I am reading Theorem 4.14 on Rudin's RCA. The assertion is here:
Let $\{u_\alpha : \alpha \in A\}$ be an orthonormal set in a Hilbert space $H$, and let $P$ be the space of all finite linear combinations of the vectors $u_\alpha$.
Then the inequality
$$ \sum_{\alpha \in A} |<x, u_\alpha>|^2\leq ||x||^2$$
holds for every $x\in H$.
Rudin said that since we have checked that inequality holds if $A$ is finite. I have understood that inequality holds if $A$ is finite but I don't know why this inequality holds if $A$ is countable and if $A$ is uncountable.
- 143
-
Bessel's inequality shows that you cannot have an uncountable number of the $\alpha$ for which $\langle x,u_{\alpha}\rangle \ne 0$. So you can always reduce to the countable case. The countable case is a limit of the finite version. So ... – Disintegrating By Parts Jul 29 '19 at 17:33
-
@DisintegratingByParts - that was the excuse for my answer below... – peter a g Jul 29 '19 at 17:36
2 Answers
By definition if $I$ is a non-empty set and $(a_i)_{i \in I}$ is a collection of non-negative numbers indexed by $I$ then $\sum_{i\in I} a_i $ is the supremum of all finite sums of $a_i$'s. So if the inequality holds when $A$ is replaced by any finite subset of $A$ then it holds as stated.
- 311,013
-
Sorry. I have understood the inequality holds if $A$ is finite set. I want to know how to prove if $A$ is countable and if $A$ is uncountable. – aaaa Jul 29 '19 at 14:25
-
-
Sorry. I don't know and I agree below answer that a sum of uncountable terms is not defined. – aaaa Jul 30 '19 at 01:16
-
-
Actually your theorem is The 4.17 in my edition. Certainly Rudin defines these sums before proving the theorem. – Kavi Rama Murthy Jul 30 '19 at 03:26
This is almost a comment given Kavi Rama Murthy's answer - however!
First of all, there used to be an argument - almost as a side-comment - in the book that 'uncountable' is not an issue. I haven't checked whether it's still there in the latest edition - but here it is, using Kavi Rama Murthy's notation:
Say that the supremum $M = \sum_{i \in I} a_i$ is finite ($a_i$ positive). Then consider the set $I_n=\{i\mid a_i > 1/n\}$, and write $I_\infty=\{i \mid a_i \not= 0\}= \cup I_n$. Then $I_\infty$ is countable, as $I_n$ must be finite, since $$(\# I_n/ n) \le \sum_{i \in I_n} a_i \le M.$$
That said, don't forget that, algebraically speaking, addition is a finite operation - one cannot 'add' infinitely many terms. In all cases, when one writes an infinite sum, one means a limit of finite sums. In particular, here, with $a_i$ positive, the expression $$M=\sum_{i \in I} a_i $$ is, by definition, the supremum of all sums with a finite number of summands - taken over $i\in I$. (Think Lebesgue integration...) Therefore if Rudin's inequality holds for all finite sums, it holds for $M$.
- 4,923
-
2"When I use a word,' Humpty Dumpty said in rather a scornful tone, 'it means just what I choose it to mean — neither more nor less." "The question is," said Alice, "whether you can make words mean so many different things." So I hesitated with my 'in all cases, when on writes an infinite sum,one means a limit of finite sums' above. But it does hold here. – peter a g Jul 29 '19 at 15:18
-
Rudin defines these sums and discusses them before coming to this theorem . – Kavi Rama Murthy Jul 30 '19 at 03:27
-
1Therefore the most useful part of my contribution is the Lewis Carroll quotation... I'll leave my answer/comment anyway - as should be clear, it was really only an addendum to your answer. – peter a g Jul 30 '19 at 12:20