A calculus of types in Isbell nuclei
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Computer Science. The day in computer science: software, systems and theory on arXiv.
Tom: We're looking at the paper: "A calculus of types in Isbell nuclei".
Jane: Here's the gist of the paper. A CALCULUS OF TYPES IN ISBELL NUCLEI identifies two constructions from different mathematical traditions as being the same objects The shared datum is minimal: an associative product, called execution, and a real-valued measurement, with no compatibility assumed between them This paper shows that, in the present setting, they produce the same objects Construction of Logical Types In linear logic and realisability, logical types are generated rather than fixed in advance one begins with a universe of realisers equipped with execution, uses orthogonality to test their interactions, and takes types to be the biorthogonally closed subsets The authors contributed equally to this work.
Tom: Construction via Isbell Duality In enriched Isbell duality, a quantitative relation induces an adjunction whose fixed points form a category, its nucleus The types of the first construction and the fixed points of the second arise by different means in different settings.
Jane: Identification and Consequences The identification pays off in both directions From the Isbell side, the finite real case inherits the projective metric and polyhedral geometry of the companion paper
GJST26: From the realisability side, the execution product contributes the oriented logical structure of insertion.
Tom: The Middle Calculus The middle closed product is defined as (16) A ⊙ B:= clmid(AB) = ‚((AB)‚ On middle types, the execution product has an associative core of the calculus The residuals are middle types and are characterized by A ⊙ B ⊆ C ⇐⇒ B ⊆ A ⊸l C ⇐⇒ A ⊆ B ⊸r C Derived Nuclei and Coherence The middle type is represented, using the profile dictionary of Sections 4 and 6, by a point (f, g) ∈ Nuc(N), where f: C → R is the presheaf coordinate and g = N∗f: C × C → R is the copresheaf coordinate The derived nucleus of a middle point is computed by the two residuals of the middle Lambek calculus The common intersection is therefore explicit: it is the locus where the three one-coordinate residual equations of M3 are simultaneously saturated Product Envelopes and Geometry The product gap (24) ∆X,Y (y; x, z):= M3(x, y, z) − f′′(y) − g′′(x, z) is independent of the chosen affine representatives of
X: and
Y: This operation is a coordinate-level operation—to get a nucleus point requires applying the relevant Isbell closure The gap decomposes into input witness gaps and outer minimization slacks Unit and Residual Structure The middle unit formula gives the middle profile The running example distinguishes the left, right, and middle unit profiles The final displayed identity is obtained by expanding M3(x, b, z) and M(x, z) The value −7 is attained in the middle unit formula by the peripheral pair (x, z) = (c, a): the multiplication changes from ca = d to cda = c Derived Nuclei in Presheaf Coordinates The second coordinate g is itself a binary profunctor g: C ↛ C, which has its own Isbell nucleus The first result of the section shows that the nucleus derived from a middle point is computed by the two residuals f ⊸l and f ⊸r of the Lambek calculus The second coordinate g is itself a binary profunctor g: C ↛ C, which has its own Isbell nucleus Boundary Arrangements The three one-variable arrangements of M3 give three ways of producing triples (u, f, v) The common intersection is therefore explicit: it is the locus where the three one-coordinate residual equations of M3 are simultaneously saturated The two-out-of-three theorem says that the three arrangements have the same pairwise intersection: compatibility with any two arrangements forces compatibility with the third Sharpness of Two-Out-of-Three Hypothesis The following example shows that the two-out-of-three hypothesis in Theorem 56 cannot be weakened to membership in a single arrangement The same example also separates derived nuclearity from binary nuclearity for the original kernel M Product Gap Identities The product gap (25) κ b,b′ X,Y (x, z):= M4(x, b, b′, z) − f(b) − f'(b') − g''(x, z) The nonnegativity of these two slacks follows from the equivalent one-sided forms of the envelope The product gap (24) is independent of the chosen affine representatives of
X: and
Y: The final displayed identity is obtained by expanding M4(x, b, b′, z) and M(b′, z) References
AL21: Tom Avery and Tom Leinster, Isbell conjugacy and the reflexive completion, Theory and Applications of Categories 36 (2021), no.
Paper summary: Tom: So we’ve covered how they identified these two constructions in "A calculus of types in Isbell nuclei." They started with linear logic and realisability where types are generated through realizers, and then looked at enriched Isbell duality where they find fixed points forming a nucleus.
Jane: And the paper shows that despite starting from those very different mathematical traditions, these two ways of constructing logical types end up producing the same objects.
Lu: The core mechanism is that the shared datum is minimal: an associative product called execution, and a real-valued measurement, with no compatibility assumed between them. This simple setup allows for things like noncommutative execution and asymmetric linear implication.
Meng: From my side, I’m curious about the practical implication of this identification. If we can rely on this structural match, does it mean something about how robust our AI models are when we try to define types?
Tom: It suggests that the oriented logical structure of insertion in realisability is deeply linked to the quantitative relation defining the Isbell nucleus. It provides a solid framework for building these types.
Jane: So, in simple terms, they are confirming that there’s a fundamental equivalence between how we generate types using logical rules and how we find fixed points using geometric tools.
Lu: The paper shows this identification pays off in both directions, validating the idea that the structures generated by different means converge on the same mathematical object.
Tom: It’s a very clean result because it doesn't require any special compatibility assumptions between the execution product and the measurement to make this work.
Jane: And they spend a lot of time detailing how they calculate things like these product gaps and derived nuclei, showing that these complex structures are stable and well-behaved.
Meng: I’m just thinking about the cost of running something like this at scale. If we need to implement this kind of type calculus, does the complexity they describe translate to a massive overhead for deployment?
Tom: The paper focuses on the structure itself, not necessarily on an optimized implementation detail, but it shows that we can define these types without immediately running into those compatibility headaches.
Lu: So, the big picture is that this calculus provides a unified mathematical language for understanding how logic and geometry interact when you look at these specific structures.
Jane: It gives us a way to see the relationship between different mathematical ideas as being fundamentally connected through this shared execution and measurement concept.
Conclusion: Tom: So we've been looking at this paper, "A calculus of types in Isbell nuclei," and now we need to talk about what all that math actually means for us on a regular Tuesday.
Jane: Right. We’ve seen how they connected two different ways of building logical types—one from linear logic and one from geometric duality—and this paper is the final wrap-up on why those two paths lead to the same thing.
Lu: The authors did a lot of deep work defining these constructions, focusing on how an associative product and a real-valued measurement interact without needing any special rules between them.
Meng: So they found that even though we started with different setups, the resulting objects—the types themselves—turn out to be identical in this specific setting.
Tom: Exactly. It’s about showing that these two seemingly separate mathematical worlds are actually describing the same underlying structure when you look at it through this lens of execution and measurement.
Jane: The authors show that this identification works both ways, which is pretty cool because it validates the logic coming from realisability by showing how it connects to enriched category theory.
Lu: It's a neat result because it doesn't require any compatibility assumptions between the execution product and that measurement; they just work together as given.
Meng: For me, what this means is that if we can find an associative product and a measurement in our AI systems, we don't have to worry about whether the way we define types logically matches how they are geometrically structured.
Tom: It simplifies things by showing that the logical structure of insertion in realisability is directly tied to the geometric properties of these Isbell nuclei.
Jane: So, basically, this paper is proving a deep structural equivalence between how we reason about logic and how we measure things geometrically.
Lu: The middle calculus they define, where you have something like "A times B" defined as the closure of A times B, is central to showing this consistency across the whole system.
Meng: So for anyone building complex systems, it suggests a very solid foundation for defining these kinds of structured types without having to guess how they should interact.
Tom: Yeah, it’s less about a new trick and more about confirming that the basic components we use—the product and the measurement—fit together in a very specific, consistent way.
Jane: It gives us confidence that when we define these kinds of complex logical structures, we aren't just making up rules; we're hitting a spot where logic and geometry naturally align.
Lu: The paper also lays out some really concrete geometry, like the product gap identities, which shows precisely how the inputs and outputs scale together in this new calculus.
Meng: I wonder if that geometric scaling stuff translates well when you try to run these things on massive datasets; we’ll have to see if those gaps stay manageable when things get big.
Tom: We'll probably need to look at those scaling aspects next, because showing it works for small cases doesn't tell us much about production deployment.
JUAN LUIS GASTALDI, SAMANTHA JARVIS, THOMAS SEILLER, JOHN TERILLA
CUNY Queens College · CNRS
cs.LO, math.CT, math.LO
Submitted: 2026-06-02
Updated: 2026-11-06
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 79/100
The gist: A CALCULUS OF TYPES IN ISBELL NUCLEI identifies two constructions from different mathematical traditions as being the same objects The shared datum is minimal: an associative product…
Key concepts
- Logical Types
- In linear logic and realisability, logical types are not fixed but generated. They arise from starting with realizers equipped with an execution mechanism, testing their interactions via orthogonality, and defining types as biorthogonally closed subsets of these realizers.
- Isbell Duality Nucleus
- This construction uses a quantitative relation to induce an adjunction, whose fixed points form a category called the nucleus. The paper shows that the types from this method arise differently than those from the logical construction, yet they are ultimately equivalent in this context.
- Middle Calculus Product ($oldsymbol{oxplus}$)
- The middle closed product is defined as $ ext{clmid}(AB)$, which has an associative core. Its residuals characterize the relationship between types: $A oxplus B ot o C$ if and only if $B ot o A o C$, showing a specific structure for how these types combine.
- Product Gap ($oldsymbol{ riangle}$)
- The product gap measures the difference between a complex operation ($ ext{M3}$) and simpler coordinate functions ($f''$ and $g''$). This gap is independent of the chosen affine representatives of the objects, decomposing into input witness gaps and outer minimization slacks.
Terminology
Summary
A CALCULUS OF TYPES IN ISBELL NUCLEI identifies two constructions from different mathematical traditions as being the same objects The shared datum is minimal: an associative product, called execution, and a real-valued measurement, with no compatibility assumed between them This paper shows that, in the present setting, they produce the same objects
Construction of Logical Types
In linear logic and realisability, logical types are generated rather than fixed in advance one begins with a universe of realisers equipped with execution, uses orthogonality to test their interactions, and takes types to be the biorthogonally closed subsets The authors contributed equally to this work.
Construction via Isbell Duality
In enriched Isbell duality, a quantitative relation induces an adjunction whose fixed points form a category, its nucleus The types of the first construction and the fixed points of the second arise by different means in different settings.
Identification and Consequences
The identification pays off in both directions From the Isbell side, the finite real case inherits the projective metric and polyhedral geometry of the companion paper [GJST26] From the realisability side, the execution product contributes the oriented logical structure of insertion.
The Middle Calculus
The middle closed product is defined as (16) A ⊙ B:= clmid(AB) = ‚((AB)‚ On middle types, the execution product has an associative core of the calculus The residuals are middle types and are characterized by A ⊙ B ⊆ C ⇐⇒ B ⊆ A ⊸l C ⇐⇒ A ⊆ B ⊸r C
Derived Nuclei and Coherence
The middle type is represented, using the profile dictionary of Sections 4 and 6, by a point (f, g) ∈ Nuc(N), where f: C → R is the presheaf coordinate and g = N∗f: C × C → R is the copresheaf coordinate The derived nucleus of a middle point is computed by the two residuals of the middle Lambek calculus The common intersection is therefore explicit: it is the locus where the three one-coordinate residual equations of M3 are simultaneously saturated
Product Envelopes and Geometry
The product gap (24) ∆X,Y (y; x, z):= M3(x, y, z) − f′′(y) − g′′(x, z) is independent of the chosen affine representatives of [X] and [Y] This operation is a coordinate-level operation—to get a nucleus point requires applying the relevant Isbell closure The gap decomposes into input witness gaps and outer minimization slacks
Unit and Residual Structure
The middle unit formula gives the middle profile The running example distinguishes the left, right, and middle unit profiles The final displayed identity is obtained by expanding M3(x, b, z) and M(x, z) The value −7 is attained in the middle unit formula by the peripheral pair (x, z) = (c, a): the multiplication changes from ca = d to cda = c
Derived Nuclei in Presheaf Coordinates
The second coordinate g is itself a binary profunctor g: C ↛ C, which has its own Isbell nucleus The first result of the section shows that the nucleus derived from a middle point is computed by the two residuals f ⊸l and f ⊸r of the Lambek calculus The second coordinate g is itself a binary profunctor g: C ↛ C, which has its own Isbell nucleus
Boundary Arrangements
The three one-variable arrangements of M3 give three ways of producing triples (u, f, v) The common intersection is therefore explicit: it is the locus where the three one-coordinate residual equations of M3 are simultaneously saturated The two-out-of-three theorem says that the three arrangements have the same pairwise intersection: compatibility with any two arrangements forces compatibility with the third
Sharpness of Two-Out-of-Three Hypothesis
The following example shows that the two-out-of-three hypothesis in Theorem 56 cannot be weakened to membership in a single arrangement The same example also separates derived nuclearity from binary nuclearity for the original kernel M
Product Gap Identities
The product gap (25) κ b,b′ X,Y (x, z):= M4(x, b, b′, z) − f(b) − f'(b') − g''(x, z) The nonnegativity of these two slacks follows from the equivalent one-sided forms of the envelope The product gap (24) is independent of the chosen affine representatives of [X] and [Y] The final displayed identity is obtained by expanding M4(x, b, b′, z) and M(b′, z)
References
[AL21] Tom Avery and Tom Leinster, Isbell conjugacy and the reflexive completion, Theory and Applications of Categories 36 (2021), no. 12, 306–347
[BGT24] Tai-Danae Bradley, Juan Luis Gastaldi, and John Terilla, The structure of meaning in language: Parallel narratives in linear algebra and category theory, Notices of the American Mathematical Society 71 (2024), no. 2, 174–185
[BTV22] Tai-Danae Bradley, John Terilla, and Yiannis Vlassopoulos, An enriched category theory of language: From syntax to semantics, La Matematica 1 (2022), no. 2, 551–580
[BV25] Tai-Danae Bradley and Juan Pablo Vigneaux, The magnitude of categories of texts enriched by language models, Theory and Applications of Categories 44 (2025), no. 37, 1256–1281
[DS04] Mike Develin and Bernd Sturmfels, Tropical convexity, Documenta Mathematica 9 (2004), 1–27
[Ell17] Jonathan Arthur Elliott, On the fuzzy concept complex, Ph.D. thesis, University of Sheffield, Sheffield, UK, 2017
[Fuj19] Soichiro Fujii, Enriched categories and tropical mathematics, arXiv e-prints (2019)
[Gir87] Jean-Yves Girard, Linear logic, Theoretical Computer Science 50 (1987), no. 1, 1–101
[Gir01], Locus solum: From the rules of logic to the logic of rules, Mathematical Structures in Computer Science 11 (2001), no. 3
[Gir11] Jean-Yves Girard, Geometry of interaction V: Logic in the hyperfinite factor., Theoretical Computer Science 412 (2011), 1860–1883
[Isb60] J. R. Isbell, Adequate subcategories, Illinois Journal of Mathematics 4 (1960), no. p. 4 4, 541–552
[Jar25] Samantha K. Jarvis, A novel closed monoidal structure on the nucleus of a profunctor, Ph. p. 4D., The Graduate Center, City University of New York, June 2025
[Kel82] G. Max Kelly, Basic concepts of enriched category theory, London Mathematical Society Lecture Note Series, vol. p. 0 64, Cambridge University Press, 1982
[Kle45] Stephen C. Kleene, On the interpretation of intuitionistic number theory, Journal of Symbolic Logic 10 (1945)
[Kre59] Georg Kreisel, Interpretation of analysis by means of constructive functionals of finite types, Constructivity in mathematics (Arend Heyting, ed.), North-Holland Pub. Co., 1959, pp. 101–128
[Kri09] Jean-Louis Krivine, Realizability in classical logic, Panoramas et synth'eses 27 (2009), 197–229
[Lam58] Joachim Lambek, The mathematics of sentence structure, The American Mathematical Monthly 65 (1958), no. 3, 154–170
[Law73] F. William Lawvere, Metric spaces, generalized logic, and closed categories, Rendiconti del Seminario Matematico e Fisico di Milano 43 (1973), 135–166
[Sei12] Thomas Seiller, Interaction graphs: Multiplicatives, Annals of Pure and Applied Logic 163 (2012), 1808–1837
[Sei16] Thomas Seiller, Interaction graphs: Additives, Annals of Pure and Applied Logic 167 (2016), 95 – 154
[Sei24] Thomas Seiller, Mathematical Informatics, Ph.D., Sorbonne Paris Nord University, 2024, Habilitation thesis
[vO08] Jaap van Oosten, Realizability, volume 152: An introduction to its categorical side, Elsevier Science, San Diego, CA, USA, 2008
[Wil13] Simon Willerton, Tight spans and Isbell completions and semi-tropical modules, Theory and Applications of Categories 28 (2013), no. 22, 696–732
[Wil14], Galois correspondences and enriched adjunctions, February 2014, Blog post
[Wil15], The Legendre-Fenchel transform from a category theoretic perspective, arXiv eprints (2015), ETH Zurich, Switzerland
How the training signal is built
The product Z:= X ⊙ Y = (f'′, g′′) is represented by the nuclear pair whose copresheaf coordinate is (22) g'' (x, z) = min b,b'∈C M4(x, b, b', z) − f(b) − f'(b') The presheaf coordinate is recovered by (23) f'' (y) = N∗g'' (y) = min x,z∈C M3(x, y, z) − g''(x, z)
Where the gains come from
The product witness is not just a minimizing pair (b, b′).
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Automated derivation of logical types from execution and measurement: The system can generate weighted subsets of C times R closed under a specific orthogonality relation, directly mapping to left and right types as defined by
the biorthogonally closed subsets for this relation.
-
Identification of nuclear fixed points: The system can compute the nucleus Nuc(M) using the Isbell duality adjunction, identifying its objects with pairs (f, g) such that
g = M∗f and f = M∗g,
which are then mapped topaired types PType(M).
-
Derivation of middle product structure: The system can compute the associative middle product symbol A⊙B by taking the
middle closure of the setwise execution product
and verifying its associativity using thetrefoil identity.
-
Computation of derived nuclei: For a middle nuclear point (f, g), the system can compute its derived nucleus Nuc(g) using
the two residuals f ⊸l and f ⊸r of the Lambek calculus,
which are shown to bethe fixed points of the binary Isbell closure operators for M.
-
Analysis of coherence theorems: The system can verify the two-out-of-three coherence theorem by checking if a triple (u, f, v) satisfies
all three equations
(Bu), (Bf), and (Bv) across the three iterated ternary constructions. -
Geometric interpretation of products: For finite real cases, the system can compute product envelopes using the formula M4 to find
the output point Z = (f'', g'') is represented by the nuclear pair whose copresheaf coordinate is (22).
Sources
Related papers
- A Non-Dichotomy for CSPs of Stable omega-Categorical Structures
- A Deeper Look at Depth: Stable Generation Accounting for Quantifier Reasoning
- Formalizing Hyperspaces and Operations on Subsets of Polish Spaces over Abstract Exact Real Numbers
- A Diagrammatic Basis for Computer Programming
- Interpolation via Generalized Splitting
- Staying Productive Under the Palm Trees: On Graded Coeffect Typing in the Tropical Semiring