We will use
(6.2.1) for the only if part.
Suppose that
\(f\) is Riemann integrable on
\(R\text{.}\) For each
\(k\in \bbN\text{,}\) we will prove that
\(D_{k}:=\{\by: \osc(f)(\by)\ge \frac 1k\}\) is a set of content
\(0\text{.}\)
Given any \(\epsilon > 0\text{.}\) There exists a partition \(\cP=\{R_{\alpha}\}\) of \(R\) such that
\begin{equation*}
\sum_{\alpha} \osc(f, R_{\alpha})\vert R_{\alpha} \vert < \frac{\epsilon}{2k}.
\end{equation*}
The rectangles in \(\cP\) are divided into two subgroups: the subgroup \(\cL_{k}\) consisting those \(R_{\alpha}\) such that \(\osc(f, R_{\alpha})\ge \frac{1}{2k}\text{,}\) and the subgroup \(\cS_{k}\) consisting those \(R_{\alpha}\) such that \(\osc(f, R_{\alpha})< \frac{1}{2k}\text{.}\) Then it follows from
\begin{equation*}
\frac{1}{2k} \sum_{R_{\alpha}\in \cL_{k} } \vert R_{\alpha} \vert \le
\sum_{R_{\alpha}\in \cL_{k} } \osc(f, R_{\alpha})\vert R_{\alpha} \vert < \frac{\epsilon}{2k}
\end{equation*}
that
\begin{equation*}
\sum_{R_{\alpha}\in \cL_{k} } \vert R_{\alpha} \vert < \epsilon.
\end{equation*}
We now claim that
\begin{equation*}
D_{k}\subset \cup_{R_{\alpha}\in \cL_{k} } R_{\alpha}.
\end{equation*}
This will show that \(D_{k}\) is a set of content \(0\text{.}\)
If the claim were not true, there would exist some
\(\bx \in D_{k} \setminus \cup_{R_{\alpha}\in \cL_{k} } R_{\alpha}\text{.}\) Thus
\(\bx\) must lie entirely in the union of the rectangles in
\(\cS_{k}\text{.}\) Because
\(\cP\) is a finite partition,
\(\cup_{R_{\alpha}\in \cL_{k}} R_{\alpha}\) is a closed set. Its complement is open, so there exists some ball
\(B(\bx, r_1)\) completely disjoint from all rectangles in
\(\cL_{k}\text{.}\)
Let
\(R_1, R_2, \dots, R_m\) be the finite collection of rectangles in
\(\cS_{k}\) that actually contain the point
\(\bx\text{.}\) Since
\(\bx\) does not belong to any other rectangle in the partition, there exists a sufficiently small radius
\(r \le r_1\) such that the ball
\(B(\bx, r)\) only intersects
\(R_1, \dots, R_m\text{.}\) Thus,
\(B(\bx, r) \subset \bigcup_{i=1}^m R_i\text{.}\)
Because
\(R_i \in \cS_k\) for each
\(i \in \{1, \dots, m\}\text{,}\) we know that
\(\osc(f, R_i) < \frac{1}{2k}\text{.}\) Let
\(c = \max_{1 \le i \le m} \osc(f, R_i)\text{.}\) We strictly have
\(c < \frac{1}{2k}\text{.}\)
Now, choose any arbitrary point \(\by \in B(\bx, r)\text{.}\) By our construction, \(\by\) must belong to at least one of these rectangles, say \(R_j\text{.}\) Because both \(\bx\) and \(\by\) belong to \(R_j\text{,}\) we have:
\begin{equation*}
\vert f(\by) - f(\bx) \vert \le \osc(f, R_j) \le c
\end{equation*}
Therefore, for any two points \(\by, \bz \in B(\bx, r)\text{,}\) we can use the triangle inequality through the anchor point \(f(\bx)\text{:}\)
\begin{equation*}
\vert f(\by) - f(\bz) \vert \le \vert f(\by) - f(\bx) \vert + \vert f(\bx) - f(\bz) \vert \le 2c
\end{equation*}
Taking the supremum over all \(\by, \bz \in B(\bx, r)\text{,}\) we find the oscillation of \(f\) on the ball:
\begin{equation*}
\osc(f, B(\bx, r)) \le 2c < \frac{1}{k}
\end{equation*}
By definition, \(\osc(f)(\bx) \le \osc(f, B(\bx, r))\text{,}\) which implies \(\osc(f)(\bx) < \frac{1}{k}\text{.}\) This strictly contradicts the assumption that \(\bx \in D_{k}\text{.}\) Thus, our claim holds, and \(D_k\) is a set of content \(0\text{.}\)
For the if part, assume that the set of discontinuities
\(D\) of
\(f\) has measure
\(0\text{.}\) Since
\(f\) is bounded on
\(R\text{,}\) there exists
\(M > 0\) such that
\(\vert f(\bx) \vert \le M\) for all
\(\bx \in R\text{.}\) Given any
\(\epsilon > 0\text{,}\) we want to find a partition
\(\cP\) of
\(R\) such that the difference between the upper and lower Darboux sums is less than
\(\epsilon\text{.}\)
Let
\(\epsilon' = \frac{\epsilon}{2 \vert R \vert}\text{.}\) Consider the set
\(D_{\epsilon'} = \{\bx \in R : \osc(f)(\bx) \ge \epsilon'\}\text{.}\) Because
\(D_{\epsilon'} \subset D\) and
\(D\) has measure
\(0\text{,}\) \(D_{\epsilon'}\) also has measure
\(0\text{.}\) Furthermore, since the oscillation function is upper semi-continuous,
\(D_{\epsilon'}\) is closed. Being a closed subset of a bounded rectangle
\(R\text{,}\) \(D_{\epsilon'}\) is compact. A compact set of measure
\(0\) has content
\(0\text{,}\) so we can cover
\(D_{\epsilon'}\) with a finite collection of open rectangles
\(\{V_1, \dots, V_m\}\) such that the sum of their volumes is less than
\(\frac{\epsilon}{4M}\text{.}\)
For every point
\(\bx \in R \setminus D_{\epsilon'}\text{,}\) we have
\(\osc(f)(\bx) < \epsilon'\text{.}\) By the definition of oscillation at a point, there exists an open rectangle
\(W_{\bx}\) containing
\(\bx\) such that
\(\osc(f, W_{\bx}) < \epsilon'\text{.}\) The collection
\(\{V_1, \dots, V_m\} \cup \{W_{\bx}\}_{\bx \in R \setminus D_{\epsilon'}}\) forms an open cover of the compact set
\(R\text{.}\) Therefore, it has a finite subcover, say
\(\{V_1, \dots, V_m, W_1, \dots, W_p\}\text{.}\)
We can form a partition
\(\cP\) of
\(R\) by extending the faces of all the rectangles in this finite subcover. For each subrectangle
\(S \in \cP\text{,}\) \(S\) is either entirely contained in some
\(W_j\text{,}\) or it is not. We divide the subrectangles of
\(\cP\) into two groups:
\(\cA\text{,}\) containing those subrectangles that intersect
\(D_{\epsilon'}\) (and thus must be covered by the union of
\(V_i\)), and
\(\cB\text{,}\) containing the remaining subrectangles which are entirely contained within some
\(W_j\text{.}\)
For any \(S \in \cA\text{,}\) the oscillation \(\osc(f, S) \le 2M\text{.}\) Since these subrectangles are covered by the rectangles \(V_i\text{,}\) the sum of their volumes is bounded by \(\sum \vert V_i \vert < \frac{\epsilon}{4M}\text{.}\) Thus,
\begin{equation*}
\sum_{S \in \cA} \osc(f, S) \vert S \vert \le 2M \sum_{S \in \cA} \vert S \vert < 2M \left( \frac{\epsilon}{4M} \right) = \frac{\epsilon}{2}.
\end{equation*}
For any \(S \in \cB\text{,}\) \(S\) is contained within some \(W_j\text{,}\) meaning \(\osc(f, S) \le \osc(f, W_j) < \epsilon' = \frac{\epsilon}{2 \vert R \vert}\text{.}\) Thus,
\begin{equation*}
\sum_{S \in \cB} \osc(f, S) \vert S \vert < \frac{\epsilon}{2 \vert R \vert} \sum_{S \in \cB} \vert S \vert \le \frac{\epsilon}{2 \vert R \vert} \vert R \vert = \frac{\epsilon}{2}.
\end{equation*}
Summing these two parts over our partition \(\cP\text{,}\) we find:
\begin{equation*}
\sum_{S \in \cP} \osc(f, S) \vert S \vert = \sum_{S \in \cA} \osc(f, S) \vert S \vert + \sum_{S \in \cB} \osc(f, S) \vert S \vert < \frac{\epsilon}{2} + \frac{\epsilon}{2} = \epsilon.
\end{equation*}
This satisfies the Riemann integrability criterion. Thus, \(f\) is Riemann integrable on \(R\text{,}\) which completes the proof.