Bisimulation by Unification

1
Bisimulation by Unification
AMAST02, La Réunion 9-13 Sept. 2002

• Roberto Bruni (Univ. Pisa Univ. Illinois)
• Paolo Baldan (Univ. Pisa Univ. Venezia)
• Andrea Bracciali (Univ. Pisa)

2
Acknowledgements

• Research Supported by
• IST Programme on FET-GC Projects
• AGILE (IST-2001-32747)
• MYTHS
• SOCS
• Thanks also to
• Italian CNR
• University of Illinois at Urbana-Champaign

3
Roadmap

• Introduction Motivation
• Running Example (toy PC with ambients)
• Symbolic Bisimulation
• Symbolic Transition Systems
• Strict Large Bisimilarity
• Bisimulation by Unification
• Conclusions
• (Related Work Future Work)

4
Mission

• Methodology for the formal analysis of open
  systems
• Algebraic Representations of Processes
• Properties as Equivalences
• Process Calculi Bisimilarity
• Closed Terms Components
• Contexts Coordinators
• Compact (Symbolic) Transition Systems

5
Open Systems are

• Interactive, Autonomous, Accessible via
  Interfaces, Dynamic, Programmable,
• Ex. Web Services, WAN Computing, Mobile Code

p
q
CX1,X2,X3
r
Components
Coordinators
6
Interaction

• Components can be dynamically connected
• Ex. Access to Network Services

(Typed) Holes constrained dynamic binding
Cp,q,r
Boundaries access policies
7
Lets Get Formal

• Process Calculi Ingredients
• Structure (?,E) Signature Structural Axioms
• Operational Semantics (SOS, LTS/RS)
• Linguistic abstraction for holes and binding
• Variables Substitutions
• Logic for expressing and proving properties
• Specification Verification
• Tool for focusing e.g. on distribution,
  communication, causal dependencies

Mostly devised for components!
8
Abstraction

• Equivalence on Components p ? q
• Bisimulation, Traces, May/Must Testing
• Equivalence on Coordinators
• CX ?univ DX iff ?p. Cp ? Dp
• (for simplicity, we consider one-holed contexts
  in most slides)
• needs universal quantification (on
  instantiations)!
• Focus on Bisimilarity (largest bisimulation) p ?
  q
• if p a? p then ? q a? q with p ? q
• (and vice versa)

9
Graphically
Components
p1
q1
a1
a1
p
q
an
an
pn
qn
10
Example Async. CCS Ambients
p 0 a a.p np open n.p in n.p
out n.p pp
(Assume AC1 parallel composition)
11
A Problem on Components
na.0a -?? n0 -/? ??
? mb.0b -?? m0 -/?
12
A Problem on Components
na.0a -?? n0 -/? ??
? mb.0b -?? m0 -/?
13
A Problem on Components
na.0a -?? n0 -/? ??
? mb.0b -?? m0 -/?
14
A Problem on Components
na.0a -?? n0 -/? ??
? mb.0b -?? m0 -/?
15
A Problem on Components
na.0a -?? n0 -/? ??
? mb.0b -?? m0 -/?
16
A Problem on Components
na.0a -?? n0 -/? ??
? mb.0b -?? m0 -/?
17
A Problem on Components
na.0a -?? n0 -/? ??
? mb.0b -?? m0 -/?
18
A Problem on Coordinators
nX ?? mX
19
Symbolic Approach

• Bisimulation Without Instantiation
• Facilitate analysis verification of
  coordinators properties
• Distinguishing Features
• Symbolic LTS
• states are coordinators
• labels are spatial/modal formulae
• Avoids universal closure
• Allows for coalgebraic techniques
• Constructive definition for Algebraic SOS
• (In general yields equivalences finer than ?univ )

20
Notation

• We start from a PC specified by
• Syntax Structural Equivalence (?,E)
• T?,E is the set of Components p,q,r
• T?,E(X) is the set of Coordinators CX, DX,
• CX1,,Xn means var(C) ? X1,,Xn
• Labels ? ranged by a,b,
• LTS L (defined on T?,E ?)
• possibly defined by SOS rules

21
Symbolic Transition Systems

• Ordinary SOS approach
• Behavior of a coordinator can depend on
• The spatial structure of the components that are
  inserted/connected/substituted
• The behavior of those components
• Idea to borrow formulae from a suitable logic
  to express the most general class of components
  that can take part in the coordinators evolution

22
What Logic Do We Need?

• Formulae must express the minimal amount of
  information on components for enabling the step
• Components that are not playing active role in
  the step
• Most general active components needed for the
  step
• Assumptions not only on the structure of
  components, but also on their behavior
• Logic L must include, as atomic formulae
• Place-holders (process variables) X q X
• Components p q p iff q ?E p

23
Symbolic Transitions
Coordinators

• CX ?(Y)?a DY
• intuitively whenever p ?(q),
• then Cp a? Dq
• ( q is to some extent the residual of p after
  satisfying ? )

Formula
Ordinary label
24
Correctness
CX ?(Y)?a DY
STS
?pi,qi. pi ?(qi)
Cp1 a? Dq1

• Cp a? Dq

Cp2 a? Dq2
LTS L
Cpn a? Dqn
components that can make a
25
Completeness
r ?E Cp a? q
LTS L
? ?,s. CX ?(Y)?a DY
STS
with p ?(s) and q ? Ds
26
Strict Bisimilarity

• Strict Bisimilarity largest (strict)
  bisimulation s.t.
• CX ?(Y)?a CY
• ?strict ?strict
• DX ?(Y)?a DY
• THEOREM
• If the STS is correct complete, then
• ?strict ? ?univ

27
Strict Bisimilarity

• Strict Bisimilarity largest (strict)
  bisimulation s.t.
• CX ?(Y)?a CY
• ?strict ?strict
• DX ?(Y)?a DY
• THEOREM
• If the STS is correct complete, then
• ?strict ? ?univ

28
Strict Bisimilarity

• Strict Bisimilarity largest (strict)
  bisimulation s.t.
• CX ?(Y)?a CY
• ?strict ?strict
• DX ?(Y)?a DY
• THEOREM
• If the STS is correct complete, then
• ?strict ? ?univ

29
Strict Bisimilarity

• Strict Bisimilarity largest (strict)
  bisimulation s.t.
• CX ?(Y)?a CY
• ?strict ?strict
• DX ?(Y)?a DY
• THEOREM
• If the STS is correct complete, then
• ?strict ? ?univ

30
Strict Bisimilarity

• Strict Bisimilarity largest (strict)
  bisimulation s.t.
• CX ?(Y)?a CY
• ?strict ?strict
• DX ?(Y)?a DY
• THEOREM
• If the STS is correct complete, then
• ?strict ? ?univ

31
Back to the Open Problem
nX Ykout n.ZW? nYkZW ?strict?
mX
32
Back to the Open Problem
nX Ykout n.ZW? nYkZW ?strict?
mX Ykout n.ZW -/?
33
Back to the Open Problem
nX Ykout n.ZW? nYkZW ?strict
mX Ykout n.ZW -/?
34
Back to the Open Problem
nX ?univ mX
(take X kout n.0)
35
A Last Problem
nmout n.X Y? n0m0 ?strict
? n0maa.X Y? n0m0
36
A Last Problem
nmout n.X Y? n0mY ?strict
n0maa.X Y? n0mY
37
A Last Problem
nmout n.X ?strict n0maa.X
nmout n.X ?univ n0maa.X
?
38
Large Bisimilarity

• What if ?strict is too fine?
• We can relax the strict bisimilarity when the
  logic L includes generic spatial formulae
• Operators f??
• q f(?1,,?n) iff ?qi. q ?E f(q1,,qn) ?
  qi ?i
• We call spatial formulae those composed by
  spatial operators and place-holders only
• Ambivalent view of Spatial Formulae as
  Coordinators

39
Large Bisimilarity

• Large Bisimilarity largest (large) bisimulation
  s.t.
• CX ?(Y)?a CY ?large D?(Y)
• ?large
• DX ?(Z)?a DZ ?(Y) ?(?(Y))
• ?(Y) spatial
• THEOREM
• If the STS is correct complete, then
• ?large ? ?univ

40
Large Bisimilarity

• Large Bisimilarity largest (large) bisimulation
  s.t.
• CX ?(Y)?a CY ?large D?(Y)
• ?large
• DX ?(Z)?a DZ ?(Y) ?(?(Y))
• ?(Y) spatial
• THEOREM
• If the STS is correct complete, then
• ?large ? ?univ

41
Large Bisimilarity

• Large Bisimilarity largest (large) bisimulation
  s.t.
• CX ?(Y)?a CY ?large D?(Y)
• ?large
• DX ?(Z)?a DZ ?(Y) ?(?(Y))
• ?(Y) spatial
• THEOREM
• If the STS is correct complete, then
• ?large ? ?univ

42
Large Bisimilarity

• Large Bisimilarity largest (large) bisimulation
  s.t.
• CX ?(Y)?a CY ?large D?(Y)
• ?large
• DX ?(Z)?a DZ ?(Y) ?(?(Y))
• ?(Y) spatial
• THEOREM
• If the STS is correct complete, then
• ?large ? ?univ

43
Large Bisimilarity

• Large Bisimilarity largest (large) bisimulation
  s.t.
• CX ?(Y)?a CY ?large D?(Y)
• ?large
• DX ?(Z)?a DZ ?(Y) ?(?(Y))
• ?(Y) spatial
• THEOREM ?strict ? ?large
• If the STS is correct complete, then
• ?large ? ?univ

44
Large Bisimilarity

• Large Bisimilarity largest (large) bisimulation
  s.t.
• CX ?(Y)?a CY ?large D?(Y)
• ?large
• DX ?(Z)?a DZ ?(Y) ?(?(Y))
• ?(Y) spatial
• THEOREM ?strict ? ?large
• If the STS is correct complete, then
• ?large ? ?univ

45
Why Use ?strict ?large

• As an approximation method for ?univ
• ?univ is not defined coinductively
• ?univ requires the verification of infinitely
  many equivalences
• Bonus Theorems
• CX ?large DX implies CEY ?univ DEY
• CX ?strict DX implies CEY ?univ DEY
• Note that in general ?large is not transitive
• Bonus Theorem
• if CX ?large DX implies CEY ?large
  DEY, then ?large is transitive and thus it is
  an equivalence relation

46
Bisimulation by Unification

• Algebraic SOS Format (spatial/modal constraints)
• (Yi is either Xi (if i?I) or Zi (if i?I))
• Formulae ? X p ?a.? f(?,,?)
• Modality ?a q ?a.? iff ?q a? p ? p ?

Xi ai? Zii?I
CX1,,Xn a? DY1,,Yn
47
The Prolog Algorithm

• trs( box(A,X) , A , X ) - !.
• trs( CX1,,Xn,a,DY1,,Yn ) -
• trs(Xi1 , ai1 , Zi1),
• ,
• trs(Xin , ain , Zin).
• The program can be seen as the specification of
  the STS
• Goals have the form ?- trs(CX1,,Xn, a , Z).
• Backtracking mechanism meta-logic ops (bagof)
  can be used to compute all symbolic transitions
  for CX
• THEOREM
• The resulting STS is correct complete

48
Conclusions

• General formal framework for open systems
• Meta-theoretic foundations
• Under suitable hypothesis
• ?strict implies ?large implies ?univ
• For the Algebraic SOS format, a minimal STS can
  be defined constructively in Prolog
• cut unification
• extension to AC1 parallel operator (see paper)

49
Dual View

• Instantiation ? Contextualization
• When ? is not a congruence
• p ? q iff ?CX. Cp ? Cq
• ? is not a bisimulation (unless ? is a
  congruence)
• (the largest congruence which is also a
  bisimulation is called dynamic bisimulation)
• Sewell, Leifer Milner minimal contexts as
  labels
• Transitions p C _ ,X1,,Xn? DX1,,Xn
• ?pi. Cp,p1,,pn -?? Dp1,,pn
• C. minimal (not necessarily minimum)
• Universal quantification moved from contexts to
  components!

50
Related Work / Source of Inspiration

• Sewell, Leifer Milner
• categorical characterization of the most general
  interaction (relative pushout)
• Caires, Cardelli Gordon
• Fiadeiro, Maibaum, Martì-Oliet, Meseguer Pita
• elegant mathematical tool for expressing
  structural temporal aspects
• Bruni, Montanari Rossi
• interactive view of Logic Programming

51
Future Work

• Deal with names
• Name restriction Logical notion of freshness
• Duality
• Categorical formulation (relative pullback?)
• Symbolic approach to the verification of infinite
  state cryptographic protocols
• Extension to meta and abductive LP
• Programmable definition of proofs
• To answer questions like under which assumptions
  can pX evolve so to satisfy a certain property?
  that are relevant in dynamic system engineering

52

• Bisimulation By Unification
• a paper by Andrea Bracciali
• Paolo Baldan
• Roberto Bruni
• AMAST presentation by Roberto Bruni
• Research supported by
• IST Programme on FET-GC Projects AGILE, MYTHS,
  SOCS
• Italian CNR
• University of Illinois at Urbana-Champaign