Someone asked me to help them with the following real analysis problem and it's been bugging me that I haven't been able to figure it out or think of something new to try for about a day now:
Suppose it holds for a sequence $\{a_j\}$ of complex numbers that for every pair of integers $N>M>0$, $$\lvert a_M-a_{M+1}\rvert+\lvert a_{M+1}-a_{M+2}\rvert+\ldots+\lvert a_{N-1}-a_{N}\rvert\le1.$$ Prove that $\{a_j\}$ converges.
It was clear to both of us that we have to show that the sequence is Cauchy. I tried contradiction a few times, though I (apart from negating the definition incorrectly a few times) couldn't begin to place where the contradiction would arise. I get that we must have $\lvert a_M-a_N\rvert\le1$, but can't see how that would imply that it's arbitrarily small. After thinking about it on my own, I realised that the smallest difference between consecutive terms must be no greater than $\frac{1}{N-M}$, but the largest one could theoretically be arbitrarily close to $1$, so I don't really see how that's helpful.
And that's about where I've been stuck for about a day. If anyone can give me hints about where to go, I'd appreciate it.
Also, yes, despite the exercise concerning a complex sequence, it is for a real analysis course.
Firstly, the sequence $(a_n)$ is bounded since \begin{align*} |a_n -a_1| &= |(a_n - a_{n-1}) + (a_{n-1} - a_{n-2}) + \cdots + (a_2 - a_1)|\\[4pt] &\le |a_n - a_{n-1}| + |a_{n-1} - a_{n-2}|+ \cdots + |a_2 - a_1|\\[4pt] &\le 1\\[4pt] \end{align*} It follows that the sequence $(a_n)$ has at least one limit point.
Suppose the sequence $(a_n)$ has two distinct limit points $u,v$, say. Let $\theta = |u-v|$.
Construct a subsequence $b_1,b_2,b_3,...$ of the sequence $(a_n)$ such that
It follows that $|b_{n+1}-b_n| > {\large{\frac{\theta}{3}}}$, for all n.
For each $n$, there are positive integers $j,k$, with $j < k$ such that $b_n = a_j$ and $b_{n+1} = a_k$. Then $$\frac{\theta}{3} < |b_{n+1}-b_n| = |(a_k-a_{k-1}) + \cdots + (a_{j+1}-a_j)| \le |a_k-a_{k-1}| + \cdots + |a_{j+1}-a_j| $$ But then, summing the LHS for all $n$ yields $\infty$, whereas summing the RHS for all $n$ yields at most $1$, contradiction.
It follows that the sequence $(a_n)$ can't have two distinct limit points, hence, since the sequence $(a_n)$ is bounded, it must converge.