[guided]We now have the integral representation $\tilde{u}(x) - \tilde{u}(y) = \int_y^x Du(t)\, d\mathcal{L}^1(t)$ and want to extract a Hölder bound. The right-hand side is the integral of a function in $L^p$ — one could try the crude bound $\int |Du| \leq |x-y| \cdot \|Du\|_{L^\infty}$, but that requires $p = \infty$. For finite $p$, the natural way to relate an $L^1$-norm of a function on a small set to its $L^p$-norm and the size of the set is Hölder's inequality: the smaller the set, the smaller the $L^1$-norm contribution, with the gain measured by an exponent depending on $p$.
Take $y \leq x$ and start with
\begin{align*}
|\tilde{u}(x) - \tilde{u}(y)| \leq \int_y^x |Du(t)|\, d\mathcal{L}^1(t).
\end{align*}
Rewrite this integral as a pairing of $|Du|$ with the indicator $\mathbb{1}_{[y,x]}$:
\begin{align*}
\int_y^x |Du|\, d\mathcal{L}^1 = \int_a^b |Du(t)| \cdot \mathbb{1}_{[y,x]}(t)\, d\mathcal{L}^1(t).
\end{align*}
Now apply [Hölder's Inequality](/theorems/???) to the pair $f := |Du|$ and $g := \mathbb{1}_{[y,x]}$ on $(a,b)$ with exponents $p$ and $p'$. The hypotheses of Hölder are: $1/p + 1/p' = 1$ (we have $p' = p/(p-1)$) and the functions lie in $L^p$ and $L^{p'}$ respectively. Both are met: $|Du| \in L^p(a,b)$ because $u \in W^{1,p}$, and $\mathbb{1}_{[y,x]} \in L^{p'}(a,b)$ since the indicator of a bounded set lies in every $L^q$. Hölder gives
\begin{align*}
\int_a^b |Du| \cdot \mathbb{1}_{[y,x]}\, d\mathcal{L}^1 \leq \||Du|\|_{L^p} \cdot \|\mathbb{1}_{[y,x]}\|_{L^{p'}}.
\end{align*}
The first factor is $\|Du\|_{L^p(a,b)}$ by definition. For the second,
\begin{align*}
\|\mathbb{1}_{[y,x]}\|_{L^{p'}}^{p'} = \int_a^b \mathbb{1}_{[y,x]}^{p'}\, d\mathcal{L}^1 = \int_a^b \mathbb{1}_{[y,x]}\, d\mathcal{L}^1 = |x - y|,
\end{align*}
so $\|\mathbb{1}_{[y,x]}\|_{L^{p'}} = |x - y|^{1/p'}$. Combining and using $1/p' = 1 - 1/p$,
\begin{align*}
|\tilde{u}(x) - \tilde{u}(y)| \leq \|Du\|_{L^p(a,b)} \cdot |x - y|^{1 - 1/p}.
\end{align*}
For $p = \infty$, the conjugate is $p' = 1$, and the formula $1 - 1/p = 1$ predicts a Lipschitz estimate. Hölder still works ($\|\mathbb{1}_{[y,x]}\|_{L^1} = |x-y|$, $\|Du\|_{L^\infty}$ as the other factor), but it is more direct to use the essential bound $|Du(t)| \leq \|Du\|_{L^\infty}$ a.e. and integrate.
Why does the proof need $p > 1$? For $p = 1$, the conjugate $p' = \infty$ and the prediction $1 - 1/p = 0$ gives no decay in $|x - y|$ — i.e. no Hölder regularity. This matches the well-known fact that $W^{1,1}$ functions in 1D are absolutely continuous but not in general (Hölder-)continuous with a uniform modulus.[/guided]