Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

How to prove that convexity is necessary condition in this question? Need to construct an example of set $A$ for which $A+A=A$ but $0 \notin cl(A)$. From the linked question follows that $A$ must be non-convex.

share|cite|improve this question
I think that this problem is somehow related with the problem of construct a set $A$ such that $A+A=A$, but there is some $\alpha>0$ such that $\alpha A\neq A$. – Tomás Feb 16 '13 at 14:33
If you have a such set $A$ and $v \in A$, then must $A$ contain $\lambda v$ for $\lambda > 0$? Because I can't imagine a such set $A$ without this property. – Diego Silvera Feb 16 '13 at 15:25
That's why I asked in the linked post if the set $A$ is a convex cone, because if $\alpha A= A$ for every $\alpha >0$ and $A+A=A$, then $A$ is a convex cone. – Tomás Feb 16 '13 at 16:23
@Tomás That would not be hard to do: take positive rational numbers, for example. – user53153 Feb 17 '13 at 2:34
up vote 7 down vote accepted

This is merely a distilled version of the answer by coffeemath (hence CW). Pick an irrational number $\alpha$ and let $A=\{(m,n)\in\mathbb Z^2:m\alpha+n>0\}$. This is a discrete set, it does not contain $(0,0)$, and obviously satisfies $A+A\subseteq A$. To prove the reverse inclusion $A\subseteq A+A$, note that for any $(m,n)\in A$ there exists $(m',n')$ such that $0<m'\alpha+n'<m\alpha+n$, and therefore $(m,n)=(m',n')+(m-m',n-n')\in A+A$.

This construction comes up in complex analysis, since the linear combinations of the functions $\{z^mw^n: (m,n)\in A\}$ form an algebra on the torus $|z|=|w|=1$ (here $z,w$ are complex). However, in this case one throws in $(0,0)$ (hence, the constant function) in order to have a unital algebra. See, e.g., Banach spaces of analytic functions by K. Hoffman, pp 54-57.

share|cite|improve this answer
Nice! Needs a lot less detail than the fibonacci construction. +1 – coffeemath Feb 17 '13 at 10:33

Let $F_n$ denote the $n^{th}$ fibonacci number, with a few terms having nonpositive subscripts, so that $$F_{-2}=-1,\ F_{-1}=1,\ F_0=0,\ F_1=F_2=1,\ F_3=2,\ F_4=3,...$$ Define the points $v_n$ in the $xy$ plane by $$v_n=[(-1)^n F_{n-3},\ (-1)^n F_{n-1}],$$ so that $$v_1=[1,0],\ v_2=[1,1],\ v_3=[0,-1],\ v_4=[1,2],\ v_5=[-1,-3],\ v_6=[2,5],...$$ Note that these points have the property that $v_n=v_{n+1}+v_{n+2}$ for each $n \ge 1$. In fact they may be generated from this relation, after $n=2$, taking $[1,0],\ [1,1]$ as the first two points and thereafter defining $v_n=v_{n-2}-v_{n-1}.$

Define the set $A$ in the plane as consisting of all (finite) liner combinations of the $v_n$ with nonnegative integer coefficients. Then $A+A=A$. This is because each $v_n$ is the sum of the next two, and $A$ is the closure of the set of $v_n$ under addition.

Now the closure of $A$ can only contain integer pairs, so to show this closure does not contain $[0,0]$ is the same as showing that $[0,0] \notin A.$

Suppose we have a nontrivial relation

[1] $\sum_{k=1}^{m+1} c_k v_k=[0,0].$

Then using the relation $v_n=v_{n+1}+v_{n+2}$ we can transform the sum until it only involves the highest two indices of the sum, namely $k=m,m+1.$ In so transforming the relation, the coefficients remain nonnegative (one only adds such coefficients). So the relation [1] would imply there are nonegative $x,y$ for which $$x v_m + y v_{m+1}=[0,0]$$ Factoring out $\pm 1$ from this relation we have the left side as $$x[F_{n-3},F_{n-1}]+y[-F_{n-2},-F_n]=[u,v],$$ where $u=xF_{n-3}-yF_{n-2}$ while using the recurrence for the fibonacci numbers, $$v=xF_{n-3}+xF_{n-2}-yF_{n-2}-yF_{n-1}.$$ Then $v-u=xF_{n-2}-yF_{n-1}.$ So from $[u,v]=[0,0]$ would follow the two equations $$xF_{n-3}-yF_{n-2}=0,\\ xF_{n-2}-yF_{n-1}=0.$$ But the determinant of this system is $-F_{n-3}F_{n-1}+F_{n-2}^2,$ which a fibonacci identity says is always $\pm 1$. This implies that $x,y$ are both zero, which means the supposed relation [1] was trivial after all.

We conclude that $[0,0]$ is not in the set $A$, so that this is an example in two dimensions of a set $A$ for which $A+A=A$, and yet $[0,0]$ is not in the closure of $A$.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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