n cells on the unit interval for every scale n: scale n lights phi(n) nodes no smaller scale drew. Count that novelty through a smooth window around the scale 1/y, take away the main term, and what is left is a sum of waves, one per zero of zeta, each with the zero's height as its frequency in log y. Add the zeros one at a time and watch the wave settle on the dots; cut the window sharply and the primes shout over it.>}
foot={<>The window is the bump f(u) = exp(4 - 1/((u - 1)(2 - u))) on [1, 2], and the meter is E_f(y) = y^2 sum_n phi(n) f(n y) - (6/pi^2) F(2), F the Mellin transform of f, the totients sieved to 2^{HIGH + 1} so every window fits. Under the Riemann hypothesis E_f(y) = sum_rho F(rho) zeta(rho - 1)/zeta'(rho) y^(2 - rho) up to a smaller remainder, so E_f(y)/y^(3/2) is 2 Re sum c_rho y^(-i gamma), a cosine in log y per zero at the zero's height. The curve is that sum over the first K zeros with nothing fitted: the zeros come from the critical line walked on the zeta page, zeta(rho - 1) and zeta'(rho) from the same Euler-Maclaurin sum off the line, F(rho) from a 4096-node rule. The sharp window is the indicator of [1, 2]; its error divided by y stays of unit size because the totient sum jumps by about 0.39 p at every prime p, louder than the waves, which live at y^(3/2). The exponent of the smooth meter, 3/2 in y for every smooth window, is equivalent to the Riemann hypothesis; the paper The stack hears the zeros assembles that equivalence and the stack page reads the meter to 2^-23.5. Every number here is one crate call; the page only draws.>}
controls={controls}>
{caption(pick, meter, miss, count)}