Let $(v_1,v_2,...)$ be linearly independent vectors in an inner product space $(V, \langle\cdot,\cdot\rangle)$.

Let $(e_1, e_2,...)$ be the gram schmidt orthonormalized version of $(v_1,v_2,...)$, so $e_1=\frac{v_1}{||v_1||}$, etc.

Let $G_n$ be the $n \times n$ Gram Matrix, so $(G_n)_{i,j}=\langle v_i,v_j\rangle$.

I want to show that $\frac{det(G_{n+1})}{det(G_{n})}=||v_{n+1}-\sum_{i=1}^{n}\langle v_{n+1},e_i\rangle e_i||^2$

I'm trying to use induction on The $n=1$ case is pretty easy to prove. The inductive hypothesis is that $\exists n \in \mathbb{N}$ such that

$\frac{det(G_{n+1})}{det(G_{n})}=||v_{n+1}-\sum_{i=1}^{n}\langle v_{n+1},e_i\rangle e_i||^2$, and from this I need to show that $$\frac{det(G_{n+2})}{det(G_{n+1})}=||v_{n+2}-\sum_{i=1}^{n+1}\langle v_{n+2},e_i\rangle e_i||^2$$

This is where I get stuck. Expanding the determinant seems to be way too messy and I wasn't able to get anywhere for it, even the $n=2$ case was hard to manage. I did manage to show that for $j<i$ we have

$$\langle v_i, v_j\rangle = \langle v_i, e_j\rangle||v_j - \sum_{k=1}^{j-1}\langle v_j,e_k\rangle e_k|| + \sum_{k=1}^{i-1}\langle v_i, e_k \rangle \langle e_k, v_i \rangle$$ And also that $$\langle v_i, v_i\rangle = ||v_i - \sum_{k=1}^{i-1}\langle v_i,e_k\rangle e_k||^2 + \sum_{k=1}^{i-1}\langle v_i, e_k \rangle \langle e_k, v_i \rangle$$

Finally after messing around a lot I think that $$det(G_n)=\prod_{i=1}^{n}(||v_i||^2-\sum_{j=1}^{i-1}\langle v_i, e_j \rangle \langle e_j, v_i \rangle)$$

If this is true, then the result I'm after becomes trivial, but I also haven't managed to get very far on proving this identity either.

Any help on this would be much appreciated, it doesn't seem like it should be this complicated! Also I can provide more details if needed. Thanks!

EDIT: I found a similar question here: Special Gram's inequality Which makes me think that this has something to do with the generalized distance, and suggests there might be a less brute force method for my problem.


HINT: You can check by induction that for each $m$, the span of the vectors $(v_1, v_2, \ldots, v_m)$ equals to the span of the vectors $(e_1, e_2, \ldots, e_m)$.

Since the vector $v'_{n+1} =v_{n+1}-\sum_{i=1}^{n}\langle v_{n+1},e_i\rangle e_i$ is perpendicular to all of the vectors $e_1, \ldots, e_n$, it is perpendicular to all of the vectors $v_1, \ldots, v_n$ .

Consider the Gram determinant $G(v_1,v_2, \ldots, v_n, v'_{n+1})$. It is easy to see that it equals the Gram determinant $G(v_1, \ldots, v_n, v_{n+1})= G_{n+1}$, since we have an equation of form $$v'_{n+1} = v_{n+1} + \sum_{m=1}^n a_m v_m$$

But using the orthogonality of $v'_{n+1}$ to $v_1$, $\ldots$, $v_n$ we get $$G(v_1,v_2, \ldots, v_n, v'_{n+1}) = G_n \cdot \|v'_{n+1}\|^2$$

  • $\begingroup$ Thanks for your help! I understand much better now, but I still can't see why $G(v_1,v_2,...,v_n,v'_{n+1})=G(v_1,v_2,...,v_n,v_{n+1})$. If you could explain why this id I'd be very grateful. $\endgroup$ – J.doe Oct 7 '17 at 3:21
  • $\begingroup$ Sorry, I think I figured it out! You use the linearity of the determinant and reduce $G(v_1,v_2,...,v'_{n+1})$ to $G(v_1,v_2,...,v_{n+1})$ and a sum of other determinants, but because $v'_{n+1}$ is $v_{n+1}$ plus a linear combination of $v_m$ for $m \leq n$, each determinant has a duplicate row or column and so is zero! Thank you so much for all your help! $\endgroup$ – J.doe Oct 7 '17 at 5:25
  • $\begingroup$ @J.doe: I am glad you got it on your own. It would be fun to see how the gram determinant changes when we modify the system by a matrix. It is in fact more convenient to consider $G(v_1, \ldots v_n; w_1, \ldots w_n)$. Now the dependence is bilinear. You can write it formally as $(\langle v_1, \ldots , \langle v_n)^t \cdot (w_1 \rangle, \ldots , w_n\rangle)$. What if we change $v_i$ by a matrix $A$ and $w_i$ by a matrix $B$. It streamlines some things. There are more fun things involving Gram dets. $\endgroup$ – orangeskid Oct 7 '17 at 11:44

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.