Can anyone show how to express complexity class P using first-order logic with LFP? (descriptive complexity)
-
$\begingroup$ It can't be done using first order logic alone, you need a fixed point operator of some sort (least fixed point is the typical one). The class of problems expressible in first order logic $FO$ is equivalent to $AC^{0}$. $\endgroup$– Luke MathiesonCommented Mar 3, 2013 at 4:04
-
$\begingroup$ corrected. forgot to write LFP. $\endgroup$– logicaCommented Mar 3, 2013 at 4:05
1 Answer
In Neil Immerman's Descriptive Complexity the proof that P = FO + LFP uses alternating graphs: an alternating graph is a directed graph $G= \{ V, E, A, s, t\}$ where $A$ is the set of nodes marked as universal ($V \setminus A$ are existential). The accessibility relation is defined in this way:
- $P_a^G(x,x)$
- if $x \in V \setminus A$ $\land P_a^G(z,y)$ holds for some edge $(x,z)$ then $P_a^G(x,y)$
- if $x \in A$ $\land \; outdeg(x)>0 \land P_a^G(z,y)$ holds for all edges $(x,z)$ then $P_a^G(x,y)$
The reachability problem $REACH_a$ is defined as $\{G | P_a^G(s,t)\}$ and is complete for P via first-order reductions.
Using LFP one can give a FO + LFP definition of $REACH_a$ (i.e. $REACH_a \subseteq FO+LFP$):
$$\varphi_{as}(P,x,y)\equiv x = y \lor [\exists z(E(x,z)\land P(z,y)) \land (A(x)\rightarrow \forall z(E(x,z)\rightarrow P(z,y)))]$$
and $REACH_a = (LFP_{P_{xy}\varphi_{ap}})(s,t)$
We have $P \subseteq REACH_a \subseteq FO+LFP$, the inverse $FO+LFP \subseteq P$ follows from the fact that a FO query can be evaluated in $L$ and at most $n^k$ evaluations of $\varphi(R,x_1,...,x_k)$ are needed for $(LFP_{R \varphi})$ ($R$ is a k-ary relation).
For further details you can see the proof on Immerman's book.