Working notes on the primes

Grant Golland’s working notes on how the prime numbers are spread out. Every figure is computed here from a sieve of all 5 761 455 primes below 100 000 000, or with mpmath, and its caption says how. A sibling site, Zeta Zeros, covers the zeros that control these patterns.

Sentences carry a flag: Settled for proved theorems or computations on a stated range, Open for conjectures. A numerical match is never a proof.

Step 5

Prime races and Chebyshev’s bias

The even split is a long-run average. Along the way, one team usually leads. In 1853 Chebyshev noticed that primes 4k + 3 almost always outnumber primes 4k + 1.

Two panels up to one million. Top: count of primes 4k+3 minus count of primes 4k+1. It stays positive almost always; it first dips below zero at p = 26861 and is negative for only 239 of the 78497 odd primes counted. Bottom: count of primes 3k+2 minus 3k+1, which never goes negative in this range.
Running differences from the sieve, every prime up to 106. Mod 4: the lead first changes hands at p = 26 861 (as found by Leech, 1957), the 4k + 1 team leads at only 239 primes in this range, and at one million 4k + 3 leads by 147. Mod 3: the 3k + 2 team never trails here; its first lead change is at 608 981 813 029 (Bays and Hudson, 1978), far beyond this plot. Motion is off if your system asks for reduced motion.
  • Settled Littlewood (1914): both races change leader infinitely often. So the bias is not permanent.
  • Settled Why 4k + 3 is favoured: squares of primes are all 1 mod 4 and mod 3, so counting prime powers rather than primes shifts the count toward the non-square classes by about √x/log x. This is the origin of the bias, and it shows up through the zeros of the L-function for that modulus.
  • Open Rubinstein and Sarnak (1994) computed that 4k + 3 is ahead for a share 0.9959 of all x in the logarithmic sense, and 3k + 2 for 0.9990. Their result assumes the Generalized Riemann Hypothesis and that the positive zero heights of these L-functions are linearly independent over the rationals, both unproved. The numbers are theirs, not recomputed here.

Reading path · step 6 of 11