I'm trying to follow a proof of the Marchenko-Pastur theorem. In particular I'm trying to show that the kth moment of the Marchenko-Pastur distribution: $$ a_k = \int_{(1-\sqrt{\gamma})^2}^{(1+\sqrt{\gamma})^2}x^k \frac{1}{2\pi\gamma x}\sqrt{((1+\sqrt{\gamma})^2-x)(x-(1-\sqrt{\gamma})^2)}dx $$ is equal to $$ \sum_{r=0}^{k-1} \frac{\gamma^r}{r+1}\binom{k}{r}\binom{k-1}{r} $$

Where I've got to so far:

I understand how we can use a substitution and rewrite the kth moment as: $$ \frac{1}{2\pi}\int_{-2}^2(\sqrt{\gamma}y+1+\gamma)^{k-1}\sqrt{4-y^2}dy $$ Then the notes I'm working through says you should expand the k-1 power, use the formula for the moments of the semicircle law (i.e. the Catalan number formula) and use Vandermonde's identity. Expanding the k-1 power we can write this as: \begin{align} \frac{1}{2\pi}\int_{-2}^2\sum_{r=0}^{k-1} \binom{k-1}{r} (\sqrt{\gamma} y)^r (1+\gamma)^{k-1-r}\sqrt{4-y^2}dy \\ =\sum_{r=0}^{k-1} \binom{k-1}{r} \gamma^{r/2} (1+\gamma)^{k-1-r}\frac{1}{2\pi}\int_{-2}^2 y^r\sqrt{4-y^2}dy \\ =\sum_{r=0, r\text{ even}}^{k-1} \binom{k-1}{r} \gamma^{r/2} (1+\gamma)^{k-1-r}C_{r/2} \end{align} where $C_i$ is the ith Catalan Number...

But I'm not sure how we can get from here to the result - or where we're going to use Vandermonde's identity? Any help is appreciated.


1 Answer 1


$$ \frac{1}{2\pi}\int_{-2}^2(\sqrt{\gamma}y+1+\gamma)^{k-1}\sqrt{4-y^2}dy \\ = \sum_{i+j\le k-1} {k-1 \choose i;j;k-1-i-j} \gamma^{i/2+j}\cdot \frac{1}{2\pi} \int_{-2}^2 y^{i}\sqrt{4-y^2} dy \\ = \sum_{2l+j\le k-1} {k-1 \choose 2l;j;k-1-2l-j} \gamma^{l+j}\cdot C_{l} \\= \sum_{2l+j\le k-1} \frac{(k-1)!}{(2l)!j! (k-1-2l-j)!}\cdot\frac {(2l)!}{l!(l+1)!}\cdot \gamma^{l+j} = \big| l+j = r\big|\\ = \sum_{l+r\le k-1}\frac{(k-1)!}{l!(r-l)! (l+1)!(k-1-l-r)!} \cdot \gamma^{r}\\ = \sum_{r = 0}^{k-1}\frac{(k-1)!}{r!(k-r)!}\cdot\gamma^r \sum_{l=0}^{k-1-r} {r\choose l} \cdot {k-r \choose k-1-r-l} \\ = \sum_{r = 0}^{k-1} \frac{(k-1)!}{r!(k-r)!}\cdot\gamma^r \cdot {k \choose k-r-1} = \sum_{r = 0}^{k-1} \frac{1}{r+1}{k \choose r} \cdot {k-1 \choose r}\cdot\gamma^r. $$


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.