Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Is it true that $$|| a R+b P||\leq\max \{|a|,|b|\},$$where $a$ and $b$ are complex numbers and $P,R$ are (orthogonal) projection operators on finite-dimensional closed subspaces of an infinite-dimensional hilbert space ?

In finite dimensions this would be true, since a projection there has always norm one.

share|cite|improve this question
up vote 8 down vote accepted

It is true when $R,P$ are orthogonal to each other (there is an ambiguity in terminology, as "orthogonal" could mean that $PR=0$, or that $P=P^*=P^2$, $R=R^*=R^2$).

If $PR=0$ is not assumed, then the answer is no: take $P=R=I$, $a=b=1$, then $\|aR+bP\|=2$.

Assuming $PR=0$, then $\|aR+bP\|=\max\{|a|,|b|\}$. Indeed, for any $\xi$ in the range of $R$ with $\|\xi\|=1$, we have $$ \|(aR+bP)\xi\|=\|aR\xi\|=\|a\xi\|=|a|; $$ similarly, $\|(aR+bP)\eta\|=|b|$ if $\eta$ is in the range of $P$ and $\|\eta\|=1$. So $\|aR+bP\|\geq\max\{|a|,|b|\}$.

For an arbitrary vector $\nu$ with $\|\nu\|=1$, we can write $\nu=\alpha\nu_1+\beta\nu_2+\gamma\nu_3$, for three unit vectors with $\nu_1$ in the range of $R$, $\nu_2$ in the range of $P$, and $\nu_3$ orthogonal to both the ranges of $R$ and $P$, and $\alpha,\beta,\gamma\geq0$, $\alpha^2+\beta^2+\gamma^2=1$ (see edit below for an explanation). Then $$ \|(aR+bP)\nu\|^2=\|a\alpha\nu_1+b\beta\nu_2\|^2=|a|^2\,\alpha^2 + |b|^2\,\beta^2\leq\max\{|a|^2,|b|^2\}, $$ so $\|(aR+bP)\nu\|=\sqrt{\alpha^2|a|^2+\beta^2|b|^2}\leq\max\{|a|,|b|\}$.

In conclusion, $\|aR+bP\|=\max\{|a|,|b|\}$.

Edit: below is a proof of the claim that, given three pairwise orthogonal subspaces $X$, $Y$, $Z$ of a Hilbert space $H$ that span the whole space, any unit vector $\nu\in H$ can be written as $$\nu=\alpha\nu_1+\beta\nu_2+\gamma\nu_3,\ \ \ \nu_1\in X,\ \nu_2\in Y, \ \nu_3\in Z,$$ with $\alpha,\beta,\gamma\geq0$ and $\alpha^2+\beta^2+\gamma^2=1$.

Let $\{x_j\}$, $\{y_j\}$, $\{z_j\}$ be orthonormal bases for $X$, $Y$, $Z$. Together, they form a basis for the whole $H$. So there exist coefficients such that $$ \nu=\sum_j a_jx_j + \sum_jb_jy_j+\sum_jc_jz_j. $$ As $\|\nu\|=1$, $\sum_j|a_j|^2+\sum_j|b_j|^2+\sum_j|c_j|^2=1$. Let $$ \alpha=(\sum_j |a_j|^2)^{1/2},\ \beta=(\sum_j |b_j|^2)^{1/2},\ \gamma=(\sum_j |c_j|^2)^{1/2},\ $$ and $$ \nu_1=\sum_j\frac{a_j}\alpha\,x_j,\ \nu_2=\sum_j\frac{b_j}\beta\,y_j,\ \nu_3=\sum_j\frac{c_j}\gamma\,z_j. $$ Then $\nu_1,\nu_2\nu_3$ are unit vectors, $\alpha^2+\beta^2+\gamma^2=1$, and $$ \nu=\alpha\nu_1+\beta\nu_2+\gamma\nu_3. $$

share|cite|improve this answer
Careful, $\sqrt{|a|^2+|b|^2} > \max\{|a|,|b|\}$, unless $ab=0$. – Erick Wong Jul 29 '12 at 18:03
Thanks, I forgot to use the convex coefficients. It's corrected now. – Martin Argerami Jul 29 '12 at 23:23
@MartinArgerami Before I accept your answer could you please explain, why "For an arbitrary vector $\nu$ with [.....] $\alpha+\beta+\gamma=1$" holds ? (Especially, why is $\alpha+\beta+\gamma=1$?) – user36772 Jul 30 '12 at 10:00
That was a small mistake. Actually, their squares add to $1$. I've corrected the mistake and added the explanation. – Martin Argerami Jul 30 '12 at 15:37
I think the detour through bases is unnecessary; also one of $\alpha$, $\beta$, $\gamma$ might be zero. See my answer to this question. – joriki Aug 24 '12 at 4:32

Let $\|x\|\le 1$. Then $$ \begin{eqnarray} \|aRx + bPx\|^2 &=& |a|^2\|Px\|^2 + |b|^2\|Rx\|^2\\ &\le& (|a|^2\wedge |b|^2)(\|Px\|^2 +\| Qx\|^2)\\ & \le& (|a|^2\wedge |b|^2)\|x\|^2. \end{eqnarray} $$

You can now conclude your result. The important thing is that $P$ and $Q$ are orthogonal so $\|Px + Qx\|^2 \le \|x\|^2.$

share|cite|improve this answer

Your Answer


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.