One line in the paper's file listing gives away how the proof works: input_csv/ag_coeffs_L5000_N65536_dec15_cert5e-15.csv. That is a table of 65,536 numbers, each carrying a certified error bound of about five parts in a quadrillion, and it sits alongside a Python verifier and eight folders of JSON output. The mathematics it supports is a claim about how a certain kind of wave collapses.
Jeongheon Park, writing in a 121-page preprint posted to arXiv on July 30, studies what is called the focusing mass-critical half-wave equation in one dimension. It is a cousin of the equations physicists use to describe how waves travel and interact, but with a twist: the operator governing propagation is nonlocal, meaning the behavior at any one point depends on the whole line at once, not just on the immediate neighborhood. That single feature is what makes the problem hard, and it shapes everything that follows.
The question is blow-up. Some solutions to equations like this one stay smooth forever. Others concentrate: the wave piles up on itself, the amplitude climbs, and at some finite moment in time the solution ceases to exist in any ordinary sense. Mathematicians want to know which starting conditions lead to which fate, and how fast the collapse happens when it does.
Park's answer covers a specific family of starting waves: those that are even (symmetric about the origin), that carry negative energy, and whose mass sits just slightly above a benchmark value called the ground-state mass. For every such starting wave, he proves, the solution blows up in finite time. He also bounds the rate. Writing $T$ for the moment of collapse, a particular measure of the solution's size stays below roughly $|\log(T-t)|^{1/4}/\sqrt{T-t}$ as $t$ climbs toward $T$. The dominant behavior is the square root: as the remaining time shrinks by a factor of a hundred, the size grows by a factor of about ten. The logarithm sitting on top is a slow correction that nudges the growth a little higher.
Park states this is the first finite-time blow-up result for negative-energy solutions to this equation in the near-ground-state regime.
The gap the computer filled
The overall strategy is not new. Park follows the modulation analysis developed by Frank Merle and Pierre Raphaël, a technique for tracking a collapsing wave by decomposing it into a rescaled ground-state profile plus a small remainder, then following how the scaling parameter evolves. Applied to the local nonlinear Schrödinger equation, that machinery is well established.
One piece did not carry over. The method needs what is called a coercivity estimate, essentially a guarantee that a particular quadratic quantity associated with the scaling direction stays positive. For the local Schrödinger equation, that estimate can be proved with ordinary differential equation techniques, because everything is governed pointwise. The half-wave equation is nonlocal, and Park writes that those ODE methods simply do not apply.
His workaround comes in two parts. First, an analytic reduction: using constrained Morse index arguments together with the Birman-Schwinger principle, a classical tool that converts questions about eigenvalues into questions about a related operator, he reduces the open-ended coercivity problem to a finite list of specific numerical inequalities. Second, he certifies that finite list by computer, using interval arithmetic. Interval arithmetic is the discipline of computing not with single numbers but with enclosing ranges, so that every rounding error is carried along explicitly and the final answer comes with a rigorous guarantee rather than a floating-point approximation. To make the ground state itself tractable, Park compactifies the real line, folding an infinite domain into a finite one so a computer can bound the profile everywhere.
The verification code, the input data, and the completed outputs are all posted as ancillary files with the paper.
Why it matters
Blow-up questions are not decoration. They mark the boundary between models that behave and models that break, and in physical settings, blow-up in the mathematics often corresponds to the moment a model stops describing reality and something else takes over. Knowing precisely which initial conditions collapse, and how quickly, tells you where a model can be trusted.
The methodological point may travel further than the specific theorem. Nonlocal equations are common and stubborn, and the standard toolkit for analyzing their collapsing solutions has a hole in exactly the place Park hit. Showing that the hole can be patched by reducing to a finite set of inequalities and then certifying them with validated numerics offers a template other analysts can reuse.
It also puts computer-assisted proof to work in a field where it remains unusual. Interval arithmetic produces genuine proof, not evidence, provided the reduction is correct and the code does what it claims. Checking that is now a matter of reading 121 pages and running the posted verifier, which is what peer review will have to do. Until then, the result stands as a careful preprint with its work shown.