How to proof the following equation, $$ \int_{0}^{x}\int_{0}^{s}d\ f_{+}^{'}(t)\ d\ s = \int_{0}^{x}(x-t)_{+}d\ f_{+}^{'}(t). $$ Here, $(x-b)_{+} = max{x-b, 0}$; $f_{+}^{'}$ denotes the right derivative of $f$.
1 Answer
I am guessing that $g_+$ means $\max\{g,0\}$. Let $\chi_E$ be the indicator function of a set $E$. If $P$ is the subset of $[0,x]$ on which $f'>0$, then $$\int_0^sdf'_+(t)=\int_0^x\chi_P(t)\ \chi_{[0,s]}(t)\ df'(t).$$ By Fubini's theorem, we get $$\int_0^x\int_0^s df_+'(t)\ ds=\int_0^x\int_0^x\chi_P(t)\ \chi_{[0,s]}(t)\ df'(t)\ ds=\int_0^x\int_0^x \chi_{[0,s]}(t)\ ds\ \chi_P(t)\ df'(t).$$ Therefore, $$\int_0^x\int_0^s df'_+(t)\ ds=\int_0^x\int_0^x\chi_{[t,x]}(s)\ ds\ \chi_P(t)\ df'(t)=\int_0^x\int_t^x ds\ \chi_P(t)\ df'(t).$$ Consequently, $$\int_0^x\int_0^s df'_+(t)\ ds=\int_0^x(x-t)\ \chi_P(t)\ df'(t)=\int_0^x(x-t)\ df'_+(t).$$ But as $x\geq t$ for all $t\in[0,x]$, $(x-t)=(x-t)_+$, and $$\int_0^x\int_0^s df'_+(t)\ ds=\int_0^x(x-t)_+\ df'_+(t).$$
Edit after OP's correction
If $P$ is the subset of $[0,x]$ on which $f'>0$, then $$\int_0^sdf'_+(t)=\int_0^x\chi_{[0,s]}(t)\ df'_+(t).$$ By Fubini's theorem, we get $$\int_0^x\int_0^s df_+'(t)\ ds=\int_0^x\int_0^x\chi_{[0,s]}(t)\ df_+'(t)\ ds=\int_0^x\int_0^x \chi_{[0,s]}(t)\ ds\ df_+'(t).$$ Therefore, $$\int_0^x\int_0^s df'_+(t)\ ds=\int_0^x\int_0^x\chi_{[t,x]}(s)\ ds\ df'_+(t)=\int_0^x\int_t^x ds\ df_+'(t).$$ Consequently, $$\int_0^x\int_0^s df'_+(t)\ ds=\int_0^x(x-t)\ df'_+(t).$$ But as $x\geq t$ for all $t\in[0,x]$, $(x-t)=(x-t)_+$, and $$\int_0^x\int_0^s df'_+(t)\ ds=\int_0^x(x-t)_+\ df'_+(t)=\int_0^\infty(x-t)_+\ df'_+(t).$$
-
$\begingroup$ Thank you for your reply. I am sorry for not stating the problem clearly. Here, $(x-b)_{+} = max{x-b, 0}$; $f_{+}^{'}$ denotes the right derivative of $f$. How should I proof with these definations? $\endgroup$– XWeiOct 25, 2018 at 13:20
-
$\begingroup$ @XinyuanWei I don't understand why you need $(x-t)_+$ when you integrate from $t=0$ to $t=x$. Shouldn't $(x-t)_+=(x-t)$ there? Is there something I missed? $\endgroup$– user593746Oct 25, 2018 at 13:22
-
$\begingroup$ @XinyuanWei But if you were writing $$\int_0^x\int_0^sdf'_+(t)\ ds=\int_0^{\color{red}\infty}(x-t)_+\ df'_+(t),$$ then everything makes sense. $\endgroup$– user593746Oct 25, 2018 at 13:27
-
$\begingroup$ I am also wondering why there is a positive part function. It is from Optimal Reinsurance under VaR and CVaR Risk Measures: A Simplified Approach, Lemma 3.1, in the following link papers.ssrn.com/sol3/papers.cfm?abstract_id=1578622 $\endgroup$– XWeiOct 25, 2018 at 13:30
-
$\begingroup$ The proof is basically unaltered. I just didn't have to use $\chi_P$ anymore (not that it was ever necessary to be there in the first proof I gave). $\endgroup$– user593746Oct 25, 2018 at 13:31