Jan Rutten - Concrete coalgebra: an introduction by examples - Lecture 1

1. Lecture one: The Method of Coalgebra – in some detail – Jan Rutten CWI Amsterdam & Radboud University Nijmegen IPM, Tehran - 13 January 2016

2. Acknowledgements My esteemed co-authors - Marcello Bonsangue (Leiden, CWI) - Helle Hansen (Delft) - Alexandra Silva (City College London) - Milad Niqui - Clemens Kupke (Glasgow) - Prakash Panangaden (McGill, Montreal) - Filippo Bonchi (ENS, Lyon) - Joost Winter (Warsaw), Jurriaan Rot (ENS, Lyon) - and many others. Samson Abramsky (Oxford) Damien Pous (ENS, Lyon)

3. Acknowledgements My esteemed co-authors - Marcello Bonsangue (Leiden, CWI) - Helle Hansen (Delft) - Alexandra Silva (City College London) - Milad Niqui - Clemens Kupke (Glasgow) - Prakash Panangaden (McGill, Montreal) - Filippo Bonchi (ENS, Lyon) - Joost Winter (Warsaw), Jurriaan Rot (ENS, Lyon) - and many others. Samson Abramsky (Oxford) Damien Pous (ENS, Lyon)

4. Overview of todays lectures Lecture one: The method of coalgebra Lecture two: A coinductive calculus of streams Lecture three: Automata and the algebra-coalgebra duality Lecture four: Coalgebraic up-to techniques

5. Overview of Lecture one 1. Category theory (where coalgebra comes from) 2. Duality (where coalgebra comes from) 3. How coalgebra works (the method in slogans) 4. Duality: induction and coinduction 5. What coalgebra studies (its subject) 6. Discussion

6. 1. Category theory (where coalgebra comes from)

7. Why categories? From Samson Abramsky’s tutorial: Categories, why and how? (Dagstuhl, January 2015)

8. Why categories? For logicians: gives a syntax-independent view of the fundamental structures of logic, opens up new kinds of models and interpretations. For philosophers: a fresh approach to structuralist foundations of mathematics and science; an alternative to the traditional focus on set theory. For computer scientists: gives a precise handle on abstraction, representation-independence, genericity and more. Gives the fundamental mathematical structures underpinning programming concepts.

9. Why categories? For mathematicians: organizes your previous mathematical experience in a new and powerful way, reveals new connections and structure, allows you to “think bigger thoughts”. For physicists: new ways of formulating physical theories in a structural form. Recent applications to Quantum Information and Computation. For economists and game theorists: new tools, bringing complex phenomena into the scope of formalisation.

10. Category Theory in 10 Slogans 1. Always ask: what are the types? 2. Think in terms of arrows rather than elements. 3. Ask what mathematical structures do, not what they are. 4. Categories as mathematical contexts. 5. Categories as mathematical structures. 6. Make deﬁnitions and constructions as general as possible. 7. Functoriality! 8. Naturality! 9. Universality! 10. Adjoints are everywhere.

20. Categories: basic deﬁnitions A category C consists of - Objects A, B, C, . . . - Morphisms/arrows: for each pair of objects A, B, a set of morphisms C(A, B) with domain A and codomain B - Composition of morphisms: g ◦ f: A g ◦ f TT f GG B g GG C - Axioms: h ◦ (g ◦ f) = (h ◦ g) ◦ f f ◦ id = f = id ◦ f

25. Categories: examples • Any kind of mathematical structure, together with structure preserving functions, forms a category. E.g. - sets and functions - groups and group homomorphisms - monoids and monoid homomorphisms - vector spaces over a ﬁeld k, and linear maps - topological spaces and continuous functions - partially ordered sets and monotone functions • Monoids are one-object categories • algebras, and algebra homomorphisms • coalgebras, and coalgebra homomorphisms

35. Arrows rather than elements A function f : X → Y (between sets) is: - injective if ∀x, y ∈ X, f(x) = f(y) ⇒ x = y - surjective if ∀y ∈ Y, ∃x ∈ X, f(x) = y - monic if ∀g, h, f ◦ g = f ◦ h ⇒ g = h - epic if ∀g, h, g ◦ f = h ◦ f ⇒ g = h Proposition • m is injective iff m is monic. • e is surjective iff e is epic.

43. Arrows rather than elements Deﬁning the Cartesian product . . . - with elements: A × B = { a, b | a ∈ A, b ∈ B } where a, b = {{a, b}, b} - with arrows (expressing a universal property): ∀C f ÐÐ g 11 f, g A A × Bπ1 oo π2 GG B

46. 2. Duality (where coalgebra comes from) An additional slogan for categories: duality is omnipresent - epi - mono - product - sum - initial object - ﬁnal object - algebra - coalgebra

47. 2. Duality (where coalgebra comes from) An additional slogan for categories: duality is omnipresent - epi - mono - product - sum - initial object - ﬁnal object - algebra - coalgebra

48. Duality: monos and epis - f is monic: ∀g, h, f ◦ g = f ◦ h ⇒ g = h • g 99 h UU • f GG • - f is epic: ∀g, h, g ◦ f = h ◦ f ⇒ g = h • • h gg g ww • foo Proposition: f is monic in C iff f is epic in Cop.

52. Duality: products and coproducts The product of A and B: ∀C f ÐÐ g 11 f, g A A × Bπ1 oo π2 GG B The coproduct of A and B: ∀C A κ1 GG f PP A + B [f, g] yy Bκ2 oo gll Proposition: O is product in C iff O is coproduct in Cop.

56. Duality: initial and final objects An object A in a category C is . . . - initial if for any object B there exists a unique arrow A ! GG B - final if for any object B there exists a unique arrow B ! GG A Proposition: A is initial in C iff A is final in Cop. Proposition: Initial and final objects are unique up-to isomorphism.

62. Where coalgebra comes from By duality. From algebra! Classically, algebras are sets with operations. Ex. (N, 0, succ), with 0 ∈ N and succ : N → N. Equivalently, 1 + N [zero, succ] N where 1 = {∗} and zero(∗) = 0.

65. Algebra Classically, algebras are sets with operations. Ex. Prog × Prog α Prog with α(P1, P2) = P1; P2.

66. Algebra, categorically F(X) α X where F is the type of the algebra.

67. Coalgebra, dually X α F(X) where F is the type of the coalgebra.

68. Example: streams Streams are our favourite example of a coalgebra: Nω head, tail N × Nω where head(σ) = σ(0) tail(σ) = (σ(1), σ(2), σ(3), . . .) for any stream σ = (σ(0), σ(1), σ(2), . . .) ∈ Nω.

69. 3. How coalgebra works (its method in slogans) • be precise about types • ask what a system does rather than what it is • functoriality • interaction through homomorphisms • aim for universality Note that all these slogans are part of the categorical approach.

70. 3. How coalgebra works (its method in slogans) • be precise about types • ask what a system does rather than what it is • functoriality • interaction through homomorphisms • aim for universality Note that all these slogans are part of the categorical approach.

71. Starting point: the system’s type A coalgebra of type F is a pair (X, α) with α : X → F(X) For instance, non-deterministic transition systems: X α X α X α X α Pf (A×X) Pf (X)A P(X)A 2 × P(X)A Formally, the type F of a coalgebra/system is a functor.

72. The importance of knowing the system’s type The type F of a coalgebra/system α : X → F(X) determines - a canonical notion of system equivalence: bisimulation - a canonical notion of minimization - a canonical interpretation: ﬁnal coalgebra semantics - (a canonical logic)

73. Doing versus being Doing Being Behaviour Construction Systems as black boxes (with internal states) Behavioural speciﬁcation

78. Example: the shufﬂe product of streams Being: (σ ⊗ τ) (n) = n k=0 n k · σ(k) · τ(n − k) Doing: σ ⊗ τ σ(0) · τ(0) GG (σ ⊗ τ) + (σ ⊗ τ )

81. Example: the Hamming numbers Being: The increasing stream h of all natural numbers that are divisible by only 2, 3, or 5: h = (1, 2, 3, 4, 5, 6, 8, 9, 10, 12, 15, 16, 18, 20, 24, . . . ) h(n) = ? Doing: h 1 GG (2 · h) (3 · h) (5 · h)

85. Homomorphisms X α h GG Y β F(X) F(h) GG F(Y) . . . are for systems/coalgebras what functions are for sets. . . . are behaviour preserving functions.

86. Functoriality X α h GG Y β F(X) F(h) GG F(Y) Note that for the deﬁnition of homomorphism, the type F needs to be a functor: F acts on sets: F(X), F(Y) and on functions: F(h)

87. Example of a homomorphism X h GG Y O × X id × h GG O × Y x0 a GG x1 b GG x2 a GG x3 b ww h GG y0 a VV y1 b xx Minimization through (canonical) homomorphism.

88. Universality Always aim at universal/canonical formulations. For instance: ﬁnal coalgebras In ﬁnal coalgebras: Being = Doing ⇒ coinduction (to be discussed shortly) ⇒ semantics

94. Semantics by ﬁnality: streams The ﬁnal homomorphism into the set of streams: X ∃! h GG Oω O × X id × h GG O × Oω maps any system X to its minimization: e.g., x0 a GG x1 b GG x2 a GG x3 b ww h GG (ab)ω a RR (ba)ω btt x0, x2 h GG (ab)ω x1, x3 h GG (ba)ω

95. 4. Duality: induction and coinduction - initial algebra - final coalgebra - congruence - bisimulation - induction - coinduction - least fixed point - greatest fixed point

96. Initial algebra The natural numbers are an example of an initial algebra: 1 + N [zero, succ] GG 1 + S β∀ N ∃ ! GG S Note: any two homomorphisms from N to S are equal. Note: id : N → N is a homomorphism. Note: [zero, succ] : 1 + N ∼= N.

100. Final coalgebra Streams are an example of a ﬁnal coalgebra: S β∀ ∃ ! GG Nω head, tail N × S GG N × Nω (Note: instead of N, we could have taken any set.) Note: any two homomorphisms from S to Nω are equal. Note: id : Nω → Nω is a homomorphism. Note: head, tail : Nω ∼= N × Nω.

104. Algebra and induction Induction = deﬁnition and proof principle for algebras. Ex. mathematical induction: for all P ⊆ N, ( P(0) and (∀n : P(n) ⇒ P(succ(n))) ) ⇒ ∀n : P(n) (Other examples: transﬁnite, well-founded, tree, structural, etc.) We show that induction is a property of initial algebras.

108. Algebras and congruences (ex. natural numbers) We call R ⊆ N × N a congruence if (i) (0, 0) ∈ R and (ii) (n, m) ∈ R ⇒ (succ(n), succ(m)) ∈ R (Note: R is not required to be an equivalence relation.) Equivalently, R ⊆ N × N is a congruence if 1 + N [zero, succ] 1 + R γ∃ oo GG 1 + N [zero, succ] N Rπ1 oo π2 GG N for some function γ : 1 + R → R.

109. Algebras and congruences (ex. natural numbers) We call R ⊆ N × N a congruence if (i) (0, 0) ∈ R and (ii) (n, m) ∈ R ⇒ (succ(n), succ(m)) ∈ R (Note: R is not required to be an equivalence relation.) Equivalently, R ⊆ N × N is a congruence if 1 + N [zero, succ] 1 + R γ∃ oo GG 1 + N [zero, succ] N Rπ1 oo π2 GG N for some function γ : 1 + R → R.

110. Initial algebras and congruences Theorem: induction proof principle Every congruence R ⊆ N × N contains the diagonal: ∆ ⊆ R where ∆ = {(n, n) | n ∈ N}. Proof: Because (N, [zero, succ]) is an initial algebra, 1 + N [zero, succ] CC 1 + R γ∃ oo GG 1 + N [zero, succ] ss N ! AA Rπ1 oo π2 GG N ! uu we have π1◦! = id = π2◦!, which implies !(n) = (n, n), all n ∈ N.

111. Initial algebras and congruences Theorem: induction proof principle Every congruence R ⊆ N × N contains the diagonal: ∆ ⊆ R where ∆ = {(n, n) | n ∈ N}. Proof: Because (N, [zero, succ]) is an initial algebra, 1 + N [zero, succ] CC 1 + R γ∃ oo GG 1 + N [zero, succ] ss N ! AA Rπ1 oo π2 GG N ! uu we have π1◦! = id = π2◦!, which implies !(n) = (n, n), all n ∈ N.

112. Initial algebras and induction Theorem: The following are equivalent: 1. For every congruence relation R ⊆ N × N, ∆ ⊆ R 2. For every predicate P ⊆ N, ( P(0) and (∀n : P(n) ⇒ P(succ(n))) ) ⇒ ∀n : P(n) Proof: Exercise. In other words: two equivalent formulations of induction!

115. Coalgebra and coinduction Coinduction = deﬁnition and proof principle for coalgebras. Coinduction is dual to induction, in a very precise way. Categorically, coinduction is a property of ﬁnal coalgebras. Algorithmically, coinduction generalises Robin Milner’s bisimulation proof method.

119. Coalgebras and bisimulations (ex. streams) We call R ⊆ Nω × Nω a bisimulation if, for all (σ, τ) ∈ R, (i) head(σ) = head(τ) and (ii) (tail(σ), tail(τ)) ∈ R Equivalently, R ⊆ Nω × Nω is a bisimulation if Nω head, tail R γ∃ π1oo π2 GG Nω head, tail N × Nω N × Roo GG N × Nω for some function γ : R → N × R.

120. Coalgebras and bisimulations (ex. streams) We call R ⊆ Nω × Nω a bisimulation if, for all (σ, τ) ∈ R, (i) head(σ) = head(τ) and (ii) (tail(σ), tail(τ)) ∈ R Equivalently, R ⊆ Nω × Nω is a bisimulation if Nω head, tail R γ∃ π1oo π2 GG Nω head, tail N × Nω N × Roo GG N × Nω for some function γ : R → N × R.

121. Final coalgebras and bisimulations Theorem: coinduction proof principle Every bisimulation R ⊆ Nω × Nω is contained in the diagonal: R ⊆ ∆ where ∆ = {(σ, σ) | σ ∈ Nω}. Proof: Because (Nω, head, tail ) is a ﬁnal coalgebra, Nω head, tail R γ∃ π1oo π2 GG Nω head, tail N × Nω N × Roo GG N × Nω we have π1 = π2, which implies σ = τ, for all (σ, τ) ∈ Nω.

122. Final coalgebras and bisimulations Theorem: coinduction proof principle Every bisimulation R ⊆ Nω × Nω is contained in the diagonal: R ⊆ ∆ where ∆ = {(σ, σ) | σ ∈ Nω}. Proof: Because (Nω, head, tail ) is a ﬁnal coalgebra, Nω head, tail R γ∃ π1oo π2 GG Nω head, tail N × Nω N × Roo GG N × Nω we have π1 = π2, which implies σ = τ, for all (σ, τ) ∈ Nω.

123. Final coalgebras and coinduction The following are equivalent: 1. For every bisimulation relation R ⊆ Nω × Nω, R ⊆ ∆ 2. ?? In other words: no obvious equivalent formulation of coinduction!

126. Congruences and bisimulations: dual? R ⊆ N × N is a congruence if 1 + N [zero, succ] 1 + R γ∃ oo GG 1 + N [zero, succ] N Rπ1 oo π2 GG N R ⊆ Nω × Nω is a bisimulation if Nω head, tail R γ∃ π1oo π2 GG Nω head, tail N × Nω N × Roo GG N × Nω

127. Congruences and bisimulations: dual? R ⊆ S × T is an F-congruence if F(S) α F(R) γ∃ oo GG F(T) β S Rπ1 oo π2 GG T R ⊆ S × T is an F-bisimulation if S α R γ∃ π1oo π2 GG T β F(S) F(R)oo GG F(T)

128. Induction and coinduction: dual? For every congruence relation R ⊆ N × N, ∆ ⊆ R For every bisimulation relation R ⊆ Nω × Nω, R ⊆ ∆

129. Induction and coinduction: dual? For every congruence relation R on an initial algebra: ∆ ⊆ R For every bisimulation relation R on a ﬁnal coalgebra: R ⊆ ∆

130. An aside: fixed points Let (P, ≤) be a preorder and f : P → P a monotone map. Classically, least fixed point induction is: ∀p ∈ P : f(p) ≤ p ⇒ µf ≤ p Classically, greatest fixed point coinduction is: ∀p ∈ P : p ≤ f(p) ⇒ p ≤ νf

133. An aside: ﬁxed points Any preorder (P, ≤) is a category, with arrows: p → q ≡ p ≤ q Any monotone map is a functor: p → q → f(p) → f(q) Lfp induction and gfp coinduction become: f(µf) GG f(p) µf GG p p GG νf f(p) GG f(νf)

136. Fixed point (co)induction = initiality and ﬁnality f(µf) GG f(p) µf GG p p GG νf f(p) GG f(νf) F(A) GG F(S) A ∃ ! GG S S ∃ ! GG Z F(S) GG F(Z)

137. Fixed point (co)induction = initiality and ﬁnality f(µf) GG f(p) µf GG p p GG νf f(p) GG f(νf) F(A) GG F(S) A ∃ ! GG S S ∃ ! GG Z F(S) GG F(Z)

138. 5. What coalgebra studies • the behaviour of – often inﬁnite, circular – systems (their equivalence, minimization, synthesis) • rather: the universal principles underlying this behaviour • these days applied in many different scientiﬁc disciplines

142. Example: dynamical systems A dynamical system is: set of states X and a transition function t : X → X Notation for transitions: x → y ≡ t(x) = y Examples: x GG y GG z poo r GG s q dd

143. Example: systems with output A system with output: o, t : X → O × X Notation: x a GG y ≡ o(x) = a and t(x) = y. x0 a GG x1 b GG x2 a GG x3 b ww y0 a VV y1 b xx

144. Example: inﬁnite data types For instance, streams of natural numbers: Nω = {σ | σ : N → N } The behaviour of streams: (σ(0), σ(1), σ(2), . . .) σ(0) GG (σ(1), σ(2), σ(3), . . .) where we call σ(0): the initial value (= head) σ = (σ(1), σ(2), σ(3), . . .): the derivative (= tail)

145. Example: streams (1, 1, 1, . . .) 1 GG (1, 1, 1, . . .) 1 GG (1, 1, 1, . . .) 1 GG · · · (1, 1, 1, . . .) 1

146. Example: streams (1, 1, 1, . . .) 1 GG (1, 1, 1, . . .) 1 GG (1, 1, 1, . . .) 1 GG · · · (1, 1, 1, . . .) 1

147. Example: streams (1, 2, 3, . . .) 1 GG (2, 3, 4, . . .) 2 GG (3, 4, 5, . . .) 3 GG · · · (1, 2, 3, . . .) 1 GG (2, 3, 4, . . .) = (1, 2, 3, . . .) + (1, 1, 1, . . .) (1, 2, 3, . . .) 1 GG 1 (1, 1, 1, . . .) 1

150. Example: non-well-founded sets Historically important: Peter Aczel’s book. x = {x} y = {y} xDD y rr x = {y} y = {z} z = {x, y} x GG y GG z yy ––

153. Example: automata A deterministic automaton GG x b a GG z b zz a }} y b yy a tt • initial state: x • ﬁnal states: y and z • L(x) = {a, b}∗ a

156. All these examples: circular behaviour GG x b a GG z b zz a }} y b yy a tt y0 a VV y1 b xx (1, 2, 3, . . .) 1 GG 1 (1, 1, 1, . . .) 1

157. Where coalgebra is used • logic, set theory • automata • control theory • data types • dynamical systems • games • economy • ecology

158. 6. Discussion • New way of thinking – give it time • Extensive example: streams (Lecture two) • Algebra and coalgebra (Lecture three and four) - bisimulation up-to - cf. CALCO • Algorithms, tools (Lecture four) - Cf. Hacking nondeterminism with induction and coinduction Bonchi and Pous, Comm. ACM Vol. 58(2), 2015

Jan Rutten - Concrete coalgebra: an introduction by examples - Lecture 1

More Related Content

Viewers also liked (13)

Similar to Jan Rutten - Concrete coalgebra: an introduction by examples - Lecture 1 (20)

Recently uploaded (20)

Jan Rutten - Concrete coalgebra: an introduction by examples - Lecture 1