# P-Product Metrics on Real Vector Space are Topologically Equivalent

## Contents

## Theorem

For $n \in \N$, let $\R^n$ be an Euclidean space.

Let $p \in \R_{\ge 1}$.

Let $d_p$ be the $p$-product metric on $\R^n$.

Let $d_\infty$ be the Chebyshev distance on $\R^n$.

Then $d_p$ and $d_\infty$ are topologically equivalent.

## Proof

Let $r, t \in \R_{\ge 1}$.

Without loss of generality, assume that $r \le t$.

For all $x, y \in \R^n$, we are going to show that:

- $\displaystyle \map {d_r} {x, y} \ge \map {d_\infty} {x, y} \ge n^{-1} \map {d_r} {x, y}$

Then we can demonstrate Lipschitz equivalence between all of these metrics, from which topological equivalence follows.

Let $d_r$ be the metric defined as:

- $\displaystyle \map {d_r} {x, y} = \paren {\sum_{i \mathop = 1}^n \size {x_i - y_i}^r}^{1/r}$

### Inequalities for Chebyshev Distance

By definition of the Chebyshev distance on $\R^n$, we have:

- $\displaystyle \map {d_\infty} {x, y} = \max_{i \mathop = 1}^n \size {x_i - y_i}$

where $x = \tuple {x_1, x_2, \ldots, x_n}$ and $y = \tuple {y_1, y_2, \ldots, y_n}$.

Let $j$ be chosen so that:

- $\displaystyle \size {x_j - y_j} = \max_{i \mathop = 1}^n \size {x_i - y_i}$

Then:

\(\displaystyle \map {d_\infty} {x, y}\) | \(=\) | \(\displaystyle \paren {\size {x_j - y_j}^p}^{1/p}\) | |||||||||||

\(\displaystyle \) | \(\le\) | \(\displaystyle \paren {\sum_{i \mathop = 1}^n \size {x_i - y_i}^p}^{1/p}\) | |||||||||||

\(\displaystyle \) | \(=\) | \(\displaystyle \map {d_p} {x, y}\) | |||||||||||

\(\displaystyle \) | \(\le\) | \(\displaystyle \paren {n \size {x_j - y_j}^p}^{1/p}\) | |||||||||||

\(\displaystyle \) | \(=\) | \(\displaystyle n^{1/p} \size {d_\infty} {x, y}\) |

$\Box$

### Inequality for General Case

We show that $\displaystyle d_r \left({x, y}\right) \ge d_t \left({x, y}\right)$, which is equivalent to proving that:

- $\displaystyle \left({\sum_{i \mathop = 1}^n \left|{x_i - y_i}\right|^r}\right)^{1/r} \ge \left({\sum_{i \mathop = 1}^n \left|{x_i - y_i}\right|^t}\right)^{1/t}$

Let $\forall i \in \left[{1 \,.\,.\, n}\right]: s_i = \left|{x_i - y_i}\right|$.

Suppose $s_k = 0$ for some $k \in \left[{1 \,.\,.\, n}\right]$.

Then the problem reduces to the equivalent one of showing that:

- $\displaystyle \left({\sum_{i \mathop = 1}^{n-1} \left|{x_i - y_i}\right|^r}\right)^{1/r} \ge \left({\sum_{i \mathop = 1}^{n-1} \left|{x_i - y_i}\right|^t }\right)^{1/t}$

that is, of reducing the index by $1$.

Note that when $n = 1$, from simple algebra $d_r \left({x, y}\right) = d_t \left({x, y}\right)$.

So, let us start with the assumption that $\forall i \in \left[{1 \,.\,.\, n}\right]: s_i > 0$.

Let $\displaystyle u = \sum_{i \mathop = 1}^n \left|{x_i - y_i}\right|^r = \sum_{i \mathop = 1}^n s_i^r$, and $v = \dfrac 1 r$.

From Derivative of Function to Power of Function, $D_r \left({u^v}\right) = v u^{v-1} D_r \left({u}\right) + u^v \ln u D_r \left({v}\right)$.

Here:

- $\displaystyle D_r \left({u}\right) = \sum_{i \mathop = 1}^n s_i^r \ln s_i$ from Derivative of Exponential Function and Sum Rule for Derivatives
- $D_r \left({v}\right) = - \dfrac 1 {r^2}$ from Power Rule for Derivatives

In the case where $r=1$, we have:

- $D_r \left({u^v}\right) = 0$

When $r > 1$, we have:

\(\displaystyle D_r \left({\left({\sum_{i \mathop = 1}^n s_i^r}\right)^{1/r} }\right)\) | \(=\) | \(\displaystyle \dfrac 1 r \left({\sum_{i \mathop = 1}^n s_i^r}\right)^{1/ \left({r - 1}\right) } \left({\sum_{i \mathop = 1}^n s_i^r \ln s_i}\right) - \dfrac {\left({\sum_{i \mathop = 1}^n s_i^r}\right)^{1/r} \ln \left({\sum_{i \mathop = 1}^n s_i^r}\right)} {r^2}\) | |||||||||||

\(\displaystyle \) | \(=\) | \(\displaystyle \dfrac {\left({\sum_{i \mathop = 1}^n s_i^r}\right)^{1/r} } r \left({\dfrac {\sum_{i \mathop = 1}^n s_i^r \ln s_i} {\sum_{i \mathop = 1}^n s_i^r} - \dfrac {\ln \left({\sum_{i \mathop = 1}^n s_i^r}\right)} r}\right)\) | |||||||||||

\(\displaystyle \) | \(=\) | \(\displaystyle \dfrac {\left({\sum_{i \mathop = 1}^n s_i^r}\right)^{1/r} } r \left({\dfrac {r \left({\sum_{i \mathop = 1}^n s_i^r \ln s_i}\right) - \left({\sum_{i \mathop = 1}^n s_i^r}\right) \ln \left({\sum_{i \mathop = 1}^n s_i^r}\right)} {r \left({\sum_{i \mathop = 1}^n s_i^r}\right)} }\right)\) | |||||||||||

\(\displaystyle \) | \(=\) | \(\displaystyle K \left({r \left({\sum_{i \mathop = 1}^n s_i^r \ln s_i}\right) - \left({\sum_{i \mathop = 1}^n s_i^r}\right) \ln \left({\sum_{i \mathop = 1}^n s_i^r}\right)}\right)\) | where $\displaystyle K = \dfrac {\left({\sum_{i \mathop = 1}^n s_i^r}\right)^{1/r} } {r^2 \left({\sum_{i \mathop = 1}^n s_i^r}\right)} > 0$ | ||||||||||

\(\displaystyle \) | \(=\) | \(\displaystyle K \left({\sum_{i \mathop = 1}^n s_i^r \ln \left({s_i^r}\right) - \left({\sum_{i \mathop = 1}^n s_i^r}\right) \ln \left({\sum_{i \mathop = 1}^n s_i^r}\right)}\right)\) | Logarithms of Powers | ||||||||||

\(\displaystyle \) | \(=\) | \(\displaystyle K \left({\sum_{j \mathop = 1}^n \left({s_j^r \left({\ln \left({s_j^r}\right) - \ln \left({\sum_{i \mathop = 1}^n s_i^r}\right)}\right)}\right)}\right)\) | |||||||||||

\(\displaystyle \) | \(=\) | \(\displaystyle K \left({\sum_{j \mathop = 1}^n \left({s_j^r \ln \left({\frac {s_j^r} {\sum_{i \mathop = 1}^n s_i^r} }\right)}\right)}\right)\) |

where $K > 0$ because all of $s_i, r > 0$.

For the same reason, $\displaystyle \dfrac{s_j^r} {\sum_{i \mathop = 1}^n s_i^r} < 1$ for all $j \in \left\{ {1, \ldots, n}\right\}$.

From Logarithm of 1 is 0 and Logarithm is Strictly Increasing, their logarithms are therefore negative.

So for $r > 1$:

- $\displaystyle D_r \left({\left({\sum_{i \mathop = 1}^n s_i^r}\right)^{1/r}}\right) < 0$

So, from Derivative of Monotone Function, it follows that (given the conditions on $r$ and $s_i$) $\displaystyle \left({\sum_{i \mathop = 1}^n s_i^r}\right)^{1/r}$ is decreasing.

As we assumed $r \le t$, we have $d_r \left({x, y}\right) \ge d_t \left({x, y}\right)$.

$\Box$

When we combine the inequalities, we have:

- $\map {d_r} {x, y} \ge \map {d_\infty} {x, y} \ge n^{-1} \map {d_1} {x, y} \ge n^{-1} \map {d_r} {x, y}$

Therefore, $d_r$ and $d_\infty$ are Lipschitz equivalent for all $r \in \R_{\ge 1}$.

$\blacksquare$

## Sources

- 1975: W.A. Sutherland:
*Introduction to Metric and Topological Spaces*... (previous) ... (next): $2.4$: Equivalent metrics: Example $2.4.5$