Linear logic

Linear logic is a substructural logic proposed by Jean-Yves Girard as a refinement of classical and intuitionistic logic, joining the
dualities of the former with many of the constructive properties of the latter.[1] Although the logic has also been studied for its own
sake, more broadly, ideas from linear logic have been influential in fields such as programming languages, game semantics, and
quantum physics (because linear logic can be seen as the logic of quantum information theory),[2] as well as linguistics,[3]
particularly because of its emphasis on resource-boundedness, duality
, and interaction.

Linear logic lends itself to many different presentations, explanations and intuitions. Proof-theoretically, it derives from an analysis
of classical sequent calculus in which uses of (the structural rules) contraction and weakening are carefully controlled. Operationally,
this means that logical deduction is no longer merely about an ever-expanding collection of persistent "truths", but also a way of
manipulating resources that cannot always be duplicated or thrown away at will. In terms of simple denotational models, linear logic
may be seen as refining the interpretation of intuitionistic logic by replacing cartesian (closed) categories by symmetric monoidal
(closed) categories, or the interpretation of classical logic by replacingboolean algebras by C*-algebras.

Contents
Connectives, duality, and polarity
Syntax
Sequent calculus presentation
Multiplicatives
Additives
Exponentials
Remarkable formulae
Encoding classical/intuitionistic logic in linear logic
The resource interpretation
Other proof systems
Proof nets
Semantics
Algebraic semantics
Decidability/complexity of entailment
Variants
See also
References
Further reading
External links

Connectives, duality, and polarity

Syntax
The language of classical linear logic (CLL) is defined inductively by theBNF notation

A ::= p ∣ p⊥
∣ A⊗A∣A⊕A
 ∣ A&A∣A⅋A
∣ 1∣0∣⊤∣⊥
∣ !A ∣ ?A
Here p and p⊥ range over logical atoms. For reasons to be explained below, the connectives ⊗, ⅋, 1, and ⊥ are called
multiplicatives, the connectives &, ⊕, ⊤, and 0 are called additives, and the connectives ! and ? are called exponentials. We can
further employ the following terminology:

⊗ is called "multiplicative conjunction" or "times" (or sometimes "tensor")

⊕ is called "additive disjunction" or "plus"
& is called "additive conjunction" or "with"
⅋ is called "multiplicative disjunction" or "par"
! is pronounced "of course" (or sometimes "bang")
? is pronounced "why not"
Binary connectives ⊗, ⊕, & and ⅋ are associative and commutative; 1 is the unit for ⊗, 0 is the unit for ⊕, ⊥ is the unit for ⅋ and
⊤ is the unit for &.

Every proposition A in CLL has a dual A⊥, defined as follows:

(p)⊥ = p⊥ (p⊥)⊥ = p
(A ⊗ B)⊥ = A⊥ ⅋ B⊥ (A ⅋ B)⊥ = A⊥ ⊗ B⊥
(A ⊕ B)⊥ = A⊥ & B⊥ (A & B)⊥ = A⊥ ⊕ B⊥
(1)⊥ = ⊥ (⊥)⊥ = 1
(0)⊥ = ⊤ (⊤)⊥ = 0
(!A)⊥ = ?(A⊥) (?A)⊥ = !(A⊥)
Observe that (-)⊥ is an involution, i.e., A⊥⊥ = A for all propositions. A⊥ is also called the linear
Classification of
negation of A. connectives

The columns of the table suggest another way of classifying the connectives of linear logic, termed add mul exp
polarity: the connectives negated in the left column (⊗, ⊕, 1, 0, !) are called positive, while their pos ⊕0 ⊗1 !
duals on the right (⅋, &, ⊥, ⊤, ?) are called negative; cf. table on the right. neg &⊤ ⅋⊥ ?

Linear implication is not included in the grammar of connectives, but is definable in CLL using linear
negation and multiplicative disjunction, by A ⊸ B := A⊥ ⅋ B. The connective ⊸ is sometimes pronounced "lollipop", owing to
its shape.

Sequent calculus presentation

One way of defining linear logic is as a sequent calculus. We use the letters Γ and Δ to range over list of propositions A1, ..., An,
also called contexts. A sequent places a context to the left and the right of the turnstile, written Γ Δ. Intuitively, the sequent asserts
that the conjunction of Γ entails the disjunction of Δ (though we mean the "multiplicative" conjunction and disjunction, as explained
below). Girard describes classical linear logic using only one-sided sequents (where the left-hand context is empty), and we follow
here that more economical presentation. This is possible because any premises to the left of a turnstile can always be moved to the
other side and dualised.

We now give inference rules describing how to build proofs of sequents.[4]

First, to formalize the fact that we do not care about the order of propositions inside a context, we add the structural rule of exchange:
 Γ, A1, A2, Δ
Γ, A2, A1, Δ
Note that we do not add the structural rules of weakening and contraction, because we do care about the absence of propositions in a
sequent, and the number of copies present.

Next we add initial sequents and cuts:

Γ, A A⊥, Δ
A, A⊥ Γ, Δ
The cut rule can be seen as a way of composing proofs, and initial sequents serve as the
units for composition. In a certain sense these
rules are redundant: as we introduce additional rules for building proofs below, we will maintain the property that arbitrary initial
sequents can be derived from atomic initial sequents, and that whenever a sequent is provable it can be given a cut-free proof.
Ultimately, this canonical form property (which can be divided into the completeness of atomic initial sequents and the cut-
elimination theorem, inducing a notion of analytic proof) lies behind the applications of linear logic in computer science, since it
allows the logic to be used in proof search and as a resource-aware lambda-calculus.

Now, we explain the connectives by giving logical rules. Typically in sequent calculus one gives both "right-rules" and "left-rules"
for each connective, essentially describing two modes of reasoning about propositions involving that connective (e.g., verification
and falsification). In a one-sided presentation, one instead makes use of negation: the right-rules for a connective (say ⅋) effectively
play the role of left-rules for its dual (⊗). So, we should expect a certain "harmony" between the rule(s) for a connective and the
rule(s) for its dual.

Multiplicatives
The rules for multiplicative conjunction ⊗
( ) and disjunction (⅋):

Γ, A Δ, B Γ, A, B
Γ, Δ, A ⊗ B Γ, A ⅋ B
and for their units:

Γ
1 Γ, ⊥
Observe that the rules for multiplicative conjunction and disjunction are admissible for plain conjunction and disjunction under a
classical interpretation (i.e., they are admissible rules inLK).

Additives
The rules for additive conjunction (&) and disjunction ⊕)
( :

Γ, A Γ, B Γ, A Γ, B
Γ, A & B Γ, A ⊕ B Γ, A ⊕ B
and for their units:

(no rule for 0)

Γ, ⊤
Observe that the rules for additive conjunction and disjunction are again admissible under a classical interpretation. But now we can
explain the basis for the multiplicative/additive distinction in the rules for the two different versions of conjunction: for the
multiplicative connective (⊗), the context of the conclusion (Γ, Δ) is split up between the premises, whereas for the additive case
connective (&) the context of the conclusion Γ
( ) is carried whole into both premises.

Exponentials
The exponentials are used to give controlled access to weakening and contraction. Specifically, we add structural rules of weakening
and contraction for ?'d propositions:[5]

Γ Γ, ?A, ?A
Γ, ?A Γ, ?A
and use the following logical rules:

?Γ, A Γ, A
?Γ, !A Γ, ?A
One might observe that the rules for the exponentials follow a different pattern from the rules for the other connectives, resembling
the inference rules governing modalities in sequent calculus formalisations of the normal modal logic S4, and that there is no longer
such a clear symmetry between the duals ! and ?. This situation is remedied in alternative presentations of CLL (e.g., the LU
presentation).

Remarkable formulae
In addition to the De Morgan dualities described above, some important equivalences in linear logic include:

Distributivity

Exponential isomorphism

(Here .)

Assume that ⅋ is any of the binary operators times, plus, with or par (but not linear implication). The following is not in general an
equivalence, only an implication:

Linear-distributions

A map which is not an isomorphism yet plays a crucial role in linear logic is:

(A ⊗ (B ⅋ C)) ⊸ ((A ⊗ B) ⅋ C)

Linear distributions are fundamental in the proof theory of linear logic. The consequence of this map were first investigated in [6] and
called a "weak distribution". In subsequent work it was renamed to "linear distribution" to reflect the fundamental connection to
linear logic.

Encoding classical/intuitionistic logic in linear logic

Both intuitionistic and classical implication can be recovered from linear implication by inserting exponentials: intuitionistic
implication is encoded as !A ⊸ B, while classical implication can be encoded as !?A ⊸ ?B or !A ⊸ ?!B (or a variety of
alternative possible translations).[7] The idea is that exponentials allow us to use a formula as many times as we need, which is
always possible in classical and intuitionistic logic.
Formally, there exists a translation of formulae of intuitionistic logic to formulae of linear logic in a way which guarantees that the
original formula is provable in intuitionistic logic if and only if the translated formula is provable in linear logic. Using the Gödel–
Gentzen negative translation, we can thus embed classical first-order logic into linear first-order logic.

The resource interpretation

Lafont (1993) first showed how intuitionistic linear logic can be explained as a logic of resources, so providing the logical language
with access to formalisms that can be used for reasoning about resources within the logic itself, rather than, as in classical logic, by
means of non-logical predicates and relations. Antony Hoare (1985)'s classical example of the vending machine can be used to
illustrate this idea.

Suppose we represent having a candy bar by the atomic proposition candy , and having a dollar by $1. To state the fact that a dollar
will buy you one candy bar, we might write the implication $1 ⇒ candy . But in ordinary (classical or intuitionistic) logic, from A
and A ⇒ B one can conclude A ∧ B. So, ordinary logic leads us to believe that we can buy the candy bar and keep our dollar! Of
course, we can avoid this problem by using more sophisticated encodings, although typically such encodings suffer from the frame
problem. However, the rejection of weakening and contraction allows linear logic to avoid this kind of spurious reasoning even with
the "naive" rule. Rather than $1 ⇒ candy , we express the property of the vending machine as a linear implication $1 ⊸ candy .
From $1 and this fact, we can concludecandy , but not $1 ⊗ candy . In general, we can use the linear logic proposition A ⊸ B to
express the validity of transforming resourceA into resource B.

Running with the example of the vending machine, let us consider the "resource interpretations" of the other multiplicative and
additive connectives. (The exponentials provide the means to combine this resource interpretation with the usual notion of persistent
logical truth.)

Multiplicative conjunction(A ⊗ B) denotes simultaneous occurrence of resources, to be used as the consumer directs. For example,
if you buy a stick of gum and a bottle of soft drink, then you are requesting gum ⊗ drink . The constant 1 denotes the absence of
any resource, and so functions as the unit of⊗.

Additive conjunction (A & B) represents alternative occurrence of resources, the choice of which the consumer controls. If in the
vending machine there is a packet of chips, a candy bar, and a can of soft drink, each costing one dollar, then for that price you can
buy exactly one of these products. Thus we write $1 ⊸ (candy & chips & drink). We do not write
$1 ⊸ (candy ⊗ chips ⊗ drink), which would imply that one dollar suffices for buying all three products together. However,
from $1 ⊸ (candy & chips & drink), we can correctly deduce $3 ⊸ (candy ⊗ chips ⊗ drink), where
$3 := $1 ⊗ $1 ⊗ $1. The unit ⊤ of additive conjunction can be seen as a wastebasket for unneeded resources. For example, we
can write $3 ⊸ (candy ⊗ ⊤) to express that with three dollars you can get a candy bar and some other stuff, without being more
specific (for example, chips and a drink, or $2, or $1 and chips, etc.).

Additive disjunction (A ⊕ B) represents alternative occurrence of resources, the choice of which the machine controls. For
example, suppose the vending machine permits gambling: insert a dollar and the machine may dispense a candy bar, a packet of
chips, or a soft drink. We can express this situation as $1 ⊸ (candy ⊕ chips ⊕ drink). The constant 0 represents a product that
cannot be made, and thus serves as the unit of ⊕ (a machine that might produce A or 0 is as good as a machine that always produces
A because it will never succeed in producing a 0).

Multiplicative disjunction (A ⅋ B) is more difficult to gloss in terms of the resource interpretation, although we can encode back
into linear implication, either asA⊥ ⊸ B or B⊥ ⊸ A.

Other proof systems

Proof nets
Introduced by Jean-Yves Girard, proof nets have been created to avoid the bureaucracy, that is all the things that make two
derivations different in the logical point of view, but not in a "moral" point of view.

For instance, these two proofs are "morally" identical:

A, B, C, D A, B, C, D
A ⅋ B, C, D A, B, C ⅋ D
A ⅋ B, C ⅋ D A ⅋ B, C ⅋ D
The goal of proof nets is to make them identical by creating a graphical representation of them.

Semantics

Algebraic semantics

Decidability/complexity of entailment
The entailment relation in full CLL is undecidable.[8] When considering fragments of CLL, the decision problem has varying
complexity:

Multiplicative linear logic (MLL): only the multiplicative connectives.MLL entailment is NP-complete, even restricting
to Horn clauses in the purely implicative fragment,[9] or to atom-free formulas.[10]
Multiplicative-additive linear logic (MALL): only multiplicatives and additives (i.e., exponential-free).
MALL entailment
is PSPACE-complete. [8]

Multiplicative-exponential linear logic (MELL): only multiplicatives and exponentials. By reduction from the
reachability problem forPetri nets,[11] MELL entailment must be at leastEXPSPACE-hard, although decidability itself
has had the status of a longstanding open problem. In 2015, a proof of decidability was published in the journal
TCS.[12]
Affine linear logic (that is linear logic with weakening, an extension rather than a fragment) was shown to be
decidable, in 1995.[13]

Variants
Many variations of linear logic arise by further tinkering with the structural rules:

Affine logic, which forbids contraction but allows global weakening (a decidable extension).
Strict logic or relevant logic, which forbids weakening but allows global contraction.
Non-commutative logicor ordered logic, which removes the rule of exchange, in addition to barring weakening and
contraction. In ordered logic, linear implication divides further into left-implication and right-implication.
Different intuitionistic variants of linear logic have been considered. When based on a single-conclusion sequent calculus
presentation, like in ILL (Intuitionistic Linear Logic), the connectives ⅋, ⊥, and ? are absent, and linear implication is treated as a
primitive connective. In FILL (Full Intuitionistic Linear Logic) the connectives ⅋, ⊥, and ? are present, linear implication is a
primitive connective and, similarly to what happens in intuitionistic logic, all connectives (except linear negation) are independent.
There are also first- and higher-order extensions of linear logic, whose formal development is somewhat standard (see first-order
logic and higher-order logic).

See also
Linear type system, a substructural type system
Logic of unity (LU)
Proof nets
Geometry of interaction
 Game semantics
Intuitionistic logic
Computability logic
Ludics
Chu spaces
Uniqueness type

References
1. Girard, Jean-Yves (1987). "Linear logic" (http://iml.univ-mrs.fr/~girard/linear.pdf) (PDF). Theoretical Computer
Science. 50 (1): 1–102. doi:10.1016/0304-3975(87)90045-4(https://doi.org/10.1016%2F0304-3975%2887%299004
5-4).
2. Baez, John; Stay, Mike (2008). Bob Coecke, ed. "Physics, Topology, Logic and Computation: A Rosetta Stone"(htt
p://math.ucr.edu/home/baez/rosetta.pdf)(PDF). New Structures of Physics.
3. de Paiva, V.; van Genabith, J.; Ritter, E. (1999). Dagstuhl Seminar 99341 on Linear Logic and Applications(http://ww
w.dagstuhl.de/Reports/99/99341.pdf)(PDF).
4. Girard (1987), p.22, Def.1.15
5. Girard (1987), p.25-26, Def.1.21
6. Cockett and Seely "Weakly distributive categories" JPAA 114(2) 133-173, 1997
7. Di Cosmo, Roberto. The Linear Logic Primer(http://www.dicosmo.org/CourseNotes/LinLog/). Course notes; chapter
2.
8. For this result and discussion of some of the fragments below
, see: Lincoln, Patrick; Mitchell, John; Scedrov
, Andre;
Shankar, Natarajan (1992). "Decision Problems for Propositional Linear Logic".Annals of Pure and Applied Logic.
56: 239–311. doi:10.1016/0168-0072(92)90075-B(https://doi.org/10.1016%2F0168-0072%2892%2990075-B) .
9. Kanovich, Max I. (1992-06-22). "Horn programming in linear logic is NP-complete".
Seventh Annual IEEE
Symposium on Logic in Computer Science, 1992. LICS '92. Proceedings . Seventh Annual IEEE Symposium on
Logic in Computer Science, 1992. LICS '92. Proceedings. pp. 200–210. doi:10.1109/LICS.1992.185533(https://doi.o
rg/10.1109%2FLICS.1992.185533).
10. Lincoln, Patrick; Winkler, Timothy (1994). "Constant-only multiplicative linear logic is NP-complete".
Theoretical
Computer Science. 135: 155–169. doi:10.1016/0304-3975(94)00108-1(https://doi.org/10.1016%2F0304-3975%289
4%2900108-1).
11. Gunter, C. A.; Gehlot, V. (1989). Tenth International Conference on Application and Theory of Petri Nets.
Proceedings. pp. 174–191.Missing or empty |title= (help)
12. Bimbó, Katalin (2015-09-13). "The decidability of the intensional fragment of classical linear logic".
Theoretical
Computer Science. 597: 1–17. doi:10.1016/j.tcs.2015.06.019(https://doi.org/10.1016%2Fj.tcs.2015.06.019) .
ISSN 0304-3975 (https://www.worldcat.org/issn/0304-3975).
13. Kopylov, A. P. (1995-06-01). "Decidability of linear affine logic" (http://citeseerx.ist.psu.edu/viewdoc/summary?doi=1
0.1.1.23.9226). , Tenth Annual IEEE Symposium on Logic inComputer Science, 1995. LICS '95. Proceedings . Tenth
Annual IEEE Symposium on Logic in Computer Science, 1995. LICS '95. Proceedings. pp. 496–504.
doi:10.1109/LICS.1995.523283(https://doi.org/10.1109%2FLICS.1995.523283) .

Further reading
Girard, Jean-Yves. Linear logic, Theoretical Computer Science, Vol 50, no 1, pp. 1–102, 1987.
Girard, Jean-Yves, Lafont, Yves, and Taylor, Paul. Proofs and Types. Cambridge Press, 1989.
Hoare, C. A. R., 1985.Communicating Sequential Processes. Prentice-Hall International.ISBN 0-13-153271-5
Lafont, Yves, 1993. Introduction to Linear Logic. Lecture notes from TEMPUS Summer School onAlgebraic and
Categorical Methods in Computer Science, Brno, Czech Republic.
Troelstra, A.S. Lectures on Linear Logic. CSLI (Center for the Study of Language and Information) Lecture Notes
No. 29. Stanford, 1992.
A. S. Troelstra, H. Schwichtenberg (1996).Basic Proof Theory. In series Cambridge Tracts in Theoretical Computer
Science, Cambridge University Press,ISBN 0-521-77911-1.
Di Cosmo, Roberto, and Danos, Vincent. The linear logic primer.
 Introduction to Linear Logic(Postscript) by Patrick Lincoln
Introduction to Linear Logicby Torben Brauner
A taste of linear logic by Philip Wadler
Linear Logic by Roberto Di Cosmo and Dale Miller. The Stanford Encyclopedia of Philosophy(Fall 2006 Edition),
Edward N. Zalta (ed.).
Overview of linear logic programmingby Dale Miller. In Linear Logic in Computer Science, edited by Ehrhard, Girard,
Ruet, and Scott. Cambridge University Press. London Mathematical Society Lecture Notes,olume V 316, 2004.
Linear Logic Wiki

External links
Media related to Linear logic at Wikimedia Commons
A Linear Logic Prover (llprover), available for use online, from: Naoyuki T
amura / Dept of CS / Kobe University /
Japan

Retrieved from "https://en.wikipedia.org/w/index.php?title=Linear_logic&oldid=859039250

"

This page was last edited on 11 September 2018, at 09:18(UTC).

Text is available under theCreative Commons Attribution-ShareAlike License ; additional terms may apply. By using this
site, you agree to the Terms of Use and Privacy Policy. Wikipedia® is a registered trademark of theWikimedia
Foundation, Inc., a non-profit organization.