7
$\begingroup$

Let a weighted graph $G(V,E)$, where the weights are real (positve and negative). Assume that $G$ has $\mathcal{O}(n\log n)$ edges.

How fast can we compute MAX-CUT on this graph? Can we compute (approximate within $\epsilon$) MAXCUT faster on sparse graphs compared to dense ones?

$\endgroup$
  • 3
    $\begingroup$ actually it's DENSE graphs on which MAX CUT is easy (i.e a PTAS exists). See for example cs.brown.edu/~ws/papers/maxcut.pdf $\endgroup$ – Suresh Venkat Dec 7 '11 at 5:22
  • 3
    $\begingroup$ That's an interesting pointer. However, from a very fast diagonal parse on the paper, it seems that they don't consider weights; the authors say that "Previous algorithms for MaxCut have been extended to weighted instances when the weights define a metric. We conjecture that such instances can also be solved by an extension of our greedy algorithms." $\endgroup$ – Dimitris Dec 7 '11 at 5:38
  • 2
    $\begingroup$ I see. there's older work on certain kinds of dense weighted MAX CUT: www.icsi.berkeley.edu/cgi-bin/pubs/publication.pl?ID=1121 $\endgroup$ – Suresh Venkat Dec 7 '11 at 5:48
7
$\begingroup$

There cannot be significantly better algorithms for graphs with $O(n\log n)$ edges than there are for general graphs. This is because the Benczur-Karger cut sparsifier can take an arbitrary graph, and output a graph with only $O(\frac{n\log n}{\epsilon})$ edges such that the value of any cut in the sparsified graph is within a $(1\pm\epsilon)$ factor of the cut value in the original graph. Therefore, a PTAS for sparse graphs would yield a PTAS for general graphs, by simply first running the cut sparsifier, and then approximating max-cut in the sparse graph. Since it is NP hard to approximate max cut in general graphs to within a factor of 16/17, this is also true for sparse graphs.

Note that there is no PTAS for max-cut even when all edge weights are non-negative. Things are even worse when negative edge weights are allowed, as in your question. See this question for the negative edge weight case: Max-cut with negative weight edges

$\endgroup$
  • $\begingroup$ “Note that there is no PTAS for max-cut” requires the usual condition “unless P=NP.” $\endgroup$ – Tsuyoshi Ito Feb 15 '12 at 16:22
4
$\begingroup$

Unless $\epsilon$ is very large, it seems that no faster algorithms are known than the exact ones. For sparse graphs, the fastest one known seems to be one with running time $O(n+m) \min\{ 2^{m/5}, 2^{(m-n)/2}\}$ for a graph with $n$ vertices and $m$ edges:

Alexander D. Scott and Gregory B. Sorkin. Faster Algorithms for MAX CUT and MAX CSP, with Polynomial Expected Time for Sparse Instances. In Proc. APPROX/RANDOM 2003, LNCS 2764, pp.382--395, Springer, 2003. Link

The title refers to the fact that if the graph is a random graph (with edge probability $c/n$ for some $c \le 1$) then there is a polynomial expected time algorithm.

$\endgroup$
0
$\begingroup$

As far as the approximation portion of your question goes... You can compute MAX-CUT with a 2 approximation in time $O(E)$ by using a simple randomized algorithm. For each edge in the graph flip a coin to determine which set to place it in. Output the larger of these two sets. In the worst case OPT will be E, and our randomized algorithm will be $|E|/2$. This is simple to see since each edge has 50% chance of being output in the cut.

$\endgroup$
  • 3
    $\begingroup$ I guess the questioner would like to know whether there is a better approximation than Goemans–Williamson (.878...) when the graph has O(n log n) edges. $\endgroup$ – MCH Dec 7 '11 at 4:44
  • 1
    $\begingroup$ thanks for the answer. As MCH mentions, I'm mainly interested in an $\epsilon$ approximation. $\endgroup$ – Dimitris Dec 7 '11 at 5:33
  • 2
    $\begingroup$ I notice that the questioner is interested in graphs that have negative edge weights -- note that this simple algorithm does not work with negative edge weights. $\endgroup$ – Aaron Roth Dec 8 '11 at 3:15
  • 1
    $\begingroup$ (And neither does the Goemans-Williamson algorithm...) $\endgroup$ – Aaron Roth Dec 8 '11 at 3:35

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.