Chapter 8 Introduction to Turing Machines (part b)

Luxemburg
1

Outline
Problems that Computers Cannot Solve The Turing Machine (TM) (the above two sections are in part a) Programming Techniques for TMs Extensions to the Basic TM Restricted TMs TMs and Computers
2

8.3 Programming Techniques for TMs

Concepts to be taught
Showing how a TM computes. Indicating that TMs are as powerful as conventional computers. Even some extended TMs can be simulated by the original TM.

8.3 Programming Techniques for TMs

Section 8.2 revisited
TMs may be used as a computer as well, not just a language recognizer. Example 8.4 (not taught in the last section)

called monus, Design a TM to compute a function _ or proper subtraction defined by

n =mn m_
=0 if m n; if m < n.
4

8.3 Programming Techniques for TMs

Section 8.2 revisited
Example 8.4 (contd) Assume input integers m and n are put on the input tape separated by a 1 as 0m10n The TM is M = ({q0, q1, , q6}, {0, 1}, {0, 1, B}, d, q0, B). No final state is needed.
5

8.3 Programming Techniques for TMs

Section 8.2 revisited
Example 8.4 (contd) M conducts the following computation steps: 1. find its leftmost 0 and replaces it by a blank; 2. move right, and look for a 1; 3. after finding a 1, move right continuously 4. after finding a 0, replace it by a 1; 5. move left until finding a blank, & then move one cell to the right to get a 0; 6. repeat the above process.

8.3 Programming Techniques for TMs

Section 8.2 revisited
symbol

state
q0 q1 q2 q3 q4 q5 q6

0
(q1, B, R) (q1, 0, R) (q3, 1, L) (q3, 0, L) (q4, 0, L) (q5, B, R) -

1
(q5, B, R) (q2, 1, R) (q2, 1, R) (q3, 1, L) (q4, B, L) (q5, B, R) -

B
(q4, B, L) (q0, B, R) (q6, 0, R) (q6, B, R) -

8.3 Programming Techniques for TMs

Section 8.2 revisited q00010 1 Bq1010 3 B0q110 4 B01q20 5 B0q311 9 Bq3011 8 q3B011 10 Bq0011 1 BBq111 4

BB1q21 6 BB11q2B 7 BB1q41 12 BBq41B 12

Bq4BBB 13 B0q6BB halt! q00100 Bq1100 B1q200 Bq3110 q3B110 Bq0110 BBq510 BBBq50 BBBBq5B BBBBBq6 halt!
8

8.3 Programming Techniques for TMs

8.3.1 Storage in the State
Technique: use the finite control of a TM to hold a finite amount of data, in addition to the state (which represents a position in a TM program). Method: think of the state as [q, A, B, C], for example, when think of the finite control to hold three data elements A, B, and C. See the figure in the next page (Figure 8.13)

8.3 Programming Techniques for TMs

Figure 8.13
q
A
Track 1 Track 2 Track 3

X Y
Z

Figure 8.13. A TM viewed as having finite control storage and multiple tracks.

10

8.3 Programming Techniques for TMs

8.3.1 Storage in the State
Example 8.6: Design a TM to recognize 01* + 10*. The set of states are of the form [qi, X] where qi = q1, q2; X = 0, 1, B. The control portion (state) remembers what the TM is doing (q0 = not read 1st symbol; q1 = reverse). The data portion remembers the first symbol seen (0, or 1).
11

8.3 Programming Techniques for TMs

8.3.1 Storage in the State
Example 8.6 (contd): The transition function d is as follows.
d([q0, B], a) = ([q1, a], a, R) for a = 0, 1. --- Copying the symbol it scanned.
d([q1, a],a) = ([q1, a],a, R) wherea is the complement of a = 0, 1. --- Skipping symbols which are complements of the 1st symbol read (stored in the state as a). d([q1, a], B) = ([q1, B], B, R) for a = 0, 1. --- Entering the 12 accepting state [q1, B] when reaching the 1st blank.

8.3 Programming Techniques for TMs

8.3.1 Storage in the State
Example 8.6 (contd): Why does not the TM designed by adding data in states in the above way increase computing power? Answer: The states [qi, X] with qi = q1, q2; X = a, b, B, is just a kind of state labeling, so they can be transformed, for example, into p1 = [q0, a], p2 = [q0, b], p3 = [q0, B], . Then, everything is the same as a common TM. 13

8.3 Programming Techniques for TMs

8.3.2 Multiple Tracks
We may think the tape of a TM as composed of several tracks. For example, if there are three tracks, we may use the tape symbol [X, Y, Z] (like that in Figure 8.13). Example 8.7 --- see the textbook. The TM recognizes the non-CFL language L = {wcw | w is in (0 + 1)+}. Why does not the power of the TM increase in this way? Answer: just a kind of tape symbol labeling.

14

8.3 Programming Techniques for TMs

8.3.3 Subroutines
The concept of subroutine may also be implemented for a TM. For details, see the textbook. Example 8.8 --- design a TM to perform multiplication on the tape in a way of transformation as follows: 0m10n1 0mn For details, see the textbook.
15

8.4 Extensions to the Basic TM

Extended TMs to be studied:
Multitape Turing machine Nondeterministic Turing machine

The above extensions make no increase of the original TMs power, but make TMs easier to use:
Multitape TM --- useful for simulating real computers Nondeterministic TM --- making TM programming easier.
16

8.4 Extensions to the Basic TM

8.4.1 Multitape TMs Finite control
Tape 1 Tape 2 Tape 3

Figure 8.16. A multitape TM.

17

8.4 Extensions to the Basic TM

8.4.1 Multitape TMs
Initially,
the input string is placed on the 1st tape; the other tapes hold all blanks; the finite control is in its initial state; the head of the 1st tape is at the left end of the input; the tape heads of all other tapes are at arbitrary positions.

A move consists of the following steps:

the finite control enters a new state; on each tape, a symbol is written; each tape head moves left or right, or stationary.
18

8.4 Extensions to the Basic TM

8.4.2 Equivalence of One tape & Multitape TMs
Theorem 8.9 Every language accepted by a multitape TM is recursive enumerable. (That is, the one-tape TM and the multitape one are equivalent) Proof: see the textbook.
19

8.4 Extensions to the Basic TM

8.4.3 Running Time and the Many-Tapes-toOne Construction
Theorem 8.10 The time taken by the one-tape TM of Theorem 8.9 to simulate n moves of the k-tape TM is O(n2). Proof: see the textbook. Meaning: the equivalence of the two types of TMs is good in the sense that their running times are roughly the same within polynomial complexity.
20

8.4 Extensions to the Basic TM

8.4.4 Nondeterministic TMs
A nondeterministic TM (NTM) has multiple choices of next moves, i.e.,

d(q, X) = {(q1, Y1, D1), (q2, Y2, D2), , (qk, Yk, Dk)}.
The NTM is not any powerful than a deterministic TM (DTM), as said by the following theorem.

Theorem 8.11
If MN is NTM, then there is a DTM MD such that L(MN) = L(MD). (for proof, see the textbook)

21

8.4 Extensions to the Basic TM

8.4.4 Nondeterministic TMs
The equivalent DTM constructed for a NTM in the last theorem may take exponentially more time than the DTM. It is unknown whether or not this exponential slowdown is necessary!

More investigation will be done in Chapter 10.

22

8.5 Restricted TMs

Restricted TMs to be studied:
the tape is infinite only to the right, and the blank cannot be used as a replacement symbol; the tapes are only used as stacks (stack machines); the stacks are used as counters only (counter machines).

The above restrictions make no decrease of the original TMs power, but are useful for theorem proving. Undecidability of the TM also applies to these restricted TMs.

23

8.5 Restricted TMs

8.5.1 TMs with Semi-infinite Tapes
Theorem 8.12 Every language accepted by a TM M2 is also accepted by a TM M1 with the following restrictions: M1s head never moves left of its initial position (so the tape is semi-infinite essential); M1 never writes a blank. (i.e., M1 and M2 are equivalent) Proof. See the textbook.

24

8.5 Restricted TMs

8.5.2 Multistack Machines
Multistack machines, which are restricted versions of TMs, may be regarded as extensions of pushdown automata (PDAs). Actually, a PDA with two stacks has the same computation power as the TM. See Fig.8.20 for a figure of a multistack TM. Theorem 8.13 If a language is accepted by a TM, then it is accepted by a two-stack machine. Proof. See the textbook. 25

8.5 Restricted TMs

8.5.3 Counter Machines
There are two ways to think of a counter machine. Way 1: as a multistack machine with each stack replaced by a counter regarded to be on a tape of a TM. A counter holds any nonnegative integer. The machine can only distinguish zero and nonzero counters. A move conducts the following operations: changing the state; add or subtract 1 from a counter which cannot becomes negative.

26

8.5 Restricted TMs

8.5.3 Counter Machines
Way 2: as a restricted multistack machine with each stack replaced by a counter implemented on a stack of a PDA.
There are only two stack symbols Z0 and X. Z0 is the initial stack symbol, like that of a PDA. Can replace Z0 only by XiZ0 for some i 0. Can replace X only by Xi for some i 0.

For an example of a counter machine of the 2nd type, do the exercise (part a) of this chapter.

27

8.5 Restricted TMs

8.5.4 The Power of Counter Machines
Every language accepted by a one-counter machine is a CFL. Every language accepted by a counter machine (of any number of counters) is recursive enumerable. Theorem 8.14 Every recursive enumerable language is accepted by a three-counter machine.

Proof. See the textbook.

28

8.5 Restricted TMs

8.5.4 The Power of Counter Machines
Theorem 8.15 Every recursive enumerable language is accepted by a two-counter machine. Proof. See the textbook.

29

8.6 Turing Machines and Computers

In this section, it is shown informally that:
a computer can simulate a TM; and that a TM can simulate a computer.

That means:
the real computer we use every day is nearly an implementation of the maximal computational model. under the assumptions that the memory space (including registers, RAM, hard disks, ) is infinite in size. the address space is infinite (not only that defined by 32 bits used in most computers today).
30

8.7 Turing Machines and Computers

8.7.1 Simulating a TM by Computer
If the previous two assumptions are not satisfied, then a real computer is actually a finite automaton! We can simulate an infinite memory space by storage swapping. Also, simulating the infinite tape of the TM by two stacks of disks, respectively for the left portion and the right portion of the tape, with the head as the middle.
31

8.6 Turing Machines and Computers

8.6.1 Simulating a TM by a Computer
Write a program on the computer to simulate the states and the symbols of the TM in the following way:
encode the states as character strings; encode the tape symbols with fixed-length character strings, too; use a table of transitions to determine each move.

By the above way, a TM may be said to be simulatable by a program of a real computer (informally)!

32

8.6 Turing Machines and Computers

8.6.2 Simulating a Computer by a TM
Meaning of this section: The TM is as powerful as a modern-day computer though it seems so simple! Sketch of the simulation using a Multitape TM (see Fig. 8.22)
use a TM tape as the computer memory; use a TM tape to simulate the instruction counter; use a TM tape for the memory address; use a TM tape as scratch to perform computation operations on it.

33

8.6 Turing Machines and Computers

8.6.2 Simulating a Computer by a TM
Meaning of this section: the TM is as powerful as a modern-day computer though it seems so simple! Sketch of using a Multitape TM (see Fig. 8.22) to simulate the sequence of instructions (described by an assembly-language program usually) of the computer:
use a TM tape as the computer memory; use a TM tape to simulate the instruction counter; use a TM tape for the memory address; use a TM tape as scratch to perform computation operations on it.

34

8.6 Turing Machines and Computers

8.6.2 Simulating a Computer by a TM
The TM simulates the instruction cycle of the computer using the above tapes. For more details, see pp. 366-367 of the textbook. Assume that the computer has an accept instruction. The TM simulate it and enters an accepting state. Essence of simulation above:
the TM has many tapes of different purposes to use; the TM can do any computation on the tapes.
35

8.6 Turing Machines and Computers

8.6.3 Comparing the Running Times of

Computers and Turing Machines

If the simulation discussed in the previous section take exponential times, then it is less meaningful. What is the fact? We hope the two types of machines are polynomially

equivalent, i.e., the computer is simulatable by the

TM in polynomial time. The answer is yes! (contd)
36

8.6 Turing Machines and Computers

8.6.3 Comparing the Running Times of Computers and Turing Machines
Theorem 8.17 If a computer: (1) has only instructions that increase the maximum word length by at most 1 and; (2) has only instructions that a multitape TM can perform on words of length k in (k2) steps or less, then the TM described in Section 8.6.2 can simulate n steps of the computer in O(n3) of its own steps. (see the textbook for a proof)
37

8.6 Turing Machines and Computers

8.6.3 Comparing the Running Times of Computers and Turing Machines
Theorem 8.18

A computer of the type described in Theorem 8.17 can be simulated for n steps by a one-tape TM, using at most O(n6) steps for the TM. Conclusion: the TM is as powerful as a real computer seen today!
38