7
$\begingroup$

So I know of the following result:

If $X$ is infinite dimensional, then number of inequivalent norms on $X$ is $2^{\dim X}$.

I was wondering if there is a similar result for complete norms. Basically I want to know how many complete inequivalent norms exist on an infinite dimensional space.

Edit 1: Is it at least possible to get a bound?

Edit 2: I now know an example of a space with no complete norms. So I would like to change my question to: If there exist a complete norm on an infinite dimensional vector space, then can we get an upper bound to number of inequivalent complete norms?

I also know that given a complete norm, we can construct an inequivalent complete norm. How to proceed further?

Any hints are appreciated.

$\endgroup$
  • $\begingroup$ Of course we must use AC here. In naked ZF, we cannot even exhibit a real vector space with two inequivalent complete norms. $\endgroup$ – GEdgar Sep 30 '16 at 17:49
  • $\begingroup$ What can we do using AC? $\endgroup$ – Sahiba Arora Sep 30 '16 at 17:52
  • $\begingroup$ Note that the bound here definitely fails. For $\dim (X) = \mathbb{N}$, there are no complete norms. $\endgroup$ – AJY Oct 5 '16 at 16:26
  • $\begingroup$ Yes, that's why I edited to ask that if there are complete norms, then what would the bound be? $\endgroup$ – Sahiba Arora Oct 5 '16 at 16:27
4
+50
$\begingroup$

Assume that $X$ is a real vector space of (infinite) dimension $\kappa$, and that there exists at least one complete norm on $X$. Let us try to show that there are $2^\kappa$ inequivalent complete norms on $X$.

Let us first show that there are at most $2^\kappa$ distinct norms on $X$. Let $(e_i)_{i\in I}$ be a basis of $X$, so that $\vert I\vert=\kappa$. A norm $N$ on $X$ is completely determined by its restriction to $X_{\mathbb Q}$, the set of all linear combinations of the vectors $e_i$ with rational coefficients. (This is because $N$ is determined by its restriction to $E_J:={\rm span}\{e_i;\; i\in J\}$ for any finite subset $J$ of $I$, and that $N$ is continuous on $E_J$ with respect to the unique vector space topology of the finite-dimensional space $E_J$.) Now, $X_{\mathbb Q}$ has the same cardinality as $\bigcup_{J\in\mathcal P_f(I)} \mathbb Q^J$, where $\mathcal P_f(I)$ is the set of all finite subsets of $I$; and since $\kappa$ is infinite, this is equal to $\kappa$. (The set $\mathcal P_f(I)$ has cardinality $\kappa$.) So, there are no more norms on $X$ than there are maps from $\kappa$ to $\mathbb R$; and since the cardinality of $\mathbb R^\kappa$ is equal to $\kappa$ (because $\mathbb R$ has cardinality $2^{\aleph_0}$), the result follows.

Now let us show that there are $2^\kappa$ pairwise inequivalent complete norms on $X$. Let $S$ be a set of cardinality $\kappa$. Then $S\times \mathbb N$ has cardinality $\kappa$ as well because $\kappa$ is infinite. Let $(e_{s,n})_{(s,n)\in S\times \mathbb N}$ be a basis of $X$. For any subset $A$ of $S$, consider the linear map $J_A:X\to X$ defined on the basis as follows: $J_A(e_{s,n})=ne_{s,n}$ if $s\in A$, and $J_A(e_{s,n})=e_{s,n}$ if $s\notin A$. Then $J_A$ is a linear automorphism of $X$ because it sends the basis $(e_{s,n})$ onto another basis of $X$. So the formula $\Vert x\Vert_A:=\Vert J_A(x)\Vert$ defines a norm on $X$, and this is a complete norm because $(X,\Vert\,\cdot\,\Vert_A)$ is by definition isometric with $(X,\Vert\,\cdot\,\Vert)$. Moreover, if $A\neq B$, then the norms $\Vert\,\cdot \Vert_A$ and $\Vert\,\cdot \Vert_A$ are not equivalent. Indeed, assume for example that $A$ is not contained in $B$, and choose $s_0\in A\setminus B$. Then $J_A(e_{s_0,n})=ne_{s_0,n}$ and $J_B(e_{s_0,n})=e_{s_0,n}$, so that $\Vert e_{s_0,n}\Vert_A=n\, \Vert e_{s0,n}\Vert_B$ for all $n\in\mathbb N$, which shows that $\Vert\,\cdot \,\Vert_A$ can not be controlled by $\Vert\,\cdot \,\Vert_B$. Therefore, we have found a family of pairwise inequivalent complete norms indexed by the set $\mathcal P(S)$, which has cardinality $2^\kappa$.

$\endgroup$

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.