BF02573590

Download as pdf or txt
Download as pdf or txt
You are on page 1of 28

Semigroup Forum Vol.

25 (1982) 83-110

RESEARCH ARTICLE

A NOETHERIAN AND CONFLUENT REWRITE SYSTEM


FOR IDEMPOTENT SEMIGROUPS

J. Siekmann a n d P. Szabb

Communicated b y K. Keimel

ABSTRACT: Let B be a semigroup with the additional relation

V w E B. ww = w

B is c a l l e d a band or an idempotent semigroup [3].

It is s h o w n in t h i s paper that the replacement rules


(rewrites) resulting from the axiom of idempotence:

xx ~ x
x ~ xx

can be replaced by the Noetherian, confluent, conditional


rewrites (i.e. a terminating replacement system having the
Church-Rosser-Property):

xx ~ x
cI' cI
xyz ~ xz if x = z and xy = z.

These rewrites are used to o b t a i n a unique normal form for


words in B a n d hence are the basis for a decision procedure
for word e q u a l i t y in B.

The proof techniques are ba~ed upon term rewriting systems


[7] rather than the usual algebraic approach. Alternative
and simpler proofs of a result reported earlier by G r e e n
and Rees [4] and Gerhardt[6] have been obtained.

0037-1912/82/0025-0083 $05.60
9 1982 Springer-Verlag New York Inc.
SIEKMANN AND SZABO

I. M O T I V A T I O N

1.1. C o n f l u e n t Noetherian rewriting systems [7] are of


considerable practical importance in c o m p u t e r science: e.g.
abstract d a t a types [5], automated theorem proving [14],
computer algebra [12] and u n i f i c a t i o n theory [16], [20].

Unfortunately any a x i o m set w h i c h includes the a x i o m of


idempotence could not so far be u s e d for a term r e w r i t i n g
s y s t e m due to the n o n - N o e t h e r i a n (i.e. n o n t e r m i n a t i n g )
nature of the r e p l a c e m e n t rule [8]:
x ~ xx.

Although it is not d i f f i c u l t to o b t a i n a c o n f l u e n t and


Noetherian rewriting system for the s i n g l e axiom of
idempotence (see s e c t i o n 4.4), the s i t u a t i o n changes if the
two axioms of i d e m p o t e n c e and a s s o c i a t i v i t y are involved.

Once a N o e t h e r i a n and c o n f l u e n t rewriting system for


associativity and i d e m p o t e n c y has b e e n o b t a i n e d , it is
possible to e x t e n d this r e s u l t to l a r g e r sets of axioms:
on the basis of [6], theorem 4.33 it is p o s s i b l e to code
the r e m a i n i n g equational axioms into a single one.

1.2. A c o n f l u e n t Noetherian rewriting system for an


equational t h e o r y T can be used to o b t a i n a unique normal
form for terms and is t h e r e f o r e o f t e n an i m p o r t a n t p r e -
r e q u i s i t e for a unification algorithm in T [19]. The T-
unification algorithm, where T consists of the a x i o m s of
associativity and idempotence [17] is b a s e d on the r e w r i t e s
proposed in this paper.

1.3. Let F A be the free semigroup generated by the


alphabet A and let BA, the free b a n d generated by A, be
FA/I, where I is the s m a l l e s t congruence on FA containing
the r e l a t i o n {(ww,w) lw 6 FA}. It is not o b v i o u s whether or
not two d i f f e r e n t w o r d s w I and w 2 in F denote the same
element in B A . An a l g o r i t h m based on the w o r k of G r e e n and
Rees [4] and G e r h a r d t [ 6 ] , which decides whether or not

84
SIEKMANN AND SZAB0

61] I = ~ 2 ] I ' is p r e s e n t e d in [9]. T h i s a l g o r i t h m is b a s e d


on the following observation: For a w o r d w s FA let {w} be
the content8 of w, i.e. the set of letters of A a p p e a r i n g
in w,

(RULE C) If {y} c {z} and {x} = {z} then

[xyz~=[xz~; x,y,z 6 FA

As RULE C is u s e d as the b a s i s for our second rewrite, we


give an a l t e r n a t i v e and simpler proof based solely on t h e
notions of t e r m r e w r i t i n g systems.

1.4. It s h o u l d be n o t e d t h a t the structure resulting


from associativity a n d the single rule

xx -~ x

(i.e. n o t r u l e x ~ xx) has been investigated under the


name of 'nonrepetitive words'[iO], chapter 1.8 . However
this is v e r y different f r o m the s t r u c t u r e investigated
here: for e x a m p l e there exists an infinite number of n o n -
repetitive words, whereas finitely generated free bands
have only a finite number of e l e m e n t s [4]. T h e r e is no
confluent rewriting system for n o n r e p e t i t i v e words either;
as d e m o n s t r a t e d below by c o u n t e r e x a m p l e 2.

1.5. In order t o p r o v i d e s a ~ e i n s i g h t i n t o t h e problems involved


we present a few e x a m p l e s , where e a c h of the n u m b e r e d l i n e s
(1.5.n) represents an i n i t i a l attempt, which is s u b -
sequently shown to be insufficient.

Given two words wl, w 2 6 FA l e t us w r i t e w[ ~ w 2 iff t h e y


constitute the same element in BA.

A word u v v w s FA m a y be r e p l a c e d by uvw, since it r e p r e s e n t s


the s a m e element in BA. We write uvvw ~ uvw for s u c h an
c
elementary c-transition a n d u ~-- w for a c h a i n of e l e m e n t a r y
c
c-transitions. Similarily: uvw ~ uvvw, and u ~ w is a
e
chain of e l e m e n t a r y c- or e - t r a n s i t i o n s . T h e n :

85
SIEKMANN AND SZABO

I * *
(I .5.1) W 1 = W 2 iff HV. W 1 ~-~ V and w 2 ~-~ v

But this is a very unsatisfactory situation. Because of the


e-transitions there is no unique element v and in general
it is not all easy to see whether such a word v exists or
not. Hence our intention to eliminate ~-~.
e

A word w 6 FA is terminal if ~ V . W ~-~ V. L e t T E R ( w ) be the


c ,
set of all terminal words obtainable from w by ~-~. T h e n
c
(1.5.1) could be replaced by:

I
(I .5.2) wI = w2 iff TER(W 1 ) D T E R ( W 2) #

Since TER is always finite this would solve the problem.

Counterexample 1:

Let w I = abc and w~_ = a b c b abc


Then TER(wl) = {abc}; T E R ( w 2) = {abc b abc}
I
and hence TER(wl) n TER(~ ) = ~, thus Wl% w 2 .
However:

w I = __abc ~-~e a b abc ~e ab abc babc ~c abc b abc = w2

and hence w I ~ w 2 , which disproves (1.5.2).

This counterexample can be generalized to the following


observation, which will be made use of later:

PROPOSITION 1.1: V U, w 6 FA
w u w ~-~ w if u is a subword of w

proof: W U W = W 1 U W2 U W 1 U W2

~e wl u wI u w2 u wI u w2

uw 2

uw2 =w

If we define w u w m--~ w as a single cc-transition, (1.5.2)


cc
can be replaced by:

86
SIEKMANN AND S Z A B O

I * *
(1.5.3) w I = w2 iff 3v. wlm-~ v and w2= ; v
where = ~ := ~ U m-~
c CC

Since both transitions terminate, this would be a basis for


a Noetherian and confluent rewriting system, but:

Counterexample 2:
w = ab abc bac acb abc acb aba
O

ab abc bac bab cac ba ba ab abc ba cac bab a

ab abc bac bab cac ba a b c ba ca cba ba

ab ab cba bca cba a b c b ac ba ba

ab ab c a c ba abc bac ba

w I = abc a cba ab cba = w2

Hence we have two words w I ,w 2 6 FA with Wo ~ wI and


Wo~ w 2 , i.e. w I I w2 however w I and w 2cannOt be further
reduced by ~ nor by r. ; and w1 * w2 .
C CC

This example leads to our final generalization, which we


shall prove later on once a notational framework has been
fixed:
V u, v, w 6 F
A
(1.5.4) UVW ~ uw if {v} c {u} = {W}

2. NOTATION AND DEFINITIONS

Let f be a binary function symbol and let A be a non-


empty set (the alphabet). We define the set of terms TERM
as the least set such that A c TERM and if s,t s TERM then
f(s,t) s TERM.

We are interested in t h e o r i e s consisting of combinations of


the following three axioms for f:

87
SIEKMANN AND SZABO

(ASS) V u, v, w 6 TERM. f(f(u,v),w) = f(u,f(v,w))


(ID) V w C TERM. f (w,w) = w
(COM) u v, w 6 TERM. f(v,w) = f(w,v)

2. I. L e t T ~ u = v denote that equation u = v is derivable


from the set of closed equations T in a suitable logical
calculus (i.e. u = v is true in T). This is abbreviated to
T
u = v; u is T-equal t o v. Since we assume (ASS) to hold
throughout the rest of this paper, we shall omit the
function symbol f and represent it as u s u a l by the
concatenation of words. That is, we identify the associative
structure on TERM with FA and represent its elements as
words i n A*, where A 9 is the least set with A c A* and if
u, v 6 A* then uv 6 A*. { (ASS), (ID), (COM) } ~ u = v is
then abbreviated to u I~ v and { (ASS), (ID) } ~ u = v is
abbreviated to u I= v, u, v s A*. Such an e q u i v a l e n c e class
method is also proposed in [13].
2.2.Let lwl be the length (number of symbols)
w 6 A*. Let V be a denumerable set, A n V = ~, (the set of
variables) and let W = A u v. A substitution o is a m a p p i n g
from V to W* with o(v) = v,v s V almost everywhere (i.e.
except for a finite number of points) and is extended to
W* by the usual endomorphism. ~ is represented as t h e finite
set of pairs ~ = { (v I lw 1 )... ( V m l W m ) } f o r v. 6 V and w 6 W*.
1 l
The set of substitutions is SUB.

We use ~ for the Occurrence relation: u occurs i n w, u~w,


iff w = w I u w2.
For w 6 W* Var(w) = {vlv s V and v ~ w} is the set of
variables occurring in w o r d w.
The set
R = {(i I - - r I if ~ T ~i ) .... ,(i n -- r n if ~n T=)_n )}
w h e r e ii , r i 6 W* and ~i _~ixiis a T-equation over W*, is
a conditional rewriting system if V a r ( r i) c_ Var(li) , for
1<_i_<n. A conditional rewriting system R is used to define
the relation - ~ c A* x A* as follows:

u -~ v iff u = u lwu 2 and if there exists ~ 6 SUB and


(i ~ r if ~ T X) s R such that w = o(i) and v = uI ~ ( r ) u 2

88
SIEKMANN AND SZABO

and s(~) T ~(~) is true, where u, Ul, u2, v, o(r), w 6 A*.

Example: Let R = {(xyx ~ x if xy C=II x ) } with t h e corresponding


relation O-~.
Then dabc b abcdO-~ dabcd, with ~ = {(xlabc) (y!b)} ,

u I = u2 = d, w = abc b abc and


~(xy) = abcb c___I a b c = o(x).

This rewriting system allows us to replace the word wvw by


w if {v} c {w}, according to (1.5.4).
Alternative definitions for conditional rewriting systems
are discussed in [I].

It should be noted that the definition of --~ r e s t s on the


fact that a 6 SUB can effectively be computed, which may
not be immediately obvious for strings in W*. However it
can be shown that the one-way unification problem for
monoids is d e c i d a b l e and there are at most a finite number
of such substitutions ~[18]. If the conditional part in R
is omitted, the set of pairs:

R = {(11 ~ rI ), (12 ~ r 2 ) , . . . , (in ~ rn )}

is c a l l e d a rewriting system, and -~ is o b t a i n e d accordingly.

-~ is the reflexive, transitive closure of -~.

We say the relation ~ is Noetherian iff there is n o infinite

sequence: u I ~ u2 ~ u3 ~ ....
A relation ~ is confluent iff

V u, v, w 6 A*: if u -~ v and u -~ w then


3z s A* with v ~ z and w ~ z.

Confluence can be illustrated diagrammatically as:

v
/A w
~, lJ

where the dotted lines represent existential quantification.

89
SIEKMANN AND SZABO

A relation ~ is locally confluent i f f

V u, v, w 6 A~: if u ~ v a n d u ~ w then
3z s A ~ w i t h v ~ z and w ~-~ z.

Diagrammatically: u

w
\ /

A Noetherian and confluent relation ~ is c a l l e d canonical


and the rewriting system R corresponding to ~ is a l s o called
Noetherian, confluent or canonical.

A canonical rewriting system allows the computation of a


unique normal form for a word w:
A word u s A~ is terminal if t h e r e d o e s n o t e x i s t a w o r d v
with u ~ v. Since ~ is N o e t h e r i a n , for every word w s A~
there exists a word v with w ~v such that v is terminal.
Since ~ is c o n f l u e n t , v is u n i q u e , v is t h e normal form of
word w, which is d e n o t e d as llw11.

If a n equational theory T has a canonical rewriting system


we have
T
U = V iff llUll= llvll

and for that reason they are an important basis for the
treatment of T - e q u a l i t y on a computer. Huet and Oppen [7]
present a survey of some general techniques of how a
canonical rewriting system can be obtained for a given T.
Unfortunately none of these general techniques apply for a
set of equational axioms which includes associativity and
idempotence, although these two axioms often occur in
practical applications [8].

In t h i s paper we have confined our attention to the special


case of associativity and idempotence, however we believe
that our method of canonicalization can be applied to o t h e r
equational theories as w e l l for which the standard methods
fail.

9@
SIEKMANN AND SZABO

3. THE PROBLEM

An obvious rewriting system for the axiom of idempotence


is:
I = { (XX ~ X), (X ~ X X ) }
o c e
where ,~ i s a step by extension
e
and ~ is a step by collapse.
C

I is used to obtain the standard relations I--,, a n d ~ and


0 C e

C e

Hence we have:

V u, V 6 A*.
I
(3.1) u = V iff H w 6 A* such that
u e-~ w and v ~ w.

But (3.1) is not a basis for a computational treatment of


I-Equality since it does not show how to compute w.

Because ~ and consequently ~ are not Noetherian, I is not


e e o
a canonical rewriting system. Hence our intention to replace
I by a canonical rewriting system I such that
o C

I
wI : W2 iff llw1! 1 : !lw211 for wl, w2 6 A*,

where the normal form nlwll is c o m p u t e d on the basis of I .


C

4. RESULTS

We now replace the obvious rewriting system I of the


O
previous section by the following conditional rewriting
system:
I := {(xx ~ x),
C

cI
(xyz ~ xz if x ~ z and
CC
cI
xy ~ z)}

Let ~-~ := O-~c U ~c' where the relations ~-~c a n d --~cco n A *


are obtained from IC The computability of ~ depends on
the decidability of CI-equality, i.e. the decidability of

91
S I E K M A N N AND SZABO

word equality in commutative bands, which is obvious:

cI
u ----- w iff {u} = {w}.

In the following paragraphs we show that I


is a c a n o n i c a l
c
rewriting system and also that the replacement of I by I
o c
is v a l i d .
On the basis of I the nontrivial problem of equality in
c
bands is reduced to the trivial problem of equality in
commutative bands.
We believe that no unconditional canonical rewriting system
exists as a valid replacement for I , which would explain
o
why the standard methods of canonicalization [7] failed in
the past [8].

4.1 Soundness of ,--~.

We have to show that I is a correct (8ound) system in


c
the sense that if w 1 o~* w 2 then w 1 ~ w2 . This is easily seen
for wI ~ w 2, but less obvious for w I c~ w2 .

LEMMA 4.1: V v, w 6 A*:


CI I
if w = wv then wvw = w.

proof: (by i n d u c t i o n on Ivl)


(i) Inductionbase: Ivl = I, i.e. v = a, a 6 A
cI
If w = wa then a ~w, let w = ~ aw 2 .
Then wvw = wlaw2awlaw 2

~ wla ~ = w by Prop. 1.1.

Hence by (3.1) :
I
wvw = w
(2) Inductionhyp.: Ivl = n, let v = ala 2 ...a n 6 A*.
CI I
If w ~- wv then wvw = w.

Suppose Ivl = n+1, i.e. let v = ua, a 6 A, lu[ = n .


wvw = wua w
I I
wuw ua wuw since w = wuw by Ind.hypothesis

92
SIEKMANN AND SZABO

~-~ w u w ua wuwu w
e
I I
= wuwu W since WU WU a WU wu = WUWU
by Ind. base

~-~ w u w
c
I
= w since w ~ wuw by Ind. hypothesis.

LEMMA 4.2: u u, V, W s A ~

If u CI W and uv CI
~- w then uvw =I u w .
cI cI
proof: If u ~ W then u ~ uw
cI
and w = wu.

Hence by lemma 4.1 we have:

(i) uwu I= u

(ii) WUW I W

Then :
I
uvw = uwu v wu w using (i) and (ii)
= ~ uvw~ defining ~ := u w
I
= W by lemma 4.1 using ~ uvw c___I
= uw

Lemma 4.2 is, in our notation, the same as Rule C, which


has also been observed in [2], [15] and [6]. We have given
an alternative proof in order to base this paper solely on
the notions of T-equality and term rewriting systems.
The standard proof (see [9]) is in contrast more complicated
and also requires substantial results from semigroup theory.

Lemma 4.1 and lemma 4.2 prove the following soundness


theorem:

THEOREM 4.1: (Soundness) V u, v 6 A~


I
If u ~-~ v then u = v.

93
SIEKMANN AND SZABO

4.2 Confluence of ~-~.


We first demonstrate the local confluence by c o n s i d e r i n g
the three possible combinations of ~ a n d ~-~. T h e p r o o f s
C CC
are rather technical and for t h a t reason are collected in
the appendix.

LEMMA 4.3: V U, v, w E A ~ .

If U ~-. V a n d u ~-~ w t h e n Bz s A ~
C C
such that v ~ z and w ~ z

LEMMA 4.4: V U, v, W 6 A ~ .

If u ' - ~ V a n d u ~-~ w t h e n Hz 6 A ~
C CC
such that v ~ z and w ~ z.

LEMMA 4.5:
V u, v, W E A ~ .

If u - -C C~ v a n d u ~-~
CC
w then Bz s A ~
e
such that v ~-~ z and w~-.-~ z.

This demonstrates the local confluence of'--~ and p r o v e s the


following theorem:

THEOREM 4.2: ,-~ is a N o e t h e r i a n , confluent relation. I

proof: It is o b v i o u s t h a t t h e r e are no i n f i n i t e c h a i n s

u ~'~c vl ~ v2 ~ c "'"

u.-. .-....
CC CC

since Ivi l>[vi+l I and I~i l>IVi+iI 9

A Noetherian, locally confluent relation is


confluent by a theorem of N e w m a n (see [ii]).

Some readers may be more familiar with the notion t h a t ~-~


"is t e r m i n a t i n g and has the Church-Rosser-Property".

94
SIEKMANN AND SZABO

4.3 A Decision Procedure B a s e d on,--*.

The following theorem states our main result; where


IIwlI is the normal form of w based o n *-~.

THEOREM 4.3: V u, v 6 A*.


I
u iff l[ull -- l[v!l

proof: "-" u ~ [lu][ by Definition,


hence u ~ ]]ull by Theorem 4.1.
v*-~ ]]VII by Definition,
hence v ~ ][vll by Theorem 4.1.
Hence v ~ IIvll= llull ~ u.

11 ml~ I1 I
If u = v then by (3.1) and due to the fact that*--* is
symmetric there exists a chain

(i) u ~ v.
The proof is by induction on n, where n = lu ~ v] i.e.
the number of rewrite steps in (i).

n = O: Then u = v and since ~-~ is confluent it follows


that ]]ul] = ]lV][.
Suppose the theorem holds for all u, w with u ~ w,
where Iu ~ w I < n.

Let u ~ v have n+1 steps, i.e.

U ~-~ W ~'-~ V .

Then by induction hypothesis llull = IIwll.

Case I : w ~-~ v. Since ~-~ is a l s o contained in .-~,


C C
~-~=~-~ , w e have by the confluence of--~:
C C
Hz. w ~ z and v ~ z. Let z be terminal, then
z = ITwll -- llvll.

Case 2: w ~-~ v. Take the reverse step v ~-~ w, and as


e c
in case I there exists z with v ~ z and w*~ z. For
z terminal, again we have z = llwlI = llvIl.

95
SIEKMANN AND SZABO

Hence in either case

llult = llwll = llvll

In applications the problem wI ~ w2 is usually not presented


in A * but in TERM. Define a homomorphism ~: TERM ~ A9 that
maps a term in its "string"-representation as mentioned
in 2.1.

Then: { (ASS) , (ID) } ~ w I = w2 iff { (ID) } ~ q0(w I ) = %0(w 2) ,


for w I, w2 6 TERM.
Hence we have as an obvious consequence:

COROLLARY: V s, t s TERM:
{(ASS),(ID)} ~ s = t iff ll~(s)II = ll~(t)II .

4.4 Idempotence without Associativity


Let 11 = {(f(x,x) ~ x)} and let ~-* b e the associated
relation on words in TERM.

~-~ is locally confluent on TERM:

PROPOSITION 4.1: V u, v, w 6 TERM with u ~-~ v and u o-~ w


there exists z 6 TERM such that

v ~-~ z and w ~ z.

proof: (by induction on the number of function symbols f


in u)

n=1: u = f(a,a) for some a s A and hence v = w = a.

Suppose the theorem is true for all terms u with at


most n function symbols f.
Let u = f(s,t) with n+1 function symbols:
I (~ ^ A
Case l:If s=t and v = s , w = f ,t) with s ~-~ s; set z = s.
A A A
Case 2: v = f(s,t) w = f(s,t) with s ~-~ s
A
t~-~ t
^ A
Set z = f(s,t).
A
A A A
Case 3: v = f(s,t), w = f(s,t) with s ~--~s

96
SIEKMANN AND SZABO

^
S ~-~ S, S # S.
Then exists by induction hypothesis A
z with
^ 9 ^ ~ ~ ^
S ~--+ Z and s o-~ z.

Set z = f(zA,t).

S i n c e o-~ i s locally confluent and Noetherian:

PROPOSITION 4.2: o-~ is confluent on TERM.

Using the confluence o f ~-~ w e have the following theorem,


which is proved analogously to Theorem 4.3:

THEOREM 4.4:

I
V u, v 6 TERM: is by ~.
{ID} ~ u: v iff llull = !Ivli
where the normal form !lull 9 obtained

Proposition 4.1 and 4.2 are also an immediate consequence


of the results in Knuth-Bendix [12]: f(x,x) ~ x is clearly
Noetherian. It is also locally confluent, since there are
no critical pairs. Hence follows confluence.
This was pointed out to u s by D. Lankford; we presented the
proof of P.4.1 in order to keep this paper selfcontained.

5. APPENDIX: PROOFS

The proofs for the three lemmas of section 4.2 are now
collected.

I n order to prove Lemma 4.3 we make use of the following four


propositions, each of which is of the form V u, v, w: if
u ~-~ v and u ~-+ w then 3z. v --~ z a n d w ~-+ z, w h e r e
c c
c cc

Diagrammatically: u = u Ixx u2 = u3yy u4

//c
v = u lx 2 u4 = w

97
SIEZ,MANN AND SZAB6

The four propositions capture the relevant situations

arising from the relative position of xx to yy. Throughout

let us assume u, u i, v, vi, x, xi, y, Yi to be words in A*.

The first proposition may be illustrated:

u I x x u2
, ~ L r ~
:x1: ,x2
I i t I
I I

u 3 Y Y u4

PJ?OPOSITION 5. I: Let u = u I xx u2 = u3 yy u4 with

u 3 = u I x I, x = x I yy x2 and u4 = x2 x u 2.
Let u~ v = u I x u2 and u~ c w
,
= u 3 y u 4.
Then: Bz. v~--~ z and w'- -" z .

proof: v = u I x u 2 with x = x I yy x2

= u I x I YY x 2 u2

u I x I y x2 u2 = z

w = u3 y u4 with u3 = u I x I

= Ul Xl Y U4 with u 4 = x2 x u 2
= u I x I y x2 x u2 with x = x I yy x2

= u I x I Y x2 x I YY x2 u2

c u I x I y x2 x I y x2 u 2~ u] x I y x2 u 2 = z

The next proposition may be illustrated:

Ul| x x u 2
I ~ 9 ~ ~, o I
,x I 'x , x~, ;x4 1
, T 2 ~ ~t I ,
i t I I
l i I t
, I ~ i $, , I
u3 Y Y u4

PROPOSITION 5.2: Let u = u I xx u 2 = u3 yy u4 with

x = x I x2 = x3 y x 4, y = x2 x 3, u3 = u I x 3, u4 = x 4 u 2.
Let u~ v = u I x u2 and u~ w = u 3 y u 4.
Then 3z. v ~-~ z and w o-~ z .

98
SIEKMANN AND SZABO

proof: w = u I x I y x4 u 2 with y = x 2 x 3

= u I x I x 2 x 3 x4 u 2 with x I x2 = x 3 y x 4

= u I x 3 Y x 4 x 3 x 4 u2 with y = x 2 x3

= u I x 3 x 2 x 3 x 4 x 3 x4 u 2 on x 3 x 4

u I x 3 x2 x 3 x4 u2 with x 2 x 3 = y

= u I x 3 Y x4 u2 with x 3 y x 4 = x

= u I x u 2 = v = z

The third proposition may be illustrated:

u I x x u 2
, : J ~ ~ It | J
IU 1 I I i
I 5 IXl I ,x21
I I I I
I t I I
I i t t
,,, L I ~ , j l
u 3 Y Y u 4

PROPOSITION 5.3: Let u = u I xx u2 = u 3 yy u 4

with u I = u 3 u5, y = u 5 Xl, x = x I y x2, u4 = x 2 x u 2.

Let u~ v = .u I x u2, u~ ~ = u 3 y u 4.
Then 3z. v~ = z and w~ ~ z.

proof: v = u I x u 2 with x = x I y x 2

= u I x I y x2 u 2 with u I = u3 u 5

= u 3 u 5 x I y x 2 u2 with u 5 x I = y

= u 3 yy x 2 u2 on y

u3 y x 2 u 2 with y = u 5 x I and u3u5=u 1

= u 3 u 5 x I x2 u2 = u I xI x 2 u 2 = z

w = u 3 y u 4 with y = u 5 x I

= u 3 u 5 x I u 4 with u 4 = x 2 x u 2

= u 3 u 5 x I x 2 x u 2 with x = x I y x2

= u 3 u 5 x I x2 x I y x2 u 2 with y = u 5 x~

= u 3 u5 x I x 2 x I u5 xI x 2 u2~u3u5xlx 2~xlx2~

~-c:U3 u5 x I x 2 u2 with u 3 u 5 = u I

= u I x I x 2 u 2 = ~ 9

The final proposition may be illustrated:

u I x x u 2
n p ! ! I
i ' I
',us :xl x2 ,'x3
! ,, 'I r'i i' i II i
u 3 Y Y u4
99
SIEKM~NN AND SZABO

PROPOSITION 5.4: Let u = u I xx u 2 = u 3 yy u 4 with

u I = u 3 u 5, y = u 5 Xl, x = x I x 2, y = x 2 x3, x = x 3 x4,

u 4 = x 4 u2 9
Let u~ v = #I x u2 and %~ w = u 3 y u 4.
Then Bz. v~ ~ z and w= ~ z.

proof: The equations x I x2 = x = x 3 x 4 of the hypoth~is

give rise to three cases:

C=8~ I: Ixll = Ix31; i.e. x I = x3, x 2 = x 4 .

Then v = u I x u2 with u I = u 3 u 5

= u 3 u5 x u 2 with x = x I x 2

= u 3 u 5 x I x 2 u 2 with x2 = x 4

= u 3 u5 x I x4 u 2 with y = u 5 xl,u 4 = x 4 u 2

= u 3 y u 4 = w

cas~ 2: Ixll > Ix31; i.e. x I : x 3 x 5 and hence

x 4 = x 5 x 2 9
Then: w = u 3 y u 4 with y = u 5 x I

= u 3 u 5 x I u 4 with u 4 = x 4 u2

= u 3 u 5 x I x~ u2 with u 3 u 5 = u 1

= u I x I x 4 u2 with x 4 = x 5 x 2

= u I x I x 5 x 2 u 2 with x I = x 3 x4

= u I x 3 x 5 x5 x 2 u2 on x 5

Ul x 3 x 5 x 2 u2 with x 3 x5 = x I

= u I x I x 2 u 2 with x I x 2 = x

= u I x u 2 = v

cas~ ~: Ixll < Ix31; i.e. x 3 = x I x5, x 2 = x 5 x4

v = u I x u 2 with u I = u 3 u 5

= u 3 u 5 x u 2 with x = x I x 2

= u3 u5 x I x 2 u2 with u 5 x I = y

= u 3 Y x 2 u 2 with y = x 2 x3,x2=x 5 x 4

= u 3 x 2 x 3 x 5 x 4 u 2 with x 3 = x I x 5

= u 3 x 2 x I x 5 x 5 x 4 u 2 on x 5

u3 x 2 x I x 5 x 4 u 2 with x I x 5 = x 3

= u 3 x 2 x 3 x 4 u 2 with x 4 u 2 = u 4

= u 3 x 2 x 3 u 4 with x 2 x 3 = y

= u 3 y u 4 = w

I00
SIEKMANN AND SZABO

These four propositions are now used to prove lemma 4.3:

proof: Let u be of the form:


u = u I xx u 2 = u 3 yy u 4 with ul,u2,u3,u4,x,y s A9
and v = u I x u2 and w = u 3 y u 4.
There are nine cases to b e considered, depending on t h e
relative position of xx and yy, i.e. depending on w h e -

ther: u I = u3 u2 = u4
lu1!>lu31 ! u 2 r < l u 4]
luII<lu31 lu21>tu41.

uI x x u2
case I: I
u~ L y 9
y u4 I ~ I~'uI = u 3, u 2 = u 4
Case 2: , u3 i, y . y . u4 , . ..... u I = UB,tU2l<lu41

Case 3: u~ t Y 9 Y ,u4 u_~-Ul = u3,1u21>lu41


i

Case 4: u 3 , iy , y u4 ,.......!uiI>lu31 , u 2 = u 4
Case 5: u 3 ~ iY Y* u4 t--.....! u I l > l u 3 1 ' 1 u 2 1 < l u 4 1
ease 6: u3 9 Y 9 Y.- u4 I- .... ! u 1 1 > l u 3 1 ' l u 2 1 > l u 4 1
Case 7: u3 9 ~ u4 i.......lu I! <!u31, u 2 = u 4
Case 8: u3 "YY"* ii u4 ,~f~.ru11 <lu3!,lu2! <lu41
Case 9: u3 k Y A [, u 4 m .... l u 1 ! < l u 3 ! , l u 2 1 > ! u 4 1

case 9: D e p e n d i n g on the relative length of y t h e r e are


exactly six cases, which c a n be p r o v e d with Prop.
5.3 and Prop. 5.4 using symmetry and exchanging
rSle of x a n d y.
Case 8: D e p e n d i n g on the relative position o f yy (insi-
de of x or overlapping) this c a s e can be proved by
Prop. 5.1 or by Prop. 5.2 respectively.

Case 7: S p e c i a l case of P r o p . 5.2 w i t h x 4 = e.


Ca8e 6: P a r a l l e l s Case 8, w h e r e the rSle of x a n d y is
exchanged.
Case 5: S y m m e t r i c to Case 9.
Case 4: S p e c i a l case of P r o p . 5.4 w i t h x 4 = e.
Case 3: S y m m e t r i c to Case 4.
Case 2: S y m m e t r i c to Case 7.
Case I: T r i v i a l , with v = w = z. 9

I01
SIEKMANN AND SZABO

In order to prove Lemma 4.4 we need two propositions:

PROPOSITION 5.5:
cI cI
Let u = rst with rs : r : t

Let r = rlr2r3, s =SlS 2 and r2 = r 3 s I.

L
r I
s !
t I

r 1 r 2 r 3 s 1 s 2 t

For u = rlr2r 3 S l S 2 t ~-~ r l r 3 SlS 2 t = v

u = rst ~-~ r t = w
cc

there exists p with v ~-~ p and w ~-~ p .

cI cI cI
Proof: S i n c e rlr2r 3 = t and r2 = r2r3: rlr2~-~-t
(5.5.1)
cI cI
Since rlr2r 3 = t : r3t : t

Now w = rt

= rlr2r 3 t
9e-c~ rlr 2 t by (5.5.1)

= p.

And v = rlr 3 SlS 2 t

= rlr 2 s2 t since r2 = r3 s1
cI
9C-C~ rlr 2 t by (5.5.1) and s2t =t

= p.

PROPOSITION 5.6 :
CI CI
Let u = ss = rxyzt with xy --~- x = z

L S L S ~ = U

r xyz t = u

For u = ss ~ s = v

u = rxyzt ~-~ rxzt = w


CC ~
there exists p with v ~-~ p and w ~-~ p .

Proof: Depending on the relative length of xyz there are

four cases:

Case 1 : s s = u
I, I I
i i i v v

r x y z t = u

102
SIEKMANN AND SZABO

u = ss ~-* s = p
c

u = rxyzt e-~ rxzt with rxy = s = zt


cc

= rx rxy

rxy = s = p.
c

Case 2: s = u
h ! i

Y x ' y ' z ' t' = u

Symmetric to case 1.

Case 3: L s ! s i = u

Yl Y2
'r ' x I ~ tz ' t " = u

Since s has Y2 as prefix and Yl as postfix,

we have s = Y2 s' Yl = rxYl = Y2 zt for some

s' 9 Hence s'y I = zt and y2 s ' = rx (5 . 6.1) .

Now u = ss ~ s = p
c

u = rxyzt--~ rxzt
cc

= y2 s' s'y I by (5.6.1)

Y2S'Yl

-- s
= p.

Case 4: l S ~ S j = U

'r ' x ' y' z I ' z2 ' t ~ = U

with z = ZlZ 2
CI CI CI
Observe: ZlZ ----- z ~ x ~ ZlX (5.6.2)

u = ss,~ s = p
c

u = rxyzt ~-~ rxzt


cc

= rxzlz2t with z2t = s = rxyz 1

= rxzlrx.y z1

~-~ rxrx y zI by (5.6.2)


cc

o-~ rx y z
c 1

= s

= p.

103
SIEKMANN AND SZABO

Proof of Lemma 4.4: The proof resembles the proof of 4.3 in


that there are the same nine cases depending o n the relative
length of Ul, u 2, u 3 and u 4. Unfortunately however,
each case has various subcases:
Each case is of the form:

U = u I SS U2~ U1 S U2 = V
CI CI
U = U 3 Xyz u 4 ~6~
c U 3 Xz U4 = W with xy ~ x ~--- z

and there exists p such that

v~-~ p and w ~-~ p.

Case I: u~ s s u~ = u
I I I i !

, D , ! !

(1.1.) u 3 x y z u 4 = u

u I ss u2 ~ uI s u2 = v = p

u 3 xyz u 4 ~c u3 xz u4 with
s=z=xy

= u 3 xxy u4

u 3 xy u4

= u 3 s u 4 = p.
(1.2.) u3 x ,y z
L u4 i

symmetric to (i.i.)
(1.3.) .u 3 x ~zY . ~
, u4 .
Y l Y2

Since xy I = s = Y2 z

we have: s = Y2 ~ Yl = Y2 ~ Yl (1.3.1.)
and hence x =

u I ss u2 ~ uI s u2 = v = p

u 3 xyz u 4 ~c u3 xz u4

= u3 Y2 xz Yl u4 by (1.3.1.)

since x =
c u3 Y2 ~ Yl u4

= u3 s u4 = p.

Case 2: uI s s u2
U ---~ I i i i I

u = ' u 3 ' xyz ~ U4 ~

]04
SIEKMANN AND SZABO

Depending on the relative length of xyz there


are three subcases each of which is an instance
of proposition 5.6 for the case r = e (i.e. r
the empty word).

u u
Case 3: u = t
1 !
s I
s !
2 I

u = ' ' xyz 'u 4 '

There are five subcases depending on the


relative length of xyz, each of which can be
proved on the basis of prop. 5.5 (for the
special case that r I is empty) and the obser-
cI
vation that if u~ v then u = v.

Case 4: Symmetric to c a s e 3.

Case 5: Symmetric to case 9.

Case 6: Depending on the relative length of xyz there


are five subcases, the only difficulty one is
shown in proposition 5.5.

Case 7: Symmetric to c a s e 2.

Case 8: There are three subcases each of which is shown


in proposition 5.6.

Case 9: Depending on the relative length of xyz there


are ten subcases:

uI s s u2

(9.1) ix,Y,Z~

(9.2) iX t yl Zl

(9.3) ,x 0 y , z ,

(9.4) ~x,y, z L

(9.5) i,
X | y i Z

(9.6) |,,,
X i Y I Z I .

(9.7) I
X
,Y i
Z
I.

(9.8) I
X
I
y , z ,

(9.9) i, x i y , Z ,1'

(9.10) I'x'y i
z
I.
)05
SIEKMANN AND SZABO

Cases 9.1, 9.2 and 9.5 are trivial.

(9.3) uI ss u2 ~ uI s u2 = v; let s = s I x Yl and

y = ylY2
uI s s I xyz u22 ~c Ul s s I xz u22 = w

w = uI s I x Yl Sl xz u22

with s = s I x Yl

e-~cc U l Sl x s I xz u22
CI cI
since y = ylY,Xy= x

uI s I xz u22

= p

v = uI s u2 = uI s I x Yl u2

= uI s I x Yl Y2 z u22

9c-~
c uI s I xz u22
= p

(9.4) This case can be shown by observing that since


y can be cut out in the second occurence of s
using a ~-~-step it can be cut out in the first
cc
occurence also.

(9.6) Similar to (9.3)


(9.7) Similar to (9.4)

(9.8) Let s = sI x YI' y = YlY2Y3 and u2 = u21 z u22.


Then

u = uI ss u2 ~ uI s u2 = v

u = u I s I xyz u22 ~c Ul Sl xz u22 = w = p.

Now v = uI s u2

= uI s I x Yl u2 " s = sI x Yl

= uI s I x Yl u21 z u22 , u 2 = u21 z u22


cI
~-~
cc
u I s I xz u22 , since Yl u21 z -~- z

= p.

(9.9) Let s = sI x YI' y = YlY2 ' s = Y2 Zl, z = ZlZ2,

u2 = z2 u22

u = u I ss u2 ~ uI s u2 = uI s I xy I z 2 u22 = v

106
SIEKMANN AND SZAB6

u = u I s I xyz u 2 2 O~c u I s I x z u22

= u I s I xz I z 2 u22 = w

if [Yl = I Zll then Yl = Zl and hence: v = w

if ]Yl > I Zll then Yl = qzl and x = xlq

hence v = u I s I x Yl z2 u22

= u I s I x I qq z I z 2 u22

uI sI xI q z I z 2 u22 = w

if ly I < IZll then zI = z I = q Yl and x = xI q

hence w = uI sI x z I z 2 u22

= Ul Sl Xl qq Yl z2 u 2 2

Ul Sl Xl q Yl z2 u22

= u I s I x Yl z2 u22 = v

(9.10) Similar to (9.7) and (9.4).

Proof of Lemma 4. 5" T h e r e are two general situations from


which all special cases can be derived by symmetry and
for interchanging the denotation of the corresponding
subwords.
Let u = u I xyz v I = u 2 rst v2, where

cI cI cI cI
x = z = yz and r = t = st.

If u--~ v = u I xz vI and
cc
u ~-~cc w = u 2 rt v2, then there exists p, such that

v ~-~ p and w~-~ p.

Case 1: y a n d s do not overlap, i.e. s is completely left


(or right) of y.
w.l.o.g, let

u = u I x yz v I = u 2 rst v 2 = u 2 rs yy z v1
then p = u2 r y z v I.

This situation may be illustrated as:


Ul x y zv
, f . i,

,
~ 2 r s t v2

107
SIEKMANN AND SZABO

u = u I xyz Vl~-~c uI xz v I = u2 rsy z vI = v and

u = u 2 rst v2 ~ c u 2 rt v2 = u 2 ryy z v I = w

(l.1) ltl-<IPl and Ix -<IYl- Then t-y, x * y and


therefore v e~c p and w ~-~ p.
cc
- cI -
(1.2) Itl-<lyl and Ix > 1 7 ! - Then s t c___I t, y = yt
- cI
implies t h a t y ----- z a n d t h e r e f o r e v ~c p and
w ~-. p.
cc
(1.3) Itl>lyl and Ix ~<191- Then xy ~cI x, x y_ e l y-
-----
cI -
implies that r ~-- y and therefore
v~-~ p and w.--~ p.
CC CC
(1.4) Itl>lyl and Ixl>IPl Then xy c___I x, xsy c_~_I s T
cI - cI -
implies that r = y and z ----- y and therefore
v--~ p and w ~-~ p.
CC CC

Case 2: y a n d s overZap; i.e. y = e and let y = Yl Y2

and s = sI s 2 with s2 = YI:

i
i
i Yl Y2
U1 X i Y ~ z v1
i a , ~ ; l o
i
i
s 2 ii s I i.
; i U
| u I I !
r s t
u2 v2

u = u I xyz v I ~c Ul xz vI = v

u = u2 rst v 2 e~c u 2 rt v2 = w

But this single cc-step can be decomposed into


two cc-steps :
u ~ v and u ~ w iff
cc cc

u = uI x Yl Y2 z vI ~ uI x Y2~Vl .-~
cc v and

u = u2 r sI s2 t v 2 "~c u 2 r sI t v 2 ~c w .

From the diagram it is easy to see that

u I x Y2 z vI = u2 r sI t v 2.

But this is just a special case of Case 1 for


= e. Hence there exists p such that

v ---* p and w ~-* p . 9


108
SIEKMANN AND SZABO

Acknowledgement: In an e a r l i e r v e r s i o n of this p a p e r we
only p r o p o s e d a specific case of RULE C as a b a s i s for the
second rewrite. We are m o s t g r a t e f u l to H. J H r g e n s e n from
the T e c h n i s c h e Hochschule Darmstadt, w h o d r e w our
attention to the r e l e v a n t literature of s e m i g r o u p theory
which greatly improved the paper 9 Thanks are also due to
P. Grund.

In p a r t i c u l a r we like to thank D. L a n k f o r d for his


interest in our w o r k and the lively c o r r e s p o n d e n c e , which
uncovered several errors and s h o r t c o m i n g s in the d r a f t of
this paper.

REFERENCES

[i] D. Brand, J. Darringer, J. J o y n e r "Completeness of


Conditional Reductions"; Proc. of the W o r k s h o p on
Automated Deduction, 1979

[2] T.C. B r o w n "On the F i n i t e n e s s of S e m i g r o u p s in w h i c h


x r = x" ; Proc C a m b r i d g e Phil. Soc. 60, 1964

[3] A.H. Clifford, G.B. Preston "The A l g e b r a i c Theory of


Semigroups"; American Math. Society, 1961

[4] J.H 9 Green, D ~ Rees "On S e m i g r o u p s in w h i c h x r = x"


Proc. Cambridge Phil. Soc. 48, 1952

[5] J.V. Guttag, E. H o r o w i t h , D.R. Musser " A b s t r a c t Data


Types and S o f t w a r e Validation"; Com. of the ACM,
vol 21, no 12, 1978

[6] J.A. Gerhardt "The L a t t i c e of E q u a t i o n a l Classes of


Idempotent Semigroups"; J. of Algebra, 15, 1970

[7] G. Huet, D. O p p e n "Equations and R e w r i t e Rules: A


Survey"; in:'Formal Language Theory: P e r s p e c t i v e s and
Open Problems'; R.V. Book (ed), A c a d e m i c Press, 1980

[8] J.M. Hullot "A C a t a l o g u e of C a n o n i c a l Term Rewriting


Systems"; Technical R e p o r t CSL-II3, Stanford Research
Institute, 1980
[9] J.M. Howie "An I n t r o d u c t i o n to S e m i g r o u p Theory",
Academic Press, 1976

109
SIEKMANN AND SZABO

[iO] M.A. Harrison "Introduction to Formal Language


Theory", Addison Wesley, 1978

[ii] G. Huet "Confluent Reductions: Abstract Properties


and Applications to Term Rewriting Systems";
JACM, vol 27, no 4, 1980

[12] D. Knuth, P. Bendix "Simple Word Problems in Universal


Algebras", in: Computational Problems in Abstract
Algebra, (ed) J. Leech, Pergamon Press, 1970

[13] D.S. Lankford, A.M. Ballantyne "Decision Procedure for


Simple Equational Theories with Permutative Axioms",
University of Texas, Report ATP-37

[14] D. Loveland "Automated Theorem Proving", North Holland


Publ. Comp., 1978

[15] D. McLean "Idempotent Semigroups", Americ. Math. Mon.


61, 1954

[16] P. Raulefs, J. Siekmann, P. Szab~, E. Unvericht "A


Short Survey on the State of the Art in Matching and
Unification Problems", Bulletin of EATC, Oct. 1978

[17] J. Siekmann, P. Szab~ "Unification in Idempotent


Semigroups"; Universit~t Karlsruhe, Institut fur Infor-
matik I (in preparation)

[18] J. Siekmann "Unification and Matching Problems",


Universit~t Karlsruhe, Institut fur Informatik I, 1978

[19] J. Siekmann, P. Szab~ "Universal Unification",


Universit~t Karlsruhe, Institut fur Informatik 1,1981

[20] P. Szab~ "Unifikationstheorie erster Ordnung",


Universit~t Karlsruhe, Ph.D. (in German)

J. Siekmann, P. Szab6
Universit~t Karlsruhe
Institut fHr Informatik I
Postfach 6380
7500 Karlsruhe 1
Federal Republic of Germany

Received January 30, 1981; in revised form April 23 and


December 17, 1981

110

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy