# Thursday August 22nd > Todo: Find notes for first 15 minutes. ## Intervals Are Not Small **Facts:** - Countable $\implies$ Cantor, all intervals are not countable - Meager $\implies$ Baire, all intervals are not meager - Null $\implies$ Borel, all intervals are not null. *Exercise:* Verify that $f$ is continuous at $x$ iff $\lim f(x_{n}) = f(x)$ for every sequence $\theset{x_{n}} \to x$. ## Discontinuities **Definition:** If $f: X \to \RR$, the *oscillation* of $f$ at $x \in X$ is given as \begin{align*} \omega_{f}(x) = \lim_{\delta \to \infty} \sup_{y \in B_{\delta}(x)} \abs{f(y) - f(z)} .\end{align*} *Exercise:* Show that $f$ is continuous at $x \iff \omega_f(x) = 0$. We can then define points of discontinuity as $$ D_f = \theset{x \in X \suchthat \omega_f(x) > 0} = \union_{n=1}^\infty\theset{x\in X \suchthat \omega_f(x) \geq \frac 1 n} $$ *Exercise:* Show that $D_f$ is closed. **Theorem 1:** $f$ is monotone $\implies D_f$ is countable. > Hint: we can't cover $\RR$ by uncountable many disjoint intervals. **Theorem 2:** $D_f$ is always an $F_\sigma$ set. > $\RR - \QQ$ is not at $F_\sigma$ set, i.e. one can not construct a function that is discontinuous on exactly this set. **Theorem 3:** $f$ is "1st class" $\implies D_f$ is meager. > $f$ is **first class** if $f(x) = \lim_{n\to\infty} f_n(x)$ pointwise and each $f_n$ is continuous. **Theorem 4 (Lebesgue Criterion):** Let $f: [a, b] \to \RR$ be bounded, then $f$ is Riemann integrable iff $D_f$ is null. > So the Dirichlet function is not Riemann integrable. *Proof of theorems 1 and 2:* **Exercise**. *Proof of Theorem 3* We want to show that $D_f$ is meager. We know it's some countable union of some sets, and it suffices to show that they are nowhere dense. So let $F_n = \theset{x \suchthat \omega_f(x) \geq 0}$ for some fixed $n$. Let $I$ be an arbitrary closed interval, we will show that there exists a subinterval $J \subseteq I$ with $J \subseteq F_n^c$. Consider $$ E_k = \intersect_{i, j \leq k} \theset{x \suchthat \abs{f_i(x) - f_j(x)} \leq \frac 1 {5n}} $$ > Motivation: this comes from working backwards from 4-5 triangle inequalities that will appear later. Some observations: $E_k$ is closed by the continuity of the $f_i$ **(good exercise)**. We also have $E_k \subseteq E_{k+1}$. Moreover, $\union_k E_k = \RR$ because the $f_i \to f$ are Cauchy. We'll now look for an interval entirely contained in the complement. Let $I \subset \RR$ be an interval, then write $$ I = \union_k (I\intersect E_k) .$$ Baire tells us that $I$ is not meager, so at least one term appearing in this union is *not* nowhere dense, i.e. there is some $k$ for which $I \intersect E_k$ is not nowhere dense, i.e. it contains an open interval (it has a nonempty interior, and its already closed, and thus it contains an interval). So let $J$ be this open interval. We want to show that $J \subseteq F_n^c$. If $x\in J$, then $x\in E_k$ as well, and so $$ \abs{f_i(x) - f_j(x)} \leq 1/5n \text{ for all } i,j \geq k .$$ So let $i\to \infty$, so $$ \abs{f(x) - f_j(x)} \leq 1/5n \text{ for all } j\geq k .$$ Now for any $x\in J$, there exists some interval $I(x) \subseteq J$ depending on $x$ such that $\abs{f(y) - f_k(x)} \leq 2/5n$. Now rewrite this as $$ \abs{f(x) - f_j(x)} = \abs{f(y) - f_k(y) + f_k(y) - f_k(x)} .$$ This implies that $\omega_f(x) \leq 4/5n$. $\qed$ ## Integrability *Proof of Theorem 4*: Suppose that $f: [a,b] \to \RR$ is bounded. Recall that $f$ is Riemann integrable iff \begin{align*} \forall ~\varepsilon \exists \text{ a partition } P_\varepsilon = \theset{a=x_1 \leq x_2 \leq \cdots x_n = b} &\text{ of } [a,b] \\ \text{ such that }U(f, P_\varepsilon) - L(f, \varepsilon) &\leq \varepsilon ,\end{align*} where $$ U(f, P_\varepsilon) - L(f, \varepsilon) \definedas \sum_n \sup_{y, z \in [x_n, x_{n+1}]} \abs{f(y) - f(z)} (x_{n+1} - x_{n}) $$ $(\Rightarrow)$: Let $\varepsilon > 0$ and $n$ be fixed, and produce a partition $P_\varepsilon$ so that this sum is less than $\varepsilon / n$. > Recall that we want to show that $F_n$ is null. Now exclude from this sum all intervals that miss $F_n$, making it no bigger. We also know that in $F_n$, the sups are no greater than $1/n$, $$ \varepsilon/n \geq \sum \text{stuff} \geq \sum \frac 1 n (x_{n+1} - x_n) $$ $\qed$ $(\Leftarrow)$: Suppose $D_f$ is null and let $\varepsilon>0$ be arbitrary, we want to construct $P_\varepsilon$. Choose $n > 1/\varepsilon$ and $F_n \subseteq D_f$ is closed and bounded and thus compact. But a compact measure zero interval can in fact be covered by *finitely* many open intervals. So $F_n$ is covered by finitely many intervals $\theset{ I_n }_{n=1}^N$ such that $\sum \abs{I_n} \leq \varepsilon$. Now if $x\not\in F_n$, then $\exists \delta(x) > 0$ where $$ \sup_{y,z \in B_\delta(x)} \abs{f(y) - f(z)} < \frac 1 n < \varepsilon. $$ Since $(\union_j I_j)^c$ is compact, there's a finite cover $I_{N+1}, \cdots I_{N'}$ covering $F_n^c$. $\qed$