MathGroup Archive 2006

[Date Index] [Thread Index] [Author Index]

Search the Archive

Re: A problem in mathematical logic

  • To: mathgroup at smc.vnet.net
  • Subject: [mg72263] Re: [mg72242] A problem in mathematical logic
  • From: Andrzej Kozlowski <akoz at mimuw.edu.pl>
  • Date: Sun, 17 Dec 2006 06:20:07 -0500 (EST)
  • References: <200612161017.FAA12165@smc.vnet.net> <70451048-2722-481D-AF0F-0899F375E840@mimuw.edu.pl>

On 16 Dec 2006, at 21:21, Andrzej Kozlowski wrote:

>
> On 16 Dec 2006, at 19:17, Bonny Banerjee wrote:
>
>> I am working on a problem that requires eliminating quantifiers  
>> from a
>> logical combination of polynomial equations and inequatlities over  
>> the
>> domain of real numbers, i.e. given a quantified logical expression  
>> E, I want
>> a quantifier-free expression F, such that E and F are equivalent.  
>> It has
>> been shown that the computational complexity of quantifier  
>> elimination is
>> inherently doubly exponential in the number of variables.
>
> Not quite true. The Basu, Pollack, Roy algorithm is double  
> exponential in the number of blocks of variables, where the blocks  
> are delimited by alternations of the existential and universal  
> quantifiers. This means you need to have at least two blocks and at  
> lest on of them must have more than one variable, for this to make  
> a difference.
>
>
>> I want to know
>> whether quantifier elimination can be done in lesser time if we  
>> take the
>> help of examples.
>>
>> Suppose I have a quantified logical expression E1 and its equivalent
>> quantifier-free expression F1. Now I am given another quantified  
>> logical
>> expression E2 and the task is to compute its equivalent quantifier- 
>> free
>> expression, say F2. Instead of spending a lot of time computing F2  
>> from E2
>> using the quatifier elimination algorithm, I want to know whether  
>> there
>> exists a mapping between E2 and E1, so that F2 can be computed  
>> from F1 by
>> reverse-mapping. This idea will be beneficial only if there exists  
>> such a
>> mapping that can be computed in time less than doubly exponential.
>>
>> Example of a mapping: If I can show that E1 and E2 are equivalent,  
>> then F1
>> and F2 have to be equivalent. So "equivalence" is an example of  
>> such a
>> mapping. Unfortunately, the general problem of equivalence  
>> checking is
>> NP-hard.
>>
>> I suspect, there might exist a mapping weaker than equivalence (and
>> computable in polynomial time) that will suffice for my purposes.  
>> Please let
>> me know if any of you are already aware of any such mapping. Any  
>> suggestion
>> regarding which book/paper to look at would also help.
>>
>
> Each of yoru two expressions E1 and E2 is equivalent to specifying  
> some semi-algebraic set. I have not given much thought to your  
> question, but at this time the only way I can imagine of  
> constructing (in the general case) your mapping is by using  
> Cylindrical Decomposition, which is indeed inherently double  
> exponential (although there are approximate versions which are only  
> single exponential - e.g.  
> Experimental`GenericCylindricalDecomposition` ). The improvements  
> in complexity of a number of algorithms (such as computing the  
> connected components of a semi-algebraic set,  quantifier  
> elimination etc.) achieved by Basu, Pollack and Roy is due to their  
> being able to avoid the need to use Cylindrical Decomposition. The  
> relevant book is
> "Algorithms in Real Algebraic Geometry" by Basu, Pollack and Roy  
> (Springer, Algorithms and Computation in mathematics, Volume 10).  
> (However, this book has 580 pages and the relevant algorithms are  
> all near the end).
>
> Andrzej Kozlowski
>
>
>

I forgot to add that the decision problem for the existential theory  
of the reals (in other words, quantifier elimination over the reals  
where we only have to deal with the existential quantifier;  this is  
equivalent to deciding whether or not a given semi-algebraic set is  
empty) can be solved with singly exponential complexity in the number  
of variables.

It seems to me that singly exponential complexity is the best one can  
expect of general algorithms of this kind. I do not know of any  
algorithms with polynomial complexity in this whole field. ( Maybe  
one can do better using numeric-symbolic methods, about which I know  
very little,  but I am pretty certain that the complexity is still  
singly exponential).

Of course a particular problem with some special structure may be  
solvable much more efficiently using special tricks rather than  
general algorithms, but that is already mathematics rather than  
"computer science" ;-)

Andrzej Kozlowski


  • Prev by Date: system of nonlinear equations in mathematica?
  • Next by Date: decrementing problem
  • Previous by thread: Re: A problem in mathematical logic
  • Next by thread: Re: Re: A problem in mathematical logic