Fibrational Versions of Dialectica Categories

Fibrational Versions of Dialectica Categories

Valeria de Paiva

Cuil, Inc.

Stanford Logic Seminar
May 2010

Outline

Introduction

Original Dialectica Categories (de Paiva, 1988)

Original Fibrational Dialectica (Hyland, 2002)

Cartesian Closed Dialectica categories (Biering, 2008)

Discussion

Introduction: What, why?

(Prove existence and describe) Cartesian Closed Dialectica
categories
Chapter 4 of Bodil Biering’s PhD thesis “Dialectica
Interpretations: A Categorical Analysis”, 2008
¨
Why? Categorical understanding of Godel’s Dialectica
Interpretation
What for?
¨
For Godel, the interpretation was a way of proving
consistency of arithmetic, an extension of Hilbert’s
programme
For me (20 years ago) a way of producing models of Linear
Logic from a proven way of understanding logic
For Biering: a way of unifying categorical structures,
brought back to proof theory?...

Biering’s Thesis Abstract

The Dialectica interpretations are remarkable syntactic
constructions
Use these constructions to develop new mathematical
structures: Dialectica categories, the Dialectica- and
Diller-Nahm triposes, and the Dialectica- and Diller-Nahm
toposes
The mathematical structures created from the functional
interpretations provide us with new models for type
theories and programming logics
Studying the mathematical structures we gain new insights
into the syntactical constructions.
Product: Biering et al “Copenhagen interpretation”

Biering’s Thesis Table of Contents

A collection of articles, want to discuss chapter 4...
Introduction
Topos Theoretic Versions of Dialectica Interpretations
A Uniﬁed View on the Dialectica Triposes
Cartesian Closed Dialectica Categories
The Copenhagen interpretation
(BI Hyperdoctrines and Higher Order Separation Logic)

Functional Interpretations

¨
Starting with Godel’s Dialectica interpretation a series of
”translation functions” between theories
Avigad and Feferman on the Handbook of Proof Theory:
¨
This approach usually follows Godel’s original
example: first, one reduces a classical theory C to
a variant I based on intuitionistic logic; then one
reduces the theory I to a quantifier-free functional
theory F.

Examples of functional interpretations:
Kleene’s realizability
Kreisel’s modified realizability
Kreisel’s No-CounterExample interpretation
Dialectica interpretation
Diller-Nahm interpretation

¨
Godel’s Dialectica Interpretation
For each formula A of HA we associate a formula of the form
AD = ∃u∀xAD (u, x) (where AD is a quantifier-free formula of
¨
Godel’s system T) inductively as follows: when Aat is an atomic
formula, then its interpretation is itself.
Assume we have already defined AD = ∃u∀x.AD (u, x) and
B D = ∃v ∀y .BD (v , y ).
We then define:
(A ∧ B)D = ∃u, v ∀x, y.(AD ∧ BD )
(A → B)D = ∃f : U → V , F : U × X → Y , ∀u, y.
( AD (u, F (u, y )) → BD (fu; y))
(∀zA)D (z) = ∃f : Z → U∀z, x.AD (z, f (z), x)
(∃zA)D (z) = ∃z, u∀x.AD (z, u, x)
The intuition here is that if u realizes ∃u∀x.AD (u, x) then f (u)
realizes ∃v ∀y.BD (v , y) and at the same time, if y is a
counterexample to ∃v ∀y .BD (v , y ), then F (u, y) is a
counterexample to ∀x.AD (u, x).

Categorical Dialectica Interpretation

Main references:
The ‘Dialectica’ Interpretation and Categories (P. J. Scott,
Zeit. fur Math Logik und Grund. der Math. 24, 1978)
The Dialectica Categories, (de Paiva, AMS vol 92, 1989)
Proof Theory in the Abstract, (Hyland, APAL 2002)
Dialectica categories are naturally symmetric monoidal
closed, but not cartesian closed categories.
We would like them to be cartesian closed.
Why?
Can we make them Cartesian closed?
Yes, in different ways.

Plan of Biering’s ‘Cartesian Closed Dialectica
Categories’

Recall the deﬁnitions and basic properties of original
dialectica categories
Discuss three approaches to classes of Cartesian closed
Dialectica categories
Preferred way explained, leads to a generalisation of
construction
Show monads and comonads in generalisation
Example of non-Girardian comonad that produces weak
exponentials
Example of extensional version of Dialectica
Conclusions

Original Dialectica Categories (V. de Paiva, 1987)

Suppose that we have a category C, with ﬁnite limits,
interpreting some type theory.
The category Dial(C) has as objects triples A = (U, X , α),
where U, X are objects of C and α is a sub-object of U × X ,
that is a monic in Sub(U × X ).
A map from A = (U, X , α) to B = (V , Y , β ) is a pair of maps
(f , F ) in C, f : U → V , F : U × X → Y such that

α(u, F (u, y )) ≤ β (f (u), y)

The predicate α is not symmetric: read (U, X , α) as
∃u.∀x.α(u, x), a proposition in the image of the Dialectica
interpretation. The functionals f and F correspond to the
dialectica interpretation of implication.

Original Dialectica Categories

My thesis has four chapters, four main definitions and four main
theorems. The first two chapters are about the “original”
dialectica categories.
Theorem (V de Paiva, 1987)
If C is a ccc with stable, disjoint coproducts, then Dial(C) has
products, tensor products, units and a linear function space
(1, ×, ⊗, I, →) and Dial(C) is symmetric monoidal closed.
This means that Dial(C) models Intuitionistic Linear Logic (ILL)
without modalities. How to get modalities? Need to define a
special comonad and lots of work to prove theorem 2...

Original Dialectica Categories
!A must satisfy !A →!A⊗!A, !A ⊗ B →!A, !A → A and !A →!!A,
together with several equations relating them.
The point is to define a comonad such that its coalgebras are
commutative comonoids and the coalgebra and the comonoid
structure interact nicely.
Theorem (V de Paiva, 1987)
Given C a cartesian closed category with free monoids
(satisfying certain conditions), we can define a comonad T on
Dial(C) such that its Kleisli category Dial(C)T is cartesian
closed.
Define T by saying A = (U, X , α) goes to (U, X ∗ , α ∗ ) where X ∗
is the free commutative monoid on X and α ∗ is the multiset
version of α.
Loads of calculations prove that the linear logic modalitiy ! is
well-defined and we obtain a full model of ILL and IL, a
posteriori of CLL.
Construction generalized in many ways, cf. dePaiva, TAC, 2006.

Fibrational Dialectica (M. Hyland, 2002)
Given C a category with finite limits and a pre-ordered fibration
p : E → C with for each I in C, a preordered collection of
predicates E(I) = (E(I), ), construct a category Dial(p). The
objects A of Dial(p) are triples (U, X , α) where U, X are objects
in C and α is in E(U × X ). Maps in Dial(p) are pairs of maps in
C,(f , F ) in C, f : U → V , F : U × X → Y such that

α(u, F (u, y)) β (f (u), y)

Theorem (Hyland, 2002)
If C is a CCC and p : E → C is (pre-ordered) fibered cartesian
closed then Dial(p) is symmetric monoidal closed.
∼
If, moreover, C has finite, distributive coproducts and E(0) = 1
and the injections X → X + Y and Y → X + Y induce an
equivalence E(X + Y ) ≡ E(X ) × E(Y ) then Dial(p) has finite
products.
This is a generalization of the original Dialectica categories, where
the fibration is the subobject fibration.

Categorical Logic in slogans
“categorical model theory” or “categorical proof theory”?
Categorical model theory: model theory, where models are
categories, instead of sets.
Categorical proof theory: models are categories,
propositions are objects, derivations are arrows in the
category. concept of two different proofs being ‘the same’
Logic connectives correspond to structure in the
categories, Lawvere and Lambek in the 60’s.
Textbook: Lambek and Scott “Introduction to higher order
categorical logic”, 1986.
Main example: Intuitionistic propositional logic modeled as
Cartesian Closed Categories (CCCs).
Works for ﬁrst-order intuitionistic logic too, models are
hyperdoctrines

Fibrations for Dummies
Fibrations can be used for modeling several kinds of type
theories. Too complicated, perhaps?
Intuition: given a set I, consider the I-indexed family of sets
(Xi )i∈I or X (i), for each i in I.
Want something similar where instead of indexing sets, we
have a base category B, creating a “family of objects of a
category indexed by objects I in the base category B”
There are two “equivalent” ways of doing this, using
“indexed categories” or “ﬁbrations”.
A ﬁbration is a structure p : E → B, where p is a functor
and E, B are categories, satisfying certain axioms.
Think of B as Sets and E as Fam(Sets) where Fam(Sets)
is the category where objects are families of sets {Xi }i∈I , I
and Xi are sets. Maps from f : {Xi }i∈I → {Yj }j∈J consist of
a function φ : I → J (a re-indexing function) together with a
family of maps {fi : Xi → Yφ (i) }i∈I

Fibrations for Dummies 2
Which certain axioms? Given a functor p : E → B when
can we think of each object X in E as a family (Xi )i∈I
indexed by I = pX in B?
There is a functor p : Fam(Sets) → (Sets) that takes the
family {Xi }i∈I to I.
Call a map f : {Xi }i∈I → {Yj }j∈J , vertical if pf = 1I – no
re-indexing going on.
Call a map f : {Xi }i∈I → {Yj }j∈J , cartesian if each fi is an
isomorphism, pure re-indexing, fi s don’t do any work.
Given a family of sets {Xi }i∈I any map α : K → I induces a
(cartesian) map f : {Xα(k) }k∈K → {Xi }i∈I where each fi is
the identity.
cartesian maps have a universal property: an arbitrary
map f factors through a cartesian map g, when φ = pf
factors through α = pg and the factorization is uniquely
determined.

Fibrations for Dummies 3

Definition
A map g : X → X in E is called cartesian if given any map
f : Y → X , each factorization of φ = pf through α = pg uniquely
determines a factorization of f through g.
[insert picture]
Definition
A functor p : E → B is a fibration, if for all X in E and maps
α : K → I = pX there exists an object X and a cartesian map
g : X → X such that pg = α.
Cartesian liftings are unique up to iso, so: If p : E → B is a
fibration, a cleavage for this fibration is a particular choice of
cartesian liftings. A fibration equipped with a particular
cleavage is called a cloven fibration. (usually to show a
functor is a fibration, we produce a cleavage)

Proof Theory in the Abstract (Troelstra Fest)

A destilation of the program of ”proof theory in the abstract” as
developed since the late 60s. Structure of paper in 2002:
Background
Dialectica
Diller-Nahm
Classical logic
Highly recommended for philosophy and clarity of explanation
of the problems with classical logic.


For Dialectica main theorems are:
Theorem (Hyland Thm 2.3, p.8, 2002)
If T is a CCC interpreting some type theory and p : P → T is
(pre-ordered) fibered cartesian closed then Dial = Dial(p) as
defined is symmetric monoidal closed.
Natural propositional structure in place
Theorem (Hyland Thm 2.5, p.9, 2002)
The fibration q : Dial → T has left and right adjoints to
re-indexing along product projections. These satisfy the
Beck-Chevalley conditions.
Predicate structure in place.
But to really interpret Dialectica we need conjunctions and
disjunctions. Here the maths turns out not be so pretty.


For conjunction, to cope with diagonals A → A ⊗ A need ‘weak
cases definition’. To cope with projections A ⊗ B → A need
inhabited types. For disjunction we need a weak coproduct as
well as a weak initial object and a codiagonal. The structure
can be made to work, but it ain’t good category theory. (These
problems and way-outs were known from the non-fibrational case).
Theorem (Hyland, Thm 2.6 p.14, 2002)
The poset reflection of our indexed category Dial → T of proofs is a
first-order hyperdoctrine: we get indexed Heyting algebras and good
quantification.
Diller-Nahm has better structure!
Hyland suggests an abstract analysis of the interpretation of set
theory studied by Burr.
Jacobs, Streicher and de Paiva first version of 2.3 in note in 1995 (Jacob’s book exercise). Streicher’ note “A

¨
Semantic Version of the Diller-Nahm Variant of Godel’s Dialectica Interpretation”, 2000.

Cartesian Closed Dialectica Categories?

Several people took up more (categorical) analysis of
Dialectica, e.g. Streicher, Rosolini, Birkedal. etc...

The natural structure of Dial(p) is smcc with ﬁnite products.
Biering: How can we make it cartesian closed?
Make the tensor product, a cartesian product
Hyland’s ‘hackery’ above
Get a Girardian comonad on the Dialectica category
Diller-Nahm variants (VdP88), (Strei?), (Burr98)?, ...
Add enough structure to deﬁne a weak function space –
without making the tensor a product, (Biering06)

Cloven Dialectica Categories
Hyland defined the category Dial(p) for pre-ordered fibrations.
Biering generalized it, using fibrations into usual categories,
requiring p to be a cloven fibration. This is needed to obtain
associativity of composition in Dial(p). She defines Dial(p) and
prove it’s a category. With appropriate conditions it has some of
the structure we want.
Theorem (Biering, 2008)
Let p : E → T be a cloven fibration. If T has finite, distributive
coproducts and products, and the injections X → X + Y and
Y → X + Y induce an equivalence µ : E(X ) × E(Y ) ≡ E(X + Y ),
natural in X and Y , then Dial(p) has binary products.
∼
Moreover, if E(0) = 1, then Dial(p) has a terminal object.
The proof is a direct generalization of Hyland’s proof.
How can one make the category Dial(p) above cartesian
closed? It has products and a terminal object, but no function
spaces. Biering will define ‘weak’ function spaces for a
particular example of fibration, the codomain fibration.

Codomain Dialectica Categories?

What is the codomain fibration?
For any category C, there is a category Arr (C), whose objects
are the arrows of C, say α : X → I.
A morphism in Arr (C) from α : X → I to β : Y → J consists of a
pair of morphisms, f : X → Y and g : I → J.
There is a functor cod : Arr (C) → C which sends an object in
Arr (C), say α : X → I to its codomain I, and a morphism, the
square, (f : X → Y , g : I → J) to g, the ‘lower’ edge.
The functor cod : Arr (C) → C is a fibration.
Theorem (Biering, Prop 3.7 p6, 2008)
Let C be a category with finite limits and finite coproducts, and
assume that the coproducts are stable and disjoint, then
Dial(cod(C)) has finite products.
Can we make Dial(cod(C)) cartesian closed? Almost...

A comonad in Dial(p)

Definition
Let C be a category with finite products and stable, disjoint
coproducts. The functor − + 1 : C → C together with families of
maps ι : X → X + 1 and µ : (X + 1) + 1 → X + 1 is a monad on
C. Define a comonad L+ on the subobject fibration
Dial(Sub(C)) using the monad − + 1 as follows. Let α be a
subobject of U × X in C , then L + (α, U, X ) = (α+, U, (X + 1))
where α+ is reindexing of α along the arrow
∼
U × (X + 1) = U × X + U → (U × X ) + 1.
This means that α + (u, x) = α(u, x), if x is in X, if x is in 1.
This comonad simply makes the second coordinate
well-pointed.

A comonad in Dial(p)
The comonad L+ just defined is not Girardian, that is we do not
have the isomorphism
∼
!(A × B) =!A⊗!B
Had this iso being satisfied, then the Kleisli category of the
comonad would be cartesian closed. As it is, the best we can
do is to have weak function spaces, ie is a retraction
C(A × B, C) C(A, [B, C])
where the weak function space is given by
B → C = (W V × (1 + Y )V ×Z ) , V × Z , γ).
Theorem (Biering, 2008, 4.7, p 12)
Let C be a cartesian closed category with finite limits, and
stable, disjoint coproducts, which is locally cartesian closed.
Then the Dialectica-Kleisli category, DialL + (cod(C)), which we
denote by Dial+, has finite products and weak function spaces.
Same is true for DialL + (Sub(C)).
Main result, proof takes 6 pages, ‘souped up’ proof of dial thm

Discussion

Examples of fibrations that meet the conditions of Theorem 4.7:

cod(PER) → PER
cod(Set) → Set
the codomain fibration cod(C) → C for a topos C and
the subobject fibration Sub(C) → C
Biering decribes example where the fibration is
Fam(PER) → PER (a per is a partial equivalence relation), the
product as as well as the weak function space in
Dial + (Fam(PER)). But calculating there is unwieldy.

Discussion
Biering concludes: Two new variants of Dialectica
categories.
First, make proof theory out of Dial + . One needs to extend
¨
Godel’s T with stable, disjoint coproducts and subset types
and interpret implication by new weak function space.
Advantage: don’t need atomic formulas to be decidable.
Second, make proof theory of Dial(p). A type-theoretic
version of Dialectica. Instead of formulas over Heyting
arithmetic (original dialectica) have dependent types over
some type system and Dialectica turns the dependent type
system into a lambda-calculus without eta-rule.
Advantage?
Further work: Other comonads for Dialectica? Oliva’s
work. Do PER model better? Didn’t describe structure for
generalized dialectica categories, apart from products.

Conclusions

A worked out example of “proof theory in the abstract”
dialectica categories Dial(C) (1988)
dialectica pre-ordered ﬁbration Dial(p) (2002)
Dialectica-Kleisli category Dial + (2008)
More to come Oliva, Hofstra, Triffonov, etc..?

THANK YOU!

References
Troelstra, A. S. (1973) Metamathematical Investigation of
Intuitionistic Arithmetic and Analysis, Springer-Verlag.
¨
W. Hodges and B. Watson: translation of K. Godel, ’Uber ¨
eine bisher noch nicht bentzte Erweiterung des ﬁniten
Standpunktes’, JPL 9 (1980)
¨
Avigad, Feferman.Godel’s Functional (Dialectica)
Interpretation
V de Paiva. The Dialectica Categories, AMS-92, 1989
Burr, W. (1999) Concepts and aims of functional
interpretations: towards a functional interpretation of
constructive set theory.
M. Hyland, Proof Theory in the Abstract, APAL, 2002.
¨
P. Oliva. An analysis of Godel’s Dialectica interpretation via
linear logic. Dialectica,2008
Biering. Dialectica Interpretations: a categorical analysis,
PHD Thesis, 2008
¨
Collected Works of Kurt Godel, vol 2 Feferman et al, 1990

Fibrational Versions of Dialectica Categories

More Related Content

What's hot (19)

Viewers also liked (6)

Similar to Fibrational Versions of Dialectica Categories (20)

More from Valeria de Paiva (19)

Recently uploaded (20)

Fibrational Versions of Dialectica Categories