Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

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

I can prove with the triangle inequality that the unit sphere in $R^n$ is convex, but how to show that it is strictly convex?

share|cite|improve this question
Are you referring to the unit ball? – Giuseppe Nov 7 '12 at 17:38
@GiuseppeTortorella I'm talking about $||x||\leq 1$ – André Nov 7 '12 at 17:39
Since there are norms for which the unit ball is not strictly convex (think 1-norm and max-norm), you should probably revise the question to specify which norm (e.g. Euclidean as in Hagen von Eitzen's Answer), or indicate that counterexamples are of possible interest. – hardmath Nov 7 '12 at 17:56
@hardmath Thanks, I meant $R^n$. – André Nov 7 '12 at 17:58
@André: But saying $\mathbb{R}^n$ only specifies the Cartesian set, not the norm involved in your Question. Wikipedia has more about strictly convex spaces. – hardmath Nov 7 '12 at 18:10
up vote 1 down vote accepted

For the euclidean metric, we see that the unit ball is strictly convex because for different vectors $a$ and $b$ we have that $$f(t):=||ta+(1-t)b||^2\\ =\langle ta+(1-t)b, ta+(1-t)b\rangle\\ = t^2||a||^2+(1-t)^2||b||^2 + 2t(1-t)\langle a,b\rangle\\= (\ldots) t^2+(\ldots)t + (\ldots)$$ is a quadratic function with positive leading term (because $f(t)\to+\infty$ as $t\to\infty$), hence $f$ assumes its maximum on the interval $[0,1]$ only at one or both of the endpoints. In other words: We make use of the fact that $f(t)=c_2t^2+c_1t+c_0$ with $c_2>0$ is strictly convex, which implies $f(t)<\max\{f(0),f(1)\}$ for $0<t<1$.

share|cite|improve this answer

I will show that if $(E,\langle\dot,\dot\rangle)$ is an inner product space, then the normed space $(E,\|\dot\|)$ is strictly convex, where $\|x\|:=\sqrt{\langle x,x\rangle}.$

Take any two point $x,y\in E,$ with $\|x\|=\|y\|=1$ and $x\neq y.$ Then for any $0<\alpha<1,$ we have $\|\alpha x+(1-\alpha) y\|<1,$ or equivalently $\|\alpha x+(1-\alpha) y\|^2-1<0.$ Infact:

$$\|\alpha x+(1-\alpha) y\|^2-1=\alpha^2+(1-\alpha)^2+2\alpha(1-\alpha)\langle x,y\rangle-1=2(1-\langle x,y\rangle)(\alpha-1)\alpha$$

Which is negative for $\alpha\in]0,1[$ because the Cauchy inequality and the hypothesis $\|x\|=\|y|=1,$ $x\neq y,$ imply $-1\leq\langle x,y\rangle<1.$

share|cite|improve this answer
+1 for pointing out the same proof given by @FlybyNight actually applies more generally, to any inner product space. – hardmath Nov 8 '12 at 15:28

To show that the closed unit ball $B$ is strictly convex we need to show that for any two points $x$ and $y$ in the boundary of $B$, the chord joining $x$ to $y$ meets the boundary only at the points $x$ and $y$.

Let $x,y \in \partial B$, then $||x|| = ||y|| = 1.$ Now consider the chord joining $x$ to $y$. We can parametrise this by $c(t) := (1-t)x + ty$. Notice that $c(0) = x$ and $c(1) = y$. We need to show that $c(t)$ only meets the boundary when $t=0$ or $t=1$. Well:

$$||c(t)||^2 = \langle c(t), c(t) \rangle = (1-t)^2\langle x, x \rangle + 2(1-t)t \, \langle x,y \rangle + t^2 \langle y,y \rangle$$

$$||c(t)||^2 = (1-t)^2||x||^2 + 2t(1-t)\langle x,y \rangle + t^2||y||^2$$

Since $x,y \in \partial B$ it follows that $||x|| = ||y|| = 1$ and so:

$$||c(t)||^2 = (1-t)^2 + 2t(1-t)\langle x,y \rangle + t^2 \, . $$

If $c(t)$ meets the boundary then $||c(t) || = 1$, so let's find the values of $t$ for which $||c(t)|| = 1$:

$$(1-t)^2 + 2t(1-t)\langle x,y \rangle + t^2 = 1 \iff 2t(1-t)(1-\langle x, y \rangle) = 0 \, .$$

Clearly $t=0$ and $t=1$ are solution since $c(0)$ and $c(1)$ lie on the boundary. Recall that $\langle x, y \rangle = \cos\theta$, where $\theta$ is the angle between vectors $\vec{0x}$ and $\vec{0y}$, because $||x|| = ||y|| = 1.$ Thus, provided $x \neq y$ we have $\langle x, y \rangle \neq 1$ and so the chord only meets the boundary at $c(0)$ and $c(1).$

share|cite|improve this answer

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.