home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!idacrd!desj@ccr-p.ida.org
- From: desj@ccr-p.ida.org (David desJardins)
- Newsgroups: sci.math
- Subject: Re: The Mathematics of Secret Santa
- Message-ID: <1818@idacrd.UUCP>
- Date: 18 Dec 92 23:21:05 GMT
- References: <1992Dec17.220416.18649@husc15.harvard.edu>
- Sender: desj@idacrd.UUCP
- Organization: IDA Center for Communications Research, Princeton
- Lines: 97
-
- Eric Blom <blom@husc15.harvard.edu> writes:
- >What is the probability that in a Secret Santa arrangement of n people, there
- >is at least one pair of participants with each other as Secret Santas?
-
- >No participant may have himself.
-
- As usual, the problem is most easily solved by inclusion-exclusion.
-
- Let D(m) be the number of derangements of m items (permutations with no
- fixed points). As is well known (also by inclusion-exclusion) this is
-
- D(m) = \sum_{k=0}^m (-1)^k m!/k!
-
- Now fix n. Let T be a collection of t distinct 2-sets from {1,...,n}.
- Let E(T) be the number of derangements of {1,...,n} which contain each
- 2-set in S as a 2-cycle (and possibly also some other 2-cycles). Then
- clearly
-
- E(T) = D (n - 2t)
-
- because 2t values of the permutation are fixed and map to themselves as
- a set, and the remainder are free to be any derangement of the remaining
- points.
-
- Now let S be a similar collection of s distinct 2-sets from {1,...,n}.
- Let F(S) be the number of derangements of {1,...,n} which contain each
- 2-set in S as a 2-cycle, and no other 2-cycles. Clearly
-
- E(T) = \sum_{S\supseteq T} F(S)
-
- because if we fix T to be the set of required 2-cycles, then the set of
- actual 2-cycles S must contain T.
-
- By the inclusion-exclusion formula:
-
- F(S) = \sum_{T\supseteq S} (-1)^{|T|-|S|} E(T).
-
- Now, the probability you asked for is 1-F(0)/D(n), where 0 is the
- empty set. The number of sets T of cardinality t which contain 0 is
-
- n! / (2^t t! (n-2t)!)
-
- Thus the formula for F(0) becomes
-
- F(0) = \sum_{t=0}^{n/2} (-1)^t n! / (2^t t! (n-2t)!) D(n-2t)
- = \sum_{t=0}^{n/2} \sum_{k=0}^{n-2t} (-1)^{t+k} n! / (2^t t! k!)
- = \sum_{k=0}^n \sum_{t=0}^{(n-k)/2} (-1)^{t+k} n! / (2^t t! k!)
-
- (You can choose whichever of the last two lines you find simpler as the
- "answer".)
-
- A quick table:
-
- n D(n) F(0) Prob
- ---- ---- ---- ------
- 2 1 0 1.0000
- 3 2 2 0.0000
- 4 9 6 0.3333
- 5 44 24 0.4545
- 6 265 160 0.3962
- 7 1854 1140 0.3851
- 8 14833 8988 0.3940
- 9 133496 80864 0.3943
- 10 1334961 809856 0.3933
-
- > For the record, here are the first few values of P(n)*!n, which should be the
- > numerator of the fraction:
- >
- > !2*P(2) = 1
- > !3*P(3) = 0
- > !4*P(4) = 3
- > !5*P(5) = 20
- > !6*P(6) = 105
- > !7*P(7) = 504
-
- I think Eric's table was right up to n=6, but wrong for n=7.
-
- Finally it is interesting to note that, from the formula, it is easy to
- compute the limit of the probability as n goes to infinity. It is well
- known (and obvious from the formula) that
-
- \lim_{n\to\infty} D(n)/n! = exp(-1)
-
- But, by placing suitable bounds on the rate of convergence (which
- details I omit here), we can show that
-
- \lim_{n\to\infty} F(0)/n!
- = \lim_{n\to\infty} \sum_{k=0}^n \sum_{t=0}^{(n-k)/2} (-1)^{t+k} / (2^t t! k!)
- = \sum_{k=0}^\infty \sum_{t=0}^infty (-1)^{t+k} / (2^t t! k!)
- = [\sum_{k=0}^\infty (-1)^k / k!] * [\sum_{t=0}^\infty (-1)^t / (2^t t!)]
- = exp(-1) exp(-1/2)
-
- So
-
- \lim_{n\to\infty} (1 - F(0)/D(n)) = 1 - exp(-1/2) = 0.393469340287
-
- David desJardins
-