Zeta Worlds

A network has its own primes, its own zeros and its own Riemann Hypothesis, and all three can be computed outright. Build one, break it, sort a thousand, then set the real zeta function on the same bench.

the whole numbersζ(s) = ∏ primes p 1 ⁄ (1 − p−s) a networkZ(u) = ∏ prime loops 1 ⁄ (1 − ulength) One recipe, used twice. A prime loop is a round trip through the network that is not a shorter round trip repeated.

In a network world all three boxes are in hand: walk the network to list its primes, ring it to get its zeros, and an exact formula carries one list to the other. For the whole numbers the same formula holds, and the middle box is open. Zeta’s zeros do turn up in a few known objects, as notes that fade or as notes that are missing. None of those objects gives a reason for the zeros to sit on one line.

Build a world

Dots joined by connections. A prime loop is a round trip that never turns straight back and is not a shorter trip repeated. Send a pulse through the same network, never turning straight back, and it breaks down into a fixed set of patterns, called modes, much as a bell’s sound is made of a fixed set of tones. Each mode grows by its own factor at every step and turns at its own pace. Those rates are the zeros.

16
Edit
Connections as a list (read, edit or paste)

The network

the prime loop you picked
Drag a dot to move it. Click one dot and then another to connect them, or to cut the connection if there is one. Double-click empty space to add a dot.
…

Size
Independent loops
Growth rate
Worst zero

Its zeros

zero on the ½ line, or inside it zero beyond the line ×main term trivial zeros

Prime loops shortest first; each one counts twice, once for each direction

Loop census the same count, done two ways

LengthPrime loopsRound trips, counted by walkingRound trips, from the zeros alone

A round trip of length m is a prime loop of that length started from any of its steps, or a shorter prime loop repeated. The last column is zero1m + zero2m + ⋯ over every zero in the picture above, and uses nothing else.

Break it

Start with two separate worlds. Each has its own main term at σ = 1. Now cut one connection in each world and rejoin the four loose ends across the gap. That makes two bridges and leaves every dot with as many connections as before. Do it again and again. One of the two main terms has to go somewhere.

8

Two worlds and the bridges between them

bridge one side of the splitting mode the other side
Dot colours show the splitting mode, the mode with the largest positive swing after the main term. While the bridge is thin it is the two halves swinging against each other, and its zero is what is left of the second main term.

What a stray zero does to the count of round trips

the wiggle in the count of round trips, for each length the limit while every zero that counts is on the line
Each dot is how far the count of round trips of one length strays from the smooth part, measured against the square root of the smooth part. Each mark up the side is ten times the one below. With every zero that counts on the line the dots stay under the grey line however long the round trips get. One zero beyond the line sends them up a ramp without end. For the count of primes the matching statement, with a limit that is allowed to rise very slowly, is equivalent to the Riemann Hypothesis.
…

Bridges
Splitting mode
Its zero

The zeros as the bridge widens

zero on the ½ line stray zero and its mirror twin ×main term
The stray zero is what is left of the second world’s main term. It slides from σ = 1 toward the ½ line as the bridge widens, and its twin comes to meet it from the other side. Now and then some other zero lies beyond the line by chance, and the verdict says so when it happens.

Sort worlds

Draw networks at random, every one with the same number of dots and the same number of connections at each dot, and sort them by whether they pass. Every world is drawn by the same recipe. Which ones pass is decided only by how the connections happen to fall.

Worlds sorted
0
 
Pass
–
 
Fail
–
 
Proven for very large worlds
69% pass
Huang, McKenzie, Yau, 2024 preprint

How close each world comes

pass fail
Stress compares the world’s widest swing with the largest it can have while every zero that counts stays on the ½ line. At 1.00 or below they all do. Above 1.00 at least one zero lies beyond the line, with its mirror twin inside.

Do the failures share anything?

share of the group that passes all worlds together
Each bar is one group of worlds, and the thin line on it is the give or take. Worlds with many squares, or with dots far apart, fail more often than the rest. No one feature decides it. Groups with fewer than 15 worlds are left out.

Closest calls and worst failures open any of them in Build to look inside

WorldStressVerdictTrianglesSquaresLength 5Widest distance

Here each triangle, square or loop of length 5 is counted once. The census in Build counts each direction separately, so it shows twice as many.

The 69% is the value the share settles toward as worlds grow very large. At sizes a browser can sort, more pass, and the share falls slowly as the worlds grow: in check runs behind this page, 86% of worlds with 100 dots and 3 connections at each passed, 80% at 400 dots and about 78% at 1,000. “Give or take” is one standard error. A world can fail on two sides. On the splitting side its worst mode has two halves of the world swinging against each other, and in the extreme the world comes apart in two. On the checkerboard side neighbours swing against each other, and in the extreme every connection joins one colour to the other.

Zeta beside it

Left, the world from Build. Right, the whole numbers. The same two formulas run in both columns: zeros rebuild the count of primes, and primes rebuild the zeros.

A network world
The whole numbers

From zeros to primes add zeros one at a time and watch the count come right

Round trips, rebuilt from zeros

counted by walking rebuilt from the zeros in use

Primes, rebuilt from zeros

counted from the primes rebuilt from the zeros in use where the true count must stay if RH is true

From primes to zeros the same formula, run backwards

Zeros, rebuilt from loop counts

built from the loop counts alone where the zeros really are

Zeros, rebuilt from primes

true count of zeros below each height built from the primes in use

The dictionary

A network worldThe whole numbers
PrimesPrime loops. Their lengths are whole numbers: 3, 4, 5, …2, 3, 5, 7, … Their lengths are log 2, log 3, log 5, …
The countRound trips of length mψ(x): add log p for every prime power up to x (a running total)
Smooth partqm, from the main term. q is one less than the connections at each dot.x, from the pole at 1
The wiggleone wave for each zero, growing like qmσone wave for each zero, growing like xσ
The hypothesisno wave has σ above ½. Where every dot has the same number of connections, every wave that counts then has σ = ½ exactly. Decided by a finite calculation; true for some worlds and false for others.every wave has σ = ½. Open since 1859.
How many zerosa finite list, repeating up the stripinfinitely many, growing more crowded with height
Where the zeros come fromthe network. They are its modes.No object is known that would hold them on the line.

One sign differs. In a network the special points are where its zeta function blows up, so round trips are counted as the smooth part plus the waves. For the whole numbers they are where zeta vanishes, so primes are counted as the smooth part minus the waves. The next station shows that sign: the echoes of plain loops point up, and zeta’s point down.

Test a candidate

If something belongs in the open box, it has to ring at the zeros of zeta, and those zeros leave fingerprints. Three of them can be measured on any list of notes. Pick a structure, or paste a list of your own, and compare it with zeta’s zeros in gold.

5
full

The notes themselves the first stretch

each loop ringing on its own the candidate’s notes

Fingerprint 1: gaps between neighbours

candidate zeta’s zeros notes that ignore each other a woven system, two-way a woven system with a direction

Fingerprint 2: echoes

candidate zeta’s zeros prime powers on the axis

Fingerprint 3: crowding

candidate zeta’s zeros

Scorecard does it match zeta?

StructureGapsEchoesCrowding

Each row is measured live with the same three instruments. Getting one fingerprint right is not hard. None of the four structures built into this bench gets all three. Getting all three would still not be enough: a proof along these lines needs an object that holds the zeros on the line, and none is known. Other lines of attack on the Riemann Hypothesis do not need such an object at all.

Twelve things to try

    Start anywhere

    Each step sets up one of the stations above and jumps to it, with a note here on what to look for. After that the controls are yours.

    What is proven, and what is not

    • ProvenFor any network, the count of round trips of each length equals what you get by raising each zero to that power and adding them up. Ihara 1966, Hashimoto 1989, Bass 1992. The Loop census checks it on every world you build.
    • ProvenA network where every dot has the same number of connections has every zero that counts on the ½ line exactly when its widest swing is within a fixed limit. Such networks are called Ramanujan graphs. The first endless families with a fixed number of connections at each dot were built by Lubotzky, Phillips and Sarnak and by Margulis in 1988.
    • ProvenAbout 69% of very large random networks of this kind pass. Huang, McKenzie, Yau, 2024 preprint. It holds for any fixed number of connections from 3 up.
    • ProvenThe versions of the Riemann Hypothesis for curves over finite fields are true. There too, the zeros are the modes of a definite object. Weil 1948 for curves; Deligne 1974 for shapes of any dimension.
    • ProvenZeta’s zeros echo at the logarithms of prime powers and nowhere else, and they crowd together like the logarithm of the height. Riemann 1859, von Mangoldt 1905, Landau 1911.
    • ProvenZeta’s zeros turn up, at half scale, as the fading notes of waves on one curved surface with a leak. Pavlov and Faddeev 1972; Lax and Phillips 1976. The Riemann Hypothesis says that all of those notes fade at the same rate. Nothing known about the surface forces that.
    • CheckedThe gaps between zeta’s zeros appear to follow the law of a woven system with a direction. Montgomery 1973 found the pattern in pairs of zeros, and Odlyzko’s computations from 1987 on agree with it far up the line. It has not been proven.
    • OpenAn object that would hold zeta’s zeros on the ½ line. Pólya recalled suggesting, around 1912 to 1914, that the zeros might be the notes of an object that keeps its energy, which would force them onto the line. Hilbert is said to have had the same idea. No such object has been found. Berry and Keating described what its loops would have to be. Connes built a setting in which the primes are round trips and the zeros appear as missing notes, and the hypothesis comes down to one formula there that nobody has proved. Having an object is not enough: every network on this page has one, and its hypothesis can still fail. A proof needs a reason why the notes cannot leave the line.
    • OpenThe Riemann Hypothesis itself. Nothing on this page is evidence for it or against it. The page shows the kind of object a proof along these lines would need, and why having one is not enough.

    Names for the parts

    Prime loop
    A round trip through a network that never turns straight back and is not a shorter round trip repeated. Where it starts does not matter. The loop and the same loop walked the other way count as two.
    Round trip
    Any closed walk that never turns straight back, counted once for each step it could start from.
    Zero
    One mode of a pulse travelling through the network without turning back. Its angle is the mode’s pitch. Its distance from the centre is how much the mode grows at each step. Strictly these are points where the network’s zeta function blows up, not where it vanishes. The page calls them zeros because they do the job that zeta’s zeros do.
    σ (sigma)
    How fast a mode grows: 1 for the main term, ½ for a mode that grows like the square root of the main term.
    Main term
    The mode that grows fastest. It sets how quickly the number of round trips grows, and plays the part of zeta’s pole at 1. Some worlds have more than one, all growing at the same rate: a world that can be two-coloured has two, mirror images of each other.
    Trivial zeros
    Zeros at +1 and −1 that every world with two or more loops has. How many there are is settled by counting dots and connections, with one extra for a world that can be two-coloured. So the hypothesis leaves them out, as Riemann’s leaves out zeta’s trivial zeros.
    Stray zero
    A zero beyond the ½ line. Where every dot has the same number of connections it comes with a mirror twin inside the line.
    Smooth part and wiggle
    The smooth part of a count is what the main term alone would give. The wiggle is what is left when the smooth part is taken off, and it is made of one wave for each zero.
    Strip and circle
    Two drawings of the same zeros. The circle is the plain picture. The strip unrolls it so that σ runs left to right, as on the s-plane.
    Swing
    One number for each mode, worked out from the connections alone: large and positive when neighbouring dots move together, large and negative when they move against each other. The widest swing is the largest in size, leaving out the main term, and its mirror image in a world that can be two-coloured.
    Stress
    A world’s widest swing divided by the most it can have while every zero that counts stays on the line. Mathematicians write the limit as 2√q.
    Note
    The pitch of a zero, which is its height on the strip.
    Woven
    Every part feeds every other, so no note belongs to one part alone.
    Echo
    Add one wave for each note. At most lengths the waves cancel. Where they pile up there is an echo. For a structure made of loops, an echo at length L means some round trip has that length.

    Everything is computed in your browser. A world’s zeros come from the eigenvalues of a table built from its connections. Round trips are counted by exact whole-number arithmetic and compared with the zeros every time. Random worlds are drawn uniformly by pairing loose ends and starting again whenever a dot would join itself or join another dot twice, or the world comes out in more than one piece. The first 10,000 zeros of zeta are stored to six decimals. They were located with the Hardy Z function and checked against a second, independent program. In the gaps charts the two curves for woven systems are Wigner’s formulas, close stand-ins for the exact laws. The 0.035 cut-off for a match is this page’s rule of thumb, not a statistical test. The first 10,000 zeros of zeta sit 0.028 from the direction law because low zeros keep slightly more even gaps than the law says, and the fit is known to improve higher up. The count of zeta’s zeros rebuilt from primes is a close approximation, not an exact formula: sums over primes do not settle on the ½ line, so the page fades the larger primes out and stops where the slider stops. An echo is a weighted sum over the notes, with the lowest notes weighted most. For the loop candidates that weighting leaves the same small offset at every length, and the page takes it off. In Break, if more than one zero lies beyond the line on the splitting side, the splitting mode is taken to be the one that best follows the two halves. The loops joined at one core use the simplest junction, one shared channel that every loop feeds and is fed by, and other junctions would give other notes. The network zeta function is named after Ihara, and the count with the smooth part taken out is the network’s version of Riemann’s explicit formula.