Consider the norm $||u||=|x|+|y|$ if $u=(x,y)$. Prove that if $T:\mathbb{R^2}\longrightarrow{\mathbb{R^2}}$ satisfies $||Tu-Tv||=||u-v||$ then $T$ is of the form $Tu=p+Au$ with $A$ linear.

I've been stuck with this problem for a while now, I know what everything is: a norm, a linear application... And their properties, however I just don't know what to do.

  • 2
    $\begingroup$ This is known as the Mazur-Ulam Theorem. $\endgroup$ – Theo Bendit Oct 11 '18 at 12:49
  • 1
    $\begingroup$ @JoséCarlosSantos I disagree. The dupe target uses the Euclidean norm, whereas this question uses the $1$-norm. $\endgroup$ – Theo Bendit Oct 11 '18 at 12:57
  • 1
    $\begingroup$ @TheoBendit You are right. I will rectract my closing vote. $\endgroup$ – José Carlos Santos Oct 11 '18 at 12:57
  • $\begingroup$ Can't we apply Mazur-Ulam Theorem as suggests by @TheoBendit? $\endgroup$ – Idonknow Jan 4 at 10:06

An isometry like $T$ must preserve spheres and balls. In this case, the sphere is diamond-shaped. Try plotting it now if you don't know what it looks like.

If $T$ is an isometry, then so is $S \circ T$ where $S$ is another isometry. If we consider $S$ to be a translation that maps $T(0)$ to $0$, then $S \circ T$ is an isometry that fixes $0$. If we can show that $S \circ T$ must be linear, then this will confirm that $T$ must be of the appropriate form, hence we assume without loss of generality that $T(0) = 0$.

Consider the unit sphere (centred at $0$, with radius $1$). Note that the four corners, $(1, 0), (0, 1), (-1, 0),$ and $(0, -1)$, are the only four points on the sphere that are all exactly $2$ distance apart. Since $T$ fixes $0$, the unit sphere must map onto itself, and the four corners must all map to $4$ distinct points that are precisely $2$ apart. That is, the four corners must be stabilised.

Moreover, $T$ must preserve opposite corner pairs (e.g. $(1, 0)$ and $(-1, 0)$), as only opposite corners have the property that only a single point is distance $1$ from each of them (the point $(0, 0)$). When considering non-opposite corners, there are infinitely many such points!

There are known linear isometries: rotation by $\pi/2$, $\pi$, and $3\pi/2$ radians, as well as reflections in the $x$-axis, $y$-axis, and the lines $x = y$ and $x = -y$. Using similar techniques to the second paragraph, we may assume without loss of generality that $T(1, 0) = (1, 0)$ using rotations. Then, since opposite corners are preserved, $(0, 1)$ and $(0, -1)$ must be preserved. By using reflections, we may further assume $(0, 1)$ and $(0, -1)$ are fixed!

Now, we have an isometry $T$ that maps $(0, 0)$ to $(0, 0)$ and fixes all four corners. It falls to us to show that $T$ is the identity map.

First, we show it fixes the axes. If we have a point $(\lambda, 0)$ on the $x$-axis with $\lambda > 0$, then it is a corner on the sphere of radius $\lambda$ centred at $(0, 0)$. By similar arguments, it must map to another corner of the sphere. But, we have $$\|T(\lambda, 0) - (1, 0)\| = \|T(\lambda, 0) - T(1, 0)\| = \|(\lambda, 0) - (1, 0)\| = |\lambda - 1|.$$ If $T(\lambda, 0) = (0, \pm \lambda)$, or indeed if $T(\lambda, 0) = (-\lambda, 0)$, then $\|T(\lambda, 0) - (1, 0)\| = 1 + \lambda$. In either case, we must have $$1 + \lambda = |\lambda - 1| \implies (1 + \lambda)^2 = (\lambda - 1)^2 \implies \lambda = 0,$$ which is a contradiction. Thus, $T(\lambda, 0) = (\lambda, 0)$, so the positive $x$-axis is fixed. Similar arguments work for the negative $x$-axis, and the $y$-axis.

Now, consider a point $(x, y) \in \mathbb{R}^2$, with $x, y \neq 0$. Note that the point of smallest distance from $(x, y)$ to the $x$-axis is $(x, 0)$. The ball of radius $|x|$, centred at $(x, y)$, intersects the $x$-axis at only one point: $(x, 0)$, and the point is on the sphere.

This ball must map to a ball of radius $|x|$, centred at the point $T(x, y)$. It too will intersect the $x$ axis at precisely one point: $T(x, 0) = (x, 0)$. Note that $(x, 0)$ is a corner point on both spheres (centred at $(x, y)$ and at $T(x, y)$). It can't be either of the corners to the left or right of $T(x, y)$, as that would imply $T(x, y)$ lies on the $x$-axis at some point $(x_0, 0)$, and since $T$ is injective, this implies $y = 0$, which is a contradiction. Thus, it must be the top or bottom corner. Hence, we must have $T(x, y)$ must lie on the vertical line passing through $(x, 0)$.

Similar reasoning shows that $T(x, y)$ must lie on the horizontal line passing through $(0, y)$. The only such point is $(x, y)$, so $T(x, y) = (x, y)$, completing the proof.


Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

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