Every tail beat sheds a pair of vortices. Where should a neighbour swim, and when should its tail beat, to use less energy?
Energy saving of one tuned follower for every position behind a single leader (green = saves, red = pays). The dot is your current spacing.
The science
Tail beat. A fish of length L swimming at U beats its tail with amplitude A = 0.2 L at frequency f; efficient swimmers sit near the Strouhal number St = f A / U ≈ 0.3, so f = St U / A.
Wake. Each half period the tail sheds a vortex of alternating sign (a reverse Kármán street, which carries a backward jet and so makes thrust). Each vortex is a Lamb-Oseen vortex, vθ = Γ/(2πr) (1 − e−r²/rc²), with Γ = cΓ U A, a core that grows as rc² = rc0² + 4νet and a circulation that decays over about 2.5 beats. The street convects at 0.85 U relative to the water, so its wavelength is λ = 0.85 U/f and the mean jet speed between the rows is Γ/λ.
Cost model. Relative to a solo fish: P/Psolo = (1 − q) ⟨(1 − ux/U)³⟩ + q ⟨(ḣ − vw)²⟩ / ⟨ḣ²⟩. The first term is body drag in the induced streamwise flow ux (averaged along the body); the second is the lateral work of the tail against the wake's lateral flow vw at the tail, with ḣ = 2πfA cos(ωt + φ) and q = 0.5 the share of lateral work in the solo cost. A tail that beats with the incoming wake flow does less work. Saving = 1 − P/Psolo.
What the model shows. A fish directly behind another swims into its backward jet and pays a lot (line, rectangle). Fish staggered between the wakes of the fish ahead (echelon, diamond) can save a few percent when their tail phase is tuned and lose when it is not. The numbers are model values: cΓ, q, the 0.85 convection factor, the decay time and the eddy viscosity are illustrative assumptions [перевірити за джерелом].
Simplifications. 2-D wake in one horizontal layer (the other layers repeat it without interaction); vortices do not move each other (frozen wake); each fish feels only the fish ahead; no body-vortex collision, no 3-D vortex rings, no turbulence, constant speed, rigid lattice. Real school savings depend on all of these. Virtual animals only; no real fish are involved.
Blender-modelled fish and seabed, Three.js r160. Educational model values.