Contraction-Free Sequent Calculi for Intuitionistic Logic

Author(s): Roy Dyckhoff

Source: The Journal of Symbolic Logic, Vol. 57, No. 3 (Sep., 1992), pp. 795-807
Published by: Association for Symbolic Logic
Stable URL: http://www.jstor.org/stable/2275431 .
Accessed: 24/02/2011 00:11

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .
http://www.jstor.org/action/showPublisher?publisherCode=asl. .

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].

Association for Symbolic Logic is collaborating with JSTOR to digitize, preserve and extend access to The
Journal of Symbolic Logic.

http://www.jstor.org
THE JOURNALOF SYMBOLIC LoGic
Volume 57, Number 3, Sept. 1992

CONTRACTION-FREE SEQUENT CALCULI

FOR INTUITIONISTIC LOGIC

ROY DYCKHOFF

?0. Prologue.Gentzen'ssequentcalculus LJ,and its variantssuch as G3 [21],

are (as is well known)convenientas a basis forautomatingproofsearchforIPC
(intuitionisticpropositionalcalculus).But a problemarises:thatof detecting
loops,
arisingfromthe use (in reverse)of the rule : => forimplicationintroductionon
the left.We describebelow an equivalentcalculus,yetanothervarianton these
systems,where the problemno longer arises: this gives a simple but effective
decisionprocedureforIPC.
The underlying methodcan be tracedback fortyyearsto Vorob'ev[33], [34].
It has been rediscovered recentlyby severalauthors(thepresentauthorin August
1990, Hudelmaier[18], [19], Paulson [27], and Lincoln et al. [23]). Since the
main idea is not plainlyapparentin Vorob'ev'swork,and thereare mathematical
applications[28], it is desirableto have a simpleproof.We presentsuch a proof,
exploitingtheDershowtiz-Mannatheorem[4] on multisetorderings.

?1. Introduction. Consider the task of constructingproofsin Gentzen's se-

quentcalculusLJ of sequentsF r G, whereF is a set of assumption
intuitionistic
formulaeand G is a formula(in thelanguageof zero-orderlogic,usingthenullary
constant f [absurdity], the unary constant m [negation, with -IA =def A D f]
and thebinaryconstants&, v, and D [conjunction,disjunction,and implication
respectively]).By the Haiptsatz [15], thereis an apparentlysimple algorithm
whichbreaksup the sequent,growingtheprooftreeuntilone reachesaxioms(of
theformF - A whereA is in F), or can make no further progressand mustback-
trackor evenabandon thesearch.(Gentzen'sargumentin factwas to use thesub-
formulapropertyderivedfromthe Hauptsatz to limitthe size of the searchtree.
Dosen [5] improveson thisargument.)
Some extraeffort is requiredto ensuretermination.Suppose we use a version
of the calculus with a contraction rule: then,at every step, we may choose to dupli-
cate a formula before breaking it up. With care, one can remove the contraction
rule almost entirely,but a trace of it remains in the rule D =>: in using this rule (in
reverse) to extend the proof tree above a node labelled A D B, F G we must

Received February 15, 1991; revised July 10, 1991.

? 1992,AssociationforSymbolicLogic
0022-4812/92/5703-0003/$02.30

795
796 ROY DYCKHOFF

duplicatethe formulaA : B, generatingchildrenA B F B.A and B, F =r G.

(Considertryingto prove- (A v m A) = f withoutdoingsuch duplication.)This
duplicationforbidstheargumentthatthesize,bysomemeasure,of thesequenthas
been diminished,and hence thatthe searchmustterminate.We could, following
standardpractice,use a stack to detectlooping-but the looping testsare ex-
pensive,and complicatethetaskof extendingthetechniqueto thefirst-order case.
In the presentwork we give a sound completecalculus UT forconstructing
suchproofs,butwithno primitive ruleforcontraction:theonlyessentialchangeis
to therule -=>. This providesa simpledecisionprocedureforIPC.
Note thatour descriptionof UT as "contraction-free" is based on our viewof
thesequentcalculusas intendedto be used fromtheroot up: we have no qualms
thereforeabout havingprimitiveruleswiththe same formulaeas antecedentsin
each of twopremisses.Thus theapparentmovefrompremissesF r A and F r B
to the conclusionF - A & B does not requirea contractionto be done on the
multisetF, F containingtwiceas manycopies of everything in F as F itselfhas.
The two copies of F are, on beingobservedto be equal as multisets,just merged.
From the viewpointof Girard's linearlogic [16], thiscounts as a contraction:
fromours,concernedwithtermination of an algorithm,it does not. Both linear
logicand thepresentworkcan be seen as meansof controllingtheuse of thecon-
tractionrule,but in different
ways.
UT does not have the subformulaproperty.Nevertheless, it has it in an ob-
vious weak sense:one can workout whatformulaeare able to appear in a proof
of a particularend-sequent.

?2. Logical preliminaries.

First,we presenta completeset of sound rulesfor
thesequentcalculusU, essentiallydue to Gentzen[15]. Of variouspresentations,
such as G3 of Kleene [21, p. 481], that by Dragalin [6] suitsus best: but that
coversthemulti-succedentversionGHPC.
Sequentsare of theform

FormulaMultiset= Formula,

wherea multiset[4] is a set in whichrepetitionsare allowed,but no account of

order is taken. A, F denotes the multisetcontainingan occurrenceof A, the
remainderbeingthemultisetF. We requireno structuralrules,otherthan those
implicitin theuse of multisets(so permutations are allowed).We easilysee (using
Dragalin's arguments, forexample)thatthe rulesof weakening,contractionand
cut are admissible.These argumentsapply when the use of the axiom rule (see
below)is confinedto atomicformulaeA: itthenfollowsthatthemoregeneralusage
is also admissible.
Our presentation(Figure 1) differs
a littlefromGentzen's,but the differences
are trivialenoughto allow us the use of the name U of his system:modulo re-
namingof some variables(in the first-order case), proofsin our systemtranslate
directly intohissystemand viceversa.Note thatwe do notrequirecontraction:its
onlyessentialuse is builtinto our formulation of the rule : A. (Our reason for
usingmultisetsratherthan sets is to avoid concealinga contractionrule in the
 CONTRACTION-FREE SEQUENT CALCULI 797

Axiom A, =A f 7J= G
ABT =G --= A T => B
IA&BI, T=G- T=A&B]
v/ =
AT - =-
_ _
G _
B, T=G -= -A => B
v _ _ _ _

IA v BLTrF=-G T=A vB F
T=IA v BI
A BT=)BF
A A - = G AF = B
D _ _ :D

IA, BI ,=G |=A:D B|

FIGURE 1

notation.)Our objectiveis contractionelimination, and we shall showhow,in the

zero-ordercase,eventhisone (disguised)use can also be eliminated.
In the presentationof theserules,in each case the formulain the conclusion
whichis designatedtheprincipalformulais boxed.A similarconventionappliesin
some of theproofsbelow.
Note the occurrenceof A D B in the major premissof the --rule.But for
this,all the rules have the propertythat each premissis simplerthan the con-
clusion.We shall replacethisruleby some similarruleswhichallow thesame set
of theorems.Let UT denote the formalsystemhaving the above rules,minus
D =>, augmentedby thefourrulesof Figure2, routinely seen to be admissiblein
U. (The T is for"terminating".)

B,AT => G

C D (D D B),T =r G
:D 2>
2 (C & D) D B, = G
D B, F =- G
C D BD
:D

(C D) BF = G
D
3
v
D D BF = c DD B F= G
(C D D) D B,F = G
FIGURE 2

We regardtheformulaf as not beingatomic.Note thattherule-D -- is sound

evenif theformulaA is not atomic:whatmattersis thatwe can show thatits use
just whenA is atomicsuffices. It is easy to showthatweakeningis (still)an admis-
sible rule of thissystem:so therefore also is the rule thatfromF = G one may
inferf D B, F -- G-a rule one mightotherwiseexpectto be primitivein LJT.
Note that the firstthreeof these fournew rules are invertible-if the con-
clusionis provablein LJ,so is thepremiss.The fourthrulehas thepropertythatif
theconclusionis provablein LJ,so is thesecondpremiss.
Our argumentswill use various formsof induction,one of whichis on the
798 ROY DYCKHOFF

"size" of sequents.We measurethesize of sequentsas follows;but firstwe define

an orderingon formulas.The weight,wt(A),of a formulaA is a positiveinteger
definedrecursivelyas follows:wt(A) = wt(f) = 1 for atomic A, wt(A v B) =
wt(A D B) = wt(A) + wt(B) + 1, and wt(A& B) = wt(A) + wt(B) + 2. This gives
us a well-founded orderrelation > on formulae,withA > B iffwt(A) > wt(B).
Note forlateruse that,forexample,wt((C & D) = B) > wt(C D (D :D B)).
Now treatthe antecedentformulaeof a sequenttogetherwithits succedentas
a multiset.Then,by definition,F ,, iffthemultisetzAis obtainedfromthemulti-
set F by replacingone or moreformulasof F by zero or moreformulas,each of
whichis of lowerweightthan one of the replacedformulas.This is the multiset
ordering of Dershowitzand Manna, known[4] to be well-founded if theordering
on formulaeis. Note thatin each of therulesof UT, theconclusionis in therela-
tion >> to each of theassociatedpremisses(if any).
Below,we shall talk informally about the"size" of a sequent:argumentsusing
thisundefinedconceptcan be formalisedas argumentsby well-founded induction
on therelation>>.To see how to convertthese(in thecases consideredbelow)into
argumentsbyinductionon thenaturalnumbers,see [4].

?3. Main results.The idea of the proof that UT is equivalentto U is to

consideran U-provable sequent,and use whereverpossibleone of theinvertible
rulesof U to replaceit by a simpler(withrespectto theordering>>) U-provable
sequent.In thecase thatthe finalstep of the U proofof thegivensequentis by
meansof D =>, we use one of the rules D =,,. . ., = =>4. The only difficultyarises
with the D=>1 rule, since its conclusion is not quite general enough: as well as the
implication A = B being introduced, the antecedent has to contain an occurrence
of the atomic left-subformulaA. We thereforeprove a lemma which implies that
in constructing the proof from the root up we may delay uses of this rule (with
A =DB the principal formula,A being atomic) until the atom A occurs on the left
of the sequent to be proved.
DEFINITION. A formula of the formA D B is said to be awkward if A is atomic
(recall that f is not regarded as atomic). A multiset F is said to be irreducibleif it
contains no pair A, A D B where A D B is awkward, and neither absurdity nor a
conjunction nor a disjunction. A sequent is irreducibleif its antecedent is irreduc-
ible. An U proof is clumsyif the principal formula of the final step is on the left
and is awkward; otherwise it is sensible.
LEMMA1. AnyU-provableirreducible sequenthas a sensibleproof.
PROOF. Otherwise, some U-provable irreducible sequents have only clumsy
proofs. Consider, among all proofs of all such sequents, one of the shortest(where
"length" of proof is measured along the leftmostbranch): let it be H, with end-
sequent F r G. So, the final step of H has an awkward principal formula A D B
on theleft.Then F =A B,F' and F is irreducible(and so A is not in F'). So the
proof looks like

(1) AF = A,F B' F= G

|A,: B 5F t= G
 CONTRACTION-FREE SEQUENT CALCULI 799

Since A does not occur in F', and cannot be identicalto A B, the proofH' is ,
nontrivial.Its end-sequenthas irreducibleF as its antecedent,so, by theinductive
assumption,it has a sensibleproofH"'. Considerits finalstep.Since A is atomic,
thisstepmustbe by *=>,where* is one of thelogicalconstantsf, &, v or v. Since
F is irreducible, ,
* mustbe D. Let D E be theprincipalformulaof thefinalstep:
sinceH"'. is sensible,D is not atomic.The proofof F G now looks as follows:
r

HO HI
AA DBDD EF"1=>D E5 ADBF"1=>A H"1
A, B1D D,E15F if-D A B, D AEF" = G
(2)
(2) AB DBE| ,DinDEF"
:, Bl5D,: E F1=>G
rmpG

J~~~A

and we permutethisinto

HO AB, E F =AA BE F"= G

A BD D EXF =r D EifA MB|,F - G

BID G
F
A3D D E|,F"

whereH2/E,B, F" G is routinelyderivedby substitutingE forD E at appro- ,

,
--

priateoccurrencesof the latterin H"/B,D E, F" G, dealingwiththe lowest

principaloccurrenceby discardingthemajorpremiss.
Thus, we have founda proof (3) of F G in which the principalformula
,
(D E) of thefinalstepis not awkward,as required. D
REMARK. Using thesame methods,it is not hard to showthatany LJ-provable
,
sequent,not havinga pair A, A B of antecedentformulaewithA atomic,has a
sensible proof. This result is stronger than we need for Theorem 1 below, but is
required when extendingthe ideas to the first-ordercase. The lemma could, in part,
be extracted from Lemma 7 in [22]. One could also go furtherand make all the
subproofs sensible.
LEMMA 2. KjFU, (C D D), B - C, DDif LJFD D, B o C D.
PROOF. Trivial [34]. E
THEOREM 1. The systemsLJ and LJT are equivalent.
PROOF. As noted earlier,it is routine to show that any sequent provable in UT
is provable in LJ, using the admissibilityof cut, contraction,and weakening in the
latter.
The important part is the converse. We argue by induction on the "size" of the
sequent being proved (not on the structureof the proof).
Suppose F-Lj T G. Several cases arise:
(a) F contains f: then there is a trivial proof (by f=i) of F r G in LJT.
(b) F contains a conjunction A & B. Let F =_ A & B F'. Then A, B, F' G is -

also provable in LJ. By the inductive hypothesis,it has a proof in LJT; combine
this proof in LJT with the use of the rule &=>, proving F e G in LJT.
(c) F contains a disjunction A v B: we deal with this in a manner similar to (b).
800 ROY DYCKHOFF

(d) F containsa pair A, A D B whereA is atomic.Let F A, A D B, F'. Then

A,B, F' => G is also provable in U, since LJB A : B. By the inductivehy-
pothesis,it has a proofin LJT; combinethisproofin UT withtheuse of therule
a 1, proving F G in UT.
(e) Otherwise,F is irreducible.By Lemma 1,F r G has an U proofwherethe
principalformulais eitheron theright,or is an implicationA : B on theleftand
A is not atomic.If the finalstep is by axiom, =>&, =>v, or => , we inductively
have a proofin UT, since in each case the premisshas (or the premisseshave)
lowersize than theconclusion.The onlydifficulty ariseswiththe D => rule.Sup-
pose thefinalstepis bymeansof thisrule,withA D B beingtheprincipalformula,
and A D B, F => G theconclusion.Considerthevariouspossibleformsof A.
(i) If A is a conjunction(C & D), then,sincein U (C & D), B iffC : (D : B),
we may also prove C D (D = B), F' = G in U. By induction,we finda proofof
thisin UT, to whichwe append an instanceof the -=2 rule.
(ii) Similarly,if A is a disjunction (C v D), we inductivelyprove C = B, D = B,
F => G in UT and append an instance of the = -3 rule.
(iii) Suppose A is an implication (C D D). So, the conclusion
A D B, F =r G

follows in U fromthe U-provable premisses

ADBF' A and B,F'-G

But, in U, A D B, F' - A iffD D B, F => A, by Lemma 2. This second sequent is

of lower size than A D B, F => G, so has, by the inductive hypothesis,a proof in
UT. Similarly,so does B, F' G. Combine these with the use of the rule = 4.
(iv) Suppose A is absurdity, f. Since f = B is vacuously provable, A = B,
F => G iffF = G in U. By the inductive hypothesis,the latterhas a proof in UT.
Combine this with the use of the weakening rule in UT. R
Since the rules of cut and contraction are admissible in U, these rules are
admissible also in LJT. (A direct proof of cut-eliminationfor UT seems difficult.)
The system UT of rules, treated in reverse as problem reduction rules, is ter-
minating: any search for a proof will terminatein either success or failure.There
is no need to use a stack and check for looping: for each rule, the move from
conclusionto premissesreplacesa sequentbyzero or moresequentsof lowersize,
where "size" is as explained in ?2. So we have a simple decision algorithm for
zero-order intuitionisticlogic. (In contrast to the case of classical logic, the order
in which we choose to use these rules is important: some choices may have to be
undone. Tennant [31] has described some heuristicsformaking such choices.)
Note that, with minor variations, this technique applies also to minimallogic.

?4. Contraction-freemulti-succedentcalculi. In practice, a multi-succedentcal-

culus is convenient for implementation: one can then share code easily with an
implementation for the classical case, and also postpone decisions about which
disjunct (of a disjunctive formula on the right)should be chosen.
Dragalin [6] gives, as noted above, a convenient formalisation of such a cal-
culus, GHPC. It looks as follows; the asterisks are to suggest "zero or more
 CONTRACTION-FREE SEQUENT CALCULI 801

formulae"on theright:

Axiom* AF = AA f F =r a

& >* AB,

AAF AA 5* F A, F
Fr BA
JA&BI,|A, BlsA
&
IA& _BIL
F jA&Bj,

V AAF=A BF =A * F ABA
LAv;B|,F za= A F='v B, A

,
ADB,=> A BF =A -, AF = B
[ADBIF=>A F=-A BINA
FIGURE 3

theaxiom*ruleto atomicA, and thenshowsthatthemore

(In fact,[6] restricts
generaluse is admissible.)The rulesare valid withrespectto U, ifwe considerthe
meaningof a sequentwithseveralformulaein the succedentto be that of the
sequentmade by disjoiningthose formulaetogether.Note that all the rules are
invertible, foremptyA. [6] shows
exceptforthelast two: but thelast is invertible
thatweakening,contractionand cut are admissiblerules.
Now, : =>* is the onlyrulewe have to change:we do itjust by changingall
occurrencesof thegoal formulametavariableG in thenew rules( D ... -4) 5

of UT intotheformulamultisetmetavariableA.
The newrulesare now as follows:

A
:DABA9T= [Abeingatomic]

D>
C (D D B), F= A
2
(C&D)DBF =A
,DB, C D =DB,F -> A
(C V D) DB,F = A
DDBF=C DD BCF = A
(C zDD)DBF= A
FIGURE 4

Our argumentsof ??2 and 3 now go throughwithno essentialchanges.We

thusobtaina contraction-freemulti-succedent logic(with
calculusforintuitionistic
no contractionhiddenin the DD- rule).
Note that,in contrastto the presentationsin, forexample,[7] and [35], the
rule v =* of GHPC has no occurrenceof zAin the succedentof themajor pre-
miss.A variantof therulehavingsuch an occurrencewould be admissible,but is
not necessaryforthe translationof LJ proofsinto the multi-succedentcalculus
GHPC. However,thereis an advantagein havingsuchan occurrenceof A: therule
802 ROY DYCKHOFF

is theninvertible.
On theotherhand,theargumentabove would thenno longerbe
correct: the difficulty
would be in replacing the sequent (C D D) ,
B, F => C, D,
A by D, B, F' C D, A in (the multi-succedent variant of) the proof of
Theorem1. So, our argumentis done in thecontextof Dragalin's systemGHPC
ratherthanin thatof [7] or [35].
However,thefollowingruleis admissible,and strengthens
therule D =
DDB,F =C DDA B F= A
(C D D)B, F= A
Using it ratherthan the weakerrule =}4* is therefore permitted, but not oblig-
atory.Unfortunately, thisruleis stillnotinvertible,
exceptwhenA is absurdityor
empty.Which of the rules D -* and v =>4**should in practice be used is unclear.

?5. Naturaldeduction rules.Gentzen'sformulation of LJallowsforthestraight-

forwardtranslation(cf.[1], [9], [16], and [29]) of LJ proofsinto NJ deductions.
For example,an instanceof therule&=- is translatedintoan instanceof therule
&E (in theformthatfromA & B we mayinferbothA and B). An instanceof the
rule D - translatesinto an instanceof the rule =DE (modus ponens)as follows:
HO Hi
A =BA, BAB.F
=} G

is translatedto thedeductionhavingA DB and theformulaein F as assumptions,

and G as conclusion:sincefromA DB and F we have (in the translationof HO)
deducedA, we maywriteB underthepair A DB and A, appeal to modus ponens,
and fillin thegap betweenB (withF) and G by enteringthetranslationof 1H1.
The new rulesof LJT correspondto thereplacement, in NJ,of modus ponens
by thefollowingrules:
ADB A
B [A is atomic] MP1
(C & D) B MP2 (C v D)B (C v D)B
MP3a MP3b
CD=)(DDB) CD=)B DD=)B

(C D)DB CD[D B] MP
B

FIGURE 5

,
wherethe parenthesised[D B] in the rule MP4 indicatesits dischargeas an
assumptionfromthesecondpremiss.
It is of interestto see how thedeductionsgot bythismeanscan be transformed
intostandardnaturaldeductions.Abramsky[1] givesa convenientformulation of
thestandardtranslationof sequentcalculusproofsintonaturaldeductionproofs.
Adoptingroughlyhis notation(in his ?2.2),and introducing appropriatenewcon-
stantsapply-atom,curry,left,right,and apply-imp,the new naturaldeduction
 CONTRACTION-FREE SEQUENT CALCULI 803

rules(withannotationsconnotingproofs)look like

p:A D B q: A is atomic]MP
apply-atom(p,q): B
p:(C&D)DB MP2
curry(p):C D (D D B)( D
vD
lef(p)C
D
-B MP3a
p:(C v D) =DB MPb
left(p): CDBP3aright(p): DD)B
p:(C D) D B q(x):C DD[x:D DB] MP
apply-imp(p,q): B
FIGURE 6

The appropriatelabellingrulesforthenew sequentcalculusrulesare then

x: B,a: A, A => g: G [ A beingatomic]

p :(A D a:
B), A, A [apply-atom(p,a)/x]g: G
x:(C D (D D B)),F = g:G
D
D-2 p: ((C & D) B),Fr [curry(p)/x]g:G
x:(C D B), y:(D D B), F r g:G
p: ((C v D) D B), F g: G
right(p)/y]
- [left(p)/x,
z:(D =DB), F = q(z):(C =DD) x:B, rF= g:G
D 4
D
F
p: ((C D D) B), - [apply-imp(p,q)/x]g: G
FIGURE 7

By thismeans,we may label UT proofswithtypedlambda calculus termsas

witnesses:thesetermscan be traversedto obtainnaturaldeductions(in NJT). To
to use thedefinitions
obtainan NJ deductionit suffices
curry(p)=def AX- Ay.apply(p,<x,y>)
left(p)=def x. apply(p,inl(x))
right(p)=def Xx.apply(p,inr(x))
apply-atom(p,q) =def apply(p,q)
apply-imp(p,q) =def apply(p,q(Ay.apply(p,Ax.y)))
thedeductionsof NJT intothoseof NJ.(These willnot necessarilybe
interpreting
normaldeductions,butcan be normalisedin theusual way.)
Not everyNJdeductioncan be obtainedin thisway,evenallowingforstandard
conversions.ByKonig'slemma,sinceUT is finitely branchingand proofsofa given
formulaare of boundeddepth, any formula can only havea finitenumberof proofs
in UT. But the formula(p D p) D (p D p), where p is many
an atom,has infinitely
804 ROY DYCKHOFF

distinctdeductionsin NJ: the Churchnumerals(forp). (In fact,in UT we can

onlyconstructproofsrepresentingthenumeralsforzero and one.) This limitation
encouragesus to look at ways of
since functionalprogramming
is regrettable,
generatingproofsmechanically.

?6. First-order calculi. These ideas extendto some extentto the

intuitionistic
case. It is easy to handleformulaeof theform(]x)A(x) a B, usingthe
first-order
equivalence(provableevenin minimallogic)
(]x)A(x) D B * (Vx)(A(x) D B),
wherethebound variablex is notfreein B. But formulaeof theform(Vx)A(x) D B
An earlyversionof thispapergaveappropriatesoundrules,which
are problematic.
are completein a weak sense: forany provableformula,a suitablechoice of a
parameter(similarto the Q-depthparameterin [12]) allowed a successfulproof
search. We referto Hudelmaier'sunpublished[18] forfulldetails of a similar
approach.

?7. Implementation. The systemUT of ruleshas been implemented in Prolog

as partof theMacLogic system[8], usedin teachingvariousfirst-orderlogics.This
systemincludessome theoremprovers(forclassical,intuitionistic, and minimal
logic),intendedto be fastat solvingproblemsbutallowedto be slowerwhenfailing,
in orderto warnstudentsof unsuccessful tacticsforproblemdecomposition.

?8. Applications.Pitts [28] reportsapplications of UT to show that, in

intuitionisticlogic,quantificationoverpropositionalvariablescan be modelledin
whatwe call thezero-ordercalculus,withniceapplicationsto thetheoryof Heyting
algebras.
Hodas and Miller [17] illustratethe use of UT in a generictheorem-prover
based on linearlogic,encodingUT in about 15 linesof Prolog.
Tennant[31] describesa theorem-prover forminimalzero-orderlogic(withan
extensionto intuitionistic relevantlogic in mind),where the looping tests are
replacedby use of a naturaldeductionversionof therulesof LJT.He reportsthat
thisled to a fourfoldincreasein performanceon a widerangeof difficult
problems.

?9. Relatedwork.Vorob'ev[33], [34] describeda decisionalgorithmforIPC

based on similarconsiderations.The presentarticlemay be regardedin part as a
restatement ofthisrelativelyancientSovietwork:itis offeredhoweveras a clarifica-
in the the is
tionand simplification, knowledgethat technique nowbeingreinvented
and exploited.The sequentcalculus lyingbehindVorob'ev'salgorithmin [34] is
concealed by the pre-processingof sequents into a normal form(using the
distributivelaws);hisalgorithm also takesadvantageoftheequivalence(fornegated
goals) of theintuitionisticdecisionproblemwiththe classical one. See [25] fora
summaryof some of therelatedSovietwork.
Hudelmaier[18], [19] inventedthe same calculus as LJT. His argumentis
similarto ours,but (like Vorob'ev)avoids theuse of multisetorderingsin favour
of constructivetechniques,fromwhichone can extractexplicit(but useless)theo-
reticalbounds on therelationshipbetweenthedepthsof the U proofand of the
 CONTRACTION-FREE SEQUENT CALCULI 805

UT proof.He has refined itintoan O(nlog n)-SPACE solution[20] to thedecision

problemforIPC. (From Statman'swork,thisproblemis knownto be PSPACE-
complete.In theworstcase, theobvious algorithmbased on UT uses space expo-
nentialin thesize of thesequentbeingproved.Our own viewis thatthisis not a
seriousproblemin practice.)
Lincoln et al. [23] reporton the use of a calculus IIL* nearlyidenticalto
LJT, fora translationof intuitionistic sequentsto linearlogic sequents(in the
calculusIMALL): thisdiffers from Girard's translation[16] in notusingthemodal
operators(!,?), butstillpreserves provability. IIL* is restricted to theimplicational
fragment of intuitionisticlogic, and differsfrom UT in the rule forintroduction of
theimplicationA v B on the left when A is atomic: rather than allowing thisonly
whenA is alreadyin theremainingcontextF, itjust requiresthederivability of A
fromF. IIL* is obviously sound, and itscompleteness follows triviallyfrom thatof
UT (but not conversely). For automated proof search, we prefer the use of LJT,
sinceit avoids workdone trying to deriveA in favourof waitinguntilitis obvious:
butIIL* is theright calculus to use forthetranslationto IMALL.
As a solutionto the looping problem, Van Gelder[32] proposesthetortoiseand
haretechnique:this requires maintenance of a stackof subproblemsbeingsolved,
and two pointers thereinto,which move at different speeds:if thereis a loop, then
thepointers will eventually point at different occurrences of thesame problem.We
have not examined this approach in any detail, believing it to have similar
disadvantages to the maintenance of a stack with a more expensive and more
frequent check, but withearlier detection of looping.
Gabbay [14] proposes the boundedrestartrule: this requiresthat a history
be keptof the search,but allows thatwhenan atomicgoal is to be proved,one
may restartthe searchat pointsafterearlieroccasions when the same goal was
investigated.
Slaney[30] reportsthathistheoremprover(forminimallogic)firsttriesto apply
the D .-rulewithoutkeepingtheprincipalformulain theantecedent, and onlyon
failuretriestheproperrule.He reportsthat"it uses a simpleloop detectorwhich
in practicedoes not seemto slow it down much,thoughon reallylargeproblems
it probablywould be expensive".
One possible decisionmethodforIPC is to use the Gbdel-McKinsey-Tarski
translationof intuitionistic problemsintoS4, and thento use a decisionprocedure
forS4. Fitting[10], [11] arguesthatalthoughone can in principleuse a periodicity
testas part of such a decisionprocedure,in practice"such a testwould be quite
expensive",and he therefore recommendsuse of a modal depthcounter.Wallen
[35] describes(forzero-orderS5) how to calculatethe multiplicity (essentiallyan
upperbound on thenumberof timesanyformulamayneed to be duplicated),and
commentsthatsimilartechniquescould be developedforthe othermodal logics.
It is not clear how effective such a multiplicity is at constrainingthe search.
Franzen[13] showshow to calculatesuchmultiplicities forIPC; buthe abandoned
thisapproachin favourof a technique(implication-locking) of retaining implicative
formulaein themajorpremissof the D - rule,but inhibiting theiruse untilmore
information is available.Tennant[31] describesa similartechnique,fettering.
Beeson[2] describesa (first-order) theorem-prover GENTZEN, implemented in
Prolog,withloop-checkingto avoid nontermination, and discussesrelatedwork.
806 ROY DYCKHOFF

to see whethera combinationof therulesforUT with

It would be interesting
themItrixmethodsof Bibel,Andrewsand Wallen [35] would be effective.
Other authorsuse techniqueswhichavoid the problemby othermeans: for
example,recentSovietwork[24] has beenbased on pre-processing to normalform,
and thentheuse of a formof "resolution".Such techniquesare outwiththescope
of thispaper'sconsiderationof techniquesforcontrollingor avoidingduplication.
Note also thatthereis no connectionwiththeworkof Dardzhania [3], or that
of Ono and Komori [26], on variantsof intuitionisticlogic withno contraction
rule:theseweakenthelogic,ratherthanreformalise thesame logic.

?10. Conclusion.We have shownthecorrectness and completenessof a variant

of Gentzen'scalculus U forzero-orderintuitionistic
logic,havingno contraction
rule and so having good terminationpropertieswhen viewed as a systemof
problemreductionrules.This can be used as thebasis fora simpledecisionproce-
dure forthislogic,which,in contrastto the directuse of U, needs no testsfor
looping. Proofsobtainedby thismeans translateeasilyinto naturaldeductions.
The same variationcan be appliedto twomulti-succedent calculi.

?11. Acknowledgements. Thanks for helpfulcommentsare due to Gianluigi

Bellin,Tony Davie, Dov Gabbay, JorgHudelmaier,Mike Livesey,Neil Leslie,
MichelLevy,Dale Miller,BirgitMoser,AndyPitts,AndreScedrov,Neil Tennant
and LincolnWallen.Special thanksare due to theauthorsof thevariousunpub-
lishedpapersand reportslistedbelow,and to GrigoriMintsforpointingme to the
Sovietliterature.Per Martin-Ldfgentlydrewtheduplicationproblemof U to my
attention,as an issue whichtheoremprovers(such as minein 1986) oftenignored.

REFERENCES

[1] SAMSON ABRAMSKY, Computational interpretationsof linear logic, technical report DOC 90/20,
Imperial College, London; Theoretical Computer Science, to appear.
[2] MICHAEL BEESON, Someapplications of Gentzen'sprooftheoryinautomateddeduction, Extensions
of logic programming workshop,1989 (P. Schroeder-Heister, editor), Lecture Notes in Computer Science,
vol. 475, Springer-Verlag,Berlin, 199 1, pp. 101- 156.
[3] G. K. DARDZHANIA, Intuitionistic systemwithoutcontraction, Polish Academyof Sciences,
Institute
of Philosophyand Sociology,Bulletinof theSectionof Logic, vol.6(1977), pp.2-8.
[4] NACHUM DERSHOWITZ and ZOHAR MANNA, Proving terminationwithmultisetorderings,Communi-
cationsof theACM, vol.22(1979),pp. 465-476; Automata,languagesandprogramming:
sixthcolloquium
(1CALP '79), Lecture Notes in Computer Science, vol. 71, Springer-Verlag, Berlin, 1979, pp. 188-202.
[5] KOSTA DOSEN, A note on Gentzen's decision procedure for intuitionisticpropositional logic,
Mathematische
Zeitschriftifir Logik undGrundlagen
derMathematik,
vol. 33 (1987),pp. 453-456.
[6] ALBERT GRIGOREVICH DRAGALIN, Mathematical intuitionism-introduction to proof theory,
Translations of Mathematical Monographs, vol. 67, American Mathematical Society, Providence,
Rhode Island, 1988.
[7] MICHAEL DUMMETT, Elements of intuitionism, Clarendon Press, Oxford, 1977.
[8] Roy DYCKHOFF, NEIL LESLIE and STEPHEN READ, MALT St Andrews (MacLogic-a proof
assistant for first-orderlogic), Computerised Logic Teaching Bulletin,vol. 2, no. 1, St Andrews University,
1989, pp. 51-60.
[9] AMY FELTY, A logicprogram fortransforming sequentproofstonaturaldeduction Extensions
proofs,
of logic programming workshop,1989 (P. Schroeder-Heister, editor), Lecture Notes in Computer Science,
vol. 475, Springer-Verlag, Berlin, 1991, pp. 157-178.
[10] MELVIN FITTING, Proof methods for modal and intuitionistic logics, Reidel, Dordrecht, 1983.
[11] , First-order
modaltableaux,Journalof Automated
Reasoning,vol. 4 (1988), pp. 191-213.
 CONTRACTION-FREE SEQUENT CALCULI 807

[12] , First-orderlogicand automatedtheorem proving,Springer-Verlag, Berlin,1990.

[13] TORKEL FRANZtN, Algorithmic aspects of intuitionistic propositionallogic I & II, Technical
reportsR87010B and R89006,SwedishInstituteof ComputerScience,Kista,Sweden,1987& 1989(see
also [13a]).
[13a] DAN SAHLIN, TORKEL FRANZtN and SEIF HARIDI, An intuitionistic predicatelogic theorem
prover, Journalof Logic and Computation, to appear.
[14] Dov M. GABBAY, Algorithmic proofwithdiminishing resourcesI, Computersciencelogic:pro-
ceedings,Heidelberg,1990 (E. Burger,H. Kleine Buning,M. M. Richterand W. Schdnfeld,editors),
LectureNotes in ComputerScience,vol. 533,Springer-Verlag, Berlin,1991,pp. 156-173.
[15] GERHARD GENTZEN, Thecollectedpapers of GerhardGentzen (M. Szabo, editor),North-Holland,
Amsterdam, 1969.
[16] JEAN-YVESGIRARD, Proofsand types,CambridgeUniversity Press,Cambridge,1989.
[17] JOSHUA HODAS and DALE MILLER, Logic programming ina fragment linearlogic,
of intuitionistic
Sixth annualIEEE symposium on Logic in ComputerScience: proceedings,Amsterdam, 1991, IEEE
ComputerSocietyPress,Los Alamitos,California,1991,pp. 32-42.
[18] JORG HUDELMAIER, A Prologprogramfor intuitionistic logic,SNS-Bericht88-28,Universitit
Tiibingen,Tubingen,January1988.
[19] XBoundsfor cutelimination inmntuitionistic propositional logic,Dissertation,Mathematics
Faculty,UniversititTubingen,Tubingen,1989;also to appear in ArchiveforMathematicalLogic.
[20] JORG HUDELMAIER, A decisionprocedurefor intuitionistic propositionallogic, submittedto
Journalof Logic and Computation.
[21] STEPHEN C. KLEENE, Introduction to metamathematics, North-Holland,Amsterdam, 1964.
[22] , Permutation of inferences in Gentzen'scalculiLK and LJ, Twopaperson thepredicate
calculus,Memoirno. 10,AmericanMathematicalSociety,Providence,Rhode Island, 1951,pp. 1-26.
[23] P. LINCOLN, A. SCEDROV and N. SHANKAR, Linearizingintuitionistic implication,Sixth annual
IEEE symposium onLogic inComputer Science:proceedings, Amsterdam, 1991,IEEE ComputerSociety
Press,Los Alamitos,California,1991.pp. 51-62.
[24] GRIGORI MINTS, Gentzen-type systems and resolution rules.Part I: Propositionallogic,COLOG-
88: proceedings, Tallinn,1988(P. Martin-Lofand G. Mints,editors), LectureNotesin ComputerScience,
vol. 417,Springer-Verlag, Berlin,1990,pp. 198-231.
[25] , Prooftheory in theUSSR 1925-1969,thisJOURNAL, vol. 56 (1991),pp. 385-424.
[26] HIROAKIRA ONO and YUICHI KOMORI, Logics without thecontraction rule,this JOURNAL, vol. 50
(1985),pp. 169-201.
[27] LARRY PAULSON,personalcommunication, concerningIsabelle (ComputerLaboratory,Cam-
bridgeUniversity), Cambridge,January1991.
[28] ANDREW PITTS, On an interpretation of secondorderquantification in first-orderintuitionistic
propositional logic,thisJOURNAL, vol. 57 (1992),pp. 33-52.
[29] DAG PRAWITZ, Naturaldeduction, Almqvist& Wiksell,Stockholm,1965.
[30] JOHN SLANEY and ROD GIRLE, Tableau and sequentcalculus methodin minimallogic theo-
remproving,ANU-Fujitsu-jointworkshopon logic and computation,Canberra,June 1990 (unpub-
lishedtalk).
[31] NEIL TENNANT, Computationallogic, seminar, A.I. Department Edinburgh,University,
Edinburgh,May 1991.
[32] ALLEN VAN GELDER, Efficient loop detectionin Prolog usingthe tortoise-and-hare technique,
Journalof Logic Programming, vol. 4 (1987),pp. 23-3 1.
[33] N. N. VOROB'EV, The derivability problemin theconstructive propositionalcalculuswithstrong
negation, DokladyAkademiiNauk SSSR, vol. 85 (1952),pp. 689-692. (Russian)
[34] , A new algorithm for derivability in the constructive propositionalcalculus, Trudy
Matematicheskogo InstitutaimeniV. A. Steklova,vol. 52 (1958), pp. 193-225; Englishtranslationin
AmericanMathematicalSocietyTranslations, ser.2, vol. 94 (1970),pp. 37-71.
[35] LINCOLN WALLEN,Automated deduction innon-classical matrixproofmethods
logics:efficient for
modaland intuitionistic logics,M.I.T. Press,Cambridge,Massachusetts,1990.

DEPARTMENT OF MATHEMATICAL AND COMPUTATIONAL SCIENCES

ST ANDREWS UNIVERSITY
ST ANDREWS, FIFE KY16 9SS, SCOTLAND

E-mail:rd~cs.st-and.ac.uk