1

Let the function $f:[ 0,\infty ) \to \mathbb{R}$ be absolutely integrable. In other words, let

$$\lim_{r\to\infty}\int_0^r | f(x)|\, dx<\infty$$

Let $g:[ 0,\infty ) \to [ 0,\infty ) $ be a bijection from $\mathbb{R}_0^+$ to $\mathbb{R}_0^+$ that is arbitrary but piece-wise differentiable.

It is true that $$\lim_{r\to\infty}\int_0^r f(x) dx= \lim_{r\to\infty}\int_0^r f(g(x)) g'(x) dx$$ for all such $g$? If so, is there a book that proves this?

anonymous
  • 598

2 Answers2

0

The question can be rephrased. The composition of f with a piece-wise differentiable function, g, is piecewise differentiable. The derivative of a piece-wise differential function is also piece-wise continuous by definition. And the limits of integration make this an improper integral. So the question is, under these conditions, can we take the improper integral. We also have that one of the functions is the derivative if the internal member of a pair of composed functions.

Suppose we have F(u) where u=g(x). Then by the chain rule, we have

$F'(x)=F'(g(x))g'(x)$ if applicable in the case that g is piece-wise differentiable. And let's suppose $f(u)=F'(u)$

Then if we have $\int f(g(x))g'(x)dx=\int F'(g(x))g'(x)dx=\int f(u) du=F(g(x))+c$

The standard rules of improper integration apply.

So I think everything hangs on 2 questions. Does the chain rule apply in the case of f(g(x)) if g is only piecewise differentiable? Can you take the improper integral of the product of a piece-wise differentiable function and a piece-wise continuous function?

The usual theorem for proving the chain rule should apply in the first case. Piece-wise integration should be available in most books addressing the issue.

https://en.wikipedia.org/wiki/Subderivative has references that seem promising.

What is d/dx[f(g(x))] if g is piecewise continuous?

$\lim_{\Delta x->\infty} \frac{f(g(x+\Delta x)-f(g(x))}{\Delta x}=\lim_{\Delta x} \frac{f(g(x+\Delta x))-f(g(x))}{g(x+\Delta x)-g(x)}\frac{g(x+\Delta x)-g(x)}{\Delta x}$

We have that g is piecewise differentiable. This means that the limit of the right factor in the right equation doesn't exist at select points. The left factor has problems if f has any discontinuities.

I take this to mean that the chain rule is applicable on the sub intervals in which g is differentiable and f is continuous.

The absolute integrability of f need not imply continuity, correct? Means continuous except at countably many points as I recall.

From links below, it appears substiution is permitted within the sub intervals on which g is differentiable.

The behavior of g at those points of discontinuity of its derivative could inject some misbehavior.

TurlocTheRed
  • 6,458
  • Well, you're saying that $\int_{D1} f(x)dx=\int_{D2} f(g)dg$ because the usual substitution law says so for proper integrals. The main issue is the fact that the integrals at hand are improper, and so the conclusion does not necessarily transfer from the proper case. Further complication may arise from the fact that both improper integrals are limits of proper integrals over the same domain $[0,r]$, whilst in proper case domains $D1$ and $D2$ are different in general. So there are definitely some subtleties involved here that are not tackled yet. – anonymous Nov 29 '18 at 03:16
  • Are you familiar with left and right Riemann sums? I believe the associated principles still apply to both improper integrals and piece wise functions.

    The proof for the chain rule is a different beast. I think you'd have to go back to the limit definition of a derivative for that.

    – TurlocTheRed Nov 29 '18 at 03:41
  • The main issue is that one cannot easily generalize the Riemannian sum to improper integrals. That's partially the reason for the introduction of Lebesgue integration. Hence the requirement in my question that $f$ be absolutely integrable. – anonymous Nov 29 '18 at 03:58
  • https://math.stackexchange.com/questions/1744250/convergence-of-riemann-sums-for-improper-integrals gets part of the way there. – TurlocTheRed Nov 29 '18 at 04:00
  • Indeed, I may have to require monotonicity, but it may not appear necessary possibly due to the fact that absolute integrability condition may prove stronger and sufficient, although not necessary. – anonymous Nov 29 '18 at 04:03
  • The derivative of a piece wise differentiable function is piecewise continuous, right? I think that restricts to only having to consider intervals that have weird behaviors at the end points. – TurlocTheRed Nov 29 '18 at 04:09
  • Found this. Has books! https://www.encyclopediaofmath.org/index.php/Absolutely_integrable_function#Improper_integral – TurlocTheRed Nov 29 '18 at 04:20
  • Related: https://math.stackexchange.com/questions/2259750/substitution-rule-in-the-sense-of-lebesgue-integral – TurlocTheRed Nov 29 '18 at 04:28
  • Another relative: https://math.stackexchange.com/questions/247833/integration-by-substitution-for-lebesgue-integration – TurlocTheRed Nov 29 '18 at 04:29
0

It doesn't holds. By example if you consider

$$g(x):=\begin{cases}x^{-1},&x>0\\ 0,&x=0\end{cases}\tag1$$

as piece-wise differentiable then $g$ is a bijection in $[0,\infty)$ and choosing $f(x):=x^{-2}\chi_{[1,\infty)}(x)$ we find that

$$\int_0^\infty |f(x)|\, dx=\int_1^\infty x^{-2}\, dx=1\tag2$$

However

$$\int_0^\infty (f\circ g)(x)g'(x)\, dx=\int_1^\infty x^2\cdot \frac{-1}{x^2}\,dx=-\int_1^\infty\, dx=-\infty\tag3$$

The following function is bijective and piece-wise differentiable in a more standard sense

$$g(x):=\begin{cases}1+\tan (\pi x/2),& x\in[0,1)\\\tanh(x-1),&x\in[1,\infty)\end{cases}\tag4$$

But using the same $f$ as before we find that

$$\int_0^\infty (f\circ g)(x) g'(x)\, dx=-\coth(x)\bigg|_{x\to 0^+}^{x\to\infty}=\infty-1=\infty\tag5$$