Fluid dynamics
The Saffman-Taylor Instability: Why Fingers Have a Selected Width
Push water into glycerine between two glass plates a fraction of a millimetre apart, and instead of a clean front you get a single, blunt-nosed finger that swallows almost exactly half the channel — width ratio λ → ½ — no matter how fast you push. That razor-sharp selection, invisible in the leading-order equations and set by a surface-tension term that formally vanishes, is the Saffman-Taylor instability: the morphological instability of the interface when a low-viscosity fluid displaces a high-viscosity one in the thin gap of a Hele-Shaw cell.
First analysed by Philip Saffman and Geoffrey Ingram Taylor in 1958, it is the canonical laboratory realization of a Laplacian growth instability, and its width-selection puzzle became a landmark test case for singular-perturbation "microscopic solvability" theory in the 1980s.
- RegimeLow-viscosity fluid displacing high-viscosity fluid in a thin gap (Hele-Shaw / porous medium)
- Key relationσ = [kV(μ₂−μ₁) − γ b² k³/12]/(μ₁+μ₂); finger ratio λ → ½
- DiscoveredP. G. Saffman & G. I. Taylor, 1958 (Proc. R. Soc. A 245)
- Characteristic scaleFastest wavelength λ* ~ b·(γ/μV)^½ ~ mm; gap b ~ 0.1–1 mm
- Realized inHele-Shaw cell (water/air into glycerine or oil); porous rock; EOR
- Matters forOil recovery, carbon sequestration, chromatography, dendritic/Laplacian growth
Interactive visualization
Press play, or step through manually. The visualization is yours to drive — try it before reading on.
Watch the 60-second explainer
A condensed visual walkthrough — narrated, captioned, under a minute.
What it is and why it matters
The Saffman-Taylor instability is the morphological breakdown of the interface between two immiscible fluids confined to the thin gap of a Hele-Shaw cell — two parallel plates separated by a gap b of order 0.1–1 mm. When a less viscous fluid (say air or water, viscosity μ₁) pushes a more viscous one (glycerine or oil, μ₂ > μ₁), the flat front is unstable: small bulges advance faster, sharpen, and the interface fractures into branching, finger-like protrusions. Push the more viscous fluid into the less viscous one and the front is perfectly stable.
Two features make it a touchstone of pattern-formation physics. First, the depth-averaged flow obeys Darcy's law, so the pressure is harmonic (∇²p = 0) and the growth is Laplacian — mathematically identical to dendritic solidification, electrodeposition, and dielectric breakdown. Second, in a long channel the chaos of competing fingers collapses onto a single steady finger whose width is a sharply selected fraction of the channel, ½. That selection turned out to hinge on a term everyone had been dropping.
The mechanism, step by step
In the thin gap, averaging the Stokes flow across b gives Darcy's law: v = −(b²/12μ)∇p. Incompressibility then makes the pressure harmonic, ∇²p = 0, in each fluid. Because v ∝ ∇p, a Hele-Shaw cell is a two-dimensional analog computer for potential flow — and for any Laplacian growth process.
Now perturb a flat front advancing at speed V with a ripple of wavenumber k. A forward bump pokes into the more-viscous fluid, where pressure gradients are steep; the local velocity ∝ ∇p rises, so the bump advances faster and grows. This is the destabilizing engine: the mobility contrast (μ₂ − μ₁) converts any protrusion into a runaway. The Laplace pressure jump across the curved interface, Δp = γ·κ, opposes it: sharp tips (high curvature κ ~ k²) cost energy, so surface tension γ preferentially damps short wavelengths. The competition sets a most-dangerous wavelength; nonlinearly, the winning bulge outruns and screens its neighbours until one finger remains.
The dispersion relation and characteristic scales
Linearizing about the flat front gives the growth rate of a mode e^(ikx+σt):
σ(k) = [ k V (μ₂ − μ₁) − γ (b²/12) k³ ] / (μ₁ + μ₂).
The first term (linear in k) is the viscous destabilization; the cubic −γk³ term is capillary stabilization. σ > 0 for 0 < k < k_c with cutoff k_c = √[12V(μ₂−μ₁)/(γb²)], and the fastest-growing mode sits at k* = k_c/√3, giving a characteristic finger spacing λ* = 2π/k* ~ b·√(γ/μV) — typically millimetres. Note the crucial point: as γ → 0, k_c → ∞ and σ diverges, so the leading-order problem is ill-posed with no scale at all.
The steady single-finger state is governed by one dimensionless group, the surface-tension parameter B = b²γ/(12μ U W²) (with W the channel width, U the finger speed), or equivalently 1/B = 12·Ca·(W/b)² where the capillary number Ca = μU/γ. Small B (fast, wide, low-tension) is the strongly nonlinear regime where λ → ½.
How it is realized and measured
The workhorse platform is the rectangular Hele-Shaw cell: two flat glass plates, gap b set by a shim, filled with glycerol or silicone oil and displaced by air or dyed water through an inlet. Saffman and Taylor's own 1958 experiments used exactly this, injecting air into glycerine, and found a single steady finger occupying about half the channel width almost independently of speed. The measured observable is the relative width λ = (finger width)/W as a function of the control parameter 1/B.
The experimental law is striking: λ decreases monotonically from near 1 at large B toward a plateau at λ ≈ 0.5 as 1/B → ∞. Tabeling, Zocchi and Libchaber's careful experiments (1987) confirmed λ → ½ and mapped the small deviations, tip-splitting and sidebranching that appear at very small B or with imposed noise. Radial cells (injection from a central hole) show the same instability but with repeated tip-splitting into dense-branched, fractal-like patterns rather than one clean finger.
Where it operates and how it differs from cousins
Any Laplacian growth with a mobility contrast and a curvature-based regularization is in the Saffman-Taylor family: two-fluid displacement in porous rock (Darcy's law holds directly, with permeability replacing b²/12), electrodeposition, and — via the mathematical isomorphism — solidification. The Mullins-Sekerka instability of a solidifying front is the diffusive analog: the harmonic pressure is replaced by a harmonic temperature/concentration field, and Laplace pressure by the Gibbs-Thomson correction. Dendrite tip selection and Saffman-Taylor width selection are governed by the same solvability mathematics.
It is distinct from Rayleigh-Taylor (density-, not viscosity-driven, and no unique tip width) and from Kelvin-Helmholtz (shear-driven). The essential Saffman-Taylor signature is the singular role of surface tension: without it, a continuous one-parameter family of finger shapes of every width λ ∈ (0,1) exists (Saffman-Taylor exact solutions), and the physical width is undetermined. Surface tension is what breaks the degeneracy.
Selection theory, applications and open questions
The deep result — resolved in the 1980s — is how a vanishingly small γ picks λ = ½. Surface tension enters as a singular perturbation: it multiplies the highest derivative, so it never appears in ordinary perturbation theory yet controls the answer. McLean and Saffman (1981) showed numerically that adding γ selects a discrete set of widths; Combescot, Hakim, Dombre, Pomeau and Pumir (1986–88), with Shraiman and with Hong and Langer, solved it analytically as a microscopic-solvability problem — a nonlinear eigenvalue condition arising from exponentially small terms 'beyond all orders' in B. The prediction: the discrete branch with λ → ½ from above, with the deviation scaling as (λ − ½) ∝ B^(2/3), matching experiment.
Practically, viscous fingering is the enemy of efficient oil recovery (water breaks through and bypasses oil) and of CO₂ sequestration and chromatography, driving mobility-control strategies (polymers, foams). Open questions remain: noise-driven tip-splitting and sidebranch statistics, selection in radial and non-Newtonian fluids, and control via time-dependent injection or lifting-plate geometries.
| Instability | Driving contrast | Stabilizing agent | Selected feature |
|---|---|---|---|
| Saffman-Taylor (viscous fingering) | Viscosity jump μ₂ > μ₁ across a displaced front | Surface tension γ (curvature) | Single finger, width ratio λ → ½ |
| Rayleigh-Taylor | Density jump, heavy fluid over light | Surface tension / viscosity | Fastest-growing wavelength λ*, no unique tip width |
| Mullins-Sekerka (solidification) | Undercooling / diffusion field | Gibbs-Thomson surface tension | Dendrite tip radius & sidebranch spacing |
| Rayleigh-Plateau | Surface area of a liquid column | Surface tension (here destabilizing) | Breakup wavelength ≈ 9.0 R |
| Saffman-Taylor, γ → 0 limit | Pure Laplacian growth | (none) | Continuous family λ ∈ (0,1) — degenerate |
Frequently asked questions
Why does the finger occupy exactly half the channel?
In the zero-surface-tension limit there is a continuous family of Saffman-Taylor exact solutions with every width λ between 0 and 1, so ½ is not special mathematically. Adding a small surface tension γ acts as a singular perturbation that selects a discrete set of shapes; the physically stable branch approaches λ = ½ from above as the surface-tension parameter B → 0. The deviation scales as (λ − ½) ∝ B^(2/3), confirmed by experiment.
Which way does the displacement have to go to be unstable?
The front is unstable only when the injected (advancing) fluid is less viscous than the fluid being displaced: μ₁ < μ₂. This makes the growth-rate term proportional to (μ₂ − μ₁) positive. Displacing a low-viscosity fluid with a high-viscosity one (μ₁ > μ₂) gives σ < 0 for all wavenumbers, so the front stays flat and stable.
What role does surface tension play — stabilizing or selecting?
Both. In the dispersion relation γ enters as a −γk³ term that stabilizes short wavelengths, setting the cutoff k_c and the fastest-growing spacing λ*. But its deeper role is selection: because it multiplies the highest spatial derivative, it is a singular perturbation that, through microscopic-solvability theory, picks the unique steady finger width out of a continuous degenerate family. Drop it and the problem is ill-posed.
Why is the flow called Laplacian growth?
Depth-averaging the Stokes flow in the thin gap gives Darcy's law v = −(b²/12μ)∇p, and incompressibility makes the pressure harmonic, ∇²p = 0, in each fluid. The interface moves at a velocity set by the gradient of a harmonic field with a curvature (Laplace-pressure) boundary condition. That same structure governs solidification, electrodeposition and dielectric breakdown, so Saffman-Taylor is the fluid archetype of Laplacian growth.
What is the connection to dendritic solidification?
They are mathematically isomorphic. In solidification the harmonic pressure is replaced by a harmonic temperature or concentration field, and the Laplace pressure γκ by the Gibbs-Thomson curvature correction to the melting point. The tip-width (or tip-radius) selection in both is set by the same singular-perturbation solvability condition, which is why the Saffman-Taylor finger became the clean two-dimensional test bed for dendrite selection theory.
What sets the finger spacing in the initial, linear stage?
Linear stability gives a growth rate σ(k) = [kV(μ₂−μ₁) − γb²k³/12]/(μ₁+μ₂). This is positive up to a cutoff k_c = √[12V(μ₂−μ₁)/(γb²)] and peaks at k* = k_c/√3, so the initial fingers appear with spacing λ* = 2π/k* ~ b√(γ/μV), typically millimetres for a sub-millimetre gap. Nonlinear competition then thins this down to a single steady finger.