Re:A Quasipolynomial Time Algorithm for Graph Isomorphism: T...
by abetusk on Thursday November 12, 2015 @03:07PM
(#10555742)
I don't know why you would think this. There was strong empirical evidence (in my opinion) for thinking Graph Isomorphism was easy. NAUTY (and SAUCY) were good practical implementations that found solutions efficiently (except for some harder class of graphs, maybe). GI was known to be in NP and co-NP which is usually a red flag that the problem is not NP-Complete. For example, linear programming, integer polynomial factorization and primality testing were all in this NP and co-NP region (also discrete logarithm but that's still open).
Though extremely informative, there's no reason that GI is polynomial (or pseudo polynomial) would give serious reason to believe that P is anywhere near NP.
Re:A Quasipolynomial Time Algorithm for Graph Isomorphism: T...
by lmkg on Thursday November 12, 2015 @04:05PM
(#10556119)
> NAUTY (and SAUCY) were good practical implementations
> that found solutions efficiently (except for some
> harder class of graphs, maybe).
There are good practical implementations of SAT solvers as well (except degenerate cases), even though SAT is the canonical NP-Complete problem.
Re:A Quasipolynomial Time Algorithm for Graph Isomorphism: T...
by abetusk on Thursday November 12, 2015 @05:24PM
(#10556581)
Fair enough, though I think the problem space is substantially different (though I didn't mention it above). It's much easier to make (random) 3-SAT difficult instances whereas for GI you needed highly structured graphs. Apparently Johnson graphs were one of the classes of graphs that caused people problems in GI.
Re:A Quasipolynomial Time Algorithm for Graph Isomorphism: T...
by wolfgke on Friday November 13, 2015 @10:28AM
(#10560310)
There are other NP-complete problems, where really hard instances are a lot more rare.
Re:A Quasipolynomial Time Algorithm for Graph Isomorphism: T...
by erik on Thursday November 12, 2015 @04:17PM
(#10556197)
I didn't think that GI was known to be in co-NP.
Re:A Quasipolynomial Time Algorithm for Graph Isomorphism: T...
by abetusk on Thursday November 12, 2015 @05:21PM
(#10556562)
Scott Aaronson's answer on MO indicates it was known under "suitable popular derandomization assumptions" [1].
Re:A Quasipolynomial Time Algorithm for Graph Isomorphism: T...
by maweki on Thursday November 12, 2015 @03:03PM
(#10555709)
This would mean that NP=Co-NP and that there is a polynomial certificate for every Co-NP problem.
That is what breaks P=NP for me and I just can't wrap my head around how that would be possible.
Re:A Quasipolynomial Time Algorithm for Graph Isomorphism: T...
by abetusk on Thursday November 12, 2015 @03:29PM
(#10555877)
Your comment is a little hard to parse. We could live in a world where there are short proofs that an NP-Complete problem is unsolvable but those short proofs are hard to find (i.e. NP=co-NP but P!=NP) [1]. I think there are other consequences but that's an area I don't know that much about.
[1] http://cstheory.stackexchange.com/questions/8087/consequence...
Re:A Quasipolynomial Time Algorithm for Graph Isomorphism: T...
by maweki on Thursday November 12, 2015 @03:37PM
(#10555924)
Well, you've read my implication-arrow the wrong way. Since I don't believe that NP=co-NP (in any case), I can't believe that P=NP because Co-P=P. So if the collapse would happen (what the OP mentioned as a possibility), one implication would be NP=co-NP. I don't believe that so I can't believe in collapse.
Re:A Quasipolynomial Time Algorithm for Graph Isomorphism: T...
by Ar-Curunir on Thursday November 12, 2015 @07:08PM
(#10557137)
Lol. The polynomial hierarchy would collapse if GI was NP-Complete, but all efforts to prove that have failed. If anything, the distance from P to NP is growing.
Re:A Quasipolynomial Time Algorithm for Graph Isomorphism: T...
by lovboat on Friday November 13, 2015 @01:36PM
(#10561489)
You can shrink an interval [a,b] by two methods [a,b'] with b'<b or [a',b] with a'> a, what you see is [a,b'] (NP-complete seem easier or lower in the difficulty scale), what I see is [a',b] that is P cover a lot more ground, I think P will cover all NP-complete problems and then the interval becomes a unique point.
If [P,NP-c] = [0,1] I see GI as 1/4, if GI is proven to be in P then I should say [P,NP-c] is now [1/4,1].
In order to reduce more that interval one should find a problem that people think has a probability of 0.5 to be in P and 0.5 to be in NP, if it is proven that problem is in P then the interval get reduced a lot more. Just trying to explain my perception of the intuitive complexity of algorithms.
Personal Opinion: Perhaps the Complexity Hierarchy is going to collapse since intuition is not so clear and perhaps the distance from P to NP is shrinking.