$(x^*Ay)^2 \leq (x^*Ax)(y^*Ay)$

Does this inequality hold for positive semi-definite matrix A? I wonder if this is equivalent to Cauchy-Schwarz Inequality. I tried to diagonalize A but it still half-way towards Cauchy-Schwarz Inequality.

Any thoughts on this inequality? Am I going on the right direction?

Thanks a lot!

  • $\begingroup$ Complex or real? If complex, you need $(x^*Ay)(y^*Ax)$ on the left. $\endgroup$ Aug 27 '11 at 16:39
  • $\begingroup$ @Jonas Meyer Real $\endgroup$
    – BVFanZ
    Aug 27 '11 at 17:00

The Cauchy-Schwarz inequality applies to any semi-definite inner product, including the one on $\mathbb{R}^n$ (as column vectors) defined by $\langle x,y\rangle=x^{\mathrm T}Ay$.

One of the simplest proofs in the real case is to consider $p(t)=\langle x-ty,x-ty\rangle=\langle x,x\rangle -2\langle x,y\rangle t +\langle y,y\rangle t^2$. Then $p(t)\geq 0$ for all $t\in \mathbb R$, and $p$ is a quadratic polynomial in $t$, so its discriminant must be nonpositive. Rearranging this discriminant inequality yields your inequality.


The matrix $A$ has a Cholesky decomposition: $A=L^* L$. Thus

$$x^* A y = (Lx)^* Ly = (Lx) \cdot (Ly),$$ where by $\cdot$ we mean the usual complex inner product: $a \cdot b = \sum_i \overline{a_i} b_i$. Now applying Cauchy-Schwarz,

$$ |x^* A y|^2 \leq ||Lx||_2^2 ||Ly||_2^2 = (x^* L^* L x) (y^* L^* L y) = (x^* A x) (y^* A y)$$

  • 1
    $\begingroup$ BVFanZ has clarified in a comment that only the real case is being considered. In the complex case, you should have $|x^*Ay|^2$ rather than $(x^*Ay)^2$. $\endgroup$ Aug 27 '11 at 17:03
  • $\begingroup$ @Jonas Meyer - indeed, thanks. $\endgroup$
    – morgan
    Aug 27 '11 at 17:05

Yes, it is equivalent to CS inequality. W.l.o.g, asumme $A$ is a diagonal matrix with positive diagonal entries, $A=diag(a_1,\cdots, a_n)$. you will see that $\left(\sum \sqrt{a_i}x_i\sqrt{a_i}y_i\right)\le \left(\sum (\sqrt{a_i}x_i)^2\right)\left(\sum (\sqrt{a_i}y_i)^2\right) $ is in the same form of CS inequality.


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.