Given the inner product space $V$ such that $\dim V > 1$.

Show that for every $\alpha \in \mathbb R$ and $v \in V$ such that $0\le \alpha \le\| v\|$ there exists a subspace $U\subset V$ such that the orthogonal projection of $v$ onto $U$ has length $\alpha$.

I think this has something to do with representing $U$ in an orthonormal basis $\{u_1\}$ for example and then saying $P_U(v)=\langle v,u_1\rangle v$. But I'm not really sure about this.

  • 3
    $\begingroup$ Let $w$ be a vector perpendicular to $v$. When you project $v$ onto $\text{span}\{w\}$, the length is 0. When you project $v$ onto $\text{span}\{v\}$, the length is $\|v\|$. What happens when you project onto $\text{span}\{u\}$, where $u$ is on the line segment connecting $v$ to $w$? $\endgroup$ – Mike Earnest Aug 4 '18 at 16:10
  • $\begingroup$ @mike I think I see that if I make the line segmant u longer than the length of the projection will get shorter? $\endgroup$ – Jason Aug 4 '18 at 17:07

As in my comment, let $w$ be a nonzero vector which is perpendicular to $w$. Such a $w$ exists; take any nonzero $x$ which is not a scalar multiple of $v$ (which exists since $\dim V>1$), and let $w=x-\frac{\langle{x,v}\rangle}{\langle{v,v}\rangle}v$.

Recall that for any $u\neq 0$, the length of the projection of $v$ onto the subspace generated by $u$ is $$ \frac{|\langle u,v\rangle |}{\|u\|} $$ Now, let $u=tw+(1-t)v$. Using the above formula, the length of the projection is $$ \frac{|\langle tw+(1-t)v,v\rangle |}{\|tw+(1-t)v\|}=\frac{|\langle tw+(1-t)v,v\rangle |}{\sqrt{\langle tw+(1-t)v,tw+(1-t)v\rangle}}=\frac{|1-t|\cdot \|v\|^2}{\sqrt{t^2\|w\|+(1-t)^2\|v\|^2}} $$ Now, consider the above as a function of $t$. We notice three things:

  • When $t=0$, the function is $\|v\|$.
  • When $t=1$, the function is $0$.
  • The function is continuous in $t$.

By the intermediate value theorem, for every $0<\alpha<\|v\|$, there exists a $t$ for which the value of the function (the length of the projection) is equal to $\alpha$, as desired.

| cite | improve this answer | |
  • $\begingroup$ For completeness you could add how to find a perpendicular vector (which is where the assumption that the dimension is more than one is necessary). $\endgroup$ – Arnaud Mortier Aug 7 '18 at 0:15
  • $\begingroup$ Thanks! I was very close but made a tiny miatake which confused me. $\endgroup$ – Jason Aug 7 '18 at 5:47

Hint: Visualize this in ${\mathbb R}^2$, projecting onto one-dimensional subspaces. The proof will generalize.

| cite | improve this answer | |
  • $\begingroup$ I think I can visualize it but I'm having troible writing down a proof $\endgroup$ – Jason Aug 4 '18 at 17:22

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.