Skip to content

Hydrodynamic Stability

Intuition

A pencil balanced on its tip satisfies the equilibrium equations perfectly — and falls anyway. Equilibrium is cheap; stability is the real question. Fluid mechanics asks it constantly: smooth laminar solutions of Navier–Stokes exist at every Reynolds number, but above some threshold the tiniest perturbation grows exponentially and the flow tips over into something else — waves, vortices, eventually turbulence.

The perturbation recipe

  1. Find an equilibrium \(\bar{\mathbf{u}}(\mathbf{x}), \bar{P}(\mathbf{x})\) (e.g. \(\sum F = 0\), steady laminar profile)
  2. Perturb: \(\mathbf{u} = \bar{\mathbf{u}} + \delta\mathbf{u}'\), \(p = \bar{p} + \delta p'\)
  3. Expand the governing equations in powers of \(\delta\)
  4. Linearize — keep \(O(\delta)\), drop \(O(\delta^2)\)
  5. Normal modes: \(\mathbf{u}' \sim \hat{\mathbf{u}}(y)\, e^{ikx + st}\)
  6. Solve the eigenvalue problem for the growth rate \(s(k)\):
  7. \(\operatorname{Re}(s) < 0\) for all \(k\) → stable
  8. \(\operatorname{Re}(s) > 0\) for some \(k\) → unstable; the fastest-growing mode is what you see in experiments

The linearization decouples wavenumbers: each \(k\) evolves independently, so instability is diagnosed one Fourier mode at a time.

Mechanical archetype

\[m\ddot{x} = \pm kx \qquad \begin{cases} -k\ (\text{valley}) & x \sim e^{\pm i\omega t}\ \text{— oscillates, stable} \\ +k\ (\text{hill}) & x \sim e^{\pm\sqrt{k/m}\,t}\ \text{— grows, unstable} \end{cases}\]

Every fluid instability is this, dressed in field theory: the sign of the "restoring force" for each mode determines its fate. (The inverted pendulum is the mechanical soul of Rayleigh–Taylor.)

Worked example: a toy reaction–diffusion model

\[\partial_t f = f - \frac{f^2}{\lambda} + \frac{1}{\lambda}\partial_y^2 f, \qquad f(0) = f(1) = 0\]

Equilibrium: \(f = 0\). Linearize (\(f = 0 + \delta f'\)): \(\partial_t f' = f' + \frac{1}{\lambda}\partial_y^2 f'\). Normal modes respecting the boundary conditions: \(f' = C\sin(n\pi y)\,e^{s t}\), giving

\[s_n = 1 - \frac{n^2\pi^2}{\lambda}\]
  • Mode \(n\) unstable iff \(\lambda > n^2\pi^2\) — a clean threshold in the control parameter.
  • For \(\lambda = 500\): \(n = 1\)–\(7\) unstable (\(49\pi^2 \approx 483 < 500\)), \(n = 8\) stable (\(64\pi^2 \approx 632 > 500\)).
  • The most unstable mode (\(n = 1\)) grows fastest and dominates what you observe.

This little model rehearses every step used on the real linearized Navier–Stokes system — including the emergence of thresholds and pattern selection.

Linearized Navier–Stokes

For a unidirectional base flow \(\bar{\mathbf{u}} = (\bar{u}_x(y), 0, 0)\), the \(O(\delta)\) momentum equation is

\[\rho\left[\partial_t \mathbf{u}' + (\bar{\mathbf{u}}\cdot\nabla)\mathbf{u}' + (\mathbf{u}'\cdot\nabla)\bar{\mathbf{u}}\right] = -\nabla p' + \mu\nabla^2\mathbf{u}', \qquad \nabla\cdot\mathbf{u}' = 0\]

with no-slip \(\mathbf{u}' = 0\) at walls. Normal modes turn this into the Orr–Sommerfeld eigenvalue problem; its unstable eigenvalues mark the transition Reynolds numbers.

Common mistakes

  • Confusing equilibrium with stability — solving \(\sum F = 0\) says nothing about what perturbations do.
  • Keeping quadratic perturbation terms in a linear analysis (they matter later — for saturation — but not for onset).
  • Forgetting that linear theory only governs onset. Once amplitudes grow, nonlinear terms take over; linear theory can't predict the final state.

Knowledge graph position

Prerequisites: Navier–Stokes, Material derivative, Fourier methods. Leads to: Kelvin–Helmholtz, Rayleigh–Taylor, Turbulence.

Quiz

Q1 (computational). In the toy model, what is the critical \(\lambda\) at which the \(n = 2\) mode first becomes unstable?

Answer

\(\lambda = 4\pi^2 \approx 39.5\) (from \(s_2 = 1 - 4\pi^2/\lambda = 0\)).

Q2 (conceptual). Why does the most unstable mode, rather than all unstable modes equally, dominate the observed pattern?

Answer

Amplitudes grow like \(e^{s_n t}\); exponential separation means the largest \(s_n\) wins by an exponentially widening margin, until nonlinearity saturates it.

Q3 (multiple choice). Linear stability analysis can predict:

  • (a) the fully developed turbulent state
  • (b) the onset threshold and initially dominant wavelength
  • (c) the final amplitude of the instability
  • (d) nothing once viscosity is included
Answer

(b). Onset and pattern selection are linear questions; saturation and the end state are nonlinear.