Mereotopology

Origin. Whitehead proposed region-based theories (1920s). Clarke formalized connection (1981). Combines mereology (parts) with topology (connection). Randell, Cui, Cohn: Region Connection Calculus (1992). Foundation for qualitative spatial reasoning without points.

Models. Regions with connection and parthood. Mereology: part-whole relations. Topology: open, closed, boundary, connection. Combined: regions that touch, overlap, contain. Points derived, not primitive. Intuitive for physical objects and space.

Formalism.

Primitives:

C(x, y): x connects with y (shares point or boundary)
Or: P(x, y): x is part of y

Connection-based (Clarke): C is reflexive and symmetric. x is connected to y iff they share a point (in any derived point topology).

Defined notions:

P(x,y) ≡ ∀z.(C(z,x) → C(z,y)) (parthood via connection)
O(x,y) ≡ ∃z.(P(z,x) ∧ P(z,y)) (overlap)
EC(x,y) ≡ C(x,y) ∧ ¬O(x,y) (external connection: touch but don't overlap)
DC(x,y) ≡ ¬C(x,y) (disconnected)

Interior and boundary:

IP(x,y): x is interior part of y (x inside y, not touching y's boundary)
TP(x,y): x is tangential part (part touching boundary)

RCC-8 relations (derived): DC, EC, PO, EQ, TPP, NTPP, TPPi, NTPPi (Disconnected, externally connected, partial overlap, equal, tangential proper part, non-tangential proper part, and inverses)

Axioms:

Connection is reflexive, symmetric
Supplementation: proper parts have complements
Closure under sum, product, complement (various strengths)

Symbols.

Symbol	Unicode	Name	Meaning
C	—	Connects	Shares boundary/point
P	—	Part	Parthood
O	—	Overlap	Share part
EC	—	External connection	Touch boundary
DC	—	Disconnected	No connection
PP	—	Proper part	Strict part
IP	—	Interior part	Inside
TP	—	Tangential part	Touching boundary

Metatheory. The intended model is the regular closed (equivalently regular open) subsets of a topological space, which form a Boolean algebra, with C(x,y) holding when two regions intersect. The full first-order theory is undecidable — Grzegorczyk (1951) proved this for topological theories of this kind — so the workable content lies in constraint fragments rather than the whole language. RCC-8 is the principal fragment, and Renz and Nebel (1999) proved deciding consistency of an RCC-8 constraint network NP-complete. They also exhibited a maximal tractable subset containing all eight base relations, on which path consistency decides consistency in polynomial time, and Renz (1999) completed the classification, showing there are exactly three maximal tractable subsets containing the base relations. Axiom strength is not a free choice: Clarke's original system with unrestricted fusion proved too strong, its models collapsing to essentially Boolean structures (Biacino and Gerla 1991), which is why Asher and Vieu and Casati and Varzi restrict the closure principles.

Applies to. Geographic information systems. Robot navigation. Natural language spatial expressions ("in," "touching," "near"). Architectural design. Medical imaging. Naive physics. Ontology (spatial entities).

Limitations. The calculus is purely qualitative: RCC-8 fixes topological relations and says nothing about distance, direction, size or shape, so a consistent network can have no realization anyone would recognize as the intended configuration. There is no canonical axiomatization — connection-based and parthood-based primitives, and differing supplementation and closure principles, yield genuinely different theories proving different things, and nothing in the subject matter settles the choice. The full first-order theory is undecidable, so tractable reasoning is confined to fragments; even RCC-8 consistency is NP-complete, with polynomial reasoning available only on the identified maximal tractable subsets, and path consistency is sound but incomplete outside them. Standard models assume crisp, self-connected regions, leaving vague boundaries, scattered regions and holes to separate machinery, and the calculus offers no way to combine its relations with metric information.