Suppose $\| \cdot \|_{\alpha}$ and $\| \cdot \|_{\beta}$ are two norms on a vector space $X.$ Show that the two norms are equivalent if and only if the normed spaces $(X,\|\cdot\|_{\alpha})$ and $(X,\|\cdot\|_{\beta})$ have the same open sets.

My attempt:

Denote $\tau_{\alpha}$ and $\tau_{\beta}$ as topology induced by $\| \cdot \|_{\alpha}$ and $\| \cdot \|_{\beta}$ respectively.

$(\Rightarrow):$ Suppose that there exists $c_1>0$ and $c_2>0$ such that $c_1 \| x \|_{\alpha} \leq \|x\|_{\beta} \leq c_2 \| x \|_{\alpha}.$ We wish to show that $\tau_{\alpha} = \tau_{\beta}.$ Let $U_{\alpha} \in \tau_{\alpha}$ and $x \in U_{\alpha}.$ Since $U_{\alpha}$ is open, there exists $\delta>0$ such that $B(x,\delta) \subseteq U_{\alpha}.$ By assumption, we have $B_{\beta}(x,\delta c_1) \subseteq B_{\alpha}(x,\delta) \subseteq U_{\alpha}.$ Hence, $U_{\alpha} \subseteq \tau_{\beta}.$

Let $U_{\beta} \in \tau_{\beta}$ and $x \in U_{\beta}.$ Since $U_{\beta}$ is open, there exists $\delta>0$ such that $B_{\beta}(x,\delta) \subseteq U_{\beta}.$ By assumption, we have $B_{\alpha}(x,\delta/c_2) \subseteq B_{\beta}(x,\delta)\subseteq U_{\beta}.$ Hence, $U_{\beta} \in \tau_{\alpha}.$ Therefore, $\tau_{\alpha} = \tau_{\beta}.$

Is my proof above correct? I do not know how to prove the other direction.

Any hint would be appreciated.

  • $\begingroup$ Can you clarify the ball inclusions? $\endgroup$ – mavavilj Jan 19 '19 at 17:57

Yes, it is correct.

In order to prove the reverse implication, suppose that both spaces have the same open sets. Consider the ball $B_\beta(0,1)$. Since it is an open set and a neighborhood of $0$, there is a $c_1>0$ such $B_\alpha(0,c)\subset B_\beta(0,1)$. In other words, $\|x\|_\alpha<c_1\Longrightarrow\|x\|_\beta<1$. For the same reason, there is a $c_2>0$ such that $\|x\|_\beta<c_2\Longrightarrow\|x\|_\alpha<1$. Therefore, the norms are equivalent. In fact, asserting that $\|x\|_\alpha<c_1\Longrightarrow\|x\|_\beta<1$ is equivalent to asserting that $\|x\|_\alpha\leqslant c_1\Longrightarrow\|x\|_\beta\leqslant1$, which is equivalent to $\|x\|_\alpha\leqslant1\Longrightarrow\|x\|_\beta\leqslant\frac1{c_1}$. So, if $x\neq0$, then $\left\|\frac x{\|x\|_\alpha}\right\|_\alpha=1$ and so $\left\|\frac x{\|x\|_\alpha}\right\|_\beta\leqslant\frac1{c_1}$. This proves that $\|x\|_\beta\leqslant\frac1{c_1}\|x\|_\alpha$, which is still true if $x=0$. The other inequality has a similar proof.

  • $\begingroup$ How to obtain $c_1\| x \|_{\beta} \leq \| x \|_{|alpha} \leq c_2 \| x \|_{\beta}?$ I do now know how to obtain my inequality based on your proof. $\endgroup$ – Idonknow Jun 6 '17 at 5:03
  • $\begingroup$ @Idonknow see my edited answer. $\endgroup$ – Henno Brandsma Jun 6 '17 at 6:04
  • $\begingroup$ @Idonknow see my edited answer, too. $\endgroup$ – José Carlos Santos Jun 6 '17 at 6:15

In this answer, I discuss in general the conditions for two metrics (so more general than norms) to be equivalent in the topology sense.

We can apply this to the norm-based metrics $d_1(x,y) = ||x-y||_\alpha$ and $d_2(x,y) = ||x-y||_\beta$, at the point $0$ and $\epsilon =1$ say, to get that if the topologies induced by $d_1$ and $d_2$ are the same there exists $\delta_1 > 0$ such that $B_{\delta_1}(0, \delta_1) \subseteq B_{d_2}(0, 1)$ and $\delta_2 > 0$ such that $B_{\delta_2}(0, \delta_1) \subseteq B_{d_1}(0, 1)$.

Now use norm properties like $||tx|| = t\cdot||x||$ for $t>0$ to use these inclusions to find your $c_1$ and $c_2$: (scaling trickery)

Suppose that we have some $x \neq 0$, then $||x||_\alpha > 0$. Then the vector $tx$ with $t = \frac{\delta_1}{2||x||_\alpha}$ has the property that $$d_\alpha(0, tx) = ||tx||_\alpha = t\cdot ||x||_\alpha = \frac{\delta_1}{2} < \delta_1$$ which says that $tx \in B_{d_1}(0,\delta_1) \subseteq B_{d_2}(0,1)$, and this again says that $||tx||_\beta < 1$. So pulling $t$ out the $||.||_\beta$ now:

$$\frac{\delta_1}{2||x||_{\alpha}}||x||_\beta < 1 \rightarrow \frac{\delta_1}{2}||x||_\beta \le ||x||_\alpha \leftrightarrow ||x||_\beta \le \frac{2}{\delta_1}||x||_\alpha $$ where the last inequality also holds for $x=0$ (hence the $\le$). So $c_2= \frac{2}{\delta_1}$ works.

Try to find $c_1$ yourself from the other ball inclusion with $\delta_2$ etc. It's a nice exercise.


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.