A necessary relation algebra for mereotopology

Ivo Dntsch
School of Information and Software

Gunther Schmidt, Michael Winter

Engineering

Department of Computer Science

University of Ulster

University of the Federal Armed Forces Munich

Newtownabbey, BT 37 0QB, Northern Ireland

85577 Neubiberg, Germany

I.Duentsch@ulst.ac.uk

{schmidt|thrash}@informatik.unibw-muenchen.de

Abstract
The standard model for mereotopological structures are Boolean subalgebras of the complete
Boolean algebra of regular closed subsets of a nonempty connected regular T0 topological space
with an additional contact relation C defined by
xCy



   0/ 

x y

A (possibly) more general class of models is provided by the Region Connection Calculus (RCC)
of Randell et al. [34]. We show that the basic operations of the relational calculus on a contact
relation generate at least 25 relations in any model of the RCC, and hence, in any standard model
of mereotopology. It follows that the expressiveness of the RCC in relational logic is much greater
than the original 8 RCC base relations might suggest. We also interpret these 25 relations in the
the standard model of the collection of regular open sets in the two-dimensional Euclidean plane.

1 Introduction
Mereotopology is an area of qualitative spatial reasoning (QSR) which aims at developing formalisms
for reasoning about spatial entities [1, 12, 31, 32]. The structures used in mereotopology consist of
three parts:
1. A relational (or mereological) part,
2. An algebraic part,
3. A topological part.

 Co-operation for this paper was supported by EU COST Action 274 Theory and Applications of Relational Structures
as Knowledge Instruments (TARSKI), www.tarski.org
Address from 2002: Department of Computer Science, Brock University, St. Catherines, Ontario, Canada, L2S 3AI

The algebraic part is often an atomless Boolean algebra, or, more generally, an orthocomplemented
lattice, both without smallest element.
Due to the presence of the binary relations part-of and contact in the relational part of mereotopology, composition based reasoning with binary relations has been of interest to the QSR community,
and the expressive power, consistency and complexity of relational reasoning has become an object
of study [24, 34]. The first time that the relational calculus has been mentioned in (modern) spatial
reasoning was in the interpretation of the 4-intersection matrix in [19], see also [39].
It has been known for some time, that the expressiveness of reasoning with basic operations on binary
relations is equal to the expressive power of the three variable fragment of first order logic with at
most binary relations [see 43, and the references therein]. Thus, it seems worthwhile to use methods
of relation algebras, initiated by Tarski [42], to study contact relations in their own right, and then
explore their expressive power with respect to topological domains.
The Region Connection Calculus (RCC) was introduced as a formal structure to reason about spatial
entities and the relationships among them [34]. Its models are basically atomless Boolean algebras
with an additional contact relation which satisfies certain axioms. A standard model of the RCC is the
Boolean algebra of regular open sets of a regular connected topological space, where two such sets
are in contact, if their boundaries intersect. However, these are not the only RCC models.
Gotts [22] explores how much topology can be defined by using the full first order RCC formalism.
Our aim is similar: We are interested which relations can be defined with relation algebra logic (i.e.
the three variable fragment of first order logic) in the algebraic setting of the RCC, interpreted in a
topological context.

2 Relation algebras
The calculus of relations has been an important component of the development of logic and algebra
since the middle of the nineteenth century. Since the mid-1970s it has become clear that the calculus
of relations is also a fundamental conceptual and methodological tool in computer science. Some
examples are program semantics [7, 35, 36, 44], program specification [5] and derivation [6], and last
but not least qualitative spatial reasoning. For a detailed overview we invite the reader to consult [9].
Let U be a nonempty set. We denote the set of all binary relations on U , i.e., the powerset of U U ,
by Rel
U  . We usually indicate the fact  x y  R for R  Rel
U  by xRy. Furthermore, we define for
R S  Rel
U 
(2.1)
(2.2)
(2.3)
(2.4)

  x y :
 z  U  xRzSy 
  x y : yRx 
def
R
def
 y : xRy 
xR
def
 x : xRy 
Ry

R S

def

Composition
Converse
Image of x under R
Inverse image of y under R
2

We also let 1 be the identity relation on U , and V U U be the universal relation. The full algebra
of binary relations on U is the algebra of type  2 2 1 0 0 2 1 0 

 Rel
U    !" 0/ V # 1 
We shall usually identify algebras with their base set. Every subalgebra of Rel
U  is called an algebra


of binary relations (BRA). If Ri : i  I %$ Rel
U  , we let  Ri : i  I & be the BRA generated by
 Ri : i  I  .
If an RA A is complete and atomic in particular, if A is finite , then each nonzero element is a
sum of atoms, and relational composition can be described by a matrix, whose rows and columns are
labelled by the atoms and an entry  P Q  is the set of atoms contained in P  Q. If 1 is an atom of A,
we omit column and row 1 .

( Ri : i  I  is a partition of V such that ' is closed under converse, and either Ri $ 1
 0/ for all i  I, we define the weak composition  w : ') "'+* 2, of ' as the mapping

If '
Ri  1 

Ri 

(2.5)

w Rj

 S -'

def

or

: S .
Ri  R j / 0/ 

Just as in the case of  , we can determine composition tables for  w . Note that Ri  R j $ Ri 
  w , we call the weak composition table extensional.
equality holds everywhere, i.e. when 

w R j;

if

An abstract relation algebra (RA) is a structure

 A 01 324 !5 0 1 # 1 

of type  2 2 1 0 0 2 1 0  which satisfies for all a b c

1.  A 01 324 !5 0 1  is a Boolean algebra (BA). Its induced ordering is denoted by

2.  A # 1  is an involuted monoid, i.e.

(a)  A # 1  is a semigroup with identity 1 ,
(b) a

a 7
a  b 

b  a.

3. The following conditions are equivalent:

a  b8 2 c 

(2.6)

a  c8 2 b 

c  b9 2 a 

0

The properties (2.6) are sometimes called the complement-free Schrder-equivalences. They are
equivalent to the Schrder-equivalences introduced by Schrder in [38]
(2.7)

a b

:<;

a 7 b 6= c

:<;

c  b 6> a 

Each BRA is an RA with the obvious operations, but not vice versa [29].

An RA A is called integral if and only if

a b

(2.8)
for all a b

:<;

0 or b

A. It is well known [25] that

A is integral if and only if 1 is an atom of the Boolean part of A.

The logic of RAs is a fragment of first order logic, and the following fundamental result is due to A.
Tarski [43]:

Theorem 2.1. If Ri : i  I ?$ Rel

U  , then  Ri : i  I & is the set of all binary relations on U which

are definable in the (language of the) relational structure  U Ri : i  I & by first order formulas using
at most three variables.
If a b are elements of a RA A, we define the right residual of a and b by
a@

(2.9)
a@

b is the largest relation c such that a  c

Analogously, we define the left residual a B

aB

(2.10)

 %
a A b 

def

b.
b of a and b as
b

a B b is the largest relation d such that d  b

defined by
(2.11)

syq
a b 


a@

def

 %
C a  b 

def

b 92D
a B

a. The symmetric quotient syq

a b  of a and b is
b

 %
a A b9 2E%
C

a  b 

In a BRA the residuals and the symmetric quotient of R and S can be characterised by
(2.12)
(2.13)
(2.14)

F  x y
   x y
F
RB S
F  x y
syq
R S 
R@

: Rx
: yS


: Rx

Sy 

xR 
Sy 

The following properties of the residual will be needed later:

Lemma 2.2. [15, 33] In every RA the following holds:
1. c @

c is reflexive and transitive.

2. If c is reflexive and symmetric, then

c @

c  A
c @

c G6 c.

In a BRA, there is an elegant way to characterise a subset of the universe U . To this end, associate the
relation
def
  x y : x  U y  M 
m

with the subset M $ U . It is easy to see that m satisfies m V  m where V is the greatest relation over

the universe U . A relation m which satisfies m V  m is called a vector. Analogously, a one-element
subset or an element of U may be described by a vector which is univalent m  m $ 1 ; these relations
are called points.
Given an ordering P, i.e., a reflexive, transitive and antisymmetric relation, one may be interested
def

in lower bounds lbP
m  of subsets characterised by a vector m. This vector is given by lbP
m 
 %
m H P   m @ P is the vector of
%
m H P  m @ P 
P B m  . Analogously, ubP
m  def
def
 lbP
m I ubP
lbP
m 3 and lubP
m  def

upper bounds of m. Last but not least, the relations glbP
m 
ubP
m J lbP
ubP
m 3 are either empty or a point, describing the greatest lower bound and the least
upper bound, if they exist, respectively. More details about the relational description of orderings,
extremal elements and their properties may be found in [37].
For properties of relation algebras not mentioned here, we refer the reader to [10, 24, 37], and for
Boolean algebras to [26].

3 Mereology
Mereology, the study of part-of relations, was given a formal framework by Lesniewski [27, 28]
as part of his programme to establish a paradox-free foundation of Mathematics. Clarke [11] has
generalised Lesniewskis classical mereology by taking a contact relation C as the basic structural
element. The axioms which C needs to fulfil are
(3.1)

C is reflexive and symmetric,

(3.2)

Cx

Cy implies x

 y

It was shown in [15] that the extensionality axiom (3.2) may be replaced by
C @ C is antisymmetric

(3.3)
i.e.

syq
C C K$ 1

(3.4)

The term mereology has nowadays become (almost) synonymous with the study of part-of and
contact relations in QSR.
If C is a contact relation we set
(3.5)

(3.6)

PP

def

def

C @ C

P . 1

part of

proper part of
5

Lemma 2.2 and (3.3) tell us that P is a partial order which we shall call the part of relation (of C). We
also write x 6 y instead of xPy. PP is called the proper part of relation.
We now define the additional relations
(3.7)

(3.8)

PO

(3.9)

EC

(3.10)

T PP

(3.11)

NT PP

(3.12)

DC

(3.13)

DR

def

def

def

def

def

def

def

P  P

overlap

O L-
P  P

partial overlap

C . O

external contact

PP M
EC  EC 

tangential proper part

PP . T PP

nontangential proper part

disconnected

discrete 

Given a contact relation C, we will use the definitions (3.5) (3.13) of the relations throughout the
remainder of the paper. An example of such relations is given in Figure 1. The domain is the set of all
 /
nonempty closed disks in the Euclidean plane, and xCy :<; x  y / 0.

TPP

NTPP

PO

EC

DC

Figure 1: Circle relations

Mereological structures also have an algebraic part: If C is a contact relation on U and 0/ O N X

U , then x is called the fusion of X , written as X , if
(3.14)

SUT y R

U VXW xCy Y<Z

S\[ z R

UQ x R

X V yCz]^

A model of mereology is a structure _ U Q C Q ` , where C is a contact relation, and the fusion exists
for all nonempty X P U . If
(3.15)

C O OQ

then _ U Q C Q ` is a model of classical mereology, since contact C is definable by part of P as

C O P a P.
Note that the definition of a model of mereology is not first order; a weak model of mereology is a
structure _ U Q C Qcb ` , where C is a contact relation, and for all x Q y R U , the fusion x b y exists.

Given a model of mereology  U C  , one can define additional operations on U as follows [11]:

x : xCx 


xd
y : y
C C  x 
 
X z : zPx for all x 

(3.16)

(3.17)
(3.18)

Universal element
Complement
X

Product

Observe that d and are partial operations - for example, 1d does not exists, and neither does x y 
if xDCy -, and that they require completeness of fusion. Biacino and Gerla [8] have shown that
the models of mereology are exactly the complete orthocomplemented lattices with the 0 element
removed, and
xCy :L;

6>
/

Models of classical mereology arise from complete Boolean algebras B with the 0 element removed
as shown in [41]; here P is the Boolean order.

4 The Region Connection Calculus

The Region Connection Calculus (RCC) was introduced in [34] as a tool for reasoning about spatial
phenomena, and has since received some prominence. It uses a contact relation C which fulfils the
conditions (3.1) and (3.2).

A model for the RCC consists of a base set U R  N, where R N are disjoint, a distinguished 1  R,
def


a unary operation d : R0 * R0 , where R0 R e 1  , a binary operation 0 : R R * R, another binary
operation 2 : R R * R  N, and a binary relation C on R. In order to avoid trivialities, we assume that
f U f&g 2.
The RCC axioms are as follows:
RCC 1.
RCC 2.
RCC 3.
RCC 4.

RCC 5.
RCC 6.
RCC 7.
RCC 8.

Uh
U
h
U
h
U
h

 R  xCx

x y  R Ei xCy ; yCxj
x  R  xC1
x  R y  R0  ,
(a) xCy d :<;lk xNT PPy
(b) xOy d :<;mk xPy

Uh x y z  R Ei xC
y 0 zn:<; xCy or xCzj

Uh x y z  R Ei xC
y 2 zo:L;p
 w  R E
wPy and wPz and xCw \j

Uh x y  R Ei4
x 2 y R :<; xOyj

If xPy and yPx, then x y.
x

q

We shall in the sequel assume without loss of generality that N

0  . Axioms RCC 1, RCC 2,
RCC 5 and RCC 8 show that  R C 0" is a weak model of mereology. It is, however, not a model of
mereology in the sense of Section 3, since it has a different definition of complement: In the RCC
models, each proper region x is connected to its complement x d , which is impossible in models of
mereology. It was shown in [14] and [40] that the algebraic part of an RCC model, i.e.
U 324 01 d  , is
a Boolean algebra. Each atomless Boolean algebra can be made into an RCC model by defining an
appropriate contact relation [13].
Notice, that some of the axioms above may be written in a relation algebraic manner as follows:
RCC 1. 1J$ C

RCC 2. C $ C

RCC 8. syq
C C K$ 1 .

In the original RCC, the relations

1 T PP T PP NT PP NT PP PO EC DC

(4.1)

were considered base relations in a system called r 8 . Somewhat earlier, Egenhofer and Franzosa [16]
arrive at a similar set of relations by purely topological considerations. Seeing that the largest element
1 is RA definable from C, it was noted in [15] that investigation of the RCC can be restricted to the


set U R L 1  , and that EC and PO split into the disjoint nonempty relations

 %
PP  PP  PP  PP 
def

ECN EC . ECD
def

PON # M
PP  PP #M
PP  PP
def

POD # M
PP  PP #.%
PP  PP

(4.2)

ECD

(4.3)
(4.4)
(4.5)
where #

 %
P 

def

P is the incomparability relation. It is not hard to see that

xECDy :<;

 ysd

 / 1


xPONy :<; x#y x 2 y / 0 x 0 y / 1
xPODy <
: ; x#y x 2 y  / 0 x 0 y  1
xECNy :<;

xECy and x 0 y

In the sequel, we shall regard

(4.6)

1 T PP T PP NT PP NT PP PON POD ECN ECD DC

as defined above as the basic relations in terms of which other relations will be defined below. The
system induced by these relations will be called r 10 . The weak composition of r 10 is given in Table
1; it is worth pointing out that the table does not have an extensional interpretation, i.e. there is no
8

RA whose composition is given by Table 1. The reason for this is that the table is not associative:
Consider for example,

T PP 

NT PPI POD

/


T PP 0 NT PP 0 PON 0 POD

T PP 0 NT PP 0 PON 0 POD 0 ECN 0 ECD

T PP t
NT PP  POD 

Nevertheless, the base relations are the atoms of a semi-associative relation algebra in the sense of
Maddux [30].
Using the relation ECD, another RCC axiom can be written in algebraic form as follows:
RCC 4. (a) C  ECD
(b)

NT PP


O  ECD

Let V be the greatest relation over R. Notice, that the property

lub u Cv

(4.7)

w
R #

R V

forces the algebraic part of a RCC model over R to be a complete BA without a least element since for
every nonempty vector m the least upper bound lub u Cv C w
m  is also nonempty. Under this assumption,
the greatest element 1 is characterized by the relation lub u Cv C w
V  . Furthermore, if we require
lub u Cv

(4.8)

w
R I C 

R  C for all relations R

then the relation algebraic counterparts of the remaining axioms RCC 3, RCC 5, RCC 6 and RCC 7
are provable.
Lemma 4.1. For all nonempty vectors m we have the following:
RCC 3: lub u Cv
RCC 5: lub u Cv
RCC 6: glb u Cv
RCC 7: glb u Cv

w
V # C  V ,
 m  C,
C w
m I C
 lb u
m I C,
Cv C w
C w
m I C
 /<
: ; lb u Cv C w
m  /
C w
m / 0
C

/
0.

Proof. RCC 3 and RCC 5 follow from 4.8. Notice, that we have

x&

Notice, that the inclusion

w
m

w
lb u Cv Cw
m 3
A proof may be found in [37]. RCC 6 and RCC 7 follow from
x& and 4.8 resp. 4.7.
glb u Cv

lub u Cv

in 4.8 can be proven.

9

TPP

TPP, NTPP

1, TPP, TPP,
PON, POD

NTPP

TPP,
NTPP,
PON, POD

TPP,
NTPP,
PON, POD

POD

TPP,
NTPP,
PON,
POD,
ECN, ECD
POD
TPP,
NTPP,
PON, ECN, DC

TPP

NTPP

NTPP

PON

POD

ECN

10

ECD
DC

TPP

ECN
DC

TPP,
NTPP,
PON,
POD,
ECN, ECD
ECN, DC

TPP,
NTPP,
PON, ECN, DC

NTPP

TPP,
NTPP,
PON, ECN, DC

TPP, NTPP

1, TPP, TPP,
PON, ECN, DC

TPP

POD
TPP,
NTPP,
PON,
POD,
ECN, ECD, DC

TPP,
NTPP,
PON, POD

POD

1, TPP, TPP,
NTPP, NTPP,
PON, POD
TPP,
NTPP,
PON, POD

NTPP

TPP,
NTPP,
PON, POD

NTPP

NTPP

DC
DC

TPP,
NTPP,
PON,
POD,
ECN, ECD, DC
DC

TPP,
NTPP,
PON, ECN, DC

1, TPP, TPP,
NTPP, NTPP,
PON, ECN, DC
NTPP

NTPP

TPP,
NTPP,
PON, ECN, DC

NTPP

Table 1: The
10

PON
TPP,
NTPP,
PON, ECN, DC

TPP,
NTPP,
PON, ECN, DC

1, TPP, TPP,
NTPP, NTPP,
PON,
POD,
ECN, ECD, DC
TPP,
NTPP,
PON, POD

TPP,
NTPP,
PON, POD

TPP,
NTPP,
PON, ECN, DC

TPP,
NTPP,
PON, POD

TPP,
NTPP,
PON, ECN, DC

PON

TPP, NTPP
NTPP

1, TPP, TPP,
NTPP, NTPP,
PON, POD
TPP, NTPP

TPP,
NTPP,
PON, POD

TPP,
NTPP,
PON,
POD,
ECN, ECD, DC
POD

TPP,
NTPP,
PON,
POD,
ECN, ECD
POD

POD

weak composition table

z
TPP
TPP,
NTPP,
PON, ECN, DC

1, TPP, TPP,
PON, ECN, DC

TPP, NTPP

TPP,
NTPP,
PON, ECN, DC

TPP,
NTPP,
PON, POD

TPP,
NTPP,
PON,
POD,
ECN, ECD
DC

ECN, DC

ECN

1
NTPP

TPP

TPP, NTPP

PON

POD

DC

POD

ECN

ECD

NTPP
1, TPP, TPP,
NTPP, NTPP,
PON, ECN, C

TPP,
NTPP,
PON, ECN, DC

NTPP

TPP,
NTPP,
PON,
POD,
ECN, ECD, DC
TPP,
NTPP,
PON, ECN, DC

DC

TPP,
NTPP,
PON, ECN, DC

DC

DC

5 Basic relational properties

In this section, we shall collect some properties of the relations listed in (4.6), which follow from the
RCC axioms. These will be used in the next section for the definition of the relation algebra. We
commence with several basic connections which were already proved in [15].
1. 1J$ NT PP  NT PP, i.e. for all z there is some x with xNT PPz.

Lemma 5.1.
2. ECN

T PP  ECD, i.e. xECNz

:<;

xT PPz d .

3. If xDCz, then xT PP
x 0 z  .
4. xNT PPz and yNT PPz

:L;
x 0 y NT PPz.

5. If xNT PPz, then

x d 2 z  T PPz.
Our second lemma deals with compositions of P DC T PP and NT PP.
1. DC  P $ DC, i.e. xDCy and z

Lemma 5.2.
2. NT PP

ECD  NTPP  ECD, i.e. xNT PPy

3. P  NT PP

NT PP, i.e. x

4. NT PP  T PP
5. NT PP  P

y imply xDCz.

:<;

y d NT PPx d .

y and yNT PPz imply xNT PPz.

NT PP.

NT PP.

6. T PP  NT PP
7. 1

NT PP.

NT PP  NT PP, i.e. for all x there is some z with xNT PPz.

Proof. 1. Consider the following computation:

DC  P $ DC :<;| C 7%
C C  C G$= C

:<;|

C  C

$=

C  C 

by definition of DC and P
by 2.7

2. Consider the following computation:

 %
C 

NT PP

ECD 

by RCC 4a

ECD 7%
ECD  C # ECD




since ECD is a bijection

ECD 7%
C  ECD   ECD

since C and ECD are symmetric

ECD  NT PP  ECD 

by RCC 4a

3. Let xPyNT PPz, and assume that k xNT PPz. Then we have xCz d and k yCzd by RCC 4a. It follows
that xCz d DCy holds, i.e. x
C } C  y and hence k xPy by the definition of P. But this is a contradiction.
11

4. $ : We prove the stronger assumption

x&

NT PP  P

NT PP

Let xNT PPyPz and assume that k xNT PPz. Then we have xCzd by RCC 4a. On the other hand,
xNT PPy implies xDCyd by RCC 4a. Since y 6 z and therefore z d 6 yd holds we conclude xDCz d by
1, a contradiction.
y : Let xNT PPz. With 5.1(1) choose some wNT PP
xd 2 z  , and set y
wNT PP
x d 2 z 

 ;
 ;
 ;
 ;
 ;
 ;
 ;

wNT PPx d

by

x&

w d 2 z. Then we have

and xd 2 z 6 xd

xNT PPw d

by 5.2(2)

xDCw

by RCC 4a

xDCw and xDCz d

by xNT PPz and RCC 4a

xDC
w 0 z d

xNT PP
w 0 z d
xNT PPy

by RCC 5

by RCC 4a
Definition of y

Furthermore, wNT PP
x d 2 z  implies wNT PPz, and by 5.1(5) we get y
5. $ was already shown in 4.
x& and y follows from 4.


wd 2 z T PPz.

6. $ : This follows from 3.

y : Let xNT PPz. With 5.1(1) choose some yNT PP

xd 2 z  . Then 5. gives us yNT PPz and yNT T Px d .
Using 5.1(4) and RCC 4a we get
x 0 y  NT PPz and yDCx. Together we conclude xT PP
x 0 y  NT PPz
by 5.1(3).
7. Let x

U . By Lemma 5.1(1), there is some y

U such that yNT PPx d , and by 2 above, xNT PPy d .

Our next lemma exhibits some new arithmetical properties involving the algebraic operations.
Lemma 5.3.
2. ECD  DC
3.
4.
5.
6.
7.

1. xNT PPy and xNT PPz

:<;

xNT PPy 2 z.

:<; zNT PPx.


PON  ECD PON, i.e. xPONz :<; xPONzd .

T PP  ECD POD .%
ECD  NTPP  .
xECN  T PPz :<; xECN
x d 2 z  T PPz.
If x#z then xT PP  T PPz :<; xT PP
x 2 z  T PPz.

If x 2 z / 0 then x ~
T PP  T PP  z :<;

x 2 z NT PPx or
x 2 z NT PPz 
NT PP, i.e. x d DCz

12

8. If x#z then xT PP  T PPz

:<;

xT PP
x 0 z  T PPz.

9. If yNT PP
x 0 z  and yDCz then yNT PPx.
Proof. 1. This follows from 5.1(4) and 5.2(2).
2. Consider the following computation:
x
ECD  DC  z :L;

:L;
: ;
L
: ;
L

3.

 ;

x d DCz
zDCx d

by RCC 2

zNT PPx 

by RCC 4a

xNT PPz 

: Suppose xPONz. Then the definition of PON implies

 z
z  x

x 2 z / 0

x 0 z / 1
x

(5.1)
(5.2)
(5.3)
(5.4)

We have to prove (5.1)-(5.4) for z d instead of z. Consider the following computation:

zd
x
x 2 zd

x 0 zd

6


g zd 0 z  1 contradicting (5.4)


z d ; x 2 z 6 z d 2 z 0 contradicting (5.3)

0 ; x 6 z contradicting (5.1)

1 ; z 6 x contradicting (5.2) 
x

 ;

x0 z

is shown analogously.

4. $ : First, we show T PP  ECD $ POD. Suppose z d T PPx. Then, we have

x

 z

 x

x 2 z / 0/

x 0 z 1
z

since x
since z

z implies z d

x is a contradiction

since x 2 z
since z d

6 xd , a contradiction

0/ implies x 6 z d , a contradiction
x

and hence xPODz. Furthermore, consider the following computation:

T PP  ECD  ECD  NTPP

T PP  ECD  ECD  ECD 


T PP  NT PPI ECD
 0

NTPP  ECD

by 5.2(2)
since ECD is a bijection

13

which shows T PP  ECD $>%

ECD  NT PP  .

y : Suppose xPODz and x d

NT PPz. Then we conclude z d x since x 0 z 1 and x 2 z
Furthermore, we have z d T PPx because z d NT PPx implies x d NT PPz by 5.2(2), a contradiction.
5. We only have to show

 ;

/

/
0.

. Let xECNyT PPz. First we have

xECNy :<;

:<;
 ;

yECNx

symmetry of ECN

yT PPx d

by 5.1(2)

y 6 xd 2 z 

since yT PPz

Assume xNT PP
x 0 z d  . Then we conclude
xNT PP
x 0 z d

o:L;
 ;
:L;

xDC
x d 2 z 

by RCC 4a

xDCy

by 5.2(1)

xNT PPy d

by RCC 4a

But xECNy gives us xT PPy d by 5.1(2), a contradiction. Since x 6 x 0 z d we conlude xT PP

x 0 zd 
and using 5.1(2) again xECN
x d 2 z  . Assume x d 2 zNT PPz. Then we aim at yNT PPz by 5.2(3),
contradicting yT PPz. Since xd 2 z 6 z we get
xd 2 z  T PPz.

6. Again, we only have to show ; . Let xT PPyT PPz. Then we have y 6 x 2 z. Since x#z we have

x 2 z PPx. Assume
x 2 z NT PPx. We conclude yNT PPx by 5.2(3), a contradiction.

7. ; : Let x 2 z 0. Suppose w.l.o.g that

x 2 y  T PPx holds. The hypothesis x
T PP  T PP  z
implies that for all y,
yT PPx

 ;

T PPz 

Thus,
x 2 z 8 T PPz and hence
x 2 z  NT PPz since x 2 z

z.

:
: Suppose w.l.o.g. that
x 2 z  NT PPz holds. Furthermore, assume that we have xT PPyT PPz for
some y  U . Then we get y 6 x 2 z and by 5.2(3) yNT PPz, a contradiction.
8. Similarly to 6.
9. Consider the following computation:
yNT PP
x 0 z  and yDCz :<;|
xd 2 z d  NT PPy d and zNT PPy d

:<;|
xd2 yd0 z NT PPyd
:<; yNT PP
zd2E
x 0 z3
:<; yNT PP
zd2 x
 ; yNT PPx

by 5.2(2) and RCC 4a

by 5.1(4)
by 5.2(2)
by 5.2(5)

which finishes the proof.

The last lemma deals with some new relation algebraic properties of the relations listed in (4.6).
14

1. ECN  T PP  ECD

Lemma 5.4.

2.
ECN  T PP   ECD
3. T PP  T PP  ECD

T PP  T PP.

T PP  T PP.
ECN  T PP.


ECN  T PP  .
-
ECN  T PP # ECD  %
T PP  T PP .
-
ECN  T PP   ECD  %
T PP  T PP  .
-
T PP  T PP# ECD  %
ECN  T PP  .
-
T PP  T PP # ECD  %
ECN  T PP  .

4. T PP  T PP  ECD
5.
6.
7.
8.

Proof. 1. Consider the following computation:

ECN  T PP  ECD

T PP  ECD  ECN

by 5.1(2)

T PP A
ECN  ECD 




since ECN and ECD are symmetric

T PP A
T PP  ECD  ECD 
T PP  T PP 

by 5.1(2)
since ECD is a bijection

2. Consider the following computation:

ECN 

T PP   ECD

T PP  ECN  ECD

T PP  T PP  ECD  ECD
T PP  T PP

since ECN is symmetric

by 5.1(2)
since ECD is a bijection

3.-8. Follow from 1. and 2. since ECD is a bijection.

6 A necessary relation algebra

We are now ready to describe the relation algebra which is a subalgebra of every BRA generated
by the contact relation of any RCC model. Note that in this section, composition means relational
composition proper, and not the weak composition of (2.5).
The relations we are going to consider are shown in Table 2. The definitions of the relations PONY B,
PONZ and PODZ give rise to some simple questions answered by the next lemma.
Lemma 6.1.
2. PONY B

1. PONZ

T PP  T PP.

T PP  T PP.
15

Table 2: Atoms of

1
T PPA
T PPA
T PPB
T PPB
NT PP
NT PP
PONXA1
PONXA2
PONXB1
PONXB2
PONYA1
PONYA2
PONYA1
PONYA2
PONY B
PONY B
PONZ
PODYA
PODY B
PODZ
ECNA
ECNB
ECD
DC

3. PONZ

% ECN T PP
% ECN T PP
" ECN T PP
" ECN T PP

T PP
T PP
T PP
T PP






"

"

"

"
5
5
"
"
5
"
5
5
5

















"

"

"

"

"
"
5
5
"
5
5
"
5

%
%
"
"
"
"
"
"
5
5

%

"

"

5

"

5

"

5

PON ECN T PP
ECN T PP T PP T PP
T PP T PP
PON ECN T PP
ECN T PP T PP T PP
T PP T PP
PON ECN T PP
ECN T PP
T PP T PP
T PP T PP
PON ECN T PP
ECN T PP
T PP T PP
T PP T PP
PON
ECN T PP
ECN T PP T PP T PP
T PP T PP
PON
ECN T PP
ECN T PP T PP T PP
T PP T PP
PON ECN T PP
ECN T PP T PP T PP
T PP T PP
PON ECN T PP
ECN T PP T PP T PP
T PP T PP
PON
ECN T PP
ECN T PP
T PP T PP
PON ECN T PP
ECN T PP
T PP T PP
PON
ECN T PP
ECN T PP
POD
ECD NTPP
T PP T PP
POD
ECD NTPP
T PP T PP
ECD NT PP
ECN T PP T PP
ECN
T PP T PP


"
5

T PP  T PP.

4. PODZ

POD.

5. PODZ

T PP  T PP.

Proof. 1. Let xPONZz and assume x

C%
T PP  T PP3 z. Since xPONz implies x#z we may apply
5.3(8) and conclude x T PP
x 0 z  or z T PP
x 0 z  . Assume w.l.o.g. that x T PP
x 0 z  holds.
Since x x 0 z we conclude xNT PP
x 0 z  . Furthermore, we have
xNT PP
x 0 z 

 ;
 ;
: ;
<
 ;

xDC
x 0 z Cd

by RCC 4a

xT PP
x 0 xd 2 z d
xT PP
x 0 z d

xECN
x d2 z 

contradicting x
C-
ECN  T PP 3 z by 5.3(5).
16

by 5.1(3)

by 5.1(2)

/ This implies PONY B %

T PP 
2. We will show that
PONY B %
T PP  T PP 3 ECD 0.

0/ since ECD is a bijection, and finally, PONY B $ T PP  T PP. The property above is proved
T PP 
by

PONY B L-
T PP  T PP 3I ECD
 PONY B  ECD .%
TPP  T PP I ECD
 PONY B  ECD .%
ECN  T PP 
 PON  ECD M
ECN  T PP   ECD
.%
ECN  T PP I ECD .%
TPP  T PPI
.%
ECN  T PP 
 PON M
T PP  T PP I.%
T PP  T PP
.%
ECN  T PP IL-
ECN  T PP 
 PONZ M
T PP  T PP I.%
T PP  T PP
 0/ 

since ECD is a bijection

by 5.4(8)
ECD
since ECD is a bijection
by 5.4(5)- (7) and 5.3(7)

by 1

3. Similarly to 2., we show

PONZ .%
T PP  T PP3# ECD

/ This property is proved by

0.

PONZ L-
T PP  T PP3I ECD
 PONZ  ECD L-
T PP  T PPI ECD
 PONZ  ECD L-
ECN  T PP 
 PON  ECD L-
ECN  T PP I ECD
L-
ECN  T PP   ECD .%
ECN  T PP 
 PON .%
T PP  T PP#.%
ECN  T PP 
L-
T PP  T PP 
$ PONY B .%
T PP  T PP 
 0/ 
6

4. Suppose xd NT PPz which implies xd

x

 z

 x

5. Again, suppose x d NT PPz which implies x 2 z

x d NT PP
x d
x d NT PPx d

by 5.4(7)
since ECD is a bijection

by 5.4(5),(2) and 5.3(7)

by 2

z. Obviously, we have

and hence xPODz.

x 2 z NT PPx :<;
:<;

since ECD is a bijection

x2 z

/

 / 0

x0 z

 1

0. Now, assume x 2 zNT PPx. Then we conclude

0 zd 

by 5.2(2)
by 5.3(9) since xd DCz d

:L;

x d NT PPz

a contradiction. Analogously, it follows that

x 2 z  NT PPz is impossible. Together we have xT PPx 2
zT PPz and hence PODZ $ T PP  T PP.
17

We now state our main result.

Theorem 6.2. The set ' of relations given in Table 2 together with the extensional interpretation of
their weak composition table (Table 4) is the set of atoms of a relation algebra .
Proof. Following [23] we have to show:
1. The relations are pairwise disjoint, their union is U
2.

'

is closed under taking converses, and either R

$ 1

U.
or R  1

0/ for all R

-'

3. Each relation is non-empty.

4. The composition of any two of them is a union of elements of ' .
1. Lemma 6.1 gives us
T PPA T PPB is a partition of T PP

ECNA ECNB is a partition of ECN

PONX A1 PONZ

is a partition of PON

PODYA PODY B PODZ is a partition of POD

which proves 1.
2. This is obvious from the definitions.

3. We shall indicate elements of U which are in the relations of Table 4.6. Notice, that we have the
following:
(a) T PPA  ECD

(b) T PPB  ECD

(c) T PPA  ECD
(d) T PPB  ECD

ECNA.

ECNB.

PODYA.

PODY B.

PONX A1.

PONYA1.

PONYA1.

PONZ.

(e) PONX A1  ECD

(f) PONX A2  ECD
(g) PONX B1  ECD
(h) PONX B2  ECD
(i) PONYA1  ECD
(j) PONYA2  ECD

PONX B1.

PONY B.
18

These equalities are a consequence of 5.1(2), 5.3(3),(4) and 5.4. Therefore, it is sufficient to show
that the relations T PPA T PPB PONX A1 PONX A2 PONXB1 PONXB2, PONYA1 and PONYA2 are
non-empty. To this end, we will use a configuration given by the Figure 2; this is only an indication in
a familiar model. First of all, we want to show that this configuration emerges in every model of RCC
 / Note, that by 5.1(1) such model is necessarily infinite.
with U / 0.
Let 1

/

U be given. Furthermore, using 5.1(1) let tNT PPs d and wNT PP

s 0 t  d . Then we have
sDCt

sDCw
tDCw

s 0 t 0 w 1

by RCC 4a
by RCC 4a and 5.2(1)
by RCC 4a and 5.2(1)
since
s 0 t I0 w

1;

g
s 0 t  d , a contradiction

Again, using 5.1(1) let aNT PPs bNT PPa dNT PPt and cNT PP
a d 2 s  . Since a d 2 s 6 s we have
cNT PPs by 5.2(5). Furthermore, cNT PP
ad 2 s  implies cDC
a 0 s d  , and hence cDCa by 5.2(1).
The required elements and their properties are listed in Table 3 on the following page. Proofs are
straightforward and left to the reader. Using 5.3(6),(7) and (9), we conclude our assumption.
4. We have generated a composition table, and have checked that Table 4 on page 21 represents a
relation algebra. Both were done with a program written in the functional language GOFER. Observe
that unlike the weak composition of Table 1 for the r 10 structures, the composition in Table 4 gives
composition of relation algebras which is associative.
To end up with a compact description, we have coded the sets of atoms by the following 5 5-matrix.
1
T PPA
T PPA
T PPB
T PPB
NT PP
NT PP PONX A1 PONX A2 PONX B1
PONX B2 PONYA1 PONYA2 PONYA1 PONYA2
PONY B PONY B
PONZ
PODYA
PODY B
PODZ
ECNA
ECNB
ECD
DC
Now, Table 4 should be read as follows. The weak composition of PODY B
, i.e. equal to union of the relations

and PODYA

T PPA T PPB T PPB PONX A1 PONX B1 PONYA1 PONYA1

PONY B PONY B PONZ PODZ 

This completes the proof.

19

is

20

+
+
+
+
+
+

+
+
+
-

+
+
+
+
+
+
+
+

x z T PPz

a s
a

+
+
+
+
+
-

t
t

s
s

t
a s
a t
b a
a
w
t

x z
a
d t
a c s
ac s
a c s
c t
a s

x z

+
+
+
+
+
+

zECN x z

s
s
s t w
s t

+
+
+
+
+
+

x z T PPx

+
+
+
+

+
+
+
+
+
+

xECN x z

T PPA
T PPB
PONXA1
PONXA2
PONXB1
PONXB2
PONYA1
PONYA2

x z T PPz

c
c

xT PP x z

a s w
a t

a
a

x z

z T PPz

t
t
s

a s
s
a t
a s
a s b
a s
s t
s

s
s
d

xT PP x

T PPA
T PPB
PONXA1
PONXA2
PONXB1
PONXB2
PONYA1
PONYA2

Relation

Relation

1 aNT PPs bNT PPa cNT PPs cDCa dNT PPt

Table 3: Elements

Figure 2: sDCt sDCw tDCw s

Table 4: The composition table of

6.1 Topological properties

Suppose that  X  is a topological space. If x $ X , we denote the closure of x by cl
x  , and its interior
by int
x  . The fringe or boundary Fr
x  of x is the set cl
x 9 int
x  . x is called regular open, if

x int
cl
x 3 . It is well known that the collection RO
X  of regular open sets is a complete Boolean

21

algebra under set inclusion where for v w

RO
X  ,

v0 w

(6.1)


v2 w

vd

(6.2)
(6.3)

int
cl
v  w 3
v  w

int
C v 

The space  X  is called regular or T3 space, if distinct points can be separated by disjoint open sets,
and if for each a  X and each closed set x not containing a, there are disjoint open sets w v such that
/
a  w x $ v.  X  is called connected if the only open-closed (clopen) sets are X and 0.
For properties of topological spaces not mentioned here, we invite the reader to consult [20].
As shown in [21], a standard RCC model is the complete Boolean algebra RO
X  of regular open sets
of a connected regular topological space  X  , where for x y  RO
X 
xCy :<;

def

(6.4)

cl
x I cl
y H/ 0/ 

Theorem 6.3 gives the topological properties of the base relations and the building blocks of the
others, from which the properties of the atoms can easily be derived.
Theorem 6.3. Let B be an atomless subalgebra of RO
X  and C be the connection relation of (6.4)



defined on U B L 0/ X  . Furthermore, let x y z  U x / y. Then,
(6.5)
(6.6)
(6.7)
(6.8)
(6.9)
(6.10)
(6.11)
(6.12)
(6.13)
(6.14)
(6.15)


xNT PPy

xPONy

xPODy

xECNy

xECDy

xT PPy


xECN T PPy

xT PP T PPy

xT PP T PPy

xECD NT PPy

xDCy

y Fr x& Fr y}  0/
cl x y

x  y y  x x  y   0/ cl x cl y G  X

x  y y  x x  y   0/ cl x cl y  X
x  y  0/ cl x & cl y K  0/ cl x & cl y K 
x  y  0/ cl x & cl y K  0/ cl x & cl y 
x

X
X

 cl y  0/
Fr x  Fr  x  z o  0/ Fr z  Fr  x  z o  0/ cl x & cl y K 
Fr x  Fr int cl x z \\}  0/ Fr z  Fr int cl x z \}  0/
Fr x  Fr x  z o  0/ Fr z  Fr x  z n  0/
x y  X
cl x

Proof. All equivalences are straightforward applications of the definitions of the Boolean operations
given in (6.1) (6.3) on page 22, and the properties of the relations given in Lemma 5.1.
We would like to close this section with an RCC model which has different properties than the one on
the full algebra RO
X  . Let K be the collection of sets of the form

p<
 2 : a f p f b 

K
a b
 p < : f p f b 
22

if 0
if a

 / a
 0

f f

where a  b ~  , and p is the Euclidian distance of p  2 to

0 0  ). We also extend
the ordering of and set a for all a . . Let B be the set of all finite unions of elements of K
/ Then B is a subalgebra of RO
2  , generated by the open disks with centre at the origin;
including 0.
by a result of [13],  B C  is a model of the RCC, where C is defined by (6.4).

Now, consider x K
0 1  . We want to show that there is no y  U B e 2 0/  with xT PPAy. Every

 K
a b : 1 a  and
element y of U with xT PPy is of the form x  K
a b  : 1 a  . Since x 2 y
 K
a b : 1 a  is disconnected to x, we conclude that xT PPBy. It follows that the BRA generated
by C on this domain is not integral. On the other hand, it is not hard to see that in RO
A , 1 $
T PPA  T PPA, so this situation cannot happen there.

7 Summary and Outlook

We have shown that each relation algebra generated by the contact relation of an RCC model contains
an integral algebra with 25 atoms as a subalgebra. Thus, the expressiveness of the RCC axioms in
the 3-variable fragment of first order logic is much greater than the original eight RCC base relations
(which are, basically, the possible relations of a pair of circles) might suggest. We have also given a
topological interpretation of the atoms of .
We have not yet found a representation of , and the problem is open as to whether there is an RCC
model with as its associated BRA. In particular, we do not know, if is the BRA generated by C
on a standard model RO
X  .
All RCC models that we know fulfil
(7.1)

NT PP

NT PP  NT PP

and the question remains, whether this is always true.

It seems also worthwhile to compare the expressivity of RA logic with that of the 9-intersection model
of Egenhofer and Herring [17], which is based only on topological properties.
Another promising area of research is to consider the expressive power of relational structures more
general than BRAs, for example, those, in which the associativity of the composition is relaxed [30].
Egenhofer and Rodrguez [18] have given a spatial interpretation of such a structure.

Acknowledgement
We would like to thank the anonymous referees for their helpful comments, and Thomas Moormann
for spotting an error in Lemma 5.3.

23

References
[1] Asher, N. and Vieu, L. (1995). Toward a geometry of common sense: A semantics and a complete axiomatization of mereotopology. In Mellish, C., editor, IJCAI 95, Proceedings of the 14th
International Joint Conference on Artificial Intelligence.
[2] Bennett, B. (1997). Logical representations for automated reasoning about spatial relationships.
Doctoral dissertation, School of Computer Studies, University of Leeds.
[3] Bennett, B. (1998). Determining consistency of topological relations. Constraints, 3:213225.
[4] Bennett, B., Isli, A., and Cohn, A. (1997). When does a composition table provide a complete
and tractable proof procedure for a relational constraint language? In IJCAI 97, Proceedings of the
Workshop of Spatial Reasoning.
[5] Berghammer, R., Gritzner, T., and Schmidt, G. (1993). Prototyping relational specifications using
higher-order objects. In Heering, J., Meinke, K., Mvller, B., and Nipkow, T., editors, Proc. International Workshop on Higher Order Algebra, Logic and Term Rewriting (HOA 93), LNCS 816,
pages 5675. Springer.
[6] Berghammer, R. and Karger, B. (1997). Algorithms from relational specifications. in [9].
[7] Berghammer, R. and Zierer, H. (1986). Relational agebraic semantics of deterministic and nondeterministic programs. Theoret. Comp. Sci., 43:123147.
[8] Biacino, L. and Gerla, G. (1991). Connection structures. Notre Dame Journal of Formal Logic,
32:242247.
[9] Brink, C., Kahl, W., and Schmidt, G., editors (1997). Relational Methods in Computer Science.
Advances in Computing Science. Springer.
[10] Chin, L. and Tarski, A. (1951). Distributive and modular laws in the arithmetic of relation
algebras. University of California Publications in Mathematics, 1:341384.
[11] Clarke, B. L. (1981). A calculus of individuals based on connection. Notre Dame Journal of
Formal Logic, 22:204218.
[12] Cohn, A. G. (1997). Qualitative spatial representation and reasoning techniques. Research
report, School of Computer Studies, University of Leeds.
[13] Dntsch, I. (2000). Contact relation algebras. In Orowska, E. and Szaas, A., editors, Relational
Methods in Computer Science Applications, pages 113134, Berlin. SpringerVerlag.
[14] Dntsch, I., Wang, H., and McCloskey, S. (1999). Relation algebras in qualitative spatial reasoning. Fundamenta Informaticae, 39:229248.
[15] Dntsch, I., Wang, H., and McCloskey, S. (2001). A relation algebraic approach to the Region
Connection Calculus. Theoretical Computer Science, 255:6383.
24

[16] Egenhofer, M. and Franzosa, R. (1991). Pointset topological spatial relations. International
Journal of Geographic Information Systems, 5(2):161174.
[17] Egenhofer, M. and Herring, J. (1991). Categorizing binary topological relationships between
regions, lines and points in geographic databases. Tech. report, Department of Surveying Engineering, University of Maine.
[18] Egenhofer, M. and Rodrguez, A. (1999). Relation algebras over containers and surfaces: An
ontological study of a room space. Spatial Cognition and Computation. To appear.
[19] Egenhofer, M. and Sharma, J. (1992). Topological consistency. In Fifth International Symposium
on Spatial Data Handling, Charleston, SC.
[20] Engelking, R. (1977). General Topology. Monografie matematyczne. Polish Scientific Publ.,
Warszawa.
[21] Gotts, N. M. (1996a). An axiomatic approach to topology for spatial information systems. Research Report 96.25, School of Computer Studies, University of Leeds.
[22] Gotts, N. M. (1996b). Topology from a single primitive relation: Defining topological properties and relations in terms of connection. Research Report 96.23, School of Computer Studies,
University of Leeds.
[23] Jnsson, B. (1984). Maximal algebras of binary relations. Contemporary Mathematics, 33:299
307.
[24] Jnsson, B. (1991). The theory of binary relations. In Andrka, H., Monk, J. D., and Nmeti,
I., editors, Algebraic Logic, volume 54 of Colloquia Mathematica Societatis Jnos Bolyai, pages
245292. North Holland, Amsterdam.
[25] Jnsson, B. and Tarski, A. (1952). Boolean algebras with operators II. American Journal of
Mathematics, 74:127162.
[26] Koppelberg, S. (1989). General Theory of Boolean Algebras, volume 1 of Handbook on Boolean
Algebras. North Holland.
[27] Lesniewski, S. (1927 1931). O podstawach matematyki. Przeglad Filozoficzny, 3034.
[28] Lesniewski, S. (1983). On the foundation of mathematics. Topoi, 2:752.
[29] Lyndon, R. C. (1950). The representation of relational algebras. Annals of Mathematics (2),
51:707729.
[30] Maddux, R. (1982). Some varieties containing relation algebras. Transactions of the American
Mathematical Society, 272:501526.
[31] Pratt, I. and Schoop, D. (1998). A complete axiom system for polygonal mereotopology of the
real plane. Journal of Philosophical Logic, 27(6):621658.
25

[32] Pratt, I. and Schoop, D. (2000). Expressivity in polygonal, plane mereotopology. Journal of
Symbolic Logic, 65(2):822838.
[33] Pratt, V. (1990). Dynamic algebras as a well behaved fragment of relation algebras. Dept of
Computer Science, Stanford.
[34] Randell, D. A., Cohn, A. G., and Cui, Z. (1992). Computing transitivity tables: A challenge for
automated theorem provers. In Kapur, D., editor, Proceedings of the 11th International Conference
on Automated Deduction (CADE-11), volume 607 of LNAI, pages 786790, Saratoga Springs, NY.
Springer.
[35] Schmidt, G. (1981a). Programs as partial graph I: Flow equivalence and correctness. Theoret.
Comp. Sci., 15:125.
[36] Schmidt, G. (1981b). Programs as partial graphs II: Recursion. Theoretical Computer Science,
15(2):159179.
[37] Schmidt, G. and Strhlein, T. (1989). Relationen und Graphen. Springer. English version:
Relations and Graphs. Discrete Mathematics for Computer Scientists, EATCS Monographs on
Comp. Sci., Springer (1993).
[38] Schrder, E. (1895). Algebra der Logik, volume 3. Teubner, Leipzig.
[39] Smith, T. P. and Park, K. K. (1992). An algebraic approach to spatial reasoning. International
Journal of Geographical Inforation Systems, 6:177192.
[40] Stell, J. (1997). Personal communication, October 30, 1997.
[41] Tarski, A. (1935). Zur Grundlegung der Booleschen Algebra, I. Fundamenta Mathematicae,
24:177198.
[42] Tarski, A. (1941). On the calculus of relations. Journal of Symbolic Logic, 6:7389.
[43] Tarski, A. and Givant, S. (1987). A formalization of set theory without variables, volume 41 of
Colloquium Publications. Amer. Math. Soc., Providence.
[44] Winter, M. (1998). Strukturtheorie heterogener Relationenalgebren mit Anwendung auf Nichtdeterminismus in Programmiersprachen. Dissertationsverlag NG, Kopierladen GmbH.

26