Rosensweig Instability (Ferrofluid Spikes)

The intuition — why a flat surface gives up

Pour a teaspoon of ferrofluid — a colloid of nanometre-scale magnetite particles suspended in oil — into a dish and it sits flat, like any other liquid. Bring a strong magnet up underneath, so the field points straight up through the surface, and within a fraction of a second the surface bristles into a regular forest of glistening black spikes. This is the Rosensweig instability, also called the normal-field instability.

The reason is an energy competition. A ferrofluid is far more magnetically permeable than the air above it. Field lines "prefer" to run through the fluid, so the system can lower its magnetic energy by letting fluid bulge up into the field — every peak that pokes into the field region concentrates field lines and is rewarded with lower magnetic energy. That magnetic reward grows with the square of the field.

Two forces resist. Gravity charges an energy cost for lifting dense fluid above the mean level. Surface tension charges a cost for the extra area of a corrugated surface. As long as the magnetic reward is smaller than these costs, the flat surface wins. Crank the field up, and at a sharp threshold the magnetic term overtakes both — the flat surface is no longer the lowest-energy state, and peaks erupt.

How it works — Maxwell stress at the interface

The physics lives at the fluid–air interface. A magnetized fluid in a field experiences a magnetic surface stress (a component of the Maxwell stress tensor) that pulls the interface outward along the field. Where the surface is perpendicular to the field, that stress points straight up, of magnitude on the order of

p_mag ≈ ½ μ₀ M²

where μ₀ = 4π×10⁻⁷ T·m/A is the permeability of free space and M is the fluid's magnetization (its magnetic moment per unit volume). This negative pressure tries to draw fluid up out of the surface.

A small ripple on the surface, of wavenumber k = 2π/λ, either grows or decays depending on whether the destabilizing magnetic stress can overcome the restoring stresses. The linear stability analysis (Cowley & Rosensweig, 1967) gives the energy balance per unit area for a ripple as a sum of three competing terms:

  gravity (stabilizing)      ~  Δρ g / k       grows as λ grows
  surface tension (stabil.)  ~  σ k            grows as λ shrinks
  magnetism (destabilizing)  ~  μ₀ M² · f(k)   peaks at intermediate k

Gravity favors long, flat undulations; surface tension favors no undulation at all; the magnetic term is strongest at an intermediate wavelength. The surface goes unstable first at the wavelength where the two stabilizers are equally weak — the capillary length — and that fixes the spike spacing.

The governing equations

Critical wavelength. The fastest-growing (and first-unstable) mode is set purely by the balance of gravity against surface tension:

k_c = √(Δρ g / σ)          (critical wavenumber)
λ_c = 2π / k_c = 2π √(σ / (Δρ g))   (critical / spike spacing)

Here Δρ is the density difference between the ferrofluid and the medium above it (≈ the fluid density, since air is negligible), g = 9.81 m/s² is gravitational acceleration, and σ is the surface tension. Notice the field does not set the spacing — only how strongly the pattern is driven.

Critical magnetization. Spikes appear once the magnetization reaches a critical value. For a fluid of relative permeability μ_r, the linear theory gives the critical condition

M_c² = (2 / μ₀) · (μ_r + 1) / μ_r · √(σ Δρ g)

The right-hand side √(σΔρg) is the same geometric mean of the two stabilizers that appears everywhere in capillary–gravity physics. The factor (μ_r+1)/μ_r → 1 for a very permeable fluid and → 2 for a weakly magnetic one.

Magnetic Bond number. Collapsing all of this into one dimensionless control parameter gives the magnetic Bond number — the ratio of the destabilizing magnetic stress to the stabilizing capillary–gravity stress:

N_Bo = μ₀ M² / (2 √(σ Δρ g))

The flat surface is stable for N_Bo below a critical value of order 1, and unstable above it. N_Bo is the magnetic analog of the ordinary Bond number Bo = Δρ g L² / σ that compares gravity to surface tension for a drop of size L.

Regimes and conditions

• Below threshold (N_Bo < 1). Flat surface. Small ripples decay. The fluid simply gets drawn slightly toward the strongest part of the field but stays smooth.
• At onset. A hexagonal pattern appears abruptly with finite amplitude — the transition is subcritical. The peaks have a definite spacing λ_c from the moment they appear.
• Hysteresis band. Raise the field to form spikes, then lower it — the spikes survive below the field where they first appeared. Flat and spiked states are bistable over a range of fields.
• Well above threshold. Spikes grow taller and sharper. At high fields the hexagonal lattice can rearrange into a square lattice, then back to hexagons as the field changes — a documented hexagon→square→hexagon re-entrant sequence.
• Field orientation matters. A field normal to the surface gives the spiky Rosensweig pattern. A field parallel (in-plane) to the surface actually stabilizes it, flattening waves rather than raising spikes.

Rosensweig vs. other interfacial instabilities

Property	Rosensweig (normal-field)	Rayleigh–Taylor	Kelvin–Helmholtz
Driving energy	Magnetic field (Maxwell stress)	Gravity on density inversion	Velocity shear between layers
Control parameter	Magnetic Bond number N_Bo	Atwood number / time	Richardson number
Onset	Sharp threshold, subcritical (first-order-like)	Any density inversion is unstable	Shear above a critical value
Final state	Stationary ordered hexagonal lattice	Runaway mushroom plumes, mixing	Rolling cat's-eye vortices, then turbulence
Length scale	Fixed: λ_c = 2π√(σ/Δρg)	Grows with time; mixing-layer width ∝ Agt²	Set by shear-layer thickness
Reversible?	Yes — remove field, spikes collapse (with hysteresis)	No — irreversible mixing	No — irreversible mixing
Stabilized by	Gravity + surface tension	Surface tension + viscosity (slows growth)	Stable stratification + surface tension

Worked example — spacing and field for a real ferrofluid

Take a typical kerosene-carrier ferrofluid: surface tension σ ≈ 0.025 N/m, density ρ ≈ 1200 kg/m³ (so Δρ ≈ 1200 kg/m³ against air), relative permeability μ_r ≈ 2.

Spike spacing.

λ_c = 2π √(σ / (Δρ g))
    = 2π √(0.025 / (1200 × 9.81))
    = 2π √(2.12×10⁻⁶)  m
    = 2π × 1.46×10⁻³  m
    ≈ 9.1 mm

So peaks should sit about 9 mm apart — exactly the centimetre-scale spacing seen in tabletop demonstrations.

Critical magnetization.

M_c² = (2/μ₀) · (μ_r+1)/μ_r · √(σ Δρ g)
√(σ Δρ g) = √(0.025 × 1200 × 9.81) ≈ √294 ≈ 17.2 Pa   (σΔρg has units Pa², so its root is in Pa)
(μ_r+1)/μ_r = 3/2 = 1.5
M_c² = (2 / 4π×10⁻⁷) × 1.5 × 17.2 ≈ 4.1×10⁷  (A/m)²
M_c  ≈ 6.4×10³ A/m  ≈ 6.4 kA/m

A few kiloamps per metre of magnetization is well within reach of a strong permanent magnet held a centimetre below the dish — which is why this is a kitchen-table demonstration, not a lab-only effect.

Real-world numbers and applications

• Loudspeakers (the biggest market). A drop of ferrofluid in the gap around a speaker's voice coil sits in the magnet's field and stays put by magnetic attraction. It conducts heat away from the coil — raising power handling by tens of percent — and damps unwanted resonances. Hundreds of millions of speakers ship with ferrofluid; here the engineering job is to keep the fluid below the Rosensweig threshold so it doesn't spike out of the gap.
• Rotary seals. Ferrofluid trapped by a permanent magnet forms a self-healing liquid O-ring around a rotating shaft, holding a vacuum or pressure difference with near-zero friction — standard in semiconductor wafer-handling and disk-drive spindles.
• Heat transfer and energy harvesting. Because magnetization drops with temperature (a ferrofluid loses its magnetism at its Curie point), a field can drive thermomagnetic convection — a pump with no moving parts. Spike formation increases surface area, which researchers exploit for enhanced evaporative and convective cooling.
• Soft robotics and adaptive optics. A controllable spike field is a millimetre-scale, electrically addressable surface. Deformable ferrofluid mirrors and reconfigurable micro-grippers use the normal-field instability as the actuation principle.
• Art and metrology. Sachiko Kodama's sculptures and countless demos use the pattern directly; physicists use the hysteresis loop and λ_c as a clean way to measure ferrofluid surface tension and magnetization.

Common misconceptions and edge cases

• "The field sets the spacing." It doesn't. Spacing λ_c is fixed by σ, Δρ, and g — properties of the fluid and gravity. The field controls whether spikes form and how tall they get, not how far apart they sit.
• "It's just iron filings standing up in a field." Iron filings are solid grains aligning with field lines. Here a continuous liquid surface deforms — it's a genuine hydrodynamic instability with surface tension and gravity, not particle alignment. The magnetite particles in a ferrofluid are only ~10 nm and stay colloidally suspended.
• "It runs away like Rayleigh–Taylor." No. The pattern saturates into a stationary, ordered lattice. Surface tension and finite magnetization cap the spike height; the system finds a new equilibrium rather than mixing irreversibly.
• "A sideways field also makes spikes." A field parallel to the surface stabilizes it. Only the field component normal to the interface drives the instability — tilt the field and the pattern weakens.
• "The transition is smooth." It's subcritical with hysteresis — spikes appear at finite amplitude and persist below the onset field once formed, so there's a bistable region where both flat and spiked surfaces are stable.
• "Any magnetic liquid does this." You need a true ferrofluid — a stable colloid with high magnetization and a normal liquid surface tension. Bulk liquid metals or paramagnetic salt solutions are far too weakly magnetic to reach N_Bo ≈ 1 with ordinary magnets.