Let $v,u\in\mathbb{R}^n$ and $A\in M_n\left(\mathbb{R}\right)$ an invertible matrix.

Prove $\left\langle v | u \right\rangle = v\cdot (A^tAu) $ is an inner product over the reals.

I was able to (allegedly) prove both symmetry, linearity and positive-definiteness without using the invertibility of $A$ at all... Linearity is straightforward; symmetry I showed as follows: $$\left\langle v | u \right\rangle = v\cdot (A^tAu) = (A^tAu)\cdot v = (A^tAu)^t v = (u^tA^t(A^t)^t)v$$ $$ = u^t(A^tAv) = u\cdot(A^tAv) = \left\langle u | v \right\rangle$$ And finally for positive-definiteness I identified $A^tA$ as a symmetric matrix $M$ with non-negative entries in the main diagonal: $$ A^tA \equiv M \begin{bmatrix}a^2_{11} & a_{12} & \dots & a_{1n}\\a_{12} & a^2_{22} & \dots & a_{1n}\\ \vdots & & \ddots & \vdots\\a_{1n}& \dots & & a^2_{nn}\end{bmatrix} $$ yielding in short $$\left\langle v | v \right\rangle = \sum_{i=1}^na_{ii}^2v_i^2 + 2\sum_{i=1}^n\sum_{j\neq i}a_{ij}v_iv_j = \left(\sum_{i,j=1}^na_{ij}v_iv_j\right)^2\geq0$$ Is there anything I'm missing in my proof?

Edit: I think I got it - A needs to be invertible for $\left\langle v | u \right\rangle = 0 $ to hold iff $v=0$, otherwise it can be the zero matrix. Is there another requirement for the invertibility?

  • $\begingroup$ The non-negative entries in the main diagonal does not ensure positive-definiteness. For example $\begin{bmatrix} 1&-2\\ -2&1\end{bmatrix}$ is not positive-definite (though it's probably not a $A^\top A$ with $A$ invertible and real since it's det is negative). $\endgroup$ – anderstood Jun 14 '16 at 22:22
  • $\begingroup$ And I don't think $\langle v|v\rangle = (\sum a_{ij}v_iv_j)^2$. For example, where's the $a_{12}v_1^2$ term in your sum? $\endgroup$ – anderstood Jun 14 '16 at 22:32
  • $\begingroup$ I understand your first comment, you are correct. Concerning your second- $a_{12}v_1^2$ shouldn't be in my sum, $a_{12}$ should only multiply $v_1v_2$, and specifically twice. But nevertheless the calculation is indeed false, as I now see... mehh... $\endgroup$ – Yoni Jun 14 '16 at 22:38
  • $\begingroup$ I assumed $a_{12}$ was the element (1,2) of $A$. If so, it should be in the sum defined by $v^\top A^\top A v$. You can easily check this with $A\in\mathbb R^{2\times 2}$, for example. $\endgroup$ – anderstood Jun 14 '16 at 22:58
  • $\begingroup$ I probably should've wrote the entries as $m_{ij}$, I marked the entries on the main diagonal a squared to show explicitly they're positive. But I see your point by the example... Thanks $\endgroup$ – Yoni Jun 15 '16 at 7:37

You need to show that if $v$ is not zero then $\langle v, v\rangle > 0$.

This requires the assumption that the null space of $A $ is trivial.

(Note that $ \langle v,v \rangle = v^T A^T Av = (Av)^T Av = \| Av \|_2^2$.)

  • $\begingroup$ Can you please clarify your note? I don't understand how you defined the norm. $\endgroup$ – Yoni Jun 14 '16 at 22:42
  • $\begingroup$ I added a little more detail, let me know if it's still unclear. $\endgroup$ – littleO Jun 14 '16 at 22:52
  • 1
    $\begingroup$ Ok, now it's totally clear... Thank you $\endgroup$ – Yoni Jun 15 '16 at 7:34

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.