First note that αβ is an irrational number (otherwise there are positive integers k1,k2 such that k1α=k2β, which leads to a contradiction).
If α=pq is a rational number, then there is a positive integer k2 such that the decimal part of qk2β is less than q1. Taking k1=p⌊qk2β⌋ gives that
k1α=q⋅⌊qk2β⌋<k2β<q⌊qk2β⌋+1,
which contradicts to the assumption. Thus α must be irrational. Similarly, β must be irrational as well.
A pair of positive integers (a,b) is called distinguished if 0<bβ−aα<1. By our assumption, there is a unique positive integer t such that aα<t<bβ. Denoting u=t−aα,v=bβ−t, we say a distinguished pair (a,b) corresponds to the intermediate number t together with a distant pair (u,v). We will prove the following.
Lemma. Assume two distinguished pairs (a1,b1), (a2,b2) correspond to distant pairs (u1,v1), (u2,v2), respectively. Then v1u1=v2u2.
Proof of Lemma. Suppose that the intermediate numbers of these two distinguished pairs are t1,t2, respectively. Assume for the sake of contradiction that v1u1>v2u2. We also take ϵ=2u1v2−u2v1>0.
Since αβ is an irrational number, there are positive integers a0,b0 such that 0<a0α−b0β<ϵ. On the other hand, our assumption implies that there is another integer t0 satisfying b0β<t0<a0α. Denote u0=a0α−t0 and v0=t0−b0β. If u0u1−v0v1>1, then one can take L=⌊v0v1⌋+1 such that u0u1>L>v0v1, i.e.,
u1−Lu0v1−Lv0=(t1+Lt0)−(a1+La0)α>0,=(t1+Lt0)−(b1+Lb0)β<0.
Set k1=a1+La0 and k2=b1+Lb0. Then ⌊k1α⌋=⌊k2β⌋=t1+Lt0−1, which leads to a contradiction. This further implies that u0u1−v0v1≤1. Using a similar argument, we can show that u0u2−v0v2≥−1. Therefore,
u1−v0u0v1≤u0,u2−v0u0v2≥−u0,⇒u1v2−u2v1≤u0(v1+v2)<2ϵ.
However, this violates our choice of ϵ. The lemma is proved.
Now we see all distinguished pairs (a,b) share the same ratio vu=bβ−tt−aα. Denote this common λ=u+vv∈(0,1); then we have an equality λ⋅aα+(1−λ)⋅bβ=t∈Z. We call linear combinations of distinguished pairs with integer coefficients nice pairs. For each nice pair (c,d), it is not hard to see that
λ⋅cα+(1−λ)⋅dβ=c⋅λα+d⋅(1−λ)β∈Z.
Take a nice pair (a,b) and set δ=bβ−aα∈(0,1). We claim that for all M>0, any pair (c,d) of integers satisfying 0<dβ−cα<M and c>δaM, is nice. Indeed, it suffices to take L=⌊δdβ−cα⌋<δM<ac so that (d−Lb)β−(c−La)α=(dβ−cα)−Lδ∈(0,δ)⊂(0,1), and therefore the pair (c−La,d−Lb) must be distinguished and the pair (c,d)=(c−La,d−Lb)+L⋅(a,b) must be a nice pair.
Now take M=α+2β, and integers c0>δaM+1 and d0=⌊βc0α⌋ satisfying the condition 0<d0β−c0α<β. Using the analysis above, (c0,d0), (c0,d0+1), and (c0−1,d0) are nice pairs, and their linear combinations (0,1) and (1,0) are nice pairs as well. This implies in particular that λα and (1−λ)β are (positive) integers. Taking positive integers m2=λα and m1=(1−λ)β, we get αm2+βm1=λ+(1−λ)=1, or equivalently m1α+m2β=αβ. We conclude the proof.