Update of question

Let $V$ be the space of real polynomials in one variable $t$ of degree less than or equal to three. Define our inner product to be: $$ \langle p,q\rangle = p(1)q(1)+p'(1)q'(1)+p''(1)q''(1)+p'''(1)q'''(1). $$

If we define $\langle\cdot,\cdot\rangle$ on the space of all continuous functions it isn’t an inner product because continuous functions are not necessarily differentiable. I know we can use the projection formula to project a continuous function onto the subspace V.

Can you show me how to project $\sqrt{t}$ onto $V$ to obtain a cubic polynomial that is “near” $\sqrt{t}$.

Here is what I did:


$$ \sqrt{t}=a+bt+ct^2+dt^3 \longrightarrow(*) $$

We need to find the coefficients $a,b,c,d$. Take the inner product of both sides of $(*)$ with the functions $1,t,t^2,t^3$ will result in a system of four equations which solving them gives the desired coefficients. For instance,

$$ <\sqrt{t},1>=< a+bt+ct^2+dt^3, 1 >. $$

How do I use the definition of the inner product I have given and the linearity property of the inner product to evaluate the above though?

I have been trying to solve this for days but could not. Can someone please show me?

  • $\begingroup$ Can someone please show me there version of how they would do it please? $\endgroup$ – user99744 Nov 5 '13 at 21:32

Substituting $\sqrt t$ to the definition of the given inner product, we have $ \langle \sqrt t,\,1\rangle = \sqrt 1\cdot1+0=1 $.
Then, $\langle \sqrt t,\,t\rangle =1+1/2$, and $\ \langle\sqrt t,\,t^2\rangle = 1+1-1/2$ (unless I miscalculated), and so on.

You will need these numbers, however, probably it's easier if we work in an orthogonal basis:

for this, by the Gram-Schmidt procedure, you can start with the standard basis $(1,t,t^2,t^3)$, and set $e_0:=1$, and look for $e_1=\lambda\cdot 1+t$ which is orthogonal to $e_0$ with respect to $\langle -,-\rangle$, i.e. $\lambda\cdot\langle 1,1\rangle+\langle t,1\rangle=0$. Then, $e_2=\lambda_0e_0+\lambda_1e_1+t^2$ needs to be orthogonal to both $e_0,\,e_1$; and so on.

Finally, by the earlier calculations we can find $\langle \sqrt t,\,e_i\rangle$ for each $i=0..3$. Then either norm these $e_i$ to have norm $1$ (setting $e_i':=\displaystyle\frac{e_i}{\sqrt{\langle e_i,e_i\rangle}}$) or get the approximating polynomial directly: $$p(t):=\sum_{i\le 3} \frac{\langle \sqrt t,\,e_i\rangle}{\langle e_i,\,e_i\rangle}\cdot e_i\,. $$

  • $\begingroup$ I was also hesitating whether we can just plug in $\sqrt t$ in the defining equation of $\langle -,-\rangle$. But.. how else to go? $\endgroup$ – Berci Nov 6 '13 at 9:07

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.