Rayleigh-Taylor instability calculator
Type the two fluids, the surface tension and the size of your box. You get the Atwood number, the growth rate of every wave the box can hold, the critical and fastest-growing wavelengths, how long a small ripple takes to grow, the late-time bubble speed, and the Courant number of your time step. Free, no sign-up, runs in your browser.
It opens on the Rayleigh-Taylor challenge of Skill-Lync's Ansys Fluent course: water over air in a 20 mm box. The second tab checks the Atwood numbers and the set-up in your report. Skill-Lync ran the same box in OpenFOAM to check the formulas.
Sharp 1984 · Lamb 1932 · Hide 1955 · Goncharov 2002 · IAPWS 2014 · checked in OpenFOAM v2412 · Updated 20 Nov 2026
Water and air as most reports take them from Fluent's database (998.2 and 1.225 kg/m³; check Materials > Fluent Database in your version). σ is water-air at 20 °C from the IAPWS formula. A box with side walls holds half-wavelengths, so its longest wave is twice the width; periodic sides hold whole ones. The ripple is the size of whatever starts the instability: a set-up perturbation, or numerical noise if you start from a flat interface.
Set the case above to what you simulated (case 2: pick it in Start from), then type the numbers from your report.
- Surface tension stabilises every wavelength shorter than 17.1 mm; the fastest-growing wavelength is 29.7 mm (37.2 1/s, two deep layers, no viscosity).
- A 20 mm box with walls holds wavelengths 2L/n: 40 mm, 20 mm, 13.3 mm… 2 of them can grow; the fastest is 40 mm at 35.3 1/s.
- From 0.001 mm, the fastest box mode needs 0.254 s to reach 10% of its wavelength (linear theory).
- Late time: on the longest box wave (40 mm) bubbles rise at about 0.144 m/s in 2-D (Goncharov); at A g, a spike falls the 20 mm light layer in 0.0639 s and reaches 0.626 m/s.
- Courant number at Δt = 0.005 s and 0.5 mm cells: 1.44 at the bubble velocity, 6.26 at the free-fall velocity. Above 1 the interface crosses more than a cell per step; cut Δt or use variable time stepping.
- Flow time 5 s = 78.2 free-fall times of the light layer. Most of it is after the fluids have swapped: the instability itself is over in the first few tenths of a second.
| Mode | λ | Cells per λ | Growth n (1/s) |
|---|---|---|---|
| 1 | 40 mm | 80 | 35.3 |
| 2 | 20 mm | 40 | 28.5 |
| 3 | 13.3 mm | 27 | stable, oscillates at 54.8 rad/s |
| 4 | 10 mm | 20 | stable, oscillates at 109.1 rad/s |
| 5 | 8 mm | 16 | stable, oscillates at 166.1 rad/s |
| 6 | 6.67 mm | 13 | stable, oscillates at 227.4 rad/s |
What is the Rayleigh-Taylor instability challenge?
A heavy fluid resting on a light one is in equilibrium, but an unstable one: any ripple on the interface lets some heavy fluid sink and some light fluid rise, which releases potential energy and makes the ripple bigger. The ripple grows exponentially, then turns into rising bubbles and falling spikes that roll up into mushrooms. That is the Rayleigh-Taylor instability. It shows up in water held upside down in a glass, in supernova remnants and in inertial-confinement fusion capsules.
It is the multiphase challenge of Skill-Lync's Introduction to GUI based CFD using Ansys Fluent course, and 663 distinct public student reports cover it (437 titled "Week 5 - Rayleigh Taylor Instability"). The brief: simulate it with the volume of fluid (VOF) model for two mesh sizes, with a 0.5 mm base mesh, and explain the effect of the mesh; run it once more with water over a user-defined fluid of 400 kg/m³ and 0.001 kg/m·s on the refined mesh; define the Atwood number for both cases and explain how it changes the behaviour. Almost every report builds the same box: two 20 mm squares, water on top, air below, a flat interface between them, walls all round.
What is the Atwood number for water over air and over the 400 kg/m³ fluid?
0.9975 for water over air and 0.4278 for water over the 400 kg/m³ fluid. The Atwood number measures how different the two densities are, from 0 (the same fluid) to 1 (a fluid over vacuum). Linear growth goes as √A, so the second case grows 1.53 times slower:
| Case | ρh, ρl (kg/m³) | A | 40 mm wave, no σ | Bubble speed, 2-D |
|---|---|---|---|---|
| 1: water over air | 998.2, 1.225 | 0.9975 | 39.1 1/s | 0.144 m/s |
| 2: water over the 400 kg/m³ fluid | 998.2, 400 | 0.4278 | 25.6 1/s | 0.112 m/s |
So in case 2 the ripple needs 1.53 times as long to reach the same size. Bubbles rise about 23% slower, and the spikes change most: at A close to 1 a spike of water falls almost freely through the air, while at A = 0.43 the light fluid it pushes aside holds it back, so bubbles and spikes look more alike (at A = 0 they would be mirror images). Most reports get the numbers right: 0.9975 and 0.4278 (0.4286 with water rounded to 1,000 kg/m³). 23 print 0.996 for case 1, which the densities do not give.
How fast does a ripple grow, and which wavelengths does surface tension stop?
Water over air grows fastest at a 29.7 mm wavelength, 37.2 1/s, and surface tension holds every wave shorter than 17.1 mm flat. In the course's 20 mm box that leaves only two waves that can grow, 40 mm and 20 mm long. Here is where those numbers come from.
Write the ripple as a wave of wavelength λ. In linear theory, for two deep fluids without viscosity, it grows as ent with (Sharp 1984, after Rayleigh and Taylor; Lamb's interfacial waves with gravity reversed):
Gravity drives it (the first term); surface tension σ resists bending the interface, and resists short waves most (the k³ term). Setting n = 0 and maximising n give the two lengths that matter:
For water over air at 20 °C (σ = 0.0727 N/m, IAPWS) λc = 17.1 mm and the fastest wave is 29.7 mm, growing at 37.2 1/s. Now compare that with the box. Side walls only allow a wave that meets them square on, a cosine with a whole number m = 1, 2, 3… of half-waves across the width L:
A 20 mm box holds 40 mm, 20 mm, 13.3 mm and shorter. With surface tension only the first two are longer than λc, so only two modes can grow, at 35.3 and 28.5 1/s; the 13.3 mm wave and everything shorter just oscillates. The box is too narrow for the fluids to choose their own wavelength: the 40 mm wave wins because it is the longest one that fits.
Two refinements, both in the calculator. The layers are only 20 mm deep, between a floor and a lid, which replaces the densities in the denominator by ρ coth(kh) (Lamb 1932, Art. 231, with surface tension as in Livescu 2004); for the 40 mm wave in this box it changes n by 0.2%. And viscosity slows short waves, which the calculator takes from Hide's (1955) approximation with ν the density-weighted mean viscosity, as given by Mikaelian (2019), who puts it within 10% of the exact viscous result and notes it is an upper bound. n0 is the inviscid rate, surface tension and depth included:
Why do Rayleigh-Taylor results change with the mesh?
Because without surface tension only viscosity holds back the short waves, and only below about 0.596 mm, so on any practical mesh the shortest wave the mesh can hold grows fastest. A finer mesh holds finer waves, so it shows finer fingers, sooner. With water-air surface tension switched on, the 0.5 and 0.25 mm meshes gave the same large-scale picture in Skill-Lync's OpenFOAM runs.
Turn surface tension off, as the course set-up does (no phase interaction is set) and as 662 of the 663 reports ran it, and the k³ term goes. Then n = √(Agk) climbs without limit as the wave gets shorter. The shortest wave a mesh can hold is two cells long, so the fastest-growing wave is set by the mesh: on 0.5 mm cells, a 1 mm wave grows at 211 1/s with viscosity, against 39.1 1/s for the 40 mm wave; on 0.25 mm cells, 225 1/s. Viscosity does stop the shortest waves, but only at about the size of these cells: by Hide's formula the fastest wave without surface tension is 0.596 mm long (227 1/s), and resolving it with ten cells would take 0.06 mm cells, about 230,000 of them in the 20 × 40 mm box. Every practical mesh is coarser, so refine it and finer fingers appear sooner (numerical diffusion in the VOF scheme also damps the shortest waves). That is what the reports see ("more spikes", "finer behaviour" on the finer mesh), and why "the results depend on the mesh" is the answer to the brief's question, not a sign the study went wrong.
The set-up adds a second problem. None of the 663 reports describes putting a ripple on the interface: it is patched flat. Something must still start the instability, and in a flat start that is round-off and the asymmetries of the mesh and solver. A ripple of 10−6 mm needs 0.45 s to reach 4 mm on the 40 mm mode; one of 10−3 mm needs 0.25 s. So when, and where, the first fingers appear depends on the noise, which changes with the mesh. Comparing onset times between meshes compares noise.
Skill-Lync ran the course box in OpenFOAM's interFoam (VOF, v2412) with the same small random ripple, the 20 longest box modes with amplitudes of 0.005 to 0.01 mm, on 0.5 and 0.25 mm meshes, with surface tension off and on. At t = 0.18 s:
At t = 0.18 s. Without surface tension the two meshes disagree on the number, size and place of the fingers, and even on which way the 40 mm wave leans (its amplitude is −5.3 mm on the 0.5 mm mesh, +2.2 mm on the 0.25 mm one). With surface tension both show that single wave leaning the same way, the finer mesh about half an e-folding time ahead (3.1 against 1.9 mm); every shorter wave in the ripple stayed under 0.5 mm. Water dark, air light, half-tone cells hold both.
Do the formulas match a CFD run?
Yes, closely. In Skill-Lync's OpenFOAM runs of the course box, a 40 mm ripple with surface tension grew at 35.5 1/s against 35.3 1/s from theory, the same on both meshes, and a 13.3 mm ripple, shorter than λc, oscillated instead of growing, as predicted.
To test the formulas, Skill-Lync gave the interface a single cosine ripple of 0.05 mm (far smaller than the wave, so linear theory applies) and measured how fast it grew. Released from rest, linear theory says the amplitude follows
so n was fitted to the amplitude of that mode (the interface height projected on the cosine) until it reached a tenth of 1/k. Side walls free-slip unless noted, floor and lid no-slip, Fluent's water and air properties:
| Run | Mesh | interFoam | Theory | Difference |
|---|---|---|---|---|
| 40 mm wave, σ on | 0.5 mm | 35.5 1/s | 35.3 1/s | +0.5% |
| 40 mm wave, σ on | 0.25 mm | 35.5 1/s | 35.3 1/s | +0.6% |
| 40 mm wave, σ on, no-slip side walls | 0.25 mm | 35.4 1/s | 35.3 1/s | +0.1% |
| 40 mm wave, σ off | 0.25 mm | 38.6 1/s | 39.1 1/s | −1.2% |
| 13.3 mm wave, σ on (shorter than λc) | 0.25 mm | ω 50.4 rad/s | ω 54.8 rad/s | −8.0% |
| 13.3 mm wave, σ off | 0.25 mm | 66.9 1/s | 67.7 1/s | −1.1% |
Box 20 mm wide and 40 mm tall, interface at mid-height, ripple amplitude 0.05 mm; Fluent database values for water (998.2 kg/m³, 0.001003 kg/m·s) and air; σ = 0.0727 N/m; interFoam with MULES, Euler time stepping, maximum Courant number 0.2 and a capillary time-step limit. Growth rates fitted while ka < 0.1; the stable wave fitted as a₀ cos ωt up to 0.1 s, after which the 40 mm wave, grown from round-off, leaks into it. Theory: the finite-depth relation with Hide's viscous correction for growing waves, the inviscid frequency for the stable one. The stable wave oscillates about 8% slower than that formula; the growth rates, which the calculator is for, agree within about 1.5%.
What time step should a Rayleigh-Taylor VOF run use?
One that keeps the Courant number near 1 or below at the speed the fluid actually reaches, about 0.15 to 0.6 m/s in the course box. On the 0.5 mm mesh that means under 1 ms (0.8 ms at the free-fall speed), not the 5 ms most reports use; finer meshes need proportionally less.
Of the 194 reports that state a time step, 142 use 0.005 s; 35 use 0.001 s or less. The Courant number says how many cells the interface moves in one step:
Take the bubble speed, 0.144 m/s for the 40 mm wave: on the 0.5 mm base mesh, Δt = 0.005 s gives Co = 1.44, and on a 0.1 mm mesh 7.2. Water falling 20 mm through air reaches 0.63 m/s, Co = 6.3 on the base mesh. Only 55 reports mention the Courant number and 7 give a value. An implicit VOF scheme runs at these numbers, but the interface smears over several cells per step, which also blurs the comparison between meshes. Variable time stepping on a Courant number of about 1, or below, keeps the interface sharp (Fluent's explicit VOF scheme sub-steps the volume fraction at a default Courant number of 0.25); with surface tension the capillary waves on the smallest cells set a further limit (interFoam needed steps of about 26 µs on the 0.25 mm mesh).
The run length matters less than it seems. 1,000 steps of 0.005 s is 5 s of flow, but water falling freely through the 20 mm air layer takes 0.0639 s. Even started from round-off, the layers swap within the first second or so; the rest of the run is sloshing and settling.
How fast do the bubbles rise after the linear stage?
For the 40 mm wave of water over air, about 0.144 m/s in 2-D. Once a bubble is about as tall as it is wide, it stops growing exponentially and rises at a near-constant speed set by the wavelength and the Atwood number, while the water spikes fall almost freely.
Linear theory holds while the amplitude is small compared with the wavelength, roughly a below 0.1λ. Then the bubbles slow to a near-constant speed. For a single mode, Goncharov's (2002) model gives:
with C = 3 in 2-D and C = 1 in 3-D. At A = 1 in 2-D it reduces to Layzer's (1955) √(g/3k). With many modes the bubbles merge into ever bigger ones and the mixing zone grows with the square of time,
with αb = 0.025 ± 0.003 in the simulations and 0.057 ± 0.008 in the experiments compared by Dimonte et al. (2004); the simulation value may be a lower bound. In a 20 mm box there is no room for that stage: the first bubble reaches the lid at once.
What does the check tab look for?
Skill-Lync read all 663 reports on 20 Nov 2026. The check tab tests:
- The Atwood number: recomputed from your densities; names the usual slips (the heavy density alone in the denominator, a density ratio instead of A, water rounded to 1,000 kg/m³, which is fine if you say so).
- Which fluid is on top: light over heavy is stable, nothing to simulate.
- Surface tension off: then the mesh, not the physics, sets the finger size. Only 1 report sets a surface tension coefficient.
- The time step: the Courant number at the bubble and free-fall speeds, and the flow time against the time the layers need to swap (set Δt and the step count above).
What this calculator does not cover
- Compressibility. The formulas are for incompressible fluids; a gas layer compressed by the fall, or the Fluent "compressible" air some reports select, is a different problem.
- The exact viscous result. Hide's formula is an upper bound on the exact viscous growth rate, within about 10% of it; no-slip side walls slow the box modes a little: 0.5% for the 40 mm wave in the OpenFOAM runs.
- Non-linear growth in detail. The bubble speed and α are models fitted to simulations and experiments, not exact results.
- Richtmyer-Meshkov and Kelvin-Helmholtz. Shock-driven and shear-driven instabilities follow different formulas.
Knowing when a CFD result reflects the physics and when it reflects the mesh is the skill this challenge is really about. It is part of Skill-Lync's Ansys Fluent course.
Frequently asked questions
What is the Atwood number?
The density contrast of two fluids, A = (ρh − ρl)/(ρh + ρl), between 0 (equal densities) and 1 (a fluid over vacuum). Water over air (998.2 and 1.225 kg/m³) gives 0.9975; water over a 400 kg/m³ fluid gives 0.4278. The linear growth rate scales with √A, so the second case grows 1.53 times slower.
What is the growth rate of the Rayleigh-Taylor instability?
For two deep inviscid fluids, n² = A g k − σk³/(ρh + ρl), with k = 2π/λ: a perturbation grows as e^(nt). Without surface tension n = √(Agk), which rises without limit for shorter waves. With water-air surface tension (σ = 0.0727 N/m) the fastest wave is 29.7 mm long and grows at 37.2 1/s, an e-folding time of 27 ms.
What is the critical wavelength for Rayleigh-Taylor instability?
λc = 2π√(σ/((ρh − ρl) g)). Shorter waves are held flat by surface tension; longer ones grow. For water over air at 20 °C it is 17.1 mm, so in a box 20 mm wide only waves 40 mm and 20 mm long (half and one wavelength across the box) can grow.
Why do my Rayleigh-Taylor results change with the mesh?
Because without surface tension only viscosity holds back short waves, and it acts only below about 0.596 mm, near the size of a 0.5 mm mesh's cells. So the shortest wave the mesh can hold, two cells, grows fastest: 211 1/s on a 0.5 mm mesh against 35.3 1/s for the 40 mm wave. Refine the mesh and finer fingers appear. Turning surface tension on in the VOF model sets a physical cut-off, and in Skill-Lync's OpenFOAM runs the two meshes then showed the same single 40 mm wave.
How fast do Rayleigh-Taylor bubbles rise?
Once the bubble is about as tall as it is wide, it rises at a near-constant speed, V = √(2A/(1+A) · g/(Ck)), C = 3 in 2-D and 1 in 3-D (Goncharov 2002). For a 40 mm wave of water over air in 2-D that is 0.144 m/s. For A close to 1 the spikes fall almost freely.
Should I turn on surface tension in Fluent for Rayleigh-Taylor?
For water and air at the scale of millimetres, yes: the capillary cut-off (17.1 mm) is comparable to the 20 mm box, so leaving it out changes which waves grow. In Fluent, enable Surface Tension Force Modeling and set the coefficient under Phase Interaction > Forces in the Multiphase Model dialog (continuum surface force is the default model; every coefficient is 0 until you set one): about 0.0727 N/m for water-air at 20 °C. It also adds a time-step limit: capillary waves on the finest cells must be resolved.
Related on Skill-Lync
- Student project: Week 5, Rayleigh-Taylor instability
- Flow over a cylinder calculator
- Mixing tee calculator
- Mesh quality calculator
- CFD courses compared
- All free engineering tools
Sources
- D. H. Sharp, "An overview of Rayleigh-Taylor instability", Physica D 12 (1984) 3-18, Eqs. 2-6: the growth rate with interfacial tension, η = η(0) cosh(αt) from rest, the critical and fastest-growing reduced wavelengths.
- H. Lamb, Hydrodynamics, 6th ed. (1932), Art. 231 Eq. 11 (two fluids between rigid horizontal planes, the coth form) and Art. 267 Eq. 2 (interfacial tension), with gravity reversed; the finite-depth form with surface tension as in D. Livescu, Phys. Fluids 16 (2004) 118, arXiv:physics/0310072, Eq. 28.
- K. O. Mikaelian, "Exact, approximate, and hybrid treatments of viscous Rayleigh-Taylor and Richtmyer-Meshkov instabilities", Phys. Rev. E 99 (2019) 023112: Hide's (1955) approximate dispersion relation, Eqs. 14-16 with ν = (μ₁ + μ₂)/(ρ₁ + ρ₂); within 10% of the exact result (Menikoff et al. 1977) and an upper bound on it; surface tension through g → g − k²T/(ρ₂ − ρ₁), Eq. 11; γ²max = 2gAkc/(3√3).
- V. N. Goncharov, "Analytical model of nonlinear, single-mode, classical Rayleigh-Taylor instability at arbitrary Atwood numbers", Phys. Rev. Lett. 88 (2002) 134502 (abstract): bubble velocity with C = 3 (2-D) and 1 (3-D). D. Layzer, Astrophys. J. 122 (1955) 1, Eq. 60: V = √(g/3k) between parallel walls.
- G. Dimonte et al., "A comparative study of the turbulent Rayleigh-Taylor instability using high-resolution three-dimensional numerical simulations: the Alpha-Group collaboration", Phys. Fluids 16 (2004) 1668 (abstract): hb ≈ αbAgt², αb 0.025 ± 0.003 (simulations), 0.057 ± 0.008 (experiments).
- IAPWS R1-76(2014), Revised Release on Surface Tension of Ordinary Water Substance: σ = Bτμ(1 + bτ), B = 235.8 mN/m, b = −0.625, μ = 1.256, Tc = 647.096 K; 72.74 mN/m at 20 °C, 71.97 at 25 °C (Table 1).
- Ansys Fluent User's Guide 2024 R2: air 1.225 kg/m³ and 1.7894 × 10⁻⁵ kg/m·s (§8.3.2, §8.4.1); surface tension under Phase Interaction > Forces, coefficients 0 by default (§26.3.10.1); explicit VOF default maximum Courant number 0.25 (§26.3.11). Water-liquid 998.2 kg/m³ as quoted in the Tutorial Guide 2024 R2 §11.4; 0.001003 kg/m·s as the reports list it.
- OpenFOAM v2412 interFoam (opencfd/openfoam-run:2412 image), run by Skill-Lync on 20 Nov 2026: the course box with single-mode and multi-mode ripples, surface tension on and off, 0.5 and 0.25 mm meshes; case files in Skill-Lync's research notes.
- Skill-Lync, all 663 public Rayleigh-Taylor reports, read 20 Nov 2026: set-up, mesh sizes, time steps, Atwood numbers, surface tension.
- Skill-Lync, scripts/verify/rti.py: re-implements every formula (cut-off by bisection, fastest wave by golden-section search, the deep-layer limit, the viscous root, the IAPWS table values) and checks the OpenFOAM runs; the page checks itself against it at build time.