Research Report 222 The Topology of Partial Metric Spaces Matthews SG RR222 t world of Scott's topological models used in the denotational semantics of programming languages may at frnt sight appear to have nothing whatever in common with the Hausdorff world of metric space theory. Can this be true though when the notion of "distance" is so important in the application of inductive proof theory to recursive definitions? This paper shows that existing work on the application of quasi metrics to denotational semantics can be taken much further than just describing Scott topologies. Using our "partial metric" we introduce a new approach by constructing each semantic domain as an Alexandrcv topology "sandwiched" The between two metric topologies. to be presented at the Eighth Summer Conference on General Topology and Applications, June 18-20 1992, Queens College, New York City. Deparrnent of Computer Science Univenity of Warwick Coventry CV4 7AL United Kingdom June 1992 The Topology of Partial Metric Spaces Steve Matthews Dept. of Computer Science University of Warwick Coventry, CV47AL email 1. sgm@dcs.warwick.ac.uk Introduction In the study of the denotational semantics of programming languages topological models are created for programming languages represented by systems of logic. More often than not this means a T o model for the lamMa calculus in the spirit of Scott. However, the inherent assumption in this approach that all suitable models must be Tp appears to remove any possibility that the theory of metric spaces ( which are all T2) can be applied in any way to domain theory in Computer Science. Rare exceptions to this rule are the use of quasi metrics in [La87]. in [Sm87] and metrics If metrics are to be used at all then the more conventional wisdom in Computer Science would dismiss Scott's T6 approach in favour of a purely metric approach ldB&7421. Unfortuantely the latter T2 approach pays the price of loosing any notion of partial order, a concept of fundamental importance in any Tarskian approach to fixed point semantics in denotational semantics. The distinct advantage of using quasi metrics is that such generalised metrics can be used to define T9 topologies with partial orders, and so allowing Tarskian semantics. Quasi metrics are not without their problems though. Being non-symmetric a quasi metric is arguably an "unnatural" notion of distance. Perhaps a more important crticism is that they shed linle light on how to develop proof theory for programs based upon metric reasoning. In [Sm87] the phrase Reconciling Domains with Metric Spaces is used to suggest that conventional wisdom from the theory of menic spaces can be applied to programming language semantics. [Ko88] contains a proof that All Topologies from Generalised Metrics. This may or may not be of interest to topologists in general as many of the more pleasant T2 properties usually come associated with metric spaces may be lost in a process of generalisation. However, the point that metrics can, if only in principle, be generalised to explore non - T2 topologies is established. Combining this with Smyth's work on quasi metrics we are naturally lead to the following question. To wlnt extent can the T2 topological metlnds applied to progrwntning langtage sennntics CUNY '92 ? of metric space tlwory be to this question can be found by showing that This report aims to demonstrate that a positive answer suggested' taken much further than has previously been the existing work on quasi metrics can be but metric be used to describe Scott Tp topologies Not onry can a generalised merric such as a quasi here the w'eighted metric' a version of a metric we can also use metrics as well. we introduce equivalent to a restricted form of quasi meric' algebraically method of constructing domains Defined using partial metics we propose here a asAlexandrovsub-topologiesofweightedmetrictopologies. of metric spaces ever closer to Scott's theory of The purpose of this work is to bring the theory with metric Spaces a reality' domains, and so make the reconciliation of domains 2. Backgound Definitions Definition A 2.L Topotogy over a set & Results U is a set tr g 2u such that ' ) @eT (T2) U e f, (T1 (T3) vsgtr user ono'eT (T4) vo,o'er Members of T are called open sets and their complements closed sets ' Definition 2.2 for a topology The T0, T1 , and T2 (i.e. Hausdorff ) separation Axioms To = V x*x'eU TI = V x*x'e T2 = V x*x'e t are, SOeT (x eO A x'* o) v( x' e o A x * o) U (3O'O'ef x e O-O' A x' e O'-O ) U (3O'O'ef x e O A x'e A OnO'= g ) O' If f is T2 then f is Tt, andif f is Tt then f is T0' CUNY '92 2 Definition A 2.3 Basis for a topology f is a set E c tr such that, r = { us I ss03 } Lemma i3 c 2u 2.1 X over U il is a basis for a topology and vo,o'eE 3 sga3 0no' = us u = uE Lemma 2.2 E , E' g 2u are each a basis for the same topology o = US' O'= US V oefJ 3 S'gtrJ' v o'e?3' 3 Sctr3 Definition A and, 2.4 Partial Order (poset) ) (POz) (PO3) (POl Definition on U if , isapair < U, Vxe(J Vx,yeU V x,l,ze(J << c U XU x<<x x<<y Ay<<x + x=! x<<y A yKz =t x<<z 2.5 An Alexandrov Topology overaposet <U,<<> isatopology T over U such that, VOetr xeO Ax<<y - V x,yeU yeO T = { { ", n+l , ... , @ } I n e u J over ( t t U { * }, S > is an example of an Alexandrov topology. The Full Topology tr [ <</ overaposet < U, << > is the topology Thetopology, x<<y - { OsU I VxeO,y€(J of all upward closures [Vi89l ft <l cuNY , '92 however , } The topology ,oo} I neuU{@} .'.'= { {n,n+l is the Full Alexandrov Topology is always T6 . yeO Abxandrov over ( <o u l {* } not every Alexandrov Topology is Tp as the indiscrete topolog;' 3 T = { A , {0,1} } shows. Definition 2.6 A Scott Topology is an Alexandrov Topology f < U , << ) such that for each chain X e @U , VOef The only T2 fubXeO over a chain complete poset (cpo) = 3k>0 Vn>k XneO Scott topologies are the "flat" trivial ones in which, V x,y e U x << y € x=J The least element J- (bottom) is not included here in our definition of a Scott topology as it is not essential to the development of the work presented in this repon. later work using our methods for frxed point semantics and reflexive domains will need to include -L. Definition A Metric 2.7 [Su75] is a function (Ml ) (M2) (M3) d : U Vx,yeU V x,y e(l V x,y,z e (J Definition 2.8 An Open Ball forametric d : U Bdr(x) ,,= t y.U X U -+ 9I such that, x=! e d( x,y)*0 d(x,y) = d(y,x) d(x,z) XU -+ fr isasetof rheform, I d(x,y) forany x eU and t>0. Lemma 2.3 Thesetofallopenballsofametric on U. d : UXU I 9t isabasisfora T2 topology ftdl Definition 2.9 A Quasi Metric isafunction q : UXU -) n (Ql ) (Q2) CUNY '92 Vx,y€(l Vx,y,ze(J x--y € q( suchrhar, q(x,y)=q(y,x)=0 x,z) 4 Lemma 2.4 q : UXU -) fr Foreachquasimetric xKo! e Vx,yeU therelation <q I UXU definedby, q(x,y)=0 is a partial ordering. Lemma 2.5 For each quasi metric q : Bs€(x) U X U -+ fr { yrU "= the set of all open balls of the form, I q(x,y) forany x e(J and €>0 isthebasisforanAlexandrov T0 topology ftql < u, over >. <<q Example 2.1 d : U X U -) fr and for each collection of closed sets Uc c 2u the function q : Uc X Uc -+ fi where, For each complete metric V X,Y e 2u q(X,Y) isaquasimetricsuchthat, Lemma 2.6 Foreachquasimetric VX,YeW q : UXU q fr V x,y eIJ isametricsuchthat .'.'= sup{ inf{ d(x,y)l(1+d(x,y)) XgY thefunction a X lyeY } I xeX l <<qY q^ : UXU + fr where, q^(x,y) ::= q(x,y) + q(Y,x) ft qI gft q^l Lemma 2.7 ( Quasi Metric Contraction Mapping Theorem ) Foreachquasimetric q : UXU -+ fr suchthat q^ iscomplete, function f :U-+U suchthat, 3 0Sc<l thereexistsaunique a e V x,y€(l andforeach q(f(x),f(y)) U suchthat a = f( a ) . Note that the unique fixed point of a quasi metric contraction mapping need not be maximal as the following simple example shows. CUNY '92 5 q : {a,b}2 -, {0,1} Q(a,b)=0 , q(b,a)- I f : {a,b} -+ {a,b} f(a) = f(b) = a Thus this theorem cannot be used to prove that recursive definitions deftne total objects Section 5 we present a second theorem specifically designed for this purpose. . In 3. Partial Metrics Definition A 3.1 PartialMetric (Pmetric) isafunction The pmetric axioms Ml suchthat, Q p(x,x)-p(x,y)=p(y,y) x=! Vx,yeU p( x,x ) V x,y e U Vx,YeU P(x,Y) = P(Y,x) p(x,z) V x,y,z eU (P1) (P2) (P3) (P4) axioms p: UXU -+91 thru Pl M3 thru P4 are intended to be a "minimal" generalisation of the metric such that objects do not necessarily have to have zero distance from themselves. In this generalisation we manage to preserve the symmetry axiom M2 to get P3 , but have to "massage" the transitivity axiom M3 to produce the generalisation P4 . Consequently a metric is precisely a pmetric p : U X U -+ 9t in which, VxeU For each p(x,x) - 0 pmetric p : U X U -> fi we preserve "half' of the metric reflexive axiom Ml giving, V x,y eU Example p(x,y) 3.1 - {0}, < > ofpositiverealnumberscanbedefinedbythepmetric p(x,y) .'.'= mer{ Ilx,lly p : (9t - {0})2 --+91 where, V x,y e U as V x,y e (J x <y e p(x,x) = p(x,y) Theposet < 9t Example 3.2 Thewellknown Baire S is defined by CUNY '92 } , Metric dB : a'S X ars + X of @-sequencesoveraset e a'S Vx+y dB(x,y) = 2-^in{neul{n)+y(n)} Thisdefinitioncanbeextendedtotheset ( U{ {0""'n-l}S finite and c^r - sequences over S by the Baire Pmetric , f I new } ) u "tS ofall -+ fr : ((U{{0,...'n-1 }S lneco })u.S)2 Inthisexample where, V neu), x € {0,-.-,n-1}S f( x, x ) ::= 2-n \f,.e can use the condition p( x, x ) = 0 to disringuish between those x which are in the Baire Space and those which are not. Example 3.3 A FlatPmetric isapmetric pt j S_rXSr+ {0,IJ for a set S, and L + S, and, pt(x,y)=0 Vx,)€S_r Here an object xe Sl is totally defined if € where, 51 .'.'= SU{l} x=yeS and only if pt( x ,x ) = 0 Definition 3.2 The OpenBalls forapmetric Bp€(x) ',= p: UXU+fr { y.A arethesetsoftheform, I p(x,y)<€ } foreach €>0 and xeA Theorem 3.1 Thesetofallopenballsof apmetricp; Proof UXU +fr isthebasisofatopology TIp] on U : Proof using Lemma 2.1 Suppose Then, A p: UXU -+fr = isapmetric. UaeA BPp1xil+l (x/ and, foranyballs BP€(x) and Bn6(l), BPt(x) n Bo5(t) = U{ Ben(z) I zeBn6(x)nBr5(t) where, q .'.'= p( z,z) + min{ t-p( x,z ), 6-p(y,z ) } D CUNY '92 7 Theorem 3.2 Foreachpmetric p: UXU+91 , xeBP6(a) Proof I openbalt BP€(a), and 3 6>0 .r € Bn6(x) I xeU, BPe(a) : Suppose p: UXU ->fr Suppose I e BPs(a) isapmetric. Then p(x,a) < € Let 5 ::= € -p(x,a) + p{ r,x) Then 6>0 as trp(x,a) Also, p(x,x) < 6 as €rp(x,a) Thus x e Be5(x) Supposenowthat y e BP5(x) p(y,x) < 6 p(y,x) < € -p(x,a) + p(x.x) p(y,x) + p(x,a) - p(x,x1 p(y,a) < 6 y (byPa) e Bn6(a) Thus Br5(x) I BP€(a). u Theorem 3.3 Pmetric topologies are To Proof . : p: UXU +fr Suppose x+y e U. Suppose isapmeric. Pl & P2 p(x,x) < p(x,)" ) or p(y,y) < p(x,y) Thenfrom p(x,x)<p(x,y) then, xeBp6(x) AylBe6(x) where, f .'.'= (p( x,x)+p(x,y))/2 Wlog suppose U Definition 3.3 Foreachpmetric p: UXU +X, Vx,yeA CUNY '92 x Kp ) (o S UXU e isthebinaryrelationdefinedby, p(x,x)=p(r,yl Theorem 3.4 Foreachpmetric Proof p: UXU -rfr, <p isapartialorder. : Suppose p: UXU -+fr isapmetric. We prove POI thru PO3 (POl x((px as p(x,x)=p(x,x) Vxe(J ) (PO2) Vx,y eU xKr! AyKpx + p(x,x)=p(x,y)= p(y,y) + x=! (byP3) (bypl) (PO3) V x,J,zeA. xKo! A y Koz + p(x,x) = p(x,y) A p(y,y) = p(y,z) (byPa) but, p(x,z) S p(x,y)+p(y,z)-p(y,y) p(x,7) < p(x, x) (byP2) p(x,z) = p(x,x/ ( by definition of ((o ) x ((o z tr Theorem 3.5 Every pmetric topology is an Alexandrov Topology. Proof: p: UXU +fr then we have to show that f Suppose isapmetric, tp I c tr[ <p ] . It is sufficient to show that, Vxe(1 Suppose , €>0 < p(x,y) + p(y,z) - p(y,y) = p(x,y) Thus, z e BPt(x) lyeBpt(x)} and €>0 aresuchthat yeBtr(x) x,J,zeU Then, p(x,z) Bp€(x) = U{{zly<<rz} andy<<pz. (byPa) as ! Kpz . tr CUNY '92 I Theorem 3.6 Foreachpmetric <p] rfandonlyif, BPr(t) = {ylx<<pY} 3€>0 Vxe(J Proof {[ p]=TI p: UXU+X, : Supposefirstthat, Then, V x e (J 3 € > 0 BP€(x) = { y I x <<ol } = urco{ylxKof voetr[ <<r] o } e rt pl TI <p] c Tt Pl {t p I : TI <, ] ( b;" Theorem 3.5 ) , TI p ] - tr[ <p ] Then, V x e(l { y I x Kol } e ft pl Thus , by Theorem 3.2 , x e BP€(x) c { y lxKr! V xeU 3 €>0 Suppose now that } But, if x e Bpt(x) then t y I x <<ol ) .g BP€(x) BPt(x) = {ylxKol} 3€>0 Thus, VxeU tr 4. Partial & Theorem Quasi Metrics 4.1 Foreachpmetric p: UXU ->fr Proof q: UXU +fr where, q(x,Y) .'.'= P(x,Y) - P(x,x) V x,y e U isaquasimetricsuchthat thefunction ftpl = Tt ql and Ko= ((n. : Suppose p: UXU -+fr Let q: UXU -+9l Vx,yeU We show first that CUNY '92 q isapnetric. besuchtha:, q(x,)'t = P(x,Y)-P(x,x) is a quasi metric by proving Ql and Q2. 10 V x,! e (J (Ql+) (Ql q(x,y) =q(y,x)=0 Vx,yeIJ - p(x,y)-p(x,x) = p(y,x)-p(y,y) = :+ p(x,x) - p(x,y) = p(y,y) (byP3) (bypl) - x =y e) (Q2) Thus x = ! - q(x,y) = 0 (bydefinirion ofq ) 0 p( x,z) S p( x,y)+p(y,z)-p(y,y) V x,y,ze(J = q(x,z) S q(x,y) + q(y,z) q is a quasi metric Now, Tt . pl = ft ql V xe(J, &s, E>p(x,x) Bpt(x) = V xeU,0<€<p(x,x) V xe(.1 , €>0 and by \*mma 2.2 Finally , ((p Be6_o1ra1 (x) BP€(x) = O Bc€(x) = BPr*4rr. (x) . = Kq V x,y e (l &s , p(x,x) = p( x,y) e u q(x,y) = 0 Definition 4.1 A Weighted Quasi Metric overaset U isapair ( q ,ll ) consistingof aquasimetric q: UXU --+9t anda Weight Function ll: (l -+fr where, (WQ) Aquasimetric Vr,!e(J q is Weightable q(x,y) +lxl ifthereexistsaweightfunction = ll q(y,x)+ lyl suchthat (q,ll) is a weighted quasi metric. Theorem 4.2 Foreachweightedquasimetric <q,ll> overaser u thefunction p: uXu -+fr where. V x,y eU isapmericsuchthat CUNY '92 p(x,y) .'.'= q(x,y) + lxl {t pl = TI ql and ($= ((0. 11 Proof : Suppose ( q ,ll ) isaweightedquasimetricoveraset U. I"rt p: UXU --rfrbesuchthat, p(x,y) .'.'= q(x,y) + lrl V x,y eU We show first that p proving is a pmetric by Pl thru P4 . (Pl =) Trivial (Ple) p(x,x) = p(x,y) = p(y,y) V x,! e(J + lxl = q(x,y)+lrl = lyl (byQl) = lxl = q(x,y) + lxl = q(y,x) + ly | = lyl (bywQ) :+ q(x,y) -- q(y,x) = = x=) (P2) Vx,yeU V x,y e U V x,y e U (P3) Vx,ye(I (P4) Vx,y,ze(J Thus p Now, (byel lrl p(x,x) q(x,y) +lxl = q(y,xi +lyl p(x,y) = p(y,x) V x,Y e(l is a pmetric Tt ) 0 (byWQ) q(x,z) V x,y,z e (J q(x,z) + lxl V x,y,z e (I p(x,z) . pI = ft qI asbyLemma2.2, V xe(J,6>lxl BP€(x) = Bq€_*t(x) V xe(J ,0< 6-<lrl V xe(J, €>0 CUNY '92 0 Bp€(x) = O Bc€(x) = Bpe*w(x) 12 Finally = Kq ((p , &S , p(x,x) =p(x,Y) e Vx,yeU q(x,Y)=0 tr Note that Theorems 4.1 & 4.2 establish an algebraic equivalence restricted form of the quasi metric Theorem between the partial metric and a . 4.3 Not every quasi metric is weightable Proof : I-et q : {a,b,c}2 I {0,1,2,3} q(a,b) = 0 Q(a,s) = I Q(b,c) = 3 Q(b,a) = 2 Q(c,a) = I Q(c,b) = 0 Supposethatthereexistsaweightfunction for q, betheuniquequasimetricsuchthat, ll : { a,b,c} + {0, 1,2,3 } then, 'b' . ', :*' "'j .i,;,"?ri,',: = lcl + q(c,a) ( lcl + q(c,b) ) + I = ( lbl + q(b,c) ) + I (bywQ) (bywQ) (bywQ) But this is impossible. U Unfortunately Theorem 4.3 does not answer the question of whether or not every quasi metric topology can be defined using a partial metric. Theroem 4.3 only shows that the method of defining a pmetric for a quasi metric topology using a weight function will not always work. The next result does answer the question for finite quasi metric topologies. Theorem 4.4 q: UXU -rS overafiniteset U thereexissapmetric suchthat Tt pl = Tt ql and Kp--<q. Foreachquasimetric p:UXU-+9t CUNY '92 13 Proof Outline : Lrt P .'.'= < U , <<q > be the partial order induced by a quasi metric q: UXU -+9t overafiniteset U. Let P" c 2u subsets of U \. be the set of all chains Lrt $ : U -+ t't VxeU be the function where $(x) Then it can be shown that the V x,y e U )tut+t in P p( ( i.e. all non - empty totally ordered , .'.'= )nax{ lcl I ceP* A lubc <<qx } function p : U X U -) o) x,y ) where , .'.'= met{ Lrr"$(x) f cep* A lubcKox A isapmetricsuchthat tubc<<, l} Ttpl = ft ql and ((p= <c. tr An importanr implication of Theorem 4.4 is that any finite partial order can be defined by a parrial metric . 5. Metrics Definition & Partial Metncs 5.1 A Weighted Metrtc overaset U isapair < d ,ll > consistingof ametric d: UXU -tfr anda Weight Function ll: -rfr where, (WM) Ametric d Vx,JeU d(x,y) isWeightable ifthereexistsaweightfunction ll suchthat <d,ll> isa weighted metric. The next two results show the algebraic equivalence between the partial metric and the weighted metric . Theorem 5.1 Foreachpmetricp:UXU+sthepair(p^:UXU+9|,||:U+fr> where, cuNY '92 1 4 Vx,ye(Jp^(x,y)::=2'p(x'y)P(x'x)-p(y'y) lxl .'.'= P(x,x) V x eU is a weighted metric such that , TtPl I tr[ P^] , and' V x,y e (J p(x,y) ::= Proof : ( p^(x'y) + lrl + lyl ) / P : UX[] 1S isaPmetric' Then, by Theorem4'l , thefunction q : SuPPose V x,y eU is a quasi metric such Tt Thus, byLemma2'6, thefunction isametricsuchthat, XU -) fr where' q(x,Y) ""= P(x'Y) - P(x'x) that tt p I = V x,y e (l U 2 q I ' { : UXU + fr where' f (x'y) ::= 2'p(x'y) - p(x'x) - p(y'y) V x'y e (J {(x'y) = q(x'y) + q(y'x) Thus, tt Pl I ft{l Finally, wMholdsasbyP2 Vx'y€u f(x'y) > lrl lyl tr Theorem 5.2 Foreachweightedmetric where overaset U thefunction p: UXLI _, fr , V x,y e (J suchthat Proof <d,||> d=p^ p(x,y) ""= ( lxl + lyl + d(x'y) ) / lrl=P(x'x) and VxeU 2 : as Ml = Pl and, WM + PZ and, MZ + P3 md, Ml&M3 + P4 tr CUNY '92 15 We now turn to constructing a contraction mapping theorem for partial metrics. The beauty of the result in Theorem 5.3 is that its formulation is virtually the same as Banach's original theorem for metric spaces. Definition 5.2 Foreachpmetric Sequence if , p : uXu -) N, V€>0 3ke(t andforeach Vn,m>k x e u(J, x isa p( Xn,X^) cauchv < t Definition 5.3 A pmetric p : UXU -) X is Complete ifforeachCauchysequence X e uIJ exists aeU suchthat, 3 limn-os p( Xn, a ) - A Note that these definitions for Cauchy sequence and complete pmetic are consistent there with the usual metric definitions. Theorem 5.3 ( Partial Metric Contraction Mapping Theorem ) Foreachcompletepmetric that, 3 0Sc<I fintly, Proof p:uXu -r N, f :U+U such p(f(x),f(y)) Vx,yeU thereexists a unique andforeachfunction a e (J such that a = f( a ), and secondly p( a, a )=0 : Suppose p,f , & c Suppose are as in the statement of the Theorem u e U. Webeginbyshowingthat A n e CUNY '92 . u .f"(u) isaCauchysequence. 16 Thusas, Vne(t p(f"(u),f"(u)) S ,"*p(u,u) weseethat ,l n e u) . f"(u) isaCauchysequence. Thus as p is complete we can choose a e U limn-* p( f"(u) , a ) - 0 But, p(f(a),a) V n e ut = 0 BS, p( f(a) , a ) ) p( f"+1(u) such that, + p( f"*I(u), yn-t(u) ) a) S c xp( a , f"(u) ) + p( f"rI(u) , a ) Thus a = f( a) eU is ,r.h thut b p(a,b) = p(f(a),f( b) Suppose b Thus, p(a,b)=0, Thus the fixed point of / Finally, p(a,a)=Q tr 6. Conclusions & andso is unique f( b), rhen, a=b . as,p(a,a) = p(f( a),f(a)) Further Work The principle conclusion from the research in this r€port is that generalised metric topology has a largely unexplored potential in the field of partial sder denorational semantics. This conclusion is justified for the following reasons. Firstly , this research supports the quasi metric approach used in [Sm87] and the metric approach in [La8fl to model panal order topologies. Secondly , distance function. Thirdly , we have shown that such work can be conducted using a symmetric we have shown that such work can be conducted within metric topology itself by adding Ore concept of a weight to points in a metric space. Finally , we have shown that Banach's Contraction Mapping Theorem can be generalised for proof rheory applications in Scon topologies. Whether or not an arbitrary quasi metric topology can be defined by a weighted merric is an open question which needs an answer in order to decide exactly which topologies can be defined using our approach. However , it is unlikely that there exists any simple algebraic method for converting an arbitrary quasi metric into a weighted quasi metric. Similarly , it is not yet known exacrly which Scott topologies can be defined using this approach. If a favourable answer to this question can be established then perhaps weighted metric models for the lamMa ialculus could be feasible. CUNY '92 17 The Wadge CycleswnTest [WaS1] for proving Kahn data flow networks free of deadlock within Kahn's fixed point semantics. By formulating the domain of all finite & infinite streams as a partial metric space we plan to use the Partial Metic Contraction Mapping Theorem to prove Wadge's test. Beyond this we hope to use partial metrics to provide a denotational semantics for intensional programming languages such as LUCID [Va&A85] which have been inspired by Kahn's work [Ka74]. For more results on partial metrics such as partial metric products , partial metric sums , and partial metric function has yet to receive a proper denotational proof sryces 7. see [Ma92] . References [dB&282) Processes and the Denotational Semantics of Concunency J. W. de Bakker & J. I. Zucker , Dept. of Computer Science report NV 249 lKa74l , /82 [Ko88] Stichting Mathematisch Centrum. The Semantics of a Simple l-anguage for Parallel Gilles Kahn All , , Proc. IFIP Conf. 1974 , Processing , pp. 471- 475 . from Generalised Metrics, Ralph Kopperman American Mathematical Monthly , Vol. 95 , No. 2, February 1988 Topologies come , . tk87l The Versatile Continuous Order, J. D. Lawson , Mathematical Foundations of Programming Language Semantics Tulane 1987 Ma92l , LNCS 298 Partial Metic Spaces, Theoretical in Research Report 212, of Warwick al., Springer 1988 . An Spaces , W. A. Sutherland , . o Metic and Topological Spaces, hess . Oxford 1975 Introduction t . of Dataflow Deadlock, W. W. Wadge , Computer Science 13 , pp. 3 - 15 Extensional Treatment Theoretical CUNY '92 Springer 1988. - Unifurmities : Reconciling Domains with Metric M. B. Smyth , Mathematical Foundations of Programming Language Semantics , 3rd Workshop , Tulane 1987 , in LNCS 298 , Clarendon [Wa81] 3rd Workshop, 8th British Colloquium for S. G. Matthews , University , Quasi eds. M. Main et. [Su75] eds. M. Main et. al. , Computer Science, March 1992 , Depr of Computer Science [Sm87] , , . 18 tWa85l Lucid , the Dataflow Programming I'anguage W. W. Wadge & E. A. Ashcroft, , APIC Studies in Data Processing No. 22 , Academic Press 1985 tvi89l CUNY '92 Topology via l-ogic, S . Vickers , . Cambridge University Press . 19