Embedding Calculus Overview

• Embedding calculus is a homotopy-theoretic framework that uses a Taylor tower of finite disc configurations to approximate spaces of embeddings.
• It integrates manifold calculus, operad theory, and configuration spaces to study intricate structures in knots, links, and diffeomorphism groups.
• The approach offers convergent approximations in high codimension and provides tools to analyze smooth, topological, and Poincaré embeddings through layered obstruction theory.

Embedding calculus usually denotes the Goodwillie–Weiss calculus of embeddings: a homotopy-theoretic “Taylor tower” for spaces of embeddings

$$\mathrm{Emb}(M,N)\longrightarrow T_\infty \mathrm{Emb}(M,N)\longrightarrow \cdots \longrightarrow T_n\mathrm{Emb}(M,N)\longrightarrow \cdots \longrightarrow T_1\mathrm{Emb}(M,N),$$

in which the $n$-th stage is assembled from embeddings of finite unions of at most $n$ discs in the source manifold, and the first stage recovers immersions (Jin, 6 Nov 2025). The subject lies at the intersection of manifold calculus, Goodwillie’s functor calculus, operad theory, configuration spaces, and homotopy theory, and it has developed into a framework for studying knots, links, diffeomorphism groups, Poincaré embeddings, graph complexes, and related obstruction theories (Krannich et al., 2024).

1. Core construction and the Taylor tower

In Weiss’s formulation, one fixes a target manifold $N$ and studies the contravariant functor $U\mapsto \mathrm{Emb}(U,N)$ on open subsets of a source manifold $M$. The tower is built by restricting to unions of at most $n$ discs and then taking a homotopy right Kan extension. In the $\infty$-categorical language used for $d$-manifolds, if $D_{\leq n,d}\subset M_d$ is the full subcategory on unions of at most $n$0 copies of $n$1, with inclusion $n$2, then for a functor $n$3 one defines the manifold $n$4-excisive approximation by

$n$5

A functor is manifold $n$6-excisive when the unit $n$7 is an equivalence, and $n$8 is determined by the values of $n$9 on configurations of up to $n$0 disjoint discs in $n$1, together with compatibility data (Jin, 6 Nov 2025).

For $n$2, the stage $n$3 recovers immersions, while the fibers of

$n$4

are describable in terms of configuration and mapping spaces, giving an inductive interpolation between immersions and embeddings (Jin, 6 Nov 2025). This is formally analogous to Goodwillie’s tower

$n$5

for homotopy functors, but Weiss calculus is tailored to contravariant functors on manifolds rather than arbitrary functors between pointed $n$6-categories (Jin, 6 Nov 2025).

A persistent theme is that the tower is informative even when convergence fails. For codimension $n$7, the map

$n$8

is a weak equivalence, but for classical knots the tower generally does not converge to the knot space itself; nevertheless, its stages, layers, and spectral sequences still encode substantial geometric and algebraic information (Kosanović, 2020, Moriya, 28 Sep 2025).

2. Operadic, module-theoretic, and categorical formulations

A major reformulation of embedding calculus interprets it through right modules over operads. In the $n$9-operadic approach, one starts from a unital $N$0-operad $N$1, forms its symmetric monoidal envelope $N$2, and identifies right $N$3-modules with presheaves

$N$4

Truncation to objects with at most $N$5 inputs gives a tower

$N$6

and for the $N$7-framed $N$8-operad this recovers the Goodwillie–Weiss tower for smooth embeddings. Replacing $N$9 by $U\mapsto \mathrm{Emb}(U,N)$0 yields topological embedding calculus, while using $U\mapsto \mathrm{Emb}(U,N)$1 yields a configuration-style or “particle” version (Krannich et al., 2024).

This operadic framework makes the tower simultaneously monoidal and Morita-theoretic. The first layer identifies $U\mapsto \mathrm{Emb}(U,N)$2 with $U\mapsto \mathrm{Emb}(U,N)$3, while higher layers are described by a recollement involving the wreath product $U\mapsto \mathrm{Emb}(U,N)$4. Changing the operad along an operadic right fibration, such as $U\mapsto \mathrm{Emb}(U,N)$5, produces pullback squares of towers and gives a smoothing-theoretic comparison between smooth and topological embedding calculi (Krannich et al., 2024).

A parallel operadic description appears for parallelized manifolds. If $U\mapsto \mathrm{Emb}(U,N)$6 is a parallelized $U\mapsto \mathrm{Emb}(U,N)$7-manifold and $U\mapsto \mathrm{Emb}(U,N)$8 is an $U\mapsto \mathrm{Emb}(U,N)$9-manifold with a section of its $M$0-frame bundle, Abramyan defines a modified embedding functor by adjoining, to an embedding $M$1, a path in the pulled-back $M$2-frame bundle from the reference framing to the derivative framing. Its Taylor tower is identified with a mapping space of right modules over the Fulton–MacPherson operad, and under rationalization the derived mapping space is modeled by a hairy graph complex (Abramyan, 8 Apr 2025).

These formulations also clarify the relation with Goodwillie’s functor calculus. Tillmann–Weiss show that for an embedding $M$3, the Goodwillie $M$4 of the complement is manifold $M$5-excisive in all codimensions, and Heuts’s categorical $M$6-excisive approximations $M$7 supply an ambient $M$8-categorical tower in which complements of $M$9-embeddings can be made functorial (Jin, 6 Nov 2025).

3. Convergence, dimension, and smooth-structure dependence

The classical convergence regime is high codimension. In the long-knot context, Goodwillie–Klein’s connectivity theorem gives that

$n$0

is $n$1-connected, except in the knot case, and if $n$2 the tower converges (Kosanović, 2020). For the modified functor on parallelized manifolds, Abramyan proves convergence under the codimension condition $n$3 (Abramyan, 8 Apr 2025).

Low-dimensional targets behave differently. For compact manifold triads $n$4 with $n$5, Krannich and Kupers prove that

$n$6

is a weak equivalence for any boundary condition $n$7. In particular, for any compact surface $n$8,

$n$9

is a weak equivalence (Krannich et al., 2021). This is a low-dimensional convergence theorem rather than a codimension-$\infty$0 theorem, and it depends on surface topology, isotopy extension, coverings, collars, and configuration-space compactifications (Krannich et al., 2021).

The $\infty$1-operadic formulation also yields topological convergence and an improved smooth convergence theorem. For $\infty$2, if the inclusion $\infty$3 is an equivalence on tangential 2-types, then

$\infty$4

is a weak equivalence, and under the same tangential 2-type condition the analogous statement holds for topological embeddings when the target is smoothable (Krannich et al., 2024).

Embedding calculus is not uniformly sensitive to smooth structure. Crowley, Ferry, and Skipper prove that embedding calculus does not distinguish exotic smooth structures in dimension $\infty$5, implying a negative answer to a question of Viro, while in higher dimensions it does distinguish certain exotic spheres (Knudsen et al., 2020). More precisely, the Taylor tower depends only on the formally smooth structure of source and target; this is enough to identify homeomorphic simply connected compact 4-manifolds, but not certain higher-dimensional exotic spheres (Knudsen et al., 2020). A plausible implication is that convergence and detectability are governed not only by codimension, but also by how much tangential information survives passage to the chosen model of the tower.

4. Layers, complements, and related calculi

One development of the subject studies complements rather than embeddings themselves. If $\infty$6 is a genuine embedding, Tillmann–Weiss show that the complement functor becomes manifold $\infty$7-excisive after applying the Goodwillie $\infty$8 of the identity on pointed spaces, and this leads to a $\infty$9-complement functor

$d$0

defined on arbitrary $d$1-embeddings, where a $d$2-embedding is a morphism of presheaves

$d$3

in $d$4. On genuine embeddings, the resulting diagram commutes with the Yoneda embedding and the $d$5-complement recovers the Goodwillie $d$6 of the ordinary complement (Jin, 6 Nov 2025).

This complement formalism interacts closely with Heuts’s categorical $d$7-excisive approximations $d$8. The $d$9-complements assemble into a lax morphism of towers, and for embeddings $D_{\leq n,d}\subset M_d$0 the resulting complement in $D_{\leq n,d}\subset M_d$1 satisfies a Stallings-type theorem: every $D_{\leq n,d}\subset M_d$2-complement is canonically equivalent to a fixed model $D_{\leq n,d}\subset M_d$3, and consequently there is a map

$D_{\leq n,d}\subset M_d$4

with $D_{\leq n,d}\subset M_d$5 (Jin, 6 Nov 2025).

A different extension, due to Klein, concerns codimension-zero embeddings of Poincaré duality spaces in disks. There one constructs a tower

$D_{\leq n,d}\subset M_d$6

interpolating between Poincaré immersions and a space of unlinked Poincaré embeddings, with $D_{\leq n,d}\subset M_d$7. The layers are expressed in terms of the coefficient spectra $D_{\leq n,d}\subset M_d$8 of the identity in Goodwillie calculus: $D_{\leq n,d}\subset M_d$9 and the appendix formulates a conjectural comparison with the manifold-calculus tower for smooth embeddings (Klein, 2014). This places embedding calculus alongside surgery-theoretic and normal-invariant methods rather than only within the manifold setting.

5. Knots, links, and finite-type phenomena

Long knots provide one of the most developed application areas. For $n$00, Goodwillie’s punctured-knots model identifies the $n$01-th stage with a homotopy limit over punctured intervals,

$n$02

and yields evaluation maps

$n$03

whose effect on components defines embedding-calculus invariants $n$04 (Kosanović, 2020). In any compact oriented $n$05-manifold $n$06, the map

$n$07

is surjective for all $n$08, which solves some remaining open cases of the Goodwillie–Klein–Weiss connectivity estimates and confirms one half of the conjecture that, for classical knots, the $n$09 are universal additive Vassiliev invariants over the integers (Kosanović, 2020).

For classical framed long knots in $n$10, Budney, Conant, Koytcheff, and Sinha identify the $n$11-th stage with a mapping-space model $n$12 and show that the map on components

$n$13

is a map of monoids whose target is an abelian group and which is invariant under Habiro’s $n$14-moves; consequently it is a finite type invariant of degree $n$15 (Budney et al., 2014). The associated tower spectral sequence has

$n$16

the $n$17-module of primitive chord diagrams on an interval with $n$18 chords, modulo 4T and SEP, which is evidence for universality over the integers (Budney et al., 2014).

The knot-theoretic layers can also be realized geometrically. For long knots in arbitrary $n$19-manifolds, the first possibly non-vanishing invariant of a knot grope cobordant to the unknot is computed by the underlying decorated tree of the grope in the associated graph complex, and the layers $n$20 satisfy

$n$21

for $n$22 (Kosanović, 2020). This identifies the first nontrivial stage of the tower with the same tree-level structures that govern finite-type theory.

The cohomological side of knot spaces is equally tightly linked to embedding calculus. For long knots $n$23, the Vassiliev spectral sequence and the Sinha spectral sequence are isomorphic at the level of $n$24-pages over a field, with the degree shift

$n$25

Since the Sinha spectral sequence degenerates rationally, this implies that the Vassiliev spectral sequence degenerates at the $n$26-page over $n$27, including the non-diagonal part (Moriya, 28 Sep 2025). The comparison is built through a Thom-space model that captures embedding calculus of the knot space in terms of fat diagonals (Moriya, 28 Sep 2025).

String links exhibit a complementary phenomenon. For $n$28, the embedding tower of $n$29 in $n$30 admits maps

$n$31

and on path components this recovers the Artin representation

$n$32

In particular, the $n$33-th stage detects all Milnor $n$34-invariants of length $n$35 (Jin, 6 Nov 2025).

6. Surfaces, mapping class groups, and graph-complex models

For surfaces, embedding calculus interacts with diffeomorphism groups and mapping class theory in a way that is unavailable in higher dimension. Krannich and Kupers show that for a compact orientable surface $n$36 with one boundary component, the $n$37-module valued tower applied to $n$38 defines a filtration

$n$39

on $n$40, and this filtration is contained in the Johnson filtration: $n$41 The comparison is mediated by Fulton–MacPherson compactifications of configuration spaces and Moriyama’s description of the Johnson filtration in terms of the action on $n$42 (Krannich et al., 2021).

At the rational level, embedding calculus for surfaces produces graph dg Lie algebras that model the Torelli side of diffeomorphism groups. Krannich–Kupers show that suitable embedding-calculus models for $n$43 and $n$44 are given by topological monoids of homotopy automorphisms of configuration-space modules over framed little-disks operads, and Felder–Naef–Willwacher identify the rational models of these automorphism spaces with semidirect products such as

$n$45

and related graph complexes (Naef et al., 10 Feb 2026).

Within this framework, the tree-level truncation of the graph complex recovers the symplectic derivation Lie algebra

$n$46

and the Johnson homomorphism is realized as the projection

$n$47

The loop filtration

$n$48

then yields a spectral sequence whose differentials are the Enomoto–Satoh traces

$n$49

with the image of the Johnson homomorphism equal to the joint kernel of all higher traces and the Johnson cokernel identified with $n$50 of the positive-loop-order part of the graph complex (Naef et al., 10 Feb 2026). This gives an embedding-calculus interpretation of Johnson theory as a passage from trees to full graph cohomology.

7. Extensions, obstructions, and terminological scope

Embedding calculus has also been adapted beyond manifolds. For a finite CW complex $n$51, Arone and Krushkal define a simplified tower by configuration-space natural transformations,

$n$52

where $n$53 is the ordered configuration space of $n$54 points in $n$55 (Arone et al., 2021). Here $n$56, $n$57 recovers the classical van Kampen obstruction, and the primary obstruction to lifting from $n$58 to $n$59 is a class

$n$60

For $n$61-complexes in $n$62, this obstruction is complete and coincides with a Whitney-disk obstruction $n$63; equivalently, it is the pullback of the Arnold class in configuration-space cohomology (Arone et al., 2021). This places van Kampen theory, Whitney towers, and configuration-space cohomology inside an explicitly calculus-like tower.

A related moral appears in Poincaré embedding theory and in complement calculus: the tower need not be restricted to smooth embeddings of manifolds, but can interpolate between immersions and embeddings, or between embeddings and complements, so long as the relevant polynomial and excisive structures are available (Klein, 2014, Jin, 6 Nov 2025). This suggests that “embedding calculus” is best regarded as a family of Taylor-type approximation procedures centered on local embedding data, configuration spaces, and excision, rather than as a single rigid model.

There is, however, a distinct and terminologically unrelated use of embeddability in the history of analysis. In work on Leibnizian infinitesimal calculus, the question is whether Leibniz’s procedures can be represented in first-order logic augmented by notions such as infinite proximity and standard part; there “embedding” is explicitly procedural rather than ontological (Blaszczyk et al., 2016). That problem belongs to logic and the foundations of analysis, not to Goodwillie–Weiss embedding calculus.

In contemporary topology, the dominant meaning of embedding calculus is therefore the Goodwillie–Weiss program and its extensions: a tower of excisive approximations to embedding spaces, reformulated through operads, right modules, complements, and graph complexes, and applied across knot theory, link invariants, surface diffeomorphisms, Poincaré duality spaces, and obstruction theory (Krannich et al., 2024).