I want to prove that the matrix $$\begin{pmatrix} 1 &\cfrac{1}{2} &\cfrac{1}{3} &\cdots &\cfrac{1}{n} \\ \cfrac{1}{2} &\cfrac{1}{3} &\cfrac{1}{4} &\cdots &\cfrac{1}{n+1} \\ &\vdots &&\ddots &\vdots \\ \cfrac{1}{n} &\cfrac{1}{n+1} &\cfrac{1}{n+2} &\cdots &\cfrac{1}{2n-1} \end{pmatrix}$$ is positive definite. Using mathematical induction, I only need to show that its determinant is positive. But I can't find the way out.

  • $\begingroup$ This is a Hilbert Matrix, and in the link the determinant is computed in the closed form; but I think this is not needed in its fullest strength, in the mere order to show that the determinant is positive. $\endgroup$ – awllower May 6 '14 at 6:35
  • $\begingroup$ oeis.org/A005249 tabulates (the inverse of) the determinant of this matrix and gives a formula, which maybe you can prove by induction. It also gives some references that might be helpful. $\endgroup$ – Gerry Myerson May 6 '14 at 10:49
  • $\begingroup$ See also: Prove the positive definiteness of Hilbert matrix. $\endgroup$ – Martin Sleziak Sep 11 '18 at 17:40

Hint. Call your matrix $H$. Then $H=\left(\int_0^1 x^{i+j}dx\right)_{i,j=1,2,\ldots,n}=\int_0^1 v(x)v(x)^Td x$, where $v(x)=(1,x,x^2,\ldots,x^{n-1})^T$. Can you see that $H$ is positive semi-definite? Now, if $u^THu=0$ for some vector $u$, then $\int_0^1 u^Tv(x)v(x)^Tu dx=0$. Can you infer that $u=0$ (think about Vandermonde matrix)?

Remark. From a wider perspective, $H$ is an instance of matrices of the form $A=\left(\frac1{1+x_i+x_j}\right)_{i,j=1,2,\ldots,n}$ where $x_i\ge0$ for all $i$ (in your case, $x_i=i-1$), which is known (q313249) to be positive semidefinite. The same trick in the above hint can be used to show that $A$ is positive definite iff all $x_i$s are distinct.

Edit. Alternatively, $H$ is a Hilbert matrix. Every leading principal submatrix of $H$ is again a Hilbert matrix. So, by Sylvester's criterion, it suffices to prove that every Hilbert matrix has a positive determinant. As Gerry Myerson has pointed in his comment, mathematical induction may be useful in this case.


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.