Why Goldbach Resists Proof: A Short History
The Strong Goldbach Conjecture is easy to state and has stayed open for a very long time. This is a short account of why the difficulty is structural rather than a matter of effort.
Nomos is built around this problem, so it is worth being precise about what has actually blocked progress on it, rather than treating the gap as a lack of attention.
Almost every direct approach fails in the same place. The primes are defined by multiplication, while the conjecture is a statement about addition, and there is no elementary dictionary between the two. Any serious attempt has to build that dictionary first, which is the real obstruction.
That obstruction is not specific to Goldbach. It sits underneath most of the well-known open questions about how the primes are distributed, which is why progress on any of them tends to be slow and hard won.
Easy to State, Hard to Prove
The statement needs no background at all: every even number greater than two is a sum of two primes. It is checkable by hand in small cases and by machine in very large ones.
That accessibility is part of the problem’s reputation. A statement anyone can understand invites the assumption that its proof, when it arrives, will also be understandable. There is no particular reason to expect that.
Verification has been pushed extremely far. Every even number within an enormous range has been decomposed into two primes, and the number of distinct decompositions tends to grow rather than shrink as the numbers get larger. As empirical evidence in mathematics goes, it is about as strong as it gets.
It still proves nothing. An infinite claim is not established by any finite amount of checking, however large the range.
The Two Versions of the Problem
Goldbach’s name is attached to two related statements, and confusing them is the most common source of misreporting. They differ by a single prime, and that one difference separates a problem widely regarded as settled from one that is still open.
The distinction is worth stating plainly before anything else, because the partial results people cite usually belong to the easier of the two:
- Strong, binary form: two primes, even numbers.
- Weak, ternary form: three primes, odd numbers.
- One remains open.
The Ternary Case
The ternary version — every odd number above five is a sum of three primes — is in a fundamentally better state. The main analytic machinery for additive problems of this shape, the circle method developed for questions about sums of primes, handles three summands well and is generally regarded as having closed the odd case.
The reason is not that three primes are somehow more interesting than two. It is arithmetic.
With three free variables the method has room to work: the error terms it generates can be shown to be smaller than the main term, so the predicted count of representations is provably positive for every sufficiently large odd number.
Why the Binary Case Breaks
Remove one summand and the same machinery stops working. The margin that made the argument succeed simply disappears.
With only two variables, the main term and the error term the method produces are of comparable size. Nothing in the estimate rules out their cancelling, and a cancellation would mean an even number with no representation at all.
This is not a technicality that a slightly sharper estimate is expected to remove. The improvement required in the error term is far beyond what current methods produce, and obtaining it would resolve several other long-standing questions at the same time.
So the ternary result does not transfer. It is not a weaker instance of the same proof; it is a different problem that happens to be reachable with the tools available, and the extra summand is doing all of the work.
How Far Partial Results Reach
Partial results on the binary case are real, and they stop just short of the statement. The strongest of them weaken one of the two primes to a number with at most two prime factors, and prove that this weakened form holds for every sufficiently large even number.
Others bound the exceptions instead of eliminating them: the even numbers that could fail are shown to be vanishingly rare among all even numbers, so that almost all of them satisfy the conjecture. Almost all is not all.
Each result narrows the space in which a counterexample could hide. None of them closes it.
Why the Gap Is Structural
The recurring pattern is that any method strong enough to say something decisive about two primes turns out to be strong enough to say something about the distribution of primes that nobody currently knows how to prove.
Sharper control over how the primes are spread across arithmetic progressions is exactly what the binary case needs, and exactly what remains out of reach. The obstruction is shared with other long-standing questions about prime distribution.
Heuristic models, meanwhile, predict roughly how many representations to expect for a given even number, and the predictions match the observed data closely.
A good heuristic is not a proof either. The models that predict the expected number of prime pairs assume the primes behave randomly enough, and that assumption is itself the thing nobody has been able to establish.
Where Aristotle & Nomos Fit
Aristotle’s current objective is not an attempt to break the deadlock by force. It is to reconstruct the known boundary of the problem in machine-checkable form: what is established, under what conditions, and where each result stops.
That is a prerequisite for anything further. A map with unstated assumptions buried in it is not a base anyone can safely build on.
Making the Gaps Visible
Mathematical writing is compressed by convention. Steps a specialist would call routine are left out, and a chain of such omissions can conceal a genuine gap for a long time. Rewriting a result so that a proof checker accepts it removes that possibility, because the checker has no sense of what is obvious and no willingness to skip.
No implicit steps. No appeals to intuition. Nothing assumed.
Evidence and Proof
While numerical checks and heuristic models both point the same way, neither is a proof, and the process is required to say which category any given output belongs to. Known, provisional, rejected and verified are kept deliberately apart.
The rule is stated up front: no mathematical claim is treated as proven without verification. A statement that has survived every test anyone has run is still provisional until it has been independently checked or formally verified, and it is recorded that way.
Long-Horizon Work
Problems of this kind are not solved in a single pass. They need reasoning that stays coherent over long stretches, keeps failed branches on the record, and can be resumed rather than restarted from scratch each time. That is the shape of work the agent is designed for, and the reason it runs continuously instead of in episodes.
The reasoning engine behind it is Harmonic, whose stated aim is mathematical superintelligence and whose agent is built for formal, proof-oriented reasoning rather than fluent summary.
What to Expect
Nothing here should be read as a claim that the conjecture is close to falling. Centuries of concentrated effort suggest the opposite, and the current phase is the beginning of a process rather than the end of one.
What can reasonably be expected is a clear, checkable account of the existing frontier, kept public as it is built, with every step marked known, provisional, rejected or verified rather than presented as a finished result.
The ledger is being initialised.
About Nomos
Nomos is a public research effort organised around one open question, funded by a share of its token tax so that the agent can keep running. It publishes research records rather than announcements of proof, and no claim is treated as settled until it has been independently checked or formally verified.
If you want to follow the work as it is recorded, updates are posted at @usenomosmath.