!
v\tur,'ii \'il \
RESOLUT|ON
METHOD,
MFCS
-2010,
Mihaieta Lr.rpea
l$v-**gEess
fi;f :'es*Ea.cEi*: ir.y tu avolu iiie u*rivaiicn ai r*cundarir
dF=iG
irrelevant elauses in arder to obtain the empty ciause.
1. tevef saturation strategy: This initial level contains the initiaf set of ciauses. We compute all the resolvents on a leve!, we add thenr to the
current level, and then we compute the next level. We continue until we
obtain the empty clause or we cannot derive any more resolvents.
Algoritkm !evet_saturstiorc-strctegjt :
Input: S- the initial set of clauses; Output: message "S-inconsistent" or "S*consistent',
ll we generate the sequence,s'0,
i.r
s1
,...,,st containing sets of clauses
(levets)
S0
:S; k:0;//initial
level
do
{ k:k- }
,Sfr =
{Res{Cl, C2}lC1e .g0 r-,.g1 u... u,gA-1 ,CZ eSrt-]
sA = sA
- (so ,.; sl u... , sA*])
ll 'we eliminate the resoivents obtained in the eunent levei but rvhich ll appear already in the previous levels
i untii (i: e Sfr or S k = A).
iflr e Sfr)
then write "S- inconsistent,'
else write "S - consistent"llno more resolvents can be derirted
end*if End alsorithrn
2. Deletian strategy:
i;y *ifi*i
the resolvents that are tautologies o!" are subsumeci
r-1
,*i,Cur:ril, i;i iilr:::iij"L;-: i, *;ciu'--sr a"-+ ci'r-li;;f;*i*C, li-':ii !'niii
"'i Lc
used fui-ther
in the
resolution prCIcess because they will produce
redundant clauses.
Example 4: Llsing the level saturatioti strategy combined with the deletion strateg! check in- consisiency/inconsistency of the set of clauses S:t p v q ,-p v q ,/ -t' .-Q'l i' ].
We buiid the sequence
,S0 , Sl
,
52 ,. . . representing tire levels of resolvents:
t0:g:{C
l-- p v
,CZ:-pv
Q
v -:t' ,C3:--q v r }
- initial set of clauses
The frrst level of resolvents is generated:
C4:Re s u{C7,C7}:
Y'--t'
C5:R.esr(C1,C3):pvr
C5:Resn(C2,C3): -''p v '-t''v r' - tautoloey
C7:Re sr{C2,C3}:---p'qt' -Q - tautolo-o1'
wiil not be inclucied in the first level' The first level of resolvents is:51:1C4: Q Y '-7',C5: p v t \
C6 and C7 The second levei of resolvents is generated:
C8:Res,. (C4,C5): q v P :C\
C9:R-e sr{,C4,C3):t v --t{ - tautology
C10:Res,. (C4,C3)=
"/
-Q
- tautologY
Cl l:Res" (C5,C2): q v r Y ---,t - tautolog}'
CIZ:Res,.(C5,C2i: -.Pv PY q - tautologY
tlie The clauses C9,C10.C11,C12 are tautalogies, and CB belongs to initial set of ciauses as C 1, so they will not be introduced in the seccnd trevel'
S7:Z
rvith the meaning that no more resolvents can be generated'
set' The emptv clause cannot be deriyed from S and thr.rs S is a consistent
3' Ffie set-of support
strategy. avoid to resolve two clauses belonging to a c*i*'si:t*ni slih:$*i ;;f ii'l'= it-t:iiai set i:f t:l*i*s.:s" L:e*ai_*** ihe r*soivenis derived from a consistent set are irrelevant in the process of derivlng
n.
This strategy was inspired by the fact that usually the set of premrses (hypotheses, facts) of a deductlon is consistent, and resolving
two
clauses from this set wlll not help us
(inconsistency).
io derive the enrpty ciause
Definition: Let S be a set of clauses A subset y of S is called the support sef of s, if the set s-y is consistent. The set_of-suppart
resolutian is the resolution of two clauses that do not belong both of them to the set S-y.
Ex*rnptre
5:
IJsing the set-of-supporl strategy pro'e that:
p-+(q-,"), rrxslt, Ul:p-+{q}r),
IJZ:'l'nS-+/,
\J3:?,t
u_>sn_.,1 l_
q)__Lt.
--) s A --t
Y:pAQi--Lt
ClfFrul):--py -g v i., C1:_-p y -=Q v t"
C}.{F(U2): ----tt" V --rs v t . C2: --r" v -r,e V I ClrFru3):-?,v (sn--r) =(--uvs)n ("-uv--r), C3:---Lt,vs and C4:--uv --.t CNF(-V): p q,^, r " and C5: p, C6: q . C7:u
5= { C 1,C2,C3,C4.C5,C6,C7}
we choose Y:{c5,c6,c7} as a suppoc set coffesponding to tire .negation of the conclusion of the deduction. s-y:{cr,c2,c3,c4} is a
consistent set corresponding to the premises of the deCuction. This srrareg)/ avoids to resolve two ciauses belonging both to the consistent set s-y.
c1 and c2 resolve upon the iiteral r, but they belong both to S-y.
The derivation of the empry clause using the set-of-support strate_ey is:
CI:---py -tQy t.,
CFp"
C2:---r V=s V l.
C3:--zlvs, C4:--uv--l
-c-.1 . i-; -!4 C8:Resp (Cl,C5): -=Q'v t.
C9:Resn (C8,C6,):r
Ci0:Re sr(C9,C2):-.,s v
C1
i:Re
su(C7.C3 S: s
Ci2:R.e su(C7,C4):--l
C13:Res, (C10,C1 1):r
C
l4:Ress
(C 1 3,C
IZ'y: o.
We conclude that S is an inconsistent set and according to Theorem
deduction holds.
the
Refinements of resCIlutlon (impose restrictions on the clashing clauses to make the resolution process more efficient). lack resafutiGF-c, ft'near
resofuffo n, sefftantic. resaIutian"
refinements and strategles of resoEution preserve the sour.ldness and completness propertles.
AEE
The problem is if we can combine these sirategies and reflnements.
All
combinations
of these
soundness property. The
cornbinations.
refinements and strategies preserve the completness is not preserved in some
Incompletness: the initial set of clauses is inconsistent, but the ernpty
clause cannot be derived because there are too many r-estrictions irnposed by those strategies and refinemenis.
general resoiution+ deletion strategy = sound and comptete generaf resolution + set-of-support sirategy= sound and complete
genera l resol utio n +defetion strateg v+set-of-su pport strateg y=sou n d a nd con'lpiete
2. Lock resoEuisn {Boyer,
',:;:,;ii *i:ci-ii'*r'i*L: with an integer.
1rfl
lg7i}
a *ei r:f cruuses is e rbitra;"y ii':dex*c
i;;le'';rr; i;-cr-"1
Resfrietian: the titerals resolved upon must have the snnallest rndrces
in their clauses.
The literals from resolvenis inherit the indices fronn their parent clauses.
!f a resolvent inherits the same literal wiih two different indices, will
keep the literal with the smallest index
It is very efficient and easy to implen'lent. We must combine lock resoEution wEth the levei saturation strategy En order to check alE
possrbEe ways of derEvinE the empty clause.
Campfetness.' Let S be a set of clauses having each llteral arbrtrary
indexed with an integer. lf S is inconsistent, then there is lock derivation
of the empty clause frorn S.
Scundness: Let S be a set of clauses having each literal arbitrary
indexed with an integer. [f from S the empty clause can be derived
usinq iock resolution. then S is inconsistent.
ExampEe
acco!^ding
6:
The set S={pvQ,-l}vQ,py-Q,-pv-q} is inconsisteni
1.
to Example
We index the literals of S as follows:
C1:11) P't
q21Q
C2: G)-,Pv (4)q
C3:(5) P'r(O) -,q, C4: J)--Pv (8)--q
C5: Reso(C1,C2)= (2)q,
C6= Resp(C3,C4)=
q6,t--'Q
C7:
Reso(C5,CO)=
11
RESO|-UTION
h/IETHOD,
MFCS -zUA
lr4ihaiela Luoea
Loek resoir.ition +deletion strategy is sound but s.lot cCImBEete. Anuin*r ;ficexlitg oi ihr* i;l*r'a;ir: *r I.
C1=
ql pv glQ ,
C2= G)-.|''v
elq
,C3=
1s;
pv
i61---Q
C4= (8) --Pv
e)--q
Using the restriction inrposed by lock resolution we can derive only the resolvents C5 and CG which are tautologies.
CS=Reso(Cl ,C4)= e)
pv {B)---p
and
C6= Reso(C2,C3)= gSQv
1q--q
If we apply also the deletion strategy, and we delete these tautologies,
then we cannot continue the derivation process to obtain n. Therefore S is inconsistent but n cannot be derived. '
.
Thus, lock reso!r.rtion+deretion strategy !s not cornplete. Continuing the lock resoiution process br.lt using also the resolvents C5 and CG we obtain:
C7= Reso{C2= (3) -r,qv
{lQ
,C5= e}
pv $}---p )=
(+) Qv
$y-p
We remark that the clauses C7 and C2 are similar but not identical, the iiterals' order (provided by the inciices) differs in these clauses. --p is the is the litera! to resolve upon form c7. Thus the role of a tutofogy as a parent clause is to modifv the iiterals, order in the other parent clause.
C8= Resq(C7=
literal to resolve upon from
c2, and
()qv
(5)
(_S)--p ,C4=
(U-p't
{1.}-Q )= (8irp,
Cg= Resp(C3=
pv(6)--g ,CB)=
(q--q
,
Ci 0= Reso(C1 =
12)
p'v giq ,C9;= {2} P
C11= Reso(C10,C8;=
".
IL
RESOLUTIOI.J
fvlETF-lOD.
h4FC$
-2010.
Mihaiela Lupea
Lock resoFritio* + set=sf-support strategy
fix;+.iir;.ie,
,,'-:
Cs
son:r'ld but nat ccrnplete"
i:..;il;i:i-, . ir'*'!'{; ii :,,., cl*tir-rr-:lrcti: p -> {,q -> r},r n s -} t,u --+ sA"--rl 1- p nQ } =u.
l=,:.i,',; i,-llk
r'1
,
U1= p -+ {q -+
u/.-t'n-s-+r. U3=u-),eA-/,
V=pr.Q-)-u
CNF([-.,1)= -,p y -Q y
|,
I
Ci= -p v -Q't/
C2=
I"
CNF(U2)=
r/
V -,s../
, -t) ,
CNF{Lj3)= (-u v s) n (-a v
-r v -s r/ I C3= -er v s,
C5=
C4=
-u v -l
CNF(-V)= p
^q
^r.
p,
C6= ts, C7=u
g={C1, C2,C3,C4,C5, C6,C7}
Using the set-of support strategy we chose the support set y={C5,C6,C7i corresponding to the conclusion of the deduction.
S-Y={C1,C2,C3,C4} is a consistent set of clauses corresponding to the premises of the deduction. This strategy a\roids to resolve two clauses
belonging both to the conslstent set S-Y
We index the literals from the clauses as follows.
C1=,., -pY r2y-Qv,.r.t'
C4=
qrol
C2= tqrrvi5;-sv14;l
C5=
1r
L,J=
,
(8)
-zlv171s
-'Lt'r 1.a1-l
r;p,
C6= yDq
C7= gilu
We renrark tlrat the restriction imposed by lock resolution (the titerals with
the smallest indices in their clauses resolve) combined with the set-ofsupport strategy (avoici to resoive two ciauses belonging both to the
consisient set S-Y={C1,C2,C3,C4}) wilt biock the resolution process efid cannct be derived.
'13
RESCLUTION tu4ETl-lOD, F,4FCS
-2010.
I,/lihaieta Lupea
C1=,r, -,pY et--gY ,,r/'
C2= (6)--rv(5) -.,s"/ 1.41t ,
C3=
qS; -.,?ty 1i-,s
C4=fl A1'-u'rSgl--t
C5=r11tP,
\ 4J -
c6= (12) Q
\/i-{lll1l
lf we don't use the set-of-support strategy we can derive
r as follows:
!L-,6r-l'V,iL-JV,r.1
\v/ \il_
Lr'*-
iI
61
1-2.1V,61 rI
C8=
16;
-rvq-s; :s
v11 6;
-ar
UV-,6,:l'VrR\-// tv,/-
ur-r3t-P'vt:,-q
f1.l-
t61!'
U lu- Gt-Pv,,,lilu {8r :l't
\L)-
{ -h=
rla tY
{\.1 ,1 U I f-
r /t!^.-DVtot-tll \o,
c5=,,',,P r/\r
C12=
(8)
-u
Ul-,tr'l.l
1.42--
Thus ihe set S={C1,C2,C3,C4,C5,CC,C7} is inconsisteni and ihe deduction
p -+ tq -+ r),r n,r ->
/. u
-+
A.-t
i- p
q --:' --Lt holds.
t+
3. Llnean reseEution- Loveland '!97S
- the re**ii..iil;:.,;-, illiilC'gs i:l *
ii;";;:*i' *i"r*l
::i
"*:,:ch ;:'l*;p'-
ai ;';esi Oiid iJ1= iil
parent clauses is the resolvent derived at the previous step'
,f
,r'
I
L)^
a"
Co iop ciause
C1,C
/ ,"
r,-..C,-t,C,, central clausgs
1
c2
Bq, 81,..., B,-t side clauses
Vi = 1...., n '. C, = Re s(C,-r, B,-r)
i -tc,,-1
)/'
*u-z
.B'-'
Linear resolution
is sound and complete
according
to the
theorem:
Theorern 4: The set S of clauses is inconsistent iff
s;ifl,n'
we can combine the linear resolution wiih the
con'rpletness is Preserved.
deleiion strategy and the
implementation level: This refinement of resoiution also provides a strategy at the clause there are backtracklng atgorithm. At each iteraiion, for ihe current centrai bui the empty ciause more possible side clauses. lt all side clauses were used, was not derived, we come back to the prevrous iteraiion,... search withoui the The consisiency of a sei of clauses is pl'oved after a complete derivation of the emPtY clause'
{J
RESCLIJTION
METHOD,
MFCS
-2010.
Mihaieta Lupea
Particular cases of linear resolution.
eee;it;'ess;'ie;"gr+F;. ii-id
uq;iiiiai a;auses frare
*i ;+a*t ** i;i-l;l cieuse as *
parent clause.
input res+felfror. all side clauses are iniiial (input) ciauses.
The equivaience between unit and input resolution is expressed in the
following theorem.
Theorern 5: Let s be a set of clauses. s Ffffi'r if and oniy
if s
f-ff1,|"
These two refinements of resolution are sound, but they are nat conrpiete:
soundness: lf sS'ff'lt''''! n then s
is inconsistent;
Encocnpletness: there are inconsisient sets of clauses from which the empty clause cannot be cjerived using input or uinit resolution.
Ex: s={ pv cr,-It'r
or uinit refutation.
Q,
pv -Q,-r},u, -q} is inconsistent
br-rt
there is no input
EESQLt.lTlOhi
METi-{CD.
MFCS
-20.j0
Mihaieta Lupea
Exarnple s: using !inear resolution prove rhe inconsistencv of the set s of ::!r;i: ;*q+:s, S= iC i .',:.:, e3, il.:., C5] C1=p v q\n r, C2= *,F, C3= C4.= --r y 14), C5=-i+,
-q,
The derivation tree is:
vt=pvqvL
Top clause: c5, side clauses:c4,c
tr
,c3,cz, central clauses: c6,c7,cg,cg.
This iinear resolution is also an input resolution and unit resolution
'ii