Euler's equations of inviscid motion are the system of partial differential equations describing fluid flow in the absence of viscosity, given by
|
(1)
|
where
is the fluid velocity,
is the pressure, and
is the fluid density.
The regularity problem for the three-dimensional incompressible Euler equations asks whether smooth divergence-free initial data can produce finite-time
blow-up. Alpöge and Buckmaster (2026a) claim finite-time blow-up on
from smooth axisymmetric initial data with swirl and a force that is smooth in space
and time and supported in a fixed solid torus. In their
construction, the circulation and meridional velocity stay bounded, while the circulation
gradient and full vorticity become unbounded and
|
(2)
|
Buckmaster (2026) reports substantial assistance from Claude and Codex across the project. An accompanying Lean 4 development formalizes the claim, but independent human verification had not been reported as of Sep. 9, 2026 (Alpöge and Buckmaster 2026b).
OpenAI (2026ab) announced a proposed resolution on Sep. 8, 2026. Its paper claims a smooth, divergence-free initial velocity having compact
support on whose maximal smooth solution has finite lifetime. The claimed
solution has unbounded velocity gradient and satisfies the divergent vorticity
integral condition in the Beale-Kato-Majda
criterion.
Coiculescu (2026ab) reported nonuniqueness for the unforced two-dimensional incompressible Euler equations on .
The construction gives a radial nonnegative initial vorticity
belonging to
for every finite
, but not to
, with two distinct renormalized weak solutions.
Weak solutions satisfy the equations after integration against smooth
functions with compact support. Renormalization
additionally requires the transport equation to hold for smooth functions of the
vorticity. One solution is stationary. The other has
three-fold rotational symmetry and preserves the distribution of vorticity
values. The solutions are distinct in
for every
on sufficiently long time intervals. The velocity is
not square-integrable, so this is not a finite-energy nonuniqueness result.
Coiculescu (2026ab) reports Claude Opus 5.5 assistance with detailed proofs, estimates, certificate computation, and formalization. Part II (Coiculescu 2026b) gives the nonlinear construction and its computer-assisted estimates. Lean formalizes the analytic implication and the soundness of the certificate checker. The recorded evaluation of the certificate also relies on the Lean compiler, runtime, and GMP. Paperlane (2026) independently checked that the Lean project compiled at the earlier revision 1df5e7953d9e.
Choi (2026) obtained finite-energy nonuniqueness in two dimensions. The initial vorticity has compact support,
has integral 0, and belongs to for every finite
.
Two distinct renormalized solutions with these same initial data have velocities
continuous in time in
and preserve the
norms of the vorticity.
This does not assert nonuniqueness for bounded initial vorticity.
Choi (2026) reports AI assistance with the technical proof
of the quadrupole-patch cascade.
Hou et al. (2026b) established a sharp envelope criterion within a specified class of stationary arrays of disjoint radial zero-circulation vortices. If the vorticity
norms are bounded by ,
with
and
finite and nondecreasing, the threshold is
|
(3)
|
Divergence gives uniqueness in that class even against bounded energy-admissible weak velocity solutions. Convergence permits a compactly supported datum in the class with two distinct unforced evolutions described by Lagrangian flows. The criterion is not asserted for arbitrary vorticity.
Hou et al. (2026a) also supplied an independent construction of nonuniqueness for each prescribed ,
with initial vorticity in
and two evolutions on
. Both are strongly continuous in these spaces and
exactly transported by their measure-preserving regular Lagrangian flows. The datum may depend
on
, so this does not strengthen Choi's all-finite-
, global-time result above. Hou et al. (2026b)
credit GPT-6 with the dyadic comparison and transfer estimates used in the envelope
analysis.