In short

The mean free path \lambda is the average distance a molecule in a gas travels between successive collisions with other molecules. For a gas in which all molecules have diameter d and the number density is n (molecules per cubic metre),

\boxed{\;\lambda = \frac{1}{\sqrt{2}\,\pi d^2 n}\;}

Derivation sketch: a single "tagged" molecule sweeps out a cylinder of cross-sectional area \pi d^2 (any other molecule whose centre lies inside that cylinder will collide with it); its volume grows as \pi d^2 v t in time t; the number of molecules inside that volume is n \pi d^2 v t, so the average number of collisions per unit time is n \pi d^2 v and \lambda = v / (\text{collision rate}) = 1/(n \pi d^2). The correction factor of \sqrt{2} comes from using the relative speed \bar{v}_{\text{rel}} = \sqrt{2}\,\bar{v} when both molecules are moving.

Equivalent forms for an ideal gas (P = n k_B T):

\lambda = \frac{k_B T}{\sqrt{2}\,\pi d^2 P}.

At room-temperature and atmospheric pressure, for nitrogen (d \approx 0.37 nm), \lambda \approx 68 nm — about 180 molecular diameters. The molecule flies freely for a distance of 180 of its own widths before colliding, a useful geometric scale. Combined with v_{\text{rms}} \approx 500 m/s, the collision frequency is \sim 7 \times 10^9 collisions per second — ten billion per molecule per second. Diffusion of smells, viscous drag, and thermal conductivity in a gas all depend quantitatively on \lambda.

Light an agarbatti in the puja room and, thirty seconds later, anyone in the next room notices the scent. Walk into a Delhi kitchen in the morning and the turmeric-and-mustard-seed smell of a fresh tempering fills the whole house in under a minute. Meanwhile — although you never see it — the molecules of sandalwood vapour are moving at about 300 metres per second. If they were travelling in straight lines, they could traverse a ten-metre apartment in about 30 milliseconds. Yet the scent takes tens of seconds to arrive. Something is slowing it down by a factor of a thousand.

That something is collisions. The air through which the sandalwood molecules must travel is a remarkably dense crowd — some 2.5 \times 10^{19} nitrogen and oxygen molecules per cubic centimetre — and each sandalwood molecule banging through it cannot go more than a tiny distance before crashing into a bystander, bouncing off in a new direction. What looks like a straightforward journey "across the room" is actually a microscopic random walk of astonishing density: the molecule takes a step, turns at a random angle, takes another step, turns, steps, turns, and so on for billions of iterations. The straight-line distance gained after all this zig-zagging is much, much less than the total path traversed.

The single geometric quantity that controls all of this is the mean free path \lambda — the average length of one of those tiny straight-line steps between collisions. It is one of the handful of numbers whose value you should carry around in your head for order-of-magnitude thinking: at standard temperature and pressure, the mean free path in air is about 70 nanometres — roughly 200 molecular diameters, or about a fifth of the wavelength of green light. That scale governs the speed of diffusion (smells crossing a room), the viscosity of gases (how a gas flows through a pipe), the thermal conductivity of air (how quickly a house cools), the minimum size at which a vacuum pump needs molecular-flow design instead of ordinary fluid-dynamics, and the altitude in the atmosphere where the "continuous fluid" picture of air gives way to "free-molecular" flight.

This article does four things. First, it derives the mean free path from a geometric collision-cross-section argument — without any fancy statistics, just a counting argument about molecules hitting cylinders. Second, it supplies the \sqrt{2} correction factor from the relative velocity of two moving targets. Third, it works out the practical dependence of \lambda on pressure and temperature. Fourth, it explores why \lambda is the master parameter of every gas-transport phenomenon.

The core idea: a molecule sweeps out a cylinder

Here is the geometric picture that everything in this chapter rests on.

Pick one molecule — call it the tagged molecule — and imagine it moving in a straight line through the gas at speed v. Treat the molecule as a hard sphere of diameter d. Treat every other molecule in the gas as a hard sphere of the same diameter, initially at rest (this is a temporary simplification; the \sqrt{2} correction fixes it later).

When does the tagged molecule collide with another? Exactly when the centre of another molecule comes within a distance d of the tagged molecule's centre. (Because both have radius d/2, the moment their surfaces touch, their centres are separated by d/2 + d/2 = d.)

So the tagged molecule is effectively a point that, as it moves, sweeps out a cylindrical shell of radius d. Any other molecule whose centre lies inside that cylinder — at any moment during the journey — is a molecule it has collided with.

Mean-free-path geometric constructionA horizontal cylinder of radius d swept out by a tagged molecule moving to the right at speed v. The tagged molecule sits at the left end as a filled red circle. Inside the cylinder, three target molecules (grey circles) are shown at their centres, all within distance d of the axis, each marked as a collision. Outside the cylinder, one molecule's centre lies just beyond d — it does not collide. Labels mark the cylinder length vt, area pi d squared, and volume pi d squared vt.swept cylinder: radius d, length vtcross-section area π d²; volume π d² v ttagged (speed v)hithithitmiss (outside)2r = d
The mean-free-path derivation. A tagged molecule moving rightward sweeps out a cylinder of radius $d$ (the effective collision-cross-section radius, since two hard spheres of diameter $d$ each collide when their centres get within $d$). Any other molecule whose centre lies inside the cylinder is a collision; those outside are misses. Counting the molecules inside the cylinder gives the collision rate.

Counting the molecules inside

Let n be the number density of molecules (molecules per unit volume) — a uniform constant for a gas in equilibrium. In time t, the tagged molecule sweeps out a cylinder of length v t and cross-sectional area \pi d^2:

\text{volume swept} = \pi d^2 \cdot v t.

Why: a cylinder's volume is base area times length. The base is a circle of radius d, area \pi d^2; the length is v t, the distance the tagged molecule travels in time t.

The number of molecular centres inside this volume is

N_{\text{inside}} = n \cdot \pi d^2 v t.

Each of these is a collision (in this stationary-target model). So the collision rate is

\frac{N_{\text{coll}}}{t} = n \pi d^2 v.

And the mean free path — the average distance between successive collisions — is

\lambda = \frac{\text{distance}}{\text{number of collisions}} = \frac{v t}{n \pi d^2 v t} = \frac{1}{n \pi d^2}.

This is the answer you would get if every molecule except the tagged one were standing perfectly still. But they are not standing still — they are moving at comparable speeds to the tagged molecule. The correct formula needs one more ingredient.

The √2 correction: targets move too

Think carefully about what speed to use when counting collisions. When two cars are approaching each other on a highway, the rate at which they close the gap is the sum of their speeds, not just one of them. The relative speed of approach is what matters for collision counting.

For two molecules with velocity vectors \vec{v}_1 and \vec{v}_2, the relative velocity is \vec{v}_{\text{rel}} = \vec{v}_1 - \vec{v}_2, and its magnitude depends on the directions. For two molecules each moving with speed v at a random relative angle \theta,

v_{\text{rel}}^2 = v_1^2 + v_2^2 - 2 v_1 v_2 \cos\theta = 2 v^2 (1 - \cos\theta).

Averaging over all directions — the angles \theta are uniformly distributed on a sphere — the average of \cos\theta is zero, so

\langle v_{\text{rel}}^2 \rangle = 2 v^2,

and taking square roots (up to a subtlety about averaging rms versus mean, which is handled carefully in the full derivation):

\langle v_{\text{rel}} \rangle = \sqrt{2} \cdot \langle v \rangle.

Why: when you average over all relative orientations of the two velocity vectors, the cross-term 2 v_1 v_2 \cos\theta averages to zero (half the time \cos\theta is positive, half the time negative), leaving just v_1^2 + v_2^2 = 2v^2. The collision rate uses the relative speed, so the correct collision rate is \sqrt{2} times larger than the stationary-target estimate.

Replacing v in the collision-rate formula by \sqrt{2}\,v gives the correct mean free path:

\boxed{\;\lambda = \frac{1}{\sqrt{2}\,\pi d^2 n}.\;}

This \sqrt{2} is a small correction — only about 41% — but it is the difference between a rough order-of-magnitude estimate and the correct expression. It was first given by James Clerk Maxwell in 1860.

Practical forms

For an ideal gas at temperature T and pressure P, the number density is n = P/(k_B T), and so

\lambda = \frac{k_B T}{\sqrt{2}\,\pi d^2 P}.

This is the form you most often use: \lambda is proportional to T and inversely proportional to P. Doubling the pressure halves the mean free path; doubling the temperature (at constant pressure) doubles it. At constant density (not pressure), \lambda is completely independent of temperature — it depends only on the crowd-density of targets, not on how fast anyone is moving.

The number for nitrogen at STP

Put in the actual numbers for air at standard temperature and pressure. Nitrogen molecules have a collision diameter d \approx 0.37 nm (this is not the molecule's "size" in any rigorous sense — it is the distance between molecular centres at which the molecules interact strongly enough to count as a collision). Pressure P = 1.013 \times 10^5 Pa, temperature T = 273.15 K, and k_B = 1.381 \times 10^{-23} J/K. Then

\lambda = \frac{k_B T}{\sqrt{2}\,\pi d^2 P} = \frac{(1.381 \times 10^{-23})(273.15)}{\sqrt{2}\,\pi\,(0.37 \times 10^{-9})^2\,(1.013 \times 10^5)}.

Compute piece by piece.

Numerator: (1.381 \times 10^{-23})(273.15) \approx 3.77 \times 10^{-21}.

Denominator: \sqrt{2} \approx 1.414; \pi \approx 3.1416; d^2 = (0.37 \times 10^{-9})^2 = 1.369 \times 10^{-19}. Product: 1.414 \times 3.1416 \times 1.369 \times 10^{-19} \times 1.013 \times 10^{5} \approx 6.16 \times 10^{-14}.

\lambda \approx \frac{3.77 \times 10^{-21}}{6.16 \times 10^{-14}} \approx 6.1 \times 10^{-8} \text{ m} \approx 61 \text{ nm}.

Textbook values usually quote \approx 68 nm (the small discrepancy depends on the exact value of d you use). Either way, the number you should carry around is: mean free path of air at STP is about 70 nanometres — about one-tenth of the wavelength of visible light.

Comparing this to the molecular diameter itself: \lambda / d \approx 70/0.37 \approx 190. A nitrogen molecule flies, on average, about 190 of its own diameters between collisions. If a molecule were the size of a marble (about 1 cm), its mean free path would be 1.9 metres — so it would fly across a bedroom, collide, fly across again, collide, and so on ten billion times a second.

Collision frequency

Given \lambda = 68 nm and v_{\text{rms}} \approx 500 m/s for nitrogen at room temperature, the time between collisions is

\tau_{\text{coll}} = \frac{\lambda}{v} \approx \frac{6.8 \times 10^{-8}}{500} \approx 1.4 \times 10^{-10} \text{ s},

so the collision frequency is

f_{\text{coll}} = \frac{1}{\tau_{\text{coll}}} \approx 7 \times 10^9 \text{ per second}.

Ten billion collisions per molecule per second. Per molecule. Multiply by Avogadro's number of molecules per mole, and you get \sim 10^{34} collisions per mole per second — a number so large that the very concept of an individual collision becomes meaningless; what you observe macroscopically is the aggregate statistical behaviour.

Exploring the parameters

Interactive: mean free path versus pressure A curve of mean free path in nanometres versus pressure in atmospheres, showing the inverse proportionality. A draggable slider on the x-axis changes the pressure; a readout displays the resulting mean free path and collision frequency. pressure P (atm) λ (nm) 300 200 100 0 1 5 10 drag the red point
Drag the red point along the horizontal axis to explore how the mean free path for nitrogen (at 273 K) depends on pressure. At 1 atm, $\lambda \approx 68$ nm and the collision frequency is about 7 GHz. Lower the pressure to 0.1 atm (a mountaintop or a poor vacuum) and $\lambda$ grows tenfold to 680 nm — still much smaller than any macroscopic length, but a tenth of the wavelength of red light. Raise to 10 atm (inside a tyre or compressed-gas cylinder) and $\lambda$ drops to 6.8 nm, only about 20 molecular diameters.

Why \lambda matters: three transport phenomena

The mean free path is not just a tidy geometric quantity — it is the master control for three macroscopic gas properties: diffusion, viscosity, and thermal conductivity. All three are called "transport phenomena" because they describe how some quantity (mass, momentum, or energy) is carried from one region of a gas to another. The kinetic-theory picture of each goes the same way: molecules moving between regions of different concentration/velocity/temperature carry their own load, and the mean free path tells you how far they can carry it before being randomized by a collision.

Diffusion — how smells cross a room

Release a small amount of scent gas in one corner of a room. Over time, the scent molecules wander outward, their concentration evening out. This is diffusion. The diffusion coefficient D — a measure of how fast this evening-out happens — comes out, from kinetic theory, as

D \approx \tfrac{1}{3} \bar{v} \lambda,

where \bar{v} is the mean molecular speed. For air at STP, \bar{v} \approx 450 m/s and \lambda \approx 70 nm, giving D \approx 10^{-5} m²/s. The typical distance a scent molecule wanders in time t is \sqrt{2 D t}; to cross ten metres, t \approx 100/(2 \cdot 10^{-5}) = 5 \times 10^6 s \approx 60 days!

That cannot be right — scents cross rooms in seconds, not months. The resolution: in a still room, air is never actually still. The slightest breath, draft, temperature difference, or convection current stirs the air at bulk speeds of centimetres per second, carrying scent parcels with it at vastly greater effective rates than pure molecular diffusion. Pure molecular diffusion is the lower bound on transport; actual transport is dominated by bulk fluid motion. The only place pure molecular diffusion wins is over very short distances (sub-millimetre) or in a truly still, isothermal gas — neither of which happens in an ordinary room.

Viscosity — why a gas resists shear

Pour honey from a spoon: it flows slowly because its molecules resist being sheared past each other. Air doesn't pour, but it too has a (much smaller) viscosity \eta, which controls how fast a dropped raindrop falls and how much drag a cyclist feels. Kinetic theory gives

\eta \approx \tfrac{1}{3} n m \bar{v} \lambda = \tfrac{1}{3} \rho \bar{v} \lambda.

An extraordinary consequence: using n \lambda = 1/(\sqrt{2}\,\pi d^2), the factor n \lambda is independent of n, so

\eta \propto \bar{v} \propto \sqrt{T},

the viscosity of an ideal gas is independent of pressure and grows as the square root of temperature. This was Maxwell's prediction, and he was so sceptical of it that he performed the experiment himself — measuring the viscosity of air at different pressures and finding it really didn't depend on pressure. He regarded the observation as the most convincing experimental support of the kinetic theory in his day.

Thermal conductivity — why gases insulate

A gas is a poor thermal conductor because the energy carriers (moving molecules) can only travel about \lambda between scattering events. Kinetic theory gives

\kappa \approx \tfrac{1}{3} n c_v \bar{v} \lambda,

where c_v is the heat capacity per molecule at constant volume. Like viscosity, thermal conductivity is independent of pressure. For air at STP, \kappa \approx 0.025 W/(m·K) — which is why still air (in a thermos, between the double panes of a window, or trapped in a wool sweater) is such an effective insulator. Anything that disrupts the molecular scale — convection, bulk currents — bypasses the mean-free-path bottleneck and ships heat much faster.

Worked examples

Example 1: Mean free path of hydrogen at Mumbai vs Leh

A hydrogen balloon is filled in Mumbai (sea level, P = 1.0 atm, T = 300 K) and then carried to Leh (altitude about 3500 m, P = 0.66 atm, T = 280 K). The collision diameter of H_2 is d \approx 0.29 nm. Find the mean free path in each location, and the ratio.

Mean free path of hydrogen at two altitudesA balloon at sea-level Mumbai (dense crowd of molecules, short MFP) next to the same balloon in Leh at 3500 m altitude (sparser molecules, longer MFP). Arrows show mean free path labels.Mumbai1.0 atm, 300 Kdense — many collisionsλ ≈ 120 nmLeh0.66 atm, 280 Ksparser — longer free flightsλ ≈ 170 nm~3500 m up
The same hydrogen balloon at two altitudes. In Leh, the air is 34% less dense (lower pressure and slightly lower temperature), so the mean free path is longer — hydrogen molecules inside the balloon fly about 40% further between collisions.

Step 1. Use \lambda = k_B T / (\sqrt{2}\,\pi d^2 P).

Common factor: \sqrt{2}\,\pi d^2 = 1.414 \cdot 3.1416 \cdot (0.29 \times 10^{-9})^2 = 4.441 \cdot 8.41 \times 10^{-20} = 3.74 \times 10^{-19} m².

Why: evaluating the geometric constant once saves arithmetic when you are comparing two locations. The \sqrt{2}\,\pi d^2 combination is the effective "cross-section area seen by a moving molecule against a moving background" and depends only on the gas species.

Step 2. Mumbai.

\lambda_{\text{Mum}} = \frac{(1.381 \times 10^{-23})(300)}{(3.74 \times 10^{-19})(1.013 \times 10^5)} = \frac{4.143 \times 10^{-21}}{3.79 \times 10^{-14}} \approx 1.09 \times 10^{-7} \text{ m} \approx 109 \text{ nm}.

Why: straight plug into the formula for 1 atm = 101,300 Pa and 300 K. The mean free path of hydrogen is a bit larger than that of nitrogen (68 nm) at the same conditions because hydrogen molecules have a smaller collision diameter (d_{H_2} = 0.29 nm vs d_{N_2} = 0.37 nm), giving a smaller cross-section.

Step 3. Leh. Pressure P = 0.66 atm = 6.69 \times 10^4 Pa, T = 280 K.

\lambda_{\text{Leh}} = \frac{(1.381 \times 10^{-23})(280)}{(3.74 \times 10^{-19})(6.69 \times 10^4)} = \frac{3.867 \times 10^{-21}}{2.50 \times 10^{-14}} \approx 1.55 \times 10^{-7} \text{ m} \approx 155 \text{ nm}.

Why: at Leh, both the temperature drop (slightly reduces \lambda, T/T_0 = 0.93) and the pressure drop (strongly increases \lambda, P/P_0 = 0.66) combine. The net effect is governed by the ratio T/P.

Step 4. Ratio.

\frac{\lambda_{\text{Leh}}}{\lambda_{\text{Mum}}} = \frac{T_{\text{Leh}}/P_{\text{Leh}}}{T_{\text{Mum}}/P_{\text{Mum}}} = \frac{280/0.66}{300/1.0} = \frac{424}{300} \approx 1.42.

Why: all the common factors (\sqrt{2}\,\pi d^2, k_B) cancel in the ratio. The mean free path scales as T/P, so the ratio is just the ratio of T/P at the two locations.

Result: Mean free path \approx 109 nm in Mumbai and 155 nm in Leh — a factor of about 1.42 longer in Leh.

What this shows: Mean free path depends more sensitively on pressure than temperature (linear in each, but pressure varies by 34% while temperature varies by only 7% in this comparison). Taken further: in the upper atmosphere at 100 km altitude, the pressure is 10^{-7} atm, and the mean free path is of order metres — larger than most apparatus. Below 100 km the "continuous-fluid" picture of air works; above it you enter the free-molecular regime, where molecules travel ballistically between rare collisions. This transition (the Knudsen number \text{Kn} = \lambda/L crossing 1) is why re-entering spacecraft need different heat-shield analysis above and below 100 km.

Example 2: How good a vacuum is a vacuum flask?

A thermos (vacuum flask) is evacuated to a pressure of 10^{-3} atm at room temperature (300 K). Estimate the mean free path of the residual air between the walls (separation 2 mm). Is the gas in the molecular-flow regime (\lambda \gtrsim wall separation) or the continuum regime (\lambda \ll wall separation)?

Mean free path in a vacuum flask gapA cross-section diagram of a thermos's double wall, with a 2 mm gap between inner and outer walls. A few molecules are drawn zigzagging in the gap with mean free path labelled 70 micrometres, much smaller than the gap — so the gas is still in the continuum regime.inner wallouter wallgap = 2 mm~10² molecules/µm³λ ≈ 70 µm (vs 2 mm gap)Knudsen numberKn = λ/L ≈ 0.035< 1 ⇒ continuum regime,heat conduction not yetsuppressed — need better vacuum
Inside a thermos gap at $10^{-3}$ atm, the mean free path is about 70 µm. The gap is 2 mm. Molecules still collide hundreds of times before crossing from wall to wall — heat can still conduct efficiently through the (now thinner) gas. A better vacuum (below $10^{-4}$ atm) is needed to reach the molecular-flow regime where conduction is dramatically suppressed.

Step 1. Scale from known: at 1 atm, \lambda \approx 70 nm. At 10^{-3} atm,

\lambda = 70 \text{ nm} \times \frac{1}{10^{-3}} = 70 \text{ nm} \times 1000 = 7.0 \times 10^4 \text{ nm} = 70 \text{ µm}.

Why: \lambda \propto 1/P at fixed temperature. A factor of 1000 reduction in pressure gives a factor of 1000 increase in \lambda. This scaling-from-known-values is much faster than a fresh calculation.

Step 2. Compare to the wall separation L = 2 mm = 2000 µm.

\frac{\lambda}{L} = \frac{70}{2000} = 0.035.

Step 3. Interpretation. The Knudsen number \text{Kn} = \lambda / L distinguishes two regimes:

  • \text{Kn} \ll 1: continuum regime — molecules collide with each other many times while crossing the gap, and the gas behaves as a fluid. Ordinary thermal conductivity applies.
  • \text{Kn} \gtrsim 1: molecular-flow regime — molecules typically travel from wall to wall without colliding with each other. Heat transport is by ballistic free-flight, and thermal conductivity effectively vanishes.

Here \text{Kn} = 0.035: still deep in the continuum regime. Heat will continue to conduct through the residual gas at a rate only modestly reduced from atmospheric.

Step 4. What pressure is needed for a good thermos?

For molecular flow, demand \text{Kn} \gtrsim 1, i.e. \lambda \gtrsim 2 mm. At 300 K, scaling from the STP number:

\lambda = \frac{70 \text{ nm}}{P(\text{atm})} \gtrsim 2 \times 10^6 \text{ nm} \quad \Rightarrow \quad P \lesssim \frac{70}{2 \times 10^6} \approx 3.5 \times 10^{-5} \text{ atm}.

Why: solving \lambda \ge L for P. A thermos that really suppresses conduction needs a vacuum of \sim 10^{-5} atm — about 100 times better than the 10^{-3} atm in this example. Cheap thermos flasks achieve 10^{-3} to 10^{-4} atm; high-quality cryogenic dewars pump down to 10^{-6} atm.

Result: At 10^{-3} atm, the mean free path is about 70 µm, much less than the 2 mm gap, so the flask is still in the continuum regime with heat conducting efficiently through the residual air. To reach molecular flow (and really suppress conduction), the pressure needs to come down to about 3 \times 10^{-5} atm or lower.

What this shows: The design of a good vacuum flask is not just about "pump out the air" but specifically about reaching a pressure where the mean free path exceeds the wall spacing. Below that threshold, each additional factor of ten of pumping down dramatically reduces heat conduction; above it, pumping helps only incrementally. This is why industrial cryogenic dewars use multi-layer "super-insulation" with very small gaps (each \lambda of the residual gas already spans the gap at moderate vacuum), rather than one big evacuated chamber.

Common confusions

If you are comfortable with \lambda = 1/(\sqrt{2}\,\pi d^2 n) and can use it to solve problems, you have what kinetic theory needs at the introductory level. What follows is the more careful statistical treatment, the distribution of path lengths, the full derivation of transport coefficients, and the quantum correction at very low temperatures where the hard-sphere model fails.

The distribution of free-path lengths

A common simplification is to speak of "the mean free path" as if every flight between collisions had length \lambda. In fact the actual free-flight distances are distributed according to a specific law.

Let P(x) dx be the probability that a molecule travels a distance between x and x + dx without colliding. In the infinitesimal slice dx, the probability of colliding is dx/\lambda (because the rate of collisions per unit distance is 1/\lambda). So the probability of surviving through dx without colliding is 1 - dx/\lambda, and the probability of surviving from the start to distance x is the product over all slices, which gives

P(\text{survive to } x) = e^{-x/\lambda}.

The probability density of colliding between x and x + dx is

P(x) \, dx = \frac{1}{\lambda} e^{-x/\lambda} \, dx.

This is an exponential distribution. Two things follow immediately.

  1. The mean of the distribution is indeed \lambda: \int_0^\infty x P(x) dx = \lambda. So \lambda is in fact the mean free path, not just a conventional name.

  2. The variance of the distribution is \lambda^2: the standard deviation equals the mean. Free flights vary wildly — some are much shorter than \lambda, a few are much longer. The probability of a free flight exceeding 3\lambda is e^{-3} \approx 0.05 — 5% of flights. The probability of exceeding 5\lambda is e^{-5} \approx 0.007. "Most" flights are within a few times \lambda, but there is a long tail.

Proper averaging and the \sqrt{2} factor redone

The quick argument for the \sqrt{2} factor used \langle v_{\text{rel}}^2 \rangle = 2 \bar{v}^2. This is approximate — it correctly captures the Maxwell–Boltzmann distribution's mean relative speed, but only because of a nice cancellation. Let us work it out properly.

The mean relative speed of two molecules of the same species at temperature T is

\langle v_{\text{rel}} \rangle = \int |\vec{v}_1 - \vec{v}_2| f(\vec{v}_1) f(\vec{v}_2) d^3 v_1 d^3 v_2,

where f is the Maxwell–Boltzmann distribution. By a change of variables to centre-of-mass velocity \vec{V} = (\vec{v}_1 + \vec{v}_2)/2 and relative velocity \vec{u} = \vec{v}_1 - \vec{v}_2, this decouples into two Gaussian integrals. The relative velocity distribution itself is Maxwell–Boltzmann with reduced mass \mu = m/2 (since both molecules have the same mass m), so

\langle v_{\text{rel}} \rangle = \sqrt{\frac{8 k_B T}{\pi \mu}} = \sqrt{\frac{8 k_B T}{\pi (m/2)}} = \sqrt{\frac{16 k_B T}{\pi m}} = \sqrt{2} \cdot \sqrt{\frac{8 k_B T}{\pi m}} = \sqrt{2} \cdot \langle v \rangle.

The \sqrt{2} factor is exact. This is the derivation that Maxwell gave in 1860.

Derivation of transport coefficients

Consider a gas with a small gradient in some quantity \phi — concentration, velocity component, or energy per molecule — along the x-direction. At the plane x = 0, molecules crossing from the +x side carry their value of \phi (typical of the point x = +\lambda where they last collided). Molecules crossing from the -x side carry \phi from x = -\lambda. The net flux of \phi across the plane is

J_\phi \approx -\tfrac{1}{3} n \bar{v} \lambda \frac{d\phi}{dx},

where the \frac{1}{3} arises from geometric averaging over molecular velocity directions (same factor as in the pressure derivation). The flux is proportional to the gradient — this is the generic form of Fick's law (for \phi = concentration, diffusion), Newton's law of viscosity (for \phi = momentum), and Fourier's law (for \phi = energy, thermal conduction). The transport coefficient is \frac{1}{3} n \bar{v} \lambda times the relevant per-molecule factor.

  • Diffusion: D = \frac{1}{3} \bar{v} \lambda (self-diffusion).
  • Viscosity: \eta = \frac{1}{3} n m \bar{v} \lambda = \frac{1}{3} \rho \bar{v} \lambda.
  • Thermal conductivity: \kappa = \frac{1}{3} n c_v \bar{v} \lambda, where c_v is the per-molecule heat capacity.

Using n \lambda = 1/(\sqrt{2} \pi d^2):

\eta = \frac{m \bar{v}}{3\sqrt{2} \pi d^2}, \qquad \kappa = \frac{c_v \bar{v}}{3\sqrt{2}\pi d^2}.

Both are independent of pressure and density — a striking and non-obvious prediction of kinetic theory, confirmed by Maxwell's own experiments.

Failure at high and low density

The derivation assumes that the mean free path is much larger than the molecular diameter (\lambda \gg d) — which is the same as saying n \pi d^2 \lambda \sim 1, or that molecules spend most of their time flying freely rather than interacting. At very high densities (liquid-like), this fails: molecules are always in contact with several neighbours, and the hard-sphere-free-flight picture breaks down. The correct treatment requires dense-gas theory or Enskog's equation.

At very low densities (spacecraft altitudes, laboratory ultra-high vacuum), the mean free path exceeds the size of the container, and molecules fly from wall to wall without intermolecular collisions. This is the free-molecular flow regime. The transport coefficients as derived above no longer apply; heat and momentum transport depend on accommodation at the walls, not on bulk gas properties. This is why satellite thermal management (at altitudes where \lambda \gg satellite size) uses very different rules than surface-atmosphere thermal management.

Relation to the Knudsen number

The dimensionless ratio

\text{Kn} = \frac{\lambda}{L},

where L is a characteristic macroscopic length (tube diameter, satellite size, aircraft chord), separates the flow regimes:

  • \text{Kn} < 0.01: continuum regime. Navier–Stokes equations apply. Air moving over a wing.
  • 0.01 < \text{Kn} < 0.1: slip regime. Navier–Stokes applies, but with boundary-layer modifications at walls.
  • 0.1 < \text{Kn} < 10: transition regime. Molecular and continuum effects both matter. This is where direct simulation Monte Carlo (DSMC) methods are used.
  • \text{Kn} > 10: free-molecular regime. Molecules are ballistic between collisions. Spacecraft re-entry above \sim 120 km altitude.

This is why MEMS devices (microfluidics, L \sim 1 µm) and satellites (altitude determines \lambda) both need non-continuum treatments at specific scales.

Quantum corrections

At very low temperatures, the thermal de Broglie wavelength \lambda_{\text{dB}} = h/\sqrt{2\pi m k_B T} becomes comparable to or larger than the classical mean free path. When this happens, molecules can no longer be treated as hard spheres — they interfere quantum-mechanically, and the collision cross-section becomes energy-dependent in ways that have no classical analogue. This is the onset of the Bose–Einstein or Fermi-degenerate regime, and the classical formula for \lambda breaks down. For helium-4 below 2.2 K, this is not a minor correction — it's the entire story; the liquid becomes superfluid.

Why the zigzag matters for the atmosphere

A nitrogen molecule at the surface moves at hundreds of metres per second but cannot escape into space — even though its kinetic energy is comparable to the gravitational binding energy required. It can't because it is in a dense crowd and its path is randomised every 70 nm. Integrating: to rise a few tens of kilometres from the random walk would take a time (L/\lambda)^2 / (\text{collision rate}) \sim (10^5/70 \times 10^{-9})^2 / 10^{10} \approx 10^{16} seconds — longer than the age of the universe. So the atmosphere stays put, held together by collisional diffusion. Only at high altitudes, where \lambda is comparable to scale height (\sim 8 km), do molecules have any real chance of escaping; this is why the boundary between "atmosphere" and "space" is conventionally placed at about 100 km. Above that, a hydrogen molecule (light, fast) actually does escape — Earth has been steadily losing hydrogen for billions of years.

Why sound travels differently in cold air

The speed of sound in an ideal gas is v_s = \sqrt{\gamma k_B T / m}. In a cold winter morning in Delhi (273 K), the speed is about 331 m/s; in a Chennai summer afternoon (303 K), it's about 348 m/s — about 5% faster. This small effect is real: voices carry a hair farther in cold air, and birds singing at dawn in winter can be heard over slightly longer distances.

But there is also a mean-free-path effect that emerges in quieter contexts. Sound attenuation (how fast sound energy is absorbed as it propagates) scales with mean free path times frequency. High-frequency sounds are attenuated more strongly in low-density air (longer \lambda) because the wavelength becomes comparable to \lambda and sound stops behaving as a continuum wave. This is one reason why ultrasonic sensors on drones have restricted range at high altitude.

Where this leads next