Stochastic differential equations (SDEs) describe systems driven by random noise, like particles buffeted by thermal fluctuations. A "strong solution" is one where the particle's path is completely determined by the specific noise realization you are given, essentially a deterministic function of the driving Brownian motion. A "weak solution" is more permissive: you only require that some probability space exists where the equation makes sense, without pinning the path to a given noise sample. Knowing when strong solutions exist matters a great deal for simulation, control, and physical modeling, but the question becomes delicate when the equations have rough or singular coefficients.
The first part of the paper shows that strong solutions do exist when the drift and diffusion coefficients are irregular but satisfy certain smallness conditions, expressed through "form bounds." Using energy estimates and a compactness argument on the space of Brownian paths, the authors construct a flow that continuously deforms space and has good regularity properties in the Sobolev sense. This flow is built in an approximation-independent way, meaning it does not matter how you smooth out the rough coefficients to get there, and it yields pathwise unique solutions within a natural class.
The second part is a sharp negative result. A long-standing open question was whether a drift that is merely bounded in time and belongs to the critical Lebesgue space in the spatial variable, combined with the physical constraint that the drift is divergence-free (incompressible), is enough to guarantee a strong solution. The authors show the answer is no: for every dimension two and above, they construct a concrete example where the equation has a unique probability law but no strong solution. The key insight comes from rescaling the problem logarithmically, which converts the question of strong solvability into whether certain random measures concentrate at a single point. Because the system has a positive Lyapunov exponent, nearby trajectories spread apart exponentially rather than clustering, and this spreading rules out the concentration needed for a strong solution.