CMPSCI 601: Theory of Computation

Posted Friday 16 May 2003

• Question 1 (15): For each statement, indicate (no justification needed) whether it is true, false, or unknown:

• (a,3) The problem CVP is L-reducible to the problem REACH.

UNKNOWN. This is probably false, but would be true if NL=P.

• (b,3) The problem REACH is L-reducible to the problem CVP.

TRUE. REACH is in P, and CVP is P-complete.

• (c,3) Every context-free language is recursively enumerable (r.e.).

TRUE. They are recursive (in P for that matter), and all recursive languages are r.e. as well.

• (d,3) The problem FACTORING is in P.

UNKNOWN. It is in NP but not believed to be NP-complete.

• (e,3) The class NL is contained in the class TC3.

TRUE. We proved NL is contained within AC1, which is certainly inside TC3. Note that the reverse implication is unknown, as for all we know TC0 might contain all of P or even NP.

• Question 2 (15): Prove that every language in FO (every language definable by a first-order formula) is in L.

Note first that we proved in class both that FO is contained in AC0 and that AC0 is contained in NC1 which is contained in L. To answer the exam question fully, however, you would have to justify these results rather than merely quoting them.

The simplest proof is by induction on the number of quantifiers in the first-order formula. If this number is zero, then evaluating the formula on input x means looking up specified bits of the input structure, evaluating numerical predicates on numbers of O(log n) bits, and applying boolean operations to the results. This can be done in O(log n) space, because the primary demand for read/write memory comes from remembering O(1) indices into the structure and these have O(log n) bits each.

Now assume that the input formula has the form ∃ x:\Φ(x) and that by the inductive hypothesis we have a log-space machine that decides Φ(x) for any index number x. Note that x ranges only over numbers with O(log n) bits. Our machine to decide ∃x:Φ(x) works as follows. It uses part of its tape to keep a counter that will range over all possible values of x. For each of these values in turn it uses the hypothesized machine to decide whether Φ(a) is true for this value a. If it ever finds an a for which Φ(a) is true it returns "true". If it finishes all values and all return "false", it returns "false".

The above argument could also be expressed in terms of a recursive algorithm, where the recursion depth is O(1) (the number of quantifiers in the first-order formula) and each recursive call needs only O(log n) space.

• Question 3 (20): Prove that ASPACE(log n) is contained in P. That is, prove that if M is an alternating Turing machine that uses space O(log n), then L(M) is in P.

Let M be a ASPACE(log n) machine and choose c so it uses at most c(log n) space and thus has O(nc) possible configurations. We must define a poly-time deterministic machine D that inputs a string x and determines whether x is in L(M).

D begins by listing all the configurations of M on input x. It then applies a labelling algorithm to these configurations, marking them as "accepting" or "rejecting". Eventually it will mark the start configuration as "accepting" or "rejecting" and this will determine its output.

On its first pass through the configurations D labels all final configurations where no more moves are possible. Then on each successive pass it looks for:

• Existential (White-move) configurations with an accepting successor, which it marks accepting,

• Existential configurations with all their successors marked rejecting, which it marks rejecting,

• Universal (Black-move) configurations with a rejecting successor, which it marks rejecting, and

• Universal configurations with all their successors marked accepting, which it marks accepting.

By the definition of alternating Turing machines, this procedure always marks nodes correctly. To be sure it eventually marks the start configuration, we need to know that M always reaches a final configuration in any sequence of legal moves from any configuration -- this is easy to enforce if M keeps a clock and never makes more than knc moves for a suitable k.

All this can be done in polynomial time because the table of configurations is polynomial length and we only have polynomially many marking rounds.

• Question 4 (10): (true/false with justification)

If A is L-reducible to B (A ≤ B), and B is NP-complete, then A must be NP-complete.

FALSE. A could be the empty language, which is not NP-complete even if P=NP. (No non-trivial language can be reduced to the empty language.)

• Question 5 (10): (true/false with justification)

If B is L-reducible to A (B ≤ A), and B is NP-complete, then A must be NP-complete.

FALSE. Now we know that every language in NP reduces to A, but we have no guarantee that A itself is in NP.

• Question 6 (10): (true/false with justification)

Assuming P is different from NP, there is no poly-time algorithm that can input an undirected graph G and approximate, within 10%, the minimum number of colors needed to color G.

TRUE. If we had such an algorithm, we could use it to solve the NP-complete 3-COLORABILITY problem in polynomial time, which is impossible unless P=NP. The given algorithm would have to return an answer of at most 3.3 on 3-colorable graphs, and of at least 3.6 on graphs that are not 3-colorable (since their minimum number of colors is at least 4).

• Question 7 (10): (true/false with justification)

The Solovay-Strassen randomized algorithm for PRIME (presented in lecture) never indicates that its input number may be prime if it is not prime.

FALSE. The Solovay-Strassen algorithm calculates two functions of its input number m and a random number a. If these functions are the same it says "m is possibly prime", if not it says "m is not prime". If m is composite it may still say "possibly prime" for some a, so the statement is not true. It is true that if Solovay-Strassen says "not prime", we know it is correct.

• Question 8 (10): (true/false with justification) If Φ is any 3-CNF formula (an OR of size-3 ANDs of literals), and x denotes a string defining a setting of all the variables occurring in Φ, then the language {x: Φ(x) is true} is in the class P.

TRUE. The string x is in this language if and only if all of the clauses in Φ are satisfied by x. We can check each clause by looking up the values of each of the three variables involved and seeing whether at least one of the given literals is true. This takes time equal to the number of clauses in Φ times the time to look up the three variables, which is clearly polynomial.

I should have asked about the language {(Φ,x,): Φ(x) is true}, which is also in P by the above argument. (If fact it is in L and even in FO given a suitable format for the input.) As I wrote the problem, there is an even easier way to justify a TRUE answer. Since Φ is fixed for the problem, so is the length of the interesting part of x. (I didn't rule out x defining other variables besides those in Φ.) This interesting part is thus of O(1) size and can be checked against a lookup table of O(1) size.

Crib Sheet:

• The following classes are defined as the languages of the following kinds of Turing machines:
• P: poly-time deterministic TM's
• NP: polytime nondetermistic TM's
• L: space O(log n) deterministic TM's
• NL: space O(log n) nondeterministic TM's
• PSPACE: poly-space deterministic TM's
• If A and B are languages, then A ≤ B (A is L-reducible to B) means that there exists a logspace-computable function f such that for any x, x is in A iff f(x) is in B.
• The following classes are defined as the languages decidable by the following kinds of circuits:
• NCi: circuits with AND, OR, and NOT gates, fan-in two, polynomial size, and O(logi) depth
• ACi: circuits with AND, OR, and NOT gates, unbounded fan-in, polynomial size, and O(logi) depth
• ThCi: circuits with threshold gates, polynomial size, and O(logi) depth
• NC is the union of NCi for all constant i
• If G is a context-free grammar, L(G) is the set of strings w such that w can be generated from G's start symbol according to G's rules. A language is a CFL iff it is equal to L(G) for some context-free grammar G.
• A language is r.e. if it is equal to L(M) for some Turing machine M.
• We refer to some or all of the following specific languages and functions:
• REACH = {(G,s,t): G is a directed graph and there is a path from s to t in G}
• CVP = {(C,x): C is a circuit, x is an input setting for C, and C outputs 1 on input x}
• PRIME = {w: w is the binary representation of a prime number}
• FACTORING is a function: the input is a positive integer n given in binary and the output is a list of n's prime factors