Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

What is the expected value of the determinant over the uniform distribution of all possible 1-0 NxN matrices? What does this expected value tend to as the matrix size N approaches infinity?

share|improve this question
    
What ring or field do the entries come from? What distribution on the entries are you assuming? –  Ben Weiss Jan 26 '10 at 4:22
    
@Ben: The field is probably the reals. The OP specified the distribution: i.i.d. uniform on 0 and 1. –  aorq Jan 26 '10 at 4:27
    
Thanks Charles, what's OP? –  Ben Weiss Jan 26 '10 at 4:29
    
does googling for the words in the title not yield anything you can work with? –  Yemon Choi Jan 26 '10 at 4:38
1  
I assume that is what Charles means by "on 0 and 1." Anyway, as for the original question it might be more interesting to compute the variance. –  Qiaochu Yuan Jan 26 '10 at 5:26
show 3 more comments

7 Answers 7

up vote 25 down vote accepted

As everyone above has pointed out, the expected value is $0$.

I expect that the original poster might have wanted to know about how big the determinant is. A good way to approach this is to compute $\sqrt{E((\det A)^2)}$, so there will be no cancellation.

Now, $(\det A)^2$ is the sum over all pairs $v$ and $w$ of permutations in $S_n$ of $$(-1)^{\ell(v) + \ell(w)} (1/2)^{2n-\# \{ i : v(i) = w(i) \}}$$

Group together pairs $(v,w)$ according to $u := w^{-1} v$. We want to compute $$(n!) \sum_{u \in S_n} (-1)^{\ell(u)} (1/2)^{2n-\# (\mbox{Fixed points of }i)}$$

This is $(n!)^2/2^{2n}$ times the coefficient of $x^n$ in $$e^{2x-x^2/2+x^3/3 - x^4/4 + \cdots} = e^x (1+x).$$

So $\sqrt{E((\det A)^2)}$ is $$\sqrt{(n!)^2/2^{2n} \left(1/n! + 1/(n-1)! \right)} = \sqrt{(n+1)!}/ 2^n$$

share|improve this answer
    
Previous version had many sign errors, I think they are fixed now. –  David Speyer Jan 26 '10 at 15:40
5  
There is another simple proof based on Cauchy Binet theorem. Turan published (in Chinese) a formula for the sum of the 4th power. (I think there are simpler proof but I dont know a reference). I am not aware of a formula for higher powers. –  Gil Kalai Jan 26 '10 at 16:15
add comment

If $N \ge 2$, then the expected value is $0$ since interchanging two rows preserves the distribution but negates the determinant.

share|improve this answer
add comment

Is it not zero whenever $n \geq 2$? Let $A$ be a $n \times n$ permutation matrix with determinant $-1$ (which requires $n \geq 2$). Then the uniform distribution of a random $n \times n$ $(0,1)$-matrix $X$ is the same as the distribution of $AX$. The determinant is multiplicative, hence Det$(AX)=$Det$(A)$Det$(X)=-$Det$(X)$. Hence the probability of Det$(X)=x$ is the same as the probability of Det$(X)=-x$.

share|improve this answer
    
The permanent isn't multiplicative. –  Qiaochu Yuan Jan 26 '10 at 4:41
1  
That's somewhat of a Freudian slip (since I typically use permanents rather than determinants). Thanks, I fixed it. –  Douglas S. Stones Jan 26 '10 at 4:54
add comment

Unless I'm missing something, this also follows immediately from linearity and multiplicativity of expectation, treating each entry as independently $0-1$ with probability $1/2$. Every permutation yields the same expected value in the sum, $\pm (1/2)^n$ depending on sign, and the number of even and odd permutations is identical (for $n \ge 2$, as noted above).

It's probably worth mentioning that an old result of Komlos shows that despite this, the probability the determinant is actually 0 is $o(1)$.

share|improve this answer
add comment

For some further results of this nature, see Exercise 5.64 of Enumerative Combinatorics, vol. 2. This exercise deals with the uniform distribution on (0,1)-matrices or $(-1,1)$-matrices, but the arguments can be carried over to other distributions where the matrix entries are i.i.d. The proofs are similar to the argument in David Speyer's comment.

share|improve this answer
add comment

Miodrag Zivkovic has actually done a classification on small orders of 0-1 matrices by rank and absolute determinant value. You may be interested in the tables in his Arxiv paper http://arxiv.org/abs/math.CO/0511636 .

Gerhard "Ask Me About System Design" Paseman, 2010.01.26

share|improve this answer
add comment

It is a little more convenient to work with random (-1,+1) matrices. A little bit of Gaussian elimination shows that the determinant of a random n x n (-1,+1) matrix is $2^{n-1}$ times the determinant of a random n-1 x n-1 (0,1) matrix. (Note, for instance, that Turan's calculation of the second moment ${\bf E} \det(A_n)^2$ is simpler for (-1,+1) matrices than for (0,1) matrices, it's just n!. It is also clearer why the determinant is distributed symmetrically around the origin.)

The log $\log |\det(A_n)|$ of a (-1,+1) matrix is known to asymptotically be $\log \sqrt{n!} + O( \sqrt{n \log n} )$ with probability $1-o(1)$; see this paper of Vu and myself. A more precise result should be that the logarithm is asymptotically normally distributed with mean $\log \sqrt{(n-1)!}$ and variance $2 \log n$. This result was claimed by Girko; the proof is unfortunately not quite complete, but the result is still likely to be true.

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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