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

Define the $\Bbb R^2$ metric $$ d\left( x,y \right) = \begin{cases} \left|x_2-y_2\right| &, x_1 = y_1 &&\text{(d1)}\\ \left|x_1-x_2\right|+\left|x_1-y_1\right|+\left|y_1-y_2\right| &, x_1 \ne y_1 && \text{(d2)} \end{cases} $$ where $x=\left(x_1, x_2 \right) \in \Bbb R^2$.
Is $d$ complete?

I think yes, but I'm stuck in the proof.
Let $\left\{ x_n = \left(x^1_n,x^2_n \right) \right\}$ be $d$-Cauchy.
By the real triangle inequality $$ \left|x^2_n - x^2_m \right| \le \left|x^1_n-x^2_n\right|+\left|x^1_n-x^1_m\right|+\left|x^1_m-x^2_m\right| , \tag{1} $$ so it's suffice to consider (d2) - Is this correct?
From the middle term in the rhs of (1), $\{x_n^1\}$ is Cauchy in $\Bbb R$, so it converges to a candidate $x^1$.
But then, what is the candidate for $\{x^2_n\}$?

share|cite|improve this question
Where is the problem from? – Jonas Meyer Feb 10 '13 at 23:38
Quite a funny metric, +1. – 1015 Feb 10 '13 at 23:52
@JonasMeyer From my university past comprehensive exam. – Nicolas Essis-Breton Feb 11 '13 at 0:36
up vote 2 down vote accepted

Yes, this is complete. But my previous answer was wrong, and it turns out to be easier than I first thought.

First observe that $$ \max\{|x_1-y_1|,|x_2-y_2|\}\leq d(x,y) $$ for all $x,y\in\mathbb{R}^2$.

Now let $x^{(n)}=(x_1^{(n)},x_2^{(n)})$ be a Cauchy sequence for $d$.

By the previous observation, it follows that both coordinates are Cauchy in $\mathbb{R}$ so they converge to some $$ \lim x_1^{(n)}=x_1\quad\mbox{and}\quad \lim x_2^{(n)}=x_2. $$

It remains to prove that $x^{(n)}$ converges to $x=(x_1,x_2)$ for $d$.

First case: there exists $N$ such that $x_1^{(n)}=x_1$ for all $n\geq N$. Then $$ \lim d(x^{(n)},x)=\lim |x_2^{(n)}-x_2|=0. $$

Second case: there does not exist such an $N$, and so there exists an injective subsequence $(x_1^{(n_k)})$. Now $$ d(x^{(n_k)},x^{(n_l)})=|x_1^{(n_k)}-x_2^{(n_k)}|+|x_1^{(n_k)}-x_1^{(n_l)}|+|x_1^{(n_l)}-x_2^{(n_l)}|. $$ Letting $k\neq l$ tend to infinity, we get: $$ 0=|x_1-x_2|+0+|x_1-x_2| $$ so $x_1=x_2$. Finally, observe that $$ \lim |x_2^{(n)}-x_2|=0 $$ and $$ \lim |x_1^{(n)}-x_2^{(n)}|+|x_1^{(n)}-x_1|+|x_1-x_2|=0 $$ so $\lim d(x^{(n)},x)=0$ follows easily.

share|cite|improve this answer

HINT: It will be convenient to let $V_x=\{x\}\times\Bbb R$ for $x\in\Bbb R$ and to let $\Delta=\{\langle x,x\rangle:x\in\Bbb R\}$, the diagonal in $\Bbb R^2$. For any $x=\langle x_1,x_2\rangle\in\Bbb R^2$, $|x_1-x_2|$ is the distance from $x$ to $V_{x_1}\cap\Delta$, the vertical distance from $x$ to the diagonal. Thus, if $x=\langle x_1,x_2\rangle,y=\langle y_1,y_2\rangle\in\Bbb R^2$, either

  • $x_1=y_1$, $X,y\in V_{x_1}=V_{y_1}$, and $d(x,y)$ is just the usual Euclidean distance between $x$ and $y$ in $V_{x_1}$;
  • or $x_1\ne y_1$, and $d(x,y)$ is the sum of the vertical distances from $x$ and $y$ to $\Delta$ and the horizontal distance between $V_{x_1}$ and $V_{y_1}$.

We can make this easier to visualize by looking at an isometric copy of the space. Define

$$\varphi:\Bbb R^2\to\Bbb R^2:\langle x,y\rangle\mapsto\langle x,y-x\rangle\;;$$

$\varphi$ is a vertical shear that holds the origin fixed and pushes $\Delta$ to where the original $x$-axis was. For $x=\langle x_1,x_2\rangle,y=\langle y_1,y_2\rangle\in\Bbb R^2$ let

$$\rho(x,y)=\begin{cases} |x_2-y_2|,&\text{if }x_1=y_1\\ |x_2|+|x_1-y_1|+|y_2|,&\text{otherwise}\;. \end{cases}$$

It’s not hard to show that $\varphi$ is a bijection and that for all $x=\langle x_1,x_2\rangle,y=\langle y_1,y_2\rangle\in\Bbb R^2$, $d(x,y)=\rho\big(\varphi(x),\varphi(y)\big)$, so that $\langle\Bbb R^2,d\rangle$ is isometric to $\langle\Bbb R^2,\rho\rangle$, and one is complete iff the other is.

But $\rho$ is easy to visualize: $\rho$ is a variant of the taxicab metric in which the only streets are the verticals $V_x$ and the $x$-axis. The $\rho$-distance from $x=\langle x_1,x_2\rangle$ to $y=\langle y_1,y_2\rangle$ is the ordinary distance if $x$ and $y$ are on the same ‘street’ (i.e., if $x_1=y_1$, and they’re on the same vertical street, or $x_2=y_2=0$, and they’re on the same horizontal street), and in all cases it’s the length of the shortest route from $x$ to $y$ along the streets.

Now show that a $\rho$-Cauchy sequence is either eventually in some $V_x$, in which case convergence is easy, or convergent to the point on the $x$-axis determined by the limit of the $x$-coordinates of the points of the $\rho$-Cauchy sequence. (If you prefer to work directly in $\langle\Bbb R^2,d\rangle$ and give $\rho$ a miss, show that a $d$-Cauchy sequence is either eventually in some $V_x$ or convergent to some point of $\Delta$. You can also, if you like, show that $d$ is a taxicab metric in which the streets are the $V_x$’s and $\Delta$, and distances along $\Delta$ are $\frac1{\sqrt2}$ times the usual Euclidean distances.)

share|cite|improve this answer
Oh, it seems that you expanded my sketch, lucky me. Nitpick: you misdefined $\rho$. There must be two cases depending on being in the same vertical or not. – savick01 Feb 11 '13 at 9:18
@savick01: Yes, I got it right in the words and wrong in the formula. Fixed now; many thanks. (Sorry to have stepped on your answer: I hadn’t actually seen yours when I wrote that, since I’d left the page open with the answer window showing and nothing else.) – Brian M. Scott Feb 11 '13 at 9:28

The answer is positive. I can give a sketch right now and will expand it with pleasure tomorrow if needed.

First, notice that there is a linear function which is an isometry between $\mathbb R^2$ with your metric and $\mathbb R^2$ with the river metric. The river metric is similar but probably easier to think about:

$$ d\left( x,y \right) = \begin{cases} |x_2-y_2|, & x_1 = y_1 &&\text{(d1)}\\ |x_2|+|y_2|+|x_1-y_1|, & x_1 \ne y_1 && \text{(d2)} \end{cases} $$

Then notice that the river metric is complete - for a Couchy sequence coordinates must converge. If the second one converges to a nonzero number, then the first must be constant for $n$ large enough (why?) and we get the convergence. If the second converges to zero, then we also easily prove the existence of the limit.

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.