It is well known that for a bilinear form over an n-dimensional vector space, $n^2$ values (on all pairs of basis-vectors) determine it uniquely.

How many values do we need to specify in order to uniquely determine a norm? The first trivial observation is that homogeneity implies it's enough to specify all the values on the unit sphere.

Then, if we indeed assume $V$ is a finite-dimensional space, every two norms on $V$ are equivalent, and in particular are continuous w.r.t each other. Hence their set of mutual agreement is closed, so it is enough to specify a dense subset of the unit sphere. since a the unit sphere of a finite dimensional normed space is separable, we conclude it is enough to specify a countable number of values.

Is it possible to use only a finite number points? (the constant may depend on the dimension ofcourse).

  • 7
    $\begingroup$ There is a detailed answer below, but the rough answer is very simple. There is a bijective correspondence between norms and bounded symmetric convex sets containing a neighbourhood of the origin. (the set is the unit ball of the norm; the norm is the Minkowski functional of the set). $\endgroup$ Apr 12, 2015 at 16:14
  • 1
    $\begingroup$ @AnthonyQuas: Minor nitpick: There should be an additional condition such as closedness or openness to make this a bijection. The open and the closed unit ball (and every set in between those two) have the same Minkowski functional. $\endgroup$ Apr 14, 2015 at 23:38

1 Answer 1


Well, as you have certainly already remarked (reading your post, I assume this), bilinearity makes a big difference. For the "only norm" case, what you are looking for, if I understand correctly your question, is a set of uniqueness for the admissible norms on a given vector space $V$. Your demonstration establishes that values on a dense set $U$ (on the unit sphere) is sufficient (and then, we can reduce to countable). You can go further by choosing a set $U$ such that $U\cup (-U)$ is dense, then you cannot go further as, on $V$, a norm $p$ is the jauge function of the balanced convex $$ C=\{x\in V|\, p(x)\leq 1\} $$ then, if your set $U\cup (-U)$ is not dense, there is a point $M\in S_V$ (the unit sphere) and a neighbourhood $W$ of $\{M,-M\}$ such that $U\cap W=\emptyset$. Now, deforming the sphere around $\{M,-M\}$ in a convex way (you can find a one parameter deformation $C_t$ such), one obtains an infinite family of norms which coincide with $p$ on $U$ and differ from it. If the space $V$ is complex, just replace the "real balanced saturation" $U\mapsto U\cup (-U)$ by its analogue (the orbit of $U$ under the group of complex numbers of modulus one $\mathbb{U}=\{z\in\mathbb{C} | |z|=1\}$).

Proposition In order $U\subset S_V$ be a uniqueness set for the collection of all norms, it is necessary and sufficient that the orbit $G.U$ be dense in $S_V$ ($G=\{−1,1\}$ for the real case and $G=\mathbb{U}$ for the complex case).

[$G.U$ dense$\Longrightarrow$ $U$ is a uniqueness set] is clear.

Now, if $G.U$ is not dense, it exists a point $w\in S_V$ which does not belong to $\overline{G.U}=G.\overline{U}$ and then $$ max_{v\in U}\,\{|\langle w|v\rangle|\}=M<1 $$ Now, the convex sets $$ C_t=\{v\in V|\, ||v||\,\leq 1\ \mathit{and}\ |\langle w|v\rangle|\leq (1-t)M+t\} $$ for $0<t\leq 1$ give the desired deformation (for each $C_t$ contains $U$).

A bit more (Local implies global ?) For $p$ a norm, we will say that set $U_p\subset S_V$ is a relative uniqueness set w.r.t. $p$ if, for any norm $q$, $p|_{U_p}=q|_{U_p}\Longrightarrow p=q$. We have just seen that, if $p$ is the euclidean (resp. hermitian) norm then each $U_p$ is a uniqueness set for all norms. This is not the case for every norm (especially if it has many linear zones) for example take the two dimensional euclidean space $\mathbb{C}$ with the norm $|z|$ (real euclidean space). It can be seen easily that, with $p=l_1$ (see below), the set $$ U_{l_1}=\{e^{i\theta}\,|\, \theta\in T=[-\frac{\pi}{6},\frac{\pi}{6}]\cup [\frac{\pi}{3},\frac{2\pi}{3}]\} $$
is a relative uniqueness set w.r.t. the $l^1$ norm $||z||=|\Re(z)|+|\Im(z)|$ ; but $U\cup\, (-U)$ is not dense in the unit sphere(for example $e^{i\frac{\pi}{4}}$ is not in the closure of it).

Remark If you want $U_{l_1}$ to be denumerable replace $T$ by $T\cap \mathbb{Q}$, with the same closure.

  • 1
    $\begingroup$ an explicit deformation that works should be the set of $v$ such that $|v \cdot w| \leq 1$ for $w \in U$. $\endgroup$
    – Will Sawin
    Apr 12, 2015 at 15:40
  • $\begingroup$ can you please elaborate on your explicit example for deformation? I understand the intuitive idea that we can change the norm in a neigbourhood of $M$. I am still not sure about why the resulting function (the norm after the change) will satisfy the triangle inequality. $\endgroup$ Apr 14, 2015 at 13:12
  • $\begingroup$ I think I understood the general idea, I can just deform the original unit ball, to some other bounded, convex, symmetric shape, changin it only in a small enough neigbourhood of $M$ where I do not have any constraints about the norm's values. $\endgroup$ Apr 14, 2015 at 13:21
  • $\begingroup$ .@asaf shachar Exactly ! You are right. [can you please elaborate on your explicit example for deformation?]--> OK, I did it in the answer above. $\endgroup$ Apr 14, 2015 at 16:35

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.