Two graph homologies and the space of long embeddings
Abstract
Graph homologies are powerful tools to compute the rational homotopy group of the space of long embeddings . Two graph homologies have been invented from two approaches to study the space of long embeddings: the hairy graph homology from embedding calculus, and BCR graph homology from configuration space integral. In this paper, we construct a monomorphism from the top hairy graph homology to the top BCR graph homology, though the latter graph homology is quite modified. This map and its left inverse are analogs of PBW isomorphism between and , the space of open Jacobi diagrams and the space of Jacobi diagrams on the unit circle, in the theory of Vassiliev knot invariants.
Introduction
A long embedding is an embedding of into , which is standard outside a disk in . We write for the space of long embeddings , and write for the homotopy fiber of the forgetful map to the space of long immersions. The aim of this paper is to compare two combinatorial objects that are introduced to study .
In 2017, Fresse, Turchin and Willwacher [FTW], following Arone and Turchin [AT1][AT2], established a beautiful framework to compute the rational homotopy group in terms of combinatorial object called graph homology. More precisely, they showed the isomorphism between and the graph homology , called hairy graph homology. This significant result is based on the theory of embedding calculus established by Goodwillie, Klein and Weiss [GW][GKW][Wei] and also the theory of little disk operad.
However, this deep framework is fully successful, so far, only for the range . Moreover, it seems difficult to know geometric generators of from generators of .
On the other hand, from 90’s to 10’s, Bott[Bot], Cattaneo, Rossi[CR], Sakai[Sak] and Watanabe[Wat] developed a framework to construct geometric (co)cycles of from another graph (co)homology. We call their graph homology BCR graph homology and write it as . To get the cocycles of , they used the operation called configuration space integral. Fortunately, this framework is applicable to .
Surprisingly, BCR graph homology has very similar graphs to those of . However, although embedding calculus allows hairy graphs with any number of loops, configuration space integral was successful only for BCR graphs with zero or one loop. This caused much less information for or given by configuration space integral.
In the author’s previous paper [Yos], we have succeeded in conducting configuration space integral for some 2-loop BCR graphs, and in giving a new non-trivial (co)cycle of the space of long embeddings (including ). However, we haven’t been able to give more examples, due to much more complicated combinatorics of than .
In this paper, we relate the two graph homologies and at the top (non-degenerate) part. We write the top part of as and write the top part of as , though is quite modified as explained later. We focus on the case is even because the author is most interested in the case . The following is our main result.
Theorem 0.1.
There is a monomorphism from to .
The proof of our main result is very similar to the proof by Bar-Natan[Bar] of PBW (Poincaré–Birkhoff–Witt) isomorphisms between , the space of open Jacobi diagrams, and , the space of Jacobi diagrams on the unit circle. As Bar-Natan showed another isomorphism between and the space of chord diagrams , we have the following. Let be the space of BCR chord diagrams defined in Definition 1.17.
Theorem 0.2.
There is an isomorphism from to .
It might be possible to apply our main result to give new nontrivial (co)cycles of by the following steps: First, compute some part (two-loop part, for example) of by relating it with (anti-)symmetric polynomials as in [CCTW]. Next, by Theorem 0.1, we obtain some nontrivial subspace of . This gives some non-trivial graph cocycle . (See the next paragraph.) So we can perform configuration space integral
associated with (with the correction term in [Yos]). Finally, conduct pairing argument in [Yos] between the cocycle and a suitably-constructed cycle of . Then might detect the non-triviality of this cycle.
However, it should be emphasized that the (top) graph homology is modified in this paper: has no relation corresponding to contractions of chords. (See Definition 1.12.) Hence, a graph cocycle constructed through a weight system on
may not be a genuine graph cocycle used in [Yos], because contractions of chords may not be canceled. One possible solution will be adding correction terms using the data of paths of immersions that has. We deal with the correction terms in the author’s future work in preparation [Yos2].
The second result holds even if we impose relations on chords to the spaces to . Let to be the quotient space.
Theorem 0.3.
There is an isomorphism from to .
At present, the author does not know there is a monomorphism from to the genuine graph homology . But it holds for all the cases the author can compute , including the simplest 2-loop case in [Yos]. Note that the map is not necessarily surjective even in the case the target is the quotient space .
Question 0.4.
Is the map from to injective?
As a final remark, we go back to Arone and Turchin’s paper [AT2]. In [AT2], the graph homology , which is written there as , is constructed from the homology of a derived mapping space between modules:
by an injective resolution of the target. In [AT2], another graph homology is introduced by a projective resolution of the source. By taking both projective and injective resolutions, one obtains one more graph homology , which does not explicitly appear in [AT2]. Since the origin is the same, the three graph homologies are isomorphic. It would be interesting to see directly the differences between and ; the duals of and have almost the same graphs. A similar question arises for differences between and the top part of . See Remark 1.21.
Acknowledgement
The author is deeply grateful to Victor Turchin for explaining to the author a lot about their work on graph homology. The author would like to thank Tadayuki Watanabe for motivating and inspiring discussion with the author. This paper was written during the author’s three-month stay at Kansas State University. The author appreciates their hospitality. This research was supported by Forefront Physics and Mathematics Program to Drive Transformation (FoPM), a World-leading Innovative Graduate Study (WINGS) Program, The University of Tokyo.
1 Definition of graph homologies
1.1 The modified BCR graph homology
Here, we define the spaces and of BCR graphs. This space is almost equivalent to the dual of the top graph cohomology defined in [Yos] as . However, we add an extra condition (the last condition in the following definition) to the graphs. Moreover, we exclude some relations on chords.
Definition 1.1 (BCR graphs).
A BCR graph is a connected graph that satisfies the following.
-
•
There are (at most) two types of vertices: external vertices drawn in black and internal vertices drawn in white. The graph must have at least one external vertex.
-
•
There are (at most) two types of edges: dashed edges and solid edges.
-
•
Each internal vertex has exactly three dashed edges and no solid edge.
-
•
Each external vertex has exactly one dashed edge and at most two solid edges.
-
•
The restriction to solid edges of the graph forms a disjoint sum of broken lines such as . In particular, neither double solid edges nor small solid loops are allowed.
Remark 1.2.
In Definition 1.1, small dashed loops , double dashed edges and multiple edges are allowed. However, some of them vanish by relations we later define.
Definition 1.3.
We define two ways of coloring BCR graphs: (I) odd case and (II) even case.
-
(I)
We orient each edge and label the vertices from to if the number of vertices is .
-
(II)
We number the edges from to if the number of edges is .
We say two colorings of the isomorphic underlying graph are the same (resp. opposite) orientation if they are transferred from each other by even (resp. odd) permutation.
Definition 1.4.
We write for the vector space over generated by colored BCR graphs.
Definition 1.5.
We write for the quotient vector space of by IHX, STU, and orientation relations. We define STU and IHX relations below.
Definition 1.6.
We define IHX relations of colored BCR graphs as in Figure 1 below. In Figure 1, graphs are allowed to have any colors so that contractions of the middle edges give three isomorphic graphs with the same orientation (after transposition to the left-hand side). See [Yos] for the sign rule for contraction of edges.
= - .
Remark 1.7.
As a consequence of an IHX relation, a small dashed loop at an internal vertex that is connected to another internal vertex vanishes:
Definition 1.8.
We define STU relations (I)(II) of colored BCR graphs as in Figure 2 below. In Figure 2, graphs are allowed to have any colors so that contractions of the middle edges (circled in red) give isomorphic graphs with the same orientation. In the case (I), the bottom vertex of the graph has one or two solid edges. In the case (II), the bottom vertex of the graph has no solid edge.

Remark 1.9.
As a consequence of an STU relation in the odd case (Definition 1.3), a small dashed loop at an internal vertex that is connected to an external vertex vanishes:
Remark 1.10.
As a consequence of an orientation relation in the even case, a double dashed edge vanishes.
Definition 1.11.
We write for the quotient vector space of by IHX, STU, orientation relations and chord relations. We define chord relations below.
Definition 1.12.
We define chord relations (I)(II) of colored BCR graphs as in Figure 3 below. In Figure 3, graphs are allowed to have any colors so that contractions of the middle edges give isomorphic graphs with the same orientation. Here, We only use chord relations in which all the graphs are modified BCR graphs. Namely, if the contracted graph has a loop consisting of only solid edges and external vertices, we do not consider the relation as a chord relation. In the case (I), the bottom vertex of the graph has one or two solid edges. In the case (II), the bottom vertex of the graph has no solid edge.

Remark 1.13.
We exclude contractions of chords that yield loops consisting of only solid edges for the following reason: if the dimension of the space of embedding is large enough comparing the order (size) , more precisely , the integral by such graphs vanish by dimensional reason. This argument is introduced in Section 8 of [Yos]. Our second result 0.2 holds even if we admit graphs with solid loops and add chord relations that include such graphs.
1.2 The hairy graph homology
We define the space of hairy graphs. The space is an analog of the space of open Jacobi diagrams modulo IHX and AS relations (and they coincide in the odd case).
Definition 1.14 (Hairy graphs).
A hairy graph is a connected graph that satisfies the following.
-
•
There are (at most) two types of vertices: external vertices drawn in black and internal vertices drawn in white. The graph must have at least one external vertex.
-
•
There is only one type of edges, dashed.
-
•
Each external vertex has exactly one dashed edge. Each internal vertex has exactly three dashed edges.
We can define color and orientation of hairy graphs by those of BCR graphs.
Definition 1.15.
We write for the vector space over generated by colored hairy graphs. We define the space as the quotient space of by orientation relations and IHX relations. In the odd case, we add the following relation (See Remark 1.9).
Remark 1.16.
1.3 The space of BCR chord diagrams
Next, we define the spaces of BCR chord diagrams and . These spaces are analogs of the space of chord diagrams , which appears in the theory of Vassiliev knot invariants.
Definition 1.17 (BCR chord diagrams).
A BCR chord diagram is a connected graph that satisfies the following.
-
•
There is only one type of vertices: external vertices drawn in black.
-
•
There are two types of edges, solid and dashed.
-
•
Each external vertex has exactly one dashed edge and at most two solid edges.
-
•
The restriction to solid edges of the graph forms a disjoint union of broken lines. See Definition 1.1.
We can define color and orientation of BCR chord diagrams by those of BCR graphs.
Definition 1.18.
We write for the vector space over generated by colored BCR chord diagrams. We define as the quotient space of by orientation relations and relations, which are defined below.
Definition 1.19.
We define relations (I)(II)(III) of colored BCR chord diagrams as in Figure 4 below. In Figure 4, graphs are allowed to have any colors so that and are both STU relations for some colored BCR graph . The graph is uniquely determined after orientation relations in .

Definition 1.20.
We define as the quotient space of by orientation relations, relations and chord relations. Here, we only use chord relations in which all the graphs are BCR chord diagrams.
Remark 1.21.
In [AT2], Arone and Turchin introduced a complex that is quasi-isomorphic to . Though is a little difficult to describe, its dual complex is described in terms of graphs [AT2]. The top (non-degenerate) part of is generated by BCR chord diagrams. However, they allow not only broken lines but also trees as the restriction to solid edges. They also impose Arnold relations that come from relations in and . As a consequence of Arnold relations (with respect to solid edges), trees of solid edges can be transformed into a linear combination of broken lines. Our relations correspond to the differential of (with Arnold relations with respect to dashed edges). So far, the author finds nothing related to chord relations in the complex .
2 Main Result
2.1 The map
Proposition 2.1.
We define the map by the inclusion. Then this map induces .
Theorem 2.2.
The map is a monomorphism.
2.2 The map
Proposition 2.3.
We define the map by the inclusion. Then this map induces the maps and .
Theorem 2.4.
Both of the maps and are isomorphisms.
3 Preliminary for proof of Main Result
Here, we introduce notations used in Section 4.
Definition 3.1.
A solid component of a BCR graph is a component of solid broken lines. That is, a solid component consists of black vertices and solid edges for any , where have no other solid edges. The solid type of is the collection (in descending order) of lengths of solid components. For example, the graph in Figure 5 has the solid type .

Notation 3.2.
Let be the subspace of spanned by colored hairy graphs with at most external vertices. Let be the subspace of spanned by colored BCR graphs with at most external vertices. Clearly restricts to . Let be the subspace of spanned by STU relations of graphs with at most external vertices.
Notation 3.3.
Let be a colored BCR graph with external vertices. Assume that solid components of are in a row as in Figure 6. For , let be the colored graph obtained from by permutating dashed edges so that th edge from the left goes to the th. Then we set
Let be the subset of which consists of permutations that preserve solid components. That is, if , the th and th edges must be connected to the same solid component. If the solid type of is , is isomorphic to

Notation 3.4.
Let be a colored BCR graph with external vertices. Let be a transposition in . We define as a colored BCR graph with vertices so that the equation
is an STU relation. The graph is determined uniquely after orientation relation in . See Figure 7.

4 Proof of Main Result
The proof of Proposition 2.1 and 2.3 is trivial by the definitions of their relations. We prove Theorem 2.2 and Theorem 2.4 in the following. The proof is very similar to the proof of [Bar, Theorem 8] and [Bar, Theorem 6].
4.1 is a monomorphism
Proof of Theorem 2.2.
We construct the left inverse of the map . We construct inductively the map which satisfies the following.
-
()
is the projection.
-
()
(orientation relation)
-
()
(IHX relation)
-
()
First, we define as the identity. Clearly satisfies ()()()(). Next we assume that is constructed so that it satisfies ()()()(). Then we define by
for a graph with external vertices.111Remember that Bar-Natan’s original construction was . We do not use part. Here is an element of determined uniquely after orientation relations. In , the element satisfies
We explain the construction of . If is the elementary transposition , we set . For a general permutation , write as the product of elementary transpositions of : . Then
Example 4.1.
Let , , , respectively. Then we have
So we have
Similarly, we have
Note that are STU relations.
We must show that does not depend on choices of the expression . Let be the monoid generated by elementary transpositions of . We have constructed the map
from the monoid ring of . To show it defines a map from , it is enough to show the kernel of includes the kernel of the map . The kernel is the both-sided ideal generated by
-
(G1)
,
-
(G2)
,
-
(G3)
.
Here the s in each relation must be elements of the same of .
Let be a colored graph of the even case
,
where the th, -th, -th, -th, -th dashed edges are drawn. Without loss of genericity, we assume that . We show that vanishes on (G1), (G2) and (G3). A completely similar proof works for the odd case.
(G1)
(G2)
By STU relation in ,
for the graph in Figure 8, which has external vertices. Since satisfies () by the induction hypothesis, we have .

(G3)
Where , and are graphs in Figure 9 and satisfies an IHX relation. We used STU relation in in the last equation, Since satisfies () by the induction hypothesis, we have .

Next we prove that satisfies () () () ().
() For a hairy graph ,
() If the orientation of is reversed (resp. preserved), the orientations of and are also reversed (resp. preserved). Hence, the orientation of is reversed (resp. preserved). By the induction hypothesis, vanishes on orientation relations. Since is the sum of for , (orientation relations).
() Let be an IHX relation. Then also satisfies an IHX relation, because the procedure does not change the part IHX occurs. Since satisfies () by the induction hypothesis,
() Let be an STU relation. Then we have
∎
4.2 is an isomorphism
Proof of Theorem 2.4.
We construct the inverse (resp. ) of the map (resp. ). Namely, we construct the map (resp. ) which vanishes on STU, IHX and orientation relations (and chord relations). Note that by Lemma 4.2 below, an IHX relation can be written a sum of several STU relations. Let be a (colored) BCR graph. By using STU relations repeatedly, we can transform the graph to a linear combination , where each is a BCR chord diagram. We call this operation resolution of . We set as the equivalence class of in . By construction, the map , if it is well-defined, induces . We can easily check and .
However, we must show that the resulting does not depend, up to relations, on choices of procedures of STU relations. We show this by induction on the number of internal vertices .
The case is trivial because no STU relation is performed. Let , and let be the unique internal vertex. Since has three adjacent dashed edges , and , there are three choices of STU relations. Each choice produces a sum of two (or one) chord diagrams. We can easily see that the differences between these combinations are exactly relations.
Next, we assume that the case is proved and show the case . Let an edge of the graph connect an external vertex and an internal vertex . Let another edge of connect an external vertex and an internal vertex . First, we assume . Consider the two procedures to perform STU relations. One performs the first STU relation to and gives a sum of graphs , while another performs the first STU relation to and gives . Note that after the first choice of STU relation, the resulting combination of chord diagrams does not depend, up to relations, on the remaining choices of STU relations. Then we can assume that the second STU relation of is performed to the edge (we write this procedure as ), while the second STU relation of is performed to the edge (we write this as ). But the two procedures yield the same combination of graphs.
So the remaining case is the case and is connected to another internal vertex. If there is an edge that connects external vertex and an internal vertex , the procedure is equal to . So let be connected to a block which consists of only internal vertices (and dashed edges). We see in Lemma 4.3 that in both cases ( first and first), the result of STU relations to after the STU relation to vanishes by relation.
This ends the proof of well-definedness of . Finally, we show that the map sends chord relations to chord relations. Let be a chord relation such that has an internal vertex connected by an edge to an external vertex . Consider performing the STU relation to and transform to . Since graphs are related by chord relations, there are corresponding edges and corresponding STU relations for each . We can observe that and are chord relations.
∎
4.3 IHX in terms of STU
Lemma 4.2.
Recall that is the space generated by BCR graphs with at least one external vertex. Then an IHX relation in can be written as a sum of STU relations.
Proof..
Let a graph have an “” part, namely two internal vertices and connected by a dashed edge . Consider the IHX relation performed on the edge . We have three graphs , and . We want to show by only using STU relations.
Using STU relation simultaneously to , and , we can assume that is connected to an external vertex by some dashed edge . Then perform the STU relation to and in this order. This yields four or two graphs for each , depending on the number of solid edges the external vertex has. We can easily see that these twelve (or six) graphs are canceled. See Figure 10.

∎
4.4 Sliding of chords
Lemma 4.3 (Sliding of chords).
Let be a BCR chord diagram. Let be an external vertex and let have one dashed edge that is connected to an external vertex . Let be other external vertices. Assume that these vertices satisfy the following. See Figure 11.
-
•
is connected by one solid edge to
-
•
are connected by chords among them.
-
•
and are connected by one solid edge.
-
•
has no other solid edge except for at most one solid edge connected to . We write this exceptional edge as .

We construct another chord diagram from by remove the solid edge and inserting a new solid edge to .
Then by using relation, and are equivalent.
Proof..
4T relation can be written as , where are four graphs obtained by four ways to connect an end of dashed edge to the sides of some fixed chord. Consider the sum of graphs obtained by all the ways to connect to the sides of the chords. By relation, this sum is equal to . On the other hand, observe that in the sum there are pairs that consist of the two isomorphic graphs with the opposite sign. (See Figure 12.)The remaining graphs are and .

∎
References
- [AT1] G. Arone, V. Turchin. On the rational homology of high-dimensional analogues of spaces of long knots. Geom. Topol. 18 (2014), no. 3, 1261–1322.
- [AT2] G. Arone, V. Turchin. Graph-complexes computing the rational homotopy of high dimensional analogues of spaces of long knots. Ann. Inst. Fourier (Grenoble) 65 (2015), no. 1, 1–62.
- [Bar] D. Bar-Natan, On the Vassiliev knot invariants. Topology 34 (1995), no. 2, 423–472.
- [Bot] R. Bott, Configuration spaces and imbedding invariants. Turkish J. Math. 20 (1996), no. 1, 1–17.
- [CCTW] J. Conant, J. Costello, V. Turchin, P. Weed, Two-loop part of the rational homotopy of spaces of long embeddings, J. Knot Theory Ramifications 23,(2014), no. 4, 1450018, 23 pp.
- [CR] A. S. Cattaneo, C. A. Rossi, Wilson surfaces and higher dimensional knot invariants. Comm. Math. Phys. 256 (2005), no. 3, 513–537
- [FTW] B. Fresse, V. Turchin, T.Willwacher, The rational homotopy of mapping space of operads, arXiv.1703.06123
- [GW] T. Goodwillie, M. Weiss, Embeddings from the point of view of immersion theory. II. Geom. Topol. 3 (1999), 103–118.
- [GKW] T. Goodwillie, J. Klein, M. Weiss. Spaces of smooth embeddings, disjunction and surgery. Surveys on surgery theory, Vol. 2, 221–284, Ann. of Math. Stud., 149, Princeton Univ. Press, Princeton, NJ, 2001.
- [Sak] K. Sakai, Configuration space integrals for embedding spaces and the Haefliger invariant. J. Knot Theory Ramifications 19 (2010), no. 12, 1597–1644.
- [SW] K. Sakai, T. Watanabe, loop graphs and configuration space integral for embedding spaces. Math. Proc. Cambridge Philos. Soc. 152 (2012), no. 3, 497–533.
- [Wat] T. Watanabe, Configuration space integral for long n-knots and the Alexander polynomial. Algebr. Geom. Topol. 7 (2007), 47–92.
- [Wei] M. Weiss, Embeddings from the point of view of immersion theory I.Geom. Topol. 3 (1999), 67–101.
- [Yos] L. Yoshioka, Cocycles of the space of long embeddings and BCR graphs with more than one loop, arXiv.2212.01573
- [Yos2] L. Yoshioka, Decorated graph complex and the space of long embeddings modulo immersions, in preparation.