0 or 1 to one whose base-4 digits are. Just below each near meeting of a power of 3 and a power of 4 the sums S miss a whole run of integers, so the share D(x) of lit integers up to x dips, and along ever closer meetings its lower limit is 0. Whether it returns above one fixed share at arbitrarily large x is open. Jump to a gap, zoom into the strip and watch.>}
controls={controls}
foot={<>The strip is one bit per integer up to 3^{level}, built in Rust: the members of A set directly, then each power of 4 folded in by one shift-or pass. Each strip pixel is shaded by the share of lit integers it covers, and the density chart draws the least and the greatest D(x) over each pixel of log x, so every dip shows at its true depth. The energy E(k, m) is summed over the 3^m digit strings a difference in B_m can be, each weighted by 2 to the zero digits it has in base 4 and in balanced ternary. Every count, density, gap and energy comes out of the crates through wasm; the page only draws. The proofs and the census to 3^22 are on two bases, section Object S.>}>
Open Is the upper density of S positive: does D(x) return above one fixed share at arbitrarily large x?
Proved No sum lands in the gap (d, min(3^k, 4^m)), d = (3^k - 1)/2 + (4^m - 1)/3 the largest sum of A_k + B_m. Where 4^m/3^k is near 1 that gap is a sixth of the scale, and iterated along ever closer coincidences it drives the lower density to 0, the answer to the question Erdos asked, checked in Lean on the problem page.
Each grey band is a gap (d, min(3^k, 4^m)), open when 4^m/3^k lies between about 3/4 and 3/2. D falls through the band and bottoms out just below min(3^k, 4^m); what it does after the last band drawn is the open question. Click to move x.
Q(k, m) = E(k, m) (d + 1)/4^(k+m) weighs how often two sums of A_k + B_m collide against a flat spread, and Cauchy-Schwarz gives card(A_k + B_m) >= (d + 1)/Q. Proved Q is unbounded, yet for every eps > 0 it is below 3^(eps k) at every large k and every m with 1/3 <= 4^m/3^k < 4, the window every pair here sits in. Conjecture Q stays bounded along one infinite chain of pairs, which would put the upper density at least 1/Q. Click a dot to pick its pair.