We will prove by induction on d that, for every positive integer N, there exist positive integers b0,b1,…,bd−1 such that, for each i=0,1,2,…,d−1, we have bi>N and
2bi+bi≡i(modd).
This yields the claim for m=b0.
The base case d=1 is trivial. Take an a>1 and assume that the statement holds for all d<a. Note that the remainders of 2i modulo a repeat periodically starting with some exponent M. Let k be the length of the period; this means that 2M+k′≡2M(moda) holds only for those k′ which are multiples of k. Note further that the period cannot contain all the a remainders, since 0 either is missing or is the only number in the period. Thus k<a.
Let d=gcd(a,k) and let a′=a/d,k′=k/d. Since 0<k<a, we also have 0<d<a. By the induction hypothesis, there exist positive integers b0,b1,…,bd−1 such that bi>max(2M,N) and
2bi+bi≡i(modd) for i=0,1,2,…,d−1.(1)
For each i=0,1,…,d−1 consider the sequence
2bi+bi,2bi+k+(bi+k),…,2bi+(a′−1)k+(bi+(a′−1)k).(2)
Modulo a, these numbers are congruent to
2bi+bi,2bi+(bi+k),…,2bi+(bi+(a′−1)k),
respectively. The d sequences contain a′d=a numbers altogether. We shall now prove that no two of these numbers are congruent modulo a.
Suppose that
2bi+(bi+mk)≡2bj+(bj+nk)(moda)(3)
for some values of i,j∈{0,1,…,d−1} and m,n∈{0,1,…,a′−1}. Since d is a divisor of a, we also have
2bi+(bi+mk)≡2bj+(bj+nk)(modd).
Because d is a divisor of k and in view of (1), we obtain i≡j(modd). As i,j∈{0,1,…,d−1}, this just means that i=j. Substituting this into (3) yields mk≡nk(moda). Therefore mk′≡nk′(moda′); and since a′ and k′ are coprime, we get m≡n(moda′). Hence also m=n.
It follows that the a numbers that make up the d sequences (2) satisfy all the requirements; they are certainly all greater than N because we chose each bi>max(2M,N). So the statement holds for a, completing the induction.