Suppose that $F,G \in BV[a, b]$, where $-\infty<a<b<\infty$. I want to prove that if there are no points on $[a,b]$ where $F$ and $G$ are both discontinuous, then $$\int_{[a,b]} F d G+\int_{[a,b]} G dF= F(b)G(b)-F(a-)G(a-)$$

I know that $$\int_{[a,b]}\frac{F(x)+F(x-)}{2}dG(x)+\int_{[a,b]} \frac{G(x)+G(x-)}{2}dF(x)=F(b)G(b)-F(a-)G(-b).$$

How I can use the hypothesis that there are no points in $[a,b]$ where $F$ and $G$ are both discontinuous??

  • 1
    $\begingroup$ Why don't you explain to the reader what NBV means? $\endgroup$
    – amsmath
    Aug 1, 2021 at 1:00
  • $\begingroup$ Why would one cares about the limit to $-\infty$ when the statement is about the interval $[a, b]$? @OliverDiaz $\endgroup$ Aug 4, 2021 at 13:30
  • $\begingroup$ @ArcticChar: I don't , since as you noted the result is of a local nature. Now, if both $F$ and $G$ were of total bounded variation over $\mathbb{R}$, then it would probably be desirable to normalize so that $G(-\infty)=\mu_G(\emptyset)=0=\mu_F(\emptyset)=F(-\infty)$, in which case the integration by parts formula takes the form \begin{align}\int_{\mathbb{R}}F(t)\mu_G(dt)&=F(\infty)G(\infty)-\int_{\mathbb{R}}G(t-)\mu_F(dt)\\&=\mu_F(\mathbb{R})\mu_G(\mathbb{R})-\int_{\mathbb{R}}G(t-)\mu_F(dt)\end{align} $\endgroup$ Aug 4, 2021 at 15:19
  • $\begingroup$ @OliverDiaz I have edited the post. I guess one could also change it to $-\infty \le a < b \le +\infty$ and keeps the N. $\endgroup$ Aug 4, 2021 at 17:11
  • $\begingroup$ @ArcticChar: under the assumtion that $F$ and $G$ are both of finite total variation over $\mathbb{R}$, that is correct. That is why the normalization is convenient; also this way we have $F(+\infty)=\mu_F(\mathbb{R})$ and $G(+\infty)=\mu_G(\mathbb{R})$, which is what one typically does in say, probability theory, when once discusses cumulative distributions. OF course, here $\mu_G$ and $\mu_F$ are in general signed measures (with local or total finite variation) $\endgroup$ Aug 4, 2021 at 17:17

1 Answer 1


Recall that any rights continuous function of finite variation $F$ on an interval $\alpha,\beta$ generates a unique (possibly signed) measure $\mu_F$ if local finite variation such that $\mu_F((a,b])=F(b)-F(a)$. This is the so called Lebesgue-Stieltjes measure associated to $F$, see for example Klenke, A. Probability theory, Universitext, Springer-Verlag, London 2008, pp 26-27.)

Theorem: Let $F$, $G$ be right--continuous functions of locally finite variation on an interval $I$ (bound or unbounded) Let $\mu_F$ and $\mu_G$ the Stieltjes-Lebesgue measures generated by $F$ and $G$ respectively. For any compact $[a,b]\subset I$ $$ \int_{(a,b]}F(t)\mu_G(dt)=F(b)G(b)-F(a)G(a)-\int_{(a,b]}G(t-)\mu_F(dt) $$ where $G(t-)=\lim_{s\nearrow t}G(s)$.

A proof can be obtained using Fubini's theorem \begin{aligned} F(b)-F(a))(G(b)-G(a))&=\int_{(a,b]\times(a,b]}\mu_F\otimes\mu_G(dt,ds)\\ &=\int_{(a,b]}\Big(\int_{(a,s]}\mu_F(dt)\Big)\mu_G(s) +\int_{(a,b]}\Big(\int_{(s,b]}\mu_F(dt)\Big)\mu_G(ds)\\ &=\int_{(a,b]}\Big(\int_{(a,s]}\mu_F(dt)\Big)\mu_G(s) +\int_{(a,b]}\Big(\int_{(a,t)}\mu_G(ds)\Big)\mu_F(dt)\\ &=\int_{(a,b]} F(s)-F(a)\mu_G(s) +\int_{(a,b]}G(t-)-G(a)\mu_F(dt)\\ \end{aligned} Algegraic simplifications yields the result in the Theorem.

If in addition $F$ and $G$ are continuous, then $F(a)=F(a-)$ and $G(a)=G(a-)$. Hence $\mu_F(\{a\})=\mu_G(\{a\})=0$, and we can substitute $(a,b]$ by $[a,b]$ in the integration.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.