13.07.2015 Views

Parameterized Regular Expressions and Their Languages

Parameterized Regular Expressions and Their Languages

Parameterized Regular Expressions and Their Languages

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

• Between each two consecutive [even] or [odd] strings there is exactly one [read] <strong>and</strong> one[head] in the slots devoted to [action] in form (1).• No other cell can change its context, except for those marked with [read] or [head], <strong>and</strong>• Thecontent thatchangesinthecellsmarkedby[read]<strong>and</strong>[head]respects thetransitionfunction δ of M.Moreover, we also add a set of expressions E 5 , accepting words such that:• The initial configuration of M is encoded as the first step of the computation representedby w.• The last computation ends in a final state of M.It is a tedious, but straightforward task to define the sets E 4 <strong>and</strong> E 5 of expressions.Furthermore, the fact that ⋂ e∈E L ✷(e) is empty if <strong>and</strong> only if M accepts on input ā followsimmediately from the remarks given along the construction. This finishes the proof. ✷The generic reduction used in the proof of Expspace-hardness of Nonemptiness ✷ alsoprovides lower bounds on the minimal sizes of words in languages L ✷ (e) (note that thelanguage L ✸ (e) always contains a word of linear size in |e|).Corollary 1. There exists a polynomial p : N → N <strong>and</strong> a sequence of parameterized regularexpressions {e n } n∈N such that each e n is of size at most p(n), <strong>and</strong> every word in the languageL ✷ (e n ) has size at least 2 2n .Before explaining the proof, we note that the single-exponential bound is easy to see(it was hinted at in the first paragraph of the introduction, <strong>and</strong> which was in fact used inconnection with querying incomplete graph data in [7]). For each n, consider an expressione n = (0|1) ∗ x 1 ...x n (0|1) ∗ . If a word w is in L ✷ (e n ), then w must contain every word in{0,1} n as a subword, which implies that its length must be at least 2 n +(n−1).Proof of Corollary 1: Clearly, for each n ∈ N, it is possible to construct a deterministicTuring machine M n over alphabet Σ = {0,1} that on input 1 n works for exactly 2 2n steps,using 2 n cells. Furthermore, it is possible to specify this machine using polynomial size withrespect to n.Next, using the construction in the reduction of Theorem 1, construct a set of parameterizedregular expressions E (Mn,1 n ) such that the single word w n ∈ ⋂ e∈E (Mn,1 n )L ✷ (e) representsa run (or, more precisely, a sequence of configurations) of M n on input 1 n . Note that eachset E (Mn,1 n ) is of size polynomial with respect to n. Moreover, according to the reductionin the proof of Theorem 1, ⋂ e∈E (Mn,1 n )L ✷ (e) contains a single word, of length greater than2 2n , representing the single run of M n on input 1 n .For each n, we define e n as the expression such that L ✷ (e n ) is empty if <strong>and</strong> only if⋂e∈E (Mn,1 n )L ✷ (e) is empty, constructed as in the proof of Lemma 2, so that e n is of size12

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!