I am trying to solve the following question which I came across when studying root system in euclidean spaces, with positive definite symmetric bilinear form.

Statement: Given a basis $\{v_1,\cdots, v_n\}$ of $\mathbb{R}^n$, $\exists$ $v\in\mathbb{R}^n$ such that $(v,v_i)>0$ for all $i$

My answer: Let $v=a_1v_1 + \cdots + a_nv_n$. Then $$(v,v_i)=a_1(v_1,v_i) + a_2(v_2,v_i) + \cdots + a_n(v_n,v_i), \hskip5mm i=1,2,\cdots,n.$$ Writing this in matrix form we get $$ \begin{bmatrix} (v,v_1) \\ \vdots \\ (v,v_n)\end{bmatrix} = \begin{bmatrix} (v_1,v_1) & \cdots & (v_n,v_1)\\ \vdots & & \vdots \\ (v_1,v_n) & \cdots & (v_n,v_n) \end{bmatrix} \begin{bmatrix} a_1 \\ \vdots \\ a_n\end{bmatrix}. $$ We want to find $a_1,\cdots,a_n$ such that the left column vector contains positive entries only. Let $A$ be the square matrix appearing above; it is matrix of positive definite (symmetric) inner product w.r.t given basis, so it is invertible over $\mathbb{R}$. Take $$ \begin{bmatrix} a_1 \\ \vdots \\ a_n\end{bmatrix} := A^{-1}\begin{bmatrix} 1 \\ \vdots \\ 1\end{bmatrix}.$$ This is non-zero vector, and for this choice of $a_i$'s we get the vector $v=a_1v_1+\cdots + a_nv_n$ with desired properties.

Q.1 Is the above statement and proof correct?

Q.2 Is there geometric way to prove the statement?

  • $\begingroup$ How do you know that $A$ is invertible? $\endgroup$ – 5xum Apr 10 '18 at 12:43
  • $\begingroup$ it is matrix of non-degenerate bilinear form (positive definite inner product should be non-degenerate, isn't it?) $\endgroup$ – Beginner Apr 10 '18 at 12:44
  • 1
    $\begingroup$ That's one way of showing it, sure. $\endgroup$ – 5xum Apr 10 '18 at 12:45

Your idea looks fine, but as a comment has pointed out, there are loose ends (such as why the inner product matrix is invertible) to tie up.

For a geometric proof, just do something akin to Gram-Schmidt orthogonalisation. Take $v=v_1$, so that $(v,v_1)>0$. Let $u_2=v_2-\frac{(v_1,v_2)}{(v_1,v_1)}v_1$ be the component of $v_2$ that is orthogonal to $v_1$. Then $u_2\ne0$, or else $v_1$ and $v_2$ are linearly dependent. Add $c_2u_2$ to $v$ for a sufficiently large $c_2>0$. Then $(v,v_i)>0$ for $i=1,2$. Continue in this manner, you get a desired $v$.


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.