Introduction to the Theory of Computation

Homework #4 Solutions

1. (Problem 3.11) Suppose I have a Turing machine M with a doubly-infinite tape, i.e. a tape
that stretches both left and right to x = . Prove that such a machine can be simulated
with a standard Turing machine M 0 whose tape is only infinite on one side, with x ranging
from 0 to . You dont have to describe the simulation in complete detail, but you should
describe it well enough to say how many states and tape symbols M 0 needs in terms of how
many M has, and how long the computation time of M 0 will be in terms of that of M .
Answer: We can simulate an doubly-infinite tape T with a singly-infinite one T 0 by
folding it. Thus positions x0 = 0, 1, 2, 3, 4 . . . on T 0 will correspond to positions x =
0, 1, 1, 2, 2, . . . on T . To simulate moving one step right on T , we move two steps right
on T 0 if x > 0, and two steps left on T 0 if x < 0; if we hit the left end of T 0 , we know that
x = 0, in which case we can move one step right to x = 1 and know that x > 0. We simulate
moving left on T similarly.
Our simulation T 0 runs half as fast as the original since it takes two steps for every step
of T . It needs about four times as many states; for each state s of T , we need a state for
each combination of s with (x is positive/negative) and (were moved both steps/were in the
middle of moving two steps). The number of tape symbols is the same as for T . (There are
other ways to do this, like using the tape symbol to mark whether x is positive or negative;
this doubles the number of tape symbols but divides the number of states we need by two.)
2. (Problem 3.9) Show that a PDA with two stacks, i.e. a finite-state automaton which has
two stacks it can examine, push, and pop, can simulate a Turing machine. (Note that it
can push one stack while popping the other, so the stacks will not necessarily have the same
depth.) Describe the simulation at the same level of detail as in the previous problem.
Now prove that a Turing machine can simulate a PDA with 3 or more stacks (if you
like you can use the fact that a single-tape Turing machine can simulate a multitape Turing
machine). Therefore, going from 1 to 2 stacks makes a PDA much more powerful, but going
to 3 or more stacks doesnt increase its power.
Answer: To show that a 2-stack PDA can simulate a Turing machine, we let the two
stacks simulate the left and right halves of the tape. Lets use the top symbol of the righthand stack to represent the symbol at our current position on the tape. To move left on
the tape, we pop a symbol from the left stack and push a (possibly re-written) symbol on
the right stack, and to move right on the tape we pop from the right and push on the left.
1

How many steps of the 2-stack PDA this takes depends on exactly how we define it, but the
computation time and the number of states both get multiplied by a small constant. The
stack alphabet is simply the tape alphabet.
To show that a 3-or-more-stack PDA is no more powerful than a 2-stack one, all we
have to do is note that a multitape Turing machine can easily simulate a multistack PDA,
with each tape simulating a stack with the top symbol at its right end. To push, we write
a symbol and move right; to pop, we simply move left. Then since multitape TMs can be
simulated by single-tape TMs were done.
3. A queue automaton is like a PDA, but with a queue rather than a stack. That is, it is
a DFA which can look at the front symbol of a queue; based on that symbol and its finite
state, it is allowed to change its state, pop the front symbol, and/or push a symbol onto the
back of the queue. Show that, unlike PDAs which can only recognize context-free languages,
a queue automaton can simulate a Turing machine.
Hint; simulating T steps of the Turing machine might take the queue automaton more
than O(T ) steps. How long does it take?
Answer: Note that unlike a PDA, a queue automaton can cycle through its queue by
popping off the front and pushing on the back. While it does this, it can look for a particular
symbol x, and change it by pushing a modified symbol y (or a word of more than one symbol)
by pushing y when it pops x. So, after each step of the TM, our queue contains a string
which consists of the entire tape, with the TMs current position marked with an additional
symbol indicating its state. (We also use markers for the left and right ends of the tape.) Its
easy to see how, by cycling through the queue, the queue automaton can carry out one step
of the Turing machine by reading its current state and the current tape symbol, popping
them off the front, pushing symbols corresponding to the new state and new symbol on the
back, and then continuing cycling through. (In order to move the machine to the left, in
which case the TM marker should be pushed before the tape symbol we read before, we can
keep a one-symbol buffer instead of immediately pushing the symbol we popped.)
If the queue has length L, each step of the TM then takes O(L) steps of the queue. Since
after T steps of the TM L is at most n + T where n is the length of the input, the total time
to simulate T steps is O(LT ) = O((n + T )T ) = O(T 2 ).
4. (Problem 3.13) Suppose that instead of being able to move left or right, a Turing machine
is only allowed to stay put or move right. Show that such a machine can only recognize
regular languages.
Answer: Clearly the TM can never go back and look at what it wrote before. It reads
the input left-to-right, and the stay put instruction corresponds to an -move which does
not read another symbol of the input. Now a DFA M which can make -moves can be
simulated by a standard DFA M 0 which cant (i.e. which reads a symbol at every step), since
at each position in the input M either 1) gets stuck and never accepts, 2) accepts there, or 3)
eventually moves on in some state, which we call the new state of M 0 . (Indeed, our definition
of NFA allowed -moves, and we saw before that they only accept regular languages.)
We also need to think about what happens if the TM moves past the input into the blank
2

section of the tape. In this case we just define A as the set of states for which, if it is in that
state when it first enters the blank part of the tape, it will eventually accept. Then we add
A to the accepting set of M 0 .
5. (Problem 3.19) Let L be the language containing the single string s on the alphabet
{a, b}, where

a if life exists on other planets within 100 light-years
s=
.
b if it doesnt
Is L decidable? Is it regular?
Answer: Either L = {a} or L = {b}; I dont know which it is, but its one of them, and
both of them are clearly regular and decidable.
This may seem like a trick question :-) but the point is that were not asking the
computer to figure out whether there is life on other planets (or whether God exists, which
was Sipsers version); were just asking it to recognize, or print, a single symbol. Consider two
programs, one which prints yes and the other which prints no. One of these programs
is right; I dont know which one, but they both run very quickly.
6. Recall that L is decidable if there is a Turing machine which always halts and, given input
w, says yes if w L and no if w
/ L. A language L is recognizable if there is a Turing
machine which halts and says yes if w L, but may or may not halt if w
/ L. Show that
a language L is decidable if and only if both L and its complement L are recognizable.
Answer: We did this in class. If L and L are both recognizable, we run both machines
in parallel. One and only one of the machines will say yes; if the recognizer for L does then
w L, and if the recognizer for L does then w
/ L.
Conversely, if L is decidable there is a machine which says yes if w L (and no if
w
/ L). This makes it a recognizer for L by definition, and switching yes and no makes
it a recognizer for L. Thus both L and L are recognizable.
7. Recall that a language L is recursively enumerable if there is a Turing machine M that
prints out a list of words in L, such that any given word w in L will eventually appear
somewhere in the list. (Another way to put this is that M takes an input i and prints out a
word f (i), and that w L if and only if w = f (i) for some i.)
Now suppose there is a Turing machine that prints out a list of words in L in increasing
order (using dictionary order). Show that L is decidable.
Answer: L is either finite or infinite. If its finite, its decidable automatically (just look
the input up in a finite list and say yes if its there).
So suppose its infinite. In that case, if w
/ L, M will eventually print out the next word
in L in dictionary order beyond w; but then we know that w
/ L since M will never come
back and print w. So we watch M , and halt and say no if M prints out a word beyond w
in dictionary order.
8. Write the pseudocode of a C program which eventually prints out every C program (which
run on their own with no input) which halts. Feel free to use words like compile and run
3

for t steps, but make sure that every program which will ever halt is printed out by your
program at some finite time.
Now, is there such a program which prints out halting programs in order according to
their length? Give the pseudocode for such a program or prove that none exists.
Answer: We did this in class. The trick is to explore the two-dimensional space of run
program i for j steps, and there are several ways to do this.
If a program existed which prints out halting programs in increase length order, then the
Halting Problem would be decidable by the argument in the previous question: if you want
to know about a program P , wait until it either prints P or a program longer than P . In
the latter case we would know that P will not halt.