcomplexes (K1 , L) trian-
gulating (M1 , ∂M1 ). We need to show that the link of any vertex v ∈ K10 is
a Gv –sphere or hemisphere. We consider the case v ∈ cl(∂M1 ∩ int M ). The
link of any other vertex of K1 is the same in both K and K1 .
Since ∂M1 is G–bicollarable, we may consider a closed Gv –invariant neigh-
borhood | stL (v)| × [−1, 1] = Uv with (x, 0) identified with x for all x ∈

18
| stL (v)| and | stL1 (v)| × [0, 1] ⊂ M1 . Triangulate Uv in the following way:
First triangulate | lnkL (v)| × [−1, 1]. Add to it by coning | lnkL (v)| × {−1}
and | lnkL (v)| × {1} with (v, −1) and (v, 1) respectively. Let J be this trian-
gulation of | lnkL (v)| × [−1, 1] ∪ | stL (v)| × {−1, 1}. Then J ∗ {v} triangulates
| stL (v)| × [−1, 1].
Now subdivide K so that it contains a subdivision of J ∗{v} as a subcom-
plex. Choose a derived Gv –subdivision K 0 near v. From the construction in
the proof of 4.4, we may assume | st0K (v)| is an –neighborhood of v in |J ∗{v}|.
Thus, by 4.5, | lnkL (v)| × [−1, 1] ∪ | stL (v)| × {−1, 1} is Gv –homeomorphic to
| lnk0K (v)|, which we know is an (n − 1)–Gv –sphere or hemisphere since M is
a G–manifold.
Consider the point w = (v, 1). Its stabilizer in Gv is all of Gv . Hence,
lnkJ (w) is an (n−2)–Gv –sphere or hemisphere. But | lnkJ (w)| is Gv –homeomorphic
to | lnkL (v)|, giving us that |J| is the suspension of | lnkL (v)| by w and (v, −1),
and thus a Gv –sphere or hemisphere. In conclusion, | lnkL (v)| × [0, 1] ∪
| stL (v)|×{1}, which is Gv –homeomorphic to | lnkK1 (v)|, is a Gv –hemisphere.
(Note that when lnkL (v)| is a Gv –hemisphere, | lnkL (v)|×[0, 1]∪| stL (v)|×{1}
is actually a quadrant of a Gv –sphere where the two nonnegative coordinates
give trivial subrepresentations. This is easily seen to be Gv –homeomorphic
to a Gv –hemisphere.)
Now assume for the remainder of the section that Y ⊂ M is a polyhedron
and M is an n–G–manifold. The following is a direct result of 4.10, 4.9, and
the non-equivariant Proposition 3.10 from [17]

Proposition 4.11. A G–regular neighborhood N of Y in M is a G–manifold

with boundary. If Y ⊂ int M, ∂N = |ṄK 0 (L)| where L and K 0 are as in the
definition of regular neighborhood.

We omit the proofs of the following two results since they are exactly the
same as the non-equivariant versions (Theorem 3.11 and Corollary 3.18) in
[17], only utilizing G–bicollarability in place of collarability of boundaries of
manifolds.

Theorem 4.12 (Simplicial G–Neighborhood Theorem). Let Y be a G–polyhedron

in int M n and N be a G–neighborhood of Y in int M n . Then N is G–regular
if and only if

(i) N is an n–manifold with boundary ∂N G–bicollarable in M ,

19
 (ii) there are admissible G–triangulations (K, L, J) of (N, Y, ∂N ) with L
full in K, K = NK (L) and J = ṄK (L)
Corollary 4.13. Suppose N1 ⊂ int N2 are two G–regular neighborhoods of
Y in int M. Then cl(N2 \ N1 ) is a G–collar of Ṅ1 .
Let Y ⊂ X be polyhedra such that X = Y ∪ Dm and Y ∩ Dm = Dm−1
where Dm and Dm−1 are disks of dimensions m ≥ 1 and m − 1 respec-
tively and there is a PL homeomorphism Dm → Dm−1 × I with Dm−1
mapping by the identity to Dm−1 × {0}. Then we say there is an (m–
dimensional) elementary collapse of X onto Y . If there is a sequence
X = X0 , X1 , X2 , . . . , Xk = Y of polyhedra such that there is an elementary
collapse of Xi−1 onto Xi , we say X collapses to Y , or X & Y . In particular,
if X = |K| and Y = |L| with K collapsing simplicially to L, then X collapses
to Y . Note that the dimensions of the individual elementary collapses in a
sequence are allowed to vary. We say that a collapse is m–dimensional if
every elementary step has dimension ≤ m.
Consider now the case that G is a finite group and X and Y are G–
polyhedra with X = Y ∪ GDm where Dm (as well as gDm for each g ∈ G) is
a disk as in the definition of an elementary collapse. Suppose we also have
that: (1) gDm 6= Dm implies that gDm ∩ Dm ⊆ Y and (2) there exists a
point y ∈ Dm−1 fixed by the stabilizer GDm = H such that Dm−1 is H–
homeomorphic to a cone with apex y on some H–complex, and (3) Dm is
H–homeomorphic to Dm−1 × I, then we say there is an elementary G–
collapse from X to Y . (Note then that any point x ∈ {y} × (0, 1] will have
stabilizer Gx = H.) A sequence of these is called a G–collapse, denoted
X &G Y .
A simplicial collapse from admissible K to L gives rise to a G–collapse if
and only if whenever (σ l τ ) is in its corresponding Morse matching, so too
is (gσ l gτ ) ∀ g ∈ G.
An elementary collapse from an n-manifold M to another n-manifold M1
where M = M1 ∪ Dn and Dn−1 ⊂ ∂M1 and Dn \ Dn−1 ⊂ ∂M is called an
elementary shelling. A sequence of such collapses is a shelling. Note that
every elementary step must be n–dimensional.
The equivariant version of shelling requires some additional conditions.
Let M1 ⊂ M be n–manifolds with an elementary G–collapse from M =
M1 ∪ GDn to M1 such that: (1) Dn ∩ M1 = Dn−1 lies in a G–collarable
subpolyhedron W ⊂ ∂M1 , (2) under the GDn –triangulation K = {y} ∗ L of
Dn−1 , if gDn 6= Dn , then gDn ∩ Dn ⊂ |L| × {0}, and (3) y ∈ ∂Dn−1 implies

20
 M
|K| x {-1}
M1
n
D
|K'|

y = (y,0)
(y, 1/2)

n-1
D x {1} Dn-1x {1}

ols
Figure 3: Shelling homeomorphism

To
d
an
that, in the G–collar on W , ({y} ∗ ∂|L|) × I ⊂ ∂M . If these three extra
er
rit
W

conditions are satisfied, this elementary G–collapse is called an elementary

ee

G–shelling. A sequence of these is called a G–shelling.

Fr
ith

Lemma 4.14. If M G–shells to M1 , then there is a G–homeomorphism

rw
ito

h : M → M1 which is the identity outside an arbitrary neighborhood of

Ed

M \ M1 .
DF
ll P

Proof. As in the corresponding proof of Lemma 3.25 in [17], we need only

Fi

to consider the case of an elementary G–shelling. Let M = M1 ∪ GDn give

PD

the elementary G–shelling. Denote GDn by H. Let K = {y} ∗ L be the

H–triangulation of Dn−1 from the definition of elementary G–collapse.
Choose a G–collar on GDn−1 in M1 within the given neighborhood of
M \ M1 . We may consider the disk Dn−1 × [−1, 1] with Dn = Dn−1 × [−1, 0],
E n = Dn−1 × [0, 1] ⊂ M1 , and Dn−1 = Dn−1 × {0}. Then if Dn 6= gDn , we
have that Dn−1 × [−1, 1] may only intersect gDn−1 × [−1, 1] in |L| × [0, 1].
We will define an H–homeomorphism from Dn−1 × [−1, 1] to E n which is
the identity on |L| × [0, 1] ∪ |K| × {1}; such a homeomorphism can then
be extended, first equivariantly to all of G(Dn−1 × [−1, 1]) and then by the
identity to the rest of M . This last extension is possible because either
|L| × I = ∂Dn−1 × I or cl(∂Dn−1 ) \ |L|) × I = |{y} ∗ ∂L| × I ⊂ ∂M .
Let K 0 be a derived H–subdivision of K near y. We have that |K|×{−1}∪
|L| × (−1, 0) is H–homeomorphic to |K 0 | = |NK 0 (y)| ∪ |NK 0 (L)| because they

21
are both H–homeomorphic to |K| with an H–collar attached outside to |L|.
Therefore, we have an H–homeomorphism from Dn−1 ×{−1, 1}∪|L|×[−1, 1]
to Dn−1 × {0, 1} ∪ |L| × [0, 1]. Coning the two polyhedra with (y, 0) and (y, 21 )
gives the desired H–homeomorphism from E n ∪ Dn to E n .

Theorem 4.15. Suppose Y ⊆ X are G–polyhedra in a G–manifold M . If

X &G Y , then a G–regular neighborhood of X G–shells to a G–regular neigh-
borhood of Y in M .

Proof. We follow the proof of the non-equivariant version, Theorem 3.26 in

[17], checking that the conditions of G–shelling are satisfied. The proof uses
induction on the dimension of the collapse from X to Y .
Suppose that the theorem holds when the G–collapse is (m−1)–dimensional.
We now consider the case where there is an m–dimensional elementary G–
collapse from X to Y . Let X = Y ∪ GDm , with Y ∩ Dm = Dm−1 × {0}
where Dm ≈GDm Dm−1 × I. For simplicity, we will from now on denote the
subgroup GDm by H.
Let K be an admissible G–triangulation of M with full subcomplexes
L2 ≤ L1 triangulating Y and X respectively. Denote by J the subcom-
plex triangulating Z = Dm−1 × {1} ⊂ Dm , and by GJ, the resulting G–
triangulation of its G–orbit, GZ. Finally, let y be the apex in the GDm –cone
structure of Dm−1 . Note then that {y} × I is fixed pointwise by H, and any
point (y, t) with t > 0 has stabilizer exactly H. Let x = (y, 12 ).
As in [17], by breaking up the collapsing into smaller steps, we may
assume that there are no vertices of K 0 in Dm−1 × (0, 1).
Now we choose a derived G–subdivision K 0 of K near L2 ∪ GJ. Choose
the derived vertices for simplices in L1 \ (L2 ∪ GJ) in Gp−1 ( 21 ) ensuring
that x is one of them, and denote by L0 the new triangulation of X. Then
NK 0 (L0 ) gives a G–regular neighborhood of X, which is the union of NK 0 (L2 )
and NK 0 (GJ), G–regular neighborhoods of Y and GZ respectively. There
is an (m − 1)–dimensional H–collapse from |J| to (y, 1), so the induction
hypothesis and 4.14 together imply that |NK 0 (J)| is an n–dimensional H–
disk. Let E n = |NK 0 (J)|. As an H–disk with x an H–fixed point, E n is seen
to be H–homeomorphic to | stṄK 0 (J) (x)| × I.
We will show that if NK 0 (gJ) 6= NK 0 (J), the two subcomplexes must
be disjoint. For such a g, suppose there exists a vertex v = vτ ∈ NK 0 (gJ) ∩
NK 0 (J). (Note that it must be a derived vertex since gJ and J are themselves
disjoint.) Then τ ∈ K contains vertices u and w of gJ and J respectively.

22
 En

Dm

Y y = (y,0)

ols
Figure 4: Regular neighborhood shelling

To
d
an
er
Thus, ρ = {u, w} ∈ L1 since L1 is a full subcomplex of K, but |ρ| is not
rit
W

contained in Y and it is not contained in GDm since a simplex in Dm is may

ee

only contain vertices from L2 and J, not gJ. This contradicts X = Y ∪GDm ,
Fr
ith

so NK 0 (gJ) ∩ NK 0 (J) must be empty. Since we have shown that gE n 6= E n

rw

implies gE n ∩ E n = ∅, it must be true that GE n = H.

ito

We next prove that |NK 0 (J)| ∩ |NK 0 (L2 )| is an (n − 1)–disk E n−1 which
Ed
DF

is H–homeomorphic to | stṄK 0 (J) (x)|, giving that E n is H–homeomorphic to

ll P

E n−1 × I as required. To see this, we show that E n−1 is an H–regular

Fi
PD

neighborhood of Dm−1 × { 21 } in the (n − 1)–H–manifold |ṄK 0 (J)|, so that we

may again invoke the induction hypothesis for (m − 1)–dimensional collapses
and ?? (since Dm−1 × { 21 } H–collapses to x and an H–regular neighborhood
of x is the desired star of x).
Let P be the subcomplex of K 0 triangulating Dm−1 × { 12 } and let Q =
ṄK 0 (J) for brevity. The claim then is that NQ (P ) = NK 0 (L2 ) ∩ NK 0 (J).
Let σ ∈ NK 0 (L2 ) ∩ NK 0 (J), we easily see that σ cannot intersect L02 or
J 0 and must consist only of derived vertices of the form vρ . Then there
must exist u ∈ L02 and w ∈ J 0 such that σ ∪ {u} and σ ∪ {w} are both
simplices of K 0 . This implies that there exists vρ ∈ σ for some ρ ∈ K
containing both u and w. But then {u, w} ∈ L1 due to the fullness of L1 .
Thus, v{u,w} is in P and can be added to σ, so σ ∈ NQ (P ). Hence, we have
NK 0 (L2 ) ∩ NK 0 (J) ⊆ NQ (P ).

23
 For the other inclusion, if σ is in NQ (P ), it means that there is a vτ ∈ P 0
such that σ ∪ {vτ } is in Q for some τ which contains vertices from both L2
and J. We note again that σ consists only of derived vertices since it is in
Q = ṄK 0 (J), so let ρ be the minimal face such that vρ ∈ σ ∪{τ }. Then ρ ≤ τ ,
so we have that ρ ∈ L1 . Since ρ was subdivided, it must contain some vertex
u ∈ L02 . Therefore, u may be added to σ to get a simplex of K 0 intersecting
L02 , i.e., σ ∈ NK 0 (L2 ), and it is already in Q ⊂ NK 0 (J). This proves that
NK 0 (L2 ) ∩ NK 0 (J) = NQ (P ). This finishes the proof that E n−1 is an H–
regular neighborhood of Dm−1 × { 21 } in |Q| and therefore H–homeomorphic
to the (n − 1)–disk | stQ (x)| as explained.
Observe that GE n ∩ |NK 0 (L2 )| ⊂ |ṄK 0 (L2 )|, which is G–collarable in
|NK 0 (L2 )| by 4.9.
There is one remaining condition to check for this to be a G–shelling.
Write E n−1 = | stQ (x)|. Then we must verify that x ∈ ∂E n−1 implies
that within the G–collar on |ṄK 0 (L2 )| in |NK 0 (L2 )|, |{x} ∗ ∂ lnkQ (x)| × I ⊂
∂|NK 0 (L0 )|. It suffices for us to show that every simplex of {x} ∗ ∂ lnkQ (x)
lies on ∂M because the G–collar is given by moving derived vertices around
with simplices of K. Thus, if a simplex σ consisting only of derived vertices
lies on ∂M , then σ × I lies on ∂M , and hence also on ∂|NK 0 (L0 )|.
Since E n−1 is an H–regular neighborhood of x ∈ |Q|, if x ∈ ∂E n−1 , then
x ∈ ∂|Q| = |Q| ∩ ∂M . Thus, x ∈ ∂M , forcing it to also belong to ∂|NK 0 (L0 )|.
Likewise, any simplex of σ ∈ {x} ∗ ∂ lnkQ (x) containing x lies in ∂E n−1 but
not in ṄQ (x), forcing σ to be in ∂M and therefore ∂|NK 0 (L0 )|. Hence, we have
proven the final condition that this constitutes a G–shelling of a G–regular
neighborhood of X to a G–regular neighborhood of Y .

Corollary 4.16. If Y G–collapses to a point, then any G–regular neighbor-

hood of Y (in a G–manifold) is a G–disk.

Corollary 4.17. A collapsible G–manifold is a G–disk.

Corollary 4.18. If X ⊂ int M and X &G Y then a G–regular neighborhood

of X in M is a G–regular neighborhood of Y in M.

Proof. Let N1 and N2 be G–regular neighborhoods of X and Y respectively

in M . By 4.15, N1 G–shells to N2 , and so by 4.14, they are G–homeomorphic
fixing Y . The homeomorphism carries any G–triangulation of N2 to a G–
triangulation of N1 . Therefore, by 4.12, N1 is a G–regular neighborhood of
Y in M .

24
 Our final goal for this section is the following result that enables us to
recognize G–regular neighborhoods from G–collapses.

Theorem 4.19 (Collapsing Criterion for G–Regular Neighborhoods). Let N

be a G–neighborhood of Y in int M. Then N is G–regular if and only if

(i) N is an n–manifold with ∂N G–bicollarable in M ,

(ii) N &G Y.

Proof. The proof follows exactly the non-equivariant case (Corollary 3.30 in
[17]), but we include it here nonetheless because this theorem is a key tool
in proving our main result. For the first implication, suppose N = |K| is
regular with K an admissible simplicial neighborhood of L; then (i) follows
from 4.11 and 4.9. Recall the map f = fL,K from the proof of 4.4. Choose
 ∈ (0, 1) and choose a G–derived K 0 of K near L with all of the new vertices
lying in f −1 () for a given  ∈ (0, 1) to obtain N1 = |NK 0 (L)| = f −1 [0, ], a
G–regular neighborhood of Y in M . We have that N ≈G N1 , and we have a
G–cell structure on N1 whose cells are obtained by intersecting the interior
simplices of K with f −1 (0), f −1 (), or f −1 [0, ]. We may collapse, along with
its orbit, each cell σ ∩ f −1 [0, ], σ ∈ K \ ṄK (L) from its face σ ∩ f −1 () in
order of decreasing dimension. That this is a G–collapse follows from the
admissibility of K.
For the other implication, suppose we have N satisfying conditions (i) and
(ii). Let C = ∂N × [−1, 1] be a G–bicollar with ∂N = ∂N × {0}. Then let
N1 = N ∪ (∂N × [0, 12 ]), which constitutes a G–regular neighborhood of N in
M because we can triangulate it to be a simplicial neighborhood. Therefore,
by 4.18, since N &G Y , N1 is also a G–regular neighborhood of Y . But
we can define a G–homeomorphism on C fixing ∂N × {−1, 1} and carrying
∂N × { 21 } to ∂N × {0}. We can extend this by the identity to all of M ,
mapping N1 to N , showing that the latter is also a G–regular neighborhood
of Y in M .

5. Main Results
In this section, G is the group {±1} unless otherwise noted. Working
inside Hom(P4 \{3}, Kn ) (G–homeomorphic to the G–manifold S n−2 ×S n−2 ),
we show that Hom{3} (P4 , Kn ) is a G–regular neighborhood of the diagonal
Hom(K2 , Kn ) using the collapsing criterion, i.e., we show that it is a manifold

25
of the correct dimension and that it (simplicially) G–collapses to the diagonal.
Recall from Section 1 how we represent elements of the posets in question as
arrays whose entries A, B, C, D are nonempty subsets of {1, . . . , n}.
Define
 
A B
M := {φ = | A ∩ B = ∅, C ∩ D = ∅}
C D

K := {φ ∈ M | B ∪ D 6= {1, . . . , n}}
L := {φ ∈ K | A ∩ C = ∅}
S := {φ ∈ K | A = C, B = D}
We reiterate that M , K, and L are the face posets of the G–regular
cell complexes Hom(P4 \ {3}, Kn ), Hom{3} (P4 , Kn ), and Hom{3} (C5 , Kn ) re-
spectively, and S that of the diagonal Hom(K2 , Kn ). By passing to order
complexes, we obtain that ∆S and ∆L are full G–subcomplexes of ∆K,
which is a full G–subcomplex of ∆M , and they are all admissible. Our goal
is to show that |∆K| is a G–regular neighborhood of |∆S| whose boundary
is |∆L|.
Proposition 5.1. ∆K is a (2n − 4)–manifold with boundary ∆L.
Proof. We show that the link of an element of K is a sphere or a disk of
A B
dimension (2n−5). For any φ = ∈ K, lnk∆K (φ) = ∆K<φ ∗∆K>φ .
C D
For any φ ∈ K, we obtain an element of its lower link by deleting proper
subsets from each of A, B, C, and D, at least one of which is nonempty.
Therefore, K<φ is isomorphic to the face poset of ∂∆A ∗ ∂∆B ∗ ∂∆C ∗ ∂∆D,
yielding that ∆K<φ is a combinatorial sphere of dimension |A| + |B| + |C| +
|D| − 5. (Recall that, if A is an unordered set, ∆A is the full simplex having
A as its vertex set, whereas, if P is a poset, ∆P is its order complex.)
When φ ∈ K \ L, we show that ∆K>φ is a sphere of dimension 2n − |A| −
|B| − |C| − |D|
 − 01, yielding that lnk∆K (φ) is a sphere of dimension 2n − 5.
0

A B
For any φ0 = ∈ M such that φ0 > φ, we have that
C 0 D0

6 A ∩ C ⊆ A0 ∩ C 0 ⊆ (B 0 ∪ D0 )c
∅=

so that φ0 ∈ K. Thus, to obtain an element of the upper link of φ, any

element of (A ∪ B)c can be added to either A or B, but not to both, and

26
similarly for elements of (C ∪ D)c . As a consequence, we have that K>φ is
isomorphic to the face poset of ∗m 0 c
i=1 S where m = |(A ∪ B) | + |(C ∪ D) |,
c

and therefore ∆K>φ is a sphere of dimension 2n − |A| − |B| − |C| − |D| − 1

as claimed.
In the case where φ ∈ L, we claim that ∆K>φ is a disk of dimension
2n − |A| − |B| − |C| − |D| − 1, meaning that lnk∆K (φ) is a disk of dimension
2n − 5. This will finish the proof that ∆K is a manifold and ∆L is its
boundary. To see that K>φ is the face poset of a subcomplex of a join of
spheres, we consider the various types of elements that we can add to one or
more of A, B, C, and D to obtain a larger element of K.
1. An element of A ∩ D cannot be added anywhere (while remaining in
M ). The same is true for elements of B ∩ D and B ∩ C. Thus, these
elements contribute nothing to the upper link.
2. An element of B \ (C ∪ D) can be added to C or to D; doing so will
give us something in K, since the element was already in B ∪ D. Thus,
each of these elements contributes a copy of S 0 to the join of spheres.
Similarly, each element of D \ (A ∪ B) contributes a copy of S 0 to the
join.
3. An element of A \ D can be added to C or to D, contributing a copy of
S 0 = {±1} to the join with +1 indicating that the element was added
to C and −1 indicating D. Similarly, an element of C \ B can be added
to A (+1) or to B (−1). Adding elements of Type 3 to B or D could
produce something not in K.
4. An element of (A ∪ B ∪ C ∪ D)c can be added to A or B (but not both)
and, at the same time, to C or D (but not both). This contributes
a copy of S 1 = {±1} ∗ {±1} (treated as a single coordinate) to the
join of spheres with the +1’s corresponding to A and C and the −1’s
corresponding to B and D. As with Type 3, adding this type of element
to B or D could yield something not in K.
To ensure that we remain in K, there must be an element of (B ∪ D)c which
is not added to B ∪ D. In terms of coordinates, this means there must be at
least one coordinate corresponding to Type 3 or 4 above that has no −1’s.
Before proceeding, we define
Fk,l ⊆ (∗ki=1 S 1 ) ∗ (∗lj=1 {±1})
to be the subcomplex whose simplices have at least one coordinate from the
join with no −1’s. (Note that, as before, each copy of S 1 = {±1} ∗ {±1} is

27
regarded as a single coordinate.) We will prove a lemma (5.2) stating that
Fk,l is a disk of dimension 2k + l − 1. Assuming that result, since K>φ is
isomorphic to the face poset of (∗m 0 c
i=1 S ) ∗ Fk,l where k = |(A ∪ B ∪ C ∪ D) |,
l = |A \ D| + |C \ B|, and m = |B \ (C ∪ D)| + |D \ (A ∪ B)|, we have that
∆K>φ is a disk of dimension

m − 1 + 2k + l = |B \ (C ∪ D)| + |D \ (A ∪ B)|
+ 2|(A ∪ B ∪ C ∪ D)c | + |A \ D| + |C \ B| − 1
= 2n − 2|A| − 2|B| − 2|C| − 2|D|
+ 2|A ∩ D| + 2|B ∩ C| + 2|B ∩ D|
+ |B \ (C ∪ D)| + |D \ (A ∪ B)|
+ |A \ D| + |C \ B| − 1
= 2n − |A| − |B| − |C| − |D| − 1

as we had claimed.
Lemma 5.2. For k, l ∈ N such that 2k+l−1 ≥ 0, Fk,l is a disk of dimension
2k + l − 1.
Proof. We proceed by induction on the dimension, 2k + l − 1. In the initial
case, F0,1 has a single S 0 coordinate which must be +1, so it is a single point,
i.e. a disk of dimension 0. To prove Fk,l is a disk, we will show that it is a
(2k + l − 1)–manifold, show it collapses to a vertex, and then apply Corollary
4.17. There are four types of vertices whose links we need to consider:
1. +1 coming from one of the k S 1 coordinates has as its link Fk−1,l+1 , a
(2k + l − 2)–disk by induction.
2. −1 coming from one of the S 1 coordinates has as its link S 0 ∗ Fk−1,l , a
(2k + l − 2)–disk.
2k+l−1 0
3. +1 coming from one of the l {±1} coordinates has as its link ∗i=1 S ,
a (2k + l − 2)–sphere.
4. −1 coming from one of the {±1} coordinates has as its link Fk,l−1 , a
(2k + l − 2)–disk.
Now we will define a matching on Fk,l . First, we order the coordinates.
In each S 1 coordinate, we also choose one of the two copies of {±1} to be
distinguished. Associate each simplex in Fk,l with the simplex obtained by
inserting or removing +1 to or from the first coordinate lacking a −1 (in
the distinguished copy of {±1} in the case the first such coordinate is S 1 ).

28
Doing this does not change which coordinate is the first without a −1, so the
pairing is well-defined. Every simplex is paired (∅ is paired with the vertex
with a +1 in the first coordinate and nothing in any other coordinate), so if
there are no cycles in this matching, Fk,l .
Suppose there were a cycle. It would have to be of the form:

σ0 l τ0 m σ1 l τ1 m σ2 l . . . l τs−1 m σs = σ0

where each σi is paired with τi . Also, for 1 ≤ i ≤ s, σi must be τi−1 minus

a vertex vi . Therefore, since this is a cycle, there must be a j such that
τj = σj ∪ {vi }. For this to be possible, vi must be a +1. Thus, all of the
simplices in the cycle must have all the same −1 coordinates, but if that is
the case, the vertex to be added in any σi l τi pair is always the same, and
vi must be the same for every i. This is a contradiction. Therefore, there are
no cycles, and we have a Morse matching with a single critical simplex.
In the proof of Proposition 5.4 below, we will use the following lemma,
which is essentially an equivariant version of Theorem 3.1 of [11].
Lemma 5.3. Let G be any group and P be a finite poset, h : P → P an
order-preserving poset map such that for any x ∈ P , h(x) ≥ x (or h(x) ≤ x).
Define Q to be the set of fixed points of h. Then ∆P collapses simplicially
to ∆Q. In the case that h is a G–poset map, P &G Q.
Proof. We prove it for the case that h(x) ≥ x, the other case being almost
identical. Since P is finite, we may choose N large enough so that for all x
in P , we have hN (x) ∈ Q. Now let σ ∈ ∆P be a chain x0 < x1 < . . . < xm .
If ∃ i : 0 ≤ i ≤ m such that xi ∈ / Q, let k be the largest such i. Then
we may insert hN (xk ) into the chain immediately following xk because xk <
hN (xk ) ≤ hN (xk+1 ) = xk+1 if k < m. Associate to σ the chain obtained by
inserting hN (xk ) or by deleting it in the case hN (xk ) = xk+1 . Since it is an
element of Q being inserted or deleted, the selection of k is not affected, and
xk uniquely determines the other chain in the pair. Therefore, this matching
is well-defined. Also, this matching is equivariant if h is a G–map.
Suppose there is a cycle

σ0 l τ0 m σ1 l τ1 m σ2 l . . . l τs−1 m σs = σ0

We have for 1 ≤ i ≤ s that σi = τi−1 \ {yi } for some yi ∈ P . Then there

must be some pair σj l τj = σ ∪ {yi }, so yi ∈ Q for all i. Thus every simplex

29
in the cycle has all the same elements of P \ Q, so ∃ x ∈ P \ Q that is
the greatest such element in every simplex. Hence τj = σj ∪ {hN (x)} for
all j, and yi = hN (x) for all i. This is a contradiction because the same
element is being added and deleted in consecutive steps. Therefore, we have
a Morse matching whose critical simplices are exactly the elements of ∆Q, a
subcomplex of ∆P . Thus, ∆P G–collapses to ∆Q.

Proposition 5.4. ∆K simplicially G–collapses to ∆S.

Proof. The collapsing will occur in three steps. Define

 
A B
K1 := {φ = ∈K | A∩C = 6 ∅}
C D

K2 := {φ ∈ K1 | A = C}
First, we collapse ∆K to ∆K1 . Let σ be a chain of the form

φ0 < φ1 < . . . < φm−1 < φm

 
Ai Bi
in ∆K where φi = . If A0 ∩ C0 = ∅, we want to pair σ with
Ci Di
another chain for which that is also true. Find the last k such that Ak ∩ Ck =
∅. Then compare Bk and Dk to Bm and Dm . If Bk = Bm and Dk = Dm , pair
σ with the
 chain obtained by adding
 to (or deleting from) the end of σ the
c
Am ∪ (Bm ∪ Dm ) Bm
element . Otherwise, find the first l > k where
Cm ∪ (Bm ∪ Dm )c Dm
Bl 6= Bk or Dl 6= Dk . Now pair σ withthe chain obtained by inserting (or
Al Bl−1
removing if it equals Xl−1 ) before φl . Nowhere are we inserting
Cl Dl−1
or deleting elements with A ∩ C = ∅, so the selection of k is not affected.
In the second case, we are inserting or deleting an element with Bl−1 = Bk
and Dl−1 = Dk , so the selection of l is not affected. Therefore, the matching
is well-defined. The critical simplices are exactly those where A0 ∩ C0 6= ∅,
forming ∆K1 , a subcomplex. Therefore, if there are no cycles, we have a
collapsing from ∆K to ∆K1 . Also, the pairings are chosen equivariantly, so
we will have a G–collapse.
Suppose we have a cycle

σ0 l τ0 m σ1 l τ1 m σ2 l . . . l τs−1 m σs = σ0

30
Again, for 1 ≤ i ≤ s, σi is obtained from τi−1 by deleting an element ψi ,
so there must be a pair σj l τj = σj ∪ {ψi } coming from our matching.
Therefore, ψi ∈ K1 for all i, which means that all the simplices in our cycle
have all of the same elements with A ∩ C = ∅. Thus, they all have the same
φk , so Bk and Dk are fixed and we know that every ψi has them as its second
column. As a result, the elements after φk that have B 6= Bk or D 6= Dk
are not changing as we move through the cycle, implying that ψi is the same
for all i. This is a contradiction, so our matching has no cycles. This proves
that ∆K G–collapses to ∆K1 .
The next two collapsingsare proved byLemma 5.3. For the first, we define
A∩C B
h1 : K1 → K1 by h1 (φ) = . This is an order-preserving G–
A∩C D
poset map, and h1 (φ) ≤ φ. The fixed point set of h1 is exactly K2 , so Lemma
5.3 implies that ∆K1 G–collapsesto ∆K2 . Forthe second collapsing, we now
A B∪D
define h2 : K2 → K2 by h2 (φ) = . This is an order-preserving
A B∪D
G–poset map, h2 (φ) ≥ φ, and the fixed point set is S. Therefore, the same
lemma implies that ∆K2 G–collapses to ∆S. Hence, ∆K G–collapses to
∆S.
Theorem 5.5. |∆K| is a G–regular neighborhood of |∆S| with boundary
|∆L|.
Proof. G acts freely outside of |∆S|, so ∂|∆K| is G–bicollarable in |∆M |.
Now the theorem follows immediately from 4.19 (the collapsing criterion for
G–regular neighborhoods) and Propositions 5.1 and 5.4.
Now our main result follows easily:
Main Theorem. The regular cell complex Hom{3} (P4 , Kn ) is a PL manifold
with boundary Hom{3} (C5 , Kn ) and is equivariantly homeomorphic (with re-
spect to the involution described above) to N := {(x, y) ∈ S n−2 × S n−2 | x ·
y ≥ 0}, where the involution on N interchanges (x, y) with (y, x). The
Stiefel manifold Vn−1,2 = ∂N is therefore equivariantly homeomorphic to
Hom{3} (C5 , Kn ), which is equivariantly homeomorphic to Hom(C5 , Kn ).
Proof. It follows from 2.1 that we have Hom{3} (P4 , K n ) ≈G |∆K| with the
subcomplex Hom{3} (C5 , K n ) ≈G |∆L|. Because |∆K| and N are both G–
regular neighborhoods of the diagonal, they are equivariantly homeomorphic
by 4.6.

31
 The Stiefel manifold Vn−1,2 has a natural action of the orthogonal group
O2 (with the Grassmannian as the quotient). The equivariant homeomor-
phism above is with respect to a single reflection of O2 . The multimorphism
complex Hom(C5 , Kn ) does not have a combinatorial O2 action; however,
there is the induced action of the dihedral group D5 (a subgroup of O2 )
which is the group of symmetries of the cycle C5 . It seems natural to ask:

Question 5.6. Is Hom(C5 , Kn ) equivariantly homeomorphic to Vn−1,2 with

respect to the action of the dihedral group D5 ?

Unfortunately, neither of the smaller models Hom{3} (C5 , Kn ) or Hom{2,4} (C5 , Kn )

is D5 –invariant, so it seems that one needs to work with Hom(C5 , Kn ) which
does not have an obvious D5 –equivariant embedding into S n−2 × S n−2 . Also,
a good (equivariant) combinatorial candidate for N is missing, which is the
obstacle to applying the methodology above to answer this question posi-
tively.

References
[1] N Alon, Splitting necklaces, Adv. in Math., 63 (1987), 247-253

[2] I Bárány, S B Shlossman, and A Szűcs, On a topological gener-

alization of a theorem of Tverberg, J. London Math. Soc., 23 (1981),
158-164

[3] E Babson and D N Kozlov, Complexes of graph homomorphisms,

Isr. J. Math., 152 (2006), 285-312

[4] E Babson and D N Kozlov, Proof of the Lovász conjecture, Annals

of Math., 165 (2007), 965-1007

[5] A Björner, M Las Vergnas, B Sturmfels, N White, and G M

Ziegler, Oriented matroids, 2nd ed., Encyc. of Math. and its Appl. 46,
Cambridge Univ. Press (1999)

[6] P Csorba Non-tidy spaces and graph colorings, Ph.D. thesis, ETH
Zürich (2005)

[7] P Csorba and F H Lutz, Graph coloring manifolds, Contemp. Math.,

423 (2006), 51-69

32
 [8] R Forman, Morse theory for cell complexes, Adv. in Math., 134 (1998),
90-145

[9] S Illman, Existence and uniqueness of equivariant triangulations of

smooth proper G–manifolds with some applications to equivariant White-
head torsion, J. reine angew. Math., 524 (2000), 129-183

[10] D N Kozlov, Cobounding odd cycle colorings, Elec. Research Announ.

AMS, 12 (2006), 53-55

[11] D N Kozlov, Collapsing along monotone poset maps, Int. J. Math. and
Math. Sciences, (2006), Art. ID 79858, 8 pp.

[12] D N Kozlov, Combinatorial algebraic topology, Springer (2007)

[13] D N Kozlov, Cohomology of colorings of cycles, Amer. J. Math., 130

(2008), 829-857

[14] L Lovász, Kneser’s conjecture, chromatic number and homotopy, J.

Comb. Theory, Ser. A, 25 (1978), 319-324

[15] A Lundell and S Weingram, The topology of CW complexes, Van

Nostrand Reinhold Company (1969)

[16] M Özaydın, Equivariant maps for the symmetric group,

Univ. of Wisconsin preprint (1987), 17pp., Available at:
http://digital.library.wisc.edu/1793/63829, Accessed 12/17/2012

[17] C P Rourke and B J Sanderson, Introduction to piecewise-linear

topology, Springer-Verlag (1982)

[18] C Schultz, Small models of graph colouring manifolds and the Stiefel
manifolds Hom(C5 , Kn ), J. Comb. Theory, Ser. A, 115 (2008), 94-104

[19] C Schultz A short proof of w1n (Hom(C2r+1 , Kn+2 )) = 0 for all n and
a graph colouring theorem by Babson and Kozlov, Isr. J. Math., 170
(2009), 125-134

[20] C Schultz Graph colorings, spaces of edges and spaces of circuits, Adv.
in Math., 221 (2009), 1733-1756

33
[21] G M Ziegler, Generalized Kneser coloring theorems with combinatorial
proofs, Invent. math., 147 (2002), 671-691. Erratum 163 (2006), 227-228

[22] R Živaljević, User’s guide to equivariant methods in combinatorics, I

and II, Publ. Inst. Math. (Beograd) (N.S.), (I) 59(73) (1996), 114-130
and (II) 64(78) (1998), 107-132

34