8

I'd like an efficient method for calculating the minimum number of transpositions needed to sort a list. I don't need to know what the transpositions actually are.

For example, the list [1, 1, 2, 0] requires 2 transpositions:

[1, 1, 2, 0] // Start
[1, 1, 0, 2] // Swap index 2 and 3
[0, 1, 1, 2] // Swap index 0 and 2

The list [0, 1, 0, 0] requires 1 transposition:

[0, 1, 0, 0] // Start
[0, 0, 0, 1] // Swap index 1 and 3

The list [2, 2, 2, 2] requires 0 transpositions because it is already sorted.

Some meta information: 1) The list may have repeated elements, so simply using the Cayley distance between the sort and the identity permutation will not work. 2) This Math Overflow question is related.

Community
  • 1
emchristiansen
  • 183
  • 4
  • 1
    Duplicate of http://cstheory.stackexchange.com/questions/4096/minimum-number-of-transpositions-to-sort-a-list – derekhh Oct 02 '12 at 07:19
  • 1
    I think that the answer(s) posted here should be merged to question 4096. – Tsuyoshi Ito Oct 02 '12 at 12:16
  • 1
    @derekhh, this is not a duplicate (or at least the interpretation of the post is different in the other question). I linked to question 4096 in the original post. – emchristiansen Oct 02 '12 at 15:59
  • 3
    Question 4096 does not state that given elements are distinct. Unfortunately, all the answers posted there silently assumed this, and this oversight can be corrected by merging the answer(s) here to there. – Tsuyoshi Ito Oct 02 '12 at 16:21
  • 1
    @emchristiansen Oops, sorry, I haven't noticed that... – derekhh Oct 03 '12 at 06:28

1 Answers1

7

Unfortunately, the problem is NP-hard in general, according to Amir, Hartman, Kapah, Levy and Porat (see http://epubs.siam.org/doi/abs/10.1137/080712969), even if each symbol appears at most three times. I didn't find a mention of a constant factor approximation algorithm, or of something reasonably fast to tackle the problem. Are there any additional restrictions you can think of in your case?

EDIT: the authors also give a 3/2-approximation algorithm, but I don't know if it's good enough for your purposes.

Anthony Labarre
  • 3,264
  • 1
  • 25
  • 33
  • Thanks for the reference. I'm dealing with pixel values, so lists of integers in [0, 255]. A common list length is 256 elements. – emchristiansen Oct 02 '12 at 15:56
  • Can you prove, that when you have a solution (minimal sorting sequence of transposition), you always swap large number on the left with a smaller number on the right? – Ivan Kuckir Jan 01 '16 at 00:44