Modification (Higher Category Theory)¶
An object-indexed family of 2-cells comparing parallel transformations of higher functors, constrained by compatibility along every source 1-cell.
Core Idea¶
A modification compares two parallel transformations of higher functors. Given pseudofunctors \(F,G:\mathcal B\to\mathcal C\) between bicategories and pseudotransformations \(\eta,\theta:F\Rightarrow G\), it assigns each object \(a\) of \(\mathcal B\) a 2-cell \(\Gamma_a:\eta_a\Rightarrow\theta_a\) in \(\mathcal C\). These cells must fit the transformations' own comparison data along every 1-cell \(f:a\to b\). Thus local 2-cells become one coherent higher comparison.[1]
The construction also has a strict form between 2-natural transformations of 2-functors. In that form it occupies the next comparison level after transformations in the 3-category whose lower levels are 2-categories and 2-functors. The strict and pseudo settings share the object-indexed family and compatibility requirement, though the equations use their respective transformation data.[2][1] A parallel pair need not admit any modification at all; existence requires suitable 2-cells satisfying the law.
How would you explain it like I'm…
Comparing Two Comparisons
Coherent Transformation Comparison
Structural Signature¶
Sig role-phrases: parallel higher functors → parallel transformations → one 2-cell per source object → compatibility along every source 1-cell.
- Two bicategories. The source supplies objects and 1-cells to index the data; the target supplies the 2-cells. Without target 2-cells there is no type for a component.[1]
- Parallel pseudofunctors \(F,G\). Both transformations must compare the same higher functors. Changing one endpoint makes \(\eta_a\) and \(\theta_a\) generally incomparable.[1]
- Parallel pseudotransformations \(\eta,\theta\). Each has an object component and its own comparison 2-cell associated with a source 1-cell. A modification compares these transformation components, rather than replacing either pseudofunctor.[1]
- Object-indexed 2-cells \(\Gamma_a\). Each cell goes from \(\eta_a\) to \(\theta_a\). A cell at just one object is insufficient unless that is the full source object set.[1]
- Compatibility law. For each source 1-cell, the two ways to combine a modification component with the pseudotransformations' comparison cells agree. This is an equation to check, not another independent datum \(\Gamma_f\).[1]
What It Is Not¶
An ordinary natural transformation relates functors by an object-indexed family of arrows satisfying naturality. A modification instead relates transformations and uses an object-indexed family of 2-cells. Merely having a 2-cell between \(\eta_a\) and \(\theta_a\) at every object is a near miss: if one compatibility equation fails, the family is not a modification.[1]
The term does not mean a general edit or alteration of an object. A diagram containing a 3-cell label is also insufficient unless its component types and compatibility law are specified.[1] The frozen encyclopedia source presents an extra morphism-indexed \(\mu_f\); the inspected definition instead specifies object-indexed components and a law for each 1-cell.[1]
Scope of Application¶
In strict higher category theory, modifications appear between 2-natural transformations of 2-functors. Lambert identifies them as one level of the 3-category \(2\mathrm{Cat}\). Miranda constructs particular ones from the source, target, identity, and composition maps of internal categories; their components vary over categories with finite limits.[2][3]
In a bicategorical setting, Ahrens and collaborators construct modifications from paths between pointwise homotopies of 2-types. The ambient bicategories are fundamental bigroupoids. Their construction shows that the same comparison rule applies to homotopy data rather than internal-category structure maps.[1] Neither setting licenses a claim that arbitrary parallel transformations have a modification.
Clarity¶
Write the types before drawing a comparison: \(\mathcal B,\mathcal C\), then \(F,G\), then \(\eta,\theta\), then \(\Gamma_a\). An object component of a pseudotransformation is a 1-cell; a component of its modification is a 2-cell between such 1-cells. The law indexed by \(f\) tells how those object components cohere; it does not give a separate component attached to \(f\).[1]
Also declare whether the transformations are strict 2-natural or pseudo. For pseudo transformations, the compatibility diagram includes their specified comparison 2-cells. Reusing the strict equation without those cells would misstate the instance.[1][2]
Manages Complexity¶
The construction packages many local 2-cell comparisons as one typed object in a higher hom structure. Instead of saying that each \(\eta_a\) can be compared with \(\theta_a\) and then hoping those comparisons fit together, it makes the fitting condition part of the definition. Miranda's source and target maps illustrate this compression: maps belonging to each internal category become components of modifications that respect a change of ambient finite-limit category.[3]
The package does not manufacture a comparison where no suitable 2-cell family exists. It organizes the proof obligation: choose the components, then check the equation for each relevant 1-cell.[1]
Abstract Reasoning¶
Given a proposed modification, first check parallelism of the functors and transformations. Type each proposed \(\Gamma_a\) as a 2-cell. Then, for an arbitrary source 1-cell \(f:a\to b\), compare the two composites in the definition's compatibility diagram. A failure there identifies the exact reason an otherwise plausible pointwise comparison cannot be promoted to a modification.[1]
In Miranda's strict example, the test is whether an ambient finite-limit-preserving functor transports the internal source, target, identity, and composition maps coherently; Theorem 3.1.9 checks the modification axiom.[3] In the 2-type example, the source constructs a modification from a supplied family of paths between two pointwise homotopies; it does not assume any two homotopies are connected.[1]
Knowledge Transfer¶
The recognition test moves between two unlike mathematical constructions. Internal-category evaluation uses natural transformations built from structural maps. Fundamental bigroupoids use paths between homotopies. In both, the essential question is whether an object-indexed 2-cell family respects the higher transformations along source 1-cells.[3][1]
This transfer stays inside higher category theory. Similar language about modifying a workflow or a physical object does not inherit the 2-cell types or coherence equation.[1]
Examples¶
Canonical: internal-category evaluation¶
Let \(\mathrm{Lex}\) be the 2-category of categories with finite limits and finite-limit-preserving functors. Miranda's \(\mathrm{Cat}(-)_1\) and \(1_{\mathrm{Lex}}\) are parallel 2-functors. The evaluation constructions \(\mathrm{ev}_1\) and \(\mathrm{ev}_0\) are parallel 2-natural transformations. The source and target maps of an internal category supply the components of two distinct modifications \(d_0,d_1:\mathrm{ev}_1\Rightarrow\mathrm{ev}_0\). At an ambient category \(\mathcal E\), each component is a natural transformation between evaluation functors; at an internal category \(C\), that component gives the source or target map \(C_1\to C_0\). Theorem 3.1.9 checks that these maps fit when \(\mathcal E\) changes by a finite-limit-preserving functor.[3]
Mapped back: \(\mathrm{Lex}\) provides source 1-cells, the two 2-functors fix endpoints, \(\mathrm{ev}_1,\mathrm{ev}_0\) are the compared transformations, and \(d_0\) or \(d_1\) is one compatible object-indexed 2-cell family. They are not modifications between each other.[3]
Applied: paths in fundamental bigroupoids¶
Let \(X,Y\) be 2-types and \(f,g:X\to Y\) be functions. Ahrens and collaborators obtain pseudofunctors \(\pi(X)\to\pi(Y)\). Pointwise homotopies \(e_1,e_2\) from \(f\) to \(g\) yield parallel pseudotransformations. Given a family of paths \(h_x:e_1(x)=e_2(x)\), the source constructs a modification whose component at \(x\) is \(h_x\). The construction depends on the supplied path family; it does not assert its existence for arbitrary \(e_1,e_2\).[1]
Mapped back: the fundamental bigroupoids provide the source objects and 1-cells and target 2-cells; \(f,g\) fix the pseudofunctor endpoints; \(e_1,e_2\) are the parallel transformations; \(h_x\) are object components satisfying the induced compatibility.[1]
Structural Tensions¶
T1: Pointwise choice vs. global compatibility. Objectwise 2-cells can be chosen locally, but the defining 1-cell equation limits which families assemble into a modification.[1] Diagnostic: Which 1-cell, if any, makes the proposed family fail the compatibility diagram?
T2: Strict calculation vs. pseudo coherence. The strict setting treats 2-natural transformation data, while pseudotransformations have specified comparison 2-cells that the modification law must respect.[2][1] Treating every comparison as an identity would lose legitimate bicategorical cases. Diagnostic: Does the instance have nonidentity comparison cells, and were they included in the modification check?
Structural–Framed Character¶
The entry sits toward the structural end of the spectrum. Its admission test is a typed family of 2-cells plus a compatibility law, not an assessment of whether the result is useful or desirable.[1] Evaluative weight therefore does not determine membership. Human practice matters when mathematicians choose a formal setting or application, but no particular practice creates the objectwise law. The definition has no institutional origin as a rule of governance or professional procedure. The word “modification” travels readily through ordinary language, yet that vocabulary travel does not carry the higher-categorical 2-cell types into unrelated domains. The abstraction is recognized in the structure of parallel higher transformations, while its presentation is imported from the formal vocabulary of higher category theory.[1][3]
Its character: structural within higher category theory; its typed coherence condition fixes membership, and its portable-sounding name does not confer broader application.
Structural Core vs. Domain Accent¶
The skeletal relation is that a family of local comparisons must respect connections among the things compared. That phrasing may help recognize a resemblance across domains, but the named modification has a domain-bound mechanism: 2-cells between transformation components, composed and whiskered against each source 1-cell's comparison data.[1] Remove those types and the compatibility law, and the object described is no longer this higher-categorical modification.
Could that broader “compatible family of comparisons” someday warrant a Prime? That is a future-Prime question, not an admission claim here. It would require several unlike cross-domain realizations with every role mapped, a stable diagnostic and failure mode across them, and a check that no live Prime already captures the same pattern. The internal-category and 2-type cases are materially unlike but both remain within higher category theory.[3][1] They establish this domain-specific entry, not the cross-domain, nonduplicate evidence for a new Prime.
Instantiates / Related Primes¶
The catalog's Relation and Comparison primes are useful loose analogues: a modification relates two parallel transformations in a typed setting. Their general signatures do not supply the necessary 2-cell and coherence data, so this paragraph records a conceptual bridge rather than a typed parent edge. The word overlap with Transformation and Identity-Preserving Modification likewise does not prove subsumption; those entries describe different operations or changes. Compatibility is a more substantial near neighbor: nontrivial modification equations require the local 2-cells to fit. Yet for a terminal source 2-category the law is automatic and a single target 2-cell is already a modification, so the full Compatibility signature is not a strict constitutive part of every case. The graph therefore records an unparented root pending later densification.[1]
Neighborhood in Abstraction Space¶
Modification (Higher Category Theory) sits in a moderately populated region (58th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.
Family — Unclustered & Miscellaneous (2835 abstractions)
Nearest neighbors
- Monoidal Natural Transformation — 0.86
- Direct Limit — 0.85
- 2-Category — 0.85
- Beck–Chevalley Condition — 0.85
- Strictification — 0.84
Computed from structural-signature embeddings · 2026-10-31
Not to Be Confused With¶
- An ordinary natural transformation: its components are arrows between functor values, one comparison level lower.
- A pseudotransformation: this is one of the two things being compared; its object components are 1-cells.
- A strict 2-category: it is one possible ambient structure, while bicategorical modifications need not be strict cases.
- An arbitrary family of 2-cells: the 1-cell compatibility equation can fail.
- A change or revision: ordinary usage lacks the higher-categorical types.
References¶
[1] Benedikt Ahrens, Dan Frumin, Marco Maggesi, Niccolò Veltri, and Niels van der Weide (2022), Bicategories in univalent foundations, Mathematical Structures in Computer Science, original open full article, Definitions 2.12–2.14 and Examples 2.6 and 2.15. Definition 2.14 gives the component and compatibility law; Example 2.15 constructs the path-family instance. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t ↩u ↩v ↩w ↩x ↩y ↩z ↩27
[2] Michael Lambert (2024), Discrete 2-Fibrations, Higher Structures 8(1), 54–96, original full journal PDF, Example 4.17 at zero-based PDF pp.32–33. It identifies modifications as the next level in strict \(2\mathrm{Cat}\), without itself supplying a worked modification example. registry ↩a ↩b ↩c ↩d
[3] Adrian Toshar Miranda (2018), Internal Categories, Macquarie University Masters by Research thesis, original full PDF, Remark 1.2.6 at zero-based PDF p.10, Proposition 3.1.1 at PDF p.27, and Theorem 3.1.9 at PDF pp.29–30. The theorem lifts evaluation maps to modifications and checks their coherence. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h