I think I remember this from either Apostol's or Rudin's principles of Analysis.

Claim: Let $\{x_n\}_{n=1}^{\infty}$ be a sequence of positive real numbers. Then $$ \limsup \frac{x_1+x_2+...+x_n+x_{n+1}}{x_n} \geq 4.$$


  1. Yes, the $4$ is attained by $x_n=2^n$ sequence.
  2. Try $x_n=a^n$, and this holds.
  3. I have already proved that UNLESS $\lim x_n =\infty $ we will have $$ \limsup \frac{x_1+x_2+...+x_n+x_{n+1}}{x_n} = \infty.$$ So, the interesting case is when we do have a sequence going to infinity, and it is now a matter of proving that something is invariant no matter how (exponentially, linearly, polynomially, a mix of them, etc.) this convergence to infinity occurs.
  • $\begingroup$ At the very least, we have \begin{align*}\limsup\frac{x_{1}+\cdots+x_{n+1}}{x_{n}} &=1+\limsup\frac{x_{1}+\cdots+x_{n-1}}{x_{n}}+\frac{x_{n+1}}{x_{n}} \\ &\geq2+\limsup\frac{x_{1}+\cdots+x_{n-1}}{x_{n}}\end{align*} in the divergent case. $\endgroup$
    – parsiad
    Jul 19 '16 at 1:03
  • $\begingroup$ Let me tell you something more interesting! Assume $x_{n+1}/x_n$ converges to $L$. Then $\lim \sup \frac{x_{1}+\cdots+x_{n+1}}{x_{n}} \geq \lim \frac{x_{n-1}}{x_n}+1+\lim \frac{x_{n+1}}{‌​x_{n}} \geq 1+L+\frac{1}{L} \geq 3.$ But still a $1$ is missing! $\endgroup$ Jul 19 '16 at 1:15
  • $\begingroup$ No. Even then, remember that I assumed the convergence of the fraction, which is NOT guaranteed. $\endgroup$ Jul 19 '16 at 1:24
  • 5
    $\begingroup$ This appears also as Problem 2.4.31 in Volume 1 of Kaczor, Nowak: Problems in Mathematical Analysis. It is given on page 47 with solution on page 207. $\endgroup$ Jul 19 '16 at 7:48

Solution(as in my lecture)
Suppose that for a positive constant $c$ and a number $N$ $$ \frac{a_1+\dots a_{n+1}}{a_n} \le c \quad \text{ for all }n \ge N $$ Let $b_n:= \sum_{i=1}^n a_n$, the inequality above is equivalent to $$b_{n+1} \le c(b_{n}-b_{n-1}) \text{ for all }n \ge N$$ This form of the inequality is more or less more easier to deal with, but let's try it further, by letting $c_{n}$ denote $\frac{b_n}{b_{n-1}}$,then $$c_{n+1}+\frac{c}{c_n} \le c$$ Thus $(c_n , n \ge N)$ is a sequence of bounded real numbers, thus we can choose a subsequence $(c_{n_k},k\ge 1)$ such that $$\lim_{k \rightarrow \infty} c_{n_k}= \underbrace{\limsup_{n \rightarrow \infty } c_n}_{=:a}$$ Thus $$c \ge \limsup_{k \rightarrow \infty} \left(c_{n_k}+\frac{c}{c_{n_k-1}} \right)=a+\limsup_{k \rightarrow \infty} \frac{c}{c_{n_k-1}} \ge a+ \frac{c}{ \limsup c_n} =a+\frac{c}{a} \ge 2\sqrt{c} $$ Or, $c$ must be bigger than $4$. This means$$\text{ if} c \ge \limsup \frac{ a_1+\dots+a_{n+1}}{a_n} \text{ then } c \ge 4$$ Hence the conclusion. $\square$

A small story: This question brings back lots of memories. The first time I encounter it was when solving the famous analysis problem books by Kaczor and Nowak. A very long time ago. Then after that, in a small tutor class with gifted highschool students, I told them that it was important to find a good presentation of your problem.
And this question was the example I used in that class.

  • 1
    $\begingroup$ Very nice! There is a similar approach on AoPS (artofproblemsolving.com/community/c7h1247405p6417340), but your's is simpler. $\endgroup$
    – Martin R
    Oct 28 at 7:53
  • $\begingroup$ Thank you, Martin $\endgroup$ Oct 28 at 7:56
  • $\begingroup$ Your argument about the sequence $(c_n)$ can be generalized in the following way: Let $(x_n)$ be a sequence of positive real numbers, $A, B > 0$, and $P$ a permutation of the positive integers. If $x_n + \frac{A}{x_{P_n}} \le B$ for all $n$ then $B^2 \ge 4A$. I wonder if that is of any use ... $\endgroup$
    – Martin R
    Oct 28 at 8:14
  • 1
    $\begingroup$ You mean $ \limsup \frac{1}{c_{n_k-1}} \ge \frac{1}{ \limsup c_n}$ ? $\endgroup$ Oct 28 at 17:14
  • 1
    $\begingroup$ Thanks a lot :). I would like to add to my previous comment that the equality is valid only for positive $c_{n_{k-1}}$'s. +1 $\endgroup$
    – Koro
    Oct 28 at 17:18

First, assume that $\lim\sup \frac{x_{1}+\cdots+x_{n+1}}{x_{n}}=L>0$ is finite. Define a sequence of polynomial $\{p_{n}(x)\}$ by $p_{0}(x)=1, p_{1}(x)=x-1$ and \begin{align} p_{n+1}(x)=(x-1)p_{n}(x)-p_{n-1}(x)-\cdots-p_{1}(x)-p_{0}(x).\,\,(n\geq 1) \end{align}

Claim. $p_{n}(L)>0$ for all $n\geq 1$.

proof. We have $x_{1}+x_{2}+\cdots+x_{n+1}\leq Lx_{n}$ for all $n\geq 1$. (Edited : Actually, we only have this for sufficiently large $n$. However, this can be modified to all $n$ with a user254665's comment.) By induction, we will prove the following : \begin{align} x_{n+1}\leq \frac{p_{n}(L)}{p_{n-1}(L)}x_{n} \end{align} (If we prove this inequality, then claim immediately follows from $x_{n+1}, x_{n}>0$.) Since $x_{1}+x_{2}\leq Lx_{1}$, it holds for $n=1$. If it holds for $n\leq N-1$, then \begin{align} x_{N+1}&\leq (L-1)x_{N}-x_{N-1}-x_{N-2}-\cdots-x_{1} \\ &\leq \left(L-1-\frac{p_{N-2}(L)}{p_{N-1}(L)}-\frac{p_{N-2}(L)}{p_{N-1}(L)}\frac{p_{N-3}(L)}{p_{N-2}(L)}-\cdots-\frac{p_{N-2}(L)}{p_{N-1}(L)}\frac{p_{N-3}(L)}{p_{N-2}(L)}\cdots \frac{p_{0}(L)}{p_{1}(L)}\right)x_{N} \\ &=\frac{p_{N}(L)}{p_{N-1}(L)}x_{N}. \end{align}

Now using the recurrence relation, we can prove $p_{n+1}(x)=x(p_{n}(x)-p_{n-1}(x))$ by subtracting two recurrence relations for $n$ and $n-1$. Then by using characteristic polynomial method, we get \begin{align} p_{n}(x)=\frac{\sqrt{x^{2}-4x}+x-2}{2\sqrt{x^{2}-4x}}\left(\frac{x+\sqrt{x^{2}-4x}}{2}\right)^{n}+\frac{\sqrt{x^{2}-4x}-x+2}{2\sqrt{x^{2}-4x}}\left(\frac{x-\sqrt{x^{2}-4x}}{2}\right)^{n} \end{align} for $x\neq 4$.

Now, suppose $0<L<4$. Let \begin{align} \frac{L+\sqrt{L^{2}-4L}}{2}=\frac{L+i\sqrt{4L-L^{2}}}{2}=\sqrt{L}e^{i\theta} \end{align} where $\cos \theta=\frac{\sqrt{L}}{2}<1$. By direct computation, we have \begin{align} p_{n}(L)=L^{n/2}\left(\cos n\theta+\frac{L-2}{\sqrt{4L-L^{2}}}\sin n\theta\right)=\frac{2L^{n/2}}{\sqrt{4L-L^{2}}} \sin (n\theta+\alpha) \end{align} where $\cos\alpha=\frac{L-2}{2}$. Since $0<\theta<\pi/2$, we can choose $n$ s.t. $\sin (n\theta+\alpha)<0\Rightarrow p_{n}(L)<0$. Contradiction. So we have $L\geq 4$.

There is an easier proof which is discovered by my friend. Let $S_{n}=x_{1}+\cdots + x_{n}$ and $t_{n}=S_{n}/S_{n+1}$. Then $0<t_{n}<1$. Suppose $\limsup\frac{x_{1}+\cdots+x_{n+1}}{x_{n}}=L<4$. Then we have \begin{align} \frac{x_{1}+\cdots+x_{n+1}}{x_{n}}\leq L\Leftrightarrow t_{n}\geq \frac{1}{L(1-t_{n-1})} \end{align} for $n\geq N$ for some $N\in \mathbb{N}$. If we put $f(x)=\frac{1}{L(1-x)}$, $t_{n}\geq f(t_{n-1})$. We can check that $f(x)>x$ and $t_{n}$ diverges, i.e. $t_{n}>1$ for some $n$. Contradiction.

  • $\begingroup$ We can show that $p_{n}(x)=\sum_{r\geq 0} (-1)^{r}\binom{n+1-r}{r}x^{n-r}$, but I think this is not helpful to prove.. $\endgroup$
    – Seewoo Lee
    Jul 19 '16 at 5:52
  • $\begingroup$ Logical flaw : $\lim \sup \sum_1^{n+1}x_j/x_n =L$ with $0<L<\infty$ does NOT imply that $\sum_1^{n+1}x_j\leq Lx_n$ for all $n. $ This can be fixed as follows: Suppose that $0<L<4.$ Then fix $m$ such that $n>m\implies \sum_1^{n+1}x_i/x_n<4. $ Now for $ j=1,..,m $ replace $ x_j $ by $ s2^j $ where $ s $ is small enough that $ s2^j\leq x_j $ for $ j=1,...,m. $ Then the rest of your A works verbatim, arriving at the desired contradiction. $\endgroup$ Jul 20 '16 at 5:35
  • $\begingroup$ @user254665 Thank you very much. I'll edit it. $\endgroup$
    – Seewoo Lee
    Jul 22 '16 at 0:10
  • $\begingroup$ Now I'm not sure whether my fix works. There may be a problem with the value of $\sum_1^m x_i/x_m . $ $\endgroup$ Jul 22 '16 at 1:01
  • 1
    $\begingroup$ Nitpicking: $\limsup\frac{x_{1}+\cdots+x_{n+1}}{x_{n}}=L$ does not imply $\frac{x_{1}+\cdots+x_{n+1}}{x_{n}}\leq L$ for sufficiently large $N$. This can be salvaged of course by choosing $L$ greater that the limes superior, and less than $4$. $\endgroup$
    – Martin R
    Oct 28 at 7:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.