2 EMMANUEL GRAFF
the case of algebraically split 5-component links (that is, 5-component links with vanishing linking
numbers).
The notion of clasper was developed by Habiro in [9]. These are surfaces in 3–manifolds with some
additional structure, on which surgery operations can be performed. In [9], Habiro describes the
clasper calculus up to isotopy, which is a set of geometric operations on claspers that yield equivalent
surgery results. It is well known to experts how clasper calculus can be refined for the study of
knotted objects up to link-homotopy (see for example [5, 26]). This homotopy clasper calculus, which
we review in Section 2, will be the key tool for proving all the main results outlined above.
The rest of this paper consists of three sections.
In Section 2, we review the homotopy clasper calculus: after briefly recalling from [9] Habiro’s
clasper theory, we recall how a fundamental lemma from [5], combined with Habiro’s work, produces
a set of geometric operations on claspers having link-homotopic surgery results.
Section 3 is dedicated to the study of braids up to link-homotopy. We start by reinterpreting braids
in terms of claspers. In Section 3.1 we define comb-claspers, a family of claspers corresponding to
braid commutators. They are next used to define a normal form on homotopy braids, thus allowing
us to rewrite any braid as an ordered product of comb-claspers. In Section 3.2 after a short algebraic
interlude, we give a presentation of the pure homotopy braid group (Corollary 3.20), using the work
of [6] and [23] as well as the technology of claspers. Finally, we define and study in Section 3.3 a
representation of the homotopy braid group which is in a sense the linearization of the homotopy
Artin representation. We give its explicit computation in Theorem 3.26 (see also Example 3.28 for
the 3-strand case) and show its injectivity in Theorem 3.31. Moreover, from the injectivity of the
representation follows the uniqueness of the normal form and thus the definition of the clasp-numbers,
a collection of braid invariant up to link-homotopy. Note that our representation has lower dimension
than Humphries one. The correspondence between the two representations has not been established
yet, but we wonder if our representation could open new leads on the torsion problem for more than
six strands.
The final Section 4 focuses on the study of links up to link-homotopy. The method used is based
on the precise description of some operations, which generate the algebraic equivalence relation men-
tioned above in the classification result of Habegger and Lin [8]; we provide them with a topological
description in terms of claspers. This new point of view allows us, for a small number of components,
to describe when two braids in normal form have link-homotopic closures. We translate in terms of
clasp-number variations the action of those operations on the normal form. In this way we recover the
classification results of Milnor [22] and Levine [17] for 4 or less components (Theorem 4.7). Moreover,
we also classify 5-component algebraically split links up to link-homotopy (Theorem 4.10).
Acknowledgement: The author thanks the referee for his/her careful reading and insightful sugges-
tions. This work is partially supported by the project AlMaRe (ANR- 19-CE40-0001-01) of the ANR.
The author thanks P. Bellingeri and J.B. Meilhan for their great advises and helpful discussions.
2. Clasper calculus up to link-homotopy
Clasper calculus has been developed by Habiro in [9] in the context of tangles up to isotopy.
Claspers turn out to be in fact a powerful tool to deal with link-homotopy. In this section we first
define claspers and their associated vocabulary. Then we describe how to handle claspers up to
link-homotopy.