The Mertens Conjecture proposed that a specific function tracking patterns in prime factorization, called the Mertens function, always stays bounded within a simple square-root limit. Believed true for nearly a century and supported by extensive numerical calculation, it was proven false in 1985 — without anyone ever actually finding, or even to this day knowing, a single explicit number where it fails. It stands as one of mathematics' most instructive cautionary tales about the difference between overwhelming evidence and genuine proof.
Some numbers have an even number of distinct prime factors, some an odd number, and some (perfect squares and similar) have a repeated factor that gets a special treatment. Tracking this "even versus odd" pattern across all the numbers up to some point n produces a running tally called the Mertens function, M(n). The Mertens Conjecture proposed that this running tally, however far you go, never strays further from zero than the square root of n.
The Mertens Conjecture had been numerically verified for every single number up to extremely large limits, with the pattern holding firmly and showing no sign whatsoever of breaking down. This made it a widely believed, well-respected open conjecture for nearly 100 years. Its eventual disproof is one of the most vivid illustrations available of a core theme running throughout this Codex: however extensive, computational verification is never a substitute for actual mathematical proof (compare the Goldbach Conjecture and Collatz Conjecture entries, both similarly well-supported numerically yet still formally unproven).
Had the Mertens Conjecture turned out to be true, it would have immediately implied the Riemann Hypothesis (see that entry) — one of the most important open problems in all of mathematics. Its disproof did not resolve the Riemann Hypothesis one way or the other (the two statements are not logically equivalent, only one-directional), but it did remove what would have been an unexpectedly elegant, comparatively simple potential pathway to proving Riemann.
The Mertens function is built from the Möbius function μ(n): μ(n)=1 if n is a product of an even number of distinct primes, μ(n)=−1 if n is a product of an odd number of distinct primes, and μ(n)=0 if n has any repeated prime factor (like 4=2², or 12=2²×3). The Mertens function is simply the running sum: M(n) = μ(1)+μ(2)+...+μ(n).
Franz Mertens conjectured in 1897 (based on his own calculations up to n=10,000, and building on an earlier, similar conjecture by Thomas Stieltjes in 1885) that |M(n)| < √n for all n>1. This is a notably strong and specific claim — considerably stronger than what's actually needed to imply the Riemann Hypothesis, which only requires the weaker bound that M(n) grows slower than n^(1/2+ε) for any small ε>0.
Rather than searching for an explicit violating number directly (which would require computation far beyond feasible reach), Odlyzko and te Riele instead analysed the mathematical relationship between M(n) and the zeros of the Riemann zeta function (see the Riemann Hypothesis entry), using extensive high-precision computation of the first several thousand zeta zeros combined with sophisticated analytic estimation techniques. This indirect analytic approach allowed them to prove that M(n)/√n must eventually exceed 1 (and, separately, must eventually fall below −1) for some sufficiently large n — establishing a rigorous existence proof of a violation without ever pinning down its specific numerical location.
Subsequent refined estimates suggest the first actual violation likely occurs somewhere around n = 10^(10^40) or possibly even larger — a number so astronomically large that directly computing M(n) at that scale is entirely beyond any conceivable computational capability, now or for the foreseeable future. The Mertens Conjecture will likely remain "disproven, but with a genuinely unknown counterexample" essentially indefinitely.
There's a precise analytic relationship between the growth rate of M(n) and the location of the zeros of the Riemann zeta function ζ(s): the Riemann Hypothesis is exactly equivalent to the statement that M(n) = O(n^(1/2+ε)) for every ε>0 (see the Big-O Notation entry) — a growth-rate bound distinctly weaker than the specific, stronger √n bound that Mertens originally proposed. Since the (now disproven) Mertens bound would have been strong enough to imply this weaker equivalent condition, the truth of Mertens's stronger conjecture would have immediately settled Riemann as a straightforward corollary.
A separate, logically weaker statement — the weak Mertens conjecture, involving an integral average of M(n)²/n² rather than the pointwise bound on M(n) itself — was also long conjectured, and would ALSO have implied the Riemann Hypothesis if true. This weak version was independently disproven as well (by Odlyzko and te Riele's same underlying methodology, extended slightly further), confirming that neither the original strong form nor this weaker relative provides a viable route to the Riemann Hypothesis after all.
The Mertens function M(n) grows so slowly, and so seemingly erratically (drifting positive and negative repeatedly, roughly like a random walk), that the specific mechanism eventually causing it to exceed √n — subtle correlations among the phases of the zeta function's complex zeros, accumulating coherently only after an astronomically large number of terms — simply couldn't manifest itself within any range of n ever directly, numerically checked by computer, even using the most powerful computational resources available at the time of the 1985 disproof (or, for that matter, since). This makes the Mertens Conjecture a particularly clean and well-understood case study in exactly why finite numerical verification, however extensive, can never substitute for a genuine mathematical proof.
The Mertens episode is frequently invoked by number theorists as a note of appropriate caution when discussing other currently open conjectures supported by extensive numerical evidence — including the Goldbach Conjecture (verified to 4×10¹⁸) and the Riemann Hypothesis itself (with over 10 trillion zeros verified to lie on the critical line). None of this diminishes the genuine evidential value such extensive computation provides, but Mertens stands as concrete proof that overwhelming numerical support, even sustained across a full century of mathematical confidence, is not logically equivalent to — and can occasionally still be overturned in contradiction to — an actual rigorous proof.