Category theory promises a unified language where abstract structures can be treated as concrete objects (via representability) and complex systems can be simplified to focus on essential behaviors (via localization). However, these powerful mathematical tools do not come for free. They are deeply tethered to foundational constraints: the set-theoretic size of the collections we work with, the structural completeness of the underlying categories, and the specific properties of the functors connecting them. When these constraints are violated, the elegant machinery of category theory can break down.
The Size and Set-Theoretic Barrier to Representability
Why does size matter in category theory? In standard applications, we assume that the collection of maps (morphisms) between any two objects forms a manageable "set." But when a category's connections grow too large—becoming a "proper class"—fundamental theorems collapse.
As shown by Neeman in Brown representability does not come for free, when hom-collections become proper classes, a category can lose its ability to support adjoint functors and representable objects. This creates a direct barrier to category simplification: if a subcategory cannot admit a Bousfield localization because its morphisms are too large, it prevents the construction of representable objects. We lose the ability to treat abstract, cohomology-like functors as concrete objects within the category.
[Morphisms expand to a Proper Class]
│
▼
[Localization Techniques Fail]
│
▼
[Cannot Construct Representable Objects]
To resolve these size issues, researchers must sometimes look beyond standard category theory to high-level logical assumptions. For instance, Bagaria et al. demonstrate in Epireflections and supercompact cardinals that large cardinal axioms from set theory (specifically supercompact cardinals) can step in to simplify complex categorical structures. They show that if a categorical property can be expressed by an "absolute formula" (one invariant across different models of set theory), the existence of supercompact cardinals guarantees that massive epireflective classes can be reduced to "small-orthogonality" classes defined by a finite set of morphisms. Without these strong set-theoretic assumptions, such structural simplifications remain unprovable.
The Structural Deficit of Incompleteness
Even when size is not an issue, a category can fail to support standard theorems if it lacks basic structural components like limits, colimits, or split idempotents. Limits (such as products or equalizers) allow us to construct intersections and equilibria, while colimits (such as coproducts or pushouts) allow us to glue objects together.
When these are missing, the category is structurally incomplete, severely limiting the applicability of general categorical theorems. For example, delle Rose et al. study this in The category of equivalence relations, demonstrating that the category of computable equivalence relations lacks an initial object and general equalizers. This structural deficit completely blocks researchers from using standard high-level categorical tools to study computability.
Similarly, in the study of feedback and recursion, Vicary et al. point out in Traced Monoidal Categories as Algebraic Structures in Prof that algebraic frameworks for traced categories are heavily restricted by the assumption of Cauchy completeness (the requirement that all idempotents split). If a category has not been explicitly completed—such as through a Karoubi envelope—modeling feedback loops becomes a major technical hurdle.
Even in applied domains like data analysis, structural obstructions arise from the underlying data math. Chowdhury et al. demonstrate in Hypergraph co-optimal transport: metric and categorical properties that measure-theoretic issues and the nature of expansive maps obstruct standard categorical limits like pullbacks and pushouts, forcing researchers to seek alternative categorical frameworks to compare hypergraphs.
How Constraints Propagate to Enriched and Homotopical Settings
When we move from standard categories to enriched or homotopical categories—where morphisms themselves carry extra algebraic, quantitative, or topological structure—these foundational dependencies become even more acute.
In computer science, building quantitative models (like a graded $\lambda$-calculus that tracks program execution time or probability) requires specialized categorical tools. Dahlqvist et al. show in A Complete V-Equational System for Graded lambda-Calculus that constructing quantitative comonads depends strictly on the existence of specific enriched limits within the underlying category. If the category lacks these limits, the quantitative framework cannot be built.
A similar dependency appears in higher representation theory and physics. Kong et al. show in Enriched monoidal categories I: Centers that the existence of generalized centers for enriched categories is never guaranteed; it hinges entirely on the precise properties of the underlying categories and functors, failing in many standard environments.
Finally, in algebraic topology, the lack of cocompleteness can break the universality of geometric models. As noted by Medina-Mardones et al. in Multisimplicial chains and configuration spaces, algebraic frameworks for multisimplicial sets often lack cocompleteness. This means that while higher homotopy structures work beautifully for specific models like configuration spaces, they cannot be automatically guaranteed for arbitrary discrete topological representations.
This forces a fundamental trade-off, much like the one described by Gutiérrez in Homotopical localizations of module spectra: researchers must constantly choose between broad, model-independent frameworks and the deeper structural properties that only emerge within highly specific, well-behaved categories.
Key Foundational Work
Epireflections and supercompact cardinals (2009) — contributed 13 concept elements to this direction.
Brown representability does not come for free (2009) — contributed 12 concept elements to this direction.
Homotopical localizations of module spectra (2004) — contributed 1 concept elements to this direction.