Maxwell-Boltzmann Velocity Distribution

Pick a single molecule out of a gas at temperature $T$ and ask: what is the probability that its velocity component along the $x$ -axis lies between $v_x$ and $v_x+dv_x$ ? The answer is one of the most consequential results in physics — a simple bell curve, $f(v_x)\propto e^{-mv_x^2/2k_BT}$ . Each Cartesian component of velocity is an independent Gaussian, centred on zero, with a width set entirely by the ratio of thermal energy to particle mass. This is the Maxwell-Boltzmann velocity distribution, and it is where thermal motion first becomes quantitative. The distribution follows directly from the Boltzmann factor $e^{-\beta E}$ applied to the kinetic energy $E=\tfrac{1}{2}m(v_x^2+v_y^2+v_z^2)$ . Because the energy splits into a sum over the three components, the exponential factorizes into a product of three independent one-dimensional Gaussians. Nothing about the potential energy, the interactions, or the container shape enters — only the kinetic term matters, so the velocity distribution of any classical gas in equilibrium, dilute or dense, is the same universal Gaussian. Two features carry enormous physical weight. First, $\langle v_x\rangle=0$ : there is no net drift, motion is isotropic, and the gas as a whole is at rest even though every molecule is flying. Second, $\langle v_x^2\rangle=k_BT/m$ , which means the average kinetic energy in each direction is exactly $\tfrac{1}{2}k_BT$ — the equipartition theorem, seen here in its cleanest form. The velocity distribution is the microscopic origin of temperature: $T$ is literally the width of this bell curve. From it flow the speed distribution, the characteristic speeds, effusion rates, reaction kinetics, and the entire kinetic theory of gases.

Key takeaways

Each velocity component of a molecule in a classical gas at temperature $T$ is an independent Gaussian: $f(v_x)=\sqrt{m/2\pi k_BT}\,e^{-mv_x^2/2k_BT}$ .
The distribution is centred on zero ( $\langle v_x\rangle=0$ , no drift) with variance $\langle v_x^2\rangle=k_BT/m$ , so $\tfrac{1}{2}m\langle v_x^2\rangle=\tfrac{1}{2}k_BT$ — equipartition, one quadratic degree of freedom.
The three components are statistically independent, so the full velocity distribution factorizes: $f(\mathbf v)=(m/2\pi k_BT)^{3/2}\,e^{-mv^2/2k_BT}$ .
Isotropy means $\langle v_x^2\rangle=\langle v_y^2\rangle=\langle v_z^2\rangle=k_BT/m$ , hence $\langle v^2\rangle=3k_BT/m$ (the root-mean-square speed).
Do not confuse the velocity distribution (a signed Gaussian peaked at $v_x=0$ ) with the speed distribution, which multiplies by the shell factor $4\pi v^2$ and peaks away from zero.
Key idea: the Boltzmann factor on kinetic energy makes each Cartesian velocity an independent, zero-mean Gaussian whose width $\sqrt{k_BT/m}$ is temperature made visible.

The core insight

Applying the Boltzmann weight $e^{-\beta E}$ to the kinetic energy $E=\tfrac12 m(v_x^2+v_y^2+v_z^2)$ makes each velocity component an independent Gaussian, $f(v_x)=\sqrt{m/2\pi k_BT}\,e^{-mv_x^2/2k_BT}$ , with $\langle v_x\rangle=0$ and $\langle v_x^2\rangle=k_BT/m$ . Temperature is the width of this bell curve.

Where this lives today

The Maxwell--Boltzmann velocity distribution underlies nearly every quantitative description of thermal motion in classical gases. Doppler broadening of spectral lines is a direct fingerprint of $f(v_x)$ : the line width along the line of sight measures the Gaussian width $\sqrt{k_BT/m}$ , and astronomers routinely read gas temperatures off stellar and nebular spectra this way. Laser cooling and atom trapping are engineered by reshaping this distribution — Doppler cooling narrows it, evaporative cooling clips its high-velocity tail. In plasma physics and fusion, the velocity distribution controls reaction rates and confinement. In molecular-dynamics simulations, thermostats work precisely by forcing the velocity components back onto the Maxwell-Boltzmann Gaussian at the target temperature. Whenever the microscopic motion of classical particles is in thermal equilibrium, their velocity components follow this Gaussian distribution. It provides the microscopic statistical meaning of temperature.

The physical picture

Imagine painting an arrow on each molecule showing its velocity, then collecting all the arrows and dropping them at a common origin. The tips form a fuzzy cloud. Because motion is isotropic, the cloud is spherically symmetric — no direction is special. Slice the cloud along the $x$ -axis and the density of tips traces a bell curve: most molecules have small $v_x$ , a few have large $|v_x|$ , and positive and negative are equally likely. That bell curve is $f(v_x)$ , and its width grows as the gas is heated (more spread) and shrinks with heavier molecules (more inertia, less spread). The payoff is that this single Gaussian encodes temperature. Increasing the temperature broadens the Gaussian and swells the cloud, since the variance satisfies $\langle v_x^2\rangle=k_BT/m\propto T$ . Cool it and the cloud contracts toward the origin. There is no other parameter: fix $T$ and $m$ and the entire velocity statistics of the gas is determined. This is why we can define temperature through the mean kinetic energy — the width of the velocity bell curve is the thermometer.

Figure

The velocity component

f(v_x)

is a Gaussian centred on zero: symmetric, peaked at v_x=0, with width √(kT/m). Heating widens it; a heavier mass narrows it. The mean is zero — no net drift.

The second figure is the contrast that trips people up: the velocity distribution and the speed distribution are not the same curve. A single component peaks at zero, but the speed $|\mathbf v|$ cannot be negative, and there are far more ways to have a moderate speed (a whole spherical shell of directions) than a near-zero one. Multiplying the Gaussian by the shell area $4\pi v^2$ pushes the peak away from zero. Both come from the same exponential, but the Jacobian makes them look different.

Figure

Contrast: the signed velocity component f(v_x) peaks at zero, but the speed distribution

f(v) =

(that Gaussian)

\times 4\pi v^2

peaks at

v_p > 0

because the shell factor

4\pi v^2

suppresses small speeds. Same exponential, different geometry.

The core equations

The starting point is the canonical probability that a molecule has a given velocity, obtained by weighting the kinetic energy with the Boltzmann factor and normalizing. Because only the kinetic energy depends on velocity, the spatial and interaction parts of the partition function cancel, leaving a universal result.

f(v_x)\,dv_x = \sqrt{\frac{m}{2\pi k_BT}}\;e^{-mv_x^2/2k_BT}\,dv_x \qquad (1)

where: $m$ — molecular mass; $k_B$ — Boltzmann constant; $T$ — absolute temperature; the prefactor $\sqrt{m/2\pi k_BT}$ is the normalization that makes $\int_{-\infty}^{\infty}f(v_x)\,dv_x=1$ . Physically, $f(v_x)$ is a Gaussian of zero mean and variance $k_BT/m$ . The first two moments carry the physics. The mean vanishes by symmetry; the variance sets the width and, through it, the temperature.

\langle v_x\rangle = 0,\qquad \langle v_x^2\rangle = \frac{k_BT}{m} \qquad (2)

where: $\langle v_x\rangle=0$ says the gas has no net momentum in any direction (isotropy); $\langle v_x^2\rangle=k_BT/m$ is the mean-square speed along one axis. Multiply by $\tfrac12 m$ and it becomes $\tfrac12 k_BT$ — equipartition for one translational degree of freedom. Because the kinetic energy is a sum over the three orthogonal directions, the Boltzmann factor factorizes and the three components are statistically independent. The full three-dimensional velocity distribution is the product.

f(\mathbf v) = f(v_x)\,f(v_y)\,f(v_z) = \left(\frac{m}{2\pi k_BT}\right)^{3/2} e^{-m(v_x^2+v_y^2+v_z^2)/2k_BT} \qquad (3)

where: the factorized form is the mathematical expression of statistical independence: knowing $v_x$ provides no information about $v_y$ or $v_z$ . Although the distribution factorizes into three one-dimensional Gaussians, the exponent depends only on $v^2=v_x^2+v_y^2+v_z^2$ , so the full velocity distribution is rotationally invariant (isotropic). Rotational symmetry therefore requires $\langle v_x^2\rangle=\langle v_y^2\rangle=\langle v_z^2\rangle$ , and summing the three variances gives $\langle v^2\rangle=3k_BT/m$ .

\langle v_x^2\rangle = \langle v_y^2\rangle = \langle v_z^2\rangle = \frac{k_BT}{m}\;\Rightarrow\; \langle v^2\rangle = \frac{3k_BT}{m} \qquad (4)

where: $\langle v^2\rangle=3k_BT/m$ is the mean-square speed, and $v_{\rm rms}\equiv\sqrt{\langle v^2\rangle} =\sqrt{\frac{3k_BT}{m}}$ is the root-mean-square speed. The factor of 3 is the number of translational degrees of freedom, each contributing $k_BT/m$ .

Derivation AThe velocity component distribution from the Boltzmann factor

The goal is to show that a single Cartesian velocity component of a molecule in a classical gas at temperature $T$ is distributed as a normalized Gaussian, $f(v_x)=\sqrt{m/2\pi k_BT}\,e^{-mv_x^2/2k_BT}$ , with no assumptions beyond the canonical Boltzmann weight on kinetic energy. The strategy is to apply $e^{-\beta E}$ to the kinetic energy, exploit the fact that the energy separates into independent quadratic terms, and fix the prefactor by normalization using the standard Gaussian integral.

Step 1. In the canonical ensemble the probability of a microstate is proportional to $e^{-\beta E}$ with $\beta=1/k_BT$ . For a free molecule the velocity-dependent part of the energy is purely kinetic, $E=\tfrac12 m(v_x^2+v_y^2+v_z^2)$ , so the velocity probability density is $P(\mathbf v)\propto e^{-\beta m(v_x^2+v_y^2+v_z^2)/2}$ .

Step 2. The exponent is a sum, so the exponential is a product: $P(\mathbf v)\propto e^{-\beta m v_x^2/2}\,e^{-\beta m v_y^2/2}\,e^{-\beta m v_z^2/2}$ . Each factor involves only one component, which is exactly what "statistically independent" means. Focus on one factor, $f(v_x)\propto e^{-\beta m v_x^2/2}$ .

Why the factorization is exact. The kinetic energy is diagonal in the Cartesian velocity components, $E=\frac{m}{2}(v_x^2+v_y^2+v_z^2)$ , so the Boltzmann factor factorizes into independent functions of $v_x$ , $v_y$ , and $v_z$ . Any interparticle potential energy depends on positions, not velocities, so it cancels between numerator and normalization and cannot spoil the velocity statistics.

Step 3. Normalize. Require $\int_{-\infty}^{\infty}f(v_x)\,dv_x=1$ . The Gaussian integral $\int_{-\infty}^{\infty}e^{-a v_x^2}\,dv_x=\sqrt{\pi/a}$ with $a=\beta m/2=m/2k_BT$ gives $\int e^{-m v_x^2/2k_BT}\,dv_x=\sqrt{2\pi k_BT/m}$ .

Step 4. Therefore the normalized density is $f(v_x)=\sqrt{m/2\pi k_BT}\,e^{-mv_x^2/2k_BT}$ , equation (1). It is a Gaussian of zero mean (the integrand $v_x f(v_x)$ is odd) and variance $\langle v_x^2\rangle=1/(2a)=k_BT/m$ , the standard Gaussian variance.

Go deeper · Derivation B · 1 minIndependence, factorization, and isotropy of the three components

Derivation BIndependence, factorization, and isotropy of the three components

The goal is to show that the Maxwell-Boltzmann distribution factorizes into three independent Cartesian components and that rotational symmetry immediately implies $\langle v_x^2\rangle = \langle v_y^2\rangle = \langle v_z^2\rangle$ , so that $\langle v^2\rangle = 3k_BT/m$ . This is the structural backbone that connects the one-component Gaussian to the speed distribution and to equipartition, and it rests entirely on the additive structure of kinetic energy.

Step 1. From the previous card, $f(\mathbf v)=f(v_x)f(v_y)f(v_z)$ because the Boltzmann factor of a sum is a product. Substituting the normalized one-dimensional forms gives $f(\mathbf v)=(m/2\pi k_BT)^{3/2}\,e^{-m(v_x^2+v_y^2+v_z^2)/2k_BT}$ , equation (3).

Step 2. Since $f(\mathbf v)=f(v_x)f(v_y)f(v_z)$ , the three Cartesian components are statistically independent. Equivalently, $P(v_x,v_y,v_z) = P(v_x)P(v_y)P(v_z)$ . The covariance $\langle v_x v_y\rangle-\langle v_x\rangle\langle v_y\rangle=0$ because the integral factorizes and each first moment vanishes.

Step 3. Isotropy: the exponent depends on the components only through $v^2=v_x^2+v_y^2+v_z^2$ , so $f(\mathbf v)$ is invariant under rotations. No direction is preferred; the velocity cloud is a spherically symmetric Gaussian ball. Why isotropy forces equal variances. Rotational invariance means the labelling of axes is arbitrary, so any statistical quantity computed for $v_x$ must equal the same quantity for $v_y$ or $v_z$ . In particular $\langle v_x^2\rangle=\langle v_y^2\rangle=\langle v_z^2\rangle$ .

Step 4. Sum the equal variances: $\langle v^2\rangle=\langle v_x^2\rangle+\langle v_y^2\rangle+\langle v_z^2\rangle=3\times(k_BT/m)=3k_BT/m$ , equation (4). The root-mean-square speed is $v_{\text{rms}}=\sqrt{3k_BT/m}$ , and the total mean kinetic energy is $\tfrac12 m\langle v^2\rangle=\tfrac32 k_BT$ — three quadratic degrees of freedom, each $\tfrac12 k_BT$ .

Limiting cases

The Gaussian velocity distribution is a purely classical, high-temperature result; its edges reveal where classical kinetic theory holds and where it must be replaced.

Case	Result	Meaning
Any classical gas, equilibrium	$f(v_x)=\sqrt{m/2\pi k_BT}\,e^{-mv_x^2/2k_BT}$	Universal — independent of interactions, density, or container.
High temperature $T\to\infty$	width $\sqrt{k_BT/m}\to\infty$	Broad, flat distribution; molecules explore a huge range of velocities.
Low temperature $T\to0$	width $\to0$ , $f\to\delta(v_x)$	Classical gas freezes toward zero velocity; but quantum statistics take over first.
Heavy vs light mass	width $\propto1/\sqrt m$	Heavy molecules move slowly; H $_2$ is far broader than SF $_6$ at the same $T$ .
Quantum-degenerate regime	Gaussian fails	When $n\lambda^3\gtrsim1$ , Fermi-Dirac or Bose-Einstein statistics replace the Maxwell-Boltzmann form.

Common trapCommon pitfall

The velocity distribution $f(v_x)$ and the speed distribution $f(v)$ are different objects and peak at different places. The single component $f(v_x)$ is a signed Gaussian centred on $v_x=0$ — the most probable value of any one component is zero. The speed $v=|\mathbf v|\ge0$ is obtained by converting to spherical shells, which multiplies by the Jacobian $4\pi v^2$ ; this factor vanishes at $v=0$ and pushes the peak of $f(v)$ out to the most probable speed $v_p=\sqrt{2k_BT/m}>0$ . Saying "molecules most commonly have zero speed" confuses the two: zero is the most common component, not the most common speed. The exponential is shared; the geometry ( $4\pi v^2$ ) is what differs.

Worked examples

Example 1 — The width of the velocity distribution for oxygen (enumerate, apply, boundary number). Take O $_2$ ( $M=31.998\,\mathrm{g/mol}$ , so $m=31.998\times1.66054\times10^{-27}\,\mathrm{kg}=5.314\times10^{-26}\,\mathrm{kg}$ ) at $T=300\,\mathrm{K}$ .

Step 1. The variance of one component is $\langle v_x^2\rangle=k_BT/m=(1.380649\times10^{-23})(300)/(5.314\times10^{-26})=7.79\times10^{4}\,\mathrm{m^2/s^2}$ .

Step 2. The standard deviation of one velocity component is $\sigma_{v_x}=\sqrt{\langle v_x^2\rangle}=\sqrt{7.79\times10^4} =279~{\rm m\,s^{-1}}$ . This is the characteristic spread of the $x$ -component of the molecular velocity.

Step 3. The root-mean-square speed uses all three components: $v_{\text{rms}}=\sqrt{\langle v^2\rangle}=\sqrt{3}\times279=\mathbf{483\,m/s}$ . As expected, $v_{\rm rms}=\sqrt3\,\sigma_{v_x}$ , because the three Cartesian components contribute equally to the total mean-square speed.

Step 4. Boundary check via equipartition: $\tfrac12 m\langle v_x^2\rangle=\tfrac12(5.314\times10^{-26})(7.79\times10^{4})=2.07\times10^{-21}\,\mathrm{J}$ , and $\tfrac12 k_BT=\tfrac12(1.380649\times10^{-23})(300)=2.07\times10^{-21}\,\mathrm{J}$ . They match — each translational degree of freedom carries $\tfrac12 k_BT$ , confirming the physics.

Example 2 — Probability that one velocity component exceeds $2\sigma$ . For nitrogen ( $m=4.65\times10^{-26}\,\mathrm{kg}$ ) at $T=300\,\mathrm{K}$ , what fraction have $v_x>2\sigma$ , where $\sigma=\sqrt{k_BT/m}$ ?

Step 1. Compute $\sigma=\sqrt{(1.380649\times10^{-23})(300)/(4.65\times10^{-26})}=\sqrt{8.91\times10^{4}}=298\,\mathrm{m/s}$ .

Step 2. Because $f(v_x)$ is a standard Gaussian, the fraction with $v_x>2\sigma$ is the upper tail of a normal distribution beyond two standard deviations: $\tfrac12\,\text{erfc}(2/\sqrt2)=\tfrac12\,\text{erfc}(1.414)=\mathbf{0.0228}$ , i.e. about $2.3\%$ .

Step 3. Interpret: roughly one molecule in $44$ has an $x$ -velocity exceeding $2\sigma\approx596\,\mathrm{m/s}$ in the $+x$ direction. The Gaussian tails are thin — extreme single-component velocities are rare, which is why the mean kinetic energy is a sharply defined quantity for a macroscopic gas.

Origins · 1 minmaxwell-boltzmann distribution and how this was discovered

maxwell-boltzmann distribution

statistical-mechanics concept · Maxwell 1860, Boltzmann 1868–77

The Maxwell-Boltzmann distribution is the equilibrium probability law for the velocities (and speeds) of particles in a classical gas. Maxwell first wrote down the velocity distribution in 1860 using a symmetry argument — that the components must be independent and isotropic, which forces the Gaussian form — and Boltzmann later derived it from his transport equation and the more general principle that equilibrium maximizes entropy. The distribution is the microscopic content of temperature: its width is $\sqrt{k_BT/m}$ , so measuring the spread of molecular velocities measures $T$ directly. It underlies kinetic theory, effusion, reaction rates, Doppler line broadening, and the very definition of the mean kinetic energy $\tfrac32 k_BT$ .

History

1859

— Maxwell presents the velocity distribution to the British Association, using an independence-plus-isotropy argument.

1860

— Maxwell publishes "Illustrations of the Dynamical Theory of Gases" with the Gaussian velocity law.

1868

— Boltzmann generalizes the distribution to include external potentials and arbitrary energies.

1872

— Boltzmann's transport equation and H-theorem show the Maxwell distribution is the unique equilibrium.

1920

— Otto Stern's molecular-beam experiments give the first direct measurement of the speed distribution.

Going further

References. Pathria & Beale, Statistical Mechanics, 3rd ed., §6.4 and Ch. 1 (Maxwell distribution) · Reif, Fundamentals of Statistical and Thermal Physics, §7.9–7.11 · Kardar, Statistical Physics of Particles, §3–4 · Blundell & Blundell, Concepts in Thermal Physics, Ch. 5–6 · Landau & Lifshitz, Statistical Physics Part 1, §29. The key results are: $f(v_x)=\sqrt{m/2\pi k_BT}\,e^{-mv_x^2/2k_BT}$ with $\langle v_x\rangle=0$ and $\langle v_x^2\rangle=k_BT/m$ ; the three-dimensional factorization $f(\mathbf v)=(m/2\pi k_BT)^{3/2}e^{-mv^2/2k_BT}$ ; and $\langle v^2\rangle=3k_BT/m$ . These satisfy: (i) $\int_{-\infty}^{\infty}f(v_x)\,dv_x=1$ ; (ii) $\langle v_x\rangle=0$ (odd integrand) and $\langle v_x^2\rangle=k_BT/m$ ; (iii) $\tfrac12 m\langle v_x^2\rangle=\tfrac12 k_BT$ ; (iv) the product factorizes to $(m/2\pi k_BT)^{3/2}e^{-mv^2/2k_BT}$ ; (v) numerically for O $_2$ at 300 K, $\sqrt{\langle v_x^2\rangle}=279\,\mathrm{m/s}$ and $v_{\text{rms}}=483\,\mathrm{m/s}$ . Constants used: $k_B=1.380649\times10^{-23}\,\mathrm{J/K}$ , atomic mass unit $1.66054\times10^{-27}\,\mathrm{kg}$ .

Builds on: Classical Ideal Gas · Equipartition Theorem · Thermal de Broglie Wavelength Leads to: Maxwell-Boltzmann Speed Distribution · Characteristic Speeds: Mean, RMS, and Most Probable · Generalized Equipartition Theorem

Why is the mean velocity zero if molecules are moving so fast?

Because velocity is a vector with direction, and the distribution is symmetric: for every molecule moving in the $+x$ direction there is, on average, one moving in $-x$ at the same speed. The individual speeds are large (hundreds of m/s), but the vector average cancels to zero, which is exactly why the gas as a whole is at rest. The non-zero quantity is $\langle v_x^2\rangle=k_BT/m$ , the mean of the square, which cannot cancel because squares are always positive.

Does this distribution depend on the interactions between molecules?

No. The velocity distribution of a classical gas in equilibrium is always the same Gaussian, whether the gas is ideal or a dense interacting fluid. The reason is that interaction potentials depend only on positions, not velocities, so they factor out of the velocity part of the Boltzmann weight. Interactions change the spatial distribution and the equation of state, but the velocity statistics are universal — a striking and useful fact.

How is the velocity distribution related to temperature?

The variance $\langle v_x^2\rangle=k_BT/m$ is directly proportional to $T$ , so temperature is literally the width of the velocity bell curve (scaled by mass). This makes the velocity distribution the operational definition of temperature for a gas: measure the spread of molecular velocities — for instance through Doppler line broadening — and you have measured $T$ . Equivalently, $T=m\langle v_x^2\rangle/k_B$ .

Is this the same as the speed distribution?

No, and confusing the two is the classic error. The velocity distribution $f(v_x)$ is a signed Gaussian peaked at zero for each component. The speed distribution $f(v)$ describes the magnitude $v=|\mathbf v|\ge0$ ; obtaining it requires multiplying the isotropic Gaussian by the spherical-shell factor $4\pi v^2$ , which vanishes at $v=0$ and shifts the peak to $v_p=\sqrt{2k_BT/m}$ . They share the same exponential but differ by the geometric Jacobian.

When does the Maxwell-Boltzmann distribution break down?

When quantum statistics matter — that is, when the thermal de Broglie wavelength becomes comparable to the interparticle spacing, $n\lambda^3\gtrsim1$ . Then indistinguishability forces Fermi-Dirac (fermions) or Bose-Einstein (bosons) statistics, and the velocity distribution is no longer Gaussian. For ordinary gases at room temperature and pressure this regime is far away ( $n\lambda^3\sim10^{-6}$ ), so the classical Gaussian is excellent; it fails in white-dwarf electrons, ultracold atomic gases, and conduction electrons in metals.

Why are all three component variances equal?

Because the distribution is isotropic — the exponent depends only on the speed $v=\sqrt{v_x^2+v_y^2+v_z^2}$ , which is rotationally invariant. Rotational invariance means the choice of axes is arbitrary, so any statistical average for $v_x$ must equal the same average for $v_y$ or $v_z$ . Hence $\langle v_x^2\rangle=\langle v_y^2\rangle=\langle v_z^2\rangle=k_BT/m$ , and their sum gives $\langle v^2\rangle=3k_BT/m$ .

Can I derive the speed distribution from this?

Yes — it is the natural next step. Take the isotropic three-dimensional Gaussian $f(\mathbf v)$ and integrate over all directions at fixed speed $v$ . The solid-angle integral gives the shell area $4\pi v^2$ , so $f(v)\,dv=4\pi v^2\,(m/2\pi k_BT)^{3/2}e^{-mv^2/2k_BT}\,dv$ . That is the Maxwell-Boltzmann speed distribution, whose peak, mean, and rms are the characteristic molecular speeds.