11
$\begingroup$

Let $K$ be a compact Hausdorff space and denote by $C(K)$ the space of all real valued and continuous functions on $K$. We endow $C(K)$ with the supremum norm topology, making it a Banach space.

Suppose now that $V \subseteq C(K)$ is a linear subspace with the following property: For every open subset $U \subseteq K$ and every $x \in U$ there exists $f \in V$ with $f(K) \subseteq [0,1]$ such that $f(x) = 1$ and $f = 0$ outside of $U$.

Does it follow that $V$ is dense in $C(K)$?

Improve this question
$\endgroup$
1
  • 1
    $\begingroup$ One may ask a further question: what is the maximum co-dimension of closed linear subspaces $V\subset C(K)$ with the stated property? $\endgroup$
    – Pietro Majer
    Nov 1 at 12:07

1 Answer 1

Reset to default
11
$\begingroup$

Here is a counterexample for $K:=[0,1]$. There exists a Borel set $B\subset K$ that meets every non-empty open set $A\subset K$ in a subset of it of positive, not full Lebesgue measure: $0<|A\cap B|<|A|$. It’s a classic exercise by Rudin (Real and Complex Analysis, Chap.2).

Consider the bounded linear form on $C(K)$ $$\psi(f):=\int_Bf(t)dt-\int_{K\setminus B}f(t)dt.$$ I claim the closed hyperplane $V:=\ker\psi$ of $C(K)$ has the stated property. In order to prove it, let's define functions $$g_{x,m}(t):=(1-m|x-t|)_+$$ for $x\in K$ and $m>0$. Note that $|\psi(g_{x,m})|\le\frac1m$, and if $B$ has density $0$ resp. $1$ at $y$, then $\psi(g_{y,m})=-\frac1m(1+o(1))$ resp. $\psi(g_{y,m})=\frac1m(1+o(1))$, as $m\to+\infty$. Then the function $$f :=g_{x,2m}+\lambda_m g_{y,m},$$ $$ \lambda_m:=-\frac{\psi(g_{x,2m})}{\psi(g_{y,m})} $$ for suitable $m$ and $y$ verifies $$0\le f\le 1,\qquad f^{-1}(1)=\{x\},\qquad \text{supp}(f)\subset U,\qquad \psi(f)=0,$$ proving the claim. In order to choose $m$ and $y$, we first fix points $y_0$ and $y_1$ in $U\setminus\{x\}$ of density $0$ resp. $1$ for $B$, which implies $\limsup_{m\to\infty}|\lambda_m|<1$, both for the choice $y:=y_0$ and $y:=y_1$. Then we fix $m$ so large that $|\lambda_m|<1$ and that $g_{x,2m}$, $g_{y_0,m}$ and $g_{y_1,m}$ have their supports in $U$, pairwise disjoint, and lastly we choose $y\in\{y_0,y_1\}$ so that $\lambda_m\ge0$.

Improve this answer
$\endgroup$
6
  • 1
    $\begingroup$ Can you explain a) how to choose $g$? b) how to fix $m, \epsilon$? $\endgroup$
    – Giorgio Metafune
    Oct 31 at 22:44
  • 1
    $\begingroup$ a) The function $g$ can be taken as well of the form $g(t):=(1-n|y-t|)_+$, with $y\in U\setminus\{x\}$ a Lebesgue point of either $B$ of $K\setminus B$. THis make the sign of $\psi(g)$ positive resp. negative, if $n$ is large enough; one also takes $n$ so large that $g$ has support in $U\setminus\{x\}$. Say we do both: a $g_-$ and a $g_+$, with $\psi(g_-)<0<\psi(g_+)$ $\endgroup$
    – Pietro Majer
    Oct 31 at 23:02
  • 1
    $\begingroup$ b) Then one fixes $m$ so large that $(1-m|x-t|)_+$ has support in $U$, disjoint from the supports of $g_+$ and $g_-$, and such that $\psi(g_-)<\psi((1-m|x-t|)_+)<\psi(g_+)$ (note that $\psi((1-m|x-t|)_+)=o(1)$ as $m\to\infty$). Finally one chooses $g_-$ or $g_+$, and fices $0\le\epsilon<1$ $\endgroup$
    – Pietro Majer
    Oct 31 at 23:12
  • 1
    $\begingroup$ And what about the counterexample with the discrete measure you added? I had a look and seemed to work? $\endgroup$
    – Giorgio Metafune
    Nov 1 at 12:12
  • 1
    $\begingroup$ Actually I see a (minor?) problem with the version with the discrete measure. The functional is now $$\Psi:=\sum_{k=1}^\infty\frac{(-1)^k}{k^2}\delta_{q_k}$$ for an enumeration $\{q_k\}_k$ of $\mathbb Q\cap [0,1]$. Then for all $k$ we have $\Psi(g_{q_k,n})=\frac{(-1)^k}{k^2}+o(1)$ as $n\to+\infty$. So the same construction with $f=g_{x,m}+\lambda g_{y,n}$ works, but for the case where $x=q_k$, for odd $k$: I do not see how to make $\Psi(f)\ge0$ and $f\ge0$ then. $\endgroup$
    – Pietro Majer
    Nov 1 at 13:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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