Under consideration for publication in Math. Struct. in Comp.

Science

An introduction to Differential Linear Logic:

proof-nets, models and antiderivatives
Thomas Ehrhard
CNRS, IRIF, UMR 8243
Univ Paris Diderot, Sorbonne Paris Cité F-75205 Paris, France

Received October 2001

Differential linear logic enriches linear logic with additional logical rules for the
exponential connectives, dual to the usual rules of dereliction, weakening and
contraction. We present a proof-net syntax for differential linear logic and a categorical
axiomatization of its denotational models. We also introduce a simple categorical
condition on these models under which a general antiderivative operation becomes
available. Last we briefly describe the model of sets and relations and give a more
detailed account of the model of finiteness spaces and linear and continuous functions.

Introduction
Extending Linear Logic (LL) with differential constructs has been considered by Gi-
rard at a very early stage of the design of this system. This option appears at various
places in the conclusion of (Gir86), entitled Two years of linear logic: selection from the
garbage collector. In Section V.2 The quantitative attempt of that conclusion, the idea
of a syntactic Taylor expansion is explicitly mentioned as a syntactic counterpart of the
quantitative semantics of the λ-calculus (Gir88). However it is contemplated there as a
reduction process rather than as a transformation on terms. In Section V.5 The exponen-
tials, the idea of reducing λ-calculus substitution to a more elementary linear operation
explicitly viewed as differentiation is presented as one of the basic intuitions behind the
exponential of LL. The connection of this idea with Krivine’s Machine (Kri85; Kri07) and
its linear head reduction mechanism (DR99) is explicitly mentioned. In this mechanism,
first considered by De Bruijn and called mini-reduction in (DB87), it is only the head
occurrence of a variable which is substituted during reduction. This restriction is very
meaningful in LL: the head occurrence is the only occurrence of a variable in a term
which is linear.
LL is based on the distinction of particular proofs among all proofs, that are linear
wrt. their hypotheses. The word linear has here two deeply related meanings.
— An algebraic meaning: a linear morphism is a function which preserves sums, linear
combinations, joins, unions (depending on the context). In most denotational models
of LL, linear proofs are interpreted as functions which are linear in that sense.
Thomas Ehrhard 2

— An operational meaning: a proof is linear wrt. an hypothesis if the corresponding

argument is used exactly once (neither erased nor duplicated) during cut-elimination.
LL has an essential operation, called dereliction, which allows one to turn a linear proof
into a non linear one, or, more precisely, to forget the linearity of a proof. Differentiation,
which in some sense is the converse of dereliction, since it turns a non linear morphism
(proof) into a linear one, has not been included in LL at an early stage of its development.
We think that there are two deep reasons for that omission.
— First, differentiation seems fundamentally incompatible with totality, a denotational
analogue of normalization usually considered as an essential feature of any reasonable
logical system. Indeed, turning a non-linear proof into a linear one necessarily leads
to a loss of information and to the production of partial linear proofs. This is typically
what happens when one takes the derivative of a constant proof, which must produce
a zero proof.
— Second, they seem incompatible with determinism because, when one linearizes a
proof obtained by contracting two linear inputs of a proof, one has to choose be-
tween these two inputs, and there is no canonical way of doing so: we take the non-
deterministic superposition of the two possibilities. Syntactically, this means that one
must accept the possibility of adding proofs of the same formula, which is standard in
mathematics, but hard to accept as a primitive logical operation on proofs (although
it is present, in a tamed version, in the additive rules of LL).
The lack of totality is compatible with most mathematical interpretations of proofs and
with most denotational models of LL: Scott domains (or more precisely, prime algebraic
complete lattices, see (Hut93; Win04; Ehr12)), dI-domains, concrete data structures,
coherence spaces, games, hypercoherence spaces etc. Moreover, computer scientists are
acquainted with the use of syntactic partial objects (fix-point operators in programming
languages, Böhm trees of the λ-calculus etc.) and various modern proof formalisms, such
as Girard’s Ludics, also incorporate partiality for enlarging the world of “proof-objects”
so as to allow the simultaneous existence of “proofs” and “counter-proofs” in order to
obtain a rich duality theory on top of which a notion of totality discriminating genuine
proofs from partial proof-objects can be developed.
It is only when we observed that the differential extension of LL is the mirror image of
the structural (and dereliction) rules of LL that we considered this extension as logically
meaningful and worth being studied more deeply. The price to pay was the necessity
of accepting an intrinsic non determinism and partiality in logic (these two extensions
being related: failure is the neutral element of non-determinism), but the gain was a new
viewpoint on the exponentials, related to the Taylor Formula of calculus.
In LL, the exponential is usually thought of as the modality of duplicable information.
Linear functions are not allowed to copy their arguments and are therefore very limited in
terms of computational expressive power, the exponential allows one to define non linear
functions which can duplicate and erase their arguments and are therefore much more
powerful. This duplication and erasure capability seems to be due to the presence of the
rules of contraction and weakening in LL, but this is not quite true: the genuinely infinite
rule of LL is promotion which makes a proof duplicable an arbitrary number of times,
Differential Linear Logic 3

and erasable. This fact could not be observed in LL because promotion is the only rule of
LL which allows one to introduce the “!” modality: without promotion, it is impossible
to build a proof object that can be cut on a contraction or a weakening rule.
In Differential LL (DiLL), there are two new rules to introduce the “!” modality: coweak-
ening and codereliction. The first of these rules allows one to introduce an empty proof
of type !A and the second one allows one to turn a proof of type A into a proof of type
!A, without making it duplicable in sharp contrast with the promotion rule. The last new
rule, called cocontraction, allows one to merge two proofs of type !A for creating a new
proof of type !A. This latter rule is similar to the tensor rule of ordinary LL with the
difference that the two proofs glued together by a cocontraction must have the same type
and cannot be distinguished anymore deterministically, whereas the two proofs glued by
a tensor can be separated again by cutting the resulting proof against a par rule. These
new rules are called costructural rules to stress the symmetry with the usual structural
rules of LL.
DiLL has therefore a finite fragment which contains the standard “?” rules (weakening,
contraction and dereliction) as well as the new “!” ones (coweakening, cocontraction and
codereliction), but not the promotion rule. Cut elimination in this system generates
sums of proofs, and therefore it is natural to endow proofs with a vector space (or
module) structure over a field (or more generally over a semi-ring1 ). This fragment has
the following pleasant properties:
— It enjoys strong normalization, even in the untyped case, as long as one considers only
proof-nets which satisfy a correctness criterion similar to the standard Danos-Regnier
criterion for multiplicative LL (MLL).
— In this fragment, all proofs are linear combinations of “simple proofs” which do not
contain linear combinations: this is possible because all the syntactic constructions
of this fragment are multilinear. So proofs are similar to polynomials or power series,
simple proofs playing the role of monomials in this algebraic analogy which is strongly
suggested by the denotational models of DiLL.
Moreover, it is possible to transform any instance of the promotion rule (which is applied
to a sub-proof π) into an infinite linear combination of proofs containing copies of π: this
is the Taylor expansion of promotion. This operation can be applied hereditarily to all
instances of the promotion rule in a proof, giving rise to an infinite linear combinations
of promotion-free DiLL simple proofs with positive rational coefficients.

Outline. We start with a syntactic presentation of DiLL, in a proof-net formalism which

uses terms instead of graphs (in the spirit of Abramsky’s linear chemical abstract ma-
chine (Abr93) or of the formalisms studied by Fernandez and Mackie, see for instance (FM99;
MS08)) and we present a categorical formalism which allows us to describe denotational
models of DiLL. We define the interpretation of proof-nets in such categories.

1 This general setting allows us to cover also “qualitative” situations where sums of proofs are lubs in
a poset.
Thomas Ehrhard 4

Then we shortly describe a differential λ-calculus formalism and we summarize some

results, giving bibliographical references.
The end of the paper is devoted to concrete models of DiLL. We briefly review the rela-
tional model, which is based on the ∗-autonomous category of sets and relations (with the
usual cartesian product of sets as tensor product) because it underlies most denotational
models of (differential) LL. Then we describe the finiteness space model which was one
of our main motivations for introducing DiLL. We provide a thorough description of this
model, insisting on various aspects which were not covered by our initial presentation
in (Ehr05) such as linear boundedness (whose relevance in this semantical setting has
been pointed out by Tasson in (Tas09b; Tas09a)), or the fact that function spaces in the
Kleisli category admit an intrinsic description.

One important step in our presentation of the categorical setting for interpreting dif-
ferential LL is the notion of an exponential structure. It is the categorical counterpart of
the finitary fragment of DiLL, that is, the fragment DiLL0 where the promotion rule is
not required to hold.
An exponential structure consists of a preadditive2 ∗-autonomous category L together
with an operation which maps any object X of L to an object !X of L equipped with
a structure of ⊗-bialgebra (representing the structural and costructural rules) as well as
a “dereliction” morphism in L(!X, X) and a “codereliction” morphism L(X, !X). The
important point here is that the operation X 7→ !X is not assumed to be functorial (it
has nevertheless to be a functor on isomorphisms). Using this simple structure, we define
in particular morphisms ∂ X ∈ L(!X ⊗ X, !X) and ∂X ∈ L(!X, !X ⊗ X).
An element of L(!X, Y ) can be considered as a non-linear morphism from X to Y (some
kind of generalized polynomial, or analytical function), but these morphisms cannot be
composed. It is nevertheless possible to define a notion of polynomial such morphism,
and these polynomial morphisms can be composed, giving rise to a category which is
cartesian if L is cartesian.
By composition with ∂ X ∈ L(!X ⊗ X, !X), any element f of L(!X, Y ) can be differen-
tiated, giving rise to an element f 0 of L(!X ⊗ X, Y ) that we consider as its derivative3 .
This operation can be performed again, giving rise to f 00 ∈ L(!X ⊗ X ⊗ X, Y ) and, as-
suming that cocontraction is commutative, this morphism is symmetric in its two last
linear parameters (a property usually known as Shwarz Lemma).
In this general context, a very natural question arises. Given a morphism g ∈ L(!X ⊗
X, Y ) whose derivative g 0 ∈ L(!X ⊗ X ⊗ X, Y ) is symmetric, can one always find a
morphism f ∈ L(!X, Y ) such that g = f 0 ? Inspired by the usual proof of Poincaré’s
Lemma, we show that such an antiderivative is always available as soon as the natural
morphism Id!X +(∂ X ∂X ) ∈ L(!X, !X) is an isomorphism for each object X. We explain

2 This means that the monoidal category is enriched over commutative monoids. Actually, we assume
more generally that it is enriched over the category of k-modules, where k is a given semi-ring.
3 Or differential, or Jacobian: by monoidal closedness, f 0 can be seen as an element of L(!X, X ( Y )
where X ( Y is the object of morphisms from X to Y in L, that is, of linear morphisms from X to
Y , and the operation f 7→ f 0 satisfies all the ordinary properties of differentiation.
Differential Linear Logic 5

how this property is related to a particular case of integration by parts. We also describe
briefly a syntactic version of antiderivatives in a promotion-free differential λ-calculus.
To interpret the whole of DiLL, including the promotion rule, one has to assume that !
is an endofunctor on L and that this functor is endowed with a structure of comonad and
a monoidal structure; all these data have to satisfy some coherence conditions wrt. the
exponential structure. These conditions are essential to prove that the interpretation of
proof-nets is invariant under the various reduction rules, among which the most com-
plicated one is an LL version of the usual chain rule of calculus. Our main references
here are the work of Bierman (Bie95), Melliès (Mel09) and, for the commutations in-
volving costructural logical rules, our concrete models (Ehr05; Ehr02), the categorical
setting developed by Blute, Cockett and Seely (BCS06) and, very importantly, the work
of Fiore (Fio07).
One major a priori methodological principle applied in this paper is to stick to Clas-
sical Linear Logic, meaning in particular that the categorical models we consider are ∗-
autonomous categories. This is justified by the fact that most of the concrete models we
have considered so far satisfy this hypothesis (with the noticeable exception of (BET12))
and it is only in this setting that the new symmetries introduced by the differential and
costructural rules appear clearly. A lot of material presented in this paper could probably
be carried to a more general intuitionistic Linear Logic setting.
Some aspects of DiLL are only alluded to in this presentation, the most significant
one being certainly the Taylor expansion formula and its connection with linear head
reduction. On this topic, we refer to (ER08; ER06; Ehr10).

Notations
In this paper, a set of coefficients is needed, which has to be a commutative semi-ring.
This set will be denoted as k. In Section 4.3, k will be assumed to be a field but this
assumption is not needed before that section.

1. Syntax for DiLL proof-structures

We adopt a presentation of proof-structures and proof-nets which is based on terms and
not on graphs. We believe that this presentation is more suitable to formalizable math-
ematical developments, although it sometimes gives rise to heavy notations, especially
when one has to deal with the promotion rule (Section 1.5). We try to provide graphical
intuitions on proof-structures by means of figures.

1.1. General constructions

1.1.1. Simple proof-structures. Let V be an infinite countable set of variables. This set is
equipped with an involution x 7→ x such that x 6= x for each x ∈ V.
Let u ⊆ V. An element x of u is bound in u if x ∈ u. One says that u is closed if all
the elements of u are bound in u. If x is not bound in u, one says that x is free in u.
Thomas Ehrhard 6

Let Σ be a set of tree constructors, given together with an arity map ar : Σ → N.

Proof trees are defined as follows, together with their associated set of variables:
— if x ∈ V then x is a tree and V(x) = {x};
— if ϕ ∈ Σn (that is ϕ ∈ Σ and ar(ϕ) = n) and if t1 , . . . , tn are trees with V(ti )∩V(tj ) = ∅
for i 6= j, then t = ϕ(t1 , . . . , tn ) is a tree with V(t) = V(ti ) ∪ · · · ∪ V(tn ). As usual,
when ϕ is binary, we often use the infix notation t1 ϕ t2 rather than ϕ(t1 , t2 ).
A cut is an expression ht | t0 i where t and t0 are trees such that V(t) ∩ V(t0 ) = ∅. We
set V(c) = V(t) ∪ V(t0 ).
A simple proof-structure is a pair p = (→ −c ; →
− →
−
t ) where t is a list of proof trees and →
−c
is a list of cuts, whose sets of variables are pairwise disjoint.
Remark : The order of the elements of → −c does not matter; we could have used multisets
instead of sequences. In the sequel, we consider these sequences of cuts up to permutation.
Bound variables of V(p) can be renamed in the obvious way in p (rename simulta-
neously x and x avoiding clashes with other variables which occur in p) and simple
proof-structures are considered up to such renamings: this is α-conversion. Let FV(p) be
the set of free variables of p. We say p is closed if FV(p) = ∅.
The simplest simple proof-structure is of course ( ; ). A less trivial closed simple proof-
structure is (hx | xi ; ) which is a loop.

1.1.2. LL types. Let A be a set of type atoms ranged over by α, β, . . . , together with an
involution α 7→ α such that α 6= α. Types are defined as follows.
— if α ∈ A then α is a type;
— if A and B are types then A ⊗ B and A`B are types;
— if A is a type then !A and ?A are types.
The linear negation A⊥ of a type A is given by the following inductive definition: α⊥ = α,
(A ⊗ B)⊥ = A⊥ `B ⊥ ; (A`B)⊥ = A⊥ ⊗ B ⊥ ; (!A)⊥ = ?A⊥ and (?A)⊥ = !A⊥ .
An MLL type is a type built using only the ⊗ and ` constructions4 .

1.2. Proof-structures for MLL

Assume that Σ2 = {⊗, `} and that ar(⊗) = ar(`) = 2.
A typing context is a finite partial function Φ (of domain D(Φ)) from V to formulas
such that Φ(x) = (Φ(x))⊥ whenever x, x ∈ D(Φ).

1.2.1. Typing rules. We first explain how to type MLL proof trees. The corresponding
typing judgments of the form Φ `0 t : A where Φ is a typing context, t is a proof tree
and A is a formula.
The rules are
Φ, x : A `0 x : A

4 We do not consider the multiplicative constants 1 and ⊥ because they are not essential for our purpose.
Differential Linear Logic 7

Φ `0 s : A Φ `0 t : B Φ `0 s : A Φ `0 t : B
Φ `0 s ⊗ t : A ⊗ B Φ `0 s`t : A`B
Given a cut c = hs | s0 i and a typing context Φ, one writes Φ `0 c if there is a type A
such that Φ `0 s : A and Φ `0 s0 : A⊥ .
Last, given a simple proof-structure p = (~c ; ~s) with ~s = (s1 , . . . , sn ) and ~c = (c1 , . . . , ck ),
a sequence Γ = (A1 , . . . , Al ) of formulas and a typing context Φ, one writes Φ `0 p : Γ if
l = n and Φ `0 si : Ai for 1 ≤ i ≤ n and Φ `0 ci for 1 ≤ i ≤ k.

1.2.2. Logical judgments. A logical judgment is an expression Φ ` p : Γ where Φ is a

typing context, p is a simple proof-structure and Γ is a list of formulas.
If one can infer that Φ ` p : Γ, this means that the proof-structure p represents a
proof of Γ. Observe that the inference rules coincide with the rules of the MLL sequent
calculus.
We give now these logical rules.
axiom
Φ, x : A, x : A⊥ ` ( ; x, x) : A, A⊥
Φ ` (~c ; t1 , . . . , tn ) : A1 , . . . , An
permutation rule, σ ∈ Sn
Φ ` (~c ; tσ(1) , . . . , tσ(n) ) : Aσ(1) , . . . , Aσ(n)
→
− → −
Φ ` (→−c ; → −s , s) : Γ, A Φ ` ( d ; t , t) : ∆, A⊥
cut rule
Φ ` (→ −c , →
−
d , hs | ti ; → − →
−
s , t ) : Γ, ∆
Φ ` (→ −c ; →
−
t , s, t) : Γ, A, B
→
− →
− `-rule
Φ ` ( c ; t , s`t) : Γ, A`B
→
− → −
Φ ` (→ −c ; → −s , s) : Γ, A Φ ` ( d ; t , t) : ∆, B
⊗-rule
Φ ` (→ −c , →− →
d ;−
→
−
s , t , s ⊗ t) : Γ, ∆, A ⊗ B
We add the mix rule for completeness because it is quite natural denotationally. Notice
however that it is not necessary. In particular, mix-free proof-nets are closed under cut
elimination.
Φ ` (~c ; ~s) : Γ Φ ` (d~ ; ~t) : ∆
mix rule
Φ ` (~c, d~ ; ~s, ~t) : Γ, ∆
Lemma 1. If Φ ` p : Γ then Φ `0 p : Γ and V(p) is closed.

Proof. Straightforward induction on derivations. 2

1.3. Reducing proof-structures

The basic reductions concern cuts, and are of the form
→
− → −
c ;cut ( d ; t )
→
− →
−
where c is a cut, d = (d1 , . . . , dn ) is a sequence of cuts and t = (t1 , . . . , tk ) is a sequence
fo trees.
With similar notational conventions, here are the deduction rules for the reduction of
MLL proof-structures.
Thomas Ehrhard 8

x1 xl
...
axiom links
p= ... ... ... ... ... ...
s1 s01 sn s0n t1 tk
... ...

Fig. 1. A simple proof-structure

x1 xl
...

p
1 k
...

Fig. 2. A synthetic representation of the proof-structure of Figure 1

→
− → −
c ;cut ( d ; t )
→
− − →
− →− − → − context
(c, b ; →s ) ;cut ( d , b ; →s, t)
x∈/ V(s)
→
− −c ; →
− ax-cut
(hx | si, c ; t ) ; (→
→
−
cutt ) [s/x]
For applying the latter rule (see Figure 3), we assume that x ∈ / V(s). Without this re-
striction, we would reduce the cyclic proof-structure (hx | xi ; ) to ( ; ) and erase the cycle
which is certainly not acceptable from a semantic viewpoint. For instance, in a model
of proof-structures based on finite dimension vector spaces, the semantics of (hx | xi ; )
would be the dimension of the space interpreting the type of x (trace of the identity).
Remark : We provide some pictures to help understand the reduction rules on proof
structures. In these pictures, logical proof-net constructors (such as tensor, par etc.) are
represented as white triangles labeled by the corresponding symbol – they correspond to
the cells of interaction nets or to the links of proof-nets – and subtrees are represented
as gray triangles.
Wires represent the edges of a proof tree. We also represent axioms and cuts as wires:
an axiom looks like and a cut looks like . In Figure 3, we indicate the variables
associated with the axiom, but in the next pictures, this information will be kept implicit.
Figure 1 represents the simple proof-structure
p = (hs1 | s01 i, . . . , hsn | s0n i ; t1 , . . . , tk ) .
with free variables x1 , . . . , xl . The box named axiom links contains axioms connecting
variables occurring in the trees s1 , . . . , sn , s01 , . . . , s0n , t1 , . . . , tk . When we do not want to
be specific about its content, we represent such a simple proof-structure as in Figure 2
by a gray box with indices 1, . . . , k on its border for locating the roots of the trees of p.
The same kind of notation will be used also for proof-structures which are not necessarily
simple, see the beginning of Paragraph 1.4.3 for this notion.
Differential Linear Logic 9

s
x x
−
→ s ;cut −
→
(−
→
c ; t) (−
→
c ; t)

Fig. 3. The axiom/cut reduction

s1 s2 t1 t2
s1 s2 t1 t2
` ⊗ ;cut

Fig. 4. The tensor/par reduction

In MLL, we have only one basic reduction (see Figure 4):

hs1 `s2 | t1 ⊗ t2 i ;cut (hs1 | t1 i, hs2 | t2 i ; )

1.4. DiLL0
This is the promotion-free fragment of differential LL. In DiLL0 , one extends the signature
of MLL with new constructors:
— Σ0 = {w, w}, called respectively weakening and coweakening.
— Σ1 = {d, d}, called respectively dereliction and codereliction.
— Σ2 = {`, ⊗, c, c}, the two new constructors being called respectively contraction and
cocontraction.
— Σn = ∅ for n > 2.

1.4.1. Typing rules. The typing rules for the four first constructors are similar to those
of MLL.
Φ `0 w : ?A Φ `0 w : !A
Φ `0 t : A Φ `0 t : A
Φ `0 d(t) : ?A Φ `0 d(t) : !A
The two last rules require the subtrees to have the same type.
Φ `0 s1 : ?A Φ `0 s2 : ?A Φ `0 s1 : !A Φ `0 s2 : !A
Φ `0 c(s1 , s2 ) : ?A Φ `0 c(s1 , s2 ) : !A

1.4.2. Logical rules. The additional logical rules are as follows.

Φ ` (→
−c ; →
−
s):Γ co-weakening
→
− →
− weakening Φ ` ( ; w) : !A
Φ ` ( c ; s , w) : Γ, ?A
Φ ` (→−c ; →
−
s , s) : Γ, A
→
− →
− dereliction
Φ ` ( c ; s , d(s)) : Γ, ?A
Thomas Ehrhard 10

? ! ;cut

Fig. 5. Weakening/coweakening reduction

s t

? ! ;cut 0 ? ! ;cut 0

Fig. 6. Dereliction/coweakening and weakening/codereliction reductions

Φ ` (→−c ; →
−
s , s) : Γ, A
→
− →
− co-dereliction
Φ ` ( c ; s , d(s)) : Γ, !A

Φ ` (→
−c ; →
−
s , s1 , s2 ) : Γ, ?A, ?A
→
− →
− contraction
Φ ` ( c ; s , c(s1 , s2 )) : Γ, ?A

→
− → −
Φ ` (→
−c ; →
−
s , s) : Γ, !A Φ ` ( d ; t , t) : ∆, !A
co-contraction
Φ ` (→−c , →
− →
d ;−
→
−
s , t , c(s, t)) : Γ, ∆, !A

1.4.3. Reduction rules. To describe the reduction rules associated with these new con-
structions, we need to introduce formal sums (or more generally k-linear combinations)
of simple proof-structures called proof-structures in the sequel, and denoted with cap-
ital letters P, Q, . . . Such an extension by linearity of the syntax was already present
in (ER03).
The empty linear combination 0 is a particular proof-structure which plays an impor-
tant role. As linear combinations, proof-structures can be linearly combined.
The typing rule for linear combinations is

∀i ∈ {1, . . . , n} Φ ` pi : Γ and µi ∈ k
Pn sum
Φ ` i=1 µi pi : Γ
Differential Linear Logic 11

s1 s2

s1 ! s2 !
? ! ;cut

t1 t2

? t1 ? t2
? ! ;cut

Fig. 7. Contraction/coweakening and weakening/cocontraction reductions

s t
s t
? ! ;cut

Fig. 8. Dereliction/codereliction reduction

The new basic reduction rules are:

hw | wi ;cut ( ; ) see Figure 5.

hd(s) | wi ;cut 0
hw | d(t)i ;cut 0 see Figure 6.
hc(s1 , s2 ) | wi ;cut (hs1 | wi, hs2 | wi ; )
hw | c(t1 , t2 )i ;cut (hw | t1 i, hw | t2 i ; ) see Figure 7.
hd(s) | d(t)i ;cut (hs | ti ; ) see Figure 8.
hc(s1 , s2 ) | d(t)i ;cut (hs1 | d(t)i, hs2 | wi ; ) + (hs1 | wi, hs2 | d(t)i ; )
hd(s) | c(t1 , t2 )i ;cut (hd(s) | t1 i, hw | t2 i ; ) + (hw | t1 i, hd(s) | t2 i ; )
see Figure 9.
hc(s1 , s2 ) | c(t1 , t2 )i
;cut (hs1 | c(x11 , x12 )i, hs2 | c(x21 , x22 )i, hc(x11 , x21 ) | t1 i, hc(x12 , x22 ) | t2 i ; )
see Figure 10.

In the last reduction rule, the four variables that we introduce are pairwise distinct and
fresh. Up to α-conversion, the choice of these variables is not relevant.
The contextual rule must be extended, in order to take sums into account.
c ;cut P
→
− → context
−
(c, b ; s ) ;
P
− →
→
− →
→ − → − →−
− P ·( c , b ; s, t )
cut p=( c ; t ) p

Remark : In the premise of this rule, P is a linear combination of proof-structures, so that

for a given proof-structure p = (→−c ; →
−
t ), Pp ∈ k is the coefficient of the proof-structure
p in this linear combination P . The sum which appears in the conclusion ranges over all
Thomas Ehrhard 12

s1 s2 t s1 t s2 s1 s2 t
! !
? ! ;cut ! + !

s t1 t2
s t1 t2 t1 s t2
? ?
? ! ;cut ? + ?

Fig. 9. Contraction/codereliction and dereliction/cocontraction reductions

s1 s2 t1 t2
s1 s2 ! ! ? ? t1 t2
? ! ;cut

Fig. 10. Contraction/cocontraction reduction

possible proof-structures p, but there are only finitely many p’s such that Pp 6= 0 so that
→
− −
this sum is actually finite. A particular case of this rule is c ;cut 0 ⇒ (c, b ; →
s ) ;cut 0.

1.5. Promotion
Let p = (→−c ; →
−
s ) be a simple proof-structure. The width of p is the number of elements
of the sequence →−s.
By definition, a proof-structure of width n is a finite linear combination of simple
proof-structures of width n.
Observe that 0 is a proof-structure of width n for all n.
Let P be a proof-structure5 of width n + 1. We introduce a new constructor6 called
promotion box, of arity n:
P !(n) ∈ Σn .
The presence of n in the notation is useful only in the case where P = 0 so it can
most often be omitted. The use of a non necessarily simple proof structure P in this
construction is crucial: promotion is not a linear construction and is actually the only
non linear construction of (differential) LL.
So if t1 , . . . , tn are trees, P !(n) (t1 , . . . , tn ) is a tree. Pictorially, this tree will typically
be represented as in Figure 11. A simple net p appearing in P is typically of the form
(→
−c ; →
−
s ) and its width is n + 1, so that → −s = (s1 , . . . , sn , s). The indices 1, . . . , n and

5 To be completely precise, we should also provide a typing environment for the free variables of P ; this
can be implemented by equipping each variable with a type.
6 The definitions of the syntax of proof trees and of the signature Σ are mutually recursive when
promotion is taken into account.
Differential Linear Logic 13

t1 tn

···
1 n
P
•
!

Fig. 11. Graphical representation of a tree whose outermost constructor is a promotion

box

• which appear on the gray rectangle representing P stand for the roots of these trees
s1 , . . . , sn and s.

1.5.1. Typing rule. The typing rule for this construction is

Φ `0 P : ?A⊥ ⊥
1 , . . . , ?An , B Φ `0 ti : !Ai (i = 1, . . . , n)
Φ `0 P !(n) (t1 , . . . , tn ) : !B

1.5.2. Logical rule. The logical rule associated with this construction is the following.
→
−
Φ ` P : ?A⊥ ⊥
1 , . . . , ?An , B Φ ` (→ −
ci ; ti , ti ) : Γi , !Ai (i = 1, . . . , n)
→
− →
−
Φ ` (→−
c1 , . . . , →
−
cn ; t1 , . . . , tn , P !(n) (t1 , . . . , tn )) : Γ1 , . . . , Γn , !B
Remark : This promotion rule is of course highly debatable. We choose this presentation
because it is compatible with our tree-based presentation of proof-structures.

1.5.3. Cut elimination rules. The basic reductions associated with promotion are as fol-
lows.
hP !(n) (t1 , . . . , tn ) | wi ;cut (ht1 | wi, . . . , htn | wi ; ) see Figure 12.
hP !(n)
(t1 , . . . , tn ) | d(s)i ;cut
Pp · (→
−c , hs | t i, . . . , hs | t i, hs0 | si ; )
X
1 1 n n
p=(→
−
c ;→
−
s ,s0 )

see Figure 13.

hP !(n)
(t1 , . . . , tn ) | c(s1 , s2 )i ;cut (hP !(n)
(x1 , . . . , xn ) | s1 i, hP !(n)
(y1 , . . . , yn ) | s2 i,
ht1 | c(x1 , y1 )i, . . . , htn | c(xn , yn )i ; )
see Figure 14.
In the second reduction rule, one has to avoid clashes of variables.
In the last reduction rules, the variables x1 , . . . , xn and y1 , . . . , yn that we introduce
together with their covariables are assumed to be pairwise distinct and fresh. Up to
α-conversion, the choice of these variables is not relevant.

1.5.4. Commutative reductions. There are also auxiliary reduction rules sometimes called
commutative reductions which do not deal with cuts — at least in the formalization of
nets we present here.
Thomas Ehrhard 14

t1 tn

···
1 n
t1 ? tn ?
P ;cut ···
•
! ?

Fig. 12. Promotion/weakening reduction

t1 tn

···
1 n
s t1 tn s
P ;cut ··· P
• n 1 •
···
! ?

Fig. 13. Promotion/dereliction reduction

The format of these reductions is

t ;com P
where t is a simple tree and P is a (not necessarily simple) proof-structure whose width
is exactly 1.
The first of these reductions is illustrated in Figure 15 and deals with the interaction
between two promotions.

P !(n+1) (t1 , . . . , ti−1 , Q!(k) (ti , . . . , tk+i−1 ), tk+i , . . . , tk+n )

;com ( ; R!(k+n) (t1 , . . . , tk+n )) (1)
where
Pp · (→
−c , hs | Q!(k) (x , . . . , x )i ;
X
R= i 1 k
p=(→
−
c ; s1 ,...,sn+1 ,s)

s1 , . . . , si−1 , x1 , . . . , xk , si+1 , . . . , sn+1 , s) .

Remark : In Figure 15 and 17, for graphical reasons, we don’t follow exactly the notations
used in the text. For instance in Figure 15, the correspondence with the notations of (1)
is given by v1 = t1 ,. . . ,vi−1 = ti−1 , u1 = ti ,. . . , uk = tk+i−1 , vi = tk+i ,. . . ,vn = tk+n .
Remark : Figure 15 is actually slightly incorrect as the connections between the “auxil-
iary ports” of the cocontraction rule within the promotion box of the right hand proof-
structure and the main ports of the trees u1 , . . . , uk are represented as vertical lines
whereas they involve axioms (corresponding to the pairs (xi , xi ) for i = 1, . . . , k in the
formula above). The same kind of slight incorrectness occurs in figure 17.
Differential Linear Logic 15

t1 tn

···
n
1 s1 s2
P ;cut
•
! ?

··· ···
1 n 1 n
t1 ? tn ?
s1 P ··· P s2
• •
! !

Fig. 14. Promotion/contraction reduction

The three last commutative reductions deal with the interaction between a promotion
and the costructural rules.
Interaction between a promotion and a coweakening, see Figure 16:

P !(n+1) (t1 , . . . , ti−1 , w, ti , . . . , tn ) ;com ( ; R!(n) (t1 , . . . , tn ))

where
Pp · (→
−c , hs | wi ; s , . . . , s , s , . . . , s
X
R= i 1 i−1 i+1 n+1 , s) .
p=(→
−
c ;→
−
s ,s)

Interaction between a promotion and a cocontraction, see Figure 17:

P !(n+1) (t1 , . . . , ti−1 , c(ti , ti+1 ), ti+2 , . . . , tn+2 ) ;com ( ; R!(n+2) (t1 , . . . , tn+2 )) (2)

where
Pp (→
−c , hs | c(x, y)i ; s , . . . , s , x, y, s , . . . , s , s) .
X
R= i 1 i−1 i n
p=(→
−
c ;→
−
s ,s)

The interaction between a promotion and a codereliction is a syntactic version of the

chain rule of calculus, see Figure 18.

P !(n+1) (t1 , . . . , ti−1 , d(u), ti+1 , . . . , tn+1 )

X
;com Pp · (hsi | d(u)i,
p=(→
−
c ; s1 ,...,sn+1 ,s)

hc(x1 , s1 ) | t1 i, . . . , hc(x\
i , si ) | ti i, . . . , hc(xn+1 , sn+1 ) | tn+1 i ;

c(P !(n+1) (x1 , . . . , xi−1 , w, xi+1 , . . . , xn+1 ), d(s)))

where we use the standard notation a1 , . . . , abi , . . . , an for the sequence

a1 , . . . , ai−1 , ai+1 , . . . , an .

We also have to explain how these commutative reductions can be used in arbitrary
Thomas Ehrhard 16

u1 uk
u1 uk

··· ···
1 k 1 k
v1 Q vn Q
• v1 • vn
! ;com !

··· ··· ···

1 i n 1 n
P P
• • i

! !

Fig. 15. Promotion/promotion commutative reduction

t1 tn
t1 tn
!
!
··· ···
1 i n ;com
P ···
n
• 1
P
! • i

Fig. 16. Promotion/coweakening commutative reduction

contexts. We deal first with the case where such a reduction occurs under a constructor
symbol ϕ ∈ Σn+1 .
t ;com P
ϕ(→−
u , t, →
−
v ) ;com p=(→ →
− →
− →
−
P
c ; w) Pp · ( c ; ϕ( u , w, v ))
−

Next we deal with the case where t occurs in outermost position in a proof-structure.
There are actually two possibilities.
t ;com P
→
− →
− →
−
( c ; u , t, v ) ;
P
→
−
−c , →
P · (→
− →
d ;−
u , w, →
−
v)
com p=( d ; w) p

t ;com P
0 →
− →
−
(ht | t i, c ; t ) ;com p=(→
P
−
−c , →
P · (→
− →
−
d , hw | t0 i ; t )
d ; w) p
We use ; for the union of the reduction relations ;cut and ;com .
This formalization of nets enjoys a subject reduction property.

Theorem 2. If Φ ` p : Γ and p ; P then Φ0 ` P : Γ for some Φ0 which extends Φ.

The proof is a rather long case analysis. We need to consider possible extensions of Φ
because of the fresh variables which are introduced by several reduction rules.
Differential Linear Logic 17

u1 u2 v1 u1 u2 vn

v1 vn
! !

··· ··· ···

1 i n ;com 1 n
P P
• • i

! !

Fig. 17. Promotion/cocontraction commutative reduction rules

u
u

t1 tn ! ! P
1 i n •
! ··· ···
··· ···
1 i n
··· ···
1 i n ;com P
• t1 ? ··· tn ?
P
•
! !
!

Fig. 18. Promotion/codereliction commutative reduction (Chain Rule)

1.6. Correctness criterion and properties of the reduction

Let P be proof-structure, Φ be a closed typing context and Γ be a sequence of formulas
such that Φ `0 P : Γ. One says that P is a proof-net if it satisfies Φ ` P : Γ. A correctness
criterion is a criterion on P which guarantees that P is a proof-net; of course, saying
that Φ ` P : Γ is a correctness criterion, but is not a satisfactory one because it is not
easy to prove that it is preserved by reduction.
Various such criteria can be found in the literature, but most of them apply to proof-
structures considered as graphical objects and are not very suitable to our term-based
approach. We rediscovered recently a correctness criterion initially due to Rétoré (Ret03)
which seems more convenient for the kind of presentation of proof-structures that we use
here, see (Ehr14). This criterion, which is presented for MLL, can easily be extended to
the whole of DiLL.
So far, the reduction relation ; is defined as a relation between simple proof-structures
and proof-structures. It must be extended to a relation between arbitrary proof-structures.
This is done by means of the following rules
p;P p;P µ ∈ k \ {0}
p+Q;P +Q µ·p;µ·P
Thomas Ehrhard 18

As it is defined, our reduction relation does not allow us to perform the reduction within
boxes. To this end, one should add the following rule.
P ;Q
P !(n) (t1 , . . . , tn ) ; Q!(n) (t1 , . . . , tn )
It is then possible to prove basic properties such as confluence and normalization7 . For
these topics, we refer mainly to the work of Pagani (Pag09), Tranquilli (PT09; Tra09;
PT11), Gimenez (Gim11). We also refer to Vaux (Vau09) for the link between the alge-
braic properties of k and the properties of ;, in a simpler λ-calculus setting.
Of course these proofs should be adapted to our presentation of proof structures. This
has not been done yet but we are confident that it should not lead to difficulties.

2. Categorical denotational semantics

We describe now the denotational semantics of DiLL in a general categorical setting. This
will give us an opportunity to provide more intuitions about the rules of this system. More
intuition about the meaning of the differential constructs of DiLL is given in Section 3.

2.1. Notations and conventions

Let C be a category. Given objects X and Y of C, we use C(X, Y ) for the set of morphisms
from X to Y . Given f ∈ C(X, Y ) and g ∈ C(Y, Z), we use g f for the composition of f
and g, which belongs to C(X, Z). In specific situations, we use also the notation g ◦ f .
When there are no ambiguities, we use X instead of IdX to denote the identity from X
to X.
Given n ∈ N and a functor F : C → D, we use the same notation F for the functor
C n → Dn defined in the obvious manner: F (X1 , . . . , Xn ) = (F (X1 ), . . . , F (Xn )) and
similarly for morphisms. If F, G : C → D are functors and if T is a natural transformation,
we use again the same notation T for the corresponding natural transformation between
the functors F, G : C n → Dn , so that TX1 ,...,Xn = (TX1 , . . . , TXn ).

2.2. Monoidal structure

A symmetric monoidal category is a structure (L, I, , λ, ρ, α, σ) where L is a category,
I is an object of L,  : L2 → L is a functor and λX ∈ L(I  X, X), ρX ∈ L(X  I, X),
αX,Y,Z ∈ L((X  Y )  Z, X  (Y  Z)) and σX,Y ∈ L(X  Y, Y  X) are natural
isomorphisms satisfying coherence conditions which can be expressed as commutative
diagrams, and that we do not recall here. Following McLane (Mac71), we present these
coherence conditions using a notion of monoidal trees (called binary words in (Mac71)).
Monoidal trees (or simply trees when there are no ambiguities) are defined by the
following syntax.

7 For confluence, one needs to introduce an equivalence relation on proof-structures which expresses
typically that contraction is associative, see (Tra09). For normalization, some conditions have to be
satisfied by k; typically, it holds if one assumes that k = N but difficulties arise if k has additive
inverses.
Differential Linear Logic 19

— hi is the empty tree

— ∗ is the tree consisting of just one leaf
— and, given trees τ1 and τ2 , hτ1 , τ2 i is a tree.
Let L(τ ) be the number of leaves of τ , defined by

L(hi) = 0
L(∗) = 1
L(hτ1 , τ2 i) = L(τ1 ) + L(τ2 ) .

Let Tn be the set of trees τ such that L(τ ) = n. This set is infinite for all n.
Let τ ∈ Tn . Then we define in an obvious way a functor τ : Ln → L. On object, it is
defined as follows:

hi = I
∗ X = X
→
− →
−
hτ1 ,τ2 i (X1 , . . . , XL(τ1 ) , Y1 , . . . , YL(τ2 ) ) = (τ1 ( X ))  (τ2 ( Y )) .

The definition on morphisms is similar.

2.2.1. Generalized associativity. Given τ1 , τ2 ∈ Tn , the isomorphisms λ, ρ and α of the

monoidal structure of L allow us to build an unique natural isomorphism ττ12 from τ1
to τ2 . We have in particular
hhi,∗i
λ X = ∗ X
h∗,hii
ρX = ∗ X
hh∗,∗i,∗i
αX,Y,Z = h∗,h∗,∗ii
X,Y,Z

The coherence commutation diagrams (which include the McLane Pentagon) allow one
→
− →
−
indeed to prove that all the possible definitions of an isomorphism τ1 ( X ) → τ2 ( X )
using these basic ingredients give rise to the same result. This is McLane coherence
Theorem for monoidal categories. In particular the following properties will be quite
useful:
ττ →
− = Id τ →
X
−
 (X )
and ττ23 →
− τ
X
1−
τ2 →
X
= ττ13 →
− .
X
(3)
→
−
We shall often omit the indexing sequence X when using these natural isomorphisms,
writing στ instead of στ →
−.
X

2.2.2. Generalized symmetry. Let n ∈ N. Let ϕ ∈ Sn , we define a functor ϕ b : Ln → Ln

by ϕ(X
b 1 , . . . , Xn ) = (Xϕ(1) , . . . , Xϕ(n) )
Assume that the monoidal category L is also symmetric. The corresponding additional
structure allows one to define a natural isomorphism  b ϕ,τ from the functor τ to the
functor τ ◦ ϕ. b The correspondence ϕ 7→  b ϕ,τ is of course functorial. Moreover, given
Thomas Ehrhard 20

σ, τ ∈ Tn and ϕ ∈ Sn , the following diagram is commutative

→
− b ϕ,σ
 →
−
σ X σ ϕ(
b X)
σ
τ σ
τ

→
− b ϕ,τ
 →−
τ X τ ϕ(
b X)
(4)
This is a consequence of McLane coherence Theorem for symmetric monoidal cate-
gories.

2.3. *-autonomous categories

A *-autonomous category is a symmetric monoidal category (L, ⊗, λ, ρ, α, σ) equipped
with the following structure:
— an endomap on the objects of L that we denote as X 7→ X ⊥ ;
— for each object X, an evaluation morphism ev⊥ ∈ L(X ⊥ ⊗ X, ⊥), where ⊥ = 1⊥ ;
— a curryfication function cur⊥ : L(U ⊗ X, ⊥) → L(U, X ⊥ )
subject to the following equations (with f ∈ L(U ⊗ X, ⊥) and g ∈ L(V, U ), so that
g ⊗ X ∈ L(V ⊗ X, U ⊗ X)):
ev⊥ (cur⊥ (f ) ⊗ X) = f
cur⊥ (f ) g = cur⊥ (f (g ⊗ X))
cur⊥ (ev⊥ ) = Id .
Then cur⊥ is a bijection. Indeed, let g ∈ L(U, X ⊥ ). Then g ⊗ X ∈ L(U ⊗ X, X ⊥ ⊗ X) and
hence ev⊥ (g ⊗ X) ∈ L(U ⊗ X, ⊥). The equations allow one to prove that the function
g 7→ ev⊥ (g ⊗ X) is the inverse of the function cur⊥ .
For any object X of L, let ηX = cur⊥ (ev⊥ σX,X ⊥ ) ∈ L(X, X ⊥⊥ ).
The operation X 7→ X ⊥ can be extended into a functor Lop → L as follows. Let
f ∈ L(X, Y ), then ηY f ∈ L(X, Y ⊥⊥ ), so ev⊥ ((ηY f ) ⊗ Y ⊥ ) ∈ L(X ⊗ Y ⊥ , ⊥) and we set
f ⊥ = cur⊥ (ev⊥ ((ηY f )⊗Y ⊥ ) σY ⊥ ,X ) ∈ L(Y ⊥ , X ⊥ ). It can be checked that this operation
is functorial.
We assume last that ηX is an iso for each object X.
One sets X ( Y = (X ⊗ Y ⊥ )⊥ and one defines an evaluation morphism ev ∈
L((X ( Y ) ⊗ X, Y ) as follows. We have
ev⊥ ∈ L((X ⊗ Y ⊥ )⊥ ⊗ (X ⊗ Y ⊥ ), ⊥)
hence
hh∗,∗i,∗i
ev⊥ ⊗h∗,h∗,∗ii ∈ L(((X ⊗ Y ⊥ )⊥ ⊗ X) ⊗ Y ⊥ , ⊥) ,
therefore
hh∗,∗i,∗i
cur⊥ (ev⊥ ⊗h∗,h∗,∗ii ) ∈ L((X ⊗ Y ⊥ )⊥ ⊗ X, Y ⊥⊥ )
and we set
hh∗,∗i,∗i
ev = η −1 cur⊥ (ev⊥ ⊗h∗,h∗,∗ii ) ∈ L((X ⊗ Y ⊥ )⊥ ⊗ X, Y ).
Differential Linear Logic 21

Let f ∈ L(U ⊗X, Y ). We have η f ∈ L(U ⊗X, Y ⊥⊥ ), hence cur⊥ −1 (η f ) ∈ L((U ⊗ X)⊗
h∗,h∗,∗ii
Y ⊥ , ⊥), so cur⊥ −1 (η f ) ⊗hh∗,∗i,∗i ∈ L(U ⊗ (X ⊗ Y ⊥ ), ⊥) and we can define a linear cur-
ryfication of f as
h∗,h∗,∗ii
cur(f ) = cur⊥ (cur⊥ −1 (η f ) ⊗hh∗,∗i,∗i ) ∈ L(U, X ( Y ) .

One can prove then that the following equations hold, showing that the symmetric
monoidal category L is closed.

ev (cur(f ) ⊗ X) = f
cur(f ) g = cur(f (g ⊗ X))
cur(ev) = Id

where g ∈ L(V, U ).
It follows as usual that cur is a bijection from L(U ⊗ X, Y ) to L(U, X ( Y ).
We set X`Y = (X ⊥ ⊗ Y ⊥ )⊥ = X ⊥ ( Y ; this operation is the cotensor product,
also called par in linear logic. Using the above properties one shows that this operation
is a functor L2 → L which defines another symmetric monoidal structure on L. The
operation X 7→ X ⊥ is an equivalence of symmetric monoidal categories from (Lop , ⊗) to
(L, `).

2.3.1. MIX. A mix *-autonomous category is a *-autonomous category L where ⊥ is

endowed with a structure of commutative ⊗-monoid8 . So we have two morphisms ξ0 ∈
L(1, ⊥) and ξ2 ∈ L(⊥⊗⊥, ⊥) and some standard diagrams must commute, which express
that ξ0 is left and right neutral for the binary operation ξ2 , and that this binary operation
is associative and commutative. Observe that (ξ2 )⊥ ∈ L(⊥⊥ , (⊥ ⊗ ⊥)⊥ ) so that

ξ20 = (ξ2 )⊥ η1 ∈ L(1, 1`1)

and (1, ξ0 , ξ20 ) is a commutative `-comonoid.

2.3.2. Vectors. Let L be a *-autonomous category and let X1 , . . . , Xn be objects of L.

A (X1 , . . . , Xn )-vector is a family (uτ )τ ∈Tn where uτ ∈ L(1, `τ (X1 , . . . , Xn )) satisfies
uτ 0 = `ττ 0 uτ for all τ, τ 0 ∈ Tn . Of course such a vector u is determined as soon as one of
the uτ ’s is given. The point of this definition is that none of these uτ ’s is more canonical
than the others, that is why we find more convenient to deal with the whole family u.
~ 1 , . . . , Xn ) be the set of these vectors. Notice that, since Tn is infinite for all n,
Let L(X
all vectors are infinite families.

~ ⊥ , X) by setting axτ =
2.3.3. MLL vector constructions. Let X ∈ L. We define ax ∈ L(X
h∗,∗i −1 hhi,∗i
`τ cur(ηX ⊗∗ ) ∈ L(1, X ⊥⊥
( X) = L(1, X `X) for all τ ∈ T2 .
⊥

8 If we see ⊥ as the object of scalars, which is compatible with the intuition that X ( ⊥ is the dual of
X, that is, the “space of linear forms on X”, then this monoid structure is an internal multiplication
law on scalars.
Thomas Ehrhard 22

~ 1 , . . . , Xn , X, Y ). We define `(u) ∈ L(X

Let u ∈ L(X ~ 1 , . . . , Xn , X`Y ) as follows. Let
τ ∈ Tn , we know that
uhτ,h∗,∗ii ∈ L(1, (`τ (X1 , . . . , Xn )`(X`Y )) = L(1, (`hτ,∗i (X1 , . . . , Xn , X`Y )) .
For any θ ∈ Tn+1 , we set
hτ,∗i
`(u)θ = `θ uhτ,h∗,∗ii ∈ L(1, `θ (X1 , . . . , Xn , X`Y )) .
One sees easily that this definition does not depend on the choice of τ : let τ 0 ∈ Tn , we
have
hτ,∗i hτ,∗i hτ 0 ,∗i
`θ `(u)hτ,h∗,∗ii = `θ `hτ,∗i `(u)hτ 0 ,h∗,∗ii
hτ 0 ,∗i
= `θ `(u)hτ 0 ,h∗,∗ii .
thanks to the definition of vectors and to Equation (3).
Let Ui , Xi be objects of L for i = 1, 2. Given
ui ∈ L(1, Ui `Xi ) = L(1, Ui⊥ ( Xi )
for i = 1, 2, we define
⊗U1 ,X1 ,U2 ,X2 (u1 , u2 ) ∈ L(1, (U1 `U2 )`(X1 ⊗ X2 ))
as follows. We have cur−1 (ui ) ⊗∗hhi,∗i ∈ L(Ui⊥ , Xi ) and hence

v = (cur−1 (u1 ) ⊗∗hhi,∗i ) ⊗ (cur−1 (u2 ) ⊗∗hhi,∗i ) ∈ L(U1⊥ ⊗ U2⊥ , X1 ⊗ X2 ) .

We have
hhi,∗i
cur(v ⊗∗ ) ∈ L(1, ((U1⊥ ⊗ U2⊥ ) ⊗ (X1 ⊗ X2 )⊥ )⊥ )
So we set
hhi,∗i
⊗U1 ,X1 ,U2 ,X2 (u1 , u2 ) = (η(U1⊥ ⊗U2⊥ ) −1 ⊗ (X1 ⊗ X2 )⊥ )⊥ cur(v ⊗∗ )
∈ L(1, (U1 `U2 )`(X1 ⊗ X2 ))
where the natural iso η is defined in Section 2.3.
This construction is natural in the sense that, given fi ∈ L(Xi , Xi0 ), gi ∈ L(Ui , Ui0 ),
one has
((g1 `g2 )`(f1 ⊗ f2 )) ⊗U1 ,X1 ,U2 ,X2 (u1 , u2 )
(5)
= ⊗U10 ,X10 ,U20 ,X20 ((f1 `g1 ) u1 , (f2 `g2 ) u2 )
~ 1 , . . . , Xn , X) and v ∈ L(Y
Let u ∈ L(X ~ 1 , . . . , Yp , Y ). Let σ ∈ Tn and τ ∈ Tp . Then we
have
uhσ,∗i ∈ L(1, (`σ (X1 , . . . , Xn ))`X) and
τ
vhτ,∗i ∈ L(1, (` (Y1 , . . . , Yp ))`Y )
and we set
⊗(u, v)hhσ,τ i,∗i = ⊗`σ (X1 ,...,Xn ),X,`τ (Y1 ,...,Yp ),Y (uhσ,∗i , vhτ,∗i )
∈ L(1, `hhσ,τ i,∗i (X1 , . . . , Xn , Y1 , . . . , Yp , X ⊗ Y )
Differential Linear Logic 23

since

(`σ (X1 , . . . , Xn )`(`τ (Y1 , . . . , Yp )))`(X ⊗ Y )

= `hhσ,τ i,∗i (X1 , . . . , Xn , Y1 , . . . , Yp , X ⊗ Y ) .

Then, given θ ∈ Tn+p+1 , one sets of course

hhσ,τ i,∗i
⊗(u, v)θ = `θ ⊗(u, v)hhσ,τ i,∗i .

One checks easily that this definition does not depend on the choice of σ and τ , using
Equations (3) and (5), and one can check that
~ 1 , . . . , Xn , Y1 , . . . , Yp , X ⊗ Y ) .
⊗(u, v) ∈ L(X
~ 1 , . . . , Xn ) and ϕ ∈ Sn . Given σ ∈ Tn , we have uσ ∈ L(1, `σ (X1 , . . . , Xn ))
Let u ∈ L(X
b ϕ,σ uσ ∈ L(1, `σ (Xϕ(1) , . . . , Xϕ(n) )). Given θ ∈ Tn , we set therefore
and hence `

b ϕ,σ uσ ∈ L(1, `θ (Xϕ(1) , . . . , Xϕ(n) ))

sym(ϕ, u)θ = `σθ `

defining an element sym(ϕ, u) of L(X ~ ϕ(1) , . . . , Xϕ(n) ) which does not depend on the
choice of σ. Indeed, let τ ∈ Tn , we know that uσ = `τσ uτ and hence sym(ϕ, u)θ =
b ϕ,σ `τ uτ = `σ `τ `
`σθ ` b ϕ,τ uτ = `τ `b ϕ,τ uτ , using Diagram (4).
σ θ σ θ
~ 1 , . . . , Xn , X ⊥ ) and v ∈ L(Y
Let u ∈ L(X ~ 1 , . . . , Yp , X). We have

~ 1 , . . . , Xn , Y1 , . . . , Yp , X ⊥ ⊗ X)
⊗(u, v) ∈ L(X

Given σ ∈ Tn and τ ∈ Tp we have

⊗(u, v)hhσ,τ i,∗i ∈ L(1, ((`σ (X1 , . . . , Xn ))`(`τ Y1 , . . . , Yp ))`(X ⊥ ⊗ X))

so that

(Id`ev⊥ ) ⊗(u, v)hhσ,τ i,∗i ∈ L(1, `hhσ,τ i,hii (X1 , . . . , Xn , Y1 , . . . , Yp )) .

Given θ ∈ Tn+p , we set

hhσ,τ i,hii
cut(u, v)θ = `θ (Id`ev⊥ ) ⊗(u, v)hhσ,τ i,∗i

and we define in that way an element cut(u, v) of L(X ~ 1 , . . . , Xn , Y1 , . . . , Yp ).

Assume now that L is a mix *-autonomous category (in the sense of Paragraph 2.3.1).
Let X1 , . . . , Xn and Y1 , . . . , Yp be objects of L. Let u ∈ L(X ~ 1 , . . . , Xn ) and v ∈
~ 1 , . . . , Yp ). Let σ ∈ Tn and τ ∈ Tp . We have uσ ∈ L(1, ` (X1 , . . . , Xn )) and vτ ∈
L(Y σ

L(1, `τ (Y1 , . . . , Yp )). Hence

(uσ `vτ ) ξ20 ∈ L(1, (`σ (X1 , . . . , Xn ))`(`τ (Y1 , . . . , Yp )))

~ 1 , . . . , Xn , Y1 , . . . , Yp ) by setting
and we define therefore mix(u, v) ∈ L(X
hσ,τ i
mix(u, v)θ = `θ (uσ `vτ ) ξ20 ∈ L(1, `θ (X1 , . . . , Xn , Y1 , . . . , Yp ))

for each θ ∈ Tn+p . As usual, this definition does not depend on the choice of σ and τ .
Thomas Ehrhard 24

2.3.4. Interpreting MLL derivations. We start with a valuation which, with each α ∈
A, associates [α] ∈ L in such a way that [α] = [α]⊥ . We extend this valuation to an
interpretation of all MLL types as objects of L in the obvious manner, so that we have a
De Morgan iso dmA ∈ L([A⊥ ], [A]⊥ ) defined inductively as follows.
We set first dmα = Id[α]⊥ . We have dmA ∈ L([A⊥ ], [A]⊥ ) and dmB ∈ L([B ⊥ ], [B]⊥ ),
therefore dmA `dmB ∈ L([(A ⊗ B)⊥ ], [A]⊥ `[B]⊥ ). We have
[A]⊥ `[B]⊥ = ([A]⊥⊥ ⊗ [B]⊥⊥ )⊥
by definition of ` and remember that η[A] ∈ L([A], [A]⊥⊥ ) and so we set
dmA⊗B = (η[A] ⊗ η[B] )⊥ (dmA `dmB ) .
We have [(A`B)⊥ ] = [A⊥ ] ⊗ [B ⊥ ] so dmA ⊗ dmB ∈ L([(A`B)⊥ ], [A]⊥ ⊗ [A]⊥ ). By
definition we have [A`B]⊥ = ([A]⊥ ⊗ [B]⊥ )⊥⊥ . So we set
dmA`B = η[A]⊥ ⊗[B]⊥ (dmA ⊗ dmB ) .
Given a sequence Γ = (A1 , . . . , An ) of types, we denote as [Γ] the sequence of objects
([A1 ], . . . , [An ]).
Given a derivation π of a logical judgment Φ ` p : Γ we define now [π] ∈ L([Γ]),~ by
induction on the structure of π.
Assume first that Γ = (A⊥ , A), p = ( ; x, x) and that π is the axiom
axiom
Φ`p:Γ
We have ax[A] ∈ L(1, [A]⊥ `[A]) and dmA ∈ L([A⊥ ], [A]⊥ ) so that we can set

[π]σ = `σh∗,∗i (dm−1

A `Id[A] ) ax[A]

~
and we have [π] ∈ L([Γ]) as required.
Assume next that Γ = (∆, A`B), that p = (→ −c ; →
−
s , s`t) and that π is the following
derivation, where λ is the derivation of the premise:
Φ ` (→
−c ; →
−
s , s, t) : ∆, A, B
→
− →
− `-rule
Φ ` ( c ; s , s`t) : ∆, A`B
then by inductive hypothesis we have [λ] ∈ L([∆,~ A, B]) and hence we set
[π] = `([λ]) ∈ L([∆,
~ A`B]) .
Assume now that Γ = (∆, Λ, A ⊗ B), that p = (→ −c , →
− →
d ;−
→
−
s , t , s ⊗ t) and π is the
following derivation, where λ is the derivation of the left premise and ρ is the derivation
of the right premise:
→
− → −
Φ ` (→−c ; →
−s , s) : ∆, A Φ ` ( d ; t , t) : Λ, B
⊗-rule
Φ ` (→−c , →
− →
d ;−
→
−
s , t , s ⊗ t) : ∆, Λ, A ⊗ B
then by inductive hypothesis we have [λ] ∈ L([∆, ~ ~
A]) and [ρ] ∈ L([Λ, B]) and hence we
set
~
[π] = ⊗([λ], [ρ]) ∈ L([∆, Λ, A ⊗ B]) .
Assume that ϕ ∈ Sn , Γ = (Aϕ(1) , . . . , Aϕ(n) ), p = (→
−c ; s
ϕ(1) , . . . , sϕ(n) ) and that π is
the following derivation, where λ is the derivation of the premise:
Differential Linear Logic 25

Φ ` (→
−c ; s , . . . , s ) : A , . . . , A
1 n 1 n
permutation rule
Φ`p:Γ
By inductive hypothesis we have [λ] ∈ L([A ~ 1 ], . . . , [An ]) and we set

[π] = sym(ϕ, [λ]) ∈ L([Γ]) ~ .

→
− →
−
Assume that Γ = (∆, Λ), that p = (→ −c , d , hs | ti ; →
−
s , t ) and that π is the following
derivation, where λ is the derivation of the left premise and ρ is the derivation of the
right premise:
→
− → −
Φ ` (→
−c ; →
−
s , s) : ∆, A⊥ Φ ` ( d ; t , t) : Λ, A
cut rule
Φ ` (→−c , →
−
d , hs | ti ; →
− →
−
s , t ) : ∆, Λ
By inductive hypothesis we have [λ] ∈ L([∆, ~ ~
A⊥ ]) and [ρ] ∈ L([Λ, A]). Let n be the length
of ∆. Let σ ∈ Tn . We have [λ]hσ,∗i ∈ L(1, (`σ ([∆]))`[A⊥ ]) and hence (Id`dmA ) [λ]hσ,∗i ∈
~
L(1, (`σ ([∆]))`[A]⊥ ). We define therefore l ∈ L([∆], [A]⊥ ) by lhσ,∗i = (Id`dmA ) [λ]hσ,∗i
(this definition of l does not depend on the choice of σ). We set
~
[π] = cut(l, [ρ]) ∈ L([∆, Λ]) .

Assume last that Γ = (∆, Λ), that p = (→ −c , →

− →
d ;−
→
−
s , t ) and that π is the following
derivation, where λ is the derivation of the left premise and ρ is the derivation of the
right premise:
→
− → −
Φ ` (→
−c ; →
−
s):∆ Φ`(d ; t ):Λ
mix rule
Φ ` (→−c , →
− →
d ;−
→
−
s , t ) : ∆, Λ
~
so that by inductive hypothesis [λ] ∈ L([∆]) ~
and [ρ] ∈ L([Λ]). We set
~
[π] = mix([λ], [ρ]) ∈ L([∆, Λ]) .

The first main property of this interpretation of derivations is that they only depend
on the underlying nets.

Theorem 3. Let π and π 0 be derivations of Φ ` p : Γ. Then [π] = [π 0 ].

The proof is a (tedious) induction on the structure of the derivations π and π 0 .

We use therefore [p] to denote the value of [π] where π is an arbitrary derivation of
` p : Γ.
Remark : It would be much more satisfactory to be able to define [p] directly, without
using the intermediate and non canonical choice of a derivation π. Such a definition would
use directly the fact that p fulfills a correctness criterion in order to build a morphism of
L. It is not very clear yet how to do that in general, though such definitions are available
in many concrete models of LL, such as coherence spaces.
The second essential property of this interpretation is that it is invariant under reduc-
tions (subject reduction)

Theorem 4. Assume that Φ ` p : Γ, Φ ` p0 : Γ and that p ; p0 . Then [p] = [p0 ].

Thomas Ehrhard 26

2.4. Preadditive models

Let L be a *-autonomous category. We say that L is preadditive if each hom-set L(X, Y )
is equipped with a structure of k-module (we use standard additive notations: 0 for
the neutral element and + for the operation), which is compatible with composition of
morphisms and tensor product:
  !
X X X
 νj tj  µi si = νj µi (tj si )
j∈J i∈I (i,j)∈I×J
!  
X X X
µi si ⊗ νj tj  = µi νj (si ⊗ tj )
i∈I j∈J (i,j)∈I×J

where the µi ’s and the νj ’s are elements of k. It follows that,

given a finite family (si )i∈I
P P
of morphisms si ∈ L(U ⊗ X, ⊥), one has cur⊥ µ s
i∈I i i = i∈I µi cur⊥ (si ), and that
the cotensor product is bilinear
!  
X X X
µi si `  νj tj  = µi νj (si `tj ) .
i∈I j∈J (i,j)∈I×J

~ 1 , . . . , Xn ) inherits canonically a
Let (Xi )ni=1 be a family of objects of L. The set L(X
k-module structure.

2.5. Exponential structure

If C is a category, we use Ciso to denote the category whose objects are those of C and
whose morphisms are the isos of C (so Ciso is a groupoid).
Let L be a preadditive *-autonomous category. An exponential structure on L is a
tuple (! , w, c, w, c, d, d) where ! is a functor Liso → Liso and the other ingredients are
natural transformations: wX ∈ L(!X, 1) (weakening), cX ∈ L(!X, !X ⊗ !X) (contraction),
wX ∈ L(1, !X) (coweakening), cX ∈ L(!X ⊗ !X, !X) (cocontraction), dX ∈ L(!X, X)
(dereliction) and dX ∈ L(X, !X) (codereliction).
These morphisms are assumed moreover to satisfy the following properties.
The structure (!X, wX , cX , wX , cX ) is required to be a commutative bialgebra. This
means that (!X, wX , cX ) is a commutative comonoid, (!X, wX , cX ) is a commutative
monoid and that the following diagrams commute (where ϕ = (1, 3, 2, 4) ∈ S4 )
cX ⊗ cX
!X ⊗ !X (!X ⊗ !X) ⊗ (!X ⊗ !X)
cX 1 wX

!X b ϕ,hh∗,∗i,h∗,∗ii
⊗ Id1 !X
cX wX
1
!X ⊗ !X (!X ⊗ !X) ⊗ (!X ⊗ !X)
cX ⊗ cX
Differential Linear Logic 27

Moreover, we also require the following commutations (in the dereliction/cocontraction

hhi,∗i h∗,hii
and codereliction/contraction diagrams, we omit the isos ⊗∗ and ⊗∗ for the sake
of readability).

X dX X dX

0 !X dX ⊗ w X + w X ⊗ dX !X
wX cX
1 !X ⊗ !X

X dX X dX

0 !X d X ⊗ wX + wX ⊗ d X !X
cX
1 wX !X ⊗ !X

and
X dX

IdX !X
X dX

2.5.1. The why not modality. We define ?X = (!(X ⊥ ))⊥ and we extend this operation
to a functor Liso → Liso in the same way (using the contravariant functoriality of ( )⊥ ).
We define
0 ⊥
wX = wX ⊥ : ⊥ → ?X .

Since cX ⊥ : !(X ⊥ ) → !(X ⊥ ) ⊗ !(X ⊥ ), we have c⊥ ⊥ ⊥ ⊥

X ⊥ : (!(X ) ⊗ !(X )) → ?X. But
η!(X ⊥ ) : !(X ) → (?X) , hence (η!(X ⊥ ) ⊗ η!(X ⊥ ) ) : ?X`?X → (!(X ) ⊗ !(X ⊥ ))⊥ and
⊥ ⊥ ⊥ ⊥

we set
c0X = c⊥ ⊥
X ⊥ (η!(X ⊥ ) ⊗ η!(X ⊥ ) ) ∈ L(?X`?X, ?X) .

Then it can be shown that (?X, wX 0

, c0X ) is a commutative `-monoid (that is, a monoid
in the monoidal category (L, `)). Of course, wX 0
and c0X are natural transformations.
Last, we have dX ⊥ : !(X ) → X and hence (dX ⊥ )⊥ : X ⊥⊥ → ?X, so we can define
⊥ ⊥

the natural morphism

d0X = (dX ⊥ )⊥ ηX : X → ?X .

2.5.2. Interpreting DiLL0 derivations. We extend the interpretation of derivations pre-

sented in Section 2.3.4 to the fragment DiLL0 presented in Section 1.4.
We first have to extend the interpretation of formulas – this is done in the obvious
way – and the definition of the De Morgan isomorphisms. We have [(!A)⊥ ] = ?[A⊥ ]
and [!A]⊥ = (![A])⊥ . By inductive hypothesis, we have the iso dmA : [A⊥ ] → [A]⊥ ,
⊥
hence ?dmA : [(!A)⊥ ] = [?(A⊥ )] → ?([A]⊥ ) = !([A]⊥⊥ ) and since we have (!η[A] )⊥ :
⊥
!([A]⊥⊥ ) → (![A])⊥ , we set

dm!A = (!η[A] )⊥ ?dmA ∈ Liso ([(!A)⊥ ], [!A]⊥ ) .

Thomas Ehrhard 28

We have [(?A)⊥ ] = ![A⊥ ] and [?A]⊥ = (!([A]⊥ ))⊥⊥ so we set

dm?A = η!([A]⊥ ) !dmA ∈ Liso ([(?A)⊥ ], [?A]⊥ )
Let π be a derivation of Φ ` p : Γ, where Γ = (A1 , . . . , An ).
Assume first that Γ = (∆, ?A), p = (→ −c ; →
−
s , w) and that π is the following derivation,
denoting with λ the derivation of the premise:
Φ ` (→−c ; →
−s):Γ
→
− →
− weakening
Φ ` ( c ; s , w) : Γ, ?A
~
By inductive hypothesis we have [λ] ∈ L([Γ]). Let τ ∈ Tn , we have [λ]τ ∈ L(1, `τ ([Γ])).
0
We have w[A] ∈ L(⊥, ?[A]) and hence

[λ]τ `w[A]
0
∈ L(1`⊥, `hτ,∗i ([Γ, ?A]))
so that we can set
hτ,∗i
[π]θ = `θ ([λ]τ `w[A]
0
) `∗h∗,hii
for any θ ∈ Tn . The fact that the family [π] defined in that way does not depend on the
~
choice of τ results from the fact that [λ] ∈ L([Γ]).
→
− →
−
Assume that Γ = (∆, ?A), p = ( c ; s , c(t1 , t2 )) and that π is the following derivation,
denoting with λ the derivation of the premise:
` (→
−c ; →
−s , t1 , t2 ) : ∆, ?A, ?A
contraction
` (→
−c ; →
−
s , c(t1 , t2 )) : ∆, ?A
We have c0 ∈ L([?A]`[?A], [?A]). By inductive hypothesis [λ] ∈ L([∆, ~ ?A, ?A]). Let
[A]
τ ∈ Tn where n is the length of ∆. We have
[λ]hτ,h∗,∗ii ∈ L(1, (`τ ([∆]))`([?A]`[?A]))
and hence, given θ ∈ Tn+1 , we set
hτ,∗i
[π]θ = `θ (`τ ([∆])`c0[A] ) [λ]hτ,h∗,∗ii
~
defining in that way [π] ∈ L([∆, ?A]).
Assume that Γ = (!A), p = ( ; w) and that π is the following derivation
co-weakening
` ( ; w) : !A
then, for θ ∈ T1 we set [π]θ = `∗θ ([!A]) w[A] defining in that way an element [π] of L([!A]).
~
→
− →
− → − →
−
Assume that Γ = (∆, Λ, !A), p = ( c , d ; s , t , c(u, v)) and that π is the following
derivation
→
− → −
Φ ` (→−c ; →
−
s , u) : ∆, !A Φ ` ( d ; t , v) : Λ, !A,
co-contraction
Φ ` (→−c , →
− →
d ;−
→
−
s , t , c(u, v)) : ∆, Λ, !A
and we denote with λ and ρ the derivations of the two premises. By inductive hypothesis,
~
we have [λ] ∈ L([∆], ~
[!A]) and [ρ] ∈ L([Λ], ~
[!A]). We have ⊗([λ], [ρ]) ∈ L([∆], [Λ], [!A] ⊗
[!A]).
Let m be the length of ∆ and n be the length of Λ. Let τ ∈ Tm+n , we have ⊗([λ], [ρ])hτ,∗i ∈
L(1, (`τ ([∆], [Λ]))`(([!A] ⊗ [!A])). Hence, given θ ∈ Tm+n+1 we set
hτ,∗i
[π]θ = `θ ((`τ ([∆], [Λ]))`cX ) ⊗([λ], [ρ])hτ,∗i .
Differential Linear Logic 29

~
so that [π] ∈ L([∆], [Λ], [!A]), and this definition does not depend on the choice of τ .
Assume that Γ = (∆, ?A), p = (→ −c ; →−
s , d(s)) and that π is the following derivation
→
− →
−
Φ ` ( c ; s , s) : ∆, A
dereliction
Φ ` (→
−c ; →
−s , d(s)) : ∆, ?A
~
Let λ be the derivation of the premise, so that [λ] ∈ L([∆], [A]).
We have d0[A] : [A] → [?A]. Let n be the length of ∆, let τ ∈ Tn and let θ ∈ Tn+1 . We
set
hτ,∗i
[π]θ = `θ ((`τ ([∆]))`d0[A] ) [λ]hτ,∗i
and we define in that way an element π of L([∆], ~ [?A]) which does not depend on the
choice of τ .
Assume that Γ = (∆, !A), p = (→ −c ; →
−s , d(s)) and that π is the following derivation
→
− →
−
Φ ` ( c ; s , s) : ∆, A
co-dereliction
Φ ` (→−c ; →
−
s , d(s)) : ∆, !A
~
Let λ be the derivation of the premise, so that [λ] ∈ L([∆], [A]). We have d[A] : [A] → [!A].
Let n be the length of ∆, let τ ∈ Tn and let θ ∈ Tn+1 . We set
hτ,∗i
[π]θ = `θ ((`τ ([∆]))`d[A] ) [λ]hτ,∗i
~
and we define in that way an element π of L([∆], [!A]) which does not depend on the
choice of τ .
Pn
Last assume that p = i=1 µi pi , that π is the following derivation
Φ ` pi : Γ ∀i ∈ {1, . . . , n}
Pn sum
Φ ` i=1 µi pi : Γ
and that λi is the derivation of the i-th premise in this derivation. Then by inductive
~
hypothesis we have [λi ] ∈ L([∆]) and we set of course
n
X
[π] = µi [λi ] .
i=1

One can prove for this extended interpretation the same results as for the MLL frag-
ment.

Theorem 5. Let π and π 0 be derivations of Φ ` p : Γ. Then [π] = [π 0 ].

Again, we set [p] = [π] where π is a derivation of Φ ` p : Γ.

Theorem 6. Assume that Φ ` p : Γ, Φ ` p0 : Γ and that p ; p0 . Then [p] = [p0 ].

2.6. Functorial exponential

Let L be a preadditive *-autonomous category with an exponential structure. A promo-
tion operation on L is given by an extension of the functor ! to all morphisms of L
and by a lax symmetric monoidal comonad structure on the “!” operation which satisfies
additional conditions. More precisely:
Thomas Ehrhard 30

— For each f ∈ L(X, Y ) we are given a morphism !f ∈ L(!X, !Y ) and the correspondence
f 7→ !f is functorial. This mapping f 7→ !f extends the action of ! on isomorphisms.
— The morphisms dX , dX , wX , wX , cX and cX are natural with respect to this extended
functor.
— There is a natural transformation pX : !X → !!X which turns (!X, dX , pX ) into a
comonad.
— There is a morphism µ0 : 1 → !1 and a natural transformation9 µ2X,Y : !X ⊗ !Y →
!(X ⊗ Y ) which satisfy the following commutations

µ0 ⊗ !X µ21,X
1 ⊗ !X !1 ⊗ !X !(1 ⊗ X)
!⊗hhi,∗i
∗
⊗hhi,∗i
∗
!X (6)

!X ⊗ µ0 µ2X,1
!X ⊗ 1 !X ⊗ !1 !(X ⊗ 1)
!⊗h∗,hii
∗
⊗h∗,hii
∗
!X (7)

hh∗,∗i,∗i
⊗h∗,h∗,∗ii !X ⊗ µ2Y,Z
(!X ⊗ !Y ) ⊗ !Z !X ⊗ (!Y ⊗ !Z) !X ⊗ !(Y ⊗ Z)
µ2X,Y ⊗ !Z µ2X,Y ⊗Z
hh∗,∗i,∗i
µ2X⊗Y,Z !⊗h∗,h∗,∗ii
!(X ⊗ Y ) ⊗ !Z !((X ⊗ Y ) ⊗ Z) !(X ⊗ (Y ⊗ Z))
(8)

σ!X,!Y
!X ⊗ !Y !Y ⊗ !X
µ2X,Y µ2Y,X
!σX,Y
!(X ⊗ Y ) !(Y ⊗ X)
(9)
— The following diagrams commute

µ2X,Y µ0
!X ⊗ !Y !(X ⊗ Y ) 1 !1
dX⊗Y d1
dX ⊗ dY 1
X ⊗Y 1

9 These morphisms are not required to be isos, whence the adjective “lax” for the monoidal structure.
Differential Linear Logic 31

µ2X,Y µ0
!X ⊗ !Y !(X ⊗ Y ) 1 !1
pX ⊗ pY pX⊗Y µ0 p1
µ2!X,!Y !µ2X,Y !µ0
!!X ⊗ !!Y !(!X ⊗ !Y ) !!(X ⊗ Y ) !1 !!1

When these conditions hold, one says that (µ0 , µ2 ) is a lax symmetric monoidal structure
on the comonad (!, d, p).

2.6.1. Monoidality and structural morphisms. This monoidal structure must also be com-
patible with the structural constructions.

µ2X,Y µ0
!X ⊗ !Y !(X ⊗ Y ) 1 !1
wX ⊗ wY wX⊗Y w1
hhi,hii 1
⊗hi
1⊗1 1 1

cX ⊗ cY
!X ⊗ !Y (!X ⊗ !Y ) ⊗ (!X ⊗ !Y ) hi
⊗hhi,hii
⊗
b ϕ,σ
1 1⊗1
µ2X,Y (!X ⊗ !Y ) ⊗ (!X ⊗ !Y ) µ 0
µ0 ⊗ µ0
c1
µ2X,Y ⊗ µ2X,Y !1 !1 ⊗ !1
cX⊗Y
!(X ⊗ Y ) !(X ⊗ Y ) ⊗ !(X ⊗ Y )

where ϕ = (1, 3, 2, 4) ∈ S4 and σ = hh∗, ∗i, h∗, ∗ii.

2.6.2. Monoidality and costructural morphisms. We need the following diagrams to com-
mute in order to validate the reduction rules of DiLL.

cX ⊗ !Y
(!X ⊗ !X) ⊗ !Y !X ⊗ !Y
wX ⊗ !Y
1 ⊗ !Y !X ⊗ !Y (!X ⊗ !X) ⊗ cY

1 ⊗ wY (!X ⊗ !X) ⊗ (!Y ⊗ !Y )

1⊗1 µ2X,Y b ϕ,σ
⊗ µ2X,Y

hhi,hii
⊗hi (!X ⊗ !Y ) ⊗ (!X ⊗ !Y )
wX⊗Y
1 !(X ⊗ Y ) µ2X,Y ⊗ µ2X,Y
cX⊗Y
!(X ⊗ Y ) ⊗ !(X ⊗ Y ) !(X ⊗ Y )

where ϕ = (1, 3, 2, 4) ∈ S4 and σ = hh∗, ∗i, h∗, ∗ii.

Thomas Ehrhard 32

dX ⊗ !Y
X ⊗ !Y !X ⊗ !Y
X ⊗ dY µ2X,Y
dX⊗Y
X ⊗Y !(X ⊗ Y )

2.6.3. Digging and structural morphisms. We assume that pX is a comonoid morphism

from (!X, wX , cX ) to (!!X, w!X , c!X ), in other words, the following diagrams commute.

pX pX
!X !!X !X !!X
w!X cX c!X
wX
pX ⊗ pX
1 !X ⊗ !X !!X ⊗ !!X

2.6.4. Digging and costructural morphisms. It is not required that pX be a monoid mor-
phism from (!X, wX , cX ) to (!!X, w!X , c!X ), but the following diagrams must commute.

wX cX pX
1 !X !X ⊗ !X !X !!X
µ0 pX pX ⊗ pX !cX
µ2!X,!X
!1 !!X !!X ⊗ !!X !(!X ⊗ !X)
!wX

In the same spirit, we need a last diagram to commute, which describes the interaction
between codereliction and digging.

dX pX
X !X !!X
⊗∗
hhi,∗i !cX
w X ⊗ dX pX ⊗ d!X
1⊗X !X ⊗ !X !!X ⊗ !!X

2.6.5. Preadditive structure and functorial exponential. Our last requirement justifies the
term “exponential” since it expresses that sums are turned into products by this functorial
operation.

!0 !(f + g)
!X !X !X !Y
cX cX
wX wX
!f ⊗ !g
1 !X ⊗ !X !Y ⊗ !Y

Remark : There is another option in the categorical axiomatization of models of Linear

Logic that we briefly describe as follows.
— One requires the linear category L to be cartesian, with a terminal object > and a
Differential Linear Logic 33

cartesian product usually denoted as X1 & X2 , projections πi ∈ L(X1 & X2 , Xi ) and

pairing hf1 , f2 i ∈ L(Y, X1 & X2 ) for fi ∈ L(Y, Xi ). This provides in particular L with
another symmetric monoidal structure.
— As above, one require the functor ! to be a comonad. But we equip it now with
a symmetric monoidal structure (m0 , m2 ) from the monoidal category (L, &) to the
monoidal category (L, ⊗). This means in particular that m0 ∈ L(1, !>) and m2X1 ,X2 ∈
L(!X1 ⊗ !X2 , !(X1 & X2 )) are isos. These isos are often called Seely isos in the litera-
ture, though Girard already stressed their significance in (Gir87), admittedly not in
the general categorical setting of monoidal comonads. An additional commutation is
required, which describes the interaction between m2 and p.

Using this structure, the comonad (! , d, p) can be equipped with a lax symmetric monoidal
structure (µ0 , µ2 ). Again, our main reference for these notions and constructions is (Mel09).
In this setting, the structural natural transformations wX and cX can be defined and it
is well known that the Kleisli category L! of the comonad ! is cartesian closed.
If we require the category L to be preadditive in the sense of Section 2.4, it is easy to
see that > is also an initial object and that & is also a coproduct. Using this fact, the
natural transformations wX and cX can also be defined.
To describe a model of DiLL in this setting, one has to require these Seely monoidality
isomorphisms to satisfy some commutations with the d natural transformation.
Here, we prefer a description which does not use cartesian products because it is closer
to the basic constructions of the syntax of proof-structures and makes the presentation
of the semantics conceptually simpler and more canonical, to our taste at least.

2.6.6. Generalized monoidality, contraction and digging. Just as the monoidal structure
of a monoidal category, the monoidal structure of ! can be parameterized by monoidal
→
−
trees. Let n ∈ N and let τ ∈ Tn . Given a family of objects X = (X1 , . . . , Xn ) of L, we
τ τ →
− τ →
−
− : ⊗ (! X ) → !⊗ ( X ) by induction on τ as follows:
define µ→
X

µhi = µ0
µ∗X = Id!X
hσ,τ i
2 σ τ
µ→ − =µ σ →
− → − − (µ→
→ − ⊗ µ→
−).
⊗ ( X ),⊗τ ( Y )
X,Y X Y

Given σ, τ ∈ Tn and ϕ ∈ Sn , one can prove that the following diagrams commute

→
−
⊗σ
τ (! X )
b ϕ,σ (!→
⊗
−
→
− →
− →
− X) →
−
⊗σ (! X ) ⊗τ (! X ) ⊗σ (! X ) ⊗σ (ϕ(!
b X ))
σ
µ→
−
τ
µ→
−
σ
µ→
− µσ →
−
X X X ϕ(
b X)
→
−
!⊗σ
τ (! X )
b ϕ,σ (→
!⊗
−
→
− →
− →
− X) →−
!⊗σ ( X ) !⊗τ ( X ) !⊗σ ( X ) !⊗σ (ϕ(
b X ))
Thomas Ehrhard 34

⊗σ (w→
−)
→
− X →
− →
−
⊗σ (! X ) ⊗σ 1 = ⊗σ 0 ⊗σ (! X ) ⊗σ (d→
−)
X
σ
µ→ σ
⊗hi0
σ
µ→ →
−
−
X
w⊗σ (→
−
−
X ⊗σ X
→
− X) →
− d⊗σ (→
−
!⊗σ ( X ) 1 = ⊗hi !⊗σ ( X ) X)

→
−
where 1 is the sequence (1, . . . , 1) (n elements) and σ0 = σ [hi/∗] ∈ T0 (the tree
obtained from σ by replacing each occurrence of ∗ by hi).
Before stating the next commutation, we define a generalized form of contraction
σ σ →
− hσ,σi →− → −
− : ⊗ !X → ⊗
c→ (! X , ! X ) as the following composition of morphisms:
X
σ
2
⊗σ c→
− b ϕ,σ ⊗hσ,σi
→
− X −
→ ⊗ →
− → − →
− → −
σ
⊗ !X ⊗ !X 0
σ2 σ2
⊗ (! X , ! X ) ⊗hσ,σi (! X , ! X )
−→
where X 0 = (X1 , X1 , X2 , X2 , . . . , Xn , Xn ), σ2 = σ [h∗, ∗i/∗] and ϕ ∈ S2n is defined by
ϕ(2i + 1) = i + 1 and ϕ(2i + 2) = i + n + 1 for i ∈ {0, . . . , n − 1}. With these notations,
one can prove that

σ
c→
−
→
− X →
− → − →
− →
−
σ
⊗ !X ⊗hσ,σi (! X , ! X ) = (⊗σ ! X ) ⊗ (⊗σ ! X )
σ σ σ
µ→
− − ⊗ µ→
µ→ −
X X X
c⊗σ (→
−
→
− σ→
− σ→
−
X)
!⊗σ X (!⊗ X ) ⊗ (!⊗ X )
σ σ →
− σ →
−
− : ⊗ ! X → !⊗ ! X as the following
We also define a generalized version of digging p→
X
composition of morphisms:
⊗σ p→
− µσ→
−
→
− X →
− !X →
−
⊗σ ! X ⊗σ !! X !⊗σ ! X

With this notation, one can prove that

σ
p→
−
→
− σ
X →
−
⊗ !X !⊗σ ! X
σ σ
µ→
− !µ→
−
X X
p⊗σ →
−
σ→
− σ→
−
X
!⊗ X !!⊗ X

We have phi = µ0 , p∗X = pX , and observe that the following generalizations of the
comonad laws hold. The two commutations involving digging and dereliction generalize
to:
→
−
⊗σ ! X
→
− σ
µ→
⊗σ ! X σ X
−
p→
−
X

→
− →
− →
−
⊗σ ! X d⊗σ !→
−
!⊗σ ! X !⊗σ d→
−
!⊗σ X
X X

→
−
The square diagram involving digging generalizes as follows. Let Y = (Y1 , . . . , Ym ) be
Differential Linear Logic 35

another list of objects and let τ ∈ Tm . One can prove that

σ
p→
−
σ→
− X →
−
⊗ !X !⊗σ ! X
σ σ
p→
− !p→
−
X X
p⊗σ !→
−
→
−σ
X →
−
σ
!⊗ ! X !!⊗ ! X

and then one can generalize this property as follows

hσ,τ i
p→
− →
−
→
− →
− X ,Y →
− →
−
(⊗ ! X ) ⊗ (⊗τ ! Y )
σ
!((⊗σ ! X ) ⊗ (⊗τ ! Y ))
σ τ →
− σ τ →
−
− ⊗ (⊗ ! Y )
p→ h∗,τ i − ⊗ (⊗ ! Y ))
!(p→
X X
p →
− → −
→
− →
− ⊗σ ! X , Y →
− →
−
!(⊗ ! X ) ⊗ (⊗τ ! Y )
σ
!(!(⊗ ! X ) ⊗ (⊗τ ! Y ))
σ
(10)

→
−
2.6.7. Generalized promotion and structural constructions. Let f : ⊗σ ! X → Y , we define
→
−
the generalized promotion f ! : ⊗σ ! X → !Y by f ! = !f p→
σ
− . Using the commutations of
X
Section 2.6.6, one can prove that this construction obeys the following commutations.

→
− f!
σ !Y
⊗ !X
σ wY
⊗ w→
−
X σ
⊗hi0
→
−
⊗σ 1 = ⊗σ0 ⊗hi = 1

with the same notations as before.

With these notations, we have

→
− f!
⊗σ ! X !Y
σ
c→
− cY
X

→
− → − f! ⊗ f!
⊗hσ,σi (! X , ! X ) !Y ⊗ !Y

The next two diagrams deal with the interaction between generalized promotion and
dereliction (resp. digging).
→
− f!
⊗σ ! X !Y
dY
f
Y
→
−
→
− →
− f ! ⊗ (⊗τ ! Y ) →
−
(⊗σ ! X ) ⊗ (⊗τ ! Y ) !Y ⊗ (⊗τ ! Y )
hσ,τ i h∗,τ i
p→
− →
− p →
−
X ,Y
! →
− τ
Y, Y
→
− →
− !(f ⊗ (⊗ ! Y )) →
−
!((⊗σ ! X ) ⊗ (⊗τ ! Y )) !(!Y ⊗ (⊗τ ! Y ))

The second diagram follows easily from (10) and allows one to prove the following prop-
Thomas Ehrhard 36
−
→ −
→ −→
erty. Let f : ⊗σ !X → Y and g : !Y ⊗ (⊗τ !Y ) → Z so that f ! : ⊗σ !X → !Y and
−
→
g ! : !Y ⊗ (⊗τ !Y ) → !Z; one has
→
− →
− −→ −
→
g ! (f ! ⊗ (⊗τ ! Y )) = (g (f ! ⊗ (⊗τ ! Y )))! : (⊗σ !X) ⊗ (⊗τ !Y ) → !Z

Remark : We actually need a more general version of this property, where f ! is not
necessarily in leftmost position in the ⊗ tree. It is also easy to obtain, but notations are
more heavy. We use the same kind of convention in the sequel but remember that the
corresponding properties are easy to generalize.

→
−
2.6.8. Generalized promotion and costructural constructions. Let f : !X ⊗ (⊗σ ! X ) → Y .
→
− →
−
Observe that f (wX ⊗ (⊗σ ! X )) ⊗σhhi,σi −→ : ⊗σ ! X → Y . The following equation holds:
!X
→
− σ →
−
f ! (wX ⊗ (⊗σ ! X )) ⊗σhhi,σi − σ
→ = (f (wX ⊗ (⊗ ! X )) ⊗hhi,σi −
!X
→)
!X
!

→
− →
−
Similarly, we have f (cX ⊗(⊗σ ! X )) : (!X ⊗ !X)⊗(⊗σ ! X ) → Y and the following equation
holds:
→
− →
−
f ! (cX ⊗ (⊗σ ! X )) = (f (cX ⊗ (⊗σ ! X )))!
This results from the commutations of Sections 2.6.2 and 2.6.4.

2.6.9. Generalized promotion and codereliction (also known as chain rule). Let f : !X ⊗
→
−
(⊗σ ! X ) → Y . We set
→
− →
−
f0 = f (wX ⊗ (⊗σ ! X )) ⊗σhhi,σi : ⊗σ ! X → Y
Then we have
→
−
→
− dX ⊗ (⊗σ ! X ) →
− f!
X ⊗ (⊗σ ! X ) !X ⊗ (⊗σ ! X ) !Y
σ
dX ⊗ c→
−
X
→
− →
−
!X ⊗ ((⊗σ ! X ) ⊗ (⊗σ ! X )) cY
h∗,h∗,∗ii
⊗hh∗,∗i,∗i
→
− →
− f ⊗ f0! dY ⊗ !Y
(!X ⊗ (⊗σ ! X )) ⊗ (⊗σ ! X ) Y ⊗ !Y !Y ⊗ !Y

This results from the commutations of Sections 2.6.2 and 2.6.4.

2.6.10. Interpreting DiLL derivations. For the sake of readability, we assume here that
the De Morgan isomorphisms (see 2.3.4) are identities, so that [A⊥ ] = [A]⊥ for each
formula A. The general definition of the semantics can be obtained by inserting De
Morgan isomorphisms at the correct places in the forthcoming expressions.
→
−
Let P be a net of arity n + 1 and let pi = (→
−ci ; ti , ti ) for i = 1, . . . , n. Consider the
following derivation π, where we denote as λ, ρ1 , . . . , ρn the derivations of the premises.
Φ ` P : ?A⊥ ⊥
1 , . . . , ?An , B Φ ` p1 : Γ1 , !A1 · · · Φn ` pn : Γn , !An
→
− →
− →
− →
− !(n)
Φ ` ( c1 , . . . , cn ; t1 , . . . , tn , P (t1 , . . . , tn )) : Γ1 , . . . , Γn , !B
Differential Linear Logic 37

~
By inductive hypothesis, we have [λ] ∈ L((![A ⊥ ⊥
1 ]) , . . . , (![An ]) , [B]) so that, picking an
element σ of Tn we have
[λ]hσ,∗i ∈ L(1, `σ ((![A1 ])⊥ , . . . , (![An ])⊥ )`[B])
= L(1, ⊗σ (![A1 ], . . . , ![An ]) ( [B])
and hence
(cur−1 ([λ]hσ,∗i ) ⊗σhhi,σi )! ∈ L(⊗σ (![A1 ], . . . , ![An ]), ![B]) .
~ i ], ![Ai ]). Let li be the length of Γi , and let us
For i = 1, . . . , n, we have [ρi ] ∈ L([Γ
choose τi ∈ Tli . We have [ρi ]hτi ,∗i ∈ L(1, `τi ([Γi ])`![Ai ]) and hence, setting
ri = cur−1 ([ρi ]hτi ,∗i ) ⊗τhhi,τ
i
ii
∈ L(⊗τi ([Γi ])⊥ , ![Ai ])

we have ⊗σ (→
−
r ) ∈ L(⊗θ ([∆]⊥ ), ⊗σ (![A1 ], . . . , ![An ])) where
∆ = Γ1 , . . . , Γn
θ = σ(τ1 , . . . , τn )
where σ(τ1 , . . . , τn ) (for σ ∈ Tn and τi ∈ Tni for i = 1, . . . , k) is the element of Tn1 +···+nk
defined inductively by
hi() = hi
∗(τ ) = τ
hσ, σ i(τ1 , . . . , τn ) = hσ(τ1 , . . . , τk ), σ 0 (τk+1 , . . . , τn )i
0

where σ ∈ Tk , σ 0 ∈ Tn−k
We have therefore
(cur−1 ([λ]hσ,∗i ) ⊗σhhi,σi )! ⊗σ (→
−
r ) ∈ L(⊗θ ([∆]⊥ ), [B])
We set
[π]θ = cur((cur−1 ([λ]hσ,∗i ) ⊗σhhi,σi )! ⊗σ (→
− hhi,θi
r ) ⊗θ ) ∈ L(1, `hθ,∗i ([∆, !B]))
~
and this gives us a definition of [π] ∈ L([∆, !B]) which does not depend on the choice of
θ.
Theorem 7. Let π and π 0 be derivations of Φ ` p : Γ. Then [π] = [π 0 ].
Again, we set [p] = [π] where π is a derivation of Φ ` p : Γ.
Theorem 8. Assume that Φ ` p : Γ, Φ ` p0 : Γ and that p ; p0 . Then [p] = [p0 ].
The proofs of these results are tedious inductions, using the commutations described
in paragraphs 2.6.7, 2.6.8 and 2.6.9.

2.7. The differential λ-calculus

Various λ-calculi have been proposed, as possible extensions of the ordinary λ-calculus
with constructions corresponding to the above differential and costructural rules of dif-
Thomas Ehrhard 38

ferential LL. We record here briefly our original syntax of (ER03), simplified by Vaux
in (Vau05)10 .
A simple term is either
— a variable x,
— or an ordinary application (M ) R where M is a simple terms and R is a term,
— or an abstraction λx M where x is a variable and M is a simple term,
— or a differential application DM · N where M and N are simple terms.
A term is a finite linear combination of simple terms, with coefficients in k. Substitution
of a term R for a variable x in a simple term M , denoted as M [R/x] is defined as usual,
whereas differential (or linear) substitution of a simple term for a variable in another
simple term, denoted as ∂M ∂x · N , is defined as follows:
(
∂y t if x = y
·t=
∂x 0 otherwise
∂λy M ∂M
· N = λy ·N
∂x  ∂x   
∂DM · N ∂M ∂N
·P =D · P · N + DM · ·P
∂x ∂x ∂x
    
∂ (M ) R ∂M ∂R
·N = · N R + DM · ·N R
∂x ∂x ∂x
All constructions are linear, except for ordinary application which is not linear in the
argument. This means that when we write e.g. (M1 + M2 ) R, what we actually intend is
(M1 ) R + (M2 ) R. Similarly, substitution M [R/x] is linear in M and not in R, whereas
differential substitution ∂M
∂x ·N is linear in both M and N . There are two reduction rules:

(λx M ) R β M [R/x]
 
∂M
D(λx M ) · N βd λx ·N
∂x
which have of course to be closed under arbitrary contexts. The resulting calculus can
be proved to be Church-Rosser using fairly standard techniques (Tait - Martin-Löf), to
have good normalization properties in the typed case etc, see (ER03; Vau05). To be more
precise, Church-Rosser holds only up to the least congruence on terms which identifies
D(DM · N1 ) · N2 and D(DM · N2 ) · N1 , a syntactic version of Schwarz Lemma: terms are
always considered up to this congruence called below symmetry of derivatives.

2.7.1. Resource calculus. Differential application can be iterated: given simple terms
M, N1 , . . . , Nn , we define Dn M · (N1 , . . . , Nn ) = D(· · · DM · N1 · · · ) · Nn ; the order on
the terms N1 , . . . , Nn does not matter, by symmetry of derivatives. The (general) re-
source calculus is another syntax for the differential λ-calculus, in which the combi-
nation (Dn M · (N1 , . . . , Nn )) R is considered as one single operation denoted e.g. as

10 Alternative syntaxes have been proposed, which are formally closer to Boudol’s calculus with multi-
plicities or with resources and are therefore often called resource λ-calculi
Differential Linear Logic 39

M [N1 , . . . , Nn , R∞ ] where the superscript ∞ is here to remind that R can be arbitrarily

duplicated during reduction, unlike the Ni ’s. This presentation of the calculus, studied
in particular by Tranquilli and Pagani, and also used for instance in (BCEM11), has
very good properties as well. It is formally close to Boudol’s λ-calculus with multiplici-
ties such as presented in (BCL99), with the difference that the operational semantics of
Boudol’s calculus is given as a rewriting strategy whereas in the differential version of
the resource λ-calculus, redexes can be reduced everywhere in terms. The price to pay is
that reduction becomes non-deterministic in the sense that it can produce formal sums
of terms.

2.7.2. The finite resource calculus. If, in the resource calculus above, one restricts one’s
attention to the terms where all applications are of the form
M [N1 , . . . , Nn , 0∞ ]
which corresponds to the differential term (D(· · · DM · N1 · · · ) · Nn ) 0, then one gets a
calculus which is stable under reduction and where all terms are strongly normalizing.
This calculus, called the finite resource calculus, can be presented as follows.
— Any variable x is a term.
— If x is a variable and s is a simple term then λx s is a simple term.
— If S is a finite multiset (also called bunch in the sequel) of simple terms then hsi S
is a simple term. Intuitively, this term stands for the application s[s1 , . . . , sn , 0∞ ] of
the resource calculus, where [s1 , . . . , sn ] = S.
A term is a (possibly infinite11 ) linear combination of finite terms. This syntax is extended
from simple terms to general terms by linearity. For instance, the term hxi [y + z, y + z]
stands for hxi [y, y] + 2 hxi [y, z] + hxi [z, z].
In the finite resource calculus, it is natural to perform several βd -reductions in one
step, and one gets
(P  
f ∈Sn s sf (1) /x1 , . . . , sf (n) /xn if degs x = n
hλx si S βd
0 otherwise
where d = degx s is the number of occurrences of x in s (which is a simple term), x1 , . . . , xd
are the occurrences of x in s and the multiset S is [s1 , . . . , sn ].
Again, this calculus enjoys confluence, and also strong normalization (even in the
untyped case). It can be used for hereditarily Taylor expanding λ-terms as explained
in (ER08; ER06; Ehr10).
Taylor expansion consists in hereditarily replacing, in a differential λ-term, any ordi-
nary application (M ) N by the infinite sum
∞
X 1
(Dn M · (N, . . . , N )) 0 .
n=0
n!

11 When considering infinite linear combinations, one has to deal with the possibility of unbounded
coefficients appearing during the reduction. One option is to accept infinite coefficients, but it is also
possible to prevent this phenomenon to occur by topological means as explained in (Ehr10).
Thomas Ehrhard 40

More precisely, it is a transformation M 7→ M ∗ from resource terms to finite resource

terms which is defined as follows:

x∗ = x
∗
(λx M ) = λx (M ∗ )
∞
∞ ∗
X 1
M [N1 , . . . , Nn , R ] = hM ∗ i [N1 ∗ , . . . , Nn ∗ , R∗ , . . . , R∗ ]
p=0
p!

so that the Taylor expansion of a resource term is a generally infinite linear combination
of finite resource terms. In the definition above, we use the extension by linearity of
the syntax of finite resource terms to arbitrary (possibly infinite) linear combinations.
The coefficients belong to the considered semi-ring k where division by positive natural
numbers must be possible.
In (ER08; ER06) we studied the behavior of this expansion with respect to differential
β-reduction in the case where the expanded terms come from the λ-calculus (that is, do
not contain differential applications; this is the uniform case), and we exhibited tight
connections between this operation and Krivine’s machine, an implementation of linear
head reduction.
There is a simple translation from resource terms (or differential terms) to DiLL proof-
nets. When restricted to the finite resource calculus, this translation ranges in DiLL0 .
This translation extends Girard’s Translation from the λ-calculus to LL proof-nets.

3. More on exponential structures

We address here two aspects of exponential structures: we study a simple condition
expressing that a map whose derivative is uniformly equal to 0 must be constant, and
we propose an axiomatization of antiderivatives in this categorical setting.
So we assume to be given a preadditive *-autonomous category L equipped with an
exponential structure in the sense of Section 2.5, and we use the same notations as in
this section.
For the sake of notational simplicity, we do as if ⊗ were strictly associative and 1 were
strictly neutral for ⊗. In other words, we do not mention the isos λ, ρ and α in our
computations, just as if they were identities (see Section 2.2). Given an object X and
n ∈ N of L, we use X ⊗n for the nth tensor power of X: X ⊗0 = 1 and X ⊗(n+1) = X ⊗n ⊗X.
Given an object X of L, we define a morphism ∂ X ∈ L(!X ⊗ X, !X) as the following
composition of morphisms

!X ⊗ dX cX
!X ⊗ X !X ⊗ !X !X
n
More generally, we define ∂ X ∈ L(!X ⊗ X ⊗n , !X):
0
∂ X = Id!X
n+1 n
∂X = ∂ X (∂ X ⊗ X)
Differential Linear Logic 41

Last we set
n n
dX = ∂ X (wX ⊗ X ⊗n ) ∈ L(X ⊗n , !X) .
n
We define dually ∂X ∈ L(!X, !X ⊗X) as ∂X = (!X ⊗dX ) cX and ∂X ∈ L(!X, !X ⊗X ⊗n )
0 n+1 n
by ∂X = Id!X and ∂X = (∂X ⊗ X) ∂X . And we set

dnX = (wX ⊗ X ⊗n ) ∂X
n
∈ L(!X, X ⊗n ) .
1
Observe that we have dX = dX and d1X = dX .
Consider now some f ∈ L(!X, Y ), to be intuitively seen as a non linear map from X to
Y (if ! were assumed to be a comonad as in Section 2.6, then f would be a morphism in
the Kleisli category L! ). Such a morphism will sometimes be called a “regular function”
in the sequel, but keep in mind that it is not even a function in general. For instance,
if X and Y are vector spaces (with a topological structure, in the infinite dimensional
case), such a regular function could typically be a smooth or an analytic function.
With these notations, f wX ∈ L(1, Y ) should be understood as the point of Y obtained
by applying the regular function f to 0. Similarly, f cX ∈ L(!X ⊗ !X, Y ) should be
understood as the regular function g : X × X → Y defined by g(x1 , x2 ) = f (x1 + x2 ).
Dually, given y ∈ L(1, Y ) considered as a point of Y , then y wX ∈ L(!X, Y ) should
be understood as the constant regular function which takes y as unique value. If g ∈
(!X ⊗ !X, Y ), to be considered as a regular function with two parameters X × X → Y ,
then f = g cX ∈ L(!X, Y ) should be understood as the regular function X → Y given by
f (x) = g(x, x). Given g ∈ L(X, Y ), to be considered as a linear function from X to Y ,
f = g dX ∈ L(!X, Y ) is g, considered now as a regular function from X to Y .
The basic idea of DiLL is that f 0 = f ∂ X ∈ L(!X ⊗ X, Y ) (that is f 0 ∈ L(!X, X ( Y ),
up to linear curryfication) represents the derivative of f .
Remember indeed that if f : X → Y is a smooth function from a vector space X to
a vector space Y , the derivative of f is a function f 0 from X to the space X ( Y of
(continuous) linear functions from X to Y : f 0 maps x ∈ X to a linear map f 0 (x) : X → Y ,
the differential (or Jacobian) of f at point x, which maps u ∈ X to f 0 (x) · u.
In particular, f dX = f ∂ X (wX ⊗ X) ∈ L(X, Y ) corresponds to f 0 (0), the differential
of f at 0.
n
More generally, f ∂ X ∈ L(!X ⊗ X ⊗n , Y ) represents the nth derivative of f , which is a
regular function from X to the space of n-linear functions from X n to Y (these n linear
functions are actually symmetric, a property called “Schwarz Lemma” and is axiomatized
here by the commutativity of the algebra structure of !X).
The axioms of an exponential structure express that this categorical definition of differ-
entiation satisfies the usual laws of differentiation. Let us give a simple example. Consider
f ∈ L(!X ⊗ !X, Y ), to be seen as a regular function f (x1 , x2 ) depending on two parame-
ters x1 and x2 in X. Remember that g = f cX ∈ L(!X, Y ) represents the map depending
on one parameter x in X given by g(x) = f (x, x).
Using the axioms of exponential structures, one checks easily that

cX ∂ X = ((!X ⊗ ∂ X ) + (∂ X ⊗ !X) (!X ⊗ σ)) (cX ⊗ X)

Thomas Ehrhard 42

which, by composing with f and using standard algebraic notations, gives

g 0 (x) · u = f10 (x, x) · u + f20 (x, x) · u

where we use fi0 for the ith partial derivative of f and “·” for the linear application of
the differential. This is Leibniz law.
Similarly, the easily proven equation wX ∂ X = 0 expresses that the derivative of a
constant map is equal to 0.
It is a nice exercise to interpret similarly the dual equations

∂X cX = (cX ⊗ X) ((!X ⊗ ∂X ) + (!X ⊗ σ) (∂X ⊗ !X))

∂X w X = 0

Consider g ∈ L(!X ⊗ X, Y ), to be considered as a regular function X × X → Y which is

linear in its second parameter. Then f = g ∂X ∈ L(!X, Y ) is the regular function X → Y
given by g(x) = f (x, x). The second equation corresponds to the fact that g(0, 0) = 0,
and the first one, to the fact that g(x1 + x2 , x1 + x2 ) = g(x1 + x2 , x1 ) + g(x1 + x2 , x2 ).

3.1. Taylor exponential structures.

Let L be an exponential structure and let f ∈ L(!X, Y ), to be considered as a “regular
function” from X to Y . The condition f ∂ X = 0 means intuitively that the derivative of
f is uniformly equal to 0, and hence, according to standard intuitions on differentiation,
f should be a constant map. In other words we should have f = f wX wX .
This property can be stated in a more general way as follows: let f1 , f2 : !X → Y , then

f1 ∂ X = f2 ∂ X ⇒ f1 + (f2 wX wX ) = f2 + (f1 wX wX )

and does not seem to be derivable from the other axioms of exponential structures. The
converse implication is easy to prove.
Remark : There is a dual condition which reads as follows: if f1 , f2 : Y → !X, then

∂X f1 = ∂X f2 ⇒ f1 + (wX wX f2 ) = f2 + (wX wX f1 )

The intuition is that, given f : Y → !X, to be considered as a generalized point of !X, if

∂X f = 0, then the range of f is included in the subspace of !X generated by the unit of
the bialgebra !X. In other words, this condition means that the kernel of ∂X is generated
by this unit.
We say that the exponential structure L is Taylor if it satisfies the first condition.
For n ∈ N, let TnX ∈ C(!X, !X) be defined by
n
X 1 i i
TnX = dX dX .
i=0
i!

Lemma 9. For any n > 0, we have

TnX ∂ X = ∂ X (Tn−1
X ⊗ IdX ) .
Differential Linear Logic 43

Proof. This results from

n n−1
dX dnX ∂ X = n∂ X ((dX dn−1
X ) ⊗ IdX )

which comes from the basic equations of exponential structures. 2

Remember that a commutative monoid M is cancellative if, in M , one has u + v =

u0 + v ⇒ u = u0 .

Proposition 10. Assume that L is Taylor and that each homset L(X, Y ) is a cancellative
n+1 n+1
monoid. Let n ∈ N and let f1 , f2 : !X → Y . If f1 ∂ X = f2 ∂ X then f1 + (f2 TnX ) =
f2 + (f1 TnX ).
n+1
In particular, if f ∂ X = 0 (that is, the (n + 1)-th derivative of f is uniformly equal to
0), then f = f TnX , meaning that f is equal to its Taylor expansion of rank n.

Proof. By induction on n. For n = 0, this is simply the hypothesis that L is Taylor.

n+2 n+2
Assume now that f1 ∂ X = f2 ∂ X and let us prove that f1 +(f2 Tn+1 n+1
X ) = f2 +(f1 TX ).
n+1 n+1
We have f1 ∂ X (∂ X ⊗ IdX ) = f2 ∂ X (∂ X ⊗ IdX ). By monoidal closeness, we have
n+1 n+1
cur(f1 ∂ X ) ∂ X = cur(f2 ∂ X ) ∂ X and hence, by inductive hypothesis, we have

cur(f1 ∂ X ) + (cur(f2 ∂ X ) TnX ) = cur(f2 ∂ X ) + (cur(f1 ∂ X ) TnX )

that is

cur(f1 ∂ X ) + (cur(f2 ∂ X (TnX ⊗ IdX ))) = cur(f2 ∂ X ) + (cur(f1 ∂ X (TnX ⊗ IdX )))

and hence

(f1 ∂ X ) + (f2 ∂ X (TnX ⊗ IdX )) = (f2 ∂ X ) + (f1 ∂ X (TnX ⊗ IdX ))

so applying Lemma 9, we get

(f1 + f2 Tn+1 n+1

X ) ∂ X = (f2 + f1 TX ) ∂ X .

Applying the hypothesis that L is Taylor, we get

f1 + f2 Tn+1
X + (f2 + f1 Tn+1 n+1
X ) wX wX = f2 + f1 TX + (f1 + f2 Tn+1
X ) w X wX

and since Tn+1

X wX wX = wX wX we get f1 + f2 Tn+1X + (f1 + f2 ) wX wX = f2 + f1 Tn+1
X +
(f2 + f1 ) wX wX and so, applying the cancellativeness hypothesis, we get finally

f1 + f2 Tn+1
X = f2 + f1 Tn+1
X

as required. 2

3.1.1. The category of polynomials. We say that f ∈ L(!X, Y ) is polynomial if there exists
n+1
n ∈ N such that f ∂ X = 0, and we call degree of f the least such n. The morphism
dX ∈ C(!X, X) is polynomial of degree 1.
Let f ∈ L(!X, Y ) and g ∈ L(!Y, Z) be polynomial of degree m and n respectively. We
Thomas Ehrhard 44

define the composition g ◦ f ∈ L(!X, Z) as follows

n
X 1 i
g◦f = g dY f ⊗i ciX .
i=0
i!
i×
z }| {
⊗i
where f = f ⊗ · · · ⊗ f .
i
Since dX dX = 0 for i 6= 1, we get dX ◦ g = g. Next observe that g ◦ dX = g TnX = g
by Proposition 10. One can prove that g ◦ f is polynomial of degree ≤ mn, that is
mn
(g ◦ f ) ∂ X = 0 by a straightforward (though boring) categorical computation using
the basic axioms of exponential structures. Using the same axioms, one shows that this
notion of composition is associative, so that we have defined a category of polynomial
morphisms.

3.1.2. Weak functoriality and the category of polynomials. We do not require this oper-
ation X 7→ !X to be functorial, but some weak form of functoriality can be derived from
the above categorical axioms. Let f ∈ L(X, Y ). By induction on n, we define a family of
morphisms f n : !X → !X as follows: f 0 = wX wX and

f n+1 = ∂ X (f n ⊗ f ) ∂X .

Proposition 11. Let f ∈ L(X, Y ) and g ∈ L(Y, Z) and let n, p ∈ N. Then

(
n!(g f )n if n = p
gp f n =
0 otherwise.

Proof. Simple calculation using the diagram commutations which define an exponential
structure. 2
Pn 1 q
So for each n ∈ N we can define !n f = q=0 q! f : !X → !X, and we have !n g !n f =
!n (g f ). So f 7→ !n f is a quasifunctor, but not a functor as it does not map IdX to Id!X ,
but to an idempotent morphism ρnX : !X → !X.
In some concrete models, this sequence (!n f )n∈N can be said to be convergent, in a
sense which depends of course on the model. The limit is then denoted as !f and the
operation defined in that way turns out often to be a true functor, defining a functorial
exponential in the sense of Section 2.6.

3.2. Computing antiderivatives.

We say that an exponential structure L has antiderivatives if the morphism

JX = IdX +(∂ X ∂X ) : !X → !X

is an isomorphism. We explain why.

We assume to be given an exponential structure L which has antiderivatives in that
sense and we set IX = JX −1 .
Differential Linear Logic 45

In the sequel, we use the following notation

ψX = (∂ X ⊗ IdX ) σ23 (∂X ⊗ IdX ) : !X ⊗ X → !X ⊗ X

where σ23 = !X ⊗ σ is an automorphism on !X ⊗ X ⊗ X, because this morphism ψX will

show up quite often. Observe in particular that

∂X ∂ X = Id!X⊗X +ψX .

Lemma 12. The following commutation holds

(IX ⊗ IdX ) ψX = ψX (IX ⊗ IdX ) (11)

−1
Proof. Since IX = (Id!X +(∂ X ∂X )) , we have IX ⊗IdX = ϕ−1 where ϕ = Id!X⊗X +((∂ X ∂X )⊗
IdX ) by functoriality of ⊗. To prove (11), it suffices therefore to prove that ϕ commutes
with ψX . For this, it suffices to show that

((∂ X ∂X ) ⊗ IdX ) ψX = ψX ((∂ X ∂X ) ⊗ IdX ) .

We have
((∂ X ∂X ) ⊗ IdX ) ψX = ((∂ X ∂X ∂ X ) ⊗ IdX ) σ23 (∂X ⊗ IdX )
but remember that ∂X ∂ X = Id!X⊗X +ψX , and hence

((∂ X ∂X ) ⊗ IdX ) ψX = ψX + ((∂ X ψX ) ⊗ IdX ) σ23 (∂X ⊗ IdX )

But ∂ X (∂ X ⊗ IdX ) σ23 = ∂ X (∂ X ⊗ IdX ) by commutativity of the bialgebra !X and by

definition of ∂ X . Therefore ∂ X ψX = ∂ X ((∂ X ∂X ) ⊗ IdX ). So we can write

((∂ X ∂X ) ⊗ IdX ) ψX
= ψX + (∂ X ⊗ IdX ) ((∂ X ∂X ) ⊗ IdX ⊗ IdX ) σ23 (∂X ⊗ IdX )
= ψX + (∂ X ⊗ IdX ) ((∂ X ∂X ) ⊗ σ) (∂X ⊗ IdX )

A similar, and completely symmetric computation, using this time the cocommutativity
of the bialgebra !X, leads to

ψX ((∂ X ∂X ) ⊗ IdX ) = ψX + (∂ X ⊗ IdX ) ((∂ X ∂X ) ⊗ σ) (∂X ⊗ IdX )

and we are done. 2

We can now prove a completely categorical version of the following proposition which
is the key step in the usual proof of Poincaré’s Lemma.

Proposition 13. Let f : !X ⊗ X → Y be such that the differential f (∂X ⊗ IdX ) :

!X ⊗ X ⊗ X → Y satisfies

f (∂X ⊗ IdX ) σ23 = f (∂X ⊗ IdX ) .

Then there exists g : !X → Y such that g ∂ X = f ; in other words, g is an “antiderivative”

of f .
Thomas Ehrhard 46

Proof. One sets

g = f (IX ⊗ IdX ) ∂X . (12)

Then we have

g ∂ X = f (IX ⊗ IdX ) ∂X ∂ X
= f (IX ⊗ IdX ) (Id!X⊗X +ψX )
= f (IX ⊗ IdX ) + f (IX ⊗ IdX ) ψX
= f (IX ⊗ IdX ) + f ψX (IX ⊗ IdX )

by Lemma 12. But

f ψX = f (∂ X ⊗ IdX ) σ23 (∂X ⊗ IdX )

= f (∂ X ⊗ IdX ) (∂X ⊗ IdX ) by our hypothesis on f
= f ((∂ X ∂X ) ⊗ IdX ) .

So we get

g ∂ X = f (IX ⊗ IdX ) + f ((∂ X ∂X IX ) ⊗ IdX )

= f (((Id!X +(∂ X ∂X )) IX ) ⊗ IdX )
=f

since IX is the inverse of Id!X +(∂ X ∂X ). 2

3.2.1. Comments. Let us give some intuition about our axiom that JX has an inverse.
Given f : !X → Y seen as a “regular function” from X to Y , we explain why the
morphism f IX : !X → Y should be understood as representing the regular function g
defined by
Z 1
g(x) = f (tx) dt
0

assuming of course that this integral makes sense. With this interpretation, g ∂ X : !X ⊗
X → Y represents the differential Dg of g, a regular function X × X → Y which maps
(x, y) to Dg(x) · y and is linear in y. Then, applying the ordinary rules of differential
calculus, and the fact that differentiation commutes with integration, we get
Z 1
Dg(x) · y = t(Df (tx) · y)dt .
0
Differential Linear Logic 47

The morphism h = g ∂ X ∂X : !X → Y corresponds to the regular function from X to Y

such that
Z 1
h(x) = (Df (tx) · (tx))dt
0
Z 1
= t(Df (tx) · x)dt
0
Z 1
df (tx)
= t dt
0 dt
Z 1
= f (x) − f (tx) dt ,
0

integrating by parts. In other words, we have seen that

f IX ∂ X ∂X = f − (f IX )
that is
f IX (Id!X +(∂ X ∂X )) = f
this is why our first axiom on IX is that IX (Id!X +(∂ X ∂X )) = Id!X . To explain why we
also require (Id!X +(∂ X ∂X )) IX = Id!X , observe that
l = f ∂ X ∂X : !X → Y
corresponds to the regular function defined by l(x) = Df (x)·x and hence l IX corresponds
to the regular function m : X → Y given by
Z 1
m(x) = (Df (tx) · (tx))dt = h(x)
0

by linearity of the differential. So we have

f ∂ X ∂X IX = f − (f IX )
that is
f (Id!X +(∂ X ∂X )) IX = f .
A remarkable and quite natural feature of this axiomatization of antiderivatives is the
fact that it is actually a mere property of the exponential structure, and not an additional
structure: it must be such that Id!X +(∂ X ∂X ) has an inverse.
Remark : The definition (12) of the antiderivative g of f in the proof above reads as
follows, if we use this intuitive interpretation of IX :
Z 1
g(x) = f (tx) · x dt (13)
0

which is exactly its definition, in the standard proof of Poincaré’s Lemma. The proof of
Proposition 13 is a rephrasing of the standard proof, which uses an integration by parts.

3.2.2. The fundamental theorem of calculus. This is the statement according to which
one can use antiderivatives for computing integrals: if f, g : R → R are such that g 0 = f ,
Thomas Ehrhard 48
Rb
then a f (t) dt = g(b) − g(a). In the present setting, it boils down to a simple categorical
equation.
Proposition 14. Let L be an exponential structure which has antiderivatives and is
Taylor. Then
∂ X (IX ⊗ IdX ) ∂X + wX wX = Id!X .

Proof. Let f1 = ∂ X (IX ⊗ IdX ) ∂X : !X → !X and let f2 = Id!X . We have

f1 ∂ X = ∂ X (IX ⊗ IdX ) ∂X ∂ X
= ∂ X (IX ⊗ IdX ) + ∂ X (IX ⊗ IdX ) (∂ X ⊗ IdX ) σ23 (∂X ⊗ IdX )
= ∂ X (IX ⊗ IdX ) + ∂ X (∂ X ⊗ IdX ) σ23 (∂X ⊗ IdX ) (IX ⊗ IdX )
by Lemma 12
= ∂ X (IX ⊗ IdX ) + ∂ X (∂ X ⊗ IdX ) (∂X ⊗ IdX ) (IX ⊗ IdX )
by commutativity of cocontraction
= ∂ X ((Id!X +(∂ X ∂X )) ⊗ IdX ) (IX ⊗ IdX )
−1
= ∂X since IX = (Id!X +(∂ X ∂X ))
= f2 ∂ X .
Since L is Taylor and since f1 wX wX = 0, we have therefore f1 + (f2 wX wX ) = f2 +
(f1 wX wX ), which is exactly the announced equation. 2

Remark : We give now an intuitive interpretation of this property. Let f : !X → Y ,

considered as a regular function from X to Y . Then f ∂ X (IX ⊗ IdX ) ∂X : !X → Y
represents the regular function g : X → Y given by
Z 1 Z 1
df (tx)
g(x) = Df (tx) · x dt = dt
0 0 dt
so that g(x) = f (x) − f (0) by the Fundamental Theorem of Calculus. In other words
g(x) + f (0) = f (x), that is f (∂ X (IX ⊗ IdX ) ∂X + wX wX ) = f .

3.3. Computing antiderivatives in the resource calculus

We can consider finite linear combinations of finite resource terms (see 2.7.2) as polyno-
mials, and with this respect, it seems natural to formally compute the antiderivative of
such a term, as one does for polynomials. This is the purpose of this short section. We
use ∆ for the set of simple resource terms and khSi for the free k-module generated by
the set S.
As with ordinary polynomials, we define first the antiderivative of a monomial, that
is, of a simple resource term. Remember that, for ordinary one variable polynomials, the
1
antiderivative of X d is d+1 X d+1 ; the definition is completely similar here. Let t ∈ ∆ be
a simple resource term and let x be a variable. We set
1
Ix (t) = t.
degx t + 1
Differential Linear Logic 49

We extend this operation by linearity to all elements u ∈ kh∆i, that is we set Ix (u) =
P
t∈∆ ut Ix (t).
(d)
For d ∈ N, let ∆x = {t ∈ ∆ | degx t = d} be the set of all simple resource terms of
(d)
degree d in x. The elements of kh∆x i are said to be homogeneous of degree d in x.
With these notations, we can write
∞
X 1 X
Ix (u) = ut t
d+1
d=0 (d)
t∈∆x
R1
Intuitively, Ix (u) stands for the integral 0 u(τ x) dτ which is the basic ingredient in the
proof above of Poincaré’s Lemma.
(1)
Let u ∈ kh∆i which is linear in the variable h, in other words u ∈ kh∆h i. Let h0 be
a variable which does not occur free in u, we assume that
∂u 0 ∂u [h0 /h]
·h = ·h
∂x ∂x
which is our symmetry hypothesis on u. In other words, for any d ∈ N, we have
X ∂t X ∂t [h0 /h]
ut · h0 = ut · h. (14)
(d)
∂x (d)
∂x
t∈∆x t∈∆x

Mimicking (13), we set

v = Ix (u) [x/h]
and we prove that
∂v
· h = u. (15)
∂x
Choose h0 as above, we prove that ∂x
∂v
· h0 = u [h0 /h] which of course implies (15). We
have, keeping in mind that h has exactly one free occurrence in each t ∈ ∆ such that
ut 6= 0
∞
∂v 0 X 1 X ∂t [x/h] 0
·h = ut ·h
∂x d+1 ∂x
d=0 (d)
t∈∆x
∞  
X 1 X ∂t
= ut · h0 [x/h] + t [h0 /h]
d+1 ∂x
d=0 (d)
t∈∆x
∞
∂t [h0 /h]
  
X 1 X
= ut · h [x/h] + t [h0 /h] by (14)
d+1 ∂x
d=0 (d)
t∈∆x
∞
∂t [h0 /h]
 
X 1 X
0
= ut · x + t [h /h]
d+1 ∂x
d=0 (d)
t∈∆x
∞
X 1 X ∂s
= ut (d t [h0 /h] + t [h0 /h]) since · x = d s for all s ∈ ∆(d)
x
d+1 ∂x
d=0 (d)
t∈∆x

= u [h0 /h]
Thomas Ehrhard 50

and we are done.

4. Concrete models
We want now to give concrete examples of categorical models of DiLL.

4.1. Products, coproducts and the Seely isomorphisms

In Section 2.6, we introduced the functorial version of the exponential without mentioning
the Seely isomorphisms: as explained in Paragraph 2.6.5, this choice is very natural
when presenting the denotational interpretation of proof-nets. But when describing the
structure of concrete models, as we want to do now, it is more natural to assume that
the linear category L is cartesian and that the ! comonad is equipped with a strong
symmetric monoidal structure from & to ⊗.
So we assume to be given a preadditive *-autonomous category L equipped with an ex-
ponential structure (Section 2.5) where ! is a monoidal comonad satisfying the conditions
of Section 2.6.
We assume moreover that L is cartesian, with terminal object >, cartesian product &,
projections πi ∈ L(X1 & X2 , Xi ). Because L is preadditive, this implies that > is also an
initial object, and that X1 & X2 (together with suitably defined injections) is also the
coproduct of X1 and X2 . In other words, L is an additive monoidal category.
We assume to be given an isomorphism m0 ∈ L(1, !>) and a natural isomorphism
2
mX1 ,X2 ∈ L(!X1 ⊗!X2 , !(X1 & X2 )) which endow the functor ! with a monoidal structure.
This means that diagrams similar to (6), (7), (8) and (9) hold.
We also require the following diagram to commute
m2X,Y
!X ⊗ !Y !(X & Y )
pX&Y

pX ⊗ pY !!(X & Y )
!h!π1 , !π2 i
m2!X,!Y
!!X ⊗ !!Y !(!X & !Y )

Of course, there is a connection between these two monoidal structures on ! . The

morphism µ0 is the following composition of morphisms:
−1
m0 p> !((m01 ) )
1 !> !!> !1

and µ2X,Y is
−1
m2X,Y pX&Y !(m2X,Y ) !(dX ⊗ dY )
!X ⊗ !Y !(X & Y ) !!(X & Y ) !(!X ⊗ !Y ) !(X ⊗ Y )

The bi-algebraic structure of !X presented in Section 2.5 is also related to this Seely
monoidal structure.
Differential Linear Logic 51

For the coalgebraic part, let ∆X ∈ L(X, X & X) be the diagonal morphism associated
with the cartesian product of X with itself. Then we have
−1
cX = m2X,X !∆X : !X → !X ⊗ !X

Similarly we set
wX = m01 !τX
where τX : X → > is the unique morphism to the terminal object. The algebraic part
satisfies similar conditions, using the codiagonal aX : X & X → X and the morphism
0
τX : > → X.

4.2. Relational semantics

We introduce now the simplest ∗-autonomous category equipped with an exponential
structure: the category of sets and relations. For this model, we assume that k = {0, 1}
with addition defined by 1 + 1 = 1.
Let Rel be the category whose objects are sets and where Rel(X, Y ) = P(X × Y ),
identities being the diagonal relations and composition being defined as follows: if R ∈
Rel(X, Y ) and S ∈ Rel(Y, Z) then

S R = {(a, c) ∈ X × Z | ∃b ∈ Y (a, b) ∈ R and (b, c) ∈ S} .

Let x ⊆ X, we set Rx = {b ∈ Y | ∃a ∈ x (a, b) ∈ R} ⊆ Y which is the direct image of

x by R. We also define R⊥ = {(b, a) ∈ Y × X | (a, b) ∈ Y } which is the transpose of R.
Given x ⊆ X and y 0 ⊆ Y , we have

(Rx) ∩ y 0 = pr2 (R ∩ (x × y 0 )) and (R⊥ y 0 ) ∩ x = pr1 (R ∩ (x × y 0 )) (16)

where pr1 and pr2 are the two projections of the cartesian product in the category Set
of sets and functions (the ordinary cartesian product “×”).
Observe that an isomorphism in Rel is a relation which is a bijection.
The symmetric monoidal structure is given by the tensor product X ⊗ Y = X × Y and
the unit 1 an arbitrary singleton. The neutrality, associativity and symmetry isomor-
phisms are defined as the obvious corresponding bijections (for instance, the symmetry
isomorphism σX,Y ∈ Rel(X ⊗ Y, Y ⊗ X) is given by σ(a, b) = (b, a)). This symmetric
monoidal category is closed, with linear function space given by X ( Y = X × Y ,
the natural bijection between Rel(Z ⊗ X, Y ) and Rel(Z, X ( Y ) being induced by the
cartesian product associativity isomorphism. Last, one takes for ⊥ an arbitrary singleton,
and this turns Rel into a ∗-autonomous category. One denotes as ? the unique element
of 1 and ⊥.
This category is additive, with cartesian product X1 & X2 of X1 and X2 defined as
{1}×X1 ∪{2}×X2 with projections πi = {((i, a), a) | a ∈ Xi } (for i = 1, 2), and terminal
object > = ∅. Then the commutative monoid structure on homsets Rel(X, Y ) is defined
by 0 = ∅ and f + g = f ∪ g and the action of k on morphisms is defined by 0 f = 0 and
1 f = f (there are no other possibilities).
Rel is also a Seely category (see Section 4.1), for a comonad ! defined as follows:
Thomas Ehrhard 52

— !X is the set of all finite multisets of elements of X;

— if R ∈ Rel(X, Y ), then we set !R = {([a1 , . . . , an ], [b1 , . . . , bn ]) | n ∈ N and ∀i (ai , bi ) ∈
R};
— dX ∈ Rel(!X, X) is dX = {([a], a) | a ∈ X};
— pX = {(m1 + · · · + mn , [m1 , . . . , mn ]) | n ∈ N and m1 , . . . , mn ∈ !X}.
The monoidality isomorphism m2X,Y ∈ Rel(!X ⊗ !Y , !(X & Y )) is the bijection which
maps ([a1 , . . . , al ], [b1 , . . . , br ]) to [(1, a1 ), . . . , (1, al ), (2, b1 ), . . . , (2, br )].
Last, we also provide a codereliction natural transformation dX ∈ Rel(X, !X) which
is simply given by dX = {(a, [a]) | a ∈ X}.
With these definitions, it is easy to see that cX = {(l + r, (l, r)) | l, r ∈ !X}, wX =
{([], ?)}, cX = {((l, r), l + r) | l, r ∈ !X} and wX = {(?, [])}. The required diagrams are
easily seen to commute.

4.2.1. Antiderivatives. This exponential structure is bicommutative and can easily seen
to be Taylor in the sense of Section 3.1. Moreover, it has antiderivatives in the sense of
Section 3.2, simply because the morphism JX = IdX +(∂ X ∂X ) : !X → !X coincides here
with the identity. Indeed ∂X = {(l+[a], (l, a)) | l ∈ !X and a ∈ X}, ∂ X = {((l, a), l+[a]) |
l ∈ !X and a ∈ X} and therefore ∂ X ∂X = {(l, l) | l ∈ !X and #l > 0}.
Concretely, saying that a morphism f ∈ Rel(!X ⊗ X, Y ) satisfies the symmetry con-
dition of Proposition 13 simply means that, given m ∈ Mfin (X), a, a0 ∈ X and b ∈ Y ,
one has ((m + [a], a0 ), b) ∈ f ⇔ ((m + [a0 ], a), b) ∈ f . In that case, the antiderivative
g ∈ Rel(!X, Y ) given by that proposition is simply
g = {(m + [a], b) | ((m, a), b) ∈ f } .

4.3. Finiteness spaces

This model can be seen as an enrichment of the model of sets and relations of Section 4.2.
It can also be described as a category of topological vector spaces and linear continuous
maps. From now on, k denotes an arbitrary field which is always endowed with the
discrete topology.

4.4. Linearly topologized vector spaces (ltvs)

Let E be a k-vector space. A linear topology on E is a topology λ such that there is a
filter L of linear subspaces of E with the following property: a subset U of E is λ-open
iff for any x ∈ U there exists V ∈ L such that x + V ⊆ U . One says that such a filter L
generates the topology L. A k-ltvs is a k-vector space equipped with a linear topology.
T
Observe that E is Hausdorff iff L = {0} (for some, and hence any, generating filter L);
from now on we assume always that this is the case.

Proposition 15. Let E be a k-ltvs. Any linear subspace U of E which is a neighborhood

of 0 is both open and closed. So E is totally disconnected (the only subsets of E which
are connected are the empty set and the one point sets).
Differential Linear Logic 53

Proof. Let L be a generating filter for the topology of E. First, let x ∈ U and let V ∈ L
be such that V ⊆ U (such a V exists because U is a neghborhood of 0), then we have
x + V ⊆ U since U is a linear subspace and hence U is open. Next let x ∈ E \ U . If
y ∈ U ∩ (x + U ) then we have y − x ∈ U and hence x ∈ U since y ∈ U and U is a linear
subspace: contradiction. Therefore U ∩ (x + U ) = ∅ and U is closed since U is open. 2

Any linear subspace which contains an open linear subspace is open.

4.4.1. Cauchy completeness. A net in E is a family (x(d))d∈D of elements of E indexed

by a directed set D. The net (x(d))d∈D converges to x ∈ E if, for any neighborhood U of
0, there exists d ∈ D such that ∀e ∈ D e ≥ d ⇒ x(e) − x ∈ U . Because E is Hausdorff, a
net converges to at most one point. As usual, one can check that a subset U of E is open
iff, for any net (x(d))d∈D which converges to a point x ∈ U , there exists d ∈ D such that
∀e ∈ D e ≥ d ⇒ x(e) ∈ U .
A net (x(d))d∈D is Cauchy if, for any neighborhood U of 0, there exists d ∈ D such
that ∀e, e0 ∈ D e, e0 ≥ d ⇒ x(e) − x(e0 ) ∈ U . This latter statement is equivalent to
∀e ∈ D e ≥ d ⇒ x(e) − x(d) ∈ U .
One says that E is complete if any Cauchy net in E converges.

4.4.2. Linear boundedness. Let E be an ltvs and let U be an open linear subspace of
E. Let πU : E → E/U be the canonical projection. This map is of course linear, and
its kernel is U which is a neighborhood of 0. This means that, endowing E/U with the
discrete topology, πU is continuous. Hence the quotient topology on E/U is the discrete
topology.
We say that a subspace B of E is linearly bounded if πU (B) is finite dimensional, for
every linear open subspace U of E. In other words, for any linear open subspace U , there
is a finite dimensional subspace A of E such that B ⊆ U + A.

Proposition 16. Any finite dimensional subspace of an ltvs E is linearly bounded. Let
B1 and B2 be subspaces of E. If B1 ⊆ B2 and B2 is linearly bounded, so is B1 . If B1
and B2 are linearly bounded, so is B1 + B2 .

A collection of subspaces of a vector space F having these properties is called a linear

bornology on F .
An ltvs E is locally linearly bounded if it has a linear open subspace which is linearly
bounded.

4.4.3. Linear and multilinear maps. Let E1 , . . . , En and F be k-ltvs’s. An n-multilinear

function f : E1 × · · · × En → F is hypocontinuous if, for any i ∈ {1, . . . , n}, any linear
open subspace V ⊆ F and any linearly bounded subspaces B1 ⊆ E1 ,. . . ,Bi−1 ⊆ Ei−1 ,
Bi+1 ⊆ Ei+1 ,. . . ,Bn ⊆ En , there exists an open linear subspace U ⊆ Ei such that
f (B1 × · · · × Bi−1 × U × Bi+1 × · · · × Bn ) ⊆ V .
We denote by (E1 , . . . , En ) ( F the k-vector space of all such multilinear maps. Given
linearly bounded subspaces B1 , . . . , Bn of E1 , . . . , En respectively and given a linear open
Thomas Ehrhard 54

subspace V of F , we define
Ann(B1 , . . . , Bn , V ) = {f ∈ (E1 , . . . , En ) ( F | f (B1 × · · · × Bn ) ⊆ V } .
This is a linear subspace of (E1 , . . . , En ) ( F and by Proposition 16 these subspaces
form a filter which defines a linear topology on (E1 , . . . , En ) ( F and this topology
is Hausdorff. Indeed, if f ∈ (E1 , . . . , En ) ( F is 6= 0, then take xi ∈ Ei such that
f (x1 , . . . , xn ) 6= 0. Since F is Hausdorff, there is a linear neighborhood V of 0 in F such
that f (x1 , . . . , xn ) ∈ / V . Let Bi = kxi ; this is a linearly bounded subspace of Ei and
f (B1 × · · · × Bn ) 6⊆ V .
In the case n = 1 (and E = E1 ), the corresponding maps f : E → F are simply called
linear, and they are continuous. The corresponding function space is denoted as E ( F .
If F = k, the corresponding maps are called (multi)linear (hypo)continuous forms.
If furthermore n = 1 the corresponding function space is denoted as E 0 and is called
topological dual of E.
Proposition 17. Let f : E1 × · · · × En → F be multilinear and hypocontinuous and
let Bi ⊆ Ei be linearly bounded subspaces for i = 1, . . . , n. Then f (B1 × · · · × Bn ) is a
linearly bounded subspace of F .

Proof. Let V be an open linear subspace of F . Let U1 be an open linear subspace of

E1 such that f (U1 × B1 × · · · × Bn ) ⊆ V . Let A1 be a finite dimensional subspace of
E1 such that B1 ⊆ U1 + A1 , we have f (B1 × · · · × Bn ) ⊆ V + f (A1 × B2 × · · · × Bn ).
Since A1 is bounded, one can find similarly a finite dimensional subspace A2 of E2
such that f (A1 × B2 × · · · × Bn ) ⊆ V + f (A1 × A2 × B3 × · · · × Bn ) and hence (since
V + V = V ) we get f (B1 × · · · × Bn ) ⊆ V + f (A1 × A2 × B3 × · · · × Bn ). Continuing
this process, we find finite dimensional subspaces Ai of Ei for i = 1, . . . , n such that
f (B1 × · · · × Bn ) ⊆ V + f (A1 × · · · × An ) and we conclude that f (B1 × · · · × Bn ) is
linearly bounded since f (A1 × · · · × An ) is finite dimensional. 2
It is tempting to think that (multi)linear continuous maps could be characterized as
those which preserve linear boundedness. This cannot be the case: think of a linear map
f : E → F where F is finite dimensional. Such a map preserves linear boundedness (any
subspace of F is linearly bounded) but has no reason to be continuous.

4.5. Finiteness spaces and the related ltvs’s

We restrict now our attention to particular ltvs’s which can be described in a simple
combinatorial way.

4.5.1. Basic definitions. Let I be a set. Given F ⊆ P(I), we define F ⊥ ⊆ P(I) by

F ⊥ = {u0 ⊆ I | ∀u ∈ F u ∩ u0 is finite} .
We have F ⊆ G ⇒ G ⊥ ⊆ F ⊥ , F ⊆ F ⊥⊥ and therefore F ⊥⊥⊥ = F ⊥ .
A finiteness space is a pair X = (|X|, F(X)) where |X| is a set and F(X) ⊆ P(|X|)
satisfies F(X) = F(X)⊥⊥ . The following properties follow easily from the definition
Differential Linear Logic 55

— if u ⊆ |X| is finite then u ∈ F(X)

— if u, v ∈ F(X) then u ∪ v ∈ F(X)
— if u ⊆ v ∈ F(X), then u ∈ F(X).
Let us prove for instance the second statement. Let u0 ∈ F(X)⊥ , then (u ∪ v) ∩ u0 =
(u ∩ u0 ) ∪ (v ∩ u0 ) is finite since both sets u ∩ u0 and v ∩ u0 are finite by our hypothesis
that u, v ∈ F(X). Since this holds for all u0 ∈ F(X)⊥ , we have u ∪ v ∈ F(X)⊥⊥ = F(X).
A strong isomorphism 12 between two finiteness spaces X and Y is a bijection ϕ : |X| →
|Y | such that, for all u ⊆ |X|, one has u ∈ F(X) iff ϕ(u) ∈ F(Y ).
Let X be a finiteness space. We define a k-vector space khXi as the set of all families
x ∈ k|X| such that the set supp(x) = {a ∈ |X| | xa 6= 0} belongs to F(X).
Given u0 ∈ F(X)⊥ , we define a linear subspace of khXi by
V X (u0 ) = {x ∈ khXi | supp(x) ∩ u0 = ∅} .
Observe first that ∀u0 , v 0 ∈ F(X)⊥ u0 ⊆ v 0 ⇔ V X (v 0 ) ⊆ V X (u0 ).
Since, given u0 , v 0 ∈ F(X)⊥ , we have V X (u0 ∪ v 0 ) = V X (u0 ) ∩ V X (v 0 ), the set {V X (u0 ) |
u ∈ F(X)⊥ } is a filter of linear subspaces of khXi. Moreover, observe that u0 ∈F(X)⊥ V X (u0 ) =
0
T

{0} (because ∀a ∈ |X| {a} ∈ F(X)⊥ ), and therefore this filter defines an Hausdorff linear
topology on khXi, that we call the canonical topology of khXi.

Proposition 18. For any finiteness space X, the ltvs khXi is Cauchy-complete.

Proof. Let (x(d))d∈D be a Cauchy net in khXi. Let a ∈ |X|. By taking u0 = {a} in
the definition of a Cauchy net, we see that there exist xa ∈ k and da ∈ D such that
∀e ≥ da x(e)a = xa . In that way we have defined x = (xa )a∈|X| ∈ k|X|
We prove first that
∀u0 ∈ F(X)⊥ ∃d ∈ D ∀e ≥ d ∀a ∈ u0 x(e)a = xa . (17)
Let u0 ∈ F(X)⊥ . Let d0 ∈ D be such that x(e) − x(d0 ) ∈ V X (u0 ) for all e ≥ d0 . Let a ∈ u0
and let da ≥ d0 be such that x(e)a = xa for all e ≥ da . Let e ≥ d0 . Let e0 ≥ e, da . We
have xa = x(e0 )a since e0 ≥ da and x(e0 )a = x(e)a since e, e0 ≥ d0 and a ∈ u0 . It follows
that ∀a ∈ u0 xa = x(e)a .
From this we deduce now that x ∈ khXi. Let u0 ∈ F(X)⊥ . Let d ∈ D be such
that ∀e ≥ d ∀a ∈ u0 x(e)a = xa . Then supp(x) ∩ u0 = supp(x(d)) ∩ u0 is finite, so
supp(x) ∈ F(X)⊥⊥ = F(X), that is x ∈ khXi.
Now Condition (17) expresses exactly that limd∈D x(d) = x and hence the net (x(d))d∈D
converges. 2
A natural question is whether the ltvs khXi, which is Hausdorff, is always metrizable.
We provide a necessary and sufficient condition under which this is the case.

12 This would coincide with the categorical notion of isomorphism if we were using morphisms which
are defined as relations. With linear continuous maps (between the associated ltvs’s) as morphisms,
the present notion of isomorphism is a particular case of the standard categorical one: we can have
more linear homeomorphisms from khXi to khY i than those which are generated by such finiteness-
preserving bijections between webs.
Thomas Ehrhard 56

Proposition 19. Let X be a finiteness space. The ltvs khXi is metrizable iff there exists
a sequence (u0n )n∈N of elements of F(X)⊥ which is monotone (n ≤ m ⇒ u0n ⊆ u0m ) and
such that ∀u0 ∈ F(X)⊥ ∃n ∈ N u0 ⊆ u0n .

Proof. Let first (u0n )n∈N be a sequence of elements of F(X)⊥ which satisfies the condition
stated above. Given x, y ∈ khXi, we define

0 if x = y


d(x, y) = 2 −n
if x 6= y and n is the least integer

such that u0 ∩ supp(x − y) 6= ∅ .


n

Indeed, if x 6= y, then supp(x − y) 6= ∅ and hence, taking a ∈ supp(x − y), we can find
n ∈ N such that {a} ⊆ u0n . This function d is easily seen to be an ultrametric distance
(that is d(x, z) ≤ max(d(x, y), d(y, z))) and it generates the canonical topology of khXi.
Indeed we have
d(x, y) < 2−n iff x − y ∈ V X (u0n )
(indeed, d(x, y) < 2−n means that the least m such that x − y ∈ / V X (u0m ) satisfies m > n)
0
and hence B2−n = V X (un ), where Bε is the open ball centered at 0 and of radius ε.
Conversely, assume that khXi is metrizable and let d be a distance defining the canon-
ical topology of khXi. For each n ∈ N, B2−n is a neighborhood of 0 and hence there
exist vn0 ∈ F(X)⊥ such that V X (vn0 ) ⊆ B2−n . Let u0n = v00 ∪ · · · ∪ vn0 ∈ F(X)⊥ . Then
V X (u0n ) ⊆ V X (vn0 ) ⊆ B2−n . Now let u0 ∈ F(X)⊥ , then V X (u0 ) is a neighborhood of 0
and hence there exists n such that B2−n ⊆ V X (u0 ), which implies V X (u0n ) ⊆ V X (u0 ) and
hence u0 ⊆ u0n . 2

It follows that there are non metrizable ltvs associated with finiteness spaces. We give
in Proposition 20 an example of this situation which arises in the semantics of LL, using
exponential constructions that will be introduced in Section 4.5.3.

Proposition 20. The ltvs kh!?1i is not metrizable

Proof. Let X = !?1, so that |X| = Mfin (N) and a subset u for |X| belongs to F(X)
iff ∃n ∈ N u ⊆ Mfin ({0, . . . , n}). The proof is a typical Cantor diagonal reasoning.
We assume towards a contradiction that khXi is metrizable, that is by Proposition 19,
we assume that there is a monotone sequence (u0n )n∈N of elements of F(X)⊥ such that
∀u0 ∈ F(X)⊥ ∃n ∈ N u0 ⊆ u0n . Let n ∈ N, we have {p[n] | p ∈ N} ∈ Mfin ({0, . . . , n})
and hence u0n ∩ {p[n] | p ∈ N} is finite. Therefore we can find a function f : N → N such
that ∀n ∈ N f (n)[n] ∈ / u0n . Let u0 = {f (n)[n] | n ∈ N}. Then u0 ∈ F(X)⊥ since, for any
0
n ∈ N, u ∩ Mfin ({0, . . . , n}) = {f (i)[i] | i ∈ [0, n]} is finite. But for all n ∈ N we have
f (n)[n] ∈ u0 \ u0n and so u0 6⊆ u0n . 2

We consider this as a very interesting phenomenon which seems to reveal a relation

between the topological complexity of the interpretation of a type with its logical com-
plexity (alternation of exponentials).
Differential Linear Logic 57

4.5.2. Linearly bounded subspaces. Let X be a finiteness space. We are interested in

characterizing the linearly bounded subspaces of khXi.
Given u ⊆ |X|, let BX (u) = {x ∈ khXi | supp(x) ⊆ u}. This is a linear subspace of
khXi.
Let u ∈ F(X). We prove that BX (u) is linearly bounded. Let u0 in F(X)⊥ . Observe
that V X (u0 ) = BX (|X| \ u0 ). We have therefore BX (u) ⊆ V X (u0 ) + BX (u ∩ u0 ), and since
u ∩ u0 is finite, the space BX (u ∩ u0 ) is finite dimensional. Let U be an open subspace
of U , let u0 ⊆ F(X)⊥ be such that V X (u0 ) ⊆ U . Then BX (u) ⊆ U + BX (u ∩ u0 ). Hence
BX (u) is linearly bounded. We show now that this condition is actually sufficient.

Proposition 21. A linear subspace B of khXi is linearly bounded iff there exists u ∈
F(X) such that B ⊆ BX (u).

S
Proof. Assume that B is linearly bounded. Let u = x∈B supp(x), so that B ⊆ BX (u),
we prove that u ∈ F(X). Let u0 ∈ F(X)⊥ . Let A be a finite dimensional subspace
of E such that B ⊂ V X (u0 ) + A. Let A0 be a finite generating subset of A and let
S
u0 = y∈A0 supp(y) ∈ F(X). Then x ∈ A ⇒ supp(x) ⊆ u0 (that is A ⊆ BX (u0 )).
Let x ∈ B, we write x = x1 + x2 where x1 ∈ V X (u0 ) and x2 ∈ A. We have supp(x) ⊆
supp(x1 ) ∪ supp(x2 ) and hence u0 ∩ supp(x) ⊆ (u0 ∩ supp(x1 )) ∪ (u0 ∩ supp(x2 )) ⊆ u0 ∩ u0
since u0 ∩ supp(x1 ) = ∅. Since this holds for all x ∈ B, we have u0 ∩ u ⊆ u0 ∩ u0 so u0 ∩ u
is finite and hence u ∈ F(X) 2

Proposition 22. The ltvs khXi is locally linearly bounded iff there exist u ∈ F(X) and
u0 ∈ F(X)⊥ such that u ∪ u0 = |X|.

This is an obvious consequence of Proposition 21.

Fink is the category whose objects are the finiteness spaces and such that Fink (X, Y )
is the set of all continuous linear maps khXi → khY i.

4.5.3. Constructions of finiteness spaces. We give a number of constructions on finiteness

spaces which allow one to interpret differential LL, starting with the most important one,
which is the linear function space.
The most striking features of these constructions can be summarized by the two fol-
lowing statements.
— In spite of the fact that these constructions are algebraic in nature (tensor product,
linear function space, topological dual etc), they are entirely performed on the webs of
the finiteness spaces and do not involve the scalar coefficients. This means in particular
that they do not depend on the choice of the field, and this is quite surprising.
— So, these constructions are performed on the webs, but they do not really depend on
them, in the following sense. Defining an intrinsic finiteness space as a k-ltvs which
is linearly homeomorphic to khXi for some finiteness space X, all these constructions
can be transferred to the category of intrinsic finiteness spaces and continuous and
linear maps.
Thomas Ehrhard 58

Let X and Y be finiteness spaces. Let X ( Y be the finiteness space such that
|X ( Y | = |X| × |Y | and
F(X ( Y ) = {u × v 0 | u ∈ F(X) and v 0 ∈ F(Y ⊥ )}⊥
= {w ⊆ |X| × |Y | | ∀u ∈ F(X) ∀v 0 ∈ F(Y )⊥ w ∩ (u × v 0 ) is finite}
Let w ∈ F(X ( Y ), u ∈ F(X) and v 0 ∈ F(X)⊥ . It follows from (16) that wu ∈ F(Y ) and
that w⊥ v 0 ∈ F(X)⊥ .
Let M ∈ khX ( Y i. If x ∈ khXi and b ∈ |Y |, then supp(M )⊥ {b} ∈ F(X)⊥ and
hence the sum a∈|X| Ma,b xa is finite. Therefore we can define M x ∈ k|Y | by M x =
P
P
( a∈|X| Ma,b xa )b∈|Y | . Since supp(M x) ⊆ supp(M ) supp(x), we have M x ∈ khY i and
hence the function fun(M ) defined by fun(M )(x) = M x is a linear map khXi → khY i.
Moreover, fun(M ) is continuous. Indeed, for any v 0 ∈ F(Y )⊥ we have V X (supp(M )⊥ v 0 ) ⊆
−1 −1
fun(M ) (V Y (v 0 )) and hence fun(M ) (V Y (v 0 )) is open since supp(M )⊥ v 0 ∈ F(X)⊥ .
Given finiteness spaces Z1 , Z2 , we define immediately the finiteness space Z1 ⊗ Z2 as
Z1 ⊗ Z2 = (Z1 ( Z2⊥ )⊥ , so that |Z1 ⊗ Z2 | = |Z1 | × |Z2 |. One of the most pleasant
features of the theory of finiteness spaces is the following property (see (Ehr05)) which
has been considerably generalized in (TV10).
Proposition 23. Let w ⊆ |Z1 | × |Z2 |. One has w ∈ F(Z1 ⊗ Z2 ) iff pri (w) ∈ F(Zi ) for
i = 1, 2.
Coming back to linear function spaces, this means in particular that, given w ⊆ |X| ×
|Y |, one has w ∈ F(X ( Y )⊥ iff there are u ∈ F(X) and v 0 ∈ F(Y )⊥ such that w ⊆ u×v 0 ,
from which we derive a simple characterization of the topology of linear function spaces.
Proposition 24. The function FunX,Y : M 7→ fun(M ) is a linear homeomorphism from
khX ( Y i to khXi ( khY i, equipped with the topology of uniform convergence on
linearly bounded subspaces.

Proof. The proof that FunX,Y is a linear isomorphism can be found in (Ehr05). We
prove that this linear isomorphism is an homeomorphism. Let B ⊆ khXi be a bounded
subspace and V ⊆ khY i is an open subspace. Let u ∈ F(X) be such that B ⊆ BX (u)
and let v 0 ∈ F(Y )⊥ be such that V Y (v 0 ) ⊆ V . Then u × v 0 ∈ F(X ( Y )⊥ and hence
V X(Y (u × v 0 ) ⊆ khX ( Y i is an open subspace. Let M ∈ V X(Y (u × v 0 ), x ∈ B and
b ∈ v 0 , we have (M x)b = 0 since supp(x) ⊆ u, which shows that FunX,Y (M )(B) ⊆ V and
hence FunX,Y is continuous.
Let now W ⊆ khX ( Y i be an open subspace. Let w ∈ F(X ( Y )⊥ be such that
V X(Y (w) ⊆ W . By Proposition 23, there are u ∈ F(X) and v 0 ∈ F(Y )⊥ such that
w ⊆ u × v 0 , and hence V X(Y (u × v 0 ) ⊆ W . Then, given M ∈ V X(Y (u × v 0 ), we have
FunX,Y (M )(BX (u)) ⊆ V Y (v 0 ), which shows that FunX,Y (W ) is an open linear subspace
of khXi ( khY i.
We have seen that FunX,Y is a continuous and open bijection and hence it is an
homeomorphism. 2
The tensor product X ⊗ Y defined above is characterized by a standard universal
property: it classifies the hypocontinuous bilinear maps.
Differential Linear Logic 59

Given vectors x ∈ khXi and y ∈ khY i, then x ⊗ y ∈ k|X⊗Y | defined by (x ⊗ y)(a,b) =

xa yb is clearly an element of khX ⊗ Y i since supp(x ⊗ y) = supp(x)×supp(y) ∈ F(X ⊗ Y ).
The map

τ : khXi × khY i → khX ⊗ Y i

(x, y) 7→ x ⊗ y

is obviously bilinear, let us check that it is hypocontinuous.

Let W be an open linear subspace of khX ⊗ Y i and let w0 ∈ F(X ⊗ Y )⊥ be such
that V X⊗Y (w0 ) ⊆ W . Let B ⊆ khXi be a linearly bounded subspace and let u ∈
F(X) be such that B ⊆ BX (u). Since w0 ∈ F(X ( Y ⊥ ) we have v 0 = w0 u ∈ F(Y )⊥ .
Let x ∈ B and y ∈ V Y (v 0 ), we have supp(x) ⊆ u and hence pr2 (supp(x ⊗ y) ∩ w0 ) ⊆
pr2 ((u × supp(y)) ∩ w0 ) = (w0 u) ∩ supp(y) = ∅ by definition of V Y (v 0 ). Therefore x ⊗ y ∈
V X⊗Y (w0 ) ⊆ W . Symmetrically, taking a linearly bounded subspace C of khY i, we show
that there is an open linear subspace U of khXi such that τ (U × C) ⊆ W . So the map
τ is bilinear and hypocontinuous.

Proposition 25. Let Z be a finiteness space and let f : khXi×khY i → khZi be bilinear
and hypocontinuous. There exists exactly one continuous linear map f˜ : khX ⊗ Y i →
khZi such that f = f˜ ◦ τ .

Proof. We define a matrix M ∈ k|X|×|Y |×|Z| by Ma,b,c = f (ea , eb )c and we show first
that supp(M ) ∈ F(X ⊗ Y ( Z).
So let u ∈ F(X), v ∈ F(Y ) and w0 ∈ F(Z)⊥ ; we must show that supp(M ) ∩ (u × v × w0 )
is finite. Let v 0 ∈ F(Y )⊥ and u0 ∈ F(X)⊥ be such that

f (BX (u) × V Y (v 0 )) ⊆ V Z (w0 ) and f (V X (u0 ) × BY (v)) ⊆ V Z (w0 ) .

Let (a, b, c) ∈ supp(M ) ∩ (u × v × w0 ), since f (ea , eb )c ∈

/ V Z (w0 ) (by definition of V Z (w0 )
and by our assumption about (a, b, c)), we must have

/ BX (u) × V Y (v 0 ) and (ea , eb ) ∈

(ea , eb ) ∈ / V X (u0 ) × BY (v).

But we know that a ∈ u and b ∈ v, that is ea ∈ BX (u) and eb ∈ BY (v). It follows that
/ V X (u0 ), that is a ∈ u0 , and similarly b ∈ v 0 .
ea ∈
Since BX (u) and BY (v) are linearly bounded, so is f (BX (u)×BY (v)) by Proposition 17
and hence there exists w ∈ F(Z) such that

f (BX (u) × BY (v)) ⊆ BZ (w) .

Therefore f (ea , eb ) ∈ BZ (w) and hence c ∈ w.

We have shown that

supp(M ) ∩ (u × v × w0 ) ⊆ (u ∩ u0 ) × (v ∩ v 0 ) × (w ∩ w0 )

and hence supp(M ) ∩ (u × v × w0 ) is finite, so M ∈ khX ⊗ Y ( Zi.

Let f˜ = fun(M ), it is a linear and continuous map from khX ⊗ Y i to khZi. We have
f˜(ea ⊗ eb ) = f (ea , eb ) for each (a, b) ∈ |X| × |Y |. Let x ∈ F(X) and y ∈ khY i, by separate
Thomas Ehrhard 60

continuity of f (which is a consequence of hypocontinuity) we have

X X
f (x, y) = f ( xa ea , yb eb )
a∈|X| b∈|Y |
X
= xa yb f (ea , eb )
(a,b)∈|X|×|Y |
X
= xa yb f˜(ea ⊗ eb )
(a,b)∈|X|×|Y |

= f˜(x ⊗ y) by continuity of f˜ .

Uniqueness of the continuous linear map f˜ results from the fact that necessarily f˜(ea ⊗
eb ) = f (ea , eb ). 2
Then one proves easily that the category Fink equipped with this tensor product
(whose neutral object is 1, which satisfies obviously kh1i = k) is ∗-autonomous, the
object of morphisms from X to Y being X ( Y and the dualizing object being ⊥ = 1
(indeed, the finiteness spaces X ( ⊥ and X ⊥ are obviously strongly isomorphic).
This category is preadditive in the sense of Section 2.4 since homsets Fink (X, Y )
have an obvious structure of k-vector space which is compatible with all the categorical
operations introduced so far.
Countable products and coproducts are available as well. Let (Xi )i∈I be a countable
˘ S
family of finiteness spaces. The finiteness space X = i∈I Xi is given by |X| = i∈I |Xi |
and F(X) = {w ⊆ |X| | ∀i ∈ I wi ∈ F(Xi )} where wi = {a ∈ |Xi | | (i, a) ∈ w}. It is easy
to check that
F(X)⊥ = {w0 ⊆ |X| | ∀i ∈ I wi0 ∈ F(Xi )⊥ and wi0 = ∅ for almost all i}
˘
and it follows that F(X)⊥⊥ = F(X). It is clear that kh i∈I Xi i = i∈I khXi i up to
Q
˘
a straightforward strong isomorphism and that i∈I Xi together with projections πj :
˘
kh i∈I Xi i → khXj i defined in the obvious way, is the cartesian product of the Xi ’s.
L ˘ ⊥ ⊥


Thanks to ∗-autonomy, the coproduct of the Xi ’s is given by i∈I Xi = i∈I Xi
L Q
and kh i∈I Xi i ⊆ i∈I khXi i is the space of all families (xi )i∈I of vectors such that
˘
xi = 0 for almost all i ∈ I. Of course, the canonical linear topology on kh i∈I Xi i is the
L
product topology, but the canonical topology on kh i∈I Xi i is much finer: it is generated
Q
by all products i∈I Vi where Vi is a linear neighborhood of 0 in khXi i.
For finite families of objects, products and coproducts coincide.
Let X be a finiteness space. We define !X by |!X| = Mfin (|X|) and
[
F(!X) = {A ⊆ |!X| | supp(m) ∈ F(X)}
m∈A

and it can be proved that indeed F(!X) = F(!X)⊥⊥ (again, see (TV10) for more general
results of this kind).
Given x ∈ khXi and m ∈ |!X|, we set
Y
xm = xm(a)
a ∈k
a∈|X|
Differential Linear Logic 61

(this is a finite product since supp(m) is a finite set), so that

x! = (xm )m∈|!X| ∈ kh!Xi

by definition of F(!X). Let M ∈ kh!X ( Y i, it is not hard to see that one defines a map
Fun(M ) : khXi → khY i by setting
 
X
Fun(M )(x) =  Mm,b xm 
m∈|!X|
b∈|Y |

all these sums are indeed finite, see (Ehr05) for the details. When the field k is infinite,
the map M 7→ Fun(M ) is injective.
In (Ehr05), it is also proven that ! is a functor. Given M ∈ khX ( Y i one defines
!M ∈ kh!X ( !Y i by setting, for m ∈ |!X| and p ∈ |!Y |,
X
(!M )m,p = [r] M r
r∈L(m,p)

where
X X
L(m, p) = {r ∈ Mfin (|X| × |Y |) | r(a, b) = m(a) and r(a, b) = p(b)}
b∈|Y | a∈|X|

(so that r ∈ L(m, p) ⇒ #m = #r = #p) and

Y p(b)!
[r] = Q ∈ N+ .
a∈|X| r(a, b)!
b∈|Y |

is a generalized multinomial coefficient.

This operation is functorial: !Id = Id and !M !N = !(M N ), and we also have

Fun(!M )(x) = !M x! = ( M x)! .

When k is infinite, this latter equation completely characterizes !M , by injectivity of the

operation Fun in that case. This functor has a comonad structure, of which we recall here
only the counit dX ∈ kh!X ( Xi given by (dX )m,a = δm,[a] .
The bijection |!(X & Y )| → |!X ⊗ !Y | which maps the element q ∈ |!(X & Y )| to the
pair (m, p) ∈ |!X ⊗ !Y | defined by m(a) = q(1, a) and p(b) = q(2, b) is a strong iso-
morphism of finiteness spaces. We also have a strong isomorphism from !⊥ to 1. These
strong isomorphisms induce natural isomorphisms m2X,Y ∈ Fink (!X ⊗ !Y , !(X & Y ))
and m0 ∈ Fink (1, !>) which endow the functor ! with a monoidality structure from
(Fink , &, >) to (Fink , ⊗, 1), satisfying moreover the coherence diagram (??): to summa-
rize, equipped with the structure described above, Fink is a Seely category, that is, a
categorical model of classical LL.
Applying the general recipe of Section 4.1, we get the contraction natural transfor-
mation cX : !X → !X ⊗ !X and the weakening morphism wX : !X → 1. We check that
(wX )m,∗ = δm,[] and that (cX )m,(p,q) = δm,p+q . We also get the cocontraction natural
transformation cX : !X ⊗ !X → !X and the coweakening morphism wX : 1 → !X. And
Thomas Ehrhard 62

we check that (wX )∗,m = δm,[] , and that (cX )(p,q),m = p+q


p δm,p+q where
 
m Y m(a)!
= ∈ N+
p p(a)!(m(a) − p(a))!
a∈|X|

is a generalized binomial coefficient.

We also have a codereliction natural transformation dX : X → !X given by (dX )a,m =
δm,[a] , which is easily seen to satisfy the conditions of Section 2.5 and 2.6, so that Fink
is a model of full differential LL.

4.5.4. An intrinsic presentation of function spaces. We have seen that a morphism from
X to Y of the linear category Fink can be seen both as an element of khX ( Y i and
as a continuous linear function from khXi to khY i.
A morphism from X to Y in the Kleisli category Relk! is an element of kh!X ( Y i.
Given M ∈ kh!X ( Y i, we have seen that we can define a function Fun M : khXi → khY i
by
X
Fun(M )(x) = M x! = ( Mm,b xm )b∈|Y | .
m∈|!X|

Moreover, the correspondence M → Fun(M ) is functorial. We provide here an intrinsic

characterization of these functions.
Let E and F be ltvs’s. Let us say that a function f : E → F is polynomial if there is
n ∈ N and hypocontinuous i-linear maps fi : E i → F (for i = 0, . . . , n) such that
f (x) = f0 + f1 (x) + · · · + fn (x, . . . , x) .
A polynomial map f of the form f (x) = fn (x, . . . , x), where fn is an n-linear hypocon-
tinuous function, is said to be homogeneous of degree n (this condition implies of course
∀t ∈ k f (tx) = tn f (x), and when k is infinite, a polynomial function is homogeneous iff
it satisfies this latter condition).
Let Polk (E, F ) be the k-vector space of polynomial functions from E to F . This space
can be endowed with the linear topology of uniform convergence on all linearly bounded
subspaces, which admits the following generating filter base of open neighborhoods of
0: the basic opens are the linear subspaces Ann(B, V ) = {f ∈ Polk (E, F ) | f (B) ⊆ V },
where B is a linearly bounded subspace of E and V is a linear open subspace of F . Let
Anak (E, F ) be the completion13 of that ltvs.

Theorem 26. Assume that k is infinite. For any finiteness spaces X and Y , the ltvs
kh!X ( Y i is linearly homeomorphic to Anak (khXi, khY i).

Proof. Let h : khXin → khY i be an hypocontinuous n-linear function of matrix M ∈

khX ⊗ · · · ⊗ X ( Y i, so that h(x1 , . . . , xn ) = M (x1 ⊗ · · · ⊗ xn ).

13 A completion of an ltvs E is a pair (Ẽ, h) where Ẽ is a complete ltvs and h : E → Ẽ is a linear and
continuous map such that, for any complete ltvs F and any linear continuous map f : E → F , there
is an unique linear and continuous map f˜ : Ẽ → F such that f˜ ◦ h = f . Using standard techniques,
one can prove that any ltvs admits a completion, which is unique up to unique isomorphism.
Differential Linear Logic 63

Remember that, using contraction and dereliction, we have defined in Section 3 the
morphism dnX ∈ kh!X ( X ⊗ · · · ⊗ Xi. Then we have N = M dnX ∈ kh!X ( Y i, and it
is easy to see that
Fun(N )(x) = h(x, . . . , x) .
In that way, we see that any polynomial map from khXi to khY i is an element of
kh!X ( Y i; we have an inclusion Polk (khXi, khY i) ⊆ kh!X ( Y i. Actually, the expo-
nential structure Fink is Taylor and this notion of polynomial map coincides with the
general notion of Section 3.1.
n
Conversely, let (m, b) ∈ |!X ( Y | with m = [a1 , . . . , an ]. The map f : khXi →
k defined by f (x(1), . . . , x(n)) = x(1)a1 . . . x(n)an is multilinear and hypocontinuous.
n
Hence the same holds for the map x 7→ f (x)eb from khXi to khY i. Therefore we have
(|!X(Y |)
k ⊆ Polk (khXi, khY i) (given a set I, remember that k(I) is the k-vector space
generated by I, that is, the space of all families (ai )i∈I of elements of k such that ai = 0
for almost all i’s).
Hence Polk (khXi, khY i) is a dense subspace of kh!X ( Y i. To show that kh!X ( Y i
is the completion of Polk (khXi, khY i) it suffices to show that the above defined linear
topology on that space (uniform convergence on all linearly bounded subspaces) is the
restriction of the topology of kh!X ( Y i.
Let B ⊆ khXi be a linearly bounded subspace and let V ⊆ khY i be linear open.
Let v 0 ∈ F(Y )⊥ be such that V Y (v 0 ) ⊆ V . By Proposition 21, supp(B) ∈ F(X), so
Mfin (supp(B)) ∈ F(!X). Let
M ∈ V !X(Y (Mfin (supp(B)) × v 0 ) ⊆ kh!X ( Y i ,
then fun(M )(x)b = 0 for each x ∈ B and b ∈ v 0 . So we have
V !X(Y (Mfin (supp(B)) × v 0 ) ∩ Polk (khXi, khY i) ⊆ Ann(B, V ) .
Conversely let U ∈ F(!X) and v 0 ∈ F(Y )⊥ , then we have u = m∈U supp(m) ∈ F(X)
S

and hence the subspace B ⊆ khXi of all vectors which vanish outside u is linearly
bounded. Let M ∈ kh!X ( Y i be such that the map Fun(M ) : khXi → khY i is polyno-
mial and belongs to Ann(B, V Y (v 0 )). Then for any m = [a1 , . . . , an ] ∈ Mfin (u) and b ∈ v 0
m(a ) m(a )
we have Mm,b = 0 because this scalar is the coefficient of the monomial ξ1 1 . . . ξn n in
the polynomial P ∈ k[ξ1 , . . . , ξn ] such that P (z1 , . . . , zn ) = Fun(M )(x)b where x ∈ khXi
is such that xa = zi if a = ai and xa = 0 if a ∈ / supp(m), and P = 0 because
0
Fun(M )(B) ⊆ V Y (v ) by assumption (we also use the fact that k is infinite). Hence
M ∈ V !X(Y (U × v 0 ) and we have shown that
Ann(B, V Y (v 0 )) ⊆ V !X(Y (U × v 0 ) ∩ Polk (khXi, khY i) ,
showing that this latter set is a neighborhood of 0 in the space of polynomials. 2
The Taylor formula proved in (Ehr05) for the morphisms of this Kleisli category shows
that actually any morphism is the sum of a converging series whose n-th term is an
homogeneous polynomial of degree n.
As an example, take E = k[ξ] ' kh!1 ( 1i. The corresponding topology on E is
the discrete topology. A typical example of generalized polynomial map is the function
Thomas Ehrhard 64

ϕ : E → k which maps a polynomial P to P (P (0)), in other words, ϕ(x0 + x1 ξ + · · · +

xn ξ n ) = x0 +x1 x0 +· · ·+xn xn0 . Considered as a generalized polynomial of infinitely many
variables x0 , x1 , . . . , we see that ϕ is not of bounded degree, and so it is not polynomial.
Nevertheless, it corresponds to a very simple and finite computation on polynomials.

4.5.5. Antiderivatives. Just as Rel, the Fink exponential structure is Taylor in the sense
of Section 3.1. Moreover, if k is of characteristic 0 (meaning that ∀n ∈ N n 1 = 0 ⇒ n = 0)
it has antiderivatives in the sense of 3.2, because the morphism JX = IdX +(∂ X ∂X ) :
!X → !X satisfies (JX )p,q = (#p + 1)δp,q for all p, q ∈ |!X| and hence is an isomorphism
1
whose inverse IX is given by (IX )p,q = #p+1 δp,q .

Acknowledgment
Part of the work reported in this article has been supported by the French-Chinese project
ANR-11-IS02-0002 and NSFC 61161130530 Locali.

References
Samson Abramsky. Computational interpretations of linear logic. Theoretical Computer Science,
111:3–57, 1993.
Antonio Bucciarelli, Alberto Carraro, Thomas Ehrhard, and Giulio Manzonetto. Full Abstrac-
tion for Resource Calculus with Tests. In Marc Bezem, editor, CSL, volume 12 of LIPIcs.
Schloss Dagstuhl - Leibniz-Zentrum fuer Informatik, 2011.
Gérard Boudol, Pierre-Louis Curien, and Carolina Lavatelli. A semantics for lambda calculi
with resource. Mathematical Structures in Computer Science, 9(4):437–482, 1999.
Richard Blute, Robin Cockett, and Robert Seely. Differential categories. Mathematical Struc-
tures in Computer Science, 16(6):1049–1083, 2006.
Richard Blute, Thomas Ehrhard, and Christine Tasson. A convenient differential category.
Cahiers de Topologie et Géométrie Différentielle Catégoriques, 53, 2012.
Gavin Bierman. What is a categorical model of intuitionistic linear logic? In Mariangiola Dezani-
Ciancaglini and Gordon D. Plotkin, editors, Proceedings of the second Typed Lambda-Calculi
and Applications conference, volume 902 of Lecture Notes in Computer Science, pages 73–93.
Springer-Verlag, 1995.
Pierre-Louis Curien, editor. Typed Lambda Calculi and Applications, 9th International Confer-
ence, TLCA 2009, Brasilia, Brazil, July 1-3, 2009. Proceedings, volume 5608 of Lecture Notes
in Computer Science. Springer, 2009.
N.G. De Bruijn. Generalizing Automath by means of a lambda-typed lambda calculus. In D.W.
Kueker, E.G.K. Lopez-Escobar, and C.H. Smith, editors, Mathematical Logic and Theoretical
Computer Science, Lecture Notes in Pure and Applied Mathematics, pages 71–92. Marcel
Dekker, 1987. Reprinted in: Selected papers on Automath, Studies in Logic, volume 133,
pages 313-337, North-Holland, 1994.
Vincent Danos and Laurent Regnier. Reversible, irreversible and optimal lambda-machines.
Theoretical Computer Science, 227(1-2):273–291, 1999.
Thomas Ehrhard. On Köthe sequence spaces and linear logic. Mathematical Structures in
Computer Science, 12:579–623, 2002.
Thomas Ehrhard. Finiteness spaces. Mathematical Structures in Computer Science, 15(4):615–
646, 2005.
Differential Linear Logic 65

Thomas Ehrhard. A finiteness structure on resource terms. In LICS, pages 402–410. IEEE
Computer Society, 2010.
Thomas Ehrhard. The Scott model of Linear Logic is the extensional collapse of its relational
model. Theoretical Computer Science, 424:20–45, 2012.
Thomas Ehrhard. A new correctness criterion for MLL proof nets. In Thomas A. Henzinger
and Dale Miller, editors, Joint Meeting of the Twenty-Third EACSL Annual Conference on
Computer Science Logic (CSL) and the Twenty-Ninth Annual ACM/IEEE Symposium on
Logic in Computer Science (LICS), CSL-LICS ’14, Vienna, Austria, July 14 - 18, 2014,
page 38. ACM, 2014.
Thomas Ehrhard and Laurent Regnier. The differential lambda-calculus. Theoretical Computer
Science, 309(1-3):1–41, 2003.
Thomas Ehrhard and Laurent Regnier. Böhm trees, Krivine machine and the Taylor expansion
of ordinary lambda-terms. In Arnold Beckmann, Ulrich Berger, Benedikt Löwe, and John V.
Tucker, editors, Logical Approaches to Computational Barriers, volume 3988 of Lecture Notes
in Computer Science, pages 186–197. Springer-Verlag, 2006. Long version available on http:
//www.pps.univ-paris-diderot.fr/~ehrhard/.
Thomas Ehrhard and Laurent Regnier. Uniformity and the Taylor expansion of ordinary lambda-
terms. Theoretical Computer Science, 403(2-3):347–372, 2008.
Marcelo P. Fiore. Differential structure in models of multiplicative biadditive intuitionistic
linear logic. In Simona Ronchi Della Rocca, editor, TLCA, volume 4583 of Lecture Notes in
Computer Science, pages 163–177. Springer, 2007.
Maribel Fernández and Ian Mackie. A Calculus for Interaction Nets. In Gopalan Nadathur,
editor, PPDP, volume 1702 of Lecture Notes in Computer Science, pages 170–187. Springer-
Verlag, 1999.
Stéphane Gimenez. Realizability proof for normalization of full differential linear logic. In
C.-H. Luke Ong, editor, TLCA, volume 6690 of Lecture Notes in Computer Science, pages
107–122. Springer-Verlag, 2011.
Jean-Yves Girard. The system F of variable types, fifteen years later. Theoretical Computer
Science, 45:159–192, 1986.
Jean-Yves Girard. Linear logic. Theoretical Computer Science, 50:1–102, 1987.
Jean-Yves Girard. Normal functors, power series and the λ-calculus. Annals of Pure and Applied
Logic, 37:129–177, 1988.
Michael Huth. Linear Domains and Linear Maps. In Stephen D. Brookes, Michael G. Main,
Austin Melton, Michael W. Mislove, and David A. Schmidt, editors, MFPS, volume 802 of
Lecture Notes in Computer Science, pages 438–453. Springer-Verlag, 1993.
Jean-Louis Krivine. Un interprteur du lambda-calcul. Unpublished note, 1985.
Jean-Louis Krivine. A call-by-name lambda-calculus machine. Higher-Order and Symbolic Com-
putation, 20(3):199–207, 2007.
Saunders Mac Lane. Categories for the Working Mathematician, volume 5 of Graduate Texts in
Mathematics. Springer-Verlag, 1971.
Paul-André Melliès. Categorical semantics of linear logic. Panoramas et Synthèses, 27, 2009.
Ian Mackie and Shinya Sato. A Calculus for Interaction Nets Based on the Linear Chemical
Abstract Machine. Electronic Notes in Theoretical Computer Science, 192(3):59–70, 2008.
Michele Pagani. The cut-elimination theorem for differential nets with promotion. In Curien
(Cur09), pages 219–233.
Michele Pagani and Paolo Tranquilli. Parallel Reduction in Resource Lambda-Calculus. In
Zhenjiang Hu, editor, APLAS, volume 5904 of Lecture Notes in Computer Science, pages
226–242. Springer, 2009.
Thomas Ehrhard 66

Michele Pagani and Paolo Tranquilli. The Conservation Theorem for Differential Nets. Mathe-
matical Structures in Computer Science, 2011. To appear.
Christian Retoré. Handsome proof-nets: perfect matchings and cographs. Theoretical Computer
Science, 294(3):473–488, 2003.
Christine Tasson. Algebraic totality, towards completeness. In Curien (Cur09), pages 325–340.
Christine Tasson. Sémantiques et syntaxes vectorielles de la logique linéaire. Thèse de doctorat,
Université Paris Diderot – Paris 7, 2009.
Paolo Tranquilli. Confluence of pure differential nets with promotion. In Erich Grädel and
Reinhard Kahle, editors, CSL, volume 5771 of Lecture Notes in Computer Science, pages
500–514. Springer-Verlag, 2009.
Christine Tasson and Lionel Vaux. Transport of finiteness structures and applications. Mathe-
matical Structures in Computer Science, 2010. To appear.
Lionel Vaux. The differential lambda-mu calculus. Theoretical Computer Science, 379(1-2):166–
209, 2005.
Lionel Vaux. The algebraic lambda-calculus. Mathematical Structures in Computer Science,
19(5):1029–1059, 2009.
Glynn Winskel. Linearity and non linearity in distributed computation. In Thomas Ehrhard,
Jean-Yves Girard, Paul Ruet, and Philip Scott, editors, Linear Logic in Computer Science, vol-
ume 316 of London Mathematical Society Lecture Notes Series. Cambridge University Press,
2004.