Day Convolution¶
Lift an indexing category's tensor to a product of functors by left Kan extending their value-wise tensor along it.
Core Idea¶
Day convolution turns functors \(F,G:\mathcal C\to\mathcal V\) into a product \(F\star G\) by first pairing their values and then left Kan extending that external tensor along the indexing category's tensor \(\mathcal C\times\mathcal C\to\mathcal C\). For covariant functors, the contribution to output \(c\) is weighted by maps \(a\otimes b\to c\); a presheaf formula has opposite variance. Suitable colimits and tensor preservation are needed for the coherent monoidal product, and symmetry requires symmetric inputs.[^ref-f2d2122bb071]
Scope of Application¶
The construction occurs literally in set-valued combinatorial species and pointed-simplicial symmetric sequences. In the first, finite label sets combine by disjoint union and values by Cartesian product. In the second, degree splits, pointed smash and symmetric-group actions replace those value operations. The symmetric-spectrum smash is a further tensor relative to the sphere monoid, not the bare symmetric-sequence convolution.[ref-054f4b919a1b][ref-f0c1c6120b45]
Clarity¶
Day convolution is not the pointwise product \(F(c)\otimes G(c)\): it gathers contributions from different indices whose tensor reaches \(c\). Nor does shared vocabulary make it the live numerical Convolution prime, whose fixed sliding-kernel signature is absent. Declaring functor variance matters, because reversing a hom weight while keeping covariant functors changes the formula.[ref-f2d2122bb071][ref-054f4b919a1b]
Manages Complexity¶
The left-Kan-extension recipe packages label partitions or degree splits into one functorial tensor, with associativity, unit and conditional symmetry inherited under appropriate hypotheses. It reduces the design question to the indexing tensor, target tensor, existence of colimits and preservation by tensor; it does not guarantee that evaluating the aggregate is computationally cheap.[ref-f2d2122bb071][ref-f0c1c6120b45]
Abstract Reasoning¶
Baez's species Cauchy product is \((F\cdot_C G)(S)=\coprod_{T\subseteq S}F(T)\times G(S-T)\): each ordered split of labels contributes. Hovey, Shipley and Smith's symmetric-sequence tensor instead combines \(X_p\wedge Y_q\) over \(p+q=n\) with induced \(\Sigma_n\)-actions. Both map the same five roles—indexing tensor, target tensor, input functors, Kan aggregation and coherent output—to different concrete carriers. The spectrum smash adds \(S\)-module balancing afterward.[ref-054f4b919a1b][ref-f0c1c6120b45]
Knowledge Transfer¶
The construction transfers literally among suitably monoidal functor categories after rechecking variance and colimit hypotheses. The broader idea of transporting an operation by universal aggregation is a future-prime question, not a license to import Day convolution into any numerical decomposition sum. No strict DAG parent is proposed: live Kan Extension is a related constructional operation, whereas live Convolution has a different full signature.[^ref-f2d2122bb071]
[^ref-f2d2122bb071]: Saul Glasman, “Day convolution for ∞-categories,” revised original author version v4, introduction and §2 Definition 2.8. [^ref-054f4b919a1b]: John C. Baez, “Dirichlet Species and Arithmetic Zeta Functions”, original author PDF, §3.1 Eq. (2). [^ref-f0c1c6120b45]: Mark Hovey, Brooke Shipley and Jeff Smith, “Symmetric Spectra”, original author PDF, §2.1 Definition 2.1.3 and Remark 2.1.5; §2.2 Definition 2.2.3.
Neighborhood in Abstraction Space¶
Day Convolution sits in a moderately populated region (56th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.
Family — Categories, Sheaves & Homotopy (21 abstractions)
Nearest neighbors
- Direct Limit — 0.87
- Mac Lane's coherence theorem — 0.87
- Monoidal Natural Transformation — 0.86
- Monoidal Monad — 0.85
- Categorical Trace — 0.85
Computed from structural-signature embeddings · 2026-10-08