# Weierstrass's Theorem

This page has been identified as a candidate for refactoring of basic complexity.In particular: Separate out lemmas into their own pagesUntil this has been finished, please leave
`{{Refactor}}` in the code.
Because of the underlying complexity of the work needed, it is recommended that you do not embark on a refactoring task until you have become familiar with the structural nature of pages of $\mathsf{Pr} \infty \mathsf{fWiki}$.To discuss this page in more detail, feel free to use the talk page.When this work has been completed, you may remove this instance of `{{Refactor}}` from the code. |

## Theorem

There exists a real function $f: \closedint 0 1 \to \closedint 0 1$ such that:

- $(1): \quad f$ is continuous
- $(2): \quad f$ is nowhere differentiable.

## Proof

Let $C \closedint 0 1$ denote the set of all real functions $f: \closedint 0 1 \to \R$ which are continuous on $\closedint 0 1$.

By Continuous Function on Closed Interval is Complete, $C \closedint 0 1$ is a complete metric space under the supremum norm $\norm {\,\cdot \,}_\infty$.

Let $X$ consist of the $f \in C \closedint 0 1$ such that:

- $\map f 0 = 0$
- $\map f 1 = 1$
- $\forall x \in \closedint 0 1: 0 \le \map f x \le 1$

Then we have the following lemma:

### Lemma 1

$X$, defined as above, is a complete metric space under $\norm {\,\cdot \,}_\infty$.

### Proof of Lemma 1

For every $n \in \N$, let $f_n \in X$.

Furthermore, suppose that in $C \closedint 0 1$:

\(\text {(1)}: \quad\) | \(\ds \lim_{n \mathop \to \infty} \norm {f_n - f}_\infty\) | \(=\) | \(\ds 0\) |

If we can prove that $f \in X$, we know $X$ contains all its limit points.

Hence by Closed Set iff Contains all its Limit Points, $X$ is closed.

This article, or a section of it, needs explaining.In particular: Provide a link to the appropriate page which defines "closed".You can help $\mathsf{Pr} \infty \mathsf{fWiki}$ by explaining it.To discuss this page in more detail, feel free to use the talk page.When this work has been completed, you may remove this instance of `{{Explain}}` from the code. |

From Topological Completeness is Weakly Hereditary, $X$ is complete.

This article, or a section of it, needs explaining.In particular: Provide a link to the appropriate page which defines "complete" -- topologically complete, or complete in the sense that a complete metric space is complete? Utmost clarity needed here.You can help $\mathsf{Pr} \infty \mathsf{fWiki}$ by explaining it.To discuss this page in more detail, feel free to use the talk page.When this work has been completed, you may remove this instance of `{{Explain}}` from the code. |

It is now to be proved that $f \in X$.

Suppose $\map f 0 \ne 0$.

Then:

- $\forall n \in \N: \norm {f_n - f}_\infty \ge \size {\map {f_n} 0 - \map f 0} = \size {\map f 0} > 0$

This would contradict equation $(1)$.

Hence $\map f 0 = 0$.

Similarly, it is necessary that $\map f 1 = 1$.

Also, for all $n \in \N$ and $x \in \closedint 0 1$, we have that:

- $0 \le \map {f_n} x \le 1$

Suppose there is an $x \in \closedint 0 1$ such that either:

- $\map f x < 0$

or:

- $\map f x > 1$

We see that it must be that:

- $\forall n \in \N: \norm {f_n - f}_\infty \ge \norm {\map {f_n} x - \map f x} > 0$

which contradicts $(1)$.

This article contains statements that are justified by handwavery.In particular: "we see that it must be that"You can help $\mathsf{Pr} \infty \mathsf{fWiki}$ by adding precise reasons why such statements hold.To discuss this page in more detail, feel free to use the talk page.When this work has been completed, you may remove this instance of `{{Handwaving}}` from the code.If you would welcome a second opinion as to whether your work is correct, add a call to `{{Proofread}}` the page. |

Therefore, $f \in X$, and hence $X$ is complete.

$\Box$

For every $f \in X$, define $\hat f: \closedint 0 1 \to \R$ as follows:

- $\map {\hat f} x = \begin{cases} \dfrac 3 4 \map f {3 x} & : 0 \le x \le \dfrac 1 3 \\ \dfrac 1 4 + \dfrac 1 2 \map f {2 - 3 x} & : \dfrac 1 3 \le x \le \dfrac 2 3 \\ \dfrac 1 4 + \dfrac 3 4 \map f {3 x - 2} & : \dfrac 2 3 \le x \le 1 \end{cases}$

We have the following lemma:

### Lemma 2

$\hat \cdot: X \to X$ is a contraction mapping.

Furthermore, we have the following inequality:

- $\forall f, g \in X: \norm {\hat f - \hat g}_\infty \le \dfrac 3 4 \norm {f - g}_\infty$

### Proof of Lemma 2

This article is incomplete.You can help $\mathsf{Pr} \infty \mathsf{fWiki}$ by expanding it.To discuss this page in more detail, feel free to use the talk page.When this work has been completed, you may remove this instance of `{{Stub}}` from the code.If you would welcome a second opinion as to whether your work is correct, add a call to `{{Proofread}}` the page. |

$\Box$

The Contraction Mapping Theorem assures existence of a unique $h \in X$ with $\hat h = h$.

We have that $h \in X \subset C \closedint 0 1$.

Thus, by definition, $h$ is a continuous real function.

It remains to be shown that $h$ is nowhere differentiable.

To do this, we establish the following lemma:

### Lemma 3

For every $n \in \N$ and $k \in \set {1, 2, 3, 4, \ldots, 3^n}$, the following inequality holds:

- $\size {\map h {\dfrac {k - 1} {3^n} } - \map h {\dfrac k {3^n} } } \ge 2^{-n}$

### Proof of Lemma 3

For all $n \in \N$ and $k \in \set {1, 2, 3, \ldots, 3^n}$:

- $1 \le k \le 3^n \implies 0 \le \dfrac{k - 1} {3^{n + 1} } < \dfrac k {3^{n + 1} } \le \dfrac 1 3$
- $3^n < k \le 2 \cdot 3^n \implies \dfrac 1 3 \le \dfrac {k - 1} {3^{n + 1} } < \dfrac k {3^{n + 1}} \le \dfrac 2 3$
- $2 \cdot 3^n < k \le 3^{n + 1} \implies \dfrac 2 3 \le \dfrac {k - 1} {3^{n + 1} } < \dfrac k {3^{n + 1} } \le 1$

This article is incomplete.In particular: inductionYou can help $\mathsf{Pr} \infty \mathsf{fWiki}$ by expanding it.To discuss this page in more detail, feel free to use the talk page.When this work has been completed, you may remove this instance of `{{Stub}}` from the code.If you would welcome a second opinion as to whether your work is correct, add a call to `{{Proofread}}` the page. |

$\Box$

Let $a \in \closedint 0 1$ be arbitrarily selected.

It is to be shown that $h$ is not differentiable at $a$.

This is to be achieved by constructing a sequence $\sequence {t_n}$ with elements in $\closedint 0 1$, which has the following limit:

- $\ds \lim_{n \mathop \to \infty} t_n = a$

To this end, let $n \in \N$ be arbitrary.

Let $k$ be the unique largest element of $\set {1, 2, 3, 4, \ldots, 3^n}$ such that:

- $\paren {k - 1} 3^{-n} \le a \le k 3^{-n}$

By the triangle inequality:

\(\ds \) | \(\) | \(\ds \size {\map h {\frac {k - 1} {3^n} } - \map h a} + \size {\map h a - \map h {\frac k {3^n} } }\) | ||||||||||||

\(\ds \) | \(\ge\) | \(\ds \size {\map h {\frac {k - 1} {3^n} } - \map h {\frac k {3^n} } }\) | ||||||||||||

\(\ds \) | \(\ge\) | \(\ds 2^{-n}\) |

Next, let $t_n$ be either $\dfrac {k - 1} {3^n}$ or $\dfrac k {3^n}$, such that the following equation is satisfied:

- $\size {\map h {t_n} - \map h a} = \max \set {\size {\map h {\dfrac {k - 1} {3^n} } - \map h a}, \size {\map h a - \map h {\dfrac k {3^n} } } }$

This implies:

- $\forall n \in \N: t_n \ne a$

Furthermore:

- $2 \size {\map h {t_n} - \map h a} \ge 2^{-n}$

and:

- $\size {t_n - a} \le 3^{-n}$

Hence, for any $n$:

- $t_n \in \closedint 0 1$

and also:

- $\ds \lim_{n \mathop \to \infty} t_n = a$

The above inequalities imply that:

- $\dfrac {\size {\map h {t_n} - \map h a} } {\size {t_n - a} } \ge \dfrac 1 2 \paren {\dfrac 3 2}^n$

But the absolute value of this expression diverges when $n$ tends to $\infty$.

Therefore $\ds \lim_{n \mathop \to \infty} \dfrac {\map h {t_n} - \map h a} {t_n - a}$ cannot exist.

From the definition of differentiability at a point, we conclude that $h$ is not differentiable at $a$.

$\blacksquare$

## Notes

As the Contraction Mapping Theorem is not constructive, the given proof is not either.

## Source of Name

This entry was named for Karl Weierstrass.

## Historical Note

The construction of a real function which is continuous, but nowhere differentiable, was first demonstrated by Karl Weierstrass.

The demonstration that such functions exist came as a profound shock to the mathematical community.

## Sources

- 1992: George F. Simmons:
*Calculus Gems*... (previous) ... (next): Chapter $\text {A}.33$: Weierstrass ($\text {1815}$ – $\text {1897}$) - 1997: Gerard Buskes and Arnoud van Rooij:
*Topological Systems: From Distance to Neighborhood*: $8.10$ (for the core proof)