See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/45908905

On Linear Information Systems

Article · March 2010

DOI: 10.4204/EPTCS.22.4 · Source: arXiv

CITATIONS READS
9 24

4 authors, including:

Antonio Bucciarelli Alberto Carraro

Paris Diderot University Università Ca' Foscari Venezia
41 PUBLICATIONS   431 CITATIONS    21 PUBLICATIONS   81 CITATIONS   

SEE PROFILE SEE PROFILE

Antonino Salibra
Università Ca' Foscari Venezia
71 PUBLICATIONS   608 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Boolean Algebras with Operators View project

Decidability of satisfiability for the logic described in "a Logic for True Concurrency" by Baldan and Crafa View project

All content following this page was uploaded by Alberto Carraro on 09 March 2014.

The user has requested enhancement of the downloaded file.

 On Linear Information Systems

A. Bucciarelli† A. Carraro◦†
† Univeristé Paris Diderot ◦ Univeristà Ca’ Foscari
Paris, France Venice, Italy
Preuves Programmes et Systèmes Dipartimento di Informatica
[email protected] [email protected]

T. Ehrhard† A. Salibra◦
[email protected] [email protected]

Scott’s information systems provide a categorically equivalent, intensional description of Scott do-
mains and continuous functions. Following a well established pattern in denotational semantics, we
define a linear version of information systems, providing a model of intuitionistic linear logic (a new-
Seely category), with a “set-theoretic” interpretation of exponentials that recovers Scott continuous
functions via the co-Kleisli construction. From a domain theoretic point of view, linear information
systems are equivalent to prime algebraic Scott domains, which in turn generalize prime algebraic
lattices, already known to provide a model of classical linear logic.

1 Introduction
The ccc of Scott domains and continuous functions, which we call SD, is the paradigmatic framework for
denotational semantics of programming languages. In that area, much effort has been spent in studying
more “concrete” structures for representing domains.
At the end of the 70’s G. Kahn and G. D. Plotkin [23] developed a theory of concrete domains
together with a representation of them in terms of concrete data structures. In the early 80’s G. Berry and
P.-L. Curien [6] defined a ccc of concrete data structures and sequential algorithms on them. At the same
time Scott [27] also developed a representation theory for Scott domains which led him to the definition
of information systems; these structures, together with the so-called approximable relations, form the
ccc Inf, which is equivalent to SD.
So it was clear that many categories of “higher-level” structures such as domains had equivalent de-
scriptions in terms of “lower-level” structures, such as concrete data structures and information systems,
which are collectively called webs.
At the end of the 80’s, J.-Y. Girard [17] discovered linear logic starting from a semantical investi-
gation of the second-order lambda calculus. His seminal work on the semantics of linear logic proofs
[18, 19], introduced a category of webs, the coherence spaces, equivalent to the category of coherent
qualitative domains and stable maps between them. Coherence spaces form a ∗-autonomous and thus a
model of classical Linear Logic (LL).
From the early 90’s on, there has been a wealth of categorical models of linear logic, arising from
different areas: we mention here S. Abramsky and R. Jagadeesan’s games [1], Curien’s sequential data
structures [12], G. Winskel’s event structures [29, 32], and Winskel and Plotkin’s bistructures [26] whose
associated co-Kleisli category is equivalent to a full-sub-ccc of Berry’s category of bidomains [5].
Remarkably, all the above-mentioned models lie outside Scott semantics.

M. Florido and I. Mackie (Eds.): First International

c A. Bucciarelli & A. Carraro & T. Ehrhard & A. Salibra
Workshop on Linearity (LINEARITY 2009) This work is licensed under the
EPTCS 22, 2010, pp. 38–48, doi:10.4204/EPTCS.22.4 Creative Commons Attribution License.
 A. Bucciarelli & A. Carraro & T. Ehrhard & A. Salibra 39

Despite the observation made by M. Barr’s in 1979 [2] that the category of complete lattices and
linear maps is ∗-autonomous, it was a common belief in the Linear Logic community that the standard
Scott semantics could not provide models of classical LL, until 1994, when M. Huth showed [20] that
the category PAL of prime-algebraic complete lattices and lub-preserving maps is ∗-autonomous and its
associated ccc PAL! (the co-Kleisli category of the “!” comonad) is a full-sub-ccc of Cpo. A few years
later, Winskel rediscovered the same model in a semantical investigation of concurrency [30, 31]: indeed
he showed that the category ScottL whose objects are preordered sets and the morphisms are functions
from downward closed to downward closed subsets which preserve arbitrary unions is ∗-autonomous;
this category is equivalent to Huth’s. T. Ehrhard [15] continues this investigation and shows that the
extensional collapse of the category Rel! , where ! is a comonad based on multi-sets over the category
Rel of sets and relations, is the category ScottL! and that both are new-Seely categories (in the sense of
Bierman [7]).
Summing up, there are several categorical models of LL in Scott semantics. In this paper we provide
a representation of these models as webs. Our starting point are information systems, of which we
provide a linear variant together with linear approximable relations: such data form the category InfL,
which we prove to be a symmetric monoidal closed category; InfL is equivalent to the category PSD of
prime algebraic Scott domains and has as full-sub-categories InfLFull (equivalent to the category PAL)
and, ultimately, Rel.
We define a comonad ! over InfL, based on sets rather than multi-sets, which makes InfL a new-Seely
category and hence a model of intuitionistic MELL: our approach is different from that of [15] in that
our comonad is not an endofunctor of Rel; we don’t need to consider multisets exactly because we work
in the bigger category InfL. We also notice that InfLFull is the largest ∗-autonomous full-subcategory
of InfL and that InfL! is a full sub-ccc of Inf.

2 The category of linear informations systems

Let A be a set. We adopt the following conventions: letters α , β , γ , . . . are used for elements of A; letters
a, b, c, . . . are used for elements of Pf (A); letters x, y, z, . . . are used for arbitrary elements of P(A).
Definition 1. A linear information system (LIS, for short) is a triple A = (A, Con, ⊢), where Con ⊆ Pf (A)
contains all singletons and ⊢ ⊆ A × A satisfies the axioms listed below.
(IS1) if a ∈ Con and ∀β ∈ b.∃α ∈ a. α ⊢ β , then b ∈ Con
(IS2) α ⊢ α
(IS3) if α ⊢ β ⊢ γ , then α ⊢ γ
The set A is called the web of A and its elements are called tokens. Let A , B be two LISs. A
relation R ⊆ A × B is linear approximable if for all α , α ′ ∈ A and all β , β ′ ∈ B:
(AR1) if a ∈ ConA and ∀β ∈ b.∃α ∈ a. (α , β ) ∈ R, then b ∈ ConB
(AR2) if α ′ ⊢A α R β ⊢B β ′ , then (α ′ , β ′ ) ∈ R
Linear approximable relations compose as usual: S ◦R = R; S = {(α , γ ) ∈ A ×C : ∃β ∈ B. α R β S γ }.
We call InfL the category with LISs as objects and linear approximable relations as morphisms.
We reserve the name InfLFull for the full-subcategory of InfL whose objects are exactly those LISs
A for which ConA = Pf (A). It is not difficult to see that the category Rel of sets and relations
is a full-subcategory of InfLFull, consisting of exactly those LISs A for which ConA = Pf (A) and
⊢A = {(α , α ) : α ∈ A}.
40 On linear information systems

2.1 The cartesian structure of InfL

We now define the cartesian product and coproduct in the category InfL, which model the additive
connectives of linear logic.

Definition 2. Let A1 , A2 be LISs. Define the LIS A1 NA2 = (A1 NA2 , Con, ⊢) as follows:
• A1 NA2 = A1 ⊎ A2
• {(i1 , γ1 ), . . . , (im , γm )} ∈ Con iff {γ j : j ∈ [1, m], i j = 1} ∈ ConA1 and {γ j : j ∈ [1, m], i j = 2} ∈ ConA2
• (i, γ ) ⊢ ( j, γ ′ ) iff i = j and γ ⊢Ai γ ′
The projections πi ∈ InfL(A1 NA2 , Ai ) are given by πi = {((i, γ ), γ ′ ) : γ ⊢Ai γ ′ }, i = 1, 2. For R ∈ InfL(C , A1 )
and S ∈ InfL(C , A2 ), the pairing hR, Si ∈ InfL(C , A1 NA2 ) is given by
hR, Si = {(γ , (1, α )) : (γ , α ) ∈ R} ∪ {(γ , (2, β )) : (γ , β ) ∈ S}. Define ⊤ = (0,
/ 0, / The LIS ⊤ is the ter-
/ 0).
minal object of InfL, since for any LIS A , the only morphism R ∈ InfL(A , ⊤) is 0. /

Definition 3. Let A1 , A2 be LISs. Define the LIS A1 ⊕ A2 = (A1 ⊕ A2 , Con, ⊢) as follows:

• A1 ⊕ A2 = A1 ⊎ A2
• {(i1 , γ1 ), . . . , (im , γm )} ∈ Con iff {γ j : j ∈ [1, m], i j = 1} ∈ ConA1 and {γ j : j ∈ [1, m], i j = 2} ∈ ConA2
• (i, γ ) ⊢ ( j, γ ′ ) iff i = j and γ ⊢Ai γ ′
Therefore cartesian products and coproducts coincide. The injections ιi ∈ InfL(Ai , A1 ⊕ A2 ) are
obtained reversing the projections, so that ιi = {(γ , (i, γ ′ )) : γ ⊢Ai γ ′ }, i = 1, 2. For R ∈ InfL(A1 , C ) and
S ∈ InfL(A2 , C ), the “co-pairing” [R, S] ∈ InfL(A1 ⊕ A2 , C ) is given by
[R, S] = {((1, α ), γ ) : (α , γ ) ∈ R} ∪ {((2, β ), γ ) : (β , γ ) ∈ S}. Define 0 = ⊤ = (0,
/ 0,
/ 0):
/ this LIS is also the
initial object of InfL and the unit of the coproduct.

2.2 The monoidal closed structure of InfL

We now define the tensor product and its dual in the category InfL, which model the multiplicative
connectives of linear logic.

Definition 4. Let A , B be LISs. Define the LIS A ⊗ B = (A ⊗ B, Con, ⊢) as follows:

• A⊗B = A×B
• {(α1 , β1 ), . . . , (αm , βm )} ∈ Con iff {α1 , . . . , αm } ∈ ConA and {β1 , . . . , βm } ∈ ConB
• (α , β ) ⊢ (α ′ , β ′ ) iff α ⊢A α ′ and β ⊢B β ′
For R ∈ InfL(A , C ) and S ∈ InfL(B, D), R ⊗ S ∈ InfL(A ⊗ B, C ⊗ D) is given by
R ⊗ S = {((α , β ), (γ , δ )) ∈ (A ⊗ B) × (C ⊗ D) : (α , γ ) ∈ R and (β , δ ) ∈ S}. It is easy to check that
⊗ : InfL × InfL → InfL is a bifunctor and that it is a symmetric tensor product, with natural isomor-
phisms
• φA⊗ ,B,C : A ⊗ (B ⊗ C ) → (A ⊗ B) ⊗ C given by

φA⊗ ,B,C = {((α , (β , γ )), ((α ′ , β ′ ), γ ′ )) : α ⊢A α ′ , β ⊢B β ′ , γ ⊢C γ ′ }

• σA⊗ ,B : A ⊗ B → B ⊗ A given by σA⊗ ,B = {((α , β ), (β ′ , α ′ )) : α ⊢A α ′ , β ⊢B β ′ }

• ρA⊗ : A ⊗ 1 → A given by ρA⊗ = {((α , ∗), α ′ ) : α ⊢A α ′ }
A. Bucciarelli & A. Carraro & T. Ehrhard & A. Salibra 41

• λA⊗ : 1 ⊗ A → A given by λA⊗ = {((∗, α ), α ′ ) : α ⊢A α ′ }

Define 1 = ({∗}, {0,/ {∗}}, {(∗, ∗)}). The LIS 1 is the unit of the tensor product.
The above data make InfL a symmetric monoidal category. As for the cartesian structure, also the
dual of the tensor product, O, coincides with the tensor in this category and thus the respective units ⊥
and 1 are equal. So we take ⊥ = ({∗}, {0,/ {∗}}, {(∗, ∗)}) to be also the unit of O.
We now proceed to define the exponential objects of InfL.
Definition 5. Let A , B be LISs. Define the LIS A ⊸ B = (A ⊸ B, Con, ⊢) as follows:
• A ⊸ B = A×B
• {(α1 , β1 ), . . . , (αm , βm )} ∈ Con iff for all J ⊆ [1, m], {α j : j ∈ J} ∈ ConA implies {β j : j ∈ J} ∈ ConB
• (α , β ) ⊢ (α ′ , β ′ ) iff α ′ ⊢A α and β ⊢B β ′
Define a natural isomorphism cur : InfL(A ⊗ C , B) → InfL(C , A ⊸ B), (the linear currying) as
cur(R) = {(γ , (α , β )) : ((α , γ ), β ) ∈ R}. Define also the (linear) evaluation morphism ev : A ⊗ (A ⊸ B) → B
as ev = {((α , (α ′ , β )), β ′ ) : α ⊢A α ′ , β ⊢B β ′ }.
The above data make InfL a symmetric monoidal closed category, and thus a model of intuitionistic
MLL proofs. The category InfL is however a rather degenerate model since the multiplicative connec-
tives O and ⊗ coincide as well as the additives N and ⊕.
We now briefly discuss the issue of duality in the category InfL. Let A be a LIS and consider the
LIS A ⊸ ⊥; an explicit description of such object is as follows:
• A ⊸ ⊥ = A × {∗}
• ConA⊸⊥ = Pf (A × {∗})
• (α , ∗) ⊢A⊸⊥ (β , ∗) iff β ⊢A α
Therefore ⊥ is not a dualizing object in InfL, but it is so in InfLFull, where the family of arrows
∂A : A → (A ⊸ ⊥) ⊸ ⊥ defined by ∂A = {(α , ((α ′ , ∗), ∗)) : α ⊢A α ′ }, for each LIS A , is a natural
isomorphism. In other words InfLFull is the largest ∗-autonomous full-subcategory of InfL.

2.3 InfL is a new-Seely category

In this section we define a comonad ! over InfL and prove that it gives a symmetric strong monoidal
functor. Finally we prove that InfL is a new-Seely category and thus a model of intuitionistic MELL
proofs, in which the exponential modality ! has a set-theoretic interpretation. For categorical notions we
refer to Melliès [25].
Definition 6. Let A be a LIS. Define the LIS !A = (!A, Con, ⊢) as follows:
• !A = ConA
• {a1 , . . . , ak } ∈ Con iff ∪ki=1 ai ∈ ConA
• a ⊢ b iff ∀β ∈ b.∃α ∈ a. α ⊢A β
Note that for X ∈!!A and a ∈!A we have X ⊢!!A {b} implies ∪X ⊢!A b but not viceversa. As an
example, consider that {α , β } ⊢!A {α , β } but in general not {{α }, {β }} ⊢!!A {{α , β }}.
Let R ∈ InfL(A , B). Define !R ∈ InfL(!A , !B) as !R = {(a, b) ∈ !A×!B : ∀β ∈ b.∃α ∈ a. (α , β ) ∈ R}.
It is an easy matter to verify that !( ) : InfL → InfL is a functor. Moreover “!” is a comonad with dig-
ging dig :! ⇒!! defined by digA = {(b,Y ) ∈ !A×!!A : b ⊢!A ∪Y } and dereliction der :! ⇒ idInfL defined
by derA = {(b, β ) ∈ !A × A : b ⊢!A {β }}.
42 On linear information systems

As a matter of fact “!” is also a monad if endowed with natural transformations codigging codig :!! ⇒!
defined by codigA = {(X , a) ∈ !!A×!A : X ⊢!!A {a}} = der!A and codereliction cod : idInfL ⇒! defined
by codA = {(α , b) ∈ A×!A : {α } ⊢!A b}. This is due to the fact that a ⊢!A b iff {a} ⊢!!A {b} and X ⊢!!A {b}
iff ∃a ∈ X . a ⊢!A b.
This also shows a further symmetry: for each object !A ,
• the digging morphism is subsumed by the codereliction morphism in the sense that
cod!A = {(a, X ) : {a} ⊢!!A X } = digA , since for X ∈!!A and a ∈!A we have {a} ⊢!!A X iff a ⊢!A ∪X;
• the codigging morphism is subsumed by the dereliction morphism in the sense that
der!A = {(X , a) : X ⊢!!A {a}} = codigA .

The forthcoming lemma shows that “!” is a symmetric strong monoidal functor. Before proving this,
we shall explicit the symmetric monoidal structure (InfL, N, ⊤) involved in the proof:
• φAN ,B,C : A N(BNC ) → (A NB)NC given by

φAN ,B,C = (1, α ), (1, (1, α ′ )) : α ⊢A α ′ ∪ (2, (1, β )), (1, (2, β ′ )) : β ⊢B β ′ ∪



∪ (2, (2, γ )), (2, γ ′ ) : γ ⊢C γ ′



• σAN ,B : A NB → BNA given by

σAN ,B = (1, α ), (2, α ′ ) : α ⊢A α ′ ∪ (2, β ), (1, β ′ ) : β ⊢B β ′




• ρAN : A N⊤ → A given by ρAN = (1, α ), α ′ : α ⊢A α ′



• λAN : ⊤NA → A given by λAN = (2, α ), α ′ : α ⊢A α ′



Lemma 7. The functor ! : (InfL, N, ⊤) → (InfL, ⊗, 1) is symmetric strong monoidal.

Proof. We give the natural isomorphisms mA ,B : !A ⊗!B ∼

=!(A NB) and n : 1 ∼
=!⊤ making ! a sym-
metric strong monoidal functor. Define
• n = {(∗, {0})}
/
• mA ,B = (a, b), {(1, α ′ ) : α ′ ∈ a′ } ∪ {(2, β ′ ) : β ′ ∈ b′ } : a, a′ ∈ !A, b, b′ ∈ !B, a ⊢!A a′ , b ⊢!B b′


We now proceed by verifying the commutation of the required diagrams. First observe that both
mA ,B ⊗ id!C ; mA NB,C ; !φAN ,B,C and φA⊗ ,B,C ; id!A ⊗ mB,C ; mA ,BNC are equal to the set of all pairs
 
((a, b), c), {(1, (1, α1 )), . . . , (1, (1, αn1 )), (2, (1, β1 )), . . . , (2, (1, βn2 )), (2, (2, γ1 )), . . . , (2, (2, γn3 ))}

where a ⊢!A {α1 , . . . , αn1 }, b ⊢!B {β1 , . . . , βn2 }, and c ⊢!C {γ1 , . . . , γn3 }. Therefore the diagram

φA⊗ ,B,C
(!A ⊗!B)⊗!C / !A ⊗ (!B⊗!C )

mA ,B ⊗id!C id!A ⊗mB,C

 
!(A NB)⊗!C !A ⊗!(BNC )
mA NB,C mA ,BNC
 !φAN,B,C 
!((A NB)NC ) / !(A N(BNC ))
A. Bucciarelli & A. Carraro & T. Ehrhard & A. Salibra 43

commutes. Finally the “units” and the “symmetry” diagrams

ρ!A
⊗
λ!B
⊗ σ!A
⊗
,!B
!A ⊗ 1 / !A 1⊗!B / !B !A ⊗!B / !B⊗!A
O O
id!A ⊗n !ρA
N
n⊗id!B !λB
N mA ,B mB,A
 mA ,⊤  m⊤,B  !σAN,B 
!A ⊗!⊤ / !(A N⊤) !⊤⊗!B / !(⊤NB) !(A NB) / !(BNA )

all commute because

• id!A ⊗ n; mA ,⊤ ; !ρAN = {((a, ∗), a′ ) : a ⊢!A a′ } = ρ!A
⊗
,
• n ⊗ id!B ; m⊤,B ; !λB
N
= {((∗, b), b′ ) : b ⊢!B b′ } = λ!B
⊗
,
• both σ!A
⊗
,!B ; mB,A and mA ,B ; !σA ,B equal the
morphism
N

(a, b), {(1, β ) : β ∈ b } ∪ {(2, α ′ ) : α ′ ∈ a′ } : b′ ∈ !B, a′ ∈ !A, b ⊢!B b′ , a ⊢!A a′ .

′ ′ ′


Proposition 8. InfL is a new-Seely category.

Proof. By Lemma 7, ! is a symmetric strong monoidal functor; it remains to check the coherence diagram
in the definition of new-Seely category. Indeed we have:

(a, b), {(1, a′ ) : a′ ∈ X ′ } ∪ {(2, b′ ) : b′ ∈ Y ′ } : a ⊢!A ∪X ′ , b ⊢!B ∪Y ′



digA ⊗ digB ; m!A ,!B =
= mA ,B ; digA NB ; !h!π1 , !π2 i

So that the following diagram commutes.

mA ,B
!A ⊗!B / !(A NB)

digA NB

digA ⊗digB !!(A NB)
!h!π1 ,!π2 i
 m!A ,!B 
!!A ⊗!!B / !(!A N!B)

2.4 A representation theorem

Not surprisingly, InfL turns out to be equivalent to the category of prime algebraic Scott domains and
linear continuous functions. In this section, we outline this equivalence.
Definition 9. An element p of a Scott domain D is prime if, whenever B ⊆f D is upper bounded and
p ≤ ∨B, there exists b ∈ B such that p ≤ b. The domain D itself is prime algebraic if every element of D
is the least upper bound of the set of prime elements below it. The set of prime elements of D is denoted
by PR(D).
A linear function between two prime algebraic Scott domains is a Scott continuos function that
commutes with all (existing) least upper bounds. The category of prime algebraic Scott domains and
linear function is denoted by PSD.
44 On linear information systems

A point of an information system A is a subset x ⊆ A satisfying the following two properties:

(PT1) if u ⊆f x then u ∈ Con (x is finitely consistent)
(PT2) if α ∈ x and α ⊢ α ′ , then α ′ ∈ x (x is closed w.r.t. ⊢)
We are now able to relate linear information systems to the corresponding categories of domains.
Definition 10. Given f , D, E , R, A , B such that f ∈ PSD(D, E ) and R ∈ InfL(A , B), we define:
• A + is the set of points of A ordered by inclusion.
• R+ (x) = {β ∈ B | ∃α ∈ x. (α , β ) ∈ R}
• D − = (PR(D), Con, ⊢) where a ∈ Con iff a is upper bounded and p ⊢ p′ iff p′ ≤D p
• f − = {(p, p′ ) ∈ PR(D) × PR(E) | f (p) ≥D p′ }
Theorem 11. The functors ( )+ , ( )− define an equivalence between the categories InfL and PSD.

Proof. It is an easy task to check that ( )+ : InfL → PSD and ,( )− : PSD → InfL are indeed full and
faithful functors and the two composite endofunctors (−)− ◦ (−)+ and (−)+ ◦ (−)− are naturally iso-
morphic to the identity functor of InfL and PSD, respectively. Moreover every hom-set InfL(A , B),
ordered by inclusion of relations, is a prime algebraic Scott domain and InfL(A , B) = (A ⊸ B)+ , so
that the functors (−)− and (−)+ preserve exponentials, i.e. (A ⊸ B)+ ∼ = PSD(A + , B + ) in the cate-
− ∼ − −
gory PSD and PSD(D, E ) = D ⊸ E in the category InfL. Finally the two functors also preserve
products, since (A NB)+ ∼ = A + × B + in the category PSD and (D × E )− ∼ = D − NE − in the category
InfL.

In view of the categorical equivalence stated in Theorem 11, Proposition 13 shows how to recover
Scott-continuous functions from linear ones. This equivalence specializes to an equivalence between
InfLFull and the category PAL of prime algebraic lattices and linear continuos functions.

3 Classical versus linear information systems

In the previous section we have treated a categorical equivalence explaining how linear information sys-
tems constitute a representation for prime algebraic Scott domains. Along the same lines Scott domains
have an appealing representation as information systems introduced by Dana Scott in [27]. More recently
[28] a more general class of information systems have also been axiomatized, namely that of continuous
information systems , defined in order to constitute a representation for continuous domains.
An information system consists of a set of tokens, over which are imposed an entailment and a
consistency relation; it determines a Scott domain with elements those sets of tokens which are consistent
and closed with respect to the entailment relation; the ordering is again just set inclusion. Vice versa a
Scott domain defines an information system through its compact elements. In this section we review the
basic notions on information systems.
Information systems are then organized in a category equivalent to that of Scott domains and con-
tinuous maps, SD, but more “concrete” and easier to work with under many respects. For example they
have been used in [11] to show that there is no reflexive object in SD whose theory is exactly the least
extensional lambda theory λ β η .
In this section we explain the relation between the “classical” Scott’s information systems and linear
information systems.
 A. Bucciarelli & A. Carraro & T. Ehrhard & A. Salibra 45

Originally ([27]) an information system (IS, for short) is a triple A = (A, Con, ⊢), where Con ⊆ Pf (A)
contains all singletons and ⊢ ⊆ Con × Con satisfies the axioms listed below.
(IS1) if a ∈ Con and a ⊢ b, then b ∈ Con
(IS2) if a′ ⊆ a, then a ⊢ a′
(IS3) if a ⊢ b ⊢ c, then a ⊢ c

A relation R ⊆ ConA × ConB is approximable if:

(AR1) if a ∈ ConA and a R b, then b ∈ ConB
(AR2) if a′ ⊢A a R b ⊢B b′ , then a′ R b′
Clearly the every approximable relation, included ⊢, is completely determined by tokens on the right-
hand side in the sense that a R b iff ∀β ∈ b. a R {β }. Hence we shall identify each approximable relation
R with its trace {(a, β ) : (a, {β }) ∈ R}.
We call Inf the category with ISs as objects and approximable relations as morphisms. It is well-
known that Inf(A , B), ordered by inclusion of relations, is a Scott domain. Let us recall the definition
of exponentials in Inf ([24]).
Definition 12. Let A , B be ISs. Define the IS A ⇒ B = (A ⇒ B, Con, ⊢) as follows:
• A ⇒ B = ConA × B
• {(a1 , β1 ), . . . , (am , βm )} ∈ Con iff for all J ⊆ [1, m], ∪{a j : j ∈ J} ∈ ConA implies {β j : j ∈ J} ∈
ConB
• {(a1 , β1 ), . . . , (am , βm )} ⊢ (a′ , β ′ ) iff {β j : a′ ⊢ a j , 1 ≤ j ≤ m} ⊢B β ′
Similarly to the functors given in Definition 10, there are functors ( )• : Inf → SD and ( )◦ : SD → Inf
which define another equivalence of categories. In particular for a given IS A , A • is the collection of
all subsets x ⊆ A satisfying the following two properties:
(PT1’) if u ⊆f x then u ∈ Con (x is finitely consistent)
(PT2’) if a ⊆ x and a ⊢ α ′ , then α ′ ∈ x (x is closed w.r.t. ⊢)
Again we have that Inf(A , B) = (A ⇒ B)• . The categorical equivalence stated in Theorem 11
mirrors perfectly the equivalence between the categories Inf and SD, so that the definition of linear
information system and linear approximable relation is exactly what is required in order to capture the
passage from the category of Scott domains and continuous functions to that of prime algebraic Scott
domains and linear functions.
In fact both linear information systems and linear approximable relations can be seen as particular
information systems and approximable relations, respectively, exactly as prime algebraic Scott domains
are Scott domains and linear functions are continuous functions. The next proposition, based on this
fact, shows that again, following a well-established pattern, the comonad ! allows to recover non-linear
approximable relations form linear ones. As usual, we denote by InfL! the co-Kleisli category of the
comonad ! over InfL.
Proposition 13. InfL! is a full-sub-ccc of Inf.

Proof. Let A , B be LISs and let A ⇒ B be the exponential object formed in the category Inf: A ⇒ B
is a linear information system and it is an easy matter to see that A ⇒ B =!A ⊸ B. Moreover
C + = C • , for any LIS C , and thus InfL(!A , B) = (!A ⊸ B)+ = (A ⇒ B)• = Inf(A , B).
46 On linear information systems

The space InfL(A , B) of clearly embeds into Inf(A , B) (exactly as the space of linear functions
embeds into that of continuous functions). The embedding is given by the map ϕ : InfL(A , B) ֒→
Inf(A , B) given by ϕ (R) = {(a, β ) : ∃α ∈ a. α R β }. In other words the linear approximable relations
are elements of Inf(A , B), i.e. exactly those approximable relations S for which (a, β ) ∈ S iff ∃α ∈
a. (α , β ) ∈ S. This is the analogue of the condition, dealing with preservation of existing suprema, that
isolates linear functions between Scott domains among the continuous ones.

4 Conclusions and future work

In this paper we defined the category InfL, whose objects and arrows result from a linearization of
Scott’s information systems. We show this category to be symmetric monoidal closed and thus a model
of MLL. We moreover prove that InfL is a new-Seely category, with a “set-theoretic” interpretation of
exponentials via a comonad !; this is made possible by the presence of the entailment relation, which
is always non-trivial in objects of the form !A : even if the entailment ⊢A is the equality we have that
InfL(!A , A ) ⊂ P(!A × A ) and this rules out Ehrhard’s counterexample for the naturality of derelic-
tion. In the purely relational model of classical MELL, Rel, the use of multisets is needed.
Indeed a comonad based on multi-sets, let’s say †, can be defined in our framework too, yelding a
different co-Kleisli category. A similar situation arises in the framework of the coherence spaces model of
LL. In that case Barreiro and Ehrhard [3] proved that the extensional collapse of the hierarchy of simple
types associated to the multi-set interpretation is the hierarchy associated to the “set” interpretation. This
means in particular that, as models of the simply typed λ -calculus, the former discriminates finerly than
the latter, being sensitive, for instance, to the number of occurrences of a variable in a term. It is likely
that, in a similar way, InfL! is the extensional collapse of InfL† .
The categories InfL and InfLFull may be themselves compared using the same paradigm. Trivializ-
ing the consitency relation boils down to add points to the underlying domains; is that another instance
of extensional collapse situation? In the case of the simple types hierarchy over the booleans, the Scott
model is actually the extensional collapse of the lattice-theoretic one [10].
Summing up it appears that, by tuning the linear information systems in different ways, one obtains
different frameworks for the interpretation of proofs, whose inter-connections remain to be investigated.
This is, in our opinion, the main advantage of the approach, with respect to the existing descriptions
of the Scott continuous models of linear logic [2, 20, 21, 30].
Linear information systems are close to several classes of webbed models of the pure the λ -calculus:
they generalize Berline’s preordered sets with coherence [4], where a set of tokens is consistent if and
only if its elements are pairwise coherent. We plan to investigate whether such a generalisation is use-
ful for studying the models of the λ -calulus in PSD. Actually, one of our motivations was to settle a
representation theory for a larger class of cartesian closed categories, whith Rel! as a particular case, in
order to provide a tool for investigating “non-standard”1 models. We get Rel as a full subcategory of
InfL, but the bunch of axioms on information-like structures making Rel! an instance of the co-Kleisli
construction remains to be found.
Another original motivation for investigating linear information systems, that we are pursuing, was
the definition of a framework suitable for the interpretation of Boudol’s λ -calculus with resources [9].
Finally we point out another research direction, suggested by the work of Ehrhard and Regnier ([16],
for example). In Köthe spaces [13] as well as in finiteness spaces [14], linear logic formulae are inter-
1 Let us call “standard” a model of the λ -calculus that is an instance of one of the “main” semantics: continuous, stable,

strongly stable.
A. Bucciarelli & A. Carraro & T. Ehrhard & A. Salibra 47

preted as topological vector spaces, and proofs of linear logic as linear continuous maps between these
spaces. Then exponentials appear as “symmetric tensor algebra” constructions [8]. In the models con-
sidered there, linear maps from !X to Y can be seen as “analytic functions” (that is, functions definable
by a power series) from the vector space X to the vector space Y and therefore can be differentiated.
Classically, the derivative of a function f : X → Y is a function f : X → (X ⊸ Y ) such that for each
x ∈ X , the linear function f ′ (x) (the derivative of f at point x) is the “best linear approximation” of
the function X → Y which maps u ∈ X to f (x + u) ∈ Y (the general definition is local). In the analytic
case, differentiation turns a linear function f :!X → Y into a linear function f ′ :!X → (X ⊸ Y ), that is,
f ′ : (!X ⊗ X ) → Y . It turns out that f ′ can be obtained from f by composing it (as a linear function from
!X to Y ) on the left with a particular linear morphism d : (!X ⊗ X ) →!X . This morphism itself can be
defined in terms of more primitive operations on !X .
This can certainly be done also in the category InfL following for [22], for example. However the
induced differential combinator in this case does not reflect the idea of approximation typical of Scott
semantics. The purpose we have in mind is to investigate the possibility of symmetric tensor bialgebra
constructions, in the category InfL, giving rise in the equivalent category PSD to a reasonable notion of
derivative and compatible with the usual idea of approximation in Scott semantics.

References
[1] S. Abramsky & R. Jagadeesan (1992): Games and Full Completeness for Multiplicative Linear Logic (Ex-
tended Abstract). In: Proceedings of the 12th Conference on Foundations of Software Technology and
Theoretical Computer Science. Springer-Verlag, London, UK, pp. 291–301.
[2] M. Barr (1979): ∗-autonomous categories. Number 752 in LNCS. Springer-Verlag.
[3] N. Barreiro & T. Ehrhard (1997). Anatomy of an extensional collapse.
[4] C. Berline (2000): From computation to foundations via functions and application: The λ -calculus and its
webbed models. Theor. Comput. Sci. 249(1), pp. 81–161.
[5] G. Berry (1979): Modèles complétement adequats et stables des lambda-calculs typés. Thèse de doctorat d‘
État, Université de Paris VII.
[6] G. Berry & P.-L. Curien (1982): Sequential algorithms on concrete data structures. Theor. Comput. Sci. 20,
pp. 265–321.
[7] G. M. Bierman (1995): What is a Categorical Model of Intuitionistic Linear Logic? In: TLCA ’95: Proceed-
ings of the Second International Conference on Typed Lambda Calculi and Applications. Springer-Verlag,
London, UK, pp. 78–93.
[8] R. Blute, P. Panangaden & R. Seely (1993): Fock space: a model of linear exponential types. In: Proceedings
of the Ninth Conference on Mathematical Foundations of Programming Semantics. Springer-Verlag, London,
UK.
[9] G. Boudol (1993): The Lambda-Calculus with Multiplicities (Abstract). In: CONCUR ’93: Proceedings of
the 4th International Conference on Concurrency Theory. Springer-Verlag, London, UK, pp. 1–6.
[10] A. Bucciarelli (1996): Logical relations and lambda theories. In: Advances in Theory and Formal Methods
of Computing, proceedings of the 3rd Imperial College Workshop. pp. 37–48.
[11] A. Carraro & A. Salibra (2009): Reflexive Scott domains are not complete for the extensional lambda calcu-
lus. In: LICS. pp. 91–100.
[12] P.-L. Curien (1994): On the Symmetry of Sequentiality. In: Proceedings of the 9th International Conference
on Mathematical Foundations of Programming Semantics. Springer-Verlag, London, UK, pp. 29–71.
[13] T. Ehrhard (2002): On Köthe sequence spaces and linear logic. Mathematical. Structures in Comp. Sci.
12(5), pp. 579–623.
 48 On linear information systems

[14] T. Ehrhard (2005): Finiteness spaces. Mathematical. Structures in Comp. Sci. 15(4), pp. 615–646.
[15] T. Ehrhard (2009). The Scott model of Linear Logic is the extensional collapse of its relational model.
[16] T. Ehrhard & L. Regnier (2006): Differential interaction nets. Theor. Comput. Sci. 364(2), pp. 166–195.
[17] J.-Y. Girard (1986): The system F of variable types, fifteen years later. Theor. Comput. Sci. 45(2), pp.
159–192.
[18] J.-Y. Girard (1987): Linear logic. Theor. Comput. Sci. 50(1), pp. 1–102.
[19] J.-Y. Girard (1988): Normal functors, power series and the lambda-calculus. Annals of pure and applied
logic 37(2), pp. 129–177.
[20] M. Huth (1994): Linear Domains and Linear Maps. In: Proceedings of the 9th International Conference on
Mathematical Foundations of Programming Semantics. Springer-Verlag, London, UK, pp. 438–453.
[21] M. Huth, A. Jung & K. Keimel (1994): Linear Types, Approximation, and Topology. In: LICS. pp. 110–114.
[22] M. Hyland & A. Schalk (2003): Glueing and orthogonality for models of linear logic. Theor. Comput. Sci.
294(1-2), pp. 183–231.
[23] G. Kahn & G. D. Plotkin (1978): Domaines Concrets. Technical Report Rapport 336, IRIA-LABORIA.
[24] K.G. Larsen & G. Winskel (1991): Using information systems to solve recursive domain equations. Inf.
Comput. 91(2), pp. 232–258.
[25] P. A. Melliès (2008). Categorical semantics of linear logic.
[26] G. D. Plotkin & G. Winskel (1994): Bistructures, Bidomains and Linear Logic. In: ICALP ’94: Proceedings
of the 21st International Colloquium on Automata, Languages and Programming. Springer-Verlag, London,
UK, pp. 352–363.
[27] D. S. Scott (1982): Domains for Denotational Semantics. In: Proceedings of the 9th Colloquium on Au-
tomata, Languages and Programming. Springer-Verlag, London, UK, pp. 577–613.
[28] Dieter Spreen, Luoshan Xu & Xuxin Mao (2008): Information systems revisited: the general continuous
case. Theor. Comput. Sci. 405(1-2), pp. 176–187.
[29] G. Winskel (1988): An introduction to event structures. In: REX Workshop. pp. 364–397.
[30] G. Winskel (1999): A Linear Metalanguage for Concurrency. In: AMAST ’98: Proceedings of the 7th
International Conference on Algebraic Methodology and Software Technology. Springer-Verlag, London,
UK, pp. 42–58.
[31] G. Winskel (2004): Linearity and non linearity in distributed computation. In: T. Ehrhard and J.-Y. Girard
and P. Ruet and P. Scott, ed. Linear Logic in Computer Science, London Mathematical Society Lecture Note
Series 316. Cambridge University Press.
[32] G.-Q. Zhang (1992): A Monoidal Closed Category of Event Structures. In: Proceedings of the 7th Interna-
tional Conference on Mathematical Foundations of Programming Semantics. Springer-Verlag, London, UK,
pp. 426–435.

View publication stats