I'm trying to understand some Functional Analysis concepts from Geometrical point of view. So I know that Metric ($d(a,b)$) is the distance between elements, Norm ($||a||$) is the length of element.

So I'd like to know how Operator Norm ($||A||$, $A: X \to Y$) could be represented as Geometrical concept.


If $A:X\to Y$ is a linear bounded operator between the normed spaces $X,Y$, then the norm of $A$ is, loosely speaking, the factor by which the unit ball of $X$ gets inflated or deflated. More precisely, it is the smallest number $\alpha>0$ by which you need to dilate the unit ball of $Y$, so that $\alpha B_Y$ will contain the image of the unit ball of $X$ under the mapping $A$.

For example, a diagonal matrix $D=\{d_i\}_{i=1}^n$ in $\mathbb{R}^n$ maps the Euclidean unit ball to an ellipsoid, whose principal axis have lengths $2|d_i|$. The maximum of $|d_i|$ is of course the norm of this diagonal operator as a map between $\ell_2^n$ and itself. It is also the smallest number by which you need to multiply the unit ball of $\ell_2^n$ in order that the inflated ball will contain the ellipsoid $D(B_2^n)$, where $B_2^n$ is the unit ball of $\ell_2^n$.


Take the (closed) unit ball $B$. It is all the points of norm at most $1$.

Now, apply the operator $T$ to the unit ball, to get a convex set $T(B)$. The operator norm of $T$ is the radius of the smallest ball centered at $0$ that will contain $T(B)$. In finite dimensions, you think of $T$ turning the unit disk into some sort of ellipse-like region and the maximal distance from the origin of the boundary of this region is the operator norm.

For example, if $\|T\| = 0$, then that means that $T$ sends $B$ to the single point $0$, because the smallest closed ball containing $T(B)$ is in fact just the set $\{0\}$. We then have by linearity that $T$ is identically zero.

This is why the inequality $\|TS\| \leq \|T\|\|S\|$ is actually intuitive. Let $r$ be the radius of the smallest closed ball containing $TS(B)$ (i.e. the operator norm of $TS$). If we look at the smallest radius that contains $S(B)$, $s$, and apply $T$ to that radius ball and ask what is the smallest ball containing that set, get $\|T\|\|S\|$, but since this set clearly contains $TS(B)$, because instead of $T(S(B))$, we took $T(sB)$ and $S(B) \subset sB$, by definition of $s$.

Likewise, you can check geometrically the triangle inequality: $\|T + S\| \leq \|T\| + \|S\|$, because $(T + S)(B) \subset T(B) + S(B)$, because the first depends on adding only the points of the image that come from the same vector (i.e. $T(v) + S(v)$, but not $T(v) + S(w)$ if $v \neq w$) and the latter adds using all vectors in $B$ for arguments of $T$ and any vector in $B$ for an argument for $S$. Then we see that if $t$ is the smallest radius containing $T(B)$ and $s$ is the smallest radius contain $S(B)$, then the smallest radius containing $T(B) + S(B)$ is at most $t+s$, because $tB + sB = (t+s)B$, as sets.


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.