2

In this answer, it is said that

If $\int_{1}^{\infty}\vert f'(x)\rvert \,\mathrm{d}x$ converges, then $\sum_{k=1}^{\infty}f(k)$ and $\int_{1}^{\infty}f(x)\,\mathrm{d}x$ converge or diverge together.

In the proof, it says that

$$\sum_{k=2}^{\infty}\left | f(k)-\int_{k-1}^{k}f(x)\,\mathrm{d}x \right |<\infty$$ implies the convergence of $$\sum_{k=2}^{\infty} f(k)-\int_{1}^{\infty}f(x)\,\mathrm{d}x$$

Why does this implication hold? If think some conditions of $f$ is missing here.

Also, this idea brings me another question: Let $(a_n)$ and $(b_n)$ be two sequences of numbers. Is it true in general that if $\sum_{n=1}^{\infty}|a_n+b_n|$ converges, then $\sum_{n=1}^{\infty}a_n$ and $\sum_{n=1}^{\infty}b_n$ converge? It is clear if those sequences were positive, but what if at least of them were negative?

UnknownW
  • 2,036

2 Answers2

2

It was shown that we have (absolute) convergence of

$$\sum_{k=2}^\infty \left|f(k) - \int_{k-1}^k f(x) \, dx \right|$$

By the comparison test, this implies convergence of

$$\sum_{k=2}^\infty \left[f(k) - \int_{k-1}^k f(x) \, dx \right],$$

and there exists a finite limit $L$ such that

$$L = \sum_{k=2}^\infty \left[f(k) - \int_{k-1}^k f(x) \, dx \right] = \lim_{n \to \infty}\left[ \sum_{k=2}^n f(k) - \sum_{k=2}^n \int_{k-1}^k f(x) \, dx\right] \\ = \lim_{n \to \infty} \left[ \underbrace{\sum_{k=2}^n f(k)}_{A_n} - \underbrace{\int_{1}^n f(x) \, dx}_{B_n} \right]$$

If either one of $A_n$ or $B_n$ converge, then the other must converge. For example, if $A_n \to A$ as $n \to \infty$, then

$$\lim_{n\to \infty} B_n = \lim_{n\to \infty}[A_n - (A_n - B_n)] = \lim_{n\to \infty}A_n - \lim_{n\to \infty}(A_n - B_n) = A- L$$

If either one of $A_n$ and $B_n$ does not converge (diverges to $\pm \infty$ or the limit fails to exist), then the other must not converge.

For example, if the limit of $A_n$ fails to exist we have

$$\limsup_{n \to \infty} \,A_n \neq \liminf_{n \to \infty} \,A_n,$$

and

$$\begin{align} \limsup_{n \to \infty} \,B_n &= \limsup_{n \to \infty} \,\, [A_n - (A_n - B_n)] \\ &= \limsup_{n \to \infty} \, A_n - \limsup_{n \to \infty}\,\, (A_n - B_n) \\&= \limsup_{n \to \infty} \, A_n - L \\ &\neq \liminf_{n \to \infty} \,\, A_n - L\\ &= \liminf_{n \to \infty} \, A_n - \liminf_{n \to \infty} \,\,(A_n - B_n) \\ &= \liminf_{n \to \infty} \,\, [A_n - (A_n - B_n)] \\ &= \liminf_{n \to \infty} \, B_n \end{align} $$

RRL
  • 92,835
  • 7
  • 70
  • 142
  • Thank you so much. I have a question. Does "X and Y converge or diverge together" not mean that "X converges if and only if Y converges"? If yes, is it not enough to put as follows: Define $L_n:=A_n-B_n$. It is shown that $(L_n)$ converges to $L$. If $(A_n)$ converges, then $(L_n+B_n)$ converges, and so $(B_n)$ converges, because $B_n=(L_n+B_n)-L_n$. The other direction is similar, i.e. if $(B_n)$ converges, then $(A_n)$ converges. Then we are done. Here, I didn't mention about the divergence, like you mentioned in the last paragraph. Is my understanding wrong? – UnknownW Nov 16 '18 at 22:45
  • That's fine -- you are just restating my first argument $B_n = (L_n + B_n) - L_n \iff B_n = A_n - (A_n - B_n)$ where $L_n = A_n - B_n$. – RRL Nov 16 '18 at 22:52
  • That's a relief. Thanks for your time! – UnknownW Nov 16 '18 at 22:56
  • @UnknownW: You're welcome. Happy to help. – RRL Nov 16 '18 at 22:57
  • By the way, did you figure this test out yourself or is there a reference that explains about this test? I am just curious if I could find something more like this. – UnknownW Nov 16 '18 at 23:05
  • @UnknownW: I figured this out some time ago when I was trying to determine convergence of series like $\sum \frac{\sin n^a}{n^b}$ where the Dirichlet test fails. By estimating the difference $f(k) - \int_{k-1}^k f(x) , dx$ I could sometimes find a comparison that helped. I have never seen this in basic calculus or analysis textbooks which only discuss the elementary integral test for nonnegative, monotone functions. Later I did some research and found old papers that talked about extensions of the integral test including one by the famous mathematician GH Hardy who worked with Ramanujan. – RRL Nov 17 '18 at 02:03
  • It is not obvious that the convergence of $\int_0^{+\infty} f$ is equivalent to the convergence of the sequence $\int_0^n f$. – Clairon Nov 03 '23 at 14:19
1

Note that, by the additivity of the integral and the triangle inequality, \begin{align} \left| \sum_{k=2}^\infty f(k)- \int_0^1 f(x) \mathrm dx \right| &= \left| \sum_{k=2}^\infty f(k)- \sum_{k=2}^\infty \int_{k-1}^k f(x) \mathrm dx \right|\\ &= \left|\sum_{k=2}^\infty \left(f(k)- \int_{k-1}^k f(x) \mathrm dx\right)\right|\\ &\leq \sum_{k=2}^\infty \left|f(k)- \int_{k-1}^k f(x) \mathrm dx\right| < \infty. \end{align} For your second question, take $a_n = n = -b_n$. Then $ a_n + b_n = 0$, so $ \sum |a_n+b_n| = 0, $ but $\sum a_n = -\sum b_n = \sum n$ which is divergent.

MSDG
  • 7,345
  • Can you explain how/why it implies the conclusion, and is from the first to the second line justified? Thanks for the counter example ... – UnknownW Nov 15 '18 at 22:49