摘要: A long-standing foundational gap in stochastic fluid dynamics is that standard
stochastic Navier–Stokes formulations are typically written with velocity noise pre-
scribed as an external forcing, rather than derived from an intrinsic mechanism. We
close this gap by showing that velocity noise admits a natural endogenous origin: it
is the martingale coefficient of the value function in a conditional stochastic optimal
control problem.
On the Sobolev group Gs = SDiffs(T3) of volume-preserving dif-
feomorphisms, the common-noise martingale coefficient Ψ induces, via the group gra-
dient, the operator Z = −∇GΨ. Its evaluation at the identity, z = Z(e), recovers
the Eulerian velocity-noise coefficient exactly. Crucially, in this construction, particle
diffusion contributes only to the second-order elliptic term of the conditional Hamilton–
Jacobi–Bellman equation and plays no role in determining this coefficient. From this
construction, we derive the conditional stochastic HJB equation, the covariant group
velocity equation, the Eulerian advection–diffusion compatibility conditions, and an
exact necessary-and-sufficient criterion for a time-reversed system to admit a standard
forward Itô representation. We further establish a quantitative gradient obstruction,
construct compatible non-trivial shear fields, and prove a locally stopped finite-rank
interacting-mode theorem with fixed-state verification identities checked by square com-
pletion. Global nonlinear well-posedness, open neighborhoods of compatible perturba-
tions, and martingale-derivative convergence remain open beyond the constructed class