You are on page 1of 25

Descriptive Complexity Theory

Walid Gomaa
Advisors: Dr. David Kueker and Dr. Bill Gasarch
INRIA Nancy Grand-Est Research Centre
CARTE team
October 1st, 2008
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
I. Logical Formalism
Logic = Syntax + Semantics
Syntax: meaningless symbols terms formulas
sentences
Semantics: meaning/interpretation by mathematical
structures (graphs, groups, elds, etc)
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
First-Order Logic
Syntax:
1
Symbols:
Logical: variables + , , , , + , + =
Non-logical: constant symbols + functions symbols +
relation symbols
2
Terms: constructed inductively from non-logical symbols
and variables
3
Formulas: constructed inductively using propositional
connectives and quantiers
4
Sentences: formulas with no free variables
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
First-Order Logic Contd
Semantics: A mathematical structure (model) that consists of
Universe
Interpretation of the constant symbols
Interpretation of the function symbols
Interpretation of the relation symbols
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
Example
Let R be a binary relation symbol. Consider
xyRxy
/ = (N, <) =/ [=
B = (Z, <) = B ,[=
( = (V, E) = B [= iff there is u V that has edges with
all the vertices in (
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
Descriptive Complexity
Computational complexity theory classies problems
based on time,space, or hardware resources
Descriptive complexity classies problems based on how
hard it is to express the problem in some logical formalism
Example: 3-colorability ( NPC)
RBG( )
x ((R(x) B(x) G(x)) )
xy ((R(x) R(y) E(x, y)) )
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
Descriptive Complexity Contd
Virtually all traditional complexity classes can be captured
by logical formalisms
Examples
NP
1
1
and hence coNP
1
1
[Fagin 74]
PH SO [Stockmeyer 77]
P (LFP+ <) [Immerman 86], hence
P NP (LFP+ <)
1
1
PSpace (PFP+ <)
It has been shown using transitive closure logic that
NL coNL
TC
0
FOC(<, +, ) [Bar and Imm 90]
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
Logical Resources
3-colorability revisited
RBG( )
x ((R(x) B(x) G(x)) )
xy ((R(x) R(y) E(x, y)) )
1
How many object variables? (2FO and 3SO)
2
Types/semantics of quantiers (FO and monadic SO)
3
Quantier rank (3SO and 2FO)
4
Alternation of quantiers (0 FO and 0 SO)
5
Length of sentences: nite/innite (nite)
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
Ehrenfeucht-Frass Games
An essential tool in descriptive complexity
Two players: Spoiler o and Duplicator T
Played over two structures / and B of the same kind
Finite number of rounds r
Roughly, the goal of o is to show that / and B can be
distinguished in at most r steps, whereas Ts goal is to
show that this can not be done
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
The EF Theorem
Theorem
The following are equivalent:
1
/
r
B
2
EF
r
(/, B) T
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
II. Expressibility in
1
1

1
1
consists exactly of
X
1
X
n

where X
i
is SO variable and is FO sentence

1
1
: X
1
X
n

1
1
(p, r ) : has at most p FO variables and quantier
depth at most r
Mon
1
1
/
1
1
: the X

i
s are sets
Bin
1
1
/
1
1
: the X

i
s are either sets or binary relations
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
Expressibility in
1
1
Contd
Inspired by Fagins result NP
1
1
, we have developed a
partial framework to investigate expressibility inside
1
1
Ultimate goal is to have a ner look into NP
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
Main Results
1
DIV
k
, DIV
k
are neither in mon
1
1
nor in mon
1
1
2
DIV
k
/ bin
1
1
(2, 2) and DIV
k
/ bin
1
1
(2, 3)
3
DIV
k
bin
1
1
(3, 3).
4
Let = bin
1
1
(3, 3): has l binary variables,
then DIV
k
for every k (4
l
1).
5
Furthermore, DIV
k
for only nitely-many k.
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
Main Results Contd
6
Hence DIV
k
creates a proper hierarchy inside the logic
bin
1
1
(3, 3).
7
An immediate consequence of the above is that
mon
1
1
bin
1
1
.
8
DIV
k
, bin
1
1
when the sizes of the interpretations of the
binary variables are bounded from above by some linear
function of the size of the universe.
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
Example: DIV
2
DIV
2
, FO and DIV
2
, mon
1
1
. However, It can be expressed
by the following bin
1
1
sentence.
R(
1
(R)
2
(R)
3
(R))
where

1
(R) xR(x, x)

2
(R) xy (R(x, y) R(y, x))

3
(R) xy (R(x, y) z (R(x, z) z = y))
denes the class of simple undirected graphs with
isolated edges (1-regular graphs). The number of vertices
in these graphs must be even
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
Open Problems
1
Finding tight lower/upper bounds for the DIV
k
hierarchy in
bin
1
1
(3, 3).
2
Get a lower bound on the expressibility of DIV
k
in
1
1
.
3
Study natural extensions of bin
1
1
(3, 3) inside
1
1
within the
framework developed above. The main goal is to get
separation results among n-ary
1
1
sublogics using
expressibiltiy of number-theoretic properties such as
primeness.
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
Open Problems Contd
4
Investigate expressibility inside
1
1
to get separation results
among n-ary
1
1
sublogics
5
Get separation results between n-ary
1
1
logics and
n-ary
1
1
logics
6
Investigate the whole of SOL and consequently the whole
of the polynomial hierarchy
7
Complexity-theoretic consequences of the above
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
III. Descriptive Comp. of Abelian Finite Groups
We want to answer the following two questions.
1
Given two AFGs G and G

, want to nd a logical sentence


that distinguishes between them?
2
How complex can this sentence be (lower/upper bounds)?
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
Abelian Finite Groups Contd
We answer these questions in FOL, hence commit to
nite length formulas
rst-order quantiers
+ as a ternary relation
Analysis done in 3 stages
1
groups of residue classes with prime orders
2
groups of residue classes with arbitrary orders (include all
cyclic groups)
3
nite abelian groups
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
Main Results
Let G
1
and G
2
be a pair of non-isomorphic nite abelian
groups, then there exists a number m that divides the order of
one of the two groups such that
1
There exists a rst-order sentence that distinguishes G
1
and G
2
such that is existential, has quantier depth
O(log m), and has at most 5 variables.
2
If is a sentence that distinguishes G
1
and G
2
, then
must have quantier depth (logm).
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
Main Results Contd
Consequences of the previous results:
1
Bounds on the rst-order distinguishability of dihedral
groups
2
FO undenability (over the class of nite groups) of
cyclicity, simplicity, nilpotency, the closure and normal
closure of a single element
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
Open Problems
1
Can the concrete lower/upper on the quantier rank of a
distinguishing sentence (for Z
n
s and D
n
s) be improved?
We have already shown that the lower bound is optimal
and we believe that the upper bound is also optimal.
2
Can the upper bound of 5 on the number of object
variables in a distinguishing sentence be improved?
3
Investigate the complexity-theoretic consequences of
these expressibility results.
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
Open Problems Contd
4
Generalize the results to all nite groups (we have already
started here with dihedral groups).
5
Study the rst-order expressibility of innite groups.
6
Use other formalisms such as xed-point logic, innitary
logics, second-order logic, allowing for generalized
quantiers, etc.
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
IV. Abstract Elementary Classes
A generalization of FO model theory
K = (K, _
K
) is a class of models K of the same
vocabulary that are partially ordered
K satises a nite set of purely semantical axioms like
closure under isomorphism and closure under unions of
chains
Example: the class of all ordinals is an AEC: it is neither
denable in FO-logic nor in L
,
, the class of dense linear
orderings, any class of structures of a FO theory
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory
Classication Theory for AECs
Given K want to answer questions about K

/

=:
1
Is K

,= ?
2
Does K

,= imply K

+ ,= ?
3
If K is -categorical, does this imply K is
+
-categorical?
4
If K is
+
-categorical, does this imply K is -categorical?
5
What are the possible functions [K

[?
6
Under what conditions on K is it possible to nd a nice
dependence relation among the subsets of / K ?
Walid Gomaa Advisors: Dr. David Kueker and Dr. Bill Gasarch Descriptive Complexity Theory

You might also like