4

Assuming $\, f_i(x), i=1,...,N\,$ are real monotone (increasing or decreasing) functions, what can we say about $g(x)=\sum_i f_i(x)$? Does this extra information gives us more insight about $g(x)$?

seyran
  • 133

2 Answers2

8

By induction on $N \ge 1$, for any reals $a_1, \dots, a_N, b_1, \dots, b_N$ with $a_i < b_i$ for all $i = 1, \dots, N$, we have: $$ \sum_{i=1}^N a_i < \sum_{i=1}^N b_i \text{.} $$ Assume first that the $f_i$ are all monotone increasing (and that this means strictly). In any case we assume that they're all "the same kind of monotone".

Given reals $x, y$ with $x < y$, letting $a_i = f_i(x)$ and $b_i = f_i(y)$, we have $a_i < b_i$ for all $i$, so: $$ g(x) = \sum_{i=1}^N a_i < \sum_{i=1}^N b_i = g(y)\text{,}\tag{*} $$ so $g$ is monotone increasing too.

Similarly if the $f_i$ are monotone decreasing (replace "$<$" with "$>$" in (*)), or if they're monotone "nondecreasing" (replace "$<$" with "$\le$") or monotone "nonincreasing".

A simple counterexample shows that the sum of monotone functions of different kinds isn't necessarily monotone: let $f_1$ be the identity on $[0,1]$ and constantly 1 on $[1,2]$; let $f_2$ be constantly $0$ on $[0,1]$ and $x \mapsto 1-x$ on $[1,2]$. Then $f_1$ (resp. $f_2$) is monotone nondecreasing (resp. nonincreasing), but $g = f_1 + f_2$ is no kind of monotone — it's a half-cycle of a triangle wave.

BrianO
  • 16,579
  • 1
    The question might have been intended to allow the possibility that some of the $f_i$ are monotone increasing and others are monotone decreasing, in which case the sum need not be monotone. – Andreas Blass Oct 28 '15 at 10:13
  • @AndreasBlass Yyyeah that may be the case, I realized after answering. Perhaps that is intended. It would help to know the working definition of "monotone". (OP?) – BrianO Oct 28 '15 at 10:19
  • If we assume that "monotone" means "monotone", then some of them could be increasing and others decreasing. In that case, I believe the answer is "of bounded variation", isn't it? – bof Oct 28 '15 at 10:22
  • @AndreasBlass yes I meant both increasing and decreasing, I think the question is not clear so what do we know about function $g(x)$ if it is a summation of monotone functions? – seyran Oct 28 '15 at 10:25
  • @seyran My comment and bof's comment together answer the question in your comment. When some of the functions are increasing and some are decreasing, you can't conclude that the sum is monotone, but you can conclude that it is of bounded variation. That's all you can conclude, because every function of bounded variation is the difference of two increasing functions. – Andreas Blass Oct 28 '15 at 10:27
  • Thanks @AndreasBlass, I need to read more about the bounded variation functions, what interests me is if the function g(x) is convex (concave) in this case. – seyran Jul 05 '16 at 08:51
4

Firstly, I begin with a very simple observation: if $f_1 , \dots , f_N$ are all increasing functions, then their sum is increasing. In fact, for all $x < y$, you have $f_i(x) \le f_i(y)$, so summing up you get $$f_1(x) + \dots +f_n(x) \le f_1(y) + \dots +f_n(y)$$ i.e. $\sum_i f_i$ is increasing. Moreover if one of them is strictly increasing, then the sum is strictly increasing (you have $<$ instead of $\le$).

Now, in the general case, you have that some of the $f_i$s are increasing, while the other ones are decreasing. Summing together increasing ones you get an increasing function $F$. Summing together decreasing ones you get a decreasing function $G$. Note that $-G$ is an increasing function, so that you have $$\sum_i f_i = \sum_{\mbox{increasing}} f_i + \sum_{\mbox{decreasing}} f_i = F+G= F-(-G)$$ so that $\sum_i f_i$ is a difference of two increasing functions. And now, about monotonicity of $\sum_i f_i$ nothing can be said. For example, if you take $F(x)=x^3$ and $G(x)=-x$, you have $F(x)+G(x)=x^3-x = x(x-1)(x+1)$ which is not monotone since it has three distinct zeroes.

Crostul
  • 36,738
  • 4
  • 36
  • 72
  • If you want some more details about functions which are difference of two increasing functions (continuous, bounded), see http://math.stackexchange.com/questions/141338/bounded-variation-difference-of-two-increasing-functions – Crostul Oct 28 '15 at 10:24