Why does the simulation strategy change so much when the continuum assumption breaks?

Continuum mechanics assumes that density, velocity and temperature can be defined at every point and uses Navier-Stokes for viscosity and heat conduction. As Kn grows, however, wall collisions dominate over molecule-molecule collisions, producing a finite slip velocity at the wall. Above Kn ~ 0.1 the velocity distribution deviates significantly from Maxwellian and Navier-Stokes itself fails to capture the physics. In that regime you must solve the Boltzmann equation directly or use DSMC, which tracks individual molecule collisions stochastically.

What exactly is slip flow?

In ordinary viscous flow (Kn < 0.01) the no-slip condition is used, so the fluid velocity at a wall matches the wall velocity. In the slip-flow regime 0.01 < Kn < 0.1 the mean free path becomes comparable to the boundary-layer thickness near the wall and the tangential velocity at the wall takes a finite value. Maxwell's first-order slip correction reads u_slip = (2-sigma)/sigma * lambda * (du/dn), where sigma is the tangential momentum accommodation coefficient (typically 0.8 to 1). This extension lets you keep Navier-Stokes a little longer and is widely used in MEMS micro-channel design.

How do I reduce the Knudsen number toward the continuum regime?

There are two ways to reduce Kn = lambda / L. (1) Shorten the mean free path. Because lambda is proportional to T / (d^2 P), you can raise the pressure, lower the temperature or use a larger molecule like CO2. (2) Enlarge the characteristic length L. For a MEMS micro-channel with L = 10 micrometers, this tool can find the pressure that brings lambda below 0.1 micrometer so that Kn 1 and guarantee straight-line flight from target to substrate.

Knudsen Number Simulator — Flow Regimes for Rarefied Gases

Q: What is the Knudsen number?

The Knudsen number Kn = lambda / L is the ratio of the molecular mean free path lambda to a characteristic device length L and tells you whether the continuum approximation of fluid mechanics is valid. For Kn 10 the gas is in the free-molecular regime where the Boltzmann equation or DSMC is required. It is particularly important in MEMS micro-channels and semiconductor process chambers where L is in the micrometer to millimeter range.

Q: How do I reduce the Knudsen number toward the continuum regime?

There are two ways to reduce Kn = lambda / L. (1) Shorten the mean free path. Because lambda is proportional to T / (d^2 P), you can raise the pressure, lower the temperature or use a larger molecule like CO2. (2) Enlarge the characteristic length L. For a MEMS micro-channel with L = 10 micrometers, this tool can find the pressure that brings lambda below 0.1 micrometer so that Kn 1 and guarantee straight-line flight from target to substrate.

Compute the mean free path lambda and the Knudsen number Kn = lambda / L in real time from temperature, pressure, molecular diameter and characteristic length, and automatically classify the flow as continuum, slip, transitional or free-molecular. Two visualizations — a scale comparison canvas and a Kn regime map — make the physics of flow-model selection intuitive.

Parameters

Temperature T

log10 Pressure P

log Pa

P = 1000 Pa

Molecular diameter d

Characteristic length L

μm

Defaults: T = 300 K, P = 1000 Pa (medium vacuum), d = 370 pm (nitrogen N2 equivalent), L = 10 μm (MEMS scale). Boltzmann constant k = 1.380649e-23 J/K.

Results

—

Mean free path lambda

—

Knudsen number Kn

—

Flow regime

—

Continuum limit L (Kn=0.01)

Scale comparison (L vs. mean free path lambda)

Blue bar = characteristic length L. Orange bar = mean free path lambda on the same log scale. Background dots = molecules with density proportional to P / T. Background tint shows the regime (green = continuum, yellow = slip, orange = transitional, red = free-molecular).

Kn regime map (flow classification)

Horizontal axis = log10 Kn from -3 to 2. Four colored bands = continuum, slip, transitional and free-molecular. Yellow circle = current Kn. Boundaries at Kn = 0.01, 0.1 and 10.

Theory & Key Formulas

The Knudsen number is the ratio of the mean free path to a device length and quantifies how rarefied the gas appears from the device viewpoint. It is computed via the hard-sphere mean free path.

Definition of the Knudsen number:

$$\mathrm{Kn} = \frac{\lambda}{L}$$

Mean free path (hard-sphere model with the relative-velocity sqrt 2 correction):

$$\lambda = \frac{k\,T}{\sqrt{2}\,\pi\,d^2\,P}$$

Flow-regime boundaries:

$$\mathrm{Kn} < 0.01 \,\,(\text{continuum}),\quad 0.01 \le \mathrm{Kn} < 0.1 \,\,(\text{slip})$$ $$0.1 \le \mathrm{Kn} < 10 \,\,(\text{transitional}),\quad \mathrm{Kn} \ge 10 \,\,(\text{free-molecular})$$

$k = 1.380649\times10^{-23}$ J/K is Boltzmann's constant, $T$ is the temperature [K], $d$ is the molecular diameter [m], $P$ is the pressure [Pa] and $L$ is the characteristic device length [m]. lambda is the mean distance a molecule travels between collisions and Kn is the rarefaction indicator from the device viewpoint.

What is the Knudsen Number Simulator?

🙋

My fluid dynamics textbook says Navier-Stokes is only valid when the continuum assumption holds. Where exactly is that boundary? Air feels like it should be a continuum in any situation.

🎓

That is exactly what the Knudsen number Kn = lambda / L decides. lambda is the molecular mean free path and L is the device length. If Kn < 0.01 textbook Navier-Stokes plus no-slip works, 0.01 to 0.1 needs slip boundary conditions, 0.1 to 10 is transitional and the continuum assumption breaks, and above 10 you are in molecular flow where Boltzmann or DSMC is mandatory. Air at 1 atm in a 1 m duct has Kn ~ 1e-10 (pure continuum), but air at 1000 Pa in a 10 μm MEMS channel gives Kn ~ 0.7 — transitional flow.

🙋

So the defaults of this tool (T = 300 K, P = 1000 Pa, d = 370 pm, L = 10 μm) are deliberately a MEMS medium-vacuum case?

🎓

Yes. lambda = kT / (sqrt 2 pi d^2 P) ~ 6.81 μm, and with L = 10 μm we get Kn = 0.68. This is exactly what happens inside a MEMS accelerometer gap or in narrow voids inside a semiconductor process chamber. In that regime you need extended Navier-Stokes with slip and temperature-jump boundary conditions, or even DSMC and Burnett equations — ordinary CFD gives large errors. Move the L slider in this tool and watch the regime cycle through continuum (green), slip (yellow), transitional (orange) and free-molecular (red) on the Kn map.

🙋

I have heard that sputter chambers in semiconductor processing deliberately run at low pressure. Are they pushing Kn up on purpose?

🎓

Exactly. Sputter deposition and MBE aim for Kn far greater than 1 — fully free-molecular flow. Setting P = 0.1 Pa (log P = -1), d = 250 pm (argon) and L = 30 cm (3e5 μm) in this tool gives lambda of tens of millimeters, Kn ~ 0.1 — surprisingly low. So real systems go down to 1e-3 to 1e-2 Pa (UHV) to push lambda into the meter range and guarantee a straight target-to-substrate flight. In practice the chamber operates around 0.5 to 2 Pa on the transitional/molecular boundary, balancing plasma density and Kn.

🙋

What about ordinary 1 atm air in a MEMS nano-channel with L = 100 nm? Can we still use the continuum assumption there?

🎓

Great catch. Even at 1 atm, lambda is about 67 nm for nitrogen, so a 100 nm nano-channel gives Kn ~ 0.67 — squarely in the transitional regime. That is why slip flow, the Knudsen correction and the Cunningham correction are everywhere in nano-fluidic research. Try P = 1e5 Pa (log P = 5) and L = 0.1 μm in this tool — you should see Kn around 0.6 to 0.7. Bird's DSMC or lattice Boltzmann is needed there, and the naive intuition that gas is always a continuum quietly breaks down.

Frequently Asked Questions

These boundaries are empirical guidelines from experiments and simulations rather than rigorous theoretical values. Kn = 0.01 is where the error of no-slip Navier-Stokes exceeds about 1% and Maxwell's first-order slip correction can extend the model. Kn = 0.1 is where even slip corrections accumulate over 10% error and Burnett equations or DSMC are required. Kn = 10 is where wall collisions completely dominate molecule-molecule collisions and the free-molecular (Knudsen) approximation is sufficient. These boundaries appear in the Schaaf-Chambré chart from the 1960s and are widely used as design guides in semiconductor and vacuum engineering.

A well-known relation is Kn = M / Re * sqrt(pi gamma / 2), where gamma is the specific-heat ratio, M is the Mach number and Re is the Reynolds number. It shows that high-speed, low-density flow (high M, low Re) automatically pushes Kn high. For an atmospheric re-entry capsule at M ~ 25 and Re ~ a few hundred, Kn lands in the 0.1 to 1 transitional/molecular regime. This is why re-entry aerothermal heating cannot be computed with ordinary CFD and requires DSMC or a direct Boltzmann solver.

DSMC was developed by G. A. Bird in the 1960s as a particle-tracking method that directly simulates the Boltzmann equation. Each simulation particle represents 1e6 to 1e20 real molecules, and three steps repeat: (1) free-flight advection over a time step, (2) Monte Carlo selection of collision pairs within each cell with stochastic collision processing, and (3) statistical treatment of wall reflection (specular or diffuse). It applies to any rarefied gas flow with Kn greater than 0.01 and, especially for Kn greater than 1, the cost is lower than continuum CFD. SPARTA, dsmcFoam (an OpenFOAM extension) and Bird's DS2V/DS3V are widely used implementations.

The most striking change is the emergence of pressure dependence in viscosity and thermal conductivity. In the continuum regime (Kn less than 0.01) Maxwell showed that viscosity is independent of pressure because eta = (1/3) rho mean v lambda and rho lambda is constant. Above Kn ~ 0.1, however, wall collisions dominate and the effective viscosity falls with pressure. The same is true of thermal conductivity, which is the working principle behind Dewar flasks (high vacuum interrupts gas conduction). In this tool, sweeping P from 1e5 Pa down to 1 Pa shows large regime changes even with T, d, L fixed. In practice, choose L so that the operating pressure range stays inside one regime.

Real-World Applications

MEMS device design: Micro-accelerometers, micro-gyroscopes and micro-pumps have moving-structure gaps of 1 to 10 μm, and even ordinary air gives Kn around 0.007 to 0.07 — the slip-to-transitional boundary. Viscous damping then deviates significantly from Navier-Stokes, affecting the resonant frequency and Q factor in design. A modified Reynolds equation with a Knudsen correction (Burgdorfer 1959, Fukui-Kaneko 1988) is the industry standard and is essential for hard-disk slider flying-height calculations and MEMS microphone design. In this tool, sweep L from 1 to 10 μm at P = 1e5 Pa and see why MEMS cannot be designed without these corrections.

Vacuum pumps and process chambers: Semiconductor sputter, CVD and epitaxial-growth processes deliberately operate in the free-molecular regime Kn greater than 1 to balance the target-to-substrate molecular flight against surface reactions. A typical PVD sputter chamber with diameter 30 cm requires lambda greater than 0.3 m, which translates to P less than about 0.02 Pa (1e-4 Torr). Set L = 3e5 μm (30 cm) in this tool and sweep the pressure to see where complete free-molecular flow appears. Even lower pressures of 1e-5 Pa are required for ultra-high-vacuum e-beam lithography and MBE.

Spacecraft re-entry aerothermodynamics: When entering the Earth's atmosphere, altitudes above about 80 km are in the rarefied regime (Kn greater than 0.01). The Space Shuttle entry trajectory passes through Kn ~ 10 (free-molecular) at 110 km, Kn ~ 0.1 (transitional) at 85 km and Kn less than 0.01 (continuum) below 60 km. Different aerodynamic models (free-molecular, bridging, Navier-Stokes) must be switched at each layer, and hybrid CFD plus DSMC computations are now standard. This tool helps build intuition for why re-entry heating prediction is hard.

Nanotechnology and gas flow in carbon nanotubes: For carbon nanotubes with inner diameter 1 to 10 nm, even atmospheric gas reaches Kn far greater than 10 — fully free-molecular flow. There, molecule-wall interactions (diffuse versus specular reflection) determine the flow rate via the Knudsen diffusion mechanism. This matters for gas separation membranes, gas sensors and catalyst supports. Try L = 0.001 to 0.1 μm in this tool to see free-molecular flow appear even at 1 atm. The Maxwell accommodation coefficient sigma is usually 0.8 to 1, but in smooth nanotube walls it can drop below 0.5, requiring further refinement of the Knudsen correction.

Common Pitfalls and Notes

The most common misconception is to assume that "atmospheric air is always a continuum in any device". In reality, once L is around 100 nm, even 1 atm air gives Kn ~ 0.7 — already transitional. MEMS, nano-fluidics, catalyst pores and micro-fabricated optics need continuum corrections even at atmospheric pressure. Fix P = 1e5 Pa (log P = 5) in this tool and move the L slider to 0.1 μm. You will see Kn shoot above 0.1. Concluding "it is air so Navier-Stokes is fine" without checking the device size is a fatal mistake at nano-scales.

The second pitfall is to forget that "large Knudsen number does not necessarily mean dilute gas". Kn = lambda / L is a relative quantity, so changing the device size at the same gas density changes Kn. A dense gas in a tiny device can have large Kn, while a dilute gas in a huge device can have small Kn. Air at 1 atm in a micro-channel breaks the continuum assumption, while the extremely dilute gas of deep space, viewed as an Earth-scale astronomical phenomenon, is still continuum. Sweep L and P independently in this tool to verify this independence.

The third pitfall is to over-trust the boundary values 0.01, 0.1 and 10 as exact physical thresholds. They are empirical guidelines and can shift by an order of magnitude depending on gas species, temperature, geometry and surface properties. Smooth walls can show significant slip even at Kn less than 0.1, while rough walls can behave continuum-like above Kn = 0.1. For critical designs, use this tool to estimate Kn, then validate with DSMC or experiments. Kn is a first-order indicator of "which model to use", but the final accuracy of physical quantities is determined by the chosen model itself and the boundary conditions.

Knudsen Number Simulator — Flow Regimes for Rarefied Gases

What is the Knudsen Number Simulator?

Frequently Asked Questions

Real-World Applications

Common Pitfalls and Notes

Related Tools