Model Homotopy Transfer

• Model homotopy transfer is a method for moving higher algebraic structures between chain complexes using explicit tree-sum formulas and homotopy retract data.
• It employs classification theorems and obstruction theory to ensure that transferred A∞, L∞, and P∞ structures retain essential coherence and functoriality.
• This framework spans applications from rational homotopy theory to effective field theory, providing canonical reductions of complex algebraic models.

Model homotopy transfer is the problem of transporting a homotopy algebraic structure—typically $A_\infty$, $L_\infty$, $C_\infty$, or more generally $P_\infty$—from one chain complex to another along a homotopy equivalence or deformation retract, while retaining the higher coherence relations in transferred form. In its modern form, the subject combines explicit tree-level formulas for transferred operations with classification results describing exactly which homotopy structures arise by transfer, and with categorical frameworks in which transfer becomes strictly functorial (Markl et al., 2020, Markl, 2024).

1. Basic homotopy-retract data and transferred operations

The foundational input is a homotopy retract. For a quadratic Koszul operad $P$ over a field of characteristic $0$, one starts with a $P_\infty$-algebra $(A,d_A,\{\mu_n^A\}_{n\ge2})$ and chain data

$$H \xrightarrow{i} A \xrightarrow{p} H,\qquad h:A\to A$$

satisfying

$$p\circ i=\mathrm{id}_H,\qquad \mathrm{id}_A-i\circ p=d_Ah+hd_A.$$

This data may be supplemented by the usual side conditions $L_\infty$0, $L_\infty$1, and $L_\infty$2. The Homotopy Transfer Theorem then produces a transferred $L_\infty$3-structure $L_\infty$4, with higher operations given by tree-sum formulas over planar rooted trees with leaves labeled by $L_\infty$5, internal edges by $L_\infty$6, vertices by the original operations $L_\infty$7, and the root by $L_\infty$8 (Markl, 2024).

In the $L_\infty$9 case, the same construction admits both recursive and combinatorial descriptions. Kopřiva formulates the transferred operations $C_\infty$0 via recursively defined $C_\infty$1 and then sets $C_\infty$2; equivalently, one sums over planar rooted trees whose internal vertices carry the original $C_\infty$3, internal edges carry the homotopy $C_\infty$4, leaves carry $C_\infty$5, and the root carries $C_\infty$6 (Kopřiva, 2017). This equivalence between recursion and tree combinatorics is one of the standard structural features of the theory.

The immediate significance of these formulas is that the higher coherence identities are not re-proved case by case. They are inherited from the original coderivation or coalgebraic data. In practical terms, this means that a complicated strict or non-strict algebraic structure on a large complex can be replaced by a homotopy-equivalent structure on a smaller complex—often homology or cohomology—at the cost of introducing higher operations.

2. The “model” classification theorem

A central refinement is the classification of those homotopy algebra structures that actually come from transfer. Markl–Rogers consider an $C_\infty$7-algebra $C_\infty$8 and a fixed chain homotopy equivalence

$C_\infty$9

After choosing a homotopy inverse $P_\infty$0 and homotopies, the classical transfer procedure yields an explicit transferred $P_\infty$1-structure $P_\infty$2. Their theorem states that for any $P_\infty$3-structure $P_\infty$4 with unary part $P_\infty$5, the following are equivalent: $P_\infty$6 is obtained by homotopy transfer from $P_\infty$7 along $P_\infty$8; $P_\infty$9 is isotopic to the explicit transferred structure $P$0; and all natural obstruction classes

$P$1

vanish (Markl et al., 2020).

This result converts transfer from an existence statement into a necessary-and-sufficient criterion. The phrase “model homotopy transfer” is naturally associated with this perspective: the transferred structure is not merely one representative among many quasi-isomorphic models, but a canonical model up to isotopy relative to the fixed linear part. The same paper also develops an obstruction theory for extending partial weak $P$2-morphisms. If $P$3 has already been constructed, the obstruction to finding $P$4 lies in

$P$5

with a specific cocycle $P$6 controlling the extension problem (Markl et al., 2020).

The classification extends from $P$7 to $P$8 for any quadratic Koszul operad $P$9 over a field of characteristic zero. This broadens the model-theoretic viewpoint from associative homotopy algebras to Lie, commutative, and other Koszul-governed homotopy structures. A plausible implication is that “coming from transfer” is itself a structured property, detectable by isotopy and obstruction classes rather than only by ad hoc explicit formulas.

3. Strict functoriality and the bifibrational reformulation

A further development replaces informal functoriality statements by a precise categorical mechanism. For $0$0-structures, one defines a category $0$1 whose objects are isotopy classes of $0$2-algebras and whose morphisms are equivalence classes of $0$3-morphisms whose linear part is a chain-homotopy equivalence. The forgetful functor

$0$4

to the category of chain complexes and chain-homotopy equivalence classes is a surjective-on-objects discrete Grothendieck bifibration (Markl, 2024).

This statement has concrete consequences. Given any homotopy equivalence class $0$5 and any lift of $0$6 to an isotopy class of $0$7-structures, there is a unique lift in $0$8 over $0$9; dually there is a unique op-lift. The same construction extends to $P_\infty$0-algebras for any quadratic Koszul operad $P_\infty$1, giving a forgetful functor

$P_\infty$2

that is again a discrete Grothendieck bifibration (Markl, 2024).

The corollary is strict functoriality of transfer under composition: if

$P_\infty$3

are composable chain-homotopy equivalences and $P_\infty$4 is a $P_\infty$5-algebra, then the transfer along $P_\infty$6 coincides canonically with the successive transfer along $P_\infty$7 and then $P_\infty$8. In the notation of the paper,

$P_\infty$9

This eliminates the higher ambiguity often associated with iterated transfer procedures (Markl, 2024).

The specializations clarify the scope of the result. For $(A,d_A,\{\mu_n^A\}_{n\ge2})$0-algebras one may work over an arbitrary commutative ring $(A,d_A,\{\mu_n^A\}_{n\ge2})$1; for $(A,d_A,\{\mu_n^A\}_{n\ge2})$2- and $(A,d_A,\{\mu_n^A\}_{n\ge2})$3-algebras one assumes $(A,d_A,\{\mu_n^A\}_{n\ge2})$4 (Markl, 2024). The significance is methodological: transfer is no longer only an explicit homological perturbation construction, but also a strictly compositional operation in a categorical setting.

4. Effective field theory, BV theory, and gauge-theoretic realizations

In mathematical physics, homotopy transfer acquires a direct field-theoretic interpretation. Arvanitakis, Hohm, Hull, and Lekeu formulate a field theory as an $(A,d_A,\{\mu_n^A\}_{n\ge2})$5-algebra $(A,d_A,\{\mu_n^A\}_{n\ge2})$6 split into light and heavy subspaces $(A,d_A,\{\mu_n^A\}_{n\ge2})$7, together with projection $(A,d_A,\{\mu_n^A\}_{n\ge2})$8, inclusion $(A,d_A,\{\mu_n^A\}_{n\ge2})$9, and a contracting homotopy $$H \xrightarrow{i} A \xrightarrow{p} H,\qquad h:A\to A$$0 satisfying

$$H \xrightarrow{i} A \xrightarrow{p} H,\qquad h:A\to A$$1

The transferred effective brackets on $$H \xrightarrow{i} A \xrightarrow{p} H,\qquad h:A\to A$$2 are then

$$H \xrightarrow{i} A \xrightarrow{p} H,\qquad h:A\to A$$3

and these are shown to govern integrating out heavy modes at tree level; in the dual picture, the effective BRST charge is obtained after a canonical change of variables generated by $$H \xrightarrow{i} A \xrightarrow{p} H,\qquad h:A\to A$$4, and the heavy-field propagator is precisely $$H \xrightarrow{i} A \xrightarrow{p} H,\qquad h:A\to A$$5 (Arvanitakis et al., 2020).

This tree-level picture has a BV realization. For a general $$H \xrightarrow{i} A \xrightarrow{p} H,\qquad h:A\to A$$6-structure one encodes the brackets $$H \xrightarrow{i} A \xrightarrow{p} H,\qquad h:A\to A$$7 in a BV master action $$H \xrightarrow{i} A \xrightarrow{p} H,\qquad h:A\to A$$8, and the classical master equation $$H \xrightarrow{i} A \xrightarrow{p} H,\qquad h:A\to A$$9 is equivalent to the $$p\circ i=\mathrm{id}_H,\qquad \mathrm{id}_A-i\circ p=d_Ah+hd_A.$$0-identities. The same transferred operations on cohomology or on a chosen subspace arise from the tree-level expansion of a BV path integral, with propagator equal to the chosen homotopy $$p\circ i=\mathrm{id}_H,\qquad \mathrm{id}_A-i\circ p=d_Ah+hd_A.$$1. In this framework, rooted-tree transfer formulas are identified with Feynman tree expansions, while unimodularity controls whether one-loop corrections appear (Maunder, 2024).

Several concrete applications are organized by this dictionary. In closed string field theory on a torus, homotopy transfer from the full $$p\circ i=\mathrm{id}_H,\qquad \mathrm{id}_A-i\circ p=d_Ah+hd_A.$$2-algebra to the subspace of doubly-massless states yields weakly constrained double field theory. The transferred 2-bracket is the Siegel-gauge projection of the closed-string cubic vertex, and the transferred 3-bracket consists of the original 4-point contact term together with single-propagator exchange terms; the induced brackets automatically satisfy the full $$p\circ i=\mathrm{id}_H,\qquad \mathrm{id}_A-i\circ p=d_Ah+hd_A.$$3-relations, so gauge closure and gauge invariance are preserved on the truncated fields (Arvanitakis et al., 2021). In gauge invariant perturbation theory, passing to gauge invariant variables is interpreted as homotopy transfer of the $$p\circ i=\mathrm{id}_H,\qquad \mathrm{id}_A-i\circ p=d_Ah+hd_A.$$4-algebra encoding the gauge theory, yielding an algorithmic procedure for higher corrections of both the invariant variables and the action (Chiaffrino et al., 2020). In massive Kaluza–Klein theory on a torus, homotopy transfer maps gauge redundant fields to gauge invariant fields, and the transferred brackets produce the physical couplings of the gauge-invariant Kaluza–Klein modes (Eloy et al., 2 Dec 2025). Tree-level scattering amplitudes also admit a transferred description: transferring a cyclic $$p\circ i=\mathrm{id}_H,\qquad \mathrm{id}_A-i\circ p=d_Ah+hd_A.$$5- or $$p\circ i=\mathrm{id}_H,\qquad \mathrm{id}_A-i\circ p=d_Ah+hd_A.$$6-structure from the full field space to the on-shell subspace yields brackets that encode tree amplitudes and satisfy generalized Jacobi identities implying Ward identities (Bonezzi et al., 2023).

Taken together, these constructions show that transfer is not merely a formal reduction of algebraic data. In the field-theoretic setting it becomes an algebraic formulation of integrating out, gauge fixing, passage to gauge invariants, and on-shell reduction, with the same rooted trees playing the role of perturbative diagrams.

5. Mathematical applications beyond the basic theorem

Outside effective field theory, model homotopy transfer plays a substantial role in rational homotopy theory. For a cocommutative dg-coalgebra $$p\circ i=\mathrm{id}_H,\qquad \mathrm{id}_A-i\circ p=d_Ah+hd_A.$$7 modeling a finite nilpotent CW-complex $$p\circ i=\mathrm{id}_H,\qquad \mathrm{id}_A-i\circ p=d_Ah+hd_A.$$8, homotopy transfer along a simplicial deformation retract onto homology $$p\circ i=\mathrm{id}_H,\qquad \mathrm{id}_A-i\circ p=d_Ah+hd_A.$$9 produces a cocommutative $L_\infty$00-coalgebra structure

$L_\infty$01

whose higher operations are the higher Massey coproducts. Applying Quillen’s functor to this transferred coalgebra recovers the Quillen minimal Lie model of $L_\infty$02. The same paper transfers a convolution $L_\infty$03-structure to

$L_\infty$04

thereby modeling the rational homotopy type of $L_\infty$05 and deriving criteria for mapping-space components to be rational $L_\infty$06-spaces (Buijs et al., 2012).

Transfer methods also extend to categorical and properadic settings. For $L_\infty$07-pre-Calabi–Yau categories, a quasi-isomorphism of dg-quivers together with contraction data yields transferred pre-Calabi–Yau structures and transferred morphisms via sum-over-trees formulas; as a consequence, every pre-Calabi–Yau category has a minimal model, and every quasi-isomorphism admits a quasi-inverse (Boucrot, 2023). This shows that transfer remains operative even when the governing operations are cyclic and multi-object rather than concentrated in a single algebra.

A different but related direction studies when transfer preserves strictness or can be combined with perturbation. Given a deformation retract of chain complexes and a dg-algebra structure on the larger complex, one generally transfers only an $L_\infty$08-structure to the smaller complex, again by summing over planar rooted trees. However, if the homotopy satisfies a generalized Leibniz rule, all higher transferred $L_\infty$09 for $L_\infty$10 vanish and the smaller complex becomes a genuine dg-algebra. After perturbing the differential, the Perturbation Lemma produces corrected data $L_\infty$11 and a transferred product

$L_\infty$12

which in the application to a truncated Koszul double complex yields an associative, unital, graded-commutative, and strict dg-algebra structure on the Buchsbaum–Eisenbud resolution (Miller et al., 2020).

These applications indicate that model homotopy transfer is best viewed as a general mechanism for replacing large algebraic or categorical objects by smaller homotopy-equivalent models while retaining the relevant higher operations explicitly enough to compute with them.

6. Alternatives, obstructions, and terminological ambiguity

The transfer paradigm is not exhaustive. Cremonini and Marotta propose an alternative construction for $L_\infty$13-algebras based on module structure rather than the standard transfer procedure. Under the full side conditions $L_\infty$14, $L_\infty$15, $L_\infty$16 and provided $L_\infty$17, the resulting $L_\infty$18-algebra is strictly quasi-isomorphic to the usual homotopy-transfer $L_\infty$19-algebra. If $L_\infty$20 or some side conditions are abandoned, the existence of an $L_\infty$21-quasi-isomorphism with the homotopy-transfer algebra can be obstructed, and the new structure is not an infinitesimal Hochschild deformation of the transferred one (Cremonini et al., 2024). This is an important correction to any misconception that all reasonable higher structures on a smaller complex must be transfer-equivalent.

A separate source of ambiguity is terminological. In legged-robot control, “Model Homotopy Transfer” names a continuation-based learning framework rather than a theorem in homotopical algebra. There, a scalar $L_\infty$22 continuously morphs a single-rigid-body model into a full-body model by redistributing mass and inertia while keeping total mass constant; PPO is warm-started along a schedule of homotopy points; and the method is reported to reach the final return in approximately $L_\infty$23 PPO iterations on a Wall-Assisted Backflip task, compared to $L_\infty$24 for Imitation Transfer, with a min–max spread of $L_\infty$25 versus $L_\infty$26 for Direct Transfer (Kang et al., 31 Dec 2025). This usage is mathematically unrelated to transfer theorems for $L_\infty$27- or $L_\infty$28-structures, although both employ a homotopy parameter to organize passage from a simpler model to a more complete one.

This suggests that the phrase “model homotopy transfer” is polysemous across disciplines. In homotopical algebra and mathematical physics it denotes a theorematic and categorical mechanism for transferring higher algebraic structure along homotopy equivalences. In robotics it denotes a continuation strategy for policy transfer across dynamical models. The shared vocabulary should not obscure the difference in objects, invariants, and proof technology.