Homology of Configuration Spaces

Bu Chenjing

ABSTRACT
We propose a method of computing the (co)homology of configuration
spaces of manifolds via a spectral sequence, and we describe this spectral se-
quence explicitly for elliptic curves.

Contents

1 Configuration spaces of euclidean spaces 2

2 Configuration spaces of manifolds 5

3 Configuration spaces of elliptic curves 10

A Proof of Theorem 3.11 18

1
Definition 0.1. Let 𝑋 be a topological space, and let 𝑛 be a non-negative integer.
The configuration space of 𝑛 points in 𝑋 is the space

Conf𝑛 (𝑋) = {𝑥1 , … , 𝑥𝑛 ∈ 𝑋 ∣ 𝑥𝑖 ≠ 𝑥𝑗 ∀𝑖 ≠ 𝑗},

equipped with the topology as an open subspace of 𝑋 𝑛 .

We aim to describe the (co)homology of Conf𝑛 (𝑋) when 𝑋 is a manifold.
Notation 0.2. Throughout this paper, when we mention the (co)homology of a
space, we always assume that a coefficient field 𝕜 with characteristic zero is chosen,
and we write 𝐻• (𝑋) for 𝐻• (𝑋; 𝕜), etc.

1 Configuration spaces of euclidean spaces

The homology of Conf𝑛 (ℝ𝑑 ) is well understood, and here we give a brief summary
of the results. For more details, the reader is referred to, for example, [Sin06].
Definition 1.1. Let 𝑛 ≥ 1 be an integer. An 𝑛-tree is a binary tree with 𝑛 leaves.
For example, and are the only possible 3-trees. For an 𝑛-tree 𝑇 , denote

|𝑇 | = 𝑛 − 1,

which is the number of internal vertices of 𝑇 . These internal vertices are canon-
ically ordered, with the ordering defined by traversing the tree in the order ‘left–
parent–right’. ⊲
Definition 1.2. Let 𝑛 ≥ 0 be an integer. An 𝑛-forest is a sequence of trees with 𝑛
leaves in total, with the leaves labelled with 1, … , 𝑛, without repeating. For a forest
𝐹 = (𝑇1 , … , 𝑇𝑚 ), let
𝑚
|𝐹 | = ∑ |𝑇𝑖 | = 𝑛 − 𝑚.
𝑖=1

The |𝐹 | internal vertices are ordered in the same way as in Definition 1.1, with the
vertices of the first tree coming first, and then the second tree, and so on. ⊲

For example, 𝐹 = 2 43 1
is a 4-forest, with |𝐹 | = 2.

Theorem 1.3. Let 𝑑 ≥ 2 be an integer. Then the homology 𝐻• (Conf𝑛 (ℝ𝑑 )) is the
quotient of the free module generated by all 𝑛-forests by the following relations:
• (anti-symmetry) For any trees 𝑇1 , 𝑇2 , one has

𝑇1 𝑇2 𝑇2 𝑇1
= −(−1)(𝑑−1)(|𝑇1 |+1)(|𝑇2 |+1) ,
𝑅 𝑅

where 𝑅 can be the root, or can be attached to any leaf of a tree.

2
 • (Jacobi identity) For any trees 𝑇1 , 𝑇2 , 𝑇3 , one has

𝑇1 𝑇2 𝑇2 𝑇3 𝑇3 𝑇1
𝑇3 𝑇1 𝑇2
+ + = 0,
𝑅 𝑅 𝑅

where 𝑅 can be the root, or can be attached to any leaf of a tree.

• (orientation) For any 𝑛-forest 𝐹 = (𝑇1 , … , 𝑇𝑚 ) and any 𝜎 ∈ 𝔖𝑚 , one has
′
(𝑇𝜎(1) , … , 𝑇𝜎(𝑚) ) = (−1)(𝑑−1)𝜎 (𝑇1 , … , 𝑇𝑚 ),

where 𝜎 ′ ∈ 𝔖|𝐹 | is the induced permutation of the internal vertices of 𝐹 .

The homology class associated to a forest 𝐹 has homological degree (𝑑 − 1)|𝐹 |.
The generators of 𝐻• (Conf𝑛 (ℝ𝑑 )) can be explicitly written down as ‘orbital
systems’. For example, the homology class

2 41 5 3
𝐹 = ∈ 𝐻3(𝑑−1) (Conf5 (ℝ𝑑 ))

is defined by the map (𝑆 𝑑−1 )3 → Conf5 (ℝ𝑑 ) depicted as follows, with 𝑑 = 2.

2 (1) 4 (2) 1 5 (3) 3

The ordering of the factors 𝑆 𝑑−1 is shown as numbers with parentheses, and is
determined by the ordering of internal vertices (Definition 1.2).
In fact, tree diagrams are closely related to Lie brackets. The previous theorem
readily implies the following.
Corollary 1.4. The space

Tree±𝑛 = (linear span of 𝑛-trees)∕∼,

where ∼ denotes anti-symmetry and the Jacobi identity, and ± is equal to (−1)𝑑−1 ,
is isomorphic to the space

Lie±𝑛 = linear span of ways to take Lie brackets of 𝑛 elements,

where the 𝑛 elements are even (+) or odd (−), and each of the 𝑛 elements is required
to be used exactly once. Let

𝑐 ∶ Tree±𝑛 ≃ Lie±𝑛

denote this isomorphism. ◻

3
 For example,
3 1
𝑐( 2
) = [[3, 1], 2] = 312 ∓ 132 − 231 ± 213.
A richer version of Corollary 1.4 is stated as follows, which we do not attempt
to prove here.
Corollary 1.5. The homology of the little disks operad 𝐸𝑛 is isomorphic to the
Poisson operad 𝑃𝑛 .
Next, let us describe the cohomology ring 𝐻 • (Conf𝑛 (ℝ𝑑 )).
Definition 1.6. Let 𝑛 ≥ 0 be an integer. By an 𝑛-graph, we mean a directed graph
with 𝑛 vertices, which are labelled 1, … , 𝑛 without repeating, and with a specified
ordering of its edges, such that it is acyclic in the sense that the induced undirected
graph contains neither cycles nor multi-edges. For an 𝑛-graph 𝛤 , denote
|𝛤 | = #edges(𝛤 ). ⊲
• 𝑑
Theorem 1.7. Let 𝑑 ≥ 2 be an integer. Then the cohomology 𝐻 (Conf𝑛 (ℝ )) is
the quotient of the free module generated by all 𝑛-graphs by the following relations:
• (anti-symmetry)
𝑗 𝑗
𝑖
= (−1)𝑑 𝑖
,

where 𝑖, 𝑗 may be connected to other edges as well.

• (Arnold identity)
𝑗 𝑗 𝑗
𝑖 𝑘
+ 𝑖 𝑘
+ 𝑖 𝑘
= 0,

where 𝑖, 𝑗, 𝑘 may be connected to other edges as well.

• (orientation) If 𝛤 ′ is obtained from 𝛤 by reordering the edges using a per-
mutation 𝜎 ∈ 𝔖|𝛤 | , then
𝛤 ′ = (−1)(𝑑−1)𝜎 𝛤 .

The cohomology class associated to a graph 𝛤 has cohomological degree (𝑑 −

1)|𝛤 |. The cup product is given by taking the disjoint union of edges (if this does
not make a valid 𝑛-graph, then the product is zero).
𝑗
The elements 𝑖
form a set of generators of the cohomology ring. These
generators can be expressed explicitly as the pullback of the fundamental class of
𝑆 𝑑−1 along the map
𝛼𝑖𝑗 ∶ Conf𝑛 (ℝ𝑑 ) → 𝑆 𝑑−1 ,
𝑥𝑖 − 𝑥 𝑗
(𝑥1 , … , 𝑥𝑛 ) ↦ .
|𝑥𝑖 − 𝑥𝑗 |
Finally, we describe the cohomology–homology pairing in terms of forests and
graphs.

4
Theorem 1.8. Let 𝐹 be an 𝑛-forest, and let 𝛤 be an 𝑛-graph. If there exists a
bijection
𝑓 ∶ {edges of 𝛤 } ≃ {internal vertices of 𝐹 },
such that for each edge 𝑒 = (𝑖 → 𝑗) of 𝛤 , the leaves 𝑖 and 𝑗 appear in the left and
right (resp. right and left) branches of the vertex 𝑓 (𝑒), respectively, then putting
𝜀(𝑒) = 1 (resp. (−1)𝑑 ), we have
⟨𝛤 , 𝐹 ⟩ = (−1)(𝑑−1)𝜎 ∏ 𝜀(𝑒),
𝑒

where 𝜎 ∈ 𝔖|𝐹 | is the permutation which relates the ordering of edges of 𝛤 to the
ordering of internal vertices of 𝐹 . If such 𝑓 does not exist, then ⟨𝛤 , 𝐹 ⟩ = 0.
Proof. This follows directly from the explicit expressions of the generators of the
(co)homology groups in terms of forests and graphs. ◻
Note that if such 𝑓 exists, then 𝑓 (𝑒) is uniquely determined, as for any two
leaves in a binary tree, there is a unique internal vertex whose left and right branches
each contain one of the two given leaves.

2 Configuration spaces of manifolds

In order to compute the homology of configuration spaces of manifolds, we use
handle theory to break down the manifold into simpler pieces.
Let us first briefly recall the basics of handle theory.
Definition 2.1. For a fixed dimension 𝑑, and for 0 ≤ 𝑟 ≤ 𝑑, an 𝑟-handle is the
manifold with corners
ℎ𝑟 = 𝐷𝑟 × 𝐷𝑑−𝑟 ,
which is to be regarded as a thickened version of the 𝑟-disk. A 𝑑-dimensional
handlebody is a sequence of manifolds (possibly with boundaries)
∅ = 𝑋−1 ⊂ 𝑋0 ⊂ 𝑋1 ⊂ ⋯ ,
either finite or infinite, such that each 𝑋𝑖 is obtained from 𝑋𝑖−1 by attaching a
handle:
𝑋𝑖 = 𝑋𝑖−1 ∪Φ𝑖 ℎ𝑟 ,
where
Φ𝑖 ∶ 𝜕𝐷𝑟 × 𝐷𝑑−𝑟 → 𝜕𝑋𝑖−1
is a smooth map, called the attaching map. The topological space
𝑋 = colim 𝑋𝑖
𝑖→∞

is the total space of the handlebody. We require local finiteness, which states that
every point of 𝑋 has a neighbourhood which only intersects with the images of
finitely many attaching maps. This ensures that 𝑋 has an induced smooth structure.
By abuse of language, we call 𝑋 a 𝑑-dimensional handlebody. ⊲

5
 It is a fundamental result in handle theory that every smooth manifold has a
handle decomposition, that is, the structure of a handlebody.
We will need our handlebodies to satisfy an extra technical condition, in order
to simplify our arguments later.
Definition 2.2. A handlebody 𝑋 is said to be perfect, if the image of each attaching
map
Φ𝑖 ∶ 𝜕𝐷𝑟 × 𝐷𝑑−𝑟 → 𝜕𝑋𝑖−1

for an 𝑟-handle is entirely contained in the boundaries of previously attached (𝑟−1)-

handles, and for each 𝑧 ∈ 𝜕𝐷𝑑−𝑟 , the (𝑟 − 1)-dimensional slice

Φ𝑖 (𝜕𝐷𝑟 × 𝑧) ⊂ 𝑋𝑖−1

is a finite union of slices of the form 𝐷𝑟−1 × 𝑧′ ⊂ ℎ𝑟−1 , where ℎ𝑟−1 is a previously
attached (𝑟 − 1)-handle, and 𝑧′ ∈ 𝜕𝐷𝑑−𝑟+1 . ⊲
In fact, every smooth manifold has the structure of a perfect handlebody, which
may be constructed from a triangulation. We omit the proof of this fact, which we
will not use anyway.
In the following, let 𝑋 be a perfect handlebody.
Definition 2.3. Let
Conf′𝑛 (𝑋) ⊂ Conf𝑛 (𝑋)

be the subspace of Conf𝑛 (𝑋) consisting of the configurations 𝑥 = (𝑥1 , … , 𝑥𝑛 ), such

that for every handle ℎ𝑟 = 𝐷𝑟 × 𝐷𝑛−𝑟 ⊂ 𝑋, and any 𝑧 ∈ 𝐷𝑛−𝑟 , the intersection
𝑥 ∩ (𝐷𝑟 × 𝑧) contains at most one point. ⊲
For example, if 𝑥 ∈ Conf′𝑛 (𝑋), then every top dimensional handle of 𝑋 is
allowed to contain at most one point of 𝑥.
Lemma 2.4. Conf′𝑛 (𝑋) is homotopy equivalent to Conf𝑛 (𝑋).
Proof. Let 𝑋̊ denote the interior of 𝑋. Then Conf𝑛 (𝑋)̊ is exactly the interior of
the manifold Conf𝑛 (𝑋) (possibly with corners). Therefore, Conf𝑛 (𝑋) is homo-
topy equivalent to Conf𝑛 (𝑋), ̊ and Conf′𝑛 (𝑋) is homotopy equivalent to Conf′𝑛 (𝑋) ∩
̊
Conf𝑛 (𝑋), which we abbreviate as Conf′𝑛 (𝑋). ̊
̊
Let 𝑈𝑟 ⊂ Conf𝑛 (𝑋) be the subspace such that the defining criterion of Conf′𝑛 (𝑋)
̊ where 𝑑 = dim 𝑋,
is satisfied for all handles of type ≥ 𝑟. Thus, 𝑈𝑑+1 = Conf𝑛 (𝑋),
′ ̊
and 𝑈0 = Conf𝑛 (𝑋). It suffices to show that 𝑈𝑟+1 is homotopy equivalent to 𝑈𝑟 for
𝑟 = 0, 1, … , 𝑑.
For each handle ℎ𝑟 of 𝑋, let 𝑌 be the vector field on ℎ𝑟 defined by

𝑌 = ∇(|𝑎|2 ) for (𝑎, 𝑏) ∈ 𝐷𝑟 × 𝐷𝑑−𝑟 ,

where the gradient is taken in euclidean space. By choosing an appropriate smooth

structure on 𝑋 (so that when two 𝑟-handles intersect at the boundary of an (𝑟 − 1)-
handle, their vector fields 𝑌 coincide on their intersection), the vector fields 𝑌

6
extend to a globally defined vector field 𝑌 on 𝑋. Using an appropriate Riemannian
metric on 𝑋, the norm of 𝑌 is bounded and 𝑋 is complete, so that the flow 𝜑𝑡 of
𝑌 is defined for all 𝑡 ∈ ℝ.
We observe that for any 𝑥 ∈ 𝑈𝑟+1 , the flow 𝜑𝑡 eventually takes 𝑥 into 𝑈𝑟 .
Precisely speaking, for any 𝑥 ∈ 𝑈𝑟+1 , there exists a unique 𝑇 ≥ 0, such that for any
𝑡 > 0, 𝜑𝑡 (𝑥) is in 𝑈𝑟 if and only if 𝑡 > 𝑇 . We define

𝜌 ∶ 𝑈𝑟+1 → 𝑉𝑟 , 𝑥 ↦ 𝜑𝑇 (𝑥),

where 𝑇 depends on 𝑥, and 𝑉𝑟 is the subset of 𝑈𝑟+1 consisting of those config-

urations 𝑥 such that for any 𝑟-handle 𝐷𝑟 × 𝐷𝑑−𝑟 ⊂ 𝑋, and any 𝑧 ∈ 𝐷𝑑−𝑟 , the
intersection 𝑥 ∩ (𝐷̊ 𝑟 × 𝑧) contains at most one point, where 𝐷̊ 𝑟 denotes the interior
of 𝐷𝑟 . Then 𝜌 is a deformation retraction map, and hence, a homotopy equivalence.
Finally, we need to show that 𝑉𝑟 is homotopy equivalent to 𝑈𝑟 . This follows
from the fact that 𝑉𝑟 is a (topological) manifold with boundary, whose interior is
𝑈𝑟 . This can be proved by inspecting the neighbourhood of every point in 𝑉𝑟 , and
it is crucial to consider 𝑋̊ instead of 𝑋. We omit the full argument. ◻
We observe that Conf′𝑛 (𝑋) has the structure of a topologically enriched CW
complex, described as follows.
Definition 2.5. A topologically enriched CW complex consists of the following
data:
• A topological space 𝑋[𝑖], called the space of 𝑖-cells, for 𝑖 = 0, 1, 2, … ;
• A sequence of topological spaces ∅ = 𝑋−1 ⊂ 𝑋0 ⊂ 𝑋1 ⊂ ⋯;
• For each 𝑖, an attaching map Φ𝑖 ∶ 𝑋[𝑖] × 𝜕𝐷𝑖 → 𝑋𝑖−1 ,
such that
• For each 𝑖, one has
𝑋𝑖 = 𝑋𝑖−1 ∪Φ𝑖 (𝑋[𝑖] × 𝐷𝑖 ).

• For each 𝑧 ∈ 𝑋[𝑖], the image

Φ𝑖 (𝑧 × 𝜕𝐷𝑖 ) ⊂ 𝑋𝑖−1

is contained in a finite union of previously attached cells, i.e., a finite union

′
of subsets of the form Φ𝑖′ (𝑧′ × 𝐷𝑖 ), where 𝑖′ < 𝑖 and 𝑧′ ∈ 𝑋[𝑖′ ].
The space
𝑋 = colim 𝑋𝑖
𝑖→∞

is called the total space of the topologically enriched CW complex. ⊲

For example, a perfect handlebody may be seen as a topologically enriched CW
complex, with the space 𝑋[𝑖] being a disjoint union of disks of dimension dim 𝑋−𝑖.
This can be seen as a special case of the following lemma, with 𝑛 = 1.

7
Lemma 2.6. The space 𝐶 = Conf′𝑛 (𝑋) has the structure of a topologically en-
riched CW complex, with homotopy equivalences
∐ (𝑛 ⋯𝑛
𝑚
1 𝑛𝑚 )
𝑑−𝑟𝑗
𝐶[𝑖] ≃ ∐ ∏ Conf𝑛𝑗 (ℝ ) ,
𝑛1 +⋯+𝑛𝑚 =𝑛
( 𝑗=1 )
𝑛1 𝑟1 +⋯+𝑛𝑚 𝑟𝑚 =𝑖

where 𝑚 is the number of handles of 𝑋, each 𝑛𝑗 is a non-negative integer, 𝑟𝑗 is the

type of the 𝑗-th handle of 𝑋, and 𝑑 is the dimension of 𝑋.
Proof. For each choice of 𝑛1 , … , 𝑛𝑚 satisfying the criterion of the coproduct, and
for each choice of 𝑧𝑗 ∈ Conf𝑛𝑗 (𝐷𝑑−𝑟𝑗 ) for 𝑗 = 1, … , 𝑚, we define

𝑒𝑧1 ,…,𝑧𝑚 ∶ (𝐷𝑟1 )𝑛1 × ⋯ × (𝐷𝑟𝑚 )𝑛𝑚 → 𝑋 𝑛

by putting 𝑛𝑗 points on the 𝑗-th handle, according the coordinates given by 𝑧𝑗 and
(𝐷𝑟𝑗 )𝑛𝑗 . Let 𝑍𝑛1 ,…,𝑛𝑚 be the space of all (𝑧1 , … , 𝑧𝑚 ) such that the image of 𝑒𝑧1 ,…,𝑧𝑚
lies in Conf′𝑛 (𝑋) (i.e., the points do not overlap). Then for each 𝑧 ∈ 𝑍𝑛1 ,…,𝑛𝑚 , the
map 𝑒𝑧 gives rise to an 𝑖-cell of Conf′𝑛 (𝑋). It follows that 𝐶[𝑖] is the disjoint union
of all the spaces 𝑍𝑛1 ,…,𝑛𝑚 , each with (𝑛 ⋯𝑛 𝑛 ) copies.
1 𝑚
We notice that
𝑚 𝑚
∏ Conf𝑛𝑗 (𝐷̊ 𝑑−𝑟𝑗 ) ⊂ 𝑍𝑛1 ,…,𝑛𝑚 ⊂ ∏ Conf𝑛𝑗 (𝐷𝑑−𝑟𝑗 ),
𝑗=1 𝑗=1

where 𝐷̊ 𝑑−𝑟𝑗 denotes the interior of 𝐷𝑑−𝑟𝑗 . However, the right hand side is a mani-
fold with corners whose interior is the left hand side, so that 𝑍𝑛1 ,…,𝑛𝑚 is homotopy
equivalent to both spaces. ◻
The (co)homology of a topologically enriched CW complex may be computed
using cellular (co)homology.
Lemma 2.7. Let 𝑋 be a topologically enriched CW complex, such that each 𝑋[𝑖]
is homotopy equivalent to a CW complex. Let
sing
𝐴𝑝,𝑞 = 𝐶𝑝 (𝑋[𝑞])

be the 𝑝-th term of the singular chain complex of 𝑋[𝑞]. Let

𝜕 ∶ 𝐴𝑝,𝑞 → 𝐴𝑝−1,𝑞

be the singular boundary map, and let

𝜕 cell ∶ 𝐴𝑝,𝑞 → 𝐴𝑝,𝑞−1

be the cellular boundary map, defined in the same way as with ordinary CW com-
plexes. Then the total complex of the double complex (𝐴𝑝,𝑞 , 𝜕, 𝜕 cell ) computes the
homology of 𝑋:
𝐻• (𝑋) ≃ 𝐻• (Tot• (𝐴•,• )).

8
Proof. Let
𝑋 ′ [𝑖] = |Sing 𝑋[𝑖]|

be the geometric realisation of the simplicial set of singular simplices of 𝑋[𝑖]. Then
the spaces 𝑋 ′ [𝑖], together with the induced attaching maps, give rise to a topolog-
ically enriched CW complex, which we denote by 𝑋 ′ . Then 𝑋 ′ carries a naturally
induced CW structure (as an ordinary CW complex).
We notice that Tot• (𝐴•,• ) is isomorphic to the cellular chain complex of 𝑋 ′ .
Moreover, we have weak homotopy equivalences 𝑋 ′ [𝑖] ≃ 𝑋[𝑖], induced by the
counit map, which are homotopy equivalences due to Whitehead’s theorem. By
induction on 𝑖, using [MP12, Lemmas 2.1.3], we see that 𝑋𝑖 and 𝑋𝑖′ are homotopy
equivalent. By [MP12, Lemmas 2.1.10], we conclude that 𝑋 and 𝑋 ′ are homotopy
equivalent. ◻

Corollary 2.8. Suppose that each 𝑋[𝑖] has the homotopy type of a CW complex.
Then there exists a spectral sequence
1
𝐸𝑝,𝑞 = 𝐻𝑝 (𝑋[𝑞]) ⇒ 𝐻𝑝+𝑞 (𝑋). ◻

Analogously, one may compute the cohomology of a topologically enriched

CW complex using cellular cohomology.

Lemma 2.9. Let 𝑋 be a topologically enriched CW complex, such that each 𝑋[𝑖]
is homotopy equivalent to a CW complex. Let
𝑝
𝐴𝑝,𝑞 = 𝐶sing (𝑋[𝑞])

be the 𝑝-th term of the singular cochain complex of 𝑋[𝑞]. Let

𝑑 ∶ 𝐴𝑝,𝑞 → 𝐴𝑝+1,𝑞

be the singular coboundary map, and let

𝑑cell ∶ 𝐴𝑝,𝑞 → 𝐴𝑝,𝑞+1

be the cellular coboundary map. Then the total complex of the double complex
(𝐴𝑝,𝑞 , 𝑑, 𝑑cell ) computes the cohomology of 𝑋:

𝐻 • (𝑋) ≃ 𝐻 • (Tot• (𝐴•,• )). ◻

Corollary 2.10. Suppose that each 𝑋[𝑖] has the homotopy type of a CW complex.
Then there exists a spectral sequence
𝑝,𝑞
𝐸1 = 𝐻 𝑝 (𝑋[𝑞]) ⇒ 𝐻 𝑝+𝑞 (𝑋). ◻

Putting these results together, we are able to compute the (co)homology of con-
figuration spaces, using topologically enriched cellular (co)homology.

9
Theorem 2.11. Let 𝑋 be a perfect handlebody. Then there exist spectral sequences

𝑚 ⊕(𝑛 ⋯𝑛
1 𝑛𝑚 )
1 𝑑−𝑟𝑗
𝐸𝑝,𝑞 = 𝐻𝑘𝑗 (Conf𝑛𝑗 (ℝ )) ⇒ 𝐻𝑝+𝑞 (Conf𝑛 (𝑋)),
⨁
𝑛1 +⋯+𝑛𝑚 =𝑛
(⨂ )
𝑗=1
𝑛1 𝑟1 +⋯+𝑛𝑚 𝑟𝑚 =𝑞
𝑘1 +⋯+𝑘𝑚 =𝑝

𝑚 ⊕(𝑛 ⋯𝑛
1 𝑛𝑚 )
𝑝,𝑞 𝑘𝑗 𝑑−𝑟𝑗
𝐸1 = 𝐻 (Conf𝑛𝑗 (ℝ )) ⇒ 𝐻 𝑝+𝑞 (Conf𝑛 (𝑋)),
𝑛1
⨁
+⋯+𝑛
(⨂ )
𝑚 =𝑛 𝑗=1
𝑛1 𝑟1 +⋯+𝑛𝑚 𝑟𝑚 =𝑞
𝑘1 +⋯+𝑘𝑚 =𝑝

where 𝑚 is the number of handles of 𝑋, each 𝑛𝑗 is a non-negative integer, 𝑟𝑗 is the

type of the 𝑗-th handle of 𝑋, and 𝑑 is the dimension of 𝑋. ◻
Since these spectral sequences come from double complexes, for any given
𝑋, it is possible to write down the differential maps explicitly, and find out the
(co)homology of Conf𝑛 (𝑋). We shall do this for elliptic curves in the next section.

3 Configuration spaces of elliptic curves

Let 𝐸 be an elliptic curve over ℂ, which is homeomorphic to a torus.
Lemma 3.1. There is a homeomorphism

Conf𝑛+1 (𝐸) ≃ 𝐸 × Conf𝑛 (𝐸 ∗ ),

where 𝐸 ∗ = 𝐸 ⧵ {∗}.
Proof. This correspondence is given by

(𝑥1 , … , 𝑥𝑛+1 ) ↦ (𝑥1 , (𝑥2 − 𝑥1 , … , 𝑥𝑛+1 − 𝑥1 )),

where we have used the group structure on 𝐸, and we identified 𝐸 ∗ ≃ 𝐸 ⧵ {0}. ◻

The space 𝐸 ∗ homeomorphic to the interior of a handlebody 𝑋, which contains
one 0-handle and two 1-handles, as shown below (edges with arrows are glued
together).

second
1-handle

first first
𝑋= 1-handle 0-handle 1-handle

second
1-handle

10
 Note that Conf𝑛 (𝐸 ∗ ) is homotopy equivalent to Conf𝑛 (𝑋), since the latter space
is a manifold (possibly with corners) whose interior is the former space. From now
on, we aim to compute

𝐻• (Conf𝑛 (𝐸 ∗ )) ≃ 𝐻• (Conf𝑛 (𝑋)),

using this handlebody structure.

1
First, we introduce a notation for generators of 𝐸𝑝,𝑞 in the spectral sequence of
∗
Theorem 2.11, for 𝑋 = 𝐸 .

Notation 3.2. We use the notation

(𝐹 , 𝑆1 , 𝑆2 ),

where

• 𝐹 is an 𝑛0 -forest with 𝑘0 inner vertices (as a generator of 𝐻𝑘0 (Conf𝑛0 (ℝ2 )) );

• 𝑆1 and 𝑆2 are ordered sets of 𝑛1 and 𝑛2 elements, respectively (as generators
of 𝐻0 (Conf𝑛𝑗 (ℝ1 )), for 𝑗 = 1, 2),

with
𝑛0 + 𝑛1 + 𝑛2 = 𝑛,

and with the leaves of 𝐹 and the elements of 𝑆1 , 𝑆2 labelled 1, 2, … , 𝑛 without

1
repeating. This denotes the corresponding generator of 𝐸𝑝,𝑞 , with

𝑝 = 𝑘0 , 𝑞 = 𝑛1 + 𝑛2 . ⊲
1
Thus, if (𝐹 , 𝑆1 , 𝑆2 ) ∈ 𝐸𝑝,𝑞 , then 𝐹 is an (𝑛 − 𝑞)-forest with |𝐹 | = 𝑝, and
|𝑆1 | + |𝑆2 | = 𝑞.
For example, the generator
2 5
(
1
, 3 , 4 1) ∈ 𝐸1,3 = 𝐻1 (Conf′5 (𝐸 ∗ )[3])

may be depicted as follows:

4 1

3 2 5 3

4 1

11
 1
Lemma 3.3. In the spectral sequence 𝐸𝑝,𝑞 , one has

𝑑1 = 𝜕 cell = 0 ∶ 𝐸𝑝,𝑞
1 1
→ 𝐸𝑝,𝑞−1 ,

so that
2 1
𝐸𝑝,𝑞 = 𝐸𝑝,𝑞 .

Proof. We have by definition

𝜕 cell (𝐹 , 𝑆1 , 𝑆2 )
= ∑ ±((𝐹 ∪ 𝑖, 𝑆1 ⧵ {𝑖}, 𝑆2 ⧵ {𝑖}) − (𝐹 ∪ 𝑖, 𝑆1 ⧵ {𝑖}, 𝑆2 ⧵ {𝑖}))
𝑖∈𝑆1 ∪𝑆2

= 0,

as homology classes. ◻

Next, we compute 𝑑2 .

Lemma 3.4. The map

2 2
𝑑2 ∶ 𝐸𝑝,𝑞 → 𝐸𝑝+1,𝑞−2

is given by
𝑖1 𝑖2
𝑑2 (𝐹 , 𝑆1 , 𝑆2 ) = ∑ ±( 𝐹 ∪ , 𝑆1 ⧵ {𝑖1 } , 𝑆2 ⧵ {𝑖2 } ),
𝑖1 ∈𝑆1
𝑖2 ∈𝑆2

where the ± sign is determined by the Koszul sign rule. Explicitly, if 𝑖𝑘 is the 𝑗𝑘 -th
element of 𝑆𝑘 (𝑘 = 1, 2), then the sign is (−1)|𝑆1 |+𝑗1 +𝑗2 −1 .

Although this lemma is a special case of Theorem 3.11 below, we write down
the construction explicitly to give the reader an idea of how the proof works.

Proof. For simplicity, we assume that (𝐹 , 𝑆1 , 𝑆2 ) = (∅, 1, 2). The general case
will follow from a similar argument.
Denote 𝛾 = (∅, 1, 2). Then

2
+

𝜕 cell 𝛾 = 1 + − 1 − 1 1 ,
−
2

12
where + and − indicate the coefficients of the points. Thus,
2
+
1 2
𝜕 cell 𝜕 −1 𝜕 cell 𝛾 = + 1 − + 1 ,
−
2

and this 1-cycle represents the points 1 and 2 orbiting each other for one counter-
clockwise rotation. ◻
Example 3.5. When 𝑛 = 2, we find that
1 2
( , ∅, ∅) = 𝑑2 (∅, 1, 2) = 𝑑2 (∅, 2, 1).
This means that the 1-cycle defined by two points orbiting each other is actually
trivial. It also means that
(∅, 1, 2) − (∅, 2, 1)
survives to 𝐸 ∞ , and defines a 2-cycle. We may write down this 2-cycle explicitly:
identifying 𝐸 ≃ ℝ2 ∕ℤ2 and 𝐸 ∗ ≃ 𝐸 ⧵ ( 21 , 21 ), we define two maps

𝑓 , 𝑔 ∶ 𝐸 ⧵ (0, 0) → Conf2 (𝐸 ∗ ),
𝑓 (𝑥, 𝑦) = ((𝑥, 0), (0, 𝑦)),
𝑔(𝑥, 𝑦) = ((0, 𝑥), (𝑦, 0)).
Putting them together (modifying them a little bit near (0, 0)), we obtain a map from
the connected sum of two tori to Conf2 (𝐸 ∗ ), as a generator of 𝐻2 (Conf2 (𝐸 ∗ )). ⊲
To describe the maps 𝑑3 , 𝑑4 , and so on, we need some preparations.
Construction 3.6. For each 𝑝, 𝑞 ≥ 0, we define a vector space
𝑛
𝑉𝑝,𝑞 = 𝕜 ⋅ (𝐹 , 𝑆)/∼,
⨁
𝐹 ,𝑆

where we sum over all pairs (𝐹 , 𝑆) such that

• 𝐹 is an (𝑛 − 𝑞)-forest with |𝐹 | = 𝑝, and with its leaves labelled by (𝑛 − 𝑞)
distinct elements of {1, … , 𝑛};
• 𝑆 is a permutation of the 𝑞 numbers not used by 𝐹 ,
and ∼ denotes the relations of forests given in Theorem 1.3. ⊲
In particular, we notice that
𝑛
𝑉𝑝,0 ≃ 𝐻𝑝 (Conf𝑛 (ℝ2 )),
𝑛
𝑉0,𝑞 ≃ 𝕜[𝔖𝑞 ],
where 𝔖𝑞 denotes the permutation group.

13
Construction 3.7. We define a map
1 𝑛
𝑎 ∶ 𝐸𝑝,𝑞 → 𝑉𝑝,𝑞
by sending
(𝐹 , 𝑆1 , 𝑆2 ) ↦ (𝐹 , [𝑆1 , 𝑆2 ])
= (𝐹 , 𝑆1 𝑆2 ) − (−1)|𝑆1 ||𝑆2 | (𝐹 , 𝑆2 𝑆1 ). ⊲
𝑛 𝑛
Construction 3.8. For each 𝑝 ≥ 0 and 𝑞 ≥ 1, we define a subspace 𝑊𝑝,𝑞 ⊂ 𝑉𝑝,𝑞 ,
and a pairing
⟨−, −⟩ ∶ 𝐻 𝑝+𝑞−1 (Conf𝑛 (ℝ2 )) ⊗ 𝑊𝑝,𝑞
𝑛
→ 𝕜,
as follows.
𝑛
Let 𝛤 be an 𝑛-graph with |𝛤 | = 𝑝 + 𝑞 − 1, and let (𝐹 , 𝑆) ∈ 𝑉𝑝,𝑞 be a generator.
The pairing ⟨𝛤 , (𝐹 , 𝑆)⟩ is only non-zero when, up to a reordering of the edges, 𝛤
can be written as
𝛤 = 𝛤 1 ∪ 𝛤2 ,
where ∪ denotes the union of edges, which coincides with the cup product, such that
the edges of 𝛤1 only connects vertices labelled with numbers used by 𝐹 , and the
edges of 𝛤2 only connects vertices labelled with numbers used by 𝑆, with |𝛤1 | = 𝑝
and |𝛤2 | = 𝑞 − 1. In this case, we define
⟨𝛤 , (𝐹 , 𝑆)⟩ = ⟨𝛤1 , 𝐹 ⟩⟨𝛤2 , 𝑆⟩,
where ⟨𝛤1 , 𝐹 ⟩ is defined in Theorem 1.8 (with 𝑑 = 2), and
⟨𝛤2 , 𝑆⟩ = det(𝜀𝑖𝑗 )1≤𝑖,𝑗≤𝑞−1 ,
where 𝜀𝑖𝑗 is given as follows: if the 𝑖-th edge of 𝛤2 is 𝑘𝑖 → 𝑙𝑖 , and the 𝑗-th element
of 𝑆 is 𝑠𝑗 , then

𝛿𝑠𝑗 𝑘𝑖 − 𝛿𝑠𝑗 𝑙𝑖 , if 𝑘𝑖 precedes 𝑙𝑖 in 𝑆,

𝜀𝑖𝑗 =
{ 0, otherwise.
𝑛 𝑛
Finally, let 𝑊𝑝,𝑞 ⊂ 𝑉𝑝,𝑞 be the subspace on which this pairing is well-defined,
that is, the relations satisfied by graphs have to be satisfied by this pairing.
Note that we could have defined (𝜀𝑖𝑗 ) as a (𝑞 − 1) × 𝑞 matrix, but all its (𝑞 − 1) ×
(𝑞 − 1) minors are equal (up to a sign), as all the row sums of (𝜀𝑖𝑗 ) are 0. Moreover,
the determinant must be 0 or ±1, as can be proved by induction on 𝑞. ⊲
𝑛
Using this pairing, one may associate homology classes to elements of 𝑊𝑝,𝑞 .
Construction 3.9. For 𝑝 ≥ 0 and 𝑞 ≥ 1, let
𝑛 𝑛
𝑏 ∶ 𝑊𝑝,𝑞 → 𝑊𝑝+𝑞−1,0 ≃ 𝐻𝑝+𝑞−1 (Conf𝑛 (ℝ2 ))
𝑛
be the map induced by the pairing ⟨−, −⟩, so that for any 𝑤 ∈ 𝑊𝑝,𝑞 and any 𝑛-graph
𝛤 with |𝛤 | = 𝑝 + 𝑞 − 1, one has
⟨𝛤 , 𝑏(𝑤)⟩ = ⟨𝛤 , 𝑤⟩. ⊲

14
 The spirit of the map 𝑏 is that it gives some sort of an ‘inverse’ to the map 𝑐
defined in Corollary 1.4, as is shown in the following example.
Example 3.10. Let 𝐹 be an (𝑛 − 𝑞)-forest with |𝐹 | = 𝑝, and let 𝑇 be a 𝑞-tree,
𝑛
with their 𝑛-leaves labelled 1, … , 𝑛. Then (𝐹 , 𝑐(𝑇 )) is in 𝑊𝑝,𝑞 , where 𝑐 is the map
defined in Corollary 1.4 (with 𝑑 = 2), and
𝑏(𝐹 , 𝑐(𝑇 )) = 𝐹 ∪ 𝑇 .
To prove this, it suffices to show that
⟨𝛤 , 𝑇 ⟩ = ⟨𝛤 , 𝑐(𝑇 )⟩
for any graph 𝛤 and any tree 𝑇 , as one can verify as an exercise in linear algebra.
⊲
𝑟
Theorem 3.11. Let 𝑟 ≥ 1, and suppose that we have an element of 𝐸𝑝,𝑞 of the form

𝛾 = ∑ 𝑐𝛼 ⋅ (𝐹 , 𝑆1𝛼 , 𝑆2𝛼 ),
𝛼

where the forest 𝐹 does not depend on 𝛼. Then

𝑑𝑟 (𝛾) = ∑ 𝑏∘𝑎 ∑ (−1)𝜎𝛼 𝑐𝛼 ⋅ (𝐹 , 𝑆1𝛼 ⧵ 𝑆1′ , 𝑆2𝛼 ⧵ 𝑆2′ ) , 𝑆1′ , 𝑆2′ ,

( ( ) )
𝑆1′ , 𝑆2′ 𝑆1′ ⊂𝑆1𝛼
𝑆2′ ⊂𝑆2𝛼

where
• The sum is taken over all sequences 𝑆1′ , 𝑆2′ of distinct elements of {1, … , 𝑛},
such that 𝑆1′ ∩ 𝑆2′ = ∅ and |𝑆1′ | + |𝑆2′ | = 𝑛 − 𝑟.
• The notation 𝑆𝑘′ ⊂ 𝑆𝑘𝛼 denotes inclusion preserving order (𝑘 = 1, 2).
• 𝜎𝛼 ∈ 𝔖𝑞 is the permutation satisfying
𝜎𝛼 (𝑆1𝛼 , 𝑆2𝛼 ) = (𝑆1𝛼 ⧵ 𝑆1′ , 𝑆2𝛼 ⧵ 𝑆2′ , 𝑆1′ , 𝑆2′ ).
𝑛
• That the expression 𝑏∘𝑎(⋯) is well-defined, i.e. the element 𝑎(⋯) lies in 𝑊𝑝,𝑞
′ ′
for each choice of 𝑆1 and 𝑆2 , is a part of the statement of the theorem.
We postpone the proof of the theorem to Appendix A.
Corollary 3.12. Suppose that 𝐹 is a forest, and 𝐿, 𝑅 are two trees, such that they
have 𝑛 leaves in total, labelled 1, … , 𝑛. Then the element
𝛾 = (𝐹 , 𝑐(𝐿), 𝑐(𝑅))
satisfies
𝑑1 (𝛾) = ⋯ = 𝑑𝑞−1 (𝛾) = 0,
𝑑𝑞 (𝛾) = (𝐹 ∪ 𝐿 𝑅, ∅, ∅),
where 𝑞 = |𝐿| + |𝑅| + 2.

15
Proof. For any tree 𝑇 , and any non-empty sequence 𝑆 ′ of distinct labels used by
𝑇 , we have
𝑐(𝑇 ) ⧵ 𝑆 ′ = 0,
where we extend the map (−) ⧵ 𝑆 ′ linearly (together with the appropriate ± sign),
𝐿 𝑅
and we define 𝑆⧵𝑆 ′ = 0 if 𝑆 ′ ⊄ 𝑆. Applying this to 𝑇 = explains why 𝑑1 (𝛾) =
⋯ = 𝑑𝑞−1 (𝛾) = 0. The expression for 𝑑𝑞 (𝛾) comes from Example 3.10. ◻
Corollary 3.13. For any 𝑝 ≥ 1, one has
𝑝+2
𝐸𝑝,0 = 0.
1
Proof. The space 𝐸𝑝,0 is generated by the elements (𝐹 , ∅, ∅), where 𝐹 is a forest.
Upon reordering the trees in 𝐹 , we may assume that 𝐹 = 𝐹 ′ ∪ 𝑇 , where 𝑇 is the
tree in 𝐹 such that it has the least leaves among the trees that have more than one
𝐿 𝑅
leaf. Let 𝑡 = |𝑇 |, and suppose that 𝑇 = . Then
(𝐹 , ∅, ∅) = 𝑑𝑡+1 (𝐹 ′ , 𝑐(𝐿), 𝑐(𝑅)).
Therefore, the element (𝐹 , ∅, ∅) is eliminated by 𝑑𝑡+1 . Since 𝑡 never exceeds |𝐹 | =
𝑝, it follows that 𝑑𝑝+1 must be surjective, whence the corollary. ◻
Corollary 3.14. For any 𝑛 ≥ 0, we have
ℎ1 (Conf𝑛 (𝐸 ∗ )) = 2𝑛,
𝑛(𝑛 − 1)(2𝑛 + 11)
ℎ2 (Conf𝑛 (𝐸 ∗ )) = ,
6
so that
ℎ1 (Conf𝑛 (𝐸)) = 2𝑛,
𝑛(𝑛 + 1)(2𝑛 + 1)
ℎ2 (Conf𝑛 (𝐸)) = .
6
Proof. The 𝐸 2 page is shown below, where each number in the grid represents the
dimension of the corresponding vector space, and the number on an arrow repre-
sents the rank of that arrow.
𝑞
4𝑛(𝑛 − 1)(𝑛 − 2) ⋯
2𝑛(𝑛−1)(𝑛−2)
3 ⋯
3𝑛(𝑛 − 1)

2𝑛 𝑛(𝑛−1) 𝑛(𝑛 − 1)(𝑛 − 2) ⋯

2
𝑛(𝑛 − 1) ⋯
1
2
𝑝

16
The rank 𝑛(𝑛 − 1)∕2 follows from Corollary 3.13, and the rank 2𝑛(𝑛 − 1)(𝑛 − 2)∕3
comes from
𝑖 𝑗 𝑖 𝑘
𝑑2 (𝐹 , 𝑖, 𝑗 𝑘) = (𝐹 ∪ , ∅, 𝑘) − (𝐹 ∪ , ∅, 𝑗),
3
where |𝐹 | = 0, so that each choice of 𝑖, 𝑗, 𝑘 gives two generators of 𝐸1,1 :

𝑖 𝑗 𝑗 𝑘 𝑘 𝑖
(⋯ , ∅, 𝑘) = (⋯ , ∅, 𝑖) = (⋯ , ∅, 𝑗),
𝑖 𝑗 𝑗 𝑘 𝑘 𝑖
(⋯ , 𝑘, ∅) = (⋯ , 𝑖, ∅) = (⋯ , 𝑗, ∅).
4
By Corollary 3.13 again, 𝐸2,0 = 0, so that the 𝐸 ∞ page looks like this:

𝑞
⋯

5𝑛(𝑛 − 1) ⋯
2
𝑛(𝑛 − 1)(𝑛 − 2)
2𝑛 ⋯
3

1 0 0

This implies the formulas for ℎ𝑝 (Conf𝑛 (𝐸 ∗ )) for 𝑝 = 1, 2. For ℎ𝑝 (Conf𝑛 (𝐸)), one
simply applies the Künneth formula. ◻

Corollary 3.15. The homology groups of Conf𝑛 (𝐸 ∗ ) and Conf𝑛 (𝐸), for small val-
ues of 𝑛, are given as follows.
𝑛 ℎ• (Conf𝑛 (𝐸 ∗ )) ℎ• (Conf𝑛+1 (𝐸))
0 1 1, 2, 1
1 1, 2 1, 4, 5, 2
2 1, 4, 5 1, 6, 14, 14, 5
3 1, 6, 17, 18 1, 8, 30, 58, 53, 18
4 1, 8, 38, 88, 81 1, 10, 55, 172, 295, 250, 81
⋯ ⋯ ⋯
Proof. Theorem 3.11 covers all the needed maps 𝑑𝑟 in order to compute these num-
bers, and the rank of the maps 𝑑𝑟 may be computed using a computer. ◻
𝑝,𝑞
Likewise, for the cohomological spectral sequence, the generators of 𝐸1 can
be written as
(𝛤 , 𝑆1 , 𝑆2 ),

where 𝛤 is an (𝑛 − 𝑞)-graph, and 𝑆1 , 𝑆2 are as above. We formulate a conjecture

on the mixed Hodge structure on the cohomology of Conf𝑛 (𝐸 ∗ ).

17
Conjecture 3.16. The mixed Hodge structure on 𝐻 𝑘 (Conf𝑛 (𝐸 ∗ ); ℂ) may be de-
𝑝,𝑞
scribed as follows. The weight filtration on 𝐸1 is given by
𝑝,𝑞
𝑊𝑙 𝐸1 = ℂ ⋅ (𝛤 , 𝑆1 , 𝑆2 ),
⨁
2|𝛤 |+|𝑆1 |+|𝑆2 |≤𝑙

and the Hodge filtration is given by

𝑝,𝑞
𝐹 𝑎 𝐸1 = ℂ ⋅ ((𝛤 , 𝑆1 , 𝑆2 ) + i(𝛤 , 𝑆2 , 𝑆1 )).
⨁
|𝛤 |+|𝑆1 |≥𝑎

𝑝,𝑞
The induced weight and Hodge filtrations on 𝐸∞ determine the mixed Hodge struc-
ture on 𝐻 𝑘 (Conf𝑛 (𝐸 ∗ ); ℂ).

Appendix A Proof of Theorem 3.11

The goal of this appendix is to study the map

𝑑𝑟 = (𝜕 cell ∘ 𝜕 −1 )𝑟−1 ∘ 𝜕 cell

𝑟
in the spectral sequence 𝐸𝑝,𝑞 . For this purpose, we make the following identifica-
1
tion, where (𝐹 , 𝑆1 , 𝑆2 ) denotes a generator of 𝐸𝑝,𝑞 .

𝑍1+ 𝑍2+
𝑆2
𝑆1 𝑆2
𝐹 𝐹
𝑆1 𝑆1 ≃
𝑆2 𝑆1
𝑆2
𝑍2− 𝑍1−

Let 𝑃 denote this rectangle, and let 𝑍1± , 𝑍2± ⊂ 𝑃 denote the four open intervals
as marked in the figure. Let 𝑍1 denote the topological space 𝑍1+ ≃ 𝑍1− , and sim-
ilarly for 𝑍2 . Let 𝑍 ± ⊂ 𝑃 denote the upper and lower horizontal edges (as closed
intervals), and let 𝑍 ⊂ 𝑃 denote their union.
Let

′
𝐶•,𝑞 = 𝐶 rect (Conf𝑛−𝑞 (𝑃 )) ⊗
⨁ ( •
𝑛1 +𝑛2 =𝑞
⊕(𝑛−𝑞, 𝑛𝑛 , 𝑛 )
𝐶•rect (Conf𝑛1 (𝑍1 )) ⊗ 𝐶•rect (Conf𝑛2 (𝑍2 ))) 1 2
,

where we take the tensor product of chain complexes, and

sing ∼ sing
𝐶•rect (Conf𝑛 (𝑋)) ⊂ 𝐶• (𝑋)⊗𝑛 → 𝐶• (𝑋 𝑛 )

18
denotes the subcomplex spanned by the elements 𝜎1 ⊗ ⋯ ⊗ 𝜎𝑛 , where each 𝜎𝑖 is a
simplex in 𝑋, such that the image of 𝜎1 × ⋯ × 𝜎𝑛 in 𝑋 𝑛 is contained in Conf𝑛 (𝑋).
We interpret the multiplicity (𝑛−𝑞, 𝑛𝑛 , 𝑛 ) as taking 𝑛 − 𝑞 points from {1, … , 𝑛} to
1 2
place on 𝑃 , 𝑛1 points to place on 𝑍1 , and 𝑛2 points on 𝑍2 .
Let
′
𝐶𝑝,𝑞 ⊂ 𝐶𝑝,𝑞

be the subspace spanned by those generators (𝑓 , 𝑧1 , 𝑧2 ), such that both 𝑓 × 𝑧+1 × 𝑧+2
and 𝑓 × 𝑧−1 × 𝑧−2 , as singular chains in 𝑃 𝑛 , are supported in Conf𝑛 (𝑃 ), where 𝑧±𝑘
denotes the push-forward of 𝑧𝑘 along the inclusion map 𝑍𝑘± ↪ 𝑃 . Then there are
boundary maps

𝜕 ∶ 𝐶𝑝,𝑞 → 𝐶𝑝−1,𝑞 , 𝜕 cell ∶ 𝐶𝑝,𝑞 → 𝐶𝑝,𝑞−1 ,

which make 𝐶•,• into a double complex, where 𝜕 cell is defined on the generators by
𝑞
𝜕 cell (𝑓 , 𝑧1 , 𝑧2 ) = ∑ (−1)𝑗−1 (𝑓 × (𝑖+∗ 𝑝𝑗 − 𝑖−∗ 𝑝𝑗 ), 𝑧1 ⧵ 𝑝𝑗 , 𝑧2 ⧵ 𝑝𝑗 ),
𝑗=1

where 𝑝𝑗 denotes the 𝑗-th point of the sequence 𝑧1 𝑧2 , and 𝑖± ∶ 𝑍1± ∪ 𝑍2± ↪ 𝑃
denotes the inclusion map. The notations 𝑖±∗ 𝑝𝑗 and 𝑧𝑘 ⧵ 𝑝𝑗 (𝑘 = 1, 2) are only
suggestive — precisely, one considers the singular chains 𝑓 ×𝑧+1 ×𝑧+2 and 𝑓 ×𝑧−1 ×𝑧−2
of Conf𝑛 (𝑃 ), and then, one regards them as elements of 𝐶𝑝,𝑞−1 by moving 𝑝𝑗 to the
first component.
Note that the double complex 𝐶•,• perfectly describes what happens to the 0-
𝑟
handle in the actual spectral sequence 𝐸𝑝,𝑞 . In fact, it is not difficult to see the
following.
Lemma A.1. The spectral sequence of the double complex 𝐶•,• is isomorphic to
•
the spectral sequence 𝐸•,• from the first page onward. ◻
•
Therefore, from now on in this appendix, we will regard 𝐸•,• as the spectral
sequence of 𝐶•,• . Note also that we have a natural forgetful map
𝑛
𝑔 ∶ 𝐶𝑝,𝑞 ↪ 𝐶𝑝 (Conf𝑛−𝑞 (𝑃 )) (𝑞) .
sing ⊕

Definition A.2. Let 𝑋 be a topological space, and let 𝛼 be a singular (resp. Borel–
Moore) 𝑛-chain of 𝑋. The canonical decomposition of 𝛼 is the unique way to write
𝛼 as a finite (resp. countable) sum

𝛼 = ∑ 𝑎𝑖 𝐴𝑖 ,
𝑖

where 𝑎𝑖 ∈ 𝕜 are distinct non-zero elements, and 𝐴𝑖 ⊂ Sing 𝑋 are simplicial

subsets, such that each (non-degenerate) 𝑛-simplex of Sing 𝑋 is contained in at
most one 𝐴𝑖 . ⊲

19
Definition A.3. Let 𝑋 be a topological manifold of dimension 𝑛. Let 𝛼 be a Borel–
Moore 𝑘-chain of 𝑋, and let 𝛽 be a singular (𝑛 − 𝑘)-chain of 𝑋. Let 𝛼 = ∑𝑖 𝑎𝑖 𝐴𝑖
and 𝛽 = ∑𝑗 𝑏𝑗 𝐵𝑗 be their canonical decompositions, and denote 𝐴 = ∐𝑖 𝐴𝑖 and
𝐵 = ∐ 𝑗 𝐵𝑗 .
We say that 𝛼 and 𝛽 intersect transversely, if there only exists finitely many
pairs (𝑥𝐴 , 𝑥𝐵 ) ∈ 𝐴 × 𝐵, such that the images of 𝑥𝐴 , 𝑥𝐵 in 𝑋 coincide, and all such
pairs (𝑥𝐴 , 𝑥𝐵 ) lie in (𝐴 ⧵ 𝜕𝐴) × (𝐵 ⧵ 𝜕𝐵).
In this case, the intersection 𝛼 ∩ 𝛽 is a well-defined 0-cycle of 𝑋. ⊲

For each 𝑘, let

𝑌 ⊂ Conf𝑘 (𝑃 )

denote the set of configurations where all the points have the same 𝑦-coordinate.
Then 𝑌 can be regarded as a submanifold of Conf𝑘 (𝑃 ) of codimension 𝑘 − 1.
The orientation of 𝑌 is given as follows. For any 𝑥 ∈ 𝑌 , the isomorphism

𝑇𝑥 Conf𝑘 (𝑃 ) ≃ 𝑇𝑥 𝑌 ⊕ ℝ ⋅ (𝑝1 moves down) ⊕ ⋯ ⊕ ℝ ⋅ (𝑝𝑘−1 moves down)

preserves orientation, where 𝑝𝑖 is the 𝑖-th point from the left to the right, according
to the configuration 𝑥.

Definition A.4. An element 𝛾 ∈ 𝐶𝑘−1,𝑛−𝑘 is said to be a good cycle, if 𝜕𝛾 = 0,

and each of the (𝑘𝑛) components of 𝑔(𝛾) intersects transversely with 𝑌 , with the
intersection (as a 0-cycle) supported in 𝑌 ∩ Conf𝑘 (𝑍) ⊂ Conf𝑘 (𝑃 ).
A good cycle 𝛾 ∈ 𝐶𝑘−1,𝑛−𝑘 is said to be simple, if precisely one of the (𝑘𝑛)
components of 𝑔(𝛾) is non-zero. ⊲

For a simple good cycle 𝛾 ∈ 𝐶𝑘−1,𝑛−𝑘 , the homology class of 𝑔(𝛾) is determined
by its intersection with 𝑌 , which lies in Conf𝑘 (𝑍).

Lemma A.5. Let 𝛾 be a simple good cycle, and suppose that

𝑚
𝑌 ∩ 𝑔(𝛾) = ∑ 𝑐𝑖 𝑥𝑖 ,
𝑖=1

with 𝑐𝑖 ∈ 𝕜 and 𝑥𝑖 ∈ Conf𝑘 (𝑃 ). Then for any (𝑘 − 1)-graph 𝛤 , if we regard 𝛤 as

an element of 𝐻 𝑘−1 (Conf𝑘 (𝑃 )), then
𝑚
⟨𝛤 , 𝑔(𝛾)⟩ = ∑ 𝑐𝑖 ⟨𝛤 , 𝑆𝑖 ⟩,
𝑖=1

where 𝑆𝑖 is the element of 𝔖𝑘 given by the horizontal arrangement of points at 𝑥𝑖 ,

and ⟨𝛤 , 𝑆𝑖 ⟩ is defined in Construction 3.8.

Proof. We observe that the inclusion Conf𝑘 (𝑃 ) ↪ Conf𝑘 (ℝ2 ) is a homotopy equiv-
alence, and we write 𝛤 for the corresponding cohomology class of Conf𝑘 (ℝ2 ). Us-
ing the duality between cohomology classes and Borel–Moore homology classes,

20
we represent 𝛤 as the Borel–Moore (𝑘+1)-cycle 𝑌𝛤 , defined as follows. Let 𝑖𝑟 → 𝑗𝑟
be the 𝑟-th edge of 𝛤 . Then 𝑌𝛤 is the union of the connected components of 𝑌
(whose definition is extended from Conf𝑘 (𝑃 ) to Conf𝑘 (ℝ2 )) in which the 𝑖𝑟 -th point
is on the left of the 𝑗𝑟 -th point for all 𝑟. The orientation of 𝑌𝛤 is given so that at
any 𝑥 ∈ 𝑌𝛤 , the isomorphism

𝑇𝑥 Conf𝑘 (ℝ2 ) ≃
𝑇𝑥 𝑌𝛤 ⊕ ℝ ⋅ (𝑖1 -th point moves down and 𝑗1 -th point moves up) ⊕
⋯ ⊕ ℝ ⋅ (𝑖𝑘−1 -th point moves down and 𝑗𝑘−1 -th point moves up)

preserves orientation. We thus have

⟨𝛤 , 𝑔(𝛾)⟩ = [𝑌𝛤 ∩ 𝑔(𝛾)] ∈ 𝐻0 (Conf𝑘 (ℝ2 )).

At each 𝑥𝑖 , either 𝑥𝑖 ∉ 𝑌𝛤 , in which case ⟨𝛤 , 𝑆𝑖 ⟩ = 0, or 𝑥𝑖 ∈ 𝑌𝛤 and the orientation

of 𝑌𝛤 differs from that of 𝑌 by a sign, which is equal to ⟨𝛤 , 𝑆𝑖 ⟩ by the definition
of the latter. ◻
Lemma A.6. Let 𝛾 ∈ 𝐶𝑘−1,𝑛−𝑘 be a good cycle, and suppose that there exists
𝛽 ∈ 𝐶𝑘,𝑛−𝑘 such that 𝛾 = 𝜕𝛽. Then 𝛽 can be chosen so that 𝜕 cell 𝛽 is a good cycle
with
1 cell
𝑔(𝑌 ∩ 𝜕 cell 𝛽) = 𝜕 (𝑌 ∩ 𝑔(𝛾))|𝑌 ,
𝑘+1
sing
where (−)|𝑌 denotes removing from an element of 𝐶0 (Conf𝑘+1 (𝑍)) the points
that are not in 𝑌 .
Proof. Let 𝑔(𝛽) = ∑𝑖 𝑎𝑖 𝐵𝑖 be the canonical decomposition, where each 𝐵𝑖 is a
simplicial subset of Sing(Conf𝑘 (𝑃 )), and let

𝐴𝑖 = 𝜕𝐵𝑖 ∩ (support of 𝑔(𝛾) in Sing(Conf𝑘 (𝑃 ))).

Then 𝑔(𝛾) is represented by ∑𝑖 𝑎𝑖 𝐴𝑖 .

We modify the (𝑘 + 1)-cycles 𝐵𝑖 , so that they intersect transversely with 𝑌 .
Denote 𝐵 = ⋃𝑖 𝐵𝑖 ⊂ Sing(Conf𝑘 (𝑃 )), and let 𝐴 = ⋃𝑖 𝐴𝑖 ⊂ 𝐵. The natural
map 𝑏 ∶ 𝐵 → Conf𝑘 (𝑃 ) is homotopic to a map 𝑏′ ∶ 𝐵 → Conf𝑘 (𝑃 ) rel 𝐴, such
that 𝑏′ (𝐵 ⧵ 𝐴) is contained in Conf𝑘 (𝑃 ⧵ 𝑍). This can be shown by induction on
the dimension of simplices. From now on, we replace 𝛽 with the singular chain
represented by 𝑏′ .
Since 𝜕 cell 𝛽 is supported in the subset of Conf𝑘+1 (𝑃 ) with at least one of the
(𝑘 + 1) points in 𝑍, it follows that 𝜕 cell 𝛽 can only intersect with 𝑌 in Conf𝑘+1 (𝑍).
At these intersection points, 𝑔(𝛽) meets with Conf𝑘 (𝑍), so 𝑔(𝛾) must meet with
Conf𝑘 (𝑍) by the construction of 𝑏′ .
Let 𝑥 ∈ Conf𝑘+1 (𝑃 ) be a point where 𝑔(𝜕 cell 𝛽) meets 𝑌 . Let 𝑣 ∶ Conf𝑘+1 (𝑃 ) →
(ℝ≥0 )𝑘+1 be the projection map which takes the (euclidean) distance from each
point to 𝑍, so that 𝑣(𝑥) = 0. Let 𝛽1 , … , 𝛽𝑘+1 be the (𝑘 + 1) of the (𝑘𝑛) components
of 𝜕 cell 𝛽 which concern 𝑘 of the (𝑘 + 1) points in question, with 𝛽𝑖 missing the

21
𝑖-th point, and let 𝜕 ′ 𝛽𝑖 be the component of 𝜕 cell 𝛽𝑖 that concerns these (𝑘 + 1)
points. Denote 𝛽 ′ = ∑𝑖 𝛽𝑖 . Then 𝑣∗ (𝜕 ′ 𝛽 ′ ) is a 𝑘-cycle supported in 𝜕((ℝ≥0 )𝑘+1 ),
with each 𝑣∗ (𝜕 ′ 𝛽𝑖 ) supported in a face (ℝ≥0 )𝑘 . Since 𝑣∗ (𝜕 ′ 𝛽 ′ ) is a cycle, each
𝜕𝑣∗ (𝜕 ′ 𝛽𝑖 ) is supported in 𝜕((ℝ≥0 )𝑘 ). If we consider its local (at 𝑥) intersection in
𝜕((ℝ≥0 )𝑘 ) with the point 0, we obtain a number 𝑡𝑖 . Replacing the point 0 by a point
in {𝑦𝑖 = 𝑦𝑗 = 0} near 0, where 𝑦𝑖 and 𝑦𝑗 are the coordinates for ℝ𝑘+1 , we see that
𝑡𝑖 = 𝑡𝑗 , since ∑𝑖 𝜕𝑣∗ (𝜕 ′ 𝛽𝑖 ) = 0. Hence, all the numbers 𝑡𝑖 are equal, and we call
this number 𝑡. It follows that the intersection of the cycle 𝑣∗ (𝜕 ′ 𝛽 ′ ), in 𝜕((ℝ≥0 )𝑘+1 ),
with the point 0, is also equal to 𝑡. Finally, we notice that 𝑡 is the coefficient of 𝑥 in
𝑌 ∩ 𝑔(𝜕 cell 𝛽) and in 𝑌 ∩ 𝑔(𝛾). We have a coefficient 1∕(𝑘 + 1), because every term
on the left corresponds to (𝑘 + 1) terms on the right. ◻

Proof of Theorem 3.11. For the sake of simplicity, we assume that 𝐹 = ∅; the
general case only involves minimal modification of the argument. We thus have
(𝑝, 𝑞) = (0, 𝑛).
Let 𝛾 ∈ 𝐶0,𝑛 be a representative of the homology class which is denoted by 𝛾
in the theorem. We prove by induction on 𝑟 that if we make appropriate choices
when taking 𝜕 −1 of an element, then the cycle

𝑑𝑟 (𝛾) = (𝜕 cell ∘ 𝜕 −1 )𝑟−1 ∘ 𝜕 cell (𝛾) ∈ 𝐶𝑟−1,𝑛−𝑟

is a good cycle, with

𝛼 ′ 𝛼 ′
𝑌 ∩ 𝑔𝑆 ′ , 𝑆 ′ (𝑑𝑟 (𝛾)) = ∑ (−1)𝜎𝛼 𝑐𝛼 ⋅ (𝑥+𝛼, 𝑆 ′ , 𝑆 ′ − (−1)|𝑆1 ⧵𝑆1 | |𝑆2 ⧵𝑆2 | 𝑥−𝛼, 𝑆 ′ , 𝑆 ′ ),
1 2 1 2 1 2
𝑆1′ ⊂𝑆1𝛼
𝑆2′ ⊂𝑆2𝛼

where the notations are as in the theorem; 𝑔𝑆 ′ , 𝑆 ′ denotes taking the components
1 2
where the second and third parts are in the connected components given by 𝑆1′ and
𝑆2′ , and then apply 𝑔; the points 𝑥±𝛼, 𝑆 ′ , 𝑆 ′ ∈ Conf𝑟 (𝑍 ± ) are given by taking the
1 2
points in 𝑆1𝛼 ⧵ 𝑆1′ and 𝑆2𝛼 ⧵ 𝑆2′ from our chosen representative 𝛾. In fact, the case
𝑟 = 1 is clear, and the induction step follows from Lemma A.6 in a straightforward
way, as long as one keeps track of the signs.
Combining this with Lemma A.5, we arrive at the formula given in the theorem.
◻

References
[MP12] J. P. May and K. Ponto (2012). More Concise Algebraic Topology. Chicago Lec-
tures in Mathematics. University of Chicago Press.
[Sin06] D. Sinha. The homology of the little disks operad. arXiv: math/0610236.

22