19.07.2014 Views

Contents - Student subdomain for University of Bath

Contents - Student subdomain for University of Bath

Contents - Student subdomain for University of Bath

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

110 CHAPTER 3. POLYNOMIAL EQUATIONS<br />

3.5.6 Complexity<br />

Let us suppose that there are s polynomials involved in the input <strong>for</strong>mula (3.54),<br />

<strong>of</strong> maximal degree d. ( Then such a cylindrical algebraic decomposition can be<br />

computed in time O (sd) 2O(k))) .<br />

There are examples [BD07, DH88], which shows that this behaviour is bestpossible,<br />

indeed the projection onto R 1 might have a number <strong>of</strong> components<br />

doubly-exponential in k.<br />

While this behaviour is intrinsic to cylindrical algebraic decomposition, it is<br />

not necessarily intrinsic to quantifier elimination as such. If a is the number <strong>of</strong><br />

alternations <strong>of</strong> quantifiers (Notation 17) in the problem (so a < k), then there<br />

are algorithms [Bas99, <strong>for</strong> example] whose ( behaviour is singly-exponential in k<br />

but doubly-exponential in a; typically O (sd) O(k2 )2 O(a))) .<br />

One particular special case is that <strong>of</strong> no alternations. Hence, using the<br />

fact that ∃x (P (x) ∨ Q(x)) is equivalent to (∃xP (x)) ∨ (∃xQ(x)), an existential<br />

problem is equivalent to a set 27 <strong>of</strong> problems <strong>of</strong> the <strong>for</strong>m<br />

⎛<br />

⎞ ⎛<br />

⎞<br />

∃x ⎝ ∧<br />

f i (x) ≥ 0⎠ ∧ ⎝ ∧<br />

( )<br />

∧<br />

g i (x) = 0⎠ ∧ h i (x) ≠ 0 . (3.65)<br />

f i∈F<br />

g i∈G<br />

h i∈H<br />

This is generally referred to as the existential theory <strong>of</strong> the reals. Since the<br />

truth <strong>of</strong> a universal problem is equivalent to the falsity <strong>of</strong> an existential problem<br />

(∀xP (x) ⇔ ¬∃x¬P (x)), this is all we need to consider.<br />

Given a problem (3.65), cylindrical algebaric decomposition will yield such<br />

an x, if one exists, and failure to yield one is a pro<strong>of</strong> that no such x exists.<br />

However, this is a somewhat unsatisfactory state <strong>of</strong> affairs in practice, since,<br />

computationally, we are relying not just on the correctness <strong>of</strong> the theory <strong>of</strong><br />

cylindrical algebraic decomposition, but also on the absence <strong>of</strong> bugs in the implementation.<br />

An alternative is provided by the Positivstellensatz approach [Ste74].<br />

Theorem 26 ([PQR09, Theorem 3]) The set <strong>of</strong> solutions to (3.65) is empty<br />

if, and only if, there are:<br />

s ∈ con(F ) where con(F ), the cone <strong>of</strong> F , is the smallest set generated by F and<br />

the set <strong>of</strong> squares <strong>of</strong> all elements <strong>of</strong> R[x] wich is closed under multiplication<br />

and addition;<br />

g ∈ (G) the ideal generated by G;<br />

m ∈ mon(H) where mon(H), the (multiplicative) monoid <strong>of</strong> H is the set <strong>of</strong> all<br />

products (inding 1 = the empty product) <strong>of</strong> elements <strong>of</strong> H;<br />

such that s + g + m 2 = 0. Furthermore, there is an algorithm to find such s, g<br />

and m (if they exist) in Q[x] provided F , G and H ⊂ Q[x].<br />

27 There may be singly-exponential blow-up here as we convert into disjunctive normal <strong>for</strong>m,<br />

but this is small compared to the other exponential issues in play!

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

Saved successfully!

Ooh no, something went wrong!