1. ::ICS 804:: Theory of Computation - Ibrahim Otieno - iotieno@uonbi.ac.ke +254-0722-429297 SCI/ICT Building Rm. G15

2. Course Outline Mathematical Preliminaries Turing Machines Recursion Theory Markov Algorithms Register Machines Regular Languages and finite-state automata Aspects of Computability

3. Last Time: Turing Machines What is Computation? Informal Description of Turing Machines Formal Description of Turing Machines Turing Machines as Language acceptors and Language recognizers Turing Machines as Computers of number- theoretic functions Modular Turing Machines Complexity Theory

4. Course Outline Mathematical Preliminaries Turing Machines ◦ Additional Varieties of Turing Machines Recursion Theory Markov Algorithms Register Machines Regular Languages and finite-state automata Aspects of Computability

5. Additional Varieties of Turing Machines Turing Machines with One-Way-Infinite tape Turing Machines that accept by terminal state Multitape Turing Machines Encoding of Turing Machines Universal Turing Machines Nondeterministic Turing Machines Turing-Computability Turing Machines and Artificial Intelligence Turing Machines and Cognitive Science

6. Turing Machines with One-Way-Infinite Tape

7. Tapes So far: Turing machines with two-way infinite tape Turing (1936): one-way infinite tape No difference in computational power A function Turing computable in a 2-way infinite tape setup is also Turing computable with a 1-way infinite tape

8. Turing Machines with One-Way-Infinite Tape Machine tape infinite only to the right Leftmost tape square containing left endmarker (use, in Deus Ex Machina) What happens on reaching left most square? One-way function computation One-way language acceptance (recognition)

9. Equivalence Result Central to theory of computability Let f be a k-ary number-theoretic function: Then there exists a Turing Machine with two-way-infinite tape that two-way computes f if and only if there exists a Turing machine with one-way-infinite tape that one-way computes f

10. Turing Machines that accept by terminal state

11. Acceptance (Recap) Word acceptance: Deterministic Turing Machine M accepts nonempty word w if, when started scanning the leftmost symbol of w on an input tape that contains w and is otherwise blank, M ultimately halts scanning a 1 on an otherwise blank tape. “Acceptance by 1”

12. Terminal State A single terminal state q t (  “one or more”) Sex-tuple  Q, , , q init, q t,   ◦ q t  Q ◦ is undefined for pairs of the form ◦ Turing Machine M can still halt in any other state ◦ Not necessarily distinct from q init ◦ Exercise: Draw a TM that accepts L = { (abc) n | n >=1} by terminal state

13. Language Acceptance by Terminal State M accepts w if halts in q t M accepts language L by terminal state provided that M accepts, by terminal state, all and only the words of L ◦ Tape content upon halting in state q t is not defined ◦ No requirement that M has read word w completely

14. Equivalence Result Theorem: Let L be a language over alphabet . Then there exists a Turing machine that accepts L “by 1” if and only if there exists a Turing machine that accepts L by terminal state. Equivalence result: forward and reverse proof involve maintaining new end markers and 

15. Multitape Turing Machines

16. Has K+2 tapes where K >= 0 (Possibly read-only) input tape at top (Possibly write-only) output tape at bottom 0 or more worktapes in between Referred to as offline TM

17. No worktapes Initially input tape: H11111111 output tape: HB Ultimately halts in the configuration input tape: H11111111 (unchanged) output tape: H1111

18. Worktapes Initially input tape: HababB worktape1: HBBBBB worktape2: HBBBBB … output tape: HBBBBB Ultimately halts in the configuration input tape: HababB (unchanged) worktape1: … worktape2: … … output tape: H1BBB

19. Clarification Let M be a multitape Turing machine. Suppose that tape t is one of M’s k + 2 tapes. In general, determination of which action is to be taken by the read/write head on tape t will depend on the symbols currently being scanned on tapes other than t and not merely on the symbol currently being scanned on tape t itself.

20. Language Acceptance Suppose that (k+2)-tape Turing Machine M is started scanning the leftmost symbol of nonempty input word w on its input tape, which is otherwise blank initially. Suppose all the other tapes are blank as well. Then if M halts with its first read-head scanning the leftmost symbol of w on its input tape, which is otherwise blank, and scanning a single 1 on its output tape, which is otherwise blank, then we shall say that M accepts word w

21. Example {w  *|n a (w) = n b (w)}

22. Example Instructions for each tape. Often: only 1 tape is reading/writing. Other tapes: “a:a” type instructions

23. Example Copies occurrences of a unto worktape1 and occurrences of b unto worktape2. Then M ascertains whether the number of a’s on worktape1 is the same as the number of b’s on worktape2

24. Example Deus Ex Machina: ◦ Machine: SAMNUM4T.TM ◦ Try tapes 6A4BMULT.TT and 9A9BMULT.TT

25. Example Calculate Time complexity Single-tape: O(n 2 ) Multi-tape: O(n)

26. Parallel Computation? Let M be a multitape Turing Machine. Suppose that tape t is one of M’s k+2 tapes. In general, determination of which action is to be taken by the read/write head on tape t will depend on the symbols currently being scanned on tapes other than t and not merely on the symbol currently being scanned No individual processors: no parallel computation Single processor controlling multiple tape heads

27. Exercise Develop a TM that accepts L = { (a n | n is prime}

28. Time/Space trade-off time M (n) space M (n) Adding tapes increases space M but generally decreases time M

29. Equivalence Results Theorem A language L is accepted by some multitape Turing machine if and only if L is accepted by some single-tape Turing machine. Theorem A language L is recognized by some multitape Turing machine if and only if L is recognized by some single-tape Turing machine.

30. Another Equivalence Result Theorem Similarly, a function f is computed by some multitape Turing machine if and only if it is computed by some single-tape Turing machine.  always trivial Proof idea for  : Tape alphabet of single-tape simulator M will be large; one symbol  will represent entire “section” through tapes of multitape M´

31. Time and Space Corollary Suppose that language L is accepted by k-tape Turing machine M. Then L is accepted by a single-tape Turing machine M 1 such that time M 1 (n) is O([time M (n)] 2 ) Corollary Suppose that language L is accepted by k-tape Turing machine M. Then L is accepted by a single-tape Turing machine M´ such that space M´ (n) is O(space M (n)).

32. Off-line Turing Machines Multitape machines: improved time analysis Singletape machines: space analysis involves counting the number of tape squares visited over the course of a computation  usually visit all squares of the input  no single-tape machine will compute in less than O(n) space

33. Encoding of Turing Machines

34. Any Turing machine can be represented by a natural number code 1. ASCII-style encoding 2. Euler-Gödel encoding

35. ASCII Describes language {a n |n  0} Slight changes to make it more general q0q1 q2 a:B B:R B:1 qq’ q’’ s’:s s:R s:s’’

36. ASCII q: is always initial state Any Turing machine: finite set S of quadruples We want to represent S by a single symbol string qq’ q’’ s’:s s:R s:s’’

37. ASCII q: is always initial state Any Turing machine: finite set S of quadruples We want to represent S by a single symbol string 3*2 different ways: q,s’,s,q’;q’,s,R,q;q,s,s’’,q’’ q,s’,s,q’;q,s,s’’,q’’;q’,s,R,q q’,s,R,q;q,s’,s,q’;q,s,s’’,q’’ q’,s,R,q;q,s,s’’,q’’;q,s’,s,q’ q,s,s’’,q’’;q’,s,R,q;q,s’,s,q’ q,s,s’’,q’’;q,s’,s,q’;q’,s,R,q

38. ASCII Each of these is a string over Turing Machine description alphabet , containing the seven symbols q ‘ s L R, ; Represent each of these options as a binary digit string: associate symbols of alphabet with binary-digit strings of e.g. length 3 Member of  Code,000 q001 s010 R011 L100 ;101 ‘110 q,s’,s,q’;q’,s,R,q;q,s,s’’,q’’ q,s’,s,q’;q,s,s’’,q’’;q’,s,R,q q’,s,R,q;q,s’,s,q’;q,s,s’’,q’’ q’,s,R,q;q,s,s’’,q’’;q,s’,s,q’ q,s,s’’,q’’;q’,s,R,q;q,s’,s,q’ q,s,s’’,q’’;q,s’,s,q’;q’,s,R,q

39. ASCII 001 000 010 110 000 010 000 001 110 101 001 110 000 010 000 011 000 001 101 001 000 010 000 010 000 010 110 110 000 001 110 110 = symbol code of machine M Encode M as a natural number: ASCII code of M = natural number denoted by the 90 digit string interpreted as a binary numeral Member of  Code,000 q001 s010 R011 L100 ;101 ‘110 q,s’,s,q’;q’,s,R,q;q,s,s’’,q’’

40. ASCII Turing machines will not have unique ASCII codes No natural number that encodes a Turing machine will be the ASCII code of more than one Turing machine Most natural numbers are not ASCII codes If n is a code, then it is the code of a unique machine (retrievability property).

41. Euler-Gödel Based on prime decomposition e.g. 18720 = 2 5.3 2.5 1.13 1 Associate each member of alphabet  with a natural number :   {1,2,3,4,5,6,7} Member of  Code,1 q2 s3 R4 L5 ;6 ‘7

42. Euler-Gödel List prime numbers for each symbol and put it to the power of the function call of that symbol Member of  Code,1 q2 s3 R4 L5 ;6 ‘7 2 2.3 1.5 3.7 7.11 1.13 3.17 1.19 2.23 7.29 6.31 2.3 7 7.41 1.43 4.47 1.53 4.59 1.61 2.67 6.71 2.73 1.79 3.83 1.89 3.97 7.101 7.103 1.107 2.109 7.113 7. q, s’, s, q’;q’, s, R, q;q, s, s’’, q’’

43. Euler-Gödel Turing machines do not have unique Gödel numbers No natural number that is the Gödel number of a Turing Machine will be the Gödel number of more than one Turing Machine Given a natural number, there is a simple algorithm for determining if it is the Gödel number of a Turing Machine and which one it is

44. Decoding 577565488500 is the alleged Gödel number of a Turing Machine Prime decomposition ◦ 577565488500 = 2 2. 144391372125 ◦ 577565488500 = 2 2. 3 1. 48130457375 ◦ … ◦ 577565488500 = 2 2. 3 1. 5 3. 7 1. 11 4. 13 1. 17 2

45. Decoding  q,B,R,q Memb er of  Code,1 q2 s3 R4 L5 ;6 ‘7 q B:R

46. Universal Turing Machines

47. Universal Turing machines So far: machines in a peculiar sense Single program, single set of instructions Change 1 instruction  change the machine Can our TM be programmed to simulate behavior of any TM whatsoever?

48. Universal Turing machines Usual sense of a machine: machine (hardware) that can run several programs (software) Digital Computer: Universal Computing device that - suitably programmed (and ignoring resource limits) - can compute any function that is in principle computable

49. Definition Turing Machine M* is universal if, for any Turing Machine M with input alphabet , when M* is started scanning the leftmost 1 in an unbroken string of n 0 +1 1s (n 0 being the encoding of machine M) followed by a single blank followed by word w over , M* transforms w exactly as machine M would transform it

50. UTMs as language acceptors / transducers

51. UTMs as Function Computers

52. UTMs Exist Turing (1936) proved this Decode/encoding of Turing Machine and copy it to one or more worktapes Arguments are copied to worktape(s) Then the program is executed Proving that UTM exists different from actually constructing it!

53. UTMs Exist ENIAC (1943): programmed manually by setting flip- flops and plugging wires into circuit boards  Store these instructions in the manner of input data Stored-program computers (EDVAC/1946) Also ACE (Turing, London) General-purpose, executive programs: operating systems ◦ Accept other programs as input and execute it hardware plasticity: ability of one general-purpose machine architecture to mimic other, diverse machine architectures by means of dedicated software (cf. emulators)

54. EDVAC (Electronic Discrete Variable Automatic Computer) more internal memory than any other computing device to date used binary rather than decimal numbers 10 years after Turing!!!

55. History of Computing on the Internet Charles Babbage Institute http://www.cbi.umn.edu/photo/cbi/gallery.htm EDVAC site at http://www.ifi.unizh.ch/se/people/hoyle/Lecture/edva c.html Most highly recommended is the Smithsonian Institute site (Landmarks in Digital Computing) at http://www.nasm.edu/NASMDOCS/DSH/LDC/ldc_c ontents.html

56. Nondeterministic Turing Machines

57. Nondeterministic Turing Machine What is happening here?

58. Nondeterministic Turing Machine L = {(ab) n | n  1}

59. Nondeterministic Turing Machine L = {(ab) n | n  1} L’ = {a n | n  1}

60. Nondeterministic Turing Machine L = {(ab) n | n  1} L’ = {a n | n  1} Alternative execution paths Non-deterministic language accepting Turing machine Should accept L  L’

61. Language Acceptance Accepts w if some computation of M results in an accepting 1 In general there will be “nonaccepting” computations even in the case of an accepted w M accepts L if and only if M accepts all and only the words of L

62. Transition Functions In the case of a nondeterministic M, we have that  M is not single-valued, e.g.,  M (q 0, a) = (B, q 1 ) and  M (q 0, a) = (B, q 6 ). We speak of a transition mapping, instead of transition function Deterministic machines of those among the nondeterministic machines with single-valued transitions mappings. Determinism = a special case of non- determinism.

63. The World of Automata

64. Turing-Computability Theorem Let f be a number- theoretic function. Then f is computed by some nondeterministic M nd if and only if f is computed by some deterministic M d.

65. Time complexity (recap) Complexity class P: class of all languages accepted in polynomial time by some Turing Machine L = {w  *| n a (w) = n b (w) } is a member of complexity class P A language L over alphabet  is said to be polynomial-time Turing-acceptable if there exist both a deterministic TM M and a polynomial p(n) such that for any w  *, we have w  L if and only if M accepts w in O(p(|w|)) steps.

66. A New Complexity Class NP = class of polynomial-time non- deterministically Turing-acceptable languages P  NP trivially Every language in NP is Turing- acceptable

67. P  NP? Is something that is easy to check, also easy to compute? Open problem worth $1M http://www.claymath.org/millennium/P_vs_NP/ Suppose that you are organizing housing accommodations for a group of four hundred university students. Space is limited and only one hundred of the students will receive places in the dormitory. To complicate matters, the Dean has provided you with a list of pairs of incompatible students, and requested that no pair from this list appear in your final choice. This is an example of what computer scientists call an NP-problem, since it is easy to check if a given choice of one hundred students proposed by a coworker is satisfactory (i.e., no pair taken from your coworker's list also appears on the list from the Dean's office), however the task of generating such a list from scratch seems to be so hard as to be completely impractical. Indeed, the total number of ways of choosing one hundred students from the four hundred applicants is greater than the number of atoms in the known universe! Thus no future civilization could ever hope to build a supercomputer capable of solving the problem by brute force; that is, by checking every possible combination of 100 students. However, this apparent difficulty may only reflect the lack of ingenuity of your programmer. In fact, one of the outstanding problems in computer science is determining whether questions exist whose answer can be quickly checked, but which require an impossibly long time to solve by any direct procedure. Problems like the one listed above certainly seem to be of this kind, but so far no one has managed to prove that any of them really are so hard as they appear, i.e., that there really is no feasible way to generate an answer with the help of a computer. Stephen Cook and Leonid Levin formulated the P (i.e., easy to find) versus NP (i.e., easy to check) problem independently in 1971.

68. P  NP? Probable answer: yes NP-complete problems: problems that are most likely not to be in P NP-hard problems: NP problems that can be reduced in polynomial time Traveling salesman: NP-complete problem. If this problem turns out to be in P, P=NP So far: no one has been able to find a polynomial time solution to an NP-hard problem Inherent problem of computation? Or human intelligence problem?

69. Turing Machines and Artificial Intelligence

70. Machine Intelligence Descartes Test: intelligence = a machine’s ability to use language: ◦ More than language recognition? ◦ Communication with humans? ◦ Weizenbaum’s ELIZA?

71. Turing Test Computing Machinery and Intelligence (1950) 3 participants: 2 men (manA and manB) and 1 woman play imitation game: ◦ Room 1: manA as the interrogator ◦ Room 2: manB + woman ◦ Communication through networked computers ◦ Goal: interrogator needs to identify male or female in room2 through questioning (questions to player X and player Y (predefined by manB and woman)) ◦ If the interrogator correctly identifies them, interrogator and woman win  Woman has to help interrogator by answering like a woman  manB has to confuse the interrogator by answering like a woman

72. Turing Test Interrogator has to ask intelligent questions: ◦ Name a shoe manufacturer ◦ Name a football team ◦ What is your shoe size If the 3 participants are of the same intelligence, the interrogator will end up merely guessing ◦ Woman+Interrogator wins half the games ◦ manB wins half the games

73. Turing Test What if the role of manB is played by a computer? Interrogator needs to guess who is the computer and who is the woman Woman will help the interrogator and computer needs to confuse him If the number of games won by the computer is at least as great as that of manB, we pronounce the computer intelligent Computer’s knowledge of the world is just as important as its knowledge of human knowledge limitations But: anthropocentric (cf. extraterrestrial intelligence)

74. Block’s critique Given 60 minutes of Imitation Game, 88 characters, … the class of imitation game conversations is finite A machine Aunt Bubbles has access to all of these conversation and simply searches for the most appropriate next answer Is this machine intelligence? “Intelligence of a jukebox” Deep Blue?

75. Turing Machines and Cognitive Science

76. Cognition Sensory perception, natural language processing, reasoning, judgment, memory, … Computation = one sort of human cognitive activity Cognitive scientists: seek plausible models of cognitive activity Theory of computability can make a contribution to cognitive science

77. Cognition and Turing Machines Strong claim: Universal Turing Machines model human cognition? ◦ Is human cognition goal-oriented? ◦ Is everything we think about computation? Weak claim: every genuinely cognitive aspect of the human mind can be modeled with a Turing Machine, given the correct tape representation. The rest is noise

78. Cognition and Turing Machines Competence models of cognition: formal analysis of overall input/output relations Performance models of cognition: includes modeling of noise Turing Machine modeling: only competence!

79. Exercise 4 Use Deus ex Machina to design a 3- tape turing machine that accepts the language {w  * | w = w R }

80. Exercise 5 Use the ASCII translation table to find the ASCII code of the Turing Machine below q s’:R

81. Exercise 6 Find the Turing Machine encoded by natural number 1,312,648,837,500 Using the Euler-Gödel scheme and the following translation table: Member of  Code,1 q2 L3 R4 S5 ;6 ‘7

82. Exercise 7 (difficult) Use Deus Ex Machina to design a non- deterministic Turing Machine that accepts language {ww|w  *} with  = {a,b} hint: design a machine that nondeterministically selects a candidate midpoint within its input word and then proceeds to compare its front half with its back half, one character at the time

83. Exercise 8 Given the following deterministic Turing Machine that should accept all and only the words over alphabet  = {a,b}, but fails Try word bbaabb

84. Exercise 8 Design a deterministic or nondeterministic Turing Machine which does accept the language L ={waaw’|w,w’   *}