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.

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

Let $X$ and $Y$ be separable Hilbert spaces and $T:X \to Y$ be linear continuous with linear continuous inverse $T^{-1}:Y \to X$.

If $x_n$ is a countable orthnormal basis of $X$, then can I say that $Tx_n$ is a basis for $Y$? Because for $y \in Y$, there is some $x \in X$ such that $$y = Tx = T\sum_{n}(x,x_n)x_n = \sum_{n}T(x,x_n)Tx_n$$ Am I allowed to do that with pulling the operator $T$ into the sum?

Does it follow that $Tx_n$ is orthonormal too? Thanks.

share|cite|improve this question
$(Tx_n)$ would be a (Schauder) basis of $Y$. It need not be orthonormal. To see why the latter is true, one may use the fact: if $X=Y$, and $(z_i)$ are such that $\sum\Vert x_i-z_i\Vert $ is sufficiently small, then $(z_i)$ is a basis of $X$ and the map defined by $x_i\mapsto z_i$ is an isomorphism. Of course, one may "peturb" $(x_i)$ so that the resulting sequence $(z_i)$ is no longer orthogonal. – David Mitra Jan 20 '13 at 12:41
Thanks @DavidMitra. Is my "proof" correct though? The wiki says that the order matter and that the sum does not converge unconditionally, so I am unsure. – george.s Jan 20 '13 at 13:47
You don't want "$T(x,x_n)$" ($(x,x_n)$ is a scalar), just $(x,x_n)$. Since $T$ is continuous, you do have $y=\sum(x,x_n) Tx_n$. You also need to show that this representation is unique. To do that, use the fact that $T^{-1}$ is bounded. What Wiki are you looking at? If $T$ is an isomorphism, then it will preserve unconditionality of sums. In particular, $(T(x_i))$ will be an unconditional basis of $Y$. – David Mitra Jan 20 '13 at 14:17
Thanks. I was looking at the wiki for Schauder basis. I really have no idea about this unconditionality of sums thing, I guess I have to read up on it. – george.s Jan 20 '13 at 14:20
up vote 2 down vote accepted

The right hand side of your displayed equation should be $\sum(x,x_n)T(x_n)$. With this correction, your argument shows that any $y\in Y$ has a representation $y=\sum\alpha_i Tx_i$. To show that $(Tx_i)$ is, in fact, a (Schauder) basis of $Y$, you need to show that this representation is unique. So here, you have to prove that if $y=\sum\alpha_iTx_i$, then necessarily, $\alpha_i=(x,x_i)$ for every $i$. This is easily done using the fact that $T^{-1}$ is continuous and the fact that $(x_i)$ is a basis of $X$.

In fact, you can show considerably more: not only is $(Tx_i)$ a basis of $Y$; but also that the basis $(Tx_i)$ is equivalent to the basis $(x_i)$. That is, $\sum\alpha_i x_i$ converges if and only if $\sum\alpha_iTx_i$ converges. It follows then that $(Tx_i)$ is an unconditional basis of $Y$ and that $\sum\alpha_iTx_i$ converges if and only if $(\alpha_i)\in\ell_2$. A normalized basis satisfying the latter condition is called a Riez basis. (Incidentally, one can show, using the parallelogram law, that a normalized basis of a separable Hilbert space is unconditional if and only if it is a Riez basis.)

So, $(Tx_i)$ shares one important property that orthonormal bases possess. But, $(Tx_i)$ is not necessarily orthonormal or even orthogonal. This follows from the fact that if one "peturbs" a basis $(x_n)$ by a sufficiently small amount, it remains a basis equivalent to $(x_n)$. More precisely: if $(x_n)$ is a normalized basis of $X$ with basis constant $K$ and if $(y_n)$ is a sequence from $X$ with $\sum\Vert x_n-y_n\Vert\le 1/2K$, then $(y_n)$ is a basis of $X$ which is equivalent to $(x_n)$ (c.f. Lindenstrauss and Tzafriri, Classical Banach Spaces I, proposition 1.a.9.).

share|cite|improve this answer
Thanks for a great answer. – george.s Jan 20 '13 at 16:14

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.