I need your help to prove the following result:

Let $v_1,\ldots,v_{n+1}$ be vectors of $\mathbb{R}^n$ such that $\langle v_i,v_j \rangle < 0 \; \forall i \neq j$, where $\langle \cdot \, , \cdot \rangle$ is the standard dot product of $\mathbb{R}^n$. Prove that any set of $n$ of these vectors is a basis of $\mathbb{R}^n$.

Thanks in advance.

Remark: I changed the notation from $e_1,\ldots,e_{n+1}$ to $v_1,\ldots,v_{n+1}$ to avoid ambiguity with the canonical basis.

I think I found a demostration for the case $n=2$, which is at least a starting point.

Proof. Let $v_1,v_2,v_3$ be vectors of $\mathbb{R}^2$ such that

\begin{equation} \left.\begin{aligned} \langle v_1,v_2 \rangle &< 0 \\ \langle v_2,v_3 \rangle &< 0 \\ \langle v_1,v_3 \rangle &< 0 \end{aligned} \right\} \text{ $\Longrightarrow v_1 \neq v_2 \neq v_3 \neq \vec{0}.$ } \end{equation}

Let's suppose $v_1,v_2,v_3$ are pairwise linearly dependent, which means

\begin{aligned} v_2 &= \alpha v_1, \; \alpha \in \mathbb{R} \\ v_3 &= \beta v_2, \; \beta \in \mathbb{R} \\ v_3 &= \gamma v_1, \; \gamma \in \mathbb{R}. \end{aligned}

So we have

\begin{aligned} \langle v_1,v_2 \rangle &= \langle v_1,\alpha v_1 \rangle = \alpha \, \langle v_1,v_1 \rangle = \alpha \, \underbrace{ \|v_1\|^2 }_{> 0} < 0 \Longleftrightarrow \alpha < 0 \\ \langle v_2,v_3 \rangle &= \langle \alpha v_1,\gamma v_1 \rangle = \alpha \gamma \, \langle v_1,v_1 \rangle = \alpha \gamma \, \underbrace{ \|v_1\|^2 }_{> 0} < 0 \Longleftrightarrow \alpha \gamma < 0 \\ \langle v_1,v_3 \rangle &= \langle v_1,\gamma v_1 \rangle = \gamma \, \langle v_1,v_1 \rangle = \gamma \, \underbrace{ \|v_1\|^2 }_{> 0} < 0 \Longleftrightarrow \gamma < 0. \end{aligned}

But, if $\alpha < 0$ and $\gamma < 0$, it can't be $\alpha \gamma < 0$, so we have a contradiction, which means that $v_1,v_2,v_3$ are pairwise linearly independent. Finally, as $\mathrm{dim} \left( \mathbb{R}^2 \right) = 2$, any set of two vectors chosen among $v_1,v_2,v_3$ is a basis of $\mathbb{R}^2$.


Ok, I got it. First of all, I want to thank Fimpellizieri, as I'm just actually going to write down more extensively what he/she suggested me in the comments.

Proof. Before starting let's notice that

$$\langle v_i,v_j \rangle < 0 \; \forall i \neq j \Longrightarrow v_k \neq \vec{0} \; \forall k \in \{1,\ldots,n+1\}.$$

Let's suppose that $n$ vectors among $v_1,\ldots,v_{n+1}$ are linearly dependent. For example, let's suppose, without loss of generality, that $v_1,\ldots,v_n$ are linearly dependent, i.e. $\exists \,(\lambda_1,\ldots,\lambda_n) \neq (0,\ldots,0)$ such as

$$\lambda_1 v_1 + \ldots + \lambda_n v_n = \vec{0}. \qquad \left(\small{1}\right)$$

Let's scalarly multiply by $v_{n+1}$ both members of $\left(\small{1}\right)$:

$$\langle \lambda_1 v_1 + \ldots + \lambda_n v_n, v_{n+1} \rangle = \langle \vec{0}, v_{n+1} \rangle,$$

i.e., for the linearity of the dot product,

$$\lambda_1 \, \underbrace{\langle v_1,v_{n+1} \rangle}_{< \, 0} + \ldots + \lambda_n \, \underbrace{\langle v_n,v_{n+1} \rangle}_{< \, 0} = 0.$$

This means that some coefficients $\lambda_i$ must be $>0$ and other ones $<0$. Let's suppose, without loss of generality, that

$$\lambda_1,\ldots,\lambda_k > 0, \quad \lambda_{k+1},\ldots,\lambda_n < 0, \qquad k< n.$$

Let's rewrite $\left(\small{1}\right)$ and separate the two categories of coefficients:

$$( \lambda_1 v_1 + \ldots + \lambda_k v_k ) + ( \lambda_{k+1} v_{k+1} + \ldots + \lambda_n v_n ) = \vec{0}$$

$\Longrightarrow ( \lambda_{k+1} v_{k+1} + \ldots + \lambda_n v_n ) = -( \lambda_1 v_1 + \ldots + \lambda_k v_k )$

$\Longrightarrow \langle (\lambda_1 v_1 + \ldots + \lambda_k v_k), (\lambda_{k+1} v_{k+1} + \ldots + \lambda_n v_n) \rangle = \\ \quad = \langle (\lambda_1 v_1 + \ldots + \lambda_k v_k), -(\lambda_1 v_1 + \ldots + \lambda_k v_k) \rangle = -\|\lambda_1 v_1 + \ldots + \lambda_k v_k\|^2. \qquad \left(\small{2}\right)$

On the other hand

$\langle (\lambda_1 v_1 + \ldots + \lambda_k v_k), (\lambda_{k+1} v_{k+1} + \ldots + \lambda_n v_n) \rangle = \\ = \lambda_1 \lambda_{k+1} \, \langle v_1,v_{k+1} \rangle + \ldots + \lambda_1 \lambda_n \, \langle v_1,v_n \rangle + \ldots + \lambda_k \lambda_{k+1} \, \langle v_k,v_{k+1} \rangle + \ldots + \lambda_k \lambda_n \, \langle v_k,v_n \rangle. \, \left(\small{3}\right)$

By putting together $\left(\small{2}\right)$ and $\left(\small{3}\right)$ we get

$\overbrace{\underbrace{\lambda_1 \lambda_{k+1}}_{<\,0} \, \underbrace{\langle v_1,v_{k+1} \rangle}_{<\,0}}^{>\,0} + \ldots + \overbrace{\underbrace{\lambda_1 \lambda_n}_{<\,0} \, \underbrace{\langle v_1,v_n \rangle}_{<\,0}}^{>\,0} + \ldots + \overbrace{\underbrace{\lambda_k \lambda_{k+1}}_{<\,0} \, \underbrace{\langle v_k,v_{k+1} \rangle}_{<\,0}}^{>\,0} + \ldots + \overbrace{\underbrace{\lambda_k \lambda_n}_{<\,0} \, \underbrace{\langle v_k,v_n \rangle}_{<\,0}}^{>\,0} \\ = \underbrace{-\|\lambda_1 v_1 + \ldots + \lambda_k v_k\|^2}_{<\,0},$

which is a contradiction. Consequently, $v_1,\ldots,v_n$ are linearly independent and, as $\mathrm{dim} \left( \mathbb{R}^n \right) = n$, they form a basis of $\mathbb{R}^n$. Finally, because of the arbitrariness of $v_1,\ldots,v_n$, any set of $n$ vectors chosen among $v_1,\ldots,v_{n+1}$ is a basis of $\mathbb{R}^n$.


  • $\begingroup$ Have you tried anything? Try to force some linear independence results here maybe. $\endgroup$ – Sean Roberson Sep 22 '16 at 20:17
  • $\begingroup$ Yes you must show that pair wise orthogonal vectors are independent $\endgroup$ – basket Sep 22 '16 at 20:24
  • $\begingroup$ Why @basket? $e_1, \ldots e_{n+1}$ are not pairwise othogonal. $\endgroup$ – Joyce Sep 22 '16 at 20:31
  • $\begingroup$ You meant the standard basis of $\mathbb{R}^{n+1}$? This set is pairwise orthogonal. Also I think you are confusing the direction; you say that they are independent yet not orthogonal, but I claim the converse is true: orthogonal implies independent. $\endgroup$ – basket Sep 22 '16 at 20:35
  • 1
    $\begingroup$ I would like to provide the answer but this is currently on hold. $\endgroup$ – Fimpellizieri Sep 24 '16 at 19:49

Here's a restatement of your claim which seems more approachable to me:

  1. Check that your set of $n+1$ vectors generates $R^n$. For example, see why this must be true in lower dimensions like 2 or 3.

  2. Show with the hypothesis that any subset of your generating set still generates, so since it has the right cardinality it is a basis.


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.