From F = m a to a field of velocities
You have spent this whole ladder learning to read PDEs as bookkeeping for change spread over space and time. Fluid dynamics is the grand payoff: the equations of a moving fluid are nothing more than the two oldest laws of physics — conservation of mass and Newton's F = m a — rewritten for a quantity that lives at every point at once. Instead of tracking one particle, you track a whole velocity field u(x, t): at each location x and time t, an arrow telling you how fast and in which direction the fluid is sliding past. Density rho(x, t) and pressure p(x, t) ride along as companion fields. The river is not a thing that moves; it is a field that, point by point, tells the next instant what to do.
The one genuinely new idea, and the source of all the difficulty, is acceleration of a fluid parcel. A clump of fluid changes its velocity for two reasons: time passes (the local field speeds up or slows down, the u_t part), and the clump is carried into a region where the velocity is different (the convective part). Following the parcel, the acceleration is u_t + (u dot grad) u. That second piece is the fluid carrying itself — velocity multiplying its own gradient — and because u multiplies a derivative of u, the equation is nonlinear. This is exactly the quasilinear structure you met when a wave first steepened into a shock: a quantity transported by its own value. Fluids are conservation laws with that same self-transport baked in.
Euler: the perfect, frictionless fluid
Start with the idealization, just as you always have. Drop all internal friction and you get the Euler equations, written down in the 1750s. Mass conservation gives the continuity equation rho_t + div(rho u) = 0 — fluid leaving a tiny box must show up as a drop in the box's density, the same conservation-law balance you wrote for heat and traffic. Newton's law gives the momentum equation: the parcel's acceleration equals the force per unit mass, and for an ideal fluid the only internal force is the push of pressure from high toward low. For an incompressible fluid (density constant, a fine model for water and slow air) the pair collapses to a clean statement: the velocity has no sources or sinks, and pressure instantly rearranges itself to enforce that.
Incompressible Euler (no friction):
u_t + (u . grad) u = - grad p / rho (momentum: F = m a)
div u = 0 (mass: nothing piles up)
Incompressible Navier-Stokes (add friction nu):
u_t + (u . grad) u = - grad p / rho + nu * Laplacian u
div u = 0
^^^^^^^^^^^^^^^^^^^ ^^^^^^^^^^^^^^
nonlinear self-transport viscous smoothing
(steepens, makes shocks) (a heat-equation term)Euler flow is gorgeous but treacherous. With no friction, the only thing opposing the self-steepening nonlinearity is pressure, and for compressible Euler — the equations of gas dynamics — that is not enough: a smooth flow can sharpen until its derivatives blow up in finite time, exactly the shock formation you saw for nonlinear conservation laws. A supersonic jet's bow shock is one of these. And just as there, once a shock appears the classical solution dies and you must pass to weak solutions selected by an entropy condition to keep the answer unique. Euler is the honest first model, but it has no built-in mechanism to dissipate energy, so it cannot, by itself, describe a fluid coming smoothly to rest.
Navier-Stokes: friction is a hidden heat equation
Real fluids stick to themselves. Fast-moving layers drag slow ones along; this internal friction is viscosity, and adding it to Euler gives the Navier-Stokes equations. The new force is proportional to nu * Laplacian u, where nu is the kinematic viscosity. Look closely at that term and you will recognize an old friend: the Laplacian is exactly the operator of the heat equation u_t = k u_xx. Viscosity makes velocity diffuse the way heat diffuses — fast and slow regions average toward each other, sharp velocity gradients are smeared out. So Navier-Stokes is a tug-of-war between two forces you already understand intimately: the nonlinear convection that tries to steepen the flow into shocks, and the diffusion that tries to smooth it back out.
It helps to track the spin of the fluid rather than its speed. Take the curl of the velocity and you get the vorticity, a field measuring how much each tiny parcel is rotating — the swirl you see when a paddle leaves a wake. Rewriting Navier-Stokes in terms of vorticity makes the pressure drop out entirely and exposes the real drama: in three dimensions the convection term can stretch a vortex tube like taffy, and stretching a spinning tube makes it spin faster (angular momentum, like a figure skater pulling in their arms). This vortex stretching is the engine that pumps energy down to ever-smaller scales, and it is unique to 3D — it simply does not exist in two dimensions. That single difference is at the heart of why 2D is tame and 3D is not.
One number decides everything: the Reynolds number
How do you know in advance whether a flow will be glassy-smooth or a churning mess? Astonishingly, it comes down to a single dimensionless ratio. Compare the strength of the nonlinear convection term to the viscous smoothing term, and you get the Reynolds number Re = U L / nu, built from a typical speed U, a typical size L, and the viscosity nu. It is the ratio of inertia trying to make trouble to friction trying to calm things down. When Re is small, diffusion wins: the flow is laminar, orderly, layer sliding over layer. When Re is large, the nonlinearity wins and the flow tips over into turbulence.
The same number explains the whole zoo of behaviours. Honey dripping (tiny Re) is described by the Stokes equations — drop the nonlinear term entirely and you are left with a linear problem nu * Laplacian u = grad p, time-reversible and gentle, the regime of swimming bacteria and lubrication films. A river or an aircraft wing (huge Re) lives at the other extreme. In between, the Reynolds number controls the thickness of the boundary layer, the thin sheet near a solid wall where the fluid is forced from full speed down to zero at the surface. That boundary layer is where drag is born, where flows separate, and where engineering lives or dies.
Turbulence and the frontier that stays open
At high Reynolds number the flow shatters into eddies of every size, and turbulence is the result: irregular, swirling, unpredictable in detail yet shot through with statistical pattern. The picture worth carrying is Richardson's energy cascade — big eddies break into smaller eddies, which break into smaller ones still, passing energy down the chain until, at the tiniest scale, viscosity finally bites and turns the motion into heat. Vortex stretching, the 3D-only effect from earlier, is the mechanism driving energy down this ladder. Turbulence is not noise added on top of a flow; it is the genuine, fully nonlinear solution of the same Navier-Stokes equations, and no closed formula for it exists.
Here is the honest, humbling truth this rung must tell. We have had the Navier-Stokes equations for nearly two centuries, we solve them numerically every day to design wings and forecast weather, and yet for the three-dimensional incompressible case nobody knows whether smooth solutions stay smooth forever. It is conceivable that vortex stretching could, starting from perfectly nice initial data, drive the velocity gradient to infinity at a single point in finite time — a finite-time blow-up — and no proof exists either way. This is the Navier-Stokes existence and smoothness problem, one of the seven Clay Millennium Prize problems, with a million-dollar bounty still unclaimed.
Why is it so hard, when 2D is fully solved? Because in 2D vorticity is just carried around and bounded, so the smoothing of viscosity always wins; in 3D, vortex stretching can in principle outrun the diffusion. The known tools — energy estimates in Sobolev spaces, the same machinery you built earlier — give global weak solutions (Leray, 1934) but cannot guarantee those weak solutions are smooth or unique. The gap between a weak solution that merely exists and a classical one that behaves is exactly the gap this Millennium Problem asks you to close. It is a perfect closing note for this rung: the most practically successful equations in all of physics still hide an open question at their core.