Sums and definition of the integral
Definition (Riemann–Stieltjes sum)
Let \(f, \phi : [a,b] \to \mathbb{R}\) with \(f\) bounded. Given a partition \(\Gamma = \{a = x_{0} < x_{1} < \dots < x_{n} = b\}\) and intermediate points \(\xi_{i} \in [x_{i-1}, x_{i}]\) for \(1 \leq i \leq n\), the Riemann–Stieltjes sum of \(f\) with respect to \(\phi\) is
\[R(f, \Gamma, \phi) = \sum_{i=1}^{n} f(\xi_{i})\,\big[\phi(x_{i}) - \phi(x_{i-1})\big].\]Definition (Riemann–Stieltjes integral)
We say that \(f\) is Riemann–Stieltjes integrable with respect to \(\phi\) on \([a,b]\) if there exists \(I \in \mathbb{R}\) such that for every \(\varepsilon > 0\) there exists \(\delta > 0\) with
\[|\Gamma| < \delta \implies |R(f, \Gamma, \phi) - I| < \varepsilon\]| for any choice of intermediate points \(\xi_{1}, \dots, \xi_{n}\), where $$ | \Gamma | = \max_{i}(x_{i} - x_{i-1})\(. In that case we write\)I = \int_{a}^{b} f \, d\phi$$. |
Exercise (Cauchy criterion for Riemann–Stieltjes integrability)
Prove that \(f\) is Riemann–Stieltjes integrable with respect to \(\phi\) if and only if for every \(\varepsilon > 0\) there exists \(\delta > 0\) such that
\[|\Gamma| < \delta \ \text{ and }\ |\Gamma'| < \delta \implies |R(f, \Gamma, \phi) - R(f, \Gamma', \phi)| < \varepsilon\]for any choices of intermediate points.
Lemma (A common discontinuity prevents integrability)
Let \(\phi : [a,b] \to \mathbb{R}\) and let \(f : [a,b] \to \mathbb{R}\) be bounded. If there exists \(z_{0} \in (a,b)\) at which both \(f\) and \(\phi\) are discontinuous, then \(f\) is not Riemann–Stieltjes integrable with respect to \(\phi\).
| Proof: We show that the Cauchy criterion fails. Since \(f\) and \(\phi\) are discontinuous at \(z_{0}\), there exists \(\varepsilon > 0\) such that, for every \(\delta > 0\), there is a point \(w_{\delta}\) with $$0 < | w_{\delta} - z_{0} | < \delta\(and\) | \phi(w_{\delta}) - \phi(z_{0}) | \geq \sqrt{\varepsilon}\(, and a point\)\xi_{\delta}\(with\) | \xi_{\delta} - z_{0} | < | w_{\delta} - z_{0} | \(and\) | f(\xi_{\delta}) - f(z_{0}) | \geq \sqrt{\varepsilon}\(; in particular\)\xi_{\delta}\(lies strictly between\)z_{0}\(and\)w_{\delta}\(. Take a partition\)\Gamma\(with\) | \Gamma | < \delta\(having\)z_{0}\(and\)w_{\delta}\(as consecutive nodes\)x_{i_{0}-1} = z_{0}\(,\)x_{i_{0}} = w_{\delta}\(, so that\)\xi_{\delta} \in (x_{i_{0}-1}, x_{i_{0}})\(. Construct two Riemann–Stieltjes sums with the same\)\Gamma\(and the same intermediate points except on the subinterval\)[x_{i_{0}-1}, x_{i_{0}}]\(, where one takes\)\xi_{i_{0}} = \xi_{\delta}\(and the other\)\xi_{i_{0}} = z_{0}$$. Then |
Thus, for every \(\delta > 0\) there are sums with norm smaller than \(\delta\) that differ by at least \(\varepsilon\), and the Cauchy criterion fails. The case in which the witnesses \(w_{\delta}\) (for \(\phi\)) and \(\xi_{\delta}\) (for \(f\)) cannot be taken on the same side of \(z_{0}\) is handled analogously and is left as an exercise.
Upper and lower sums
Definition (Lower and upper Darboux–Stieltjes sums)
Let \(f : [a,b] \to \mathbb{R}\) be bounded and \(\Gamma = \{a = x_{0} < \dots < x_{n} = b\}\). For \(1 \leq i \leq n\) let
\[m_{i} = \inf_{x_{i-1} \leq \xi \leq x_{i}} f(\xi), \qquad M_{i} = \sup_{x_{i-1} \leq \xi \leq x_{i}} f(\xi).\]The lower and upper sums with respect to \(\phi\) are
\[L(f, \Gamma, \phi) = \sum_{i=1}^{n} m_{i}\,\big[\phi(x_{i}) - \phi(x_{i-1})\big], \qquad U(f, \Gamma, \phi) = \sum_{i=1}^{n} M_{i}\,\big[\phi(x_{i}) - \phi(x_{i-1})\big].\]Note (Sandwiching of the sums and the Riemann case)
If \(\phi\) is increasing, then \(\phi(x_{i}) - \phi(x_{i-1}) \geq 0\) and for every choice of intermediate points
\[L(f, \Gamma, \phi) \leq R(f, \Gamma, \phi) \leq U(f, \Gamma, \phi).\]If \(\phi(x) = x\) we recover the usual Riemann sums and the usual Riemann integral.
Lemma (Monotonicity properties of the Darboux–Stieltjes sums)
Let \(f : [a,b] \to \mathbb{R}\) be bounded and \(\phi : [a,b] \to \mathbb{R}\) increasing.
- If \(\Gamma_{1} \subseteq \Gamma_{2}\) (refinement), then \(L(f, \Gamma_{1}, \phi) \leq L(f, \Gamma_{2}, \phi)\) and \(U(f, \Gamma_{2}, \phi) \leq U(f, \Gamma_{1}, \phi)\).
- For any partitions \(\Gamma_{1}, \Gamma_{2}\), \(\ L(f, \Gamma_{1}, \phi) \leq U(f, \Gamma_{2}, \phi)\).
Proof: (1). It suffices to insert a point \(y\) into a subinterval \([x_{i-1}, x_{i}]\) of \(\Gamma_{1}\). Since
\[\sup_{[x_{i-1}, y]} f, \ \sup_{[y, x_{i}]} f \ \leq\ \sup_{[x_{i-1}, x_{i}]} f,\]and \(\phi(y) - \phi(x_{i-1}) \geq 0\), \(\phi(x_{i}) - \phi(y) \geq 0\) with sum \(\phi(x_{i}) - \phi(x_{i-1})\), we obtain
\[\big(\textstyle\sup_{[x_{i-1}, y]} f\big)\big(\phi(y) - \phi(x_{i-1})\big) + \big(\textstyle\sup_{[y, x_{i}]} f\big)\big(\phi(x_{i}) - \phi(y)\big) \leq \big(\textstyle\sup_{[x_{i-1}, x_{i}]} f\big)\big(\phi(x_{i}) - \phi(x_{i-1})\big),\]that is, refinement can only decrease \(U\). Iterating over the added points, \(U(f, \Gamma_{2}, \phi) \leq U(f, \Gamma_{1}, \phi)\). The argument for \(L\) (with infima) is symmetric and gives \(L(f, \Gamma_{1}, \phi) \leq L(f, \Gamma_{2}, \phi)\).
(2). Let \(\Gamma = \Gamma_{1} \cup \Gamma_{2}\), a common refinement. By (1) and the previous note,
\[L(f, \Gamma_{1}, \phi) \leq L(f, \Gamma, \phi) \leq U(f, \Gamma, \phi) \leq U(f, \Gamma_{2}, \phi).\]Existence of the integral
Note (Reduction to an increasing integrator)
From the definition one checks that if \(\int_{a}^{b} f\,d\phi_{1}\) and \(\int_{a}^{b} f\,d\phi_{2}\) exist and \(\phi = \phi_{1} - \phi_{2}\), then \(\int_{a}^{b} f\,d\phi\) exists and
\[\int_{a}^{b} f\,d\phi = \int_{a}^{b} f\,d\phi_{1} - \int_{a}^{b} f\,d\phi_{2}.\]Since every function of bounded variation is a difference of two increasing functions (Jordan’s characterisation), the study of integrability with respect to an integrator of bounded variation reduces to the case in which \(\phi\) is increasing.
Theorem (Existence of the integral for a continuous integrand and an integrator of bounded variation)
Let \(f : [a,b] \to \mathbb{R}\) be continuous and \(\phi : [a,b] \to \mathbb{R}\) of bounded variation. Then \(\int_{a}^{b} f\,d\phi\) exists and
\[\left| \int_{a}^{b} f\,d\phi \right| \leq \Big(\sup_{[a,b]} |f|\Big)\,\operatorname{Var}(\phi, [a,b]).\]Proof: By the previous note it suffices to prove existence when \(\phi\) is increasing; if \(\phi\) is constant every sum vanishes and the integral equals \(0\), so we assume \(\phi(b) > \phi(a)\). Let \(\varepsilon > 0\). Since \(f\) is uniformly continuous, there exists \(\delta_{1} > 0\) such that
\[|x - y| < \delta_{1} \implies |f(x) - f(y)| < \frac{\varepsilon}{2\big(\phi(b) - \phi(a)\big)}.\]| The upper and lower sums approach each other. If $$ | \Gamma | < \delta_{1}\(, by continuity there exist\)\xi_{i}, \eta_{i} \in [x_{i-1}, x_{i}]\(with\)f(\xi_{i}) = M_{i}\(,\)f(\eta_{i}) = m_{i}\(; since\) | \xi_{i} - \eta_{i} | \leq | \Gamma | < \delta_{1}$$, |
| Stability of \(U\). If $$ | \Gamma | < \delta_{1}\(and\) | \Gamma’ | < \delta_{1}\(, using\)U - L \leq \varepsilon/2\(and\)L(f, \cdot, \phi) \leq U(f, \cdot, \phi)$$ between distinct partitions, |
| and symmetrically, so that $$ | U(f, \Gamma, \phi) - U(f, \Gamma’, \phi) | \leq \varepsilon/2$$. |
| Construction of the limit. Take partitions \(\{\Gamma_{k}\}_{k=1}^{\infty}\) with $$ | \Gamma_{k} | \to 0\(and let\)\Gamma_{k}’ = \bigcup_{j=1}^{k} \Gamma_{j}\(, so that\)\Gamma_{k}’ \subseteq \Gamma_{k+1}’\(and\)\Gamma_{k} \subseteq \Gamma_{k}’\(. By monotonicity (1),\){U(f, \Gamma_{k}’, \phi)}_{k}$$ is decreasing and bounded below (by any lower sum), so that |
| exists. Given \(\varepsilon\), let \(k_{0}\) be such that \(0 \leq U(f, \Gamma_{k}', \phi) - U < \varepsilon/2\) for \(k \geq k_{0}\), and \(k_{1}\) such that $$ | \Gamma_{k} | < \delta_{1}\(for\)k \geq k_{1}\(. Since\)\Gamma_{k} \subseteq \Gamma_{k}’\(and both have norm\)< \delta_{1}\(, the stability of\)U\(gives\) | U(f, \Gamma_{k}, \phi) - U(f, \Gamma_{k}’, \phi) | \leq \varepsilon/2\(, hence for\)k \geq \max{k_{0}, k_{1}}$$, |
| Independence of the sequence. If \(\{\widetilde{\Gamma}_{k}\}\) is another sequence with $$ | \widetilde{\Gamma}_{k} | \to 0\(, for large\)k\(both norms are\)< \delta_{1}\(and the stability of\)U\(gives\) | U(f, \Gamma_{k}, \phi) - U(f, \widetilde{\Gamma}_{k}, \phi) | \leq \varepsilon/2\(; hence\)U(f, \widetilde{\Gamma}_{k}, \phi) \to U\(as well. Thus,\)U(f, \Gamma, \phi) \to U\(as\) | \Gamma | \to 0$$. |
| Conclusion. It has already been shown that \(U(f, \Gamma, \phi) \to U\) as $$ | \Gamma | \to 0\(. Moreover, for\) | \Gamma | < \delta_{1}\(we have that\)L(f, \Gamma, \phi) \geq U(f, \Gamma, \phi) - \tfrac{\varepsilon}{2}\(, so that also\)L(f, \Gamma, \phi) \to U\(. The sandwiching\)L(f, \Gamma, \phi) \leq R(f, \Gamma, \phi) \leq U(f, \Gamma, \phi)\(then gives\)R(f, \Gamma, \phi) \to U\(as\) | \Gamma | \to 0\(. Hence\)\int_{a}^{b} f\,d\phi = U\(exists. Finally, for\)\phi\(of bounded variation and any\)\Gamma$$, |
and passing to the limit we obtain the stated bound.
Properties: linearity, mean value and integration by parts
Theorem (Linearity of the Riemann–Stieltjes integral)
Let \(f_{1}, f_{2} : [a,b] \to \mathbb{R}\) be Riemann–Stieltjes integrable with respect to \(\phi_{1}, \phi_{2} : [a,b] \to \mathbb{R}\), \(c \in \mathbb{R}\) and \(a < c' < b\). Then:
- \(\displaystyle \int_{a}^{b}(c f_{1} + f_{2})\,d\phi_{1} = c\int_{a}^{b} f_{1}\,d\phi_{1} + \int_{a}^{b} f_{2}\,d\phi_{1}\);
- \(\displaystyle \int_{a}^{b} f_{1}\,d(c\phi_{1}) = c\int_{a}^{b} f_{1}\,d\phi_{1}\);
- \(\displaystyle \int_{a}^{b} f_{1}\,d(\phi_{1} + \phi_{2}) = \int_{a}^{b} f_{1}\,d\phi_{1} + \int_{a}^{b} f_{1}\,d\phi_{2}\);
- \(\displaystyle \int_{a}^{b} f_{1}\,d\phi_{1} = \int_{a}^{c'} f_{1}\,d\phi_{1} + \int_{c'}^{b} f_{1}\,d\phi_{1}\).
Proof: Exercise.
Note (Bounds by the supremum and the infimum of the integrand)
If \(\phi\) is increasing, from \(L(f, \Gamma, \phi) \leq R(f, \Gamma, \phi) \leq U(f, \Gamma, \phi)\), \(U(f, \Gamma, \phi) \leq (\phi(b) - \phi(a))\sup_{[a,b]} f\) and \(L(f, \Gamma, \phi) \geq (\phi(b) - \phi(a))\inf_{[a,b]} f\) we obtain, passing to the limit,
\[\Big(\inf_{[a,b]} f\Big)\big(\phi(b) - \phi(a)\big) \leq \int_{a}^{b} f\,d\phi \leq \Big(\sup_{[a,b]} f\Big)\big(\phi(b) - \phi(a)\big).\]Lemma (Mean value theorem for the Riemann–Stieltjes integral)
Let \(f : [a,b] \to \mathbb{R}\) be continuous and \(\phi : [a,b] \to \mathbb{R}\) increasing, with \(f\) Riemann–Stieltjes integrable with respect to \(\phi\). Then there exists \(\xi \in [a,b]\) such that
\[\int_{a}^{b} f\,d\phi = f(\xi)\big(\phi(b) - \phi(a)\big).\]Proof: If \(\phi(b) = \phi(a)\), the bounds of the previous note give \(\int_{a}^{b} f\,d\phi = 0\) and any \(\xi\) will do. If \(\phi(b) > \phi(a)\), dividing the same bounds by \(\phi(b) - \phi(a)\),
\[\inf_{[a,b]} f \ \leq\ \frac{1}{\phi(b) - \phi(a)}\int_{a}^{b} f\,d\phi \ \leq\ \sup_{[a,b]} f.\]Since \(f\) is continuous on the compact set \([a,b]\), it attains its infimum and its supremum, and by the intermediate value theorem it takes every value between them; in particular there exists \(\xi \in [a,b]\) at which \(f(\xi)\) equals the quotient above, which gives the identity.
Theorem (Integration by parts)
If \(\int_{a}^{b} f\,d\phi\) exists, then \(\int_{a}^{b} \phi\,df\) exists and
\[\int_{a}^{b} f\,d\phi + \int_{a}^{b} \phi\,df = f(b)\phi(b) - f(a)\phi(a).\]Proof: Let \(\Gamma = \{a = x_{0} < \dots < x_{n} = b\}\) with intermediate points \(\xi_{i} \in [x_{i-1}, x_{i}]\). Consider the points \(\xi_{0} = a\), \(\xi_{n+1} = b\) and the partition \(\Gamma' = \{\xi_{0}, \xi_{1}, \dots, \xi_{n+1}\}\), which satisfies \(x_{i-1} \leq \xi_{i} \leq x_{i} \leq \xi_{i+1}\), so that each \(x_{i}\) is a valid intermediate point of \(\Gamma'\). By summation by parts (Abel),
\[R(f, \Gamma, \phi) = \sum_{i=1}^{n} f(\xi_{i})\big(\phi(x_{i}) - \phi(x_{i-1})\big) = -\sum_{i=0}^{n} \phi(x_{i})\big(f(\xi_{i+1}) - f(\xi_{i})\big) + f(b)\phi(b) - f(a)\phi(a),\]| that is, \(R(f, \Gamma, \phi) = -R(\phi, \Gamma', f) + f(b)\phi(b) - f(a)\phi(a)\), where \(R(\phi, \Gamma', f)\) is a Riemann–Stieltjes sum of \(\phi\) with respect to \(f\) with partition \(\Gamma'\) and intermediate points \(x_{i}\). Since \(\xi_{i} \in [x_{i-1}, x_{i}]\), each gap satisfies $$\xi_{i+1} - \xi_{i} \leq x_{i+1} - x_{i-1} \leq 2 | \Gamma | \(, so that\) | \Gamma’ | \leq 2 | \Gamma | \to 0\(. Since\)\int_{a}^{b} f\,d\phi\(exists,\)R(f, \Gamma, \phi) \to \int_{a}^{b} f\,d\phi\(as\) | \Gamma | \to 0\(, and the algebraic identity above forces\)R(\phi, \Gamma’, f) \to f(b)\phi(b) - f(a)\phi(a) - \int_{a}^{b} f\,d\phi\(. Since\) | \Gamma’ | \to 0\(and this limit does not depend on the chosen sequence of partitions,\)\int_{a}^{b} \phi\,df\(exists and equals\)f(b)\phi(b) - f(a)\phi(a) - \int_{a}^{b} f\,d\phi$$, which is the desired identity. |
Passing to the limit under the integral sign
Example (An increasing sequence of Riemann-integrable functions with a non-integrable limit)
Let \(\{r_{n}\}_{n=1}^{\infty}\) be an enumeration of \(\mathbb{Q} \cap [0,1]\) and
\[f_{n} : [0,1] \to \mathbb{R}, \qquad f_{n}(x) = \begin{cases} 1, & x \in \{r_{1}, \dots, r_{n}\},\\ 0, & \text{otherwise.}\end{cases}\]Then \(f_{n} \leq f_{n+1}\) and \(f_{n} \to f = \mathbf{1}_{\mathbb{Q} \cap [0,1]}\) pointwise. It is an exercise to check that \(\int_{0}^{1} f_{n}(x)\,dx = 0\) for every \(n\), whereas \(f\) is not Riemann integrable. Thus, the pointwise limit of Riemann-integrable functions need not be one, which motivates a more flexible notion of integral.
Lemma (Passing to the limit under uniform convergence)
Let \(\phi : [a,b] \to \mathbb{R}\) be of bounded variation and \(\{f_{n}\}_{n=1}^{\infty}\) a sequence of functions Riemann–Stieltjes integrable with respect to \(\phi\). If \(f_{n} \to f\) uniformly on \([a,b]\) and \(f\) is Riemann–Stieltjes integrable with respect to \(\phi\), then
\[\lim_{n \to \infty} \int_{a}^{b} f_{n}\,d\phi = \int_{a}^{b} f\,d\phi.\]Proof: For any function \(g\) Riemann–Stieltjes integrable with respect to \(\phi\) and every partition \(\Gamma\),
\[|R(g, \Gamma, \phi)| \leq \Big(\sup_{[a,b]}|g|\Big)\sum_{i}|\phi(x_{i}) - \phi(x_{i-1})| \leq \Big(\sup_{[a,b]}|g|\Big)\operatorname{Var}(\phi, [a,b]),\]| and passing to the limit, $$\big | \int_{a}^{b} g\,d\phi\big | \leq (\sup_{[a,b]} | g | )\operatorname{Var}(\phi, [a,b])\(. Applying this to\)g = f_{n} - f\((integrable, since\)f_{n}\(and\)f$$ are) and using linearity, |
| Since \(f_{n} \to f\) uniformly, $$\sup_{[a,b]} | f_{n} - f | \to 0$$, and the right-hand side tends to zero. |