Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

In Luenberger book Cauchy-Schwarz Inequality is defined like this: For all $x,y$ in an inner product space $|(x|y)| \le \|x\|\|y\|$. Equality holds if and only if $x = \lambda y$ or $y = \theta$.

Proof starts for all scalars $\lambda$,

$$ 0 \le (x-\lambda y | x-\lambda y) = (x|x) - \lambda(y|x) - \bar{\lambda}(x|y) + |\lambda|^2 (y|y) $$

I understand this expansion. But then, it will select a particular $\lambda = (x|y)/(y|y)$, and obtains

$$ 0 \le (x|x) - \frac{|(x|y)|^2}{(y|y)} $$

I dont understand how he chose that particular $\lambda$. I guess I understand why, he chose it to get rid of it in the main equation, but is it okay to chose any $\lambda$ that will clean up the equation like this?

share|improve this question
1  
I find the notation in this question a bit hard to read. (Indeed, I think it has induced some typos.) Is that the notation from the source you cite? Using $|$ both as a separator in the notation for an inner product and for the standard modulus $|\cdot|$ makes things more difficult to parse than need be. I would suggest either $\langle \cdot,\cdot\rangle$ or, if you insist on parentheses, $(\cdot,\cdot)$. Cheers. –  cardinal Feb 7 '12 at 14:37
2  
@cardinal It seems that the author of the book really uses this notation, see the Google Books preview: books.google.com/books?id=lZU0CAH4RccC&pg=PA47 –  Martin Sleziak Feb 7 '12 at 14:41
    
@Martin: (+1) Yikes. –  cardinal Feb 7 '12 at 14:43
    
Is the condition for equality to hold stated correctly in Luenberger's book? What the OP has posted above is what is in the book in a chapter titled Hilbert Space, but it is in a section titled Pre-Hilbert Spaces. I believe that in Hilbert space, $x$ need not equal $\lambda y$ for equality to hold but rather only the slightly weaker condition $||x - \lambda y|| = 0$ is needed. But perhaps the condition is intended for spaces in which $||z||=0$ exactly when $z$ is identically $0$, and the more general version is proved later. I don't have the book, only the couple of pages in the link. –  Dilip Sarwate Feb 7 '12 at 15:12
    
Aimed at physics students, I guess, where $(x|y)$ or $\langle x | y \rangle$ is the standard. –  GEdgar Feb 7 '12 at 15:31
show 3 more comments

3 Answers 3

up vote 3 down vote accepted

I dont understand how he chose that particular $\lambda$. I guess I understand why, he chose it to get rid of it in the main equation, but is it okay to chose any $\lambda$ that will clean up the equation like this?

Yes, the inequality that you display holds for all $\lambda$ and so "it is okay to choose any $\lambda$" that you like. Luenberger's choice (it might well be the one used originally by Cauchy and/or Schwarz) "cleans up the equation" as you note, and provides motivation for its use. But if you have another value for $\lambda$ in mind that allows you to reach the conclusion $$0 \le (x|x) - \frac{|(x|y)|^2}{(y|y)},$$ (which is just a re-arrangement of the Cauchy-Schwarz Inequality), by all means, go for it.

See Appendix B of this Lecture Note for a more prolix proof of the Cauchy-Schwarz Inequality than the one in the Luenberger book.

share|improve this answer
add comment

As said before, we always have $$0\leq\|x-\lambda y\|^2=(x-\lambda y|x-\lambda y)$$ Of course the choice $\lambda=\frac{(x|y)}{\|y\|^2}$ is neither arbitrary or accidental luck. Geometrically the specific $\lambda$ chosen is the one that gives the orthogonal projection of $x$ along the vector $y$.

The product $(x|y)$ removes the part of $x$ that is orthogonal to $y$, and dividing by $\|y\|^2$ normalizes $y$.

Consider the picture below where

  • the black arrow is to picture $x$

  • the blue arrow is to picture $y$

  • the red arrow is to picture $\lambda y$

  • the green arrow is to picture $x-\lambda y$

OrthogonalProjOfXontoY

share|improve this answer
    
@JonasMeyer Yes, of course. Thanks Jonas :) –  AD. Feb 12 '12 at 7:50
add comment

You have a wrong idea. Actually, you don't need to select $\lambda$. For every $\lambda$,

$0 \le (x-\lambda y | x-\lambda y)$, that means

$0 \le (x|x)-\lambda(y|x)-\bar{\lambda}(x|y) + |\lambda|^2 (y|y)$,Consider it a function with the independent variable $\lambda$. So if $\lambda$ is R

$\Delta= 4(x|y)^2-4(x|x)(y|y) \le 0$. If $\lambda is C$, you need to use other functions.

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.