The guided path: from primes to the proof
Four stops, about 30 minutes, one idea at each. You start by finding primes with your own hands and finish at the exact idea the Lean proof turns on: why an upper limit on primes is enough to rule out zeros off the critical line. Each stop ends with a question that the next stop answers.
The path in one breath
- Find the primes. Sieve them out. They get rarer as you go.
- Count them. The count follows a smooth curve, and RH says how tightly.
- Put a ceiling on them. Chebyshev's trick: one whole number that is divisible by lots of primes, but not too big.
- See why a ceiling rules out zeros. A zero off the line would push a prime count through the ceiling.
You need: whole numbers and powers for stops 1 and 3; graphs for stop 2; nothing new for stop 4. Formulas are optional extras.
Stop 1 of 4 · Years 5 and up · about 5 minutes
Find the primes
A prime has no factors except 1 and itself. Sieving crosses out everything else.
The Riemann Hypothesis is a statement about prime numbers, so start by making some. This is the same sieve as Activity 1.
Sieve of Eratosthenes
Tap the smallest number that isn't crossed out yet: it's prime, so it gets circled, and every one of its multiples is crossed out in one go. Can you guess how many it will remove? Keep going until the sieve tells you to stop, and find out why it can stop early.
What you just saw
- Every number that isn't prime has a prime factor no bigger than its square root. That's why the sieve can stop early.
- Each new prime removes fewer numbers than the one before. Look at the tally: 2 removes 59, 3 removes 19, 5 removes 7, 7 removes 4.
- Primes get rarer: 25 of the first 100 numbers are prime, but only 21 of the next 100.
Check yourself. Sieving up to 120, the sieve stops after 7, before you ever use 11. Why is that safe?
Next question Primes get rarer. How fast? Can we predict how many there are up to a million without finding them all? Go to stop 2 →
Stop 2 of 4 · Years 9 and up · about 8 minutes
Count them
The number of primes up to \(x\) hugs a smooth curve. RH says it never strays far.
Write \(\pi(x)\) for the number of primes up to \(x\). (It's just a name, nothing to do with 3.14.) Gauss guessed that near \(x\), about one number in \(\ln x\) is prime, which gives the estimate \(x/\ln x\), and the better curve \(\operatorname{li}(x)\). Here are the real counts, from the same code as Activity 2.
Counting primes: \(\pi(x)\) versus \(x/\ln x\)
Try it: choose 1,000,000 and tick show li(x). Compare the \(\pi(x)\) column with the \(\operatorname{li}(x)\) column, and the difference with the \(\sqrt x\) column.
What you just saw
- The ratio column creeps towards 1: \(\pi(x)\) really does grow like \(x/\ln x\). That is the Prime Number Theorem (1896).
- \(\operatorname{li}(x)\) is much closer. Up to a million, the gap between \(\pi(x)\) and \(\operatorname{li}(x)\) is about 130, while \(\sqrt x\) is 1,000.
- That is what RH is about. RH is equivalent to: the gap between \(\pi(x)\) and \(\operatorname{li}(x)\) is never much bigger than \(\sqrt x\) (precisely, below \(\tfrac1{8\pi}\sqrt x\ln x\) from \(x=2657\) on; Schoenfeld 1976).
Check yourself. Up to \(x=1{,}000{,}000\), how does the gap between \(\pi(x)\) and \(\operatorname{li}(x)\) compare with \(\sqrt x=1{,}000\)?
A table up to a million is evidence, not proof. To prove a statement about every \(x\), you need a way to bound primes without counting them one by one.
Next question How can you prove there aren't too many primes, without listing them? Go to stop 3 →
Stop 3 of 4 · Years 7 and up · about 8 minutes
Put a ceiling on them: Chebyshev's trick
Divisible by lots of primes + not too big = not too many primes.
In the 1850s Chebyshev found a way to bound primes from above using one whole number. Here is the simplest version, due to Erdős. Take the number of ways to choose 10 things from 20:
$$\binom{20}{10}=\frac{20!}{10!\,10!}=184{,}756=2^2\times 11\times 13\times 17\times 19.$$- What divides it. Every prime between 10 and 20 (11, 13, 17, 19) divides it. Each appears once on top, in \(20!\), and not at all underneath, so it can't cancel.
- How big it is. It is at most \(4^{10}\), because choosing 10 from 20 is one of the \(2^{20}=4^{10}\) ways to choose any subset of 20 things.
- Put them together. So \(11\times13\times17\times19\le4^{10}\). The same works for every \(n\): the primes between \(n\) and \(2n\) multiply to at most \(4^n\). That is a ceiling: there can't be too many of them.
Move the slider and watch both facts hold for every \(n\). This is the same activity as on How the argument works.
Try it: Chebyshev's number
The bars use a logarithmic scale (number of digits). The teal bar is always shorter than the coral one. That gap is how much the trick "wastes".
What you just saw
- The trick only ever gives an upper limit. Divisibility says "these primes fit inside this number", and size says "the number isn't bigger than this". Neither can say there are at least so many primes.
- Chebyshev's ceiling is in the right ballpark but too loose to prove RH: the rounding in factorials wastes too much (that's the gap between the bars).
- The Lean proof keeps the same two facts, what divides it and how big it is, but applies them to a far cleverer object (built from "theta lifts") whose size is known almost exactly. Its ceiling is sharp enough to reach the \(\sqrt x\) scale.
Check yourself. Why must 13 divide \(\binom{20}{10}=\frac{20!}{10!\,10!}\)?
Next question We now have a way to put a ceiling on primes. But RH is about zeros of the zeta function. Why would a ceiling on primes say anything about zeros? Go to stop 4 →
Stop 4 of 4 · Years 10 and up · about 8 minutes
Why a ceiling rules out zeros
Every zero adds a wave to the prime count. A zero off the line makes a wave that grows too fast and breaks through the ceiling.
Riemann found that the error in a prime count is a sum of waves, one for each zero \(\rho=\beta+i\gamma\) of the zeta function. Zeros come in pairs, \(\beta+i\gamma\) and \(\beta-i\gamma\), and each pair adds a wave of roughly this shape:
$$x^{\beta}\cos(\gamma\ln x+c).$$Two things matter. The wave's height grows like \(x^{\beta}\): the further right the zero (bigger \(\beta\)), the faster it grows. And the wave swings up as far as it swings down.
Zeros on the critical line have \(\beta=\tfrac12\), so their waves grow like \(\sqrt x\): exactly the size of error you saw at stop 2. Now imagine a zero off the line, with \(\beta\) bigger than \(\tfrac12\). (The proof shows no such zero exists. We imagine one to see why.)
Try it: a wave against the ceiling
The graph shows one imagined zero's wave and the ceiling \(x^{1/2+\varepsilon}\) that the proof establishes. Both are divided by \(\sqrt x\) so they fit on one graph, and \(x\) is on a log scale: each step along is ten times bigger.
What you just saw
- With \(\beta=\tfrac12\) the wave stays under the ceiling forever.
- With \(\beta\) bigger than \(\tfrac12+\varepsilon\), the wave eventually breaks through, and then keeps breaking through, higher each time. A proved ceiling makes that impossible, so no such zero exists.
- A zero between \(\tfrac12\) and \(\tfrac12+\varepsilon\) slips under. That is why the proof establishes the ceiling for every \(\varepsilon\gt0\), however small: then no \(\beta\) above \(\tfrac12\) can hide.
- Because the wave swings up as much as down, a ceiling alone is enough. No floor is needed. That is lucky, because Chebyshev-style tricks only ever give ceilings (stop 3).
Check yourself. Why is an upper limit on the prime count enough, with no lower limit?
One more step, for the record. The ceiling rules out zeros to the right of the line. Zeros to the left come in mirror pairs with zeros on the right (\(\rho\) and \(1-\rho\), by the functional equation), so they're ruled out too. That short argument is the first step of the Lean proof (Activity 8). The step from "ceiling" to "no zeros" is a classical theorem of Landau.
Where the path ends, and where the proof begins
You now have the whole shape of the argument:
- RH is about how tightly the prime count hugs its smooth curve (stop 2).
- A zero off the line would make the count break through a ceiling of about \(\sqrt x\) (stop 4).
- So it's enough to prove that ceiling. Chebyshev's trick proves ceilings, just not sharp enough (stop 3).
- The Lean proof builds a far better "number" whose divisibility and size are both known almost exactly, and repeats the step about \(\sqrt x\) times. Balancing the errors gives exactly the exponent \(\tfrac12\): the critical line.
Step 4 is where everyone, experts included, currently stops. The proof's main inequality, called (M), is checked by computer but hasn't yet been explained by hand. Nobody understands all of this proof yet. You've just walked the part that people can.
Read how the argument works Find your rung on the maths ladder More activities