ON THE COMPUTATION OF COARSE COHOMOLOGY

ARKA BANERJEE

Abstract. The purpose of this article is to relate coarse cohomology of metric spaces
arXiv:2401.02231v1 [math.MG] 4 Jan 2024

with a more computable cohomology. We introduce a notion of boundedly supported

cohomology and prove that coarse cohomology of many spaces are isomorphic to the boundedly
supported cohomology. Boundedly supported cohomology coincides with compactly supported
Alexander–Spanier cohomology if the space is proper and contractible. Our work generalizes
an earlier result of Roe which says that the coarse cohomology is isomorphic to the compactly
supported Alexander-Spanier cohomology if the space is uniformly contractible. As an
application of our main theorem, we obtain that coarse cohomology of the complement can
be computed in terms of Alexander-Spanier cohomology for many spaces.

1. Introduction
John Roe [5] introduced the notion of coarse cohomology H X∗ (X) of a metric space X to
study large scale geometry of the space. This cohomology roughly measures the way in which
uniformly large bounded sets fit together. Coarse cohomology is an invariant of the large
scale geometry of the space: if two spaces are coarsely equivalent, then they have isomorphic
coarse cohomology. In general, coarse cohomology is hard to compute. For nice spaces Roe
proved the following.
Theorem 1.1 (Roe [5]). If X is uniformly contractible, then the coarse cohomology H X∗ (X)
is isomorphic to the compactly supported Alexander–Spanier cohomology H∗c (X).
A space is called uniformly contractible if there exists a function ρ : R+ → R+ such that
any set of diameter r is contractible inside a set of diameter ρ(r). Examples of uniformly
contractible spaces include universal cover of compact aspherical complexes. For example,
Euclidean space of any dimension is uniformly contractible and proper. By Roe’s Theorem it
follows that H X∗ (Rn ) = H∗c (Rn ).
In this article, our goal is to generalize the above theorem to facilitate computation of
coarse cohomology for more spaces. Our main motivation for doing that is to compute coarse
cohomology of the complement of a subspace A ⊂ X denoted by H X∗ (X − A). This notion
was introduced and studied in [1]. One way to describe this cohomology isSas follows: consider
complements of expanding neighborhoods X − Nf (A), where Nf (A) = x∈A B(x, f (x)) for
proper functions f : A → (0, ∞). These complements form a directed system of metric spaces
under inclusion (in the direction of slower growth of f ), and coarse (co)chain complexes of
the complement X − A is defined to be the inverse limits of the usual coarse complexes of
these spaces.
In [1], it was shown that the coarse cohomology of the complement can be realized as the
coarse cohomology of a single space. Equip X with the following distance function
dA (x, y) = min{d(x, A) + d(y, A), d(x, y)}.
1
2 ARKA BANERJEE

dA defines a pseudometric on X. It induces a metric on the space X/Ā which is formed by

identifying the closure of A to a point.
Proposition 1.2 ([1]). Let X be a metric space, A ⊂ X and X/Ā is the metric space
equipped with the metric induced from dA . Then H X∗ (X/Ā) is isomorphic to H X∗ (X − A).
So, one way to get the coarse cohomology of the complement of A is to understand
the coarse cohomology of (X/Ā, dA ). Unfortunately, Theorem 1.1 does not apply to the
computation of H X∗ (X/Ā) because the space (X/Ā, dA ) is rarely uniformly contractible even
if X is. On the other hand, if X is uniformly contractible, then (X/Ā, dA ) retains those
properties at infinity in some sense which we describe next.
We say X is uniformly contractible at infinity if there exist two non-decreasing control
function ρ, µ : R+ → R+ and a basepoint b ∈ X such that any set B of diameter r is
contractible inside a set of diameter ρ(r) if d(b, B) ≥ µ(r).
Our main theorem in this article is the following.
Theorem 1.3. If X is uniformly contractible at infinity, then its coarse cohomology H X∗ (X)
is isomorphic to its boundedly supported cohomology H B∗ (X).
We will define H B∗ (X) later (Definition 2) in the paper. In particular, when X is
proper and contractible, H B∗ (X) coincides with the compactly supported Alexander–Spanier
cohomology (Example 2.2). Hence, Theorem 1.3 generalizes Theorem 1.1. In general, H B∗ (X)
is isomorphic to the reduced Alexander–Spanier cohomology of X at infinity with degree
shifted down by one (Proposition 2.3). As a consequence of Theorem 1.3, we prove the
following.
Corollary 1.4. Suppose X is uniformly contractible at infinity. Let A ⊂ X so that X ̸= Nr (A)
for any r. Then H X∗ (X − A) = lim H̃∗−1 (X − Nr (A)) for ∗ ≥ 1.
−→
The hypothesis of Theorem 1.3 is much weaker than the hypothesis of Roe’s Theorem 1.1.
In particular, the property of uniform contractibility at infinity is more robust compared to
uniform contractibility. For example, deleting a bounded set does not affect the property
of being uniformly contractible at infinity. This is not true for uniformly contractible
spaces because such spaces are necessarily contractible and complement of bounded set in a
contractible set might not be contractible anymore. On the other hand, spaces satisfying
uniform contractibility at infinity can be very far from being contractible (see Figure 1).
Our approach to prove Theorem 1.3 is different from the proof of Roe’s Theorem 1.1 in [5].
The main part of the proof of Theorem 1.1 in [5] involve showing that coarse cohomology is
isomorphic to the C̆ech cohomology of certain open covers. In order to show that Roe relates
this C̆ech cohomology to certain presheaf cohomology using double complex. Our approach
to prove Theorem 1.3 follows an idea from [6] which is more geometric and direct in the sense
that we explicitly construct (co)chain homotopy to establish the isomorphism between the
concerned cohomology groups.
A space X is called uniformly acyclic at infinity if there exist a functions ρ : [0, ∞) →
[0, ∞) and a basepoint b ∈ X such that any set B of diameter r, the inclusion B ,→ Nρ(r) (B)
induces trivial map between the singular homology. Another interesting aspect of the
Theorem 1.3 is that it does not hold if we replace uniform contractibility at infinity by
uniform acyclicity. That means any proof of this theorem needed to be able to distinguish
between uniform acyclicity and uniform contractibility. This makes the proof more subtle
than one might initially expect it to be.
 ON THE COMPUTATION OF COARSE COHOMOLOGY 3

Figure 1. A subspace of R2 that consists of countable union of circles {Ci }i∈N

and the ray r := [0, ∞) × {0} such that the ith circle has radius i and distance
between two consecutive circle grows to infinity. This is an example of a space
that is uniformly contractible at infinity.

Overview. In section 2, we set some terminologies for various cochain complexes. In

particular, we define boundedly supported cohomology H B∗ (X) and recall coarse cohomology
H X∗ (X). In section 3, we prove a warm-up theorem that says H X∗ (X) and H B∗ (X) are
isomorphic when X is uniformly acyclic at infinity and locally acyclic at infinity. This proof
gives the main ideas that will go into the proof of the Theorem 1.3. Section 4 contains the
proof of the Theorem 1.3. Finally in section 5, we prove Corollary 1.4.
Acknowledgement. Part of this paper was written when the author was a graduate student
in University of Wisconsin–Milwaukee. The author would like to thank his advisor Boris
Okun for his insights and advice during the project. In particular, the author is indebted to
him for sharing his idea on the proof of Theorem 3.1 from which this whole project originated.
The author would also like to thank Kevin Schreve for helpful conversations.

2. Preliminaries
We introduce some notation for various cochain complexes and corresponding cohomology
groups associated to X. Let R be a ring. The basic complex is the complex of all cochains
with d : Cn−1 (X; R) → Cn (X; R) being the boundary maps as follows:

Cn (X; R) = {ϕ : X n+1 → R}
n
X
(dϕ)(x0 , . . . , xn ) = (−1)i ϕ(x0 , . . . , x̂i , . . . , xn )
i=0

It is an acyclic complex.
4 ARKA BANERJEE

We will refer to points in X n+1 as n-simplices. We will refer to a continuous map

f : ∆n → X as a singular n-simplex. We will often need to measure distances between
simplices of different dimensions. A convenient way to do this is to stabilize simplices
by repeating the last coordinate, as follows. Let X ∞ denotes the set of all eventually
constant sequence in X equipped with the sup metric. Let i : X n+1 → X ∞ denote the
map (x0 , . . . , xn ) 7→ (x0 , . . . , xn , xn , xn , . . . ). For a cochain ϕ define its stabilized support
|ϕ| = {i(σ) | σ ∈ X n+1 and ϕ(σ) ̸= 0} ⊂ X ∞ .
Let ∆ = i(X) denote the diagonal of X ∞ . Let ||ϕ|| be the intersection of the diagonal
∆ = {(x, x, . . . , x) | x ∈ X} ⊂ X ∗+1 and the closure of the support of the function
ϕ : X ∗+1 → R.

Alexander–Spanier cohomology. Let C∗0 (X; R) be the complex of locally zero cochains:
C∗0 (X; R) = {ϕ ∈ C∗ (X; R) | ∥ϕ∥ = ∅}.
Note that restriction of d gives a well defined map C∗0 (X) → C∗+1
0 (X). Consequently, d
induces a well defined map C∗as (X; M ) → C∗+1
as (X; R), where
C∗as (X; R) = C∗ (X; R)/ C∗0 (X; R).
Alexander–Spanier cohomology, denoted by H∗ (X; M ), is the cohomology of the complex
(C∗as (X; M ), d).
Example 2.1. Alexander–Spanier cohomology groups coincide with the singular cohomology
groups for locally finite complexes. On the other hand, they may be different for the spaces
that are not locally connected. For instance, degree 1 Alexander–Spanier cohomology group
of the Warsaw circle with coefficient in Z is Z whereas degree 1 singular cohomology group is
trivial.
Compactly supported Alexander–Spanier cohomology. Compactly supported cochains
are the cochains from the following complex
C∗c (X; R) = {ϕ ∈ C∗ (X; R) | ∥ϕ∥ is compact}.
Compactly supported Alexander–Spanier cohomology H∗c (X; R) is the cohomology of the
following cochain complex
C∗cas (X; R) = C∗c (X; R)/ C∗0 (X; R).
with the usual boundary operator induced by d.

Boundedly supported cohomology. Now we define a natural generalization of compactly

supported cochains, called boundedly supported cochains:
C B∗ (X; R) := {ϕ ∈ C∗ (X; R) | ||ϕ|| is bounded}
Boundedly supported cohomology H B∗ (X; R) is the cohomology of C B∗ (X; R) with the
usual boundary operator d.
Example 2.2. If X is proper and contractible, then H B∗ (X; R) = H∗c (X; R). Indeed, if X is
contractible, then C∗0 (X; R) is an acyclic complex. In this case, the sequence
0 → C∗0 → C∗ → C∗cas → 0
 ON THE COMPUTATION OF COARSE COHOMOLOGY 5

gives H∗c (X; R) = H∗ (C∗c (X; R)). Moreover, if X is proper, then C∗c (X; R) = C B∗ (X; R).
Hence we get H∗c (X; R) = H∗ (C B∗ (X; R)) = H B∗ (X; R). For example,
(
Z if ∗ = n
H B∗ (Rn ; Z) = H∗c (Rn ; Z) =
0 otherwise
In general, the boundedly supported cohomology is the same as the (reduced) Alexander–
Spanier cohomology at infinity with degree shifted down by one. More precisely we have the
following.
Proposition 2.3. (1) If X is bounded, then
(
R if ∗ = 0
H B∗ (X; R) =
0 otherwise
(2) If X is unbounded and b ∈ X, then
(
0 if ∗ = 0
H B∗ (X; R) = ∗−1
lim H̃ (X − Nr (b); R) otherwise
−→
Proof. (1) If X is bounded then C B∗ (X; R) = C∗ (X; R). The cohomology of the latter
complex is trivial except in degree 0. Hence, H B∗ (X; R) = 0 for ∗ ≥ 1. H B0 (X; R) =
{constant functions on X} ∼ = R.
(2) Elements in H B0 (X; R) consists of constant functions defined on X with support
contained in a neighborhood of b. This means, if X is not bounded, then H B0 (X; R) is
trivial. Hence in this case H B0 (X; R) = 0.
Consider the following maps between the cochain complexes where j is the inclusion map
and i is induced by canonical restriction maps (followed by quotient maps) ir : C∗ (X; R) →
C∗ (X − Nr (b); R) → C∗as (X − Nr (b); R).
j i
− C∗ (X; R) →
0 → C B∗ (X; R) → − lim C∗as (X − Nr (b); R) → 0
−→
The above is a short exact sequence because of the following
ker(i) = {ϕ ∈ C∗ (X; R) | ϕ ∈ C∗0 (X − Nr (A); R) for some r}
= {ϕ ∈ C∗ (X; R) | ||ϕ|| is bounded}
= im(j)
The above short exact sequence of cochain complexes induces a long exact sequence of the
corresponding reduced cohomology groups. The reduced cohomology group of the middle
complex is trivial in all degrees. Hence, the long exact sequence implies that
H B∗ (X; R) ∼= lim H̃∗−1 (X − Nr (b); R) for ∗ ≥ 1
−→
□
Example 2.4. Suppose X is the space appearing in figure 1. Since X is unbounded, H B0 (X)
is trivial. Suppose b ∈ X. For ∗ ≥ 1, we have the following
H B∗ (X, Z) = lim H̃∗−1 (X − Nr (b); Z)
−
(→
Π∞ ∞
i=0 Z/ ⊕i=0 Z ∗=2
=
0 otherwise
6 ARKA BANERJEE

We finish this section by briefly recalling the coarse cohomology.

Coarse cohomology. Coarse cohomology of a space X is the cohomology of the following
cochain complex

C X∗ (X; R) := {ϕ ∈ C∗ (X; R) | |ϕ| ∩ Nr (∆) is bounded for all r}

Example 2.5. According to Theorem 1.1, coarse cohomology is isomorphic to compactly
supported Alexander–Spanier cohomology for uniformly contractible spaces. In particular,
coarse cohomology of the universal cover of any compact aspherical space is isomorphic to
its compactly supported Alexander–Spanier cohomology. For example, H X∗ (Rn ; Z) = Z for
∗ = n and is trivial otherwise.

3. A warm-up theorem
A space is called locally acyclic at infinity with coefficient in R if complement of some
bounded subset in the space is locally acyclic with coefficients in R. A space X is uniformly
acyclic at infinity with coefficients in R if there exist two non-decreasing control functions
µ, ρ : [0, ∞) → [0, ∞) and a basepoint b ∈ X such that any set B of diameter r and
d(b, B) ≥ µ(r), the inclusion B ,→ Nρ(r) (B) induces trivial map between the singular
homology with coefficients in R. Sometimes, we will refer to such space as (µ, ρ)-uniformly
acyclic at infinity.
In this section we prove the following Theorem.
Theorem 3.1. If X is uniformly acyclic at infinity and locally acyclic at infinity with
coefficients in R, then the inclusion C X∗ (X; R) ,→ C B∗ (X; R) induces an isomorphism on
cohomology:
H X∗ (X; R) ∼
= H B∗ (X; R).
For a uniformly acyclic at infinity space, we can perform a version of the standard “connect
the dots” construction and the proof of the above theorem relies on that. Suppose X is
(µ, ρ)-uniformly contractible at infinity. Every 1-simplex σ of diameter r outside of the ball
of radius µ(r) is fillable, so we can pick a singular 1-chain f (σ), such that |f (σ)| ⊂ Nρ(r) (σ)
and ∂f (σ) = ∂σ. Proceeding by induction on the dimension, if a simplex σ is sufficiently far
from the base point, its boundary is already filled by a singular cycle that bounds a singular
chain f (σ) contained in a controlled neighborhood of the simplex. Moreover, if the space is
locally acyclic at infinity, we can choose f (σ) to have small diameter whenever diameter of σ
is small and is outside some bounded set. In order to do that we can take the chain f (σ) to
be of diameter 2k(σ) where
k(σ) := inf{diam(|c|) | ∂c = f (∂σ)}.
Here we are multiplying by 2 to ensure existence of such chain and that whenever X is locally
acyclic at x, for any open neighborhood U of x, there exist another neighborhood V ⊂ U of
x, such that |f (σ)| ⊂ U for all σ ∈ V n+1 .
Note that not every simplex is fillable, and that the diameter of fillings grow with dimension,
as well as the size of the balls that we have to avoid. To formalize the notion of sufficiently
far, we make the following definition. Given an increasing sequence of increasing control
functions µn : [0, ∞) → [0, ∞), denote
CFn (X; R) = ⟨σ n | d(σ n , b) ≥ µn (diam σ n )⟩ ⊂ Cn (X; R)
 ON THE COMPUTATION OF COARSE COHOMOLOGY 7

Since µn is increasing, this defines a subcomplex of the chain complex of finitely supported
chains. Let V : Cs∗ (X; R) → C∗ (X; R) denote the forgetful map, which maps a singular
simplex to its vertices. For the rest of the section we will suppress the coefficient R from the
notation. The previous discussion then gives us the following.
Lemma 3.2. Suppose X is uniformly acyclic at infinity and is locally acyclic at infinity.
Then there exist two non-decreasing sequence of control functions ρn , µn : [0, ∞) → [0, ∞)
and a map M : CF∗ (X) → Cs∗ (X) where
CFn (X) = ⟨σ n | d(σ n , A) ≥ µn (diam σ n )⟩ ⊂ Cn (X)
such that
(1) M is a chain map.
(2) |M (σ n )| ⊂ Nρn (diam(σn )) (σ n ).
(3) There is bounded set B ⊂ X such that for any x ∈ / B, and a neighborhood U of x,
there is a neighborhood W of x such that |M (σ n )| ⊂ U for all σ n ∈ W n+1 .
Lemma 3.3. Assume that X ⊂ Y . Let f : C∗ (X) → C∗ (Y ) be a chain map such that for
any n-simplex σ n , |f (σ n )| ⊂ Nρn (diam(σn )) (σ n ) for some non-decreasing sequence of function
ρn : [0, ∞) → [0, ∞). Then f and the inclusion map i : C∗ (X) → C∗ (Y ) are homotopic via a
chain homotopy D such that |D(σ)| ⊂ Nρn (diam(σn )) (σ n ).
Proof. We can define D by induction on the dimension of σ. Define D0 (x) := (x, f (x)). Note
that f (x)−x = ∂D(x) and D0 (x) ⊂ Nρ0 (diam{x}) (x). Suppose we have defined Dm : Cm (X) →
Cm+1 (X) such that Dm (σ) ⊂ Ndiam(σ) (σ) and ∂Dm (σ) + Dm−1 ∂(σ) = i(σ) − f (σ) for any
m ≤ n. To define Dn+1 (σ) for an (n + 1)-simplex σ, consider c = i(σ) − f (σ) − Dn ∂(σ). By
induction hypothesis, c is a cycle and |c| ⊂ Nρ(diam(σ)) (σ). Take a vertex b from the chain c,
and consider the cone operator Tb : Cn+1 (X) → Cn+2 (X), (x0 , . . . , xn+1 ) 7→ (b, x0 , . . . , xn+1 ).
Define Dn+1 (σ) := Tb (c). By construction, ∂Dn+1 (σ) = c = i(σ) − f (σ) − Dn ∂(σ) and
|Dn+1 (σ)| ⊂ Nρn (diam(σ)) (σ). □

Let CU∗ (X) be the complex of singular chains supported by U. Combining the previous
two lemmas we can now prove the following which will be the main ingredient to prove
Theorem 3.1.
Proposition 3.4. Suppose X is uniformly acyclic at infinity and locally acyclic at infinity. Let
U be an open cover of X. For each x ∈ X, fix a set Ux ∈ U that contains x. Then there exist
two increasing sequences of control functions µn and ρn and a chain map S : CF∗ (X) → CU∗ (X)
where
CFn (X) = ⟨σ n | d(σ n , b) ≥ µn (diam σ n )⟩ ⊂ Cn (X)
and a map G : CF∗ (X) → C∗+1 (X) so that
(1) G is a chain homotopy between V S and the inclusion map i : CF∗ (X) → C∗ (X):
∂G(σ) = i(σ) − V S(σ) − G∂(σ).
(2) |G(σ n )| ⊂ Nρn (diam(σn )) (σ n ).
(3) There exists a bounded set B such that for any point x ∈/ B and a neighborhood U of
n n+2
x, there is a neighborhood W of x so that |G(σ )| ⊂ U for any σ ∈ W n+1 .
8 ARKA BANERJEE

Proof. Let µ∗ , ρ∗ and CF∗ (X) be the ones provided by the lemma 3.2. We choose a barycentric
subdivision map p : Cs∗ (X) → CU∗ (X) and compose it with the map M from lemma 3.2 to get
the map S : CF∗ (X) → CU∗ (X). Since |M (σ n )| ⊂ Nρn (diam(σn )) (σ n ) and |p(M (σ n ))| ⊂ |M (σ n )|,
it follows that |S(σ n )| ⊂ Nρn (diam(σ)) (σ n ). Applying the Lemma 3.3 to the map V S, we get
the homotopy G between V S and i with property (2). Finally property (3) follows from the
property (3) of the map M in Lemma 3.2. □

We are now ready to prove Theorem 3.1.

Proof of 3.1. By the long exact sequence, we need to show that H∗ (C B∗ (X)/ C X∗ (X)) = 0.
That means for ϕ ∈ C Bn (X) with dϕ ∈ C Xn+1 (X), we need to find ψ ∈ C Bn−1 (X) so that
ϕ − dψ ∈ C Xn (X). Let U be a bounded neighborhood of ∥ϕ∥ in X and for each x ∈ X − ∥ϕ∥
choose a neighborhood Ux such that Uxn+1 ∩ |ϕ| = ∅. Let U denote the cover of X that consist
of the collection of Ux together with U .
Proposition 3.4 produces the chain homotopy G : CF∗ (X) → C∗+1 (X), which we use to
define a linear map D : C∗ (X) → C∗+1 (X) by setting
(
G(σ n ) if σ n ∈ CFn (X),
D(σ n ) =
0 otherwise.

We define
T = id − ∂D − D∂,
Dually we have
T ∗ = id − D∗ d − dD∗ .
We claim that T ∗ ϕ ∈ C Xn (X). By Proposition 3.4(1), T (σ) = V S(σ) for all σ ∈ CFn (X).
If diam(σ n ) ≤ k and d(σ n , b) > µn (k) for some k ≥ 0, then σ n ∈ CF∗ (X) and hence
T (σ n ) = V S(σ n ). Moreover, if σ n is outside of the ρn (k)-neighborhood of U n+1 , then |T (σ n )|
does not touch U n+1 because |T (σ n )| = |V S(σ n )| ⊂ Nρn (k) (σ n ). This implies a simplex in
|T (σ)| belongs to Uxn+1 for some x. Therefore (T ∗ ϕ)(σ n ) = ϕ(T (σ n )) = 0. In other words,
σn ∈/ |T ∗ ϕ|. It follows that,
|T ∗ ϕ| ∩ Nk (∆) ⊂ Nµn (k) (b) ∪ Nρn (k) (U n+1 ).
Since U is bounded by assumption, |T ∗ ϕ| ∩ Nk (∆) is bounded. This proves the claim.
Next we claim that D∗ d(ϕ′ ) is coarse. We have |Gσ| ⊂ Nρn (diam(σ)) (σ) from Proposi-
tion 3.4(2). Therefore, by construction |Dσ| ⊂ Nρn (diam(σ)) (σ). So, D∗ preserves coarseness,
and the claim follows since d(ϕ′ ) is coarse.
Finally we claim that D∗ (ϕ) ∈ C Bn−1 (X). By Proposition 3.4(3), we can choose a bounded
set B containing U such that for any point x ∈ / B, and a Ux ∈ U containing x, there is a
neighborhood W of x so that |G(σ )| ⊂ Ux for any σ n ∈ V n+1 . Hence for x ∈
n n+2
/ B ∪ ||ϕ||,
n+1 ∗
we have |D∗ (σ)| ∈/ |ϕ| for all σ ∈ W . Therefore, ||D (ϕ)|| ⊂ ||ϕ|| ∪ B. The claim follows
∗
since ϕ ∈ C B (X).
Since ϕ + dD∗ (ϕ) = T ∗ (ϕ) − D∗ dϕ, our desired cochain is ψ = −D∗ (ϕ). □

Combining Theorem 1.3 and Proposition 2.3 we get the following

 ON THE COMPUTATION OF COARSE COHOMOLOGY 9

Corollary 3.5. If X is unbounded, uniformly acyclic at infinity and locally acyclic at infinity,
then for any b ∈ X
(
0 if ∗ = 0
H X∗ (X; R) = ∗−1
lim H̃ (X − Nr (b); R) otherwise.
−→
Example 3.6. Let X be the space appearing in figure 1. X is unbounded, uniformly acyclic
at infinity and locally acyclic. By Theorem 3.1, we have H X∗ (X) = H B∗ (X). It follows from
Example 2.4 that
(
∗ Π∞ ∞
i=0 Z/ ⊕i=0 Z ∗=2
H X (X; Z) =
0 otherwise.
Remark 3.7. Theorem 3.1 does notFhold if we drop the locally acyclic at infinity condition. For
instance, consider the space X = ∞ i=1 Si of disjoint union of countably infinite Warsaw circles.
2
We can embed X into R in a way so that diam(Si ) = 1 for all i and X is coarsely equivalent to
a ray [0, ∞) with the subspace metric. H X2 (X; Z) = H X2 (X; Z) = H X2 ([0, ∞); Z) = 0. On
the other hand, each Warsaw circle has nontrivial Alexander–Spanier cohomology in degree 1.
It follows that, lim H̃1 (X − Nr (b); Z) ̸= 0. So, the conclusion of the above Corollary 3.5 fails
−→
in this case.
4. Proof of the main theorem
In this section we prove our main theorem. We will continue to suppress the coefficient
ring R from the notation.
Theorem 4.1. If X is uniformly contractible at infinity, then the inclusion C X∗ (X) ,→
C B∗ (X) induces an isomorphism on cohomology:
H X∗ (X) ∼
= H B∗ (X).
The underlying idea of the proof is similar to the proof of Theorem 3.1. However, recall
from Remark 3.7 that Theorem 4.1 does not hold if we replace uniform contractibility by
uniform acyclicity. This means any proof of the Theorem 4.1 needs to be able to distinguish
between uniform acyclicity and uniform contractibility.
Most of the work will go into proving Proposition 4.2 which is an analog of Proposition 3.4
when X is uniformly contractible at infinity. The main difficulty of getting such analog is
the absence of local acyclicity of X which was present in Proposition 3.4. To get around
that, we exploit the Kuratowski’s embedding theorem [3] that says any metric space X can
be isometrically embedded inside the locally acyclic space ℓ∞ (X) which is the space of all
bounded sequence {xn } in X with the metric d({xn }, {yn }) = supn∈N {d(xn , yn )}. We then
take neighborhood of X inside ℓ∞ (X). This neighborhood will be locally acyclic but might
not be uniformly acyclic which is another condition we absolutely need. This tension make
the proof of Proposition 4.2 more subtle than Proposition 3.4.
For convenience, we will use ℓ∞ to mean ℓ∞ (X) for the rest of the paper.
Proposition 4.2. Suppose X ⊂ ℓ∞ is uniformly contractible at infinity. Let N (X) be an
open neighborhood of X in ℓ∞ . Let U be an open cover of N (X). Then there exist two
non-decreasing sequence of control functions ρn , µn : R≥0 → R≥0 and a subcomplex CF∗ (X) of
C∗ (X) of the following form
CFn (X) = ⟨σ n | d(σ n , b) > µn (diam(σ n ))⟩ ⊂ Cn (X),
10 ARKA BANERJEE

a chain map S : CF∗ (X) → CU∗ (N (X)) and a map G : CF∗ (X) → C∗+1 (N (X)) with the
following properties.
(1) G is a chain homotopy between V S and the inclusion map i : CF∗ (X) ,→ C∗ (N (X)).
∂G(σ) = V S(σ) − i(σ) − G∂(σ)
(2) |G(σ n )| ⊂ Nρn (diam(σn )) (σ n ).
(3) For any k ∈ N, there exists a bounded set B ⊂ X such that for any x ∈ / B and a
neighborhood U of x in ℓ∞ , there is a neighborhood W of x in X such that |G(σ n )| ⊂
U n+2 for all σ n ∈ W n+1 if n ≤ k.
Sketch of the proof: As stated before, the main idea of the proof of Proposition 4.2 is
similar to the Proposition 3.4. To get the map S, we first fill simplices in X by singular
chains and then use barycentric subdivision on these chains until they fall inside CU∗ (X).
Using uniform contractibility of X at infinity, most of this filling process can be done inside
X with the necessary control on the support as required by property (2).
But to have property (3), we need to fill small simplices by small singular chains. Unless
X is locally acyclic this cannot be achieved by staying inside X.
This is where we use the ambient space ℓ∞ . Since N (X) is an open subset of ℓ∞ , we can fill
small enough simplices of X in N (X) by taking the convex hull of its vertices. In summary,
we fill ‘big’ simplices in X and ‘small’ simplices (small ones) in N (X). The main difficulty is
to choose these fillings in a compatible way so that it gives a chain map CF∗ (X) → Cs∗ (N (X)):
meaning we have to ensure that boundary of the filling is same as filling of the boundary. If
N (X) is uniformly contractible, then we can go as before by induction on the dimension: to
fill in a simplex σ, choose a chain that bounds the filling of ∂σ which is already defined. The
problem is N (X) might not be even contractible, even if X is uniformly contractible.
To get around this problem, we first construct a chain map CF∗ (X) → Cs∗ (X) that sends
small simplices to singular chains which can be homotoped to the convex filling by staying
inside N (X) (Lemma 4.3). Convex filling of a simplex is the image of that simplex under the
following chain map
c : C∗ (X) → Cs∗ (ℓ∞ )
n
X
(x0 , . . . , xn ) 7→ c(σ) : (s0 , . . . , sn ) 7→ s i xi .
i=0

It is in the construction of this chain map CF∗ (X) → Cs∗ (X) where we crucially use the fact
that our space is uniformly contractible at infinity, not just uniformly acyclic at infinity.
Figure 2 illustrates that this construction cannot be performed when X is the Warsaw circle
which is an acyclic space.
In the second step, we modify the chain map CF∗ (X) → Cs∗ (X) produced in Lemma 4.3
to get a chain map CF∗ (X) → Cs∗ (N (X)) that sends small enough simplices to its convex
filling. The idea here is to glue the filling of small simplices obtained in Lemma 4.3 with the
associated homotopy of this filling with the convex filling. This is done in Lemma 4.5.
Finally, to prove Proposition 4.2, we post compose the map CF∗ (X) → Cs∗ (N (X)) from
Lemma 4.5 with a suitable subdivision operator to get the map S into CU∗ (X) and the
homotopy G is obtained by applying Lemma 3.3 to V S : C∗ (X) → C∗ (N (X)).
 ON THE COMPUTATION OF COARSE COHOMOLOGY 11

Figure 2. In the above figure, the space X ⊂ R2 ⊂ ℓ∞ is the Warsaw circle

and N (X) is a tubular neighborhood of the grey region in ℓ∞ that deforms
retract to the grey region. Red line is the convex filling of the simplex (x, y).
Any filling of (x, y) in X has to be around the circular part. Hence, there is no
homotopy in N (X) between the red convex filling and any filling of (x, y) in
X. Furthermore, any neighborhood of x contains such a y where this happens.

Lemma 4.3. Suppose X ⊂ ℓ∞ is uniformly contractible at infinity. Let N (X) be an open

neighborhood of X in ℓ∞ . Then there exist two non-decreasing sequence of control functions
ρn , µn : R≥0 → R≥0 , a subcomplex CF∗ (X) of C∗ (X) of the following form
CFn (X) = ⟨σ n | d(σ n ; b) > µn (diam(σ n ))⟩ ⊂ Cn (X),
a chain map f : CF∗ (X) → Cs∗ (X) and a map D : CF∗ (X) → Cs∗+1 (ℓ∞ ) with the following
properties.
(1) D is a chain homotopy between f and the convex filling c(σ)
∂D(σ) = f (σ) − c(σ) − D∂(σ)
(2) |D(σ n )| ⊂ Nρn (diam(σn )) (σ n ) where the tubular neighborhood is taken in ℓ∞ .
(3) For any k ∈ N, there exists a bounded set Z ⊂ X such that for each point x ∈ X − Z,
there is a neighborhood W of x in X such that for all σ n ∈ W n+1 with n ≤ k,
D(σ n ) ∈ Csn+1 (N (X)) .
Cone construction. We will need to use certain cone operators to define f and D of the
above lemma. Recall that a standard singular n-simplex ∆n is the convex hull of the n + 1
unit vectors in Rn+1 . A singular n-simplex in X is a continuous function α : ∆n → X and
the image of α is called the support of the singular simplex.
Let Ht be a homotopy that contracts some set B ⊂ ℓ∞ to a point in ℓ∞ . We call
such homotopy to be a contracting homotopy. To Ht , we can associate a cone operator
H : Cs∗ (B) → Cs∗+1 (ℓ∞ ). The construction goes as follows. Let α be a singular n-simplex
12 ARKA BANERJEE

supported in B ⊂ ℓ∞ . Consider the following map I × ∆n → ℓ∞ where I = [0, 1].

(t, (s0 , . . . , sn )) 7→ Ht (α(s0 , . . . , sn ))
Since H1 is a constant map, the above map induces a map from (I × ∆n )/({1} × ∆n ) to
ℓ∞ . (I × ∆n )/({1} × ∆n ) can be identified with a singular (n + 1)-simplex. Hence the above
map gives a singular (n + 1)-simplex and we define H(α) to be this simplex. The ith face of
H(α) corresponds to the above map restricted to the set of points that have zero in their ith
coordinates. When dim(α) ≥ 1, one can check that that ∂(H(α)) = α − H(∂α).
Now, we will define a different cone operator H̄ associated to the homotopy Ht . The inputs
of H̄ will be generated by convex fillings of simplices whose vertices lie in B (we will later
extend the domain of H̄). If σ is a 0-simplex, we define H̄(σ) := H(σ). If σ = (x0 , . . . , xn )
with n ≥ 1, then we consider the following map I × ∆n → ℓ∞ .
n
X
(t, (s0 , . . . , sn )) 7→ si Ht (xi )
i=0

The above map restricted to {1} × ∆n is a constant map and hence induces a map from
(I × ∆n )/({1} × ∆n ). That means we can realize the above map as a singular (n + 1)-simplex
in ℓ∞ , which we define to be H̄(c(σ)) (see figure 3). The ith face of this simplex corresponds
to the above map restricted to the set of points whose ith coordinates are zero. In particular,
the 0th face is c(σ) and hence H̄(c(σ)) can be viewed as a cone on c(σ).
We can extend the domain of H̄ to simplices of the form Ḡ(c(σ)) where Ḡ is the associated
cone operator to some contracting homotopy Gs such that the composition Ht ◦ Gs is defined
on the vertices of an σ for all [t, s] ∈ [0, 1]2 .
If σ is a 0-simplex, then we define H̄(Ḡ(c(σ))) := H(G(σ)). If σ = (x0 , . . . , xn ) with n ≥ 1,
then we consider the following map from I 2 × ∆n to ℓ∞ .
n
X
(t, s, (s0 , . . . , sn )) 7→ si Ht (Gs (xi ))
i=0

Since H1 and G1 are constant maps, the above map induce a map from two fold cone on ∆n
and hence gives a singular (n + 2)-simplex. We define H̄(Ḡ(c(σ))) to be this simplex.
We can iterate the above process. Suppose σ = (x0 , . . . , xn ) is an n-simplex for some
n ∈ N. Let {Htkk , . . . , Ht11 } be a set of k contracting homotopies such that the domain of the
composition Htkk ◦ . . . Ht11 contains the vertices of σ. When n = 0, we define
H̄ k (. . . (H̄ 1 (σ))) . . . ) := H k (. . . (H 1 (σ))) . . . )
For n ≥ 1, we consider the following map I k × ∆n → ℓ∞
n
X
(tk , . . . , t1 , (s0 , . . . , sn )) 7→ si Htkk (. . . (Ht11 (xi )))
i=0

Since H1i
is a constant map for each i ∈ {1, 2, . . . , k}, the above map induces a map from
k-fold cone on ∆n . So the above map gives a singular (n + k)-simplex and we define this
simplex to be H̄ k (. . . (H̄ 1 ((c(σ))) . . . ). Again note that, the above map restricted to the set
of points whose first coordinates are zero gives the simplex H̄ k−1 (. . . (H̄ 1 (c(σ))) . . . ).
Let us now summarize the above discussion. H̄ is a linear map defined on a module
generated by singular simplices of the form H̄ k (. . . (H̄ 1 (c(σ))) . . . ) where the domain of the
composition Ht ◦ Htkk ◦ · · · ◦ Ht11 contains the vertices of σ for all (t, tk . . . . , t1 ) ∈ [0, 1]1+k .
 ON THE COMPUTATION OF COARSE COHOMOLOGY 13

Figure 3. Ht and Gs are two contracting homotopy with p and q being their
contracting point respectively. The first two pictures are of H̄(c(x, y)) and
H̄(c(x, y, z)). The third picture is of Ḡ(H̄(c(x, y))).

Suppose β = H̄ k (. . . (H̄ 1 (c(σ))) . . . ) is such a singular simplex in the domain of H̄. Then
H̄(β) is the simplex defined by the following map I k+1 × ∆n → ℓ∞
n
X
(t, tk , . . . , t1 , (s0 , . . . , sn )) 7→ si Ht (Htkk (. . . (Ht11 (xi )) . . . ))
i=0

In what follows, we will blur the distinction between the above map and the corresponding
singular simplex for convenience. We will denote both of them by H̄(β). The following is
immediate from the construction.
Lemma 4.4. If dim(β) ≥ 1, then ∂ H̄(β) = β − H̄(∂β).

Proof of 4.3. Since X is uniformly contractible at infinity, there exist two non-decreasing
control functions ρ, µ : R+ → R+ and a basepoint b ∈ X such that any set B of diameter r
is contractible inside a set of diameter ρ(r) if d(b, B) ≥ µ(r). For each bounded set B ⊂ X
with diameter r and d(b, B) ≥ µ(r), we fix a homotopy HtB that contracts B inside Nρ(r) (B).
Since X is is a metric space it is paracompact. Using paracompactness of X, we pick a locally
finite open cover U = {Uα } of X such that diameters of all Uα are uniformly bounded. In
particular, for each 1-simplex (x, y) ∈ ∪Uα2 , the set S(x, y) := {α | (x, y) ∈ Uα2 } is finite.

Construction of the complex CF∗ (X): We can assume CF0 (X) to be C0 (X) by letting
µ0 = 0. For each 1-simplex σ = (x, y) in ∪Uα2 , we pick an α so that σ ∈ Uα2 and we let
B(σ) := Uα . For other 1-simplices, we let B(σ) to be {x, y}. Proceeding inductively on
S n-simplex σ, suppose B(τ ) is already defined for any τ ∈ |∂σ|. Let r be
dimension, for an
the diameter of τ ∈|∂σ| B(τ ). We define
B(σ) = Nρ(r) (∪τ ∈|∂σ| B(τ )).
14 ARKA BANERJEE

We observe that for any S, there exist R such that diam(B(σ)) ≤ S whenever diam(σ) ≤ R.
Therefore by uniform contractibility at infinity, we can choose an increasing sequence of
functions µn : [0, ∞) → [0, ∞) indexed by natural numbers so that µn ≥ R for n ≥ 1 and
that B(σ) is contractible inside Nρ(diam(B(σ)) (B(σ)) if d(σ n , b) ≥ µn (diam(σ n )). For n ≥ 1,
we let
CFn (X) = ⟨σ n | d(σ n ; b) > µn (diam(σ n ))⟩.

Construction of the chain map f : We let f to be the identity map on CF0 (X) = C0 (X).
B(σ)
For every σ ∈ CF∗ (X), we have an associated homotopy Ht that contracts B(σ) inside
Nρ(diam(B(σ)) (B(σ)). We define f inductively to be the following (see figure 4):

f (σ) := H̄ B(σ) (f (∂σ)).

In order for the above definition to make sense, f (∂σ) has to be in the domain of H̄ B(σ) .
We can show that by induction on dimension of σ. By induction hypothesis, suppose
f (∂σ) is well defined and hence |f (∂σ)| consists of singular simplices of the form H̄ B(τ ) (β)
where τ is a codimension one subsimplex of σ. Note that for any subsimplex τ of σ,
B(τ )
Ht (B(τ )) ⊂ Nρ(diam(B(τ )) B(τ ) ⊂ B(σ). That implies any simplex of the form H̄ B(τ ) (β) is
in the domain of H̄ B(σ) and hence f is well defined.

Now we show that f is a chain map by induction. Notice that f |CF0 (X) is the inclusion map
CF0 (X) ,→ C0 (X). For a 1-simplex (x, y), we have

∂f ((x, y)) = ∂(H̄ B((x,y)) (∂(x, y)))

= ∂(H B((x,y)) (y − x))
= ∂(x, y) = f (∂((x, y)))

Inducting on the dimension of σ, let us assume that f (∂(σ)) = ∂(f (σ)) if 1 ≤ dim(σ) < n.
Recall from Lemma 4.4 that ∂(H̄ B(σ) (α)) = α − H̄ B(σ) (∂α) if dim(α) ≥ 1. If dim(σ) ≥ 2,
then we have the following.
∂f (σ) = ∂ H̄ B(σ) (f (∂(σ)))
= f (∂(σ)) − H̄ B(σ) (∂f (∂(σ))) [by Lemma 4.4]
= f (∂(σ))

Construction of the chain homotopy D: We first let D0 to be trivial on CF0 (X). We

then define D∗ (σ) inductively to be the following.

D∗ (σ) := −H̄ B(σ) (D∗ (∂σ)) − H̄ B(σ) (c(σ))

In order for the above definition to make sense, we need to check that D∗ (∂σ) does be-
long to the domain of the H̄ B(σ) . The proof of this is same as the well definedness of f , so
we will skip it. (See 4 for an illustration of f and D applied to 1 and 2-dimensional simplices.)
 ON THE COMPUTATION OF COARSE COHOMOLOGY 15

Figure 4. In the left picture, the straight line between x and y is c((x, y))
and the red singular 1-simplices give the filling f ((x, y)) = H̄ B((x,y)) (∂(x, y))
B((x,y)
where Ht contracts the set B((x, y)) to the point p. The striped region
is D((x, y)) = H̄ B(x,y) (c((x, y))). The picture on the right is the support
of H̄ B((x,y,z)) (D(∂((x, y, z))) which is made of four singular 3-simplices. The
one in the center is H̄ B(x,y,z) ((x, y, z)). Other three belong to the support of
H̄ B((x,y,z)) (D(∂((x, y, z)))).

Proof of (1): This can be proved by induction on the dimension of σ. Since D0 is trivial,
the claim is true for the base case n = 1. For a 1-simplex (x, y), we have

∂D((x, y)) = −∂ H̄ B((x,y)) (c((x, y))) = −c((x, y)) + H̄ B((x,y)) (∂(c(x, y)))
= −c((x, y)) + H̄ B((x,y)) (∂(x, y))
= −c((x, y)) + f ((x, y))
16 ARKA BANERJEE

Assume that ∂D∗ (σ) = f (σ) − c(σ) − D∗ (∂σ) if dim(σ) < n where n ≥ 2. Suppose σ is an
n-simplex with n ≥ 2. Then we have
∂D∗ (σ)
= −∂ H̄ B(σ) (D∗ (∂(σ))) − ∂ H̄ B(σ) (c(σ))
= H̄ B(σ) (∂(D∗ (∂σ))) − D∗ (∂(σ)) − ∂ H̄ B(σ) (c(σ)) [by Lemma 4.4]
B(σ) B(σ) B(σ) 2
= H̄ (f (∂(σ))) − H̄ (c(∂σ))) − H̄ (D∗ (∂ (σ)))
B(σ)
− D∗ ∂(σ) + H̄ (∂(c(σ))) − c(σ) [by induction hypothesis]
B(σ)
= H̄ (f (∂(σ))) − c(σ) − D∗ (∂(σ))
= f (σ) − c(σ) − D∗ ∂(σ)

Proof of (2): Note that vertices of simplices of |f (σ)| and |D(σ)| live in Nρ(diam(B(σ)) (B(σ)).
As observed earlier, diam(B(σ)) depends only on dim(σ) and diam(σ). That means, we
can find a non-decreasing sequence of function ρn : [0, ∞) → [0, ∞) such that |f (σ n )| ≤
ρn (diam(σ n )) and |D(σ n )| ≤ ρn (diam(σ n )). The claim follows.

Proof of (3): To prove (3), we will take a closer look at the support of D(σ). Consider the
following set
P(σ) := {(σ n , σ n−1 , . . . , σ j ) | σ j ⊂ σ j+1 ⊂ · · · σ n−1 ⊂ σ n = σ, 1 ≤ j ≤ n}
For each s = (σ n , . . . , σ j ) ∈ P(σ) and σ j = (x0 , . . . , xj ), consider the following simplex
n n−1 ) j
Ds (σ) := H̄ B(σ ) (H̄ B(σ (. . . (H̄ B(σ ) (c(σ j ))) . . . )).
We can check that
|D(σ)| = ∪s∈P(σ) |Ds (σ)|
If s = (σ n , . . . , σ j ) ∈ P(σ) and σ j = (x0 , . . . , xj ), then Ds (σ) is supported in the following
set
j j
B(σ n ) B(σ n−1 ) B(σ j )
X X
{ si · Htn (Htn−1 (..(Htj (xi )))..)} | si = 1, (tn , . . . , tj ) ∈ [0, 1]n−j+1 }
i=0 i=0

Let
B(σ n ) B(σ n−1 ) B(σ j )
rs (σ) = max n−j+1
{diam Htn (Htn−1 (..(Htj {v(σ j )})}},
(tn ,...,tj )∈[0,1]

where v(σ j ) denotes the set of vertices of σ j . Since all the homotopies HtB take place in X,
it follows that |Ds (σ)| ⊂ Nrs (σ) (X).
Therefore to prove (3), it is enough to show that there is a bounded set Z so that for any
ϵ > 0 and k ∈ N, we can choose small neighborhood Wx of x ∈ / Z so that for any simplex
n+1
σ ∈ Wx , n ≤ k, s ∈ P(σ), we have rs (σ) < ϵ.
Recall that we picked a locally finite cover U of X at the beginning of the proof. Hence
there are only finitely many Uα ∈ U containing any given x ∈ X. Let Vx be the intersection
of those finitely many Uα . Then for any 1-simplex σ ∈ Vx2 , B(σ) is one of such B(Uα ) by
construction. In particular #{B(σ) | σ ∈ Vx2 } is finite. Note that for any simplex σ, B(σ) is
 ON THE COMPUTATION OF COARSE COHOMOLOGY 17

determined by ∪τ ∈S B(τ ) and dim(σ), where S is the set of 1-subsimplices of σ. This implies
that, for any given k ∈ N, the set {B(σ n ) | σ n ∈ Vxn+1 , n ≤ k} is finite.
This implies that for any ϵ > 0, we can take small enough neighborhood Wx ⊂ Vx of
B(σ n ) B(σ n−1 ) B(σ j )
every x ∈ / Z, such that diam{Htn (Htn−1 ..(Htj (Wx ))..)} < ϵ for all σ ∈ Wxn+1 , for
all (σ n , . . . , σ j ) ∈ P(σ), (tn , . . . , tj ) ∈ [0, 1]n−j+1 and n ≤ k. That implies rs (σ) < ϵ for all
s ∈ P(σ), σ ∈ Wxn+1 , n ≤ k, and x ∈ / Z. That finishes the proof.
□
In the next lemma we use the map f and D from Lemma 4.3 to obtain a filling CF∗ (X) →
Cs∗ (N (X)) that sends small enough simplices to their convex fillings, in particular to small
singular simplices.
Lemma 4.5. Suppose X ⊂ ℓ∞ is uniformly contractible at infinity. Let N (X) be an open
neighborhood of X in ℓ∞ . Then there exists two non-decreasing sequence of control functions
ρn , µn : R≥0 → R≥0 and a subcomplex CF∗ (X) of C∗ (X) of the following form
CFn (X) = ⟨σ n | d(σ n ; b) > µn (diam(σ n ))⟩ ⊂ Cn (X),
a chain map g : CF∗ (X) → Cs∗ (N (X)) with the following properties.
(1) |g(σ n )| ⊂ Nρn (diam(σn )) (σ n ) where the tubular neighborhood is taken in ℓ∞ .
(2) There exists a bounded set Z ⊂ X such that for each x ∈ X − Z, there is an open
neighborhood Wx of x in X such that g(σ) = c(σ) for all σ ∈ Wxn+1 with n ≤ k.
Proof. Let f , D, Z and Wx be as in the Lemma 4.3. We define g(σ) := c(σ) if σ is supported
in Wx for all x ∈ X − Z. If a i-simplex σ i is not in Wxi+1 for any x but there is a (i − 1)-
subsimplex σ ′ of σ that lives in some Wxi , then we define g(σ) := f (σ) − [σ : σ ′ ]D(σ ′ ) (see
figure 5), where [σ : σ ′ ] = ±1 tells us whether orientations of σ and σ ′ agree or not. For such
σ, we will now check that g(∂σ) = ∂g(σ). For that we will use the fact that D is a chain
homotopy between f and c. We first observe the following.
∂g(σ) = ∂f (σ) − [σ : σ ′ ]∂D(σ ′ )
= f (∂σ) + [σ : σ ′ ]D(∂σ ′ ) − [σ : σ ′ ]f (σ ′ ) + [σ : σ ′ ]c(σ ′ )

Now we will show that the above is the same as g∂(σ). First we observe that any (i − 1)-
subsimplex τ ̸= σ ′ of σ has a maximal (i − 2)-subsimplex τ ′ supported in V . Hence for such
τ , we have g(τ ) = f (τ ) − [τ : τ ′ ]D(τ ′ ). We use this to obtain the following
X
g(∂(σ)) = [σ : τ ]g(τ ) + [σ : σ ′ ]g(σ ′ )
τ ∈∂σ
τ ̸=σ ′
X
= [σ : τ ]g(τ ) + [σ : σ ′ ]c(σ ′ )
τ ∈∂σ
τ ̸=σ ′
X X
= [σ : τ ]f (τ ) − [σ : τ ][τ : τ ′ ]D(τ ′ ) + [σ : σ ′ ]c(σ ′ )
τ ∈∂σ τ ∈∂σ
τ ̸=σ ′ τ ̸=σ ′
X
= f (∂σ) − [σ : σ ′ ]f (σ ′ ) − [σ : τ ][τ : τ ′ ]D(τ ′ ) + [σ : σ ′ ]c(σ ′ )
τ ∈∂σ
τ ̸=σ ′
18 ARKA BANERJEE

Figure 5. This is a picture of g(σ) where σ = (x, y, z). Here p is the

contracting point of the homotopy H B((x,y)) . The grey part is f (σ) and striped
part is D(σ ′ ) where σ ′ = (x, y)

To get ∂g(σ) = g(∂σ), it is now enough to show that

X
[σ : σ ′ ]D(∂σ ′ ) = − [σ : τ ][τ : τ ′ ]D(τ ′ )
τ ∈∂σ
τ ̸=σ ′

For that, we need to show that for a common (i−2)-subsimplex β of σ ′ and τ ′ , [σ : σ ′ ][σ ′ : β] =
−[σ : τ ][τ : β]. This is true because coefficient of β in ∂(∂σ) is [σ : σ ′ ][σ ′ : β] + [σ : τ ][τ : β]
and so this has to be zero.
In other cases, an i-simplex σ has some maximal subsimplices {σ ′ } in Wx for some x with
dimension less than i − 1. In this case, by induction on dimension, we can see that g(∂σ)
deforms retract to f (∂σ) = ∂(f (σ)) by deforming along D(σ ′ ) for each maximal subsimplex
σ ′ supported in V . Then we use f (σ) to fill this deformed g(∂σ). This process gives a cell
whose boundary is g(∂σ) and we define g(σ) to be that simplex.
Since |f (σ n )| ⊂ Nρn (diam(σn )) (σ n ) and |D(σ n )| ⊂ Nρn (diam(σn )) (σ n ) by Lemma 4.3, it follows
that |g(σ n )| ⊂ Nρn (diam(σn )) (σ n ) giving us property (1).
Finally, property (2) follows directly from the construction of g. □
We are now ready to prove 4.2. For the convenience of the reader, we recall the statement
first.
Proposition 4.6. Suppose X ⊂ ℓ∞ is uniformly contractible at infinity. Let N (X) be an
open neighborhood of X in ℓ∞ . Let U be an open cover of N (X). Then there exist two
non-decreasing sequence of control functions ρn , µn : R≥0 → R≥0 and a subcomplex CF∗ (X) of
 ON THE COMPUTATION OF COARSE COHOMOLOGY 19

C∗ (X) of the following form

CFn (X) = ⟨σ n | d(σ n ; b) > µn (diam(σ n ))⟩ ⊂ Cn (X),
a chain map S : CF∗ (X) → CU∗ (N (X)) and a map G : CF∗ (X) → C∗+1 (N (X)) with the
following properties.
(1) G is a chain homotopy between V S and the inclusion map i : CF∗ (X) ,→ C∗ (N (X)).
∂G(σ) = V S(σ) − i(σ) − G∂(σ)
(2) |G(σ n )| ⊂ Nρn (diam(σn )) (σ n ).
(3) For any k ∈ N, there exists a bounded set B ⊂ X such that for any x ∈ X − B
and a neighborhood U of x in ℓ∞ , there is a neighborhood W of x in X such that
|G(σ n )| ⊂ U n+2 for all σ n ∈ W n+1 if n ≤ k.
Proof. Let us first define the maps S and G. We take the filling g : CF∗ (X) → Cs∗ (N (X)) from
the Lemma 4.5 and then apply barycentric subdivision on g to define S : CF∗ (X) → CU∗ (N (X)).
Since |g(σ n )| ⊂ Nρn (diam(σn )) (σ n ) by Lemma 4.5, it follows that |S(σ n )| ⊂ Nρn (diam(σ)) (σ n ) and
consequently |V S(σ n )| ⊂ Nρn (diam(σ)) (σ n ). Applying the Lemma 3.3 to the map V S, we get
the homotopy G between V S and the inclusion map i : CF∗ (X) → C∗ (N (X)) with property
(2). Property (3) follows directly from the property (2) of the map g from Lemma 4.5. □
Recall that the Proposition 3.4 was the main ingredient to prove the Theorem 3. Likewise
Proposition 4.6 is ‘almost’ enough to prove Theorem 4.1. The only caveat is that the maps
S and G in Proposition 4.6 take images in C∗ (N (X)) instead of C∗ (X) (compare with
Proposition 3.4). Fortunately, thin enough neighborhoods of X are as good as X from
the point of view of boundedly supported cohomology. More precisely, we will show (in
Proposition 4.9) that H B∗ (X) is taut in the sense that any class in H B∗ (X) is a restriction
of a class from H B∗ (N (X)) for some neighborhood N (X) of X inside ℓ∞ .
Let U be a collection of open sets in X and A be a subset of X. The star of A with respect
to U, denoted by st(A, U), is defined to be the union of those elements of U whose intersection
with A is nonempty. An open covering of A in X is a collection U of open sets of X such
that A ⊂ st(A, U). We now state the following lemma from [8] which roughly says that there
is a projection map from st(X, V) to V that does not move close points too far apart.
Lemma 4.7 ([8]). If X ⊂ ℓ∞ , then for every open covering U of X in ℓ∞ , there is an open
covering V of X in ℓ∞ and a function f : st(X, V) → X such that
(1) f (x) = x for all x ∈ X
(2) For each V ∈ V with V ∩ X ̸= ∅ there is a U ∈ U such that V ∪ f (V ) ⊂ U .
Definition 4.8. A map f : X → Y between metric spaces is called coarse map, if inverse
image of a bounded set is bounded and there exist a non decreasing function ρ : R+ → R+
such that
d(f (x), f (y)) ≤ ρ(d(x, y))
The next proposition basically says that a boundedly supported cochain in X can be
extended to a boundedly supported cochain in N (X) for an appropriate neighborhood N (X)
of X in ℓ∞ .
Proposition 4.9. Let X be a subspace of ℓ∞ . Then for any ϕ ∈ C B∗ (X), there exists
a neighborhood N (X) of X in ℓ∞ and a coarse map f : N (X) → X such that f ∗ (ϕ) ∈
C B∗ (N (X)) and i∗ f ∗ (ϕ) = ϕ where i : X → N (X) is the inclusion.
20 ARKA BANERJEE

Proof. The proof is similar to the proof of tautness of Alexander-Spanier cohomology [8].
Suppose, ϕ ∈ C B∗ (X). That means there is a bounded set B such that for any a ∈ X − B,
we can choose a neighborhood Ua so that Ua∗+1 ∩ |ϕ| = ∅. Cover B with a bounded open set
UB in X. Let U = UB ∪ {Ua }a∈X−B be the open cover of X. We can choose these open sets
so that their diameters are less than r for some r ≥ 0. Then Lemma 4.7 yields a refinement
of V of U, and a map f : st(X, V) → X such that f (a) = a for all a ∈ X and for each V ∈ V
with V ∩ X ̸= ∅, there is a U ∈ U such that V ∪ f (V ) ⊂ U .
By the second property of f in Lemma 4.7, it follows that d(x, f (x)) ≤ r for all x ∈
st(X, V). Hence, d(f (x), f (y)) ≤ 2r + d(x, y) by triangle inequality which means f has an
upper control function. Furthermore f sends unbounded sets to unbounded sets because
d(y, f (x)) ≥ d(y, x) − r by triangle inequality. Hence f is a coarse map.
We let N (X) := st(X, V). To show that f ∗ (ϕ) ∈ C B∗ (N (X)), we notice that if an element
V ∈ V is far from UB , then V ∪ f (V ) does not touch UB , and hence f ∗ (ϕ)|V ⊂ Ua for some
a ∈ X − B. Finally i∗ f ∗ (ϕ) = ϕ because f |X = idX . □
Now we are ready to prove our main theorem which basically mimics the proof of Theo-
rem 3.1.
Proof of 4.1. By the long exact sequence, we need to show H∗ (C B∗ (X)/ C X∗ (X)) = 0.
In other words, for ϕ ∈ C Bn (X) with dϕ ∈ C Xn+1 (X) we need to find ψ ∈ C Bn−1 (X) so
that ϕ − dψ ∈ C Xn . By Kuratowski’s embedding theorem, X can be embedded inside ℓ∞ .
Proposition 4.9 produces a neighborhood of N (X) and a coarse map f : N (X) → X such
that ϕ′ = f ∗ (ϕ) ∈ C B∗ (N (X)). Let U be a bounded neighborhood of ∥ϕ′ ∥ in X and for
each x ∈ X − ∥ϕ′ ∥ choose a neighborhood Ux such that Uxn+1 ∩ |ϕ| = ∅. Let U denote the
collection of Ux together with U . Moreover, we have d(ϕ′ ) = d(f ∗ (ϕ)) = f ∗ (d(ϕ)) is coarse
since d(ϕ) is coarse and f is a coarse map.
Proposition 4.2 produces the chain homotopy G : CF∗ (X) → C∗+1 (N (X)), which we use to
define a linear map D : C∗ (X) → C∗+1 (N (X)) by setting
(
G(σ n ) if σ n ∈ CFn (X)
D(σ n ) =
0 otherwise
Let i : C∗ (X) → C∗ (N (X)) be the inclusion map. We now define
T = i − ∂D − D∂.
And dually

T ∗ = i∗ − D∗ d − dD∗
By construction
ϕ − dD∗ (ϕ′ ) = i∗ (ϕ′ ) − dD∗ (ϕ′ ) = T ∗ (ϕ′ ) + D∗ d(ϕ′ ).
Thus ψ = D∗ (ϕ′ ) would be the desired cochain if we can prove T ∗ (ϕ′ ) and D∗ d(ϕ′ ) are coarse
cochains and D∗ (ϕ′ ) is a boundedly supported cochain.
We claim that T ∗ (ϕ′ ) is coarse. By Proposition 4.2(1), T (σ) = V S(σ) for all σ ∈ CFn (X).
Let σ n be of some fixed diameter r. If d(σ n , b) > µn (r), then σ n ∈ CFn (X) and it follows that
T (σ) is supported by U. If all vertices of σ n are outside of ρn (r)-neighborhood of U , then
|T (σ)| does not meet U n+1 and therefore T ∗ (ϕ′ )(σ) = ϕ′ (T (σ)) = 0, since ϕ|Uxn+1 = 0 for all
Ux . Since U is bounded, the claim follows.
 ON THE COMPUTATION OF COARSE COHOMOLOGY 21

We claim that D∗ d(ϕ′ ) is coarse. Recall from Lemma 4.2(2) that |Gσ| ⊂ Nρn (diam(σ)) (σ).
Therefore, by construction |Dσ| ⊂ Nρn (diam(σ)) (σ). So, D∗ preserves coarseness, and the claim
follows since d(ϕ′ ) is coarse.
Finally, we claim that D∗ (ϕ′ ) ∈ C B∗+1 (X). Because of the property 4.2(3) of G, we can
choose a neighborhood V of x for all x ∈ / ||ϕ′ || except points in some bounded set B, such
that D∗ (V ∗+1 ) does not intersect |ϕ′ |. That implies x ∈ / ||D∗ ϕ′ || for all x ∈
/ ||ϕ′ || ∪ B. Since
′ ∗
ϕ ∈ C B (N (X)), the claim follows.
□
Combining Theorem 4.1 and Proposition 2.3 we get the following.
Corollary 4.10. If X is unbounded, uniformly contractible at infinity, then for any b ∈ X
(
0 if ∗ = 0
H X∗ (X; R) = ∗−1
lim H̃ (X − Nr (b); R) otherwise
−→
5. Computation of Coarse cohomology of the complement
We start by briefly reviewing the notion of coarse complement from [1]. Roughly the idea
is that coarse complement of a subset A in X is determined by the collection of subsets S of
X which are coarsely disjoint from A:
S := {B ⊂ X | Nr (B) ∩ Nr (A) is bounded for any r ≥ 0}
Coarse cohomology of the complement of A is defined to be the cohomology of the following
complex
C Xn (X − A) = {ϕ ∈ Cn (X) | ∀B ∈ S ϕ|B ∈ C Xn (B)}
with the usual coboundary operator d. We denote the cohomology of the above complex by
H X∗ (X − A). It was shown in [1] that H X∗ (X − A) can be computed as coarse cohomology
of a single space which we now recall from [1]. Consider the following pseudometric dA on X.
dA (x, y) := min{d(x, A) + d(y, A), d(x, y)}
Note that dA (x, y) = 0 if and only if (x, y) ∈ Ā × Ā where Ā is the closure of A in X. Hence
the pseudometric dA becomes a metric on the quotient space X/Ā.
Proposition 5.1 ([1]). The quotient map q : X → X/Ā induces isomorphism H X∗ (X − A) ∼ =
∗
H X (X/Ā).
Proposition 5.1 allows us to use Theorem 4.1 to compute coarse cohomology of the
complement.
Theorem 5.2. If X/Ā is uniformly contractible at infinity and unbounded, then
(
0 if ∗ = 0
H X∗ (X − A) = ∗−1
lim H̃ (X − Nr (A)) otherwise
−→
Proof. If X/Ā is uniformly contractible at infinity, then Theorem 4.1 yields H X∗ (X/Ā) =
H B∗ (X/Ā). Combining this with Proposition 5.1 gives us H X∗ (X − A) = H B∗ (X/Ā).
Since X/Ā is unbounded, for any b ∈ X/Ā, Proposition 2.3 gives us the following
(
0 if ∗ = 0
H B∗ (X/Ā) = ∗−1
lim H̃ (X − Nr (A)) otherwise
−→
22 ARKA BANERJEE

Finally we observe that lim H̃∗ (X/Ā − Nr (b)) = lim H̃∗ (X − Nr (A)). That finishes the
−→ −→
proof. □
X is called uniformly contractible away from A if there are two non-decreasing function
µ, ρ : [0, ∞) → [0, ∞) such that any ball B(x, r) of radius r centered at x ∈ X with
d(x, A) ≥ µ(r) is contractible inside B(x, ρ(r)).
Let q : X → X/Ā be the quotient map. We observe that the ball Br (x) in (X, d) with
center at x is isometric to the ball Br (q(x)) in (X/Ā, dA ) if d(x, A) > 2r. That means, if X
is uniformly contractible away from A, then so is X/Ā. If X − Nr (A) ̸= ∅ for any r, then
X/Ā id unbounded. Hence, as a consequence of Theorem 5.2, we get the following.
Corollary 5.3. If X is uniformly contractible away from A, and X − Nr (A) ̸= ∅ for any r,
then
(
0 if ∗ = 0
H X∗ (X − A) = ∗−1
lim H̃ (X − Nr (A)) otherwise
−→
References
[1] Arka Banerjee and Boris Okun. Coarse cohomology of the complement. arXiv 2308.13965, 2023.
[2] M. Kapovich and B. Kleiner. Coarse Alexander duality and duality groups. J. Differential Geom. 69
(2):279–352, 2005.
[3] C. Kuratowski. Quelques problèmes concernant les espaces métriques non-séparables. Fundamenta Mathe-
maticae 25 (1):534-545, 1935.
[4] L. Mosher, M. Sageev, and K. Whyte. Quasi-actions on trees II: Finite depth Bass-Serre trees. Mem.
Amer. Math. Soc. 214 (1008):vi+105, 2011.
[5] J. Roe. Coarse cohomology and index theory on complete Riemannian manifolds. Mem. Amer. Math. Soc.
104 (497):x+90, 1993.
[6] J. Roe. Lectures on coarse geometry. University Lecture Series, vol. 31. American Mathematical Society,
Providence, RI, 2003.
[7] E. H. Spanier. Algebraic topology. Springer-Verlag, New York-Berlin, 1981.
[8] E. H. Spanier. Tautness for Alexander-Spanier cohomology. Pacific Journal of Mathematics 75 (2):561 –
563, 1978.

Department of Mathematics and Statistics, Auburn University, Auburn, AL 36849

Email address: [email protected]