Fluid Physics · Anno MMXXVI

The D-Bridge

Where real water ends a Navier–Stokes singularity, and why heavy water ends it elsewhere
Sabrina Hackfort · Independent Researcher · ORCID 0009-0007-7458-2487
❦ ❦ ❦

In September 2026 it was reported that the Navier–Stokes equations, driven by a smooth force, can form a singularity in finite time: a vortex that collapses until its speed becomes infinite. The proofs are still being checked. They concern the equations, which treat water as a smooth, silent, incompressible continuum. Real water is none of these on the smallest scales. It trembles with heat, its molecules are held in a network of hydrogen bonds that springs back like a gel for about a picosecond, sound travels through it at a finite speed, and it is made of molecules 0.28 nm apart. Long before infinity, one of these walls must take over. Which one comes first, and does it matter whether the bridges are made of hydrogen or of deuterium?

The Map

Each line is a wall: below it, the Navier–Stokes description of water no longer holds. The collapsing vortex starts at the top and shrinks downward. Its strength is the local Reynolds number c = U·L/ν, which the known constructions keep constant, let grow, or let fall. Where its path first meets a wall, real water takes over.

Water at 25 °C. The faint lines are the walls of the other isotope, for comparison. "c constant" follows the scaling seen in Hou's computations, "c grows" the forced construction of September 2026 (growth exaggerated for visibility; the paper's exponent is below 1/100), "c falls" the constructions with bounded velocity. Heat estimate ΔT ≈ U²/cp, an order of magnitude only.

What the Map Shows

Weak collapses are taken over by thermal noise: the molecules tremble faster than the vortex turns. Strong collapses outrun the speed of sound and stop being incompressible, for very strong ones already at tens of nanometres. Between them lies a narrow window, roughly 0.13 < c < 2, where the hydrogen-bond network is the first to give way, at one to four molecular spacings.

The isotope leaves a fingerprint with a sign. In heavy water the bridges relax about a quarter more slowly and the viscosity is higher, so strong collapses leave the Navier–Stokes world about 18–19 % earlier. Weak collapses leave it about 27 % later, because the denser, more viscous heavy water trembles less relative to the flow. One substance, two kinds of singularity, opposite shifts.

What It Does Not Claim

This is an order-of-magnitude map, not a theorem. That the equations stop describing water does not mean the singularity is healed: for elastic liquid models no such regularisation is proven, and relaxation can even create blow-up. The strength c of the September 2026 constructions is not given as a number in their papers. In the hydrogen-bond window, the scales are so close to the molecular spacing that the continuum picture is doubtful anyway. The relaxation time of heavy water is scaled from its measured isotope ratio, and its speed of sound is interpolated between measurements. In a first search no earlier work treating hydrogen bonds or the H₂O/D₂O difference as the physical end of a Navier–Stokes singularity was found; the search was not exhaustive.

Sources

Hydrogen-bond relaxation 0.84 ps: Schulz, Schlaich, Heyden, Netz, Kappler, Phys. Rev. Fluids 5, 103301 (2020), arXiv 2003.08309 · Viscosity of heavy water: IAPWS R16-17 and Assael et al., J. Phys. Chem. Ref. Data 50, 033102 (2021) · Speed of sound in D₂O: Lago et al., Fluid Phase Equilib. (2020) · Isotope ratio of relaxation ≈ 1.2: Ryzhov et al., arXiv 2509.09462 · Thermal noise: Bandak, Eyink, Mailybaev, Goldenfeld, arXiv 2107.03184 · Physical completion of Navier–Stokes: Braunstein, arXiv 2605.21357 · Continuum limit of water: Bocquet & Charlaix, arXiv 0909.0628 · Blow-up constructions: OpenAI (2026), Alpöge–Buckmaster–Coiculescu arXiv 2609.16470, Córdoba & Martínez-Zoroa arXiv 2309.08495, Hou arXiv 2107.06509 · Viscoelastic blow-up: Hu & Racke arXiv 2307.00987.

Archived

Note, code, data and sources: doi:10.5281/zenodo.23301679 (version 0.1, 11 October 2026). Archive SHA-256 1382b0fb7d89766e9a4c0a8388b43a639fb241d09ef3f7213dc00cfbc80c051c, submitted to OpenTimestamps for a Bitcoin timestamp.