ProofComplete

Proof of Schnirelmann's Theorem

Schnirelmann's theorem was the first result showing that every integer is a sum of a bounded number of primes, providing crucial motivation for the Goldbach conjecture. The proof uses the concept of Schnirelmann density and an additive combinatorics argument.


Statement

Theorem8.3Schnirelmann's Theorem

There exists a constant CC such that every integer nβ‰₯2n \geq 2 is the sum of at most CC primes. Equivalently, the set of primes is an additive basis of finite order.


Proof

Proof

Step 1: Schnirelmann density. For AβŠ†NA \subseteq \mathbb{N} with 0∈A0 \in A, define the Schnirelmann density Οƒ(A)=inf⁑nβ‰₯1A(n)n\sigma(A) = \inf_{n \geq 1} \frac{A(n)}{n} where A(n)=∣A∩[1,n]∣A(n) = |A \cap [1, n]|.

Schnirelmann's lemma: If Οƒ(A)β‰₯1/2\sigma(A) \geq 1/2, then A+A=N0A + A = \mathbb{N}_0 (every non-negative integer is a sum of two elements of AA).

Proof of lemma: For Οƒ(A)β‰₯1/2\sigma(A) \geq 1/2 and any nn, consider Aβ€²=A∩[0,n]A' = A \cap [0, n] and nβˆ’Aβ€²={nβˆ’a:a∈Aβ€²}n - A' = \{n - a : a \in A'\}. We have ∣Aβ€²βˆ£β‰₯n/2+1|A'| \geq n/2 + 1 (counting 0), so ∣Aβ€²βˆ£+∣nβˆ’Aβ€²βˆ£=2∣Aβ€²βˆ£>n+1|A'| + |n - A'| = 2|A'| > n + 1, meaning Aβ€²βˆ©(nβˆ’Aβ€²)β‰ βˆ…A' \cap (n - A') \neq \emptyset by pigeonhole. Hence n=a+aβ€²n = a + a' for some a,aβ€²βˆˆAa, a' \in A.

Schnirelmann's sum theorem: If Οƒ(A)+Οƒ(B)β‰₯1\sigma(A) + \sigma(B) \geq 1, then A+B=N0A + B = \mathbb{N}_0. More generally, Οƒ(A+B)β‰₯Οƒ(A)+Οƒ(B)βˆ’Οƒ(A)Οƒ(B)\sigma(A + B) \geq \sigma(A) + \sigma(B) - \sigma(A)\sigma(B).

Step 2: The Goldbach set has positive density. Let G={0,1}βˆͺ{n:n=p+pβ€²Β forΒ primesΒ p,pβ€²}G = \{0, 1\} \cup \{n : n = p + p' \text{ for primes } p, p'\}. By the Selberg sieve upper bound and explicit computations, the even integers not in GG (if any) are rare. More precisely, Vinogradov's theorem (or even earlier results of Schnirelmann using Brun's sieve) shows Οƒ(G)>0\sigma(G) > 0.

The key input: by Brun's sieve, the number of even integers ≀2N\leq 2N that are sums of two primes is β‰₯cN\geq cN for some c>0c > 0. Combined with Goldbach for odd numbers (every odd number >1> 1 is either prime or 2+2 + prime -- wait, this needs more care).

Actually, Schnirelmann proved: Οƒ(P+P)>0\sigma(\mathcal{P} + \mathcal{P}) > 0 where P={0}βˆͺ{primes}\mathcal{P} = \{0\} \cup \{\text{primes}\}. This follows from the lower bound: ∣{n≀N:n=p+pβ€²}∣β‰₯cN|\{n \leq N : n = p + p'\}| \geq cN for even nn (by the circle method or sieve estimates) combined with odd n=2+(nβˆ’2)n = 2 + (n-2) where nβˆ’2n - 2 is handled similarly.

Step 3: Iterating. Since Οƒ(P+P)=Ξ±>0\sigma(\mathcal{P} + \mathcal{P}) = \alpha > 0, repeated application of the sum theorem gives Οƒ(k(P+P))β†’1\sigma(k(\mathcal{P} + \mathcal{P})) \to 1 as kk increases. Once Οƒ(k(P+P))β‰₯1/2\sigma(k(\mathcal{P}+\mathcal{P})) \geq 1/2, one more summing gives the full N0\mathbb{N}_0. This shows every nn is a sum of C=2k+2C = 2k + 2 primes.

More precisely: Οƒ(A+A)β‰₯2Οƒ(A)βˆ’Οƒ(A)2=1βˆ’(1βˆ’Οƒ(A))2\sigma(A + A) \geq 2\sigma(A) - \sigma(A)^2 = 1 - (1-\sigma(A))^2. By induction: Οƒ(kA)β‰₯1βˆ’(1βˆ’Οƒ(A))2kβˆ’1\sigma(kA) \geq 1 - (1-\sigma(A))^{2^{k-1}}. For kβ‰₯⌈log⁑2(1/Οƒ(A))βŒ‰+1k \geq \lceil\log_2(1/\sigma(A))\rceil + 1: Οƒ(kA)β‰₯1/2\sigma(kA) \geq 1/2, and Οƒ((k+1)A)=1\sigma((k+1)A) = 1. Thus C≀2(k+1)=O(log⁑(1/Ξ±))C \leq 2(k+1) = O(\log(1/\alpha)). β–‘\square

β– 

RemarkQuantitative Bounds

Schnirelmann's original proof gave an astronomical value of CC. The best known bound is C=4C = 4 for sufficiently large nn (from Vinogradov-Helfgott: every odd n>5n > 5 is a sum of 3 primes, plus possibly one more prime for even numbers). If binary Goldbach holds, C=3C = 3 suffices for all nβ‰₯4n \geq 4.

ExampleSchnirelmann Density Examples
  • Perfect squares: Οƒ=0\sigma = 0 (density A(n)/n∼1/nβ†’0A(n)/n \sim 1/\sqrt{n} \to 0)
  • Primes: Οƒ=0\sigma = 0 (density ∼1/log⁑nβ†’0\sim 1/\log n \to 0)
  • Primes βˆͺ{1}\cup \{1\}: Οƒ=1/2\sigma = 1/2 (since A(2)=1A(2) = 1, giving σ≀1/2\sigma \leq 1/2)
  • {0,1,2,4,5,6,8,…}\{0, 1, 2, 4, 5, 6, 8, \ldots\} (all except 3β€Šmodβ€Š43 \bmod 4 gaps): Οƒ=1/2\sigma = 1/2