computational fluid dynamics

Air streaming over a wing, blood pulsing through an artery, smoke curling off a candle — fluids move in ways that are easy to watch but maddeningly hard to predict. There is no formula you can plug numbers into to get the flow around a new aircraft. Computational fluid dynamics, CFD, is the craft of predicting that flow by solving the equations of motion of a fluid on a computer, cell by cell and step by step, instead of building a wind tunnel.

The recipe is the whole subject in miniature. You take the governing equations (usually the Navier-Stokes equations, expressing conservation of mass and momentum), chop the region of space into a mesh of small cells, and approximate the equations on that mesh — most commonly with the finite-volume method, which tracks the flux of mass and momentum across each cell face so that what leaves one cell exactly enters its neighbor, or with spectral methods for smooth flows. Then you march forward in time, solving a large sparse linear system at each step. The CFL condition ties the time step to the cell size and flow speed; turbulence and thin boundary layers demand fine meshes and clever models. The output is velocity, pressure, and temperature at every cell, from which you read off lift, drag, heating, and mixing.

CFD designs aircraft and Formula 1 cars, predicts how pollutants disperse in a city, models combustion in engines, and is the engine inside weather and climate codes. Its honest limits are real: a simulation can only resolve features larger than its mesh, turbulence is modeled rather than fully computed in almost all practical runs, and a flow that looks beautifully smooth on screen may be hiding under-resolved physics. CFD results must be validated against experiment and checked for mesh-independence — refine the mesh and confirm the answer stops changing.