2
$\begingroup$

How to prove the following inequality: Let $X$ and $Y$ be $n\times m$ matrices with real entries. Prove that \begin{equation} \det\left(XY^T\right)^2 \leq \det\left(XX^T\right)\det\left(YY^T\right) . \end{equation}

$\endgroup$

3 Answers 3

10
$\begingroup$

If you replace determinants by traces, then this inequality is just Cauchy-Schwarz for the inner product $(X,Y)=\mathop{\mathrm{tr}}(XY^T)$ on the space of matrices. Now, we just have to recall that determinants are traces: for a $n\times n$-matrix $A$ we have $\det(A)=\mathop{\mathrm{tr}}(\Lambda^n(A))$, where $\Lambda^n(A)$ is the $n$-th exterior power of the linear operator $A$.

(Now that I took time to look at answers above I believe this is more or less precisely Wadim Zudilin's proof made a tiny bit more invariant.)

$\endgroup$
2
  • 1
    $\begingroup$ The answers are not above any more! :-) $\endgroup$ May 25, 2010 at 11:19
  • $\begingroup$ True - I keep forgetting about the built-in nonlinearity :-) $\endgroup$ May 25, 2010 at 12:46
5
$\begingroup$

If $n>m$ all the determinants are zero.

The Cauchy-Binet formula allows you to write the determinant of $XY^t$ as the sum $$ \sum_{I\subset\lbrace1,2,\dots,m\rbrace}\det(X_IY_I^t) =\sum_{I\subset\lbrace1,2,\dots,m\rbrace}\det(X_I)\det(Y_I^t) $$ where the sum is over all $n$-element subsets of $\lbrace1,2,\dots,m\rbrace$. Here $X_I$ and $Y_I$ are quadratic $n\times n$ matrices with columns of $X$ and $Y$ from the set $I$. The similar formulas are available for $\det(XX^t)$ and $\det(YY^t)$, so the wanted inequality reduces to verification of $$ \left(\sum_{I}x_Iy_I\right)^2\le \sum_{I}x_I^2\sum_Iy_I^2. $$ This a known inequality.

$\endgroup$
1
  • $\begingroup$ Yep, I think this is the easiest answer. Thank you very much! $\endgroup$
    – Marine
    May 28, 2010 at 4:12
2
$\begingroup$

Denotes vectors-rows of $X$ by $x_1$, $\dots$, $x_n$, of $Y$ by $y_1$, $\dots$, $y_n$, all $x_i$, $y_i$ belong to $\mathbb{R}^m$. Let $y_i=z_i+w_i$, where $z_i$ belongs to linear span of $x_1$,$\dots$,$x_n$, and $w_i$ is orthogonal to this linear span. If we replace $y_i$ to $z_i$, the left hand side does not change, while the right hand side may only decrease ($YY^t=ZZ^t+WW^t$ with natural notations, and all matrices are non-negative definite, so all eigenvalues of $YY^t$ are at least the same as those of $ZZ^t$, hence $\det(YY^t)\geq \det(ZZ^t)$.) But if all $x_i$, $y_i$ belong to the same subspace of dimension $n$, then equality in inicial inequality occurs, as we may think that $m=n$.

$\endgroup$

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.