[Date Index] [Thread Index] [Author Index]

Re: random derangement

• To: mathgroup at smc.vnet.net
• Subject: [mg37435] Re: [mg37416] random derangement
• From: Daniel Lichtblau <danl at wolfram.com>
• Date: Tue, 29 Oct 2002 00:09:36 -0500 (EST)
• References: <200210271133.GAA12182@smc.vnet.net>
• Sender: owner-wri-mathgroup at wolfram.com

"DIAMOND Mark R." wrote:
> 
> This is not quite aposite to either NG, but there appear to be none better
> ... My apologies.
> 
> I am searching for an algorithm for producing a random derangement of, for
> instance, the integers 1 to approx 10000.
> 
> I thought Skiena's site might have such an algorithm, but I could not locate
> one. ... Producing all derangements and choosing one at random is marginally
> beyond the capacity of my machine :-)
> 
> Cheers,
> 
> Mark R. Diamond

One thing that will work efficiently is a modification of a basic random
shuffle. The basic shuffle can be found at

http://forums.wolfram.com/mathgroup/archive/2001/Apr/msg00263.html

The modification is that at step j we insist on moving something between
position j+1 (rather than j) and the end into position j.

derangement = Compile[{{n,_Integer}}, Module[
  {deck=Range[n], newj},
  Do[
    newj = Random[Integer, {j+1,n}];
    deck[[{j,newj}]] = deck[[{newj,j}]],
    {j,n-1}];
  deck
  ]]

In[4]:= Timing[dd = derangement[10^6];]
Out[4]= {5.25 Second, Null}

Check that this is indeed a derangement:

In[5]:= Select[Transpose[{dd,Range[10^6]}], #[[1]]==#[[2]]&]
Out[5]= {}

(Or you can use MapIndexed for this test):
In[12]:= Apply[Or, MapIndexed[#1==#2[[1]]&, dd]]
Out[12]= False

I think this will give random derangements with equal probabilities
though I don't have a proof of that off the top of my head.


Daniel Lichtblau
Wolfram Research

• Follow-Ups:
  • Re: Re: random derangement
    • From: Daniel Lichtblau <danl@wolfram.com>

• References:
  • random derangement
    • From: "DIAMOND Mark R." <dot@dot.dot>