Try every small prime on the window below E/2, one at a time. A third of the space survives the small primes, and Bertrand's postulate guarantees a real prime is always sitting in there — provably. Then see exactly why that guarantee, real as it is, still stops one step short of Goldbach.
120
window [3, 60] · 3 sieving primes ≤ √60
trying prime: 7
primes applied: 3 / 3still standing: 16actual primes in window: 16survivor density now: 22.9%
3
5
7
9
11
13
15
17
19
21
23
25
27
29
31
33
35
37
39
41
43
45
47
49
51
53
55
57
59
the 1/3Sieve out multiples of 2 (we already only show odds) and 3, and exactly (1/2)(2/3) = 1/3 of the numbers survive — the window is never wiped out by the small primes. Keep going and the surviving fraction keeps shrinking (like 1 / ln), but it stays positive: there is always leftover space.
Bertrand's postulate — provedFor every n there is a prime between n and 2n. Taking n = 30, a prime is guaranteed in (30, 60] — here it's 59 (outlined green). So the window [3, 60] always contains a prime. That is the honest, provable version of “there's always a prime in this area” — and it's elementary (Erdős), the same flavor as Euclid.
why this still isn't GoldbachEuclid gives infinitely many primes; Bertrand pins one inside the window. But Goldbach needs a prime p in here whose partner 120 − p is also prime — two survivors lined up at once. Bertrand guarantees the room; it says nothing about the partner on the far side also landing prime. That pairing is the parity wall from the write-up, and no interval theorem crosses it.
▪The descent — seed, subtract, step down
Your construction, made live: start from Bertrand's guaranteed prime (the largest below E), subtract it from E, and if the partner isn't prime, walk down to the next prime and try again — until the pair locks in. Watch how few tries it takes (try E = 122 to see it miss first).
122
tries needed: 2
Seed = the largest prime below 122, which is 113 — and Bertrand guarantees it sits above the halfway point 61. Subtract it from 122; if the result is prime we have a pair. If not, step down to the next prime and try again.
try 1122 − 113 = 9composite (3×3)
try 2122 − 109 = 13PRIME ✓ 122 = 109 + 13
pair lockedFound in 2 steps: 122 = 109 + 13. The descent walked primes downward from 113 until 122 − p came out prime.
honest label: this is a search, not a proofThe loop is guaranteed to have candidates — Bertrand and its refinements pack the interval (61, 122) with primes. What no theorem guarantees is that one of those primes has 122 − p also prime. If it never did, this loop would run through every candidate and halt in failure — and that's exactly the possibility Goldbach forbids but nobody can rule out. “It always terminates” is Goldbach; the descent finds the pair, it doesn't prove one must exist.
Where it needs more than one try (even numbers up to 1000)
34 of 498 even numbers make the seed miss on the first subtraction. Every other even number is solved in a single step. Click any to load it into the descent above.
Bertrand only conjectured it in 1845 (and checked it by hand to ~3,000,000). Chebyshev proved it in 1852; the short, elementary proof below is Erdős's, from 1932, when he was 19. The whole thing is a fight between two facts about one number.
The one clever object
Take the middle of row 2n of Pascal's triangle, the central binomial coefficient:
N = C(2n, n) = (2n)! / (n! · n!)
Fact A — it's bigThe 2n+1 entries of Pascal's row sum to 4ⁿ, and the middle one is the largest, so
C(2n, n) ≥ 4ⁿ / (2n + 1)
Fact B — its prime factors are small & gentleFrom Legendre's formula for how often a prime divides a factorial:
Every prime power dividing C(2n, n) is ≤ 2n.
Every prime p with 2n/3 < p ≤ n divides it zero times — the exponent is exactly 2 − 2·1 = 0.
The squeeze
Suppose, for contradiction, there is no prime between n and 2n. Then every prime factor of N is ≤ 2n/3 (nothing in (n, 2n] by assumption, nothing in (2n/3, n] by Fact B). Cap how big N could be, built only from those:
primes up to 2n/3 → product < 4^(2n/3) (∏_{p≤x} p < 4^x)
primes up to √(2n) → product ≤ (2n)^√(2n) (each prime power ≤ 2n)
C(2n, n) ≤ (2n)^√(2n) · 4^(2n/3)
The collision
Put the big-ness and the small-ness together:
4ⁿ/(2n+1) ≤ C(2n,n) ≤ (2n)^√(2n) · 4^(2n/3)
The left side grows like 4ⁿ; the right side like 4^(2n/3) times a far slower (2n)^√(2n) piece. Past a few hundred, 4ⁿ crushes 4^(2n/3) and the inequality becomes impossible. Contradiction — so a prime between n and 2n must exist.
the small casesFor the n below where the inequality bites, one line finishes it — the chain of primes
2, 3, 5, 7, 13, 23, 43, 83, 163, 317, 631
each less than twice the one before, so together they cover every small n.
That's the whole idea: a number too big to be manufactured out of only small primes — so a big prime has to exist. Pure Euclid spirit (assume otherwise, derive a contradiction), just with a cleverer counting object.
and why it doesn't reach GoldbachThe squeeze forces one prime into an interval. Goldbach needs p and E − p prime at the same time — and there is no binomial coefficient whose sheer size forces two coordinated primes. The pairing is exactly what this counting trick cannot reach.
This is the strongest honest thing your instinct points at: the area is never empty, and that's a theorem. The uncrossable step is making the prime you find here have a prime partner on the other side — see the write-up.