2

Let us say $f$ is an integrable function on $[a,b]$ and we want to evaluate $\int_a^b f(x)dx$ but often the calculation is not easy.So,we have a method of substitution.We substitute $x=\phi(t)$ where $\phi$ is a differentiable function on $[\alpha,\beta]$ such that $\phi(\alpha)=a$ and $\phi(\beta)=b$.Also $\phi'$ is integrable on $[\alpha,\beta]$ and $\phi'(x)\neq 0$ for all $x\in [\alpha,\beta]$.Then we can evaluate the above integral by $\int _a^b f(x)dx=\int_\alpha^\beta f(\phi(t))\phi'(t)dt$.

But I am a little but troubled with so many conditions,I could do the proof but I am having a hard time use the theorem to problems as I often forget the conditions required.So,can someone help me to get a better insight about this theorem of substitution in Riemann integrals?I would also like some counterexamples that show each condition to be essential.

2 Answers2

6

Strong sufficient conditions are that $f$ is continuous and $\phi'$ is integrable. A straightforward proof uses the FTC, and monotonicity of $\phi$ is not needed.

Defining $F(t) = \int_{\phi(\alpha)}^{\phi(t)}f(x) \, dx$, we have $F'(t) = f(\phi(t)) \phi'(t)$ since $f$ is continuous, and

$$\int_a^b f(x) \, dx = \int_{\phi(\alpha)}^{\phi(\beta)}f(x) \, dx = F(\beta)= \int_\alpha^\beta F'(t) \, dt = \int_\alpha^\beta f(\phi(t))\phi'(t) \, dt$$


On the other hand, we can drop the condition that $f$ is continuous and assume only integrability. To facilitate an easy proof using Riemann sums, we need to assume that $\phi$ is both continuously differentiable and monotone.

Take a partition $\alpha = t_0 < t_1 < \ldots < t_n = \beta$ and form the sum

$$\tag{*}S(P,f\circ\phi \, \phi')= \sum_{j=1}^n f(\phi(\xi_j))\phi'(\xi_j)(t_j - t_{j-1})$$

where we use intermediate points $\xi_j \in [t_{j-1},t_j]$ and which will converge to $\int_\alpha^\beta f(\phi(t)) \phi'(t) \, dt$ as the partition is refined.

If $\phi$ is increasing then a partition $P'$ of $[\phi(\alpha),\phi(\beta)]$ is induced by

$$\phi(\alpha) = \phi(t_0) < \phi(t_1) < \ldots < \phi(t_n) = \phi(\beta),$$

and using the intermediate points $\phi(\xi_j)$, we have a Riemann sum for the integral of $f$ over $[\phi(\alpha),\phi(\beta)]$ of the form

$$S(P',f) = \sum_{j=1}^n f(\phi(\xi_j))(\,\phi(t_j) - \phi(t_{j-1})\,)$$

Note that we need the monotonicity of $\phi$ to ensure that $\phi(\xi_j) \in [\phi(t_{j-1}), \phi(t_j)]$.

Applying the mean value theorem, there exist points $\eta_j \in (t_{j-1},t_j))$ such that

$$\tag{**}S(P',f) = \sum_{j=1}^n f(\phi(\xi_j))\phi'(\eta_j)(t_j - t_{j-1})$$

Notice the similarity between the sums in (*) and (**). Aside from the distinction between $\eta_j$ and $\xi_j$, they are identical. Using the continuity (and, hence, uniform continuity) of $\phi'$ we can show that as the partition is refined and both $\|P\|, \|P'\| \to 0$ we have

$$\lim_{\|P|| \to 0}|S(P,f\circ \phi\,\phi') - S(P',f)| = 0$$

Therefore, $S(P',f)$ converges to both integrals and we have

$$\lim_{\|P'\| \to 0}S(P',f) = \int_{\phi(\alpha)}^{\phi(\beta)} f(x) \, dx = \int_a^b f(\alpha(t)) \alpha'(t) \, dt$$

Again, there are a number of ways to prove the change-of-variables theorem -- without the assumption that $\phi$ is monotone -- that avoid this association with Riemann sums. In the most general form only integrability and not continuity of $f$ and $\phi'$ is assumed.


The conditions can be weakened further. The result holds if both $f$ and $\phi'$ are integrable, without any assumptions of continuity. This is much more difficult to prove. Here is where you might begin to search for counterexamples.

RRL
  • 90,707
1

Think of this as the fundamental theorem applied to a composition. By the chain rule it holds that $(f \circ \phi)'=f'(\phi) \circ \phi'$ so, roughly, $f \circ \phi=\int (f'(\phi) \circ \phi')$. The remainder conditions over the limits of integration are a result of changing variables $\phi (x)= t$.

astro
  • 649
  • No your answer is not correct.$f$ may be a function which is Riemann integrable but yet it may not have any antiderivatice $\int f$ ,so it is not always application of fundamental theorem...it is not that simple....For example See Volterra function....a function which is differentiable but its derivative is not Riemann integrable. – Kishalay Sarkar Jul 05 '20 at 14:37
  • Your method is useful if $f$ is continuous or if $f$ is an integrable function and $f$ has an antiderivative on $[a,b]$. – Kishalay Sarkar Jul 05 '20 at 14:38
  • He asked for a way to grab the idea, and I think this approach works for that purpose. If he wanted to add more patological behaviour he probably would have turned to a book and not a forum.Also, $f$ integrable is among his hypothesis. So your comment is out of place. – astro Jul 05 '20 at 14:39
  • I am studying real analysis.To me,grabbing the idea means knowing everything about a theorem....each condition....I did not ask for an intuitive way...of just thinking.... – Kishalay Sarkar Jul 05 '20 at 14:41
  • Then you shoouldn't have asked for a better insight but instead for a thoroug explanation. – astro Jul 05 '20 at 14:45
  • @astro The original poster explicitly included "$f$ is integrable" and not "$f$ is differentiable". Don't blame him that you don't know the answer. –  Jul 05 '20 at 15:23
  • @user762914 thanks...we can use tagged partitions to prove this theorem...we acutally do not need $f$ to be differentiable.Actually my main objective was to see that all the conditions are essential,none is redundant..and for that we require to eliminate the conditions one by one....keeping the others....It would be nice if someone did this. – Kishalay Sarkar Jul 05 '20 at 17:17
  • Whatever. To be honest, by reading at the original post all I get is he is having trouble to keep in mind how to apply the substitution theorem. All I was trying to do was giving him some mnemonics to help him keep the concept in mind at ease. The post is not clear about what he wants to know.. – astro Jul 06 '20 at 00:10
  • And also, in case you people can not read well, my post begins with ''think of this as...'' which clearly stands for some kind of idea of a matter. Exactly not at all I did say ''what happens is...'' or ''the theorem is...'' or anything that attempted to be formal. I really think this was rude and out of place. Specially since IT IS NOT AT ALL CLEAR what the author of the post wanted since his request is VERY poorly stated. Good bye. – astro Jul 06 '20 at 01:51