Bottomed Stream Representations of spaces induced by dynamical systems Hideki Tsuiki Kyoto University ! Representing Streams II 2014, Jan. 20 ー31, Leiden %9%H%j!<%‘%W%m%0% Computation over infinite sequences 0 1 0 • 1 Stream Programming (Type2 machine [Weihrauch, et.al]) 0 1 0 0 1 0 1 Input Type2 Machine (Stream Programming) Output % % 1 • Real number computation via binary expansion into infinite sequences. L F J 9 0 0 Unnatural computational notion on R. %\%H%‘F~$jJ8;zNs$rMQ$$$?!" 0LAj6u4V>e$N7W;; Admissible Representation 5 4 3 5 4 3 2 2 0 1 0.0 1 0.5 1 1.0 0.0 0 1 0.5 1.0 • One cannot determine the first digit of 1/2 from any approximation. Intervals of 0 and 1 are overlapping only at one point. • Giving more redundancy so that one can output 0 or 1 when enough good approximation is given. • Admissible representation. (Signed digit representation, etc.) [Weihrauch, Schroder,…]. Type2-machie + Adm. rep [0,1] admissible representa.on (Signed-­‐digit, Cauchy sequence, …) {0,1}ω • • Type2-machine • This idea dates back to [Turing 36]. (One of the) standard ways of defining the notion of computation over the reals. I have a different approach. Gray Representation 5 4 3 5 4 3 2 2 1 1 0.0 • • 0.5 1.0 0.0 Taking the “exclusive or” of a digit and the next digit. Expansion based on Gray-code. 0.5 1.0 n 0 1 2 3 4 5 6 7 8 … Binary 0 1 10 11 100 101 110 111 1000 Gray 0 1 11 10 110 111 101 100 1100 Gray Representation 5 4 3 1 0 1 0 1 0 5 4 3 0 2 1 0 2 1 1 01 1 0 1 0.0 • • • • • • 0.5 1.0 0.0 0 0 0.5 1.0 For T = {0,1,⊥}, unique representation of [0,1] with T- sequences. ⊥ means undefinedness or non-terminating computation. Each sequence contains at most one ⊥. Define Tωn as the set of T-sequences with at most n copies of ⊥. Tωn has the subspace topology of the Scott topology of Tω . This rep. is a topological embedding φ of [0,1] into Tω1 . How to Output Gray-code 5 4 3 2 0 1 1 0.0 0.5 1.0 1 How to Output Gray-code 5 4 3 ⊥ 1 ⊥ ⊥ ⊥ ⊥ ... 2 0 1 ⊥ ⊥ ⊥ ⊥ ... 1 1 ⊥ ⊥ ⊥ ⊥ ... ⊥ 1 0 ⊥ 0 ⊥ 0 ⊥ 0 ... 1 0.0 0.5 1.0 ⊥:undefined currently. IM2-machine 0 1 • • • Two-head access to input/ output tapes. 0 1 Indeterministic behavior depending on how to input when both input heads have values. More generally, defined as an n-head machine. 0 1 0 Input IM2-Machine (two-heads I/O) Output 1 0 0 1 0 1 0 IM2-machine + Gray-code IM21-machine + Gray-rep. and Type2-machine + Signed-digit rep. induce the same computability notion on [0,1] {0,1, }ω [0,1] [0,1] admissible representation IM2 machine {0,1}ω • A machine is operating directly on points. • Computation which reflect the structure Type2-machine of the space. Bottomed Stream and related works Relation between regularities of them Combinatorial Dynamical System (joint work with Izumi property FullTakeuti and Shuji folding maps Yamada) Generalization Gray Independent [Ohta,T,Y amada, 2010] Subbase 5 4 3 Dimension of bottomed seq. 2 Gray 1 0.0 0.5 1.0 [T, Yamada, 2009] 0 1 0 1 0 1 0 1 0 IM2 machine Input (two-heads I/O) Output 1 0 [T, 2000] 0 IM2-Machine [T, 2002] 1 0 Programming Language Computation Implementation similar to REAL as an extension PCF of Haskell [Terayama,T, 2013] [T, 2004] [T, 2012] of Conversion from Binary to topology [T, Sugihara 2005] Domain Theory Domain Representation Implementation in Concurrent Logic Languages [T, 2004] [T, Tsukamoto, 2013] [T, 2005] n BD : • • • • • domain of n⊥-sequences 0̅, 1̅ : output of 0 or 1 from the blue head. Finite/infinite-time output-tape states of an IM2-machine: ∞ ̅ ̅ quotient of {0, 1, 0, 1} with a ̅ b = b a for a, b ∈ {0, 1}. —> BD1 : the domain of n⊥-sequences. Similarly, we define the domain BDn of n⊥-sequences as the quotient of {0(k), 1(k) | k < n} ∞with a(k+1) b(l) = b(l) a(k) for a, b ∈ {0, 1}. The set of limit elements of BDn is Tωn . 010ω 110ω Tω1 0⊥0 ω 01 ⊥ ⊥ ⊥10ω 1 ........................................................... 0⊥1 00 ̅ 1 01 0⊥0 0 ⊥10 0 ̅ 0 1 0 0 ̅ 0 1 0 ̅ 1 0⊥ ⊥ ⊥ ⊥00 ̅ 0 … 0 ̅ 0 1 … … 0 ⊥ 1⊥ ⊥ ̅ 0 ⊥⊥⊥⊥ 1 1 … … Finite/infinte-time state of a tape for the Gray-embedding • • • We need ⊥1, but not ⊥0. There is a point x such that φ(x) = ⊥1…., but not ⊥0…. Finite-tme states : {φ(x)|n : x ∈ X, n ∈ N}. (here, p|n is the prefix of a with length n.) • • Infinite-time states: ideal completion. c.f. signed digit representation. The space of minimal limit elements homeomorphic to I 010ω 110ω ω ⊥10 .............................................................................. 010 100 110 00 01 ̅ 1 0 1 0 0 0⊥1 0 11 1⊥1 ⊥10 ̅ 1 1 0 ̅ 1 1 1 1 ̅ 1 10 0 For every n-­‐dimensional separable metrizable space X, there exists an embedding φ of X in BDn such that X is homeomorphic to the minimal-­‐limit set of the domain constructed in this way. Regular n -domain …… ̅ 0 0⊥1 00 ̅ 1 0 ⊥10 01 1 0 0 0 ̅ 1 0 ⊥1 ̅ 1 1⊥1 11 1 1 ̅ 1 10 0 1 B ̅ 1 0,1 • A language A ⊆ {0, 1, 0̅, 1}̅ * recognizable by a finite state automaton F which satisfies 𝛿(s, ab̅ ) = 𝛿(s, ba) for every state s of F. Take a quotient of A with ab̅ =ba . Can be generalized to a subdomain of BDn. We call such a domain a regular n⊥-domain. • a n⊥-domain is full if {0,1}* ⊆ A. • • • A 0,1 Regular n -domain • Definition. For n ∈ N, consider the alphabet {0(k), 1(k) | k < n} . Let A be a language recognizable by a finite-state automaton F such that 𝛿(s, a(k+1) b(l)) = 𝛿(s, b(l) a(k)) for every state s of F. A regular n⊥-domain D is the ideal completion of a poset K obtained as the quotient of such a language A with a(k+1) b(l) = b(l) a(k) for a, b ∈ {0, 1}. ! • A n⊥-domain is a subdomain of BDn. Gray-code: itinerary of the tent function. • Tent function: ⇢ 2x f (x) = 1 2x (0 x 1/2) (1/2 < x 1) • Dynamical system f: I → I. • Itinerary of the tent function: Infinite sequence of {0,1,⊥} which codes whether x is mapped to [0,1/2), (1/2, 1], or {1/2} by its repetition. • 8 < 0 1 'f (x)(n) = : ? (0 f n (x) < 1/2) n (1/2 < f (x) 1) n n (1/2 < f (x) 1) (f (x) = 1/2) 1 0 0 1/2 S0,0 S0,1 S1,0 S1,1 S2,0 S2,1 S3,0 S3,1 Gray-expansion is nothing but itinerary of the tent function. 1 Generalization to other dynamical systems. • A dynamical system with the notion of full-itinerary. ! • Each point x is identified by its itinerary φ(x) . ! • Full-folding map Exact full-folding map {φ(x)|k : x ∈ X, k ∈ N} generates a regular n⊥-domain for some n. Exact full-folding map with this property. Full-folding map • X X: compact metrizable space. f flip X • Definition: A continuous map f: X →X is a full-folding map if, for a regular open set X0 and X1 = X \ X0, f |X̄i : X̄i ⇠ = X (homeo) for i = 0, 1. U is regular open if it is the interior of its closure. • • • We define C = X \ (X0 ∪ X1 ) . f is 2-1 on X0 ∪ X1 and 1-1 on C. We define homeomorphisms go : X → Xi (i = 0, 1) as gi = f |X̄i 1 . Real Kneading [Wikipedia] Itinerary of a full-folding map If 𝑓:𝑋→𝑋 is a full-­‐folding map, we define its i6nerary 𝜑𝑓:𝑋→{0,1,⊥}𝜔 as • Exact full-folding map flip • • • • • flip flip X is divided into 2𝑛 regions 𝑅(𝑝) (𝑝∈{0,1}𝑛) by taking n inverses. De[ine 𝑔𝑝=𝑔𝑝(0)∘𝑔𝑝(1)∘…∘𝑔𝑝(𝑛−1) . cl 𝑅(𝑝) =𝑔𝑝(𝑋) and cl 𝑅(𝑝) is homeomorphic to 𝑋. We say that a f-­‐f. map 𝑓 is exact if max{diam cl 𝑅(𝑝) | 𝑝∈{0,1}𝑛 } shrinks to 0 as 𝑛→∞ We need exactness for coding! • It is easy to show that the tent map is the only exact f.f. map on I=[0,1] modulo conjugacy. ! • Characteriza6on of exact full-­‐folding 2 maps on D=𝐼 which induce regular n⊥-­‐domains. f : X ! X is conjugate to g : Y ! Y if there is a homeomorphism h : X ! Y such that E f =g h X h Y f g X h: homeo. Y First, consider the case X is homeomorphic to the unit disk D and g0 and g1 are similarity maps. • Recutangular full-­‐folding map. • Triangular full-­‐folding map. Rectangular Full-­‐folding map. 1 011 001 0 0 010 000 1 110 100 111 101 ! ! 1 0 0 1 Itinerary of (x, y): interleaving inverted itineraries of x and y with tent map flip : ⊥⊥00111… Triangular Full-­‐folding map 011 010 001 100 101 110 f(C) 000 111 g0 11 X0 f X X1 g1 flip : ⊥0⊥⊥0111… Full-­‐Folding maps of 𝐷 flip 11 Categories of Full-­‐folding maps on D A(, A) B(, B) C(, C) E(, E) G J I H F Acute f.f. map D(, D) K Only A, B, J contain exact full-­‐folding maps. Unimodal maps • • • 1 X1 h:I → I is a unimodal map if h(0) = h(1) and h is increasing on [0,1/2] and decreasing on [1/2, 1]. I6neray is defined and plays an important role in Kneading Theory [Milnor-­‐Thurston, …] 𝜑(1/2) determine approx. behavior of S the dynamical system. S I6nerary can be exact, but not full S except for the tent func6on and its conjugates. X0 • 0 0,0 S0,1 1,0 S1,1 2,0 S2,1 1/2 000… does not exist. 1 Boundary restriction of a f.f. map. • f • Conversely, we can form a f.f. map from a unimodal map, if it is not conjugate to the tent map. Itineraries of f.f. map and unimodal map. • X1 011 010 X0 001 011 010 110 111 000, 100… do not exist. 101 000, 100… exist inside the disk. 000 001 100 101 110 111 Itineraries of a and b determine exact f.f. map. • A B J Cutting/Co-­‐cutting time • (Hofbauer, Keller, Bruin, …) There a re f our c ases. For a f.f. map of category A, ! an = a for some n. The number of boaoms in the i6nerary of a is infinite. Do not induce a n⊥-­‐domain for any n. • Both cubng-­‐6me and co-­‐cubng 6mes are finite. Kneading sequence pω for an odd p. No exact f. f. map. • Cubng 6me is infinite, co-­‐cubng 6me is finite. Kneading sequence pω for an even p. No exact f. f. map. • Both cubng and co-­‐cubng 6mes are infinite. • Post cri6cally finite. ( ∪fn(a) is a finite set) Kneading sequence qpω for nonempty p, q (ultimately periodic). Exact f.f. maps exist, construc6on via complex dynamical system. • Otherwise… f.f. map with dense Inverse image of the boundary. Similar result for categories B, taking care of both cri6cal points. • Theorems • A conformal full-folding map (i.e., both g0 and g1 (z̄) are conformal maps from the upper half disk) on I 2 is semi-conjugate to a complex quadratic rational map with real coefficients. Therefore, modulo conjugacy, it has the form az 2 + b f (z) = 2 . cz + d with a, b, c, d 2 R. The conjugacy map identifies complex conjugacy to map to the closure of the upper half plane and make a Mobius transformation to the unit disk. • It is exact and induces a full rational n?-domain if f is postcritically finite and has no super attractive periodic orbit. Correspondence between regularity of tape-states of a machine and regularity of the dynamical system which determine the code. 34 Conclusion Bottomed sequences enable symbolic manipulation of continuous objects. • • • The space of bottomed sequences is continuous and, at the same time, can be manipulated symbolically. • Separable metrizable spaces can be embedded into the space of bottomed sequences. Studied how properties of continuous objects (with respect to coding) are reflected to the structure of an IM2-machine operating on it. • Dimension as the minimal number of heads. • Regularity of the dynamical system which induces the coding as regularity of the machine behavior with respect to the next input. We are interested in logical/coalgebraic treatment of bottomed sequences for programming over them.