Once I was working on simple integrals, and I decided to break the system by counting the integral. Let $f(x)=xe^x$. Then if we take not $\{x=u;\ e^xdx=dv\}$ but $\{xdx=dv;\ e^x=u\}$. As soon as I did so, the primitive took the form of $$ \int xe^xdx = \sum\limits_{n=2}^{+\infty}(-1)^n\frac{x^n}{n!}e^x + C $$ Can this be considered the right decision?
-
ah yes. thx for this – Арес Аресович Jun 17 '21 at 22:08
-
This might be a language/translation problem, but it's not clear what you mean by "counting the integral." (It seems clear, though, that you decided to see what would happen if you integrated by parts the "wrong" way, and just kept at until you wound up with an infinite series.) – Barry Cipra Jun 17 '21 at 22:19
-
Your term "the right decision" should be "a correct solution". When solving problems in math, решение = solution, not decision. I have no idea what exactly "counting" is supposed to mean. Что вы имели в виду на русском? – KCd Jun 17 '21 at 22:40
-
Your primitive is elementary. Being elementary is the property of a function having a particular type of representation. Writing it in a way that does not appear to fit the condition of being elementary does not make the function non-elementary. For comparison, we don't say $3$ is irrational from the equation $3 = \sqrt{2} + (3 - \sqrt{2})$ where some terms are irrational. – KCd Jun 17 '21 at 22:43
1 Answers
Your answer is:
$$\begin{align}e^x\sum_{n=2}^\infty(-1)^n\frac{x^n}{n!}&=e^x\left(e^{-x}-(1-x)\right)\\ &=1+(x-1)e^x\end{align}$$
The usual integration by parts gives you:
$$e^x(x-1)$$
These two answers differ by a constant, so they can be (and are) both correct.
Technically, this is a risky approach, because what you really get, inductively, is, for any $N$:
$$\begin{align}\int xe^x\,dx &=e^x\sum_{n=2}^N(-1)^n\frac{x^n}{n!} +(-1)^{N+1}\frac{1}{N!}\int x^{N}e^x\,dx\end{align}$$
Now you need to know what it means for $$\frac1{N!}\int x^{N}e^x\,dx$$ to converge to zero, since the integral is an indefinite integral.
It works if you replace all the integrals with: $$I_N(x)=\int_{0}^x t^Ne^{t}\,dt$$
But you still need to know $\frac1{N!}I_N(x)\to 0$ for all $x.$
For any $x\geq 0$ $$0<I_N(x)=\int_0^x t^Ne^t\,dt<x^N\int_0^x e^t\,dt=x^N(e^x-1)$$ So we know $\frac1{N!}I_N(x)\to 0$ as $N\to\infty.$
When $x<0$ then $$|I_N(x)|\leq\int_0^{|x|}|x|^{N}=\frac{|x|^{N+1}}{N+1}$$
So again $\frac{1}{N!}I_N(x)\to 0.$
More generally, if $f_1(x),f_2(x),\dots$ have the property that $f_{n+1}’(x)=f_n(x)$ for all $n$ then:
$$\int f_1(x)e^{-x}\,dx=e^{-x}\sum_{n=2}^N f_n(x) + \int f_N(x) e^{-x}\,dx$$ (I’ve switched to $e^{-x}$ here to remove the alternating signs, but you can get them back easily.)
But if $f_1(x)=\cos x$ you don’t get $$I_N(x)=\int f_N(x)e^{-x}\,dx\to 0.$$
Amusingly, if $f_1(x)=\cos \alpha x$, with $\alpha>1,$ you get $I_n(x)\to 0,$ so you get an infinite series.
$$e^{-x}\sum_{n=1}^{\infty}(-1)^n\alpha^{-2n}\left(\alpha\sin \alpha x-\cos\alpha x\right)\\=\frac{\alpha \sin \alpha x -\cos \alpha x}{1+\alpha^2}e^{-x}$$
The right side is the correct value for any $\alpha,$ but the infinite series doesn’t converge for $|\alpha|\leq 1.$

- 177,126
-
Good answer! This result might be worth mentioning to affirm that the resulting limit, when it exists (only at a single $x$ is required), is indeed a primitive of the original function: https://math.stackexchange.com/a/1194164/355036up – jawheele Jun 17 '21 at 23:58