Particle-Number Fluctuations

Draw an imaginary box inside a jar of gas and count the molecules in it. Come back a moment later and count again — the number has changed. Nothing was added or removed from the jar; the molecules simply wandered across the invisible boundary. In the grand canonical ensemble this is not a nuisance but the whole point: a system exchanging particles with a reservoir at fixed temperature $T$ and chemical potential $\mu$ has a particle number $N$ that genuinely fluctuates, and the size of those fluctuations is a window into the physics of the fluid. The remarkable thing is what controls the fluctuations. You might expect the variance $\langle\Delta N^2\rangle$ to be some intractable microscopic quantity, but it turns out to equal a familiar laboratory response function — the isothermal compressibility $\kappa_T$ , the very softness you feel when you squeeze the fluid. A stiff, nearly incompressible liquid holds its density fixed, so $N$ barely fluctuates; a soft, compressible gas lets density wander, so fluctuations are large. This is the density-fluctuation twin of the energy fluctuation–response relation $\langle\Delta E^2\rangle=k_BT^2C_V$ : a spontaneous jitter equals a measured susceptibility. Two consequences dominate everything that follows. First, for a macroscopic system the relative fluctuation is only $\sim 1/\sqrt N$ , so $N$ is effectively sharp and the grand canonical ensemble reproduces the fixed- $N$ canonical thermodynamics — the formal license to use whichever ensemble is convenient. Second, at a critical point $\kappa_T$ diverges, fluctuations swell to macroscopic and optical scales, and the fluid turns milky: critical opalescence. Every numeric and sign claim below is python-verified.

⚡ Key Takeaways

In the grand canonical ensemble $N$ fluctuates; its variance is the second derivative of $\ln\Xi$ with respect to $\beta\mu$ : $\langle\Delta N^2\rangle=(z\partial_z)^2\ln\Xi=k_BT(\partial\langle N\rangle/\partial\mu)_{T,V}$ .
That susceptibility is the isothermal compressibility: $\langle\Delta N^2\rangle=\dfrac{\langle N\rangle^2 k_BT}{V}\kappa_T$ — the density-fluctuation analogue of $\langle\Delta E^2\rangle=k_BT^2C_V$ .
Relative fluctuations vanish as $1/\sqrt{\langle N\rangle}$ , so the grand canonical and canonical ensembles agree in the thermodynamic limit — the justification for ensemble equivalence.
The classical ideal gas is Poissonian: $\langle\Delta N^2\rangle=\langle N\rangle$ , the $\kappa_T=1/(nk_BT)$ special case.
Quantum statistics shift this: bosons bunch ( $\langle\Delta N^2\rangle>\langle N\rangle$ ) and fermions anti-bunch ( $\langle\Delta N^2\rangle<\langle N\rangle$ ).
Near a critical point $\kappa_T\to\infty$ : fluctuations become macroscopic, ensemble equivalence fails, and the fluid opalesces.
Key idea: one second derivative of $\ln\Xi$ delivers both the variance of $N$ and the number susceptibility; converting variables turns it into the compressibility, so $\langle\Delta N^2\rangle=(k_BT\langle N\rangle^2/V)\kappa_T$ .

💡 The Core Insight

A single second derivative of $\ln\Xi$ with respect to $\beta\mu$ gives simultaneously the variance of the particle number and the susceptibility $\partial\langle N\rangle/\partial\mu$ . Converting to laboratory variables turns that susceptibility into the isothermal compressibility, so $\langle\Delta N^2\rangle=(k_BT\langle N\rangle^2/V)\kappa_T$ — density fluctuations are governed by how easily the fluid can be squeezed.

The physical picture

Hold a fixed, imaginary sub-box inside a large fluid and watch the count $N$ of particles inside it. Because particles drift freely across the invisible walls, $N$ is never exactly its average — it jitters. How much it jitters is set by how easily the local density can change, and that is exactly what the compressibility measures. A nearly incompressible liquid resists density changes, so the count stays tight; a dilute, soft gas lets the density roam, so the count swings widely. The same "softness" that governs the response to a pressure squeeze governs the spontaneous fluctuations — that is the fluctuation–response principle in its number-fluctuation guise.

Figure

A fixed imaginary box inside a fluid: particles cross the boundary so N fluctuates. Stiff fluid → tight count; soft (compressible) fluid → wide swings.

The engine that turns this picture into equations is the fugacity $z=e^{\beta\mu}$ acting as a counting variable. In the grand partition function $\Xi=\sum_N z^N Q_N$ , each extra power of $z$ tags one more particle, so $\Xi$ is a moment-generating function for $N$ exactly as in elementary probability theory. Differentiate $\ln\Xi$ with respect to $\ln z$ once and you get the mean; twice and you get the variance. No new physics — just the same generating-function calculus that gave energy fluctuations from $\ln Z$ , now applied to the particle-number direction. The punchline is what happens when you approach a critical point. Away from transitions the Fano factor $\langle\Delta N^2\rangle/\langle N\rangle$ sits at its Poisson value of $1$ (for a classical gas), and relative fluctuations are microscopic. But the compressibility $\kappa_T$ diverges at the liquid–gas critical point, so the Fano factor shoots up: density fluctuations grow to the scale of visible light, scatter it strongly, and a transparent fluid abruptly turns milky. The figure below shows this knee — the observable fingerprint of a diverging susceptibility.

Figure

The Fano factor ⟨ΔN²⟩/⟨N⟩ vs temperature: flat at the Poisson value 1 away from T_c, diverging as T → T_c because κ_T → ∞ — this knee is critical opalescence.

The core equations

An open system's state is weighted by the Gibbs factor $e^{-\beta(E-\mu N)}$ , and summing over all states and all particle numbers gives the grand partition function. The fugacity $z=e^{\beta\mu}$ organizes the sum by particle number, so $\Xi$ is literally a power series in $z$ whose $N$ -th coefficient is the fixed- $N$ canonical partition function:

\Xi(T,V,\mu)=\sum_{N=0}^{\infty} z^{N}\,Q_N(T,V),\qquad z\equiv e^{\beta\mu} \qquad (1)

where: $\Xi$ — grand partition function; $z$ — fugacity; $Q_N$ — canonical partition function at fixed $N$ ; $\beta=1/k_BT$ ; $\mu$ — chemical potential. Reading it as a generating function is the key move: each factor $z^N$ carries the "particle count" $N$ , so derivatives with respect to $z$ pull down powers of $N$ . The probability of finding exactly $N$ particles is the corresponding term normalized by $\Xi$ , and the mean number is the first logarithmic derivative:

P(N)=\frac{z^{N}Q_N}{\Xi},\qquad \langle N\rangle=z\frac{\partial \ln\Xi}{\partial z}=\frac{1}{\beta}\frac{\partial \ln\Xi}{\partial \mu}\Big|_{T,V} \qquad (2)

where: $P(N)$ — probability of $N$ particles; $\langle N\rangle$ — mean particle number. The operator $z\partial_z$ and the operator $\beta^{-1}\partial_\mu$ are the same thing, since $z=e^{\beta\mu}$ means $\partial_\mu = \beta z\partial_z$ . Applying that operator a second time produces the variance — the connected (cumulant) form in which the mean-squared subtracts off cleanly. This is the central identity of the article:

\langle\Delta N^2\rangle=\langle N^2\rangle-\langle N\rangle^2=\Big(z\frac{\partial}{\partial z}\Big)^2\ln\Xi=\frac{1}{\beta^2}\frac{\partial^2 \ln\Xi}{\partial \mu^2}\Big|_{T,V} \qquad (3)

where: $\langle\Delta N^2\rangle$ — variance of $N$ . Because $\ln\Xi$ is the cumulant-generating function of the particle-number distribution (in the variable $\beta\mu$ ), its first derivative is the mean and its second is the variance — the generic fluctuation–response structure. Recognizing the second derivative as a thermodynamic response gives the fluctuation–response relation for particle number: the variance equals the temperature times the number susceptibility $\partial\langle N\rangle/\partial\mu$ .

\langle\Delta N^2\rangle=k_BT\left(\frac{\partial \langle N\rangle}{\partial \mu}\right)_{T,V} \qquad (4)

where: $(\partial\langle N\rangle/\partial\mu)_{T,V}$ — number susceptibility, the response of $N$ to a change in chemical potential. Because a variance is non-negative, this guarantees $(\partial\langle N\rangle/\partial\mu)_{T,V}\ge 0$ : adding chemical potential can only pull particles in — a stability statement. The susceptibility is not directly measurable, but a short chain of thermodynamic identities converts it into the isothermal compressibility, which is measurable:

\langle\Delta N^2\rangle=\frac{\langle N\rangle^2 k_BT}{V}\,\kappa_T,\qquad \kappa_T\equiv -\frac{1}{V}\Big(\frac{\partial V}{\partial p}\Big)_{T,N} \qquad (5)

where: $\kappa_T$ — isothermal compressibility; $p$ — pressure. This is the density-fluctuation analogue of $\langle\Delta E^2\rangle=k_BT^2 C_V$ : a spontaneous fluctuation set equal to a laboratory response function. Two immediate readings follow. For a macroscopic system the variance is extensive ( $\propto N$ ) while the mean is $\propto N$ too, so the relative fluctuation shrinks as $N^{-1/2}$ ; and for independent single-particle modes the total variance is a sum of per-mode variances that reveal quantum statistics:

\frac{\sqrt{\langle\Delta N^2\rangle}}{\langle N\rangle}\sim\frac{1}{\sqrt N}\xrightarrow{N\to\infty}0,\qquad \langle\Delta n_k^2\rangle=\langle n_k\rangle(1\pm\langle n_k\rangle) \qquad (6)

where: $n_k$ — occupation of single-particle mode $k$ ; upper sign $+$ — bosons (bunching, variance enhanced); lower sign $-$ — fermions (anti-bunching, variance suppressed). The classical Poisson result $\langle\Delta n_k^2\rangle\approx\langle n_k\rangle$ is recovered when $\langle n_k\rangle\ll1$ .

Derivation AFrom the number susceptibility to the compressibility — the owned bridge

This is the article's owned structural result: the microscopic variance $\langle\Delta N^2\rangle=z\partial_z\langle N\rangle$ read off the grand partition function and the thermodynamic form $\langle\Delta N^2\rangle=(k_BT\langle N\rangle^2/V)\kappa_T$ read off the compressibility are the same number. The goal is to convert the number susceptibility $(\partial\langle N\rangle/\partial\mu)_{T,V}$ into $\kappa_T$ using homogeneity and Gibbs–Duhem, so the two derivations of the variance meet. The strategy: exploit that for a homogeneous fluid $\mu$ depends only on density and temperature, then chain-rule through the pressure.

Step 1. Trade $N$ for density. For a homogeneous fluid the chemical potential depends only on $T$ and the density $n=N/V$ , not separately on $V$ , so $(\partial\langle N\rangle/\partial\mu)_{T,V}=V(\partial n/\partial\mu)_T$ . Why: extensivity means the intensive state is fixed by intensive variables — $\mu(n,T)$ — so scaling out the volume isolates the density response.

Step 2. Use Gibbs–Duhem at fixed $T$ . The Gibbs–Duhem relation $\mathrm d\mu=-s\,\mathrm dT+v\,\mathrm dp$ (with $s$ entropy per particle, $v=1/n$ volume per particle) gives, at constant $T$ , $(\partial\mu/\partial p)_T=v=1/n$ , hence $(\partial p/\partial\mu)_T=n$ . Why: at fixed temperature the only way to change the chemical potential of a one-component fluid is to change its pressure, and the exchange rate is exactly the volume per particle.

Step 3. Chain-rule the density onto the pressure. Write $(\partial n/\partial\mu)_T=(\partial n/\partial p)_T(\partial p/\partial\mu)_T=n\,(\partial n/\partial p)_T$ , using the result of Step 2. Why: the density responds to $\mu$ only through the pressure it induces, so the two rates multiply.

Step 4. Identify the compressibility. Since $n=N/V$ at fixed $N$ , a pressure change moves $V$ and hence $n$ : $(\partial n/\partial p)_T=n\kappa_T$ with $\kappa_T=-V^{-1}(\partial V/\partial p)_T$ . Therefore $(\partial n/\partial\mu)_T=n^2\kappa_T$ . Why: squeezing the fluid ( $\kappa_T$ large) is exactly what lets the local density — and thus the count in the box — respond, which is why softness and fluctuations are the same thing.

Step 5. Assemble. Combining, $(\partial\langle N\rangle/\partial\mu)_{T,V}=Vn^2\kappa_T=\langle N\rangle^2\kappa_T/V$ ; multiplying by $k_BT$ from Eq. (4) gives $\langle\Delta N^2\rangle=(k_BT\langle N\rangle^2/V)\kappa_T$ . Dimensionally $[\kappa_T][k_BT][N^2/V]=\mathrm{Pa}^{-1}\cdot\mathrm{J}\cdot\mathrm{m}^{-3}$ , which is dimensionless — correct for a variance. Why: the microscopic generating-function variance and the macroscopic compressibility are two faces of the same curvature of the grand potential. *Verified: the chain $(\partial N/\partial\mu)_{T,V}=Vn^2\kappa_T=N^2\kappa_T/V$ is checked symbolically, and both the microscopic $(z\partial_z)^2\ln\Xi$ and the compressibility form reduce to $\langle\Delta N^2\rangle=\langle N\rangle$ for the ideal gas, in verify_aq-sm-particle-number-fluctuations.py.*

Derivation BQuantum mode variances: bunching and anti-bunching

The goal is to show that for ideal quantum gases the number variance splits into a sum of independent per-mode variances, each of the form $\langle n_k\rangle(1\pm\langle n_k\rangle)$ , and that the sign of the correction to Poisson statistics is fixed by the statistics: bosons enhance fluctuations (bunching), fermions suppress them (anti-bunching). The strategy is to treat each single-particle level as its own tiny grand canonical system and apply the fluctuation–response relation mode by mode.

Step 1. One mode is an open subsystem. For an ideal gas each single-particle level $k$ of energy $\varepsilon_k$ exchanges particles with the reservoir independently, with grand sum $\Xi_k=\sum_{n_k} e^{-\beta(\varepsilon_k-\mu)n_k}$ — a geometric series (bosons, $n_k=0,1,2,\dots$ ) or a two-term sum (fermions, $n_k=0,1$ ). Why: in a non-interacting gas the occupation of one level says nothing about another, so the grand partition function factorizes and each mode can be analyzed alone.

Step 2. Mean occupation. Summing gives $\Xi_k=(1-ze^{-\beta\varepsilon_k})^{-1}$ (Bose) or $1+ze^{-\beta\varepsilon_k}$ (Fermi), and $\langle n_k\rangle=z\partial_z\ln\Xi_k=1/(e^{\beta(\varepsilon_k-\mu)}\mp1)$ , the Bose–Einstein ( $-$ ) and Fermi–Dirac ( $+$ ) distributions. Why: the same $z\partial_z$ that counted total particles now counts the occupation of a single level.

Step 3. Variance from the response. Applying $\langle\Delta n_k^2\rangle=k_BT\,\partial\langle n_k\rangle/\partial\mu$ and differentiating the distribution gives $\langle\Delta n_k^2\rangle=\langle n_k\rangle(1+\langle n_k\rangle)$ for bosons and $\langle n_k\rangle(1-\langle n_k\rangle)$ for fermions. Why: the fluctuation–response relation applies to each mode, and the derivative of the $\mp1$ occupation function carries the sign that distinguishes the two statistics.

Step 4. Sum over modes and read the physics. Independence lets variances add: $\langle\Delta N^2\rangle=\sum_k\langle n_k\rangle(1\pm\langle n_k\rangle)=\langle N\rangle\pm\sum_k\langle n_k\rangle^2$ . The extra positive term for bosons (bunching) makes $\langle\Delta N^2\rangle>\langle N\rangle$ ; the extra negative term for fermions (anti-bunching, capped by Pauli exclusion at $\langle n_k\rangle\le1$ ) makes $\langle\Delta N^2\rangle<\langle N\rangle$ . In the dilute limit $\langle n_k\rangle\ll1$ the correction vanishes and both recover Poisson. Why: exchange symmetry makes identical bosons want to share a mode (positive correlation) and forbids fermions from doing so (negative correlation), and that is exactly what the $\pm\langle n_k\rangle^2$ term encodes. *Verified: $\langle\Delta n_k^2\rangle=k_BT\,\partial\langle n_k\rangle/\partial\mu=\langle n_k\rangle(1\pm\langle n_k\rangle)$ is confirmed symbolically for both statistics, and the numeric instances ( $\langle n_k\rangle=2$ boson $\to$ variance $6>2$ ; $\langle n_k\rangle=0.5$ fermion $\to$ variance $0.25<0.5$ ) in the verify script.*

Limiting cases

The general relation $\langle\Delta N^2\rangle=(k_BT\langle N\rangle^2/V)\kappa_T$ specializes cleanly across regimes:

Case	Result	Meaning
Classical ideal gas	$\langle\Delta N^2\rangle=\langle N\rangle$	$\kappa_T=1/(nk_BT)$ makes the variance equal the mean — Poisson statistics; relative fluctuation $=1/\sqrt{\langle N\rangle}$ .
Ideal Bose gas	$\langle\Delta N^2\rangle>\langle N\rangle$	Bunching: $\langle n_k\rangle(1+\langle n_k\rangle)$ per mode; grows without bound approaching condensation as $\kappa_T\to\infty$ .
Ideal Fermi gas	$\langle\Delta N^2\rangle<\langle N\rangle$	Anti-bunching: $\langle n_k\rangle(1-\langle n_k\rangle)$ ; filled levels ( $\langle n_k\rangle\to1$ ) cannot fluctuate, so Pauli exclusion strongly suppresses fluctuations below $T_F$ .
Critical point	$\kappa_T\to\infty\Rightarrow\langle\Delta N^2\rangle/\langle N\rangle\to\infty$	Macroscopic, long-ranged density fluctuations; ensemble equivalence breaks; critical opalescence.

⚠️ “In an open system N is essentially fixed, so fluctuations never matter”

This is right in the bulk and dangerously wrong near a critical point. In the thermodynamic limit the relative fluctuation is $\sim1/\sqrt N\sim10^{-12}$ for a mole, so $N$ is sharp and the ensemble choice is immaterial — but that smallness is entirely due to $\kappa_T$ being finite. At the liquid–gas critical point $\kappa_T$ diverges, the variance grows super-extensively, density fluctuations reach optical wavelengths, and ensemble equivalence genuinely fails. Two further traps: the fluctuation formula $\langle\Delta N^2\rangle=k_BT(\partial N/\partial\mu)_{T,V}$ holds at fixed $T,V$ — do not silently swap in $(\partial N/\partial\mu)_{T,p}$ , which is a different derivative — and the tidy result $\langle\Delta N^2\rangle=\langle N\rangle$ is a special property of the classical ideal gas, not a universal identity; quantum statistics and interactions both change the Fano factor away from $1$ .

Worked examples

1 · Number fluctuations of a mole of ideal gas (a real number, and why it is invisible)

Take a fixed sub-volume of a classical ideal gas holding on average one mole, $\langle N\rangle=6.022\times10^{23}$ . Compute the RMS particle-number fluctuation and its relative size, and say what it means for measurement.

Step 1. Use the ideal-gas result. For the classical ideal gas the distribution is Poisson, so $\langle\Delta N^2\rangle=\langle N\rangle$ and the RMS fluctuation is $\sigma_N=\sqrt{\langle N\rangle}$ .

Step 2. Plug in the mole. $\sigma_N=\sqrt{6.022\times10^{23}}=\mathbf{7.76\times10^{11}}$ particles. That is a huge absolute number — hundreds of billions of molecules flicker in and out.

Step 3. Now take the relative size. $\sigma_N/\langle N\rangle=1/\sqrt{\langle N\rangle}=1/(7.76\times10^{11})=\mathbf{1.29\times10^{-12}}$ .

Step 4. Interpret. Even though $\sim 10^{12}$ molecules fluctuate, that is only about one part in $10^{12}$ of the total — undetectable by any ordinary density measurement, which is precisely why we can pretend $N$ is fixed and use the canonical ensemble. The absolute fluctuation is enormous; the relative fluctuation is what vanishes as $1/\sqrt N$ , and it is the relative fluctuation that governs whether the ensemble choice matters.

2 · From an invisible fluctuation to a measurable one, to a diverging one

(a) Shrink the sub-volume until it holds $\langle N\rangle=10^{6}$ particles. What is the relative fluctuation now? (b) A fluid is driven near its critical point where the compressibility is $10^{4}$ times the ideal-gas value at the same density. What is the Fano factor $\langle\Delta N^2\rangle/\langle N\rangle$ ?

Step 1. Small subvolume. With $\langle N\rangle=10^6$ and Poisson statistics, $\sigma_N=\sqrt{10^6}=1000$ and the relative fluctuation is $1000/10^6=\mathbf{10^{-3}}$ — one part in a thousand, now genuinely measurable with single-particle-resolved imaging.

Step 2. Set up the critical case. The Fano factor is $F=\langle\Delta N^2\rangle/\langle N\rangle$ . For the ideal gas $F=1$ (Poisson). The compressibility form $\langle\Delta N^2\rangle=(k_BT\langle N\rangle^2/V)\kappa_T$ shows $F$ scales linearly with $\kappa_T$ at fixed density.

Step 3. Apply the enhancement. If $\kappa_T$ is $10^4$ times the ideal-gas value, then $F=1\times10^4=\mathbf{10^{4}}$ .

Step 4. Interpret. The count in the box now swings a hundred times more widely than Poisson would allow ( $\sqrt{10^4}=100$ ), because a nearly incompressible response has been replaced by an almost infinitely soft one. As $\kappa_T\to\infty$ at the exact critical point, $F$ diverges, the fluctuations reach the wavelength of visible light, and the fluid opalesces — the naked-eye signature of a diverging thermodynamic susceptibility.

jean perrin

1870–1942 · Lille / Paris

Jean Perrin was the French physicist who turned fluctuation theory from a prediction into proof that atoms are real. Working from 1908 to 1913 with painstakingly size-graded colloidal suspensions, he measured exactly the quantities that the Einstein–Smoluchowski fluctuation theory had predicted: the mean-square Brownian displacement of a suspended grain, and — most directly relevant here — the equilibrium density fluctuations of grains distributed in the gravitational field. From the exponential sedimentation profile $n(h)\propto e^{-mgh/k_BT}$ and the scatter of grain counts about it, Perrin extracted Avogadro's number from more than a dozen independent methods, all agreeing near $6\times10^{23}$ . Because the grand-canonical result $\langle\Delta N^2\rangle=\langle N\rangle$ (Poisson) fixes the expected count fluctuations, his measured scatter was a measurement of $k_B$ and hence of the molecular scale: the statistical spread of a fluctuating number, made visible under a microscope, decided the century-long debate over atomism in favour of the atom. Perrin received the 1926 Nobel Prize in Physics "for his work on the discontinuous structure of matter." His experiments are why the fluctuation formulas of this article are not merely bookkeeping but a bridge between the microscopic and the measurable — the same bridge Smoluchowski and Einstein built theoretically and Perrin walked across.

⏳ History

1869

— Andrews measures the liquid–gas critical point of carbon dioxide and reports the opalescent glow near it.

1906

— Smoluchowski publishes his kinetic theory of Brownian motion, independently of Einstein.

1908

— Smoluchowski explains critical opalescence quantitatively via diverging density fluctuations.

1910

— Einstein derives the fluctuation-scattering formula, linking opalescence and sky brightness to compressibility.

1913

— Perrin's colloid experiments confirm the fluctuation and Brownian predictions, pinning Avogadro's number and settling atomism.

1914

— Ornstein and Zernike introduce the direct correlation function to handle the long-range critical correlations.

1956

— Hanbury Brown and Twiss observe photon bunching, the optical counterpart of bosonic number fluctuations.

Where this lives today

The fluctuation–compressibility relation is the working tool behind modern scattering experiments. In light and neutron scattering the measured intensity in the forward direction is set by $\langle\Delta N^2\rangle$ , so a scattering experiment is a direct read-out of the isothermal compressibility; near a critical point the divergence of $\kappa_T$ shows up as a dramatic surge in small-angle scattering, and the resulting critical opalescence remains a textbook demonstration performed with a sealed tube of a near-critical fluid. The same sum rule, $nk_BT\kappa_T=1+n\int[g(r)-1]\,\mathrm d^3r$ , ties number fluctuations to the pair correlation function $g(r)$ and is a cornerstone of the theory of simple liquids and of density-functional approaches to inhomogeneous fluids. In quantum-gas and quantum-optics laboratories the per-mode variance formula is measured directly. Single-atom-resolved imaging of trapped Fermi gases sees sub-Poissonian number fluctuations — the density profile is smoother than a classical gas allows, a direct visualization of Pauli suppression and the $\langle n_k\rangle(1-\langle n_k\rangle)$ law. Bosonic gases near condensation show super-Poissonian fluctuations and bunching, the matter-wave analogue of the Hanbury Brown–Twiss photon-bunching effect that founded quantum optics. Engineered "number-squeezed" states, in which fluctuations are pushed below the Poisson limit on purpose, underpin atom interferometry and precision sensing. In every case the experiment is quoting a Fano factor and comparing it to the grand-canonical prediction derived here.

Going further

References. Pathria & Beale, Statistical Mechanics, 3rd ed., §4.5 (energy and density fluctuations in the grand canonical ensemble) · Kardar, Statistical Physics of Particles, Ch. 4 & 7 · Reif, Fundamentals of Statistical and Thermal Physics, §9 (grand canonical ensemble) · Sethna, Statistical Mechanics, Ch. 6 · Landau & Lifshitz, Statistical Physics Part 1, §35, §111–114 (fluctuations and the compressibility) · Smoluchowski, Ann. Phys. (1908); Einstein, Ann. Phys. (1910).

Provenance. All relations are standard grand-canonical results. A verifier should reproduce: $\langle\Delta N^2\rangle=(z\partial_z)^2\ln\Xi=k_BT(\partial\langle N\rangle/\partial\mu)_{T,V}$ (Pathria §4.5, Reif §9); the Maxwell/Gibbs–Duhem chain converting the susceptibility to $\kappa_T$ (Landau–Lifshitz §111–114); the ideal-gas Poisson result $\langle\Delta N^2\rangle=\langle N\rangle$ with $\kappa_T=1/(nk_BT)$ ; and the per-mode variances $\langle n_k\rangle(1\pm\langle n_k\rangle)$ for Bose/Fermi (Pathria Ch. 6, L&L §113). Numerical instances (one-mole $\sigma_N=7.76\times10^{11}$ , relative $1.29\times10^{-12}$ ; subvolume $10^6\to$ relative $10^{-3}$ ; near-critical Fano $10^4$ ) are recomputed in verify_aq-sm-particle-number-fluctuations.py. $\lambda=h/\sqrt{2\pi mk_BT}$ and $k_B$ (CODATA) are standard.

Builds on: Grand Canonical Ensemble · Grand Partition Function · Gibbs Factor and Chemical Potential · Heat Capacity and Fluctuation-Response Relation. Leads to: Energy Fluctuations in the Canonical Ensemble · Thermodynamic Quantities from the Partition Function · Critical Opalescence.

Why should the variance of N have anything to do with the compressibility?

Because both are second derivatives of the same free energy. The number susceptibility $(\partial\langle N\rangle/\partial\mu)_{T,V}$ measures how strongly the particle count responds when you nudge the chemical potential, and a short chain of thermodynamic identities (Gibbs–Duhem at fixed temperature, then the definition of $\kappa_T$ ) shows this response is the isothermal compressibility scaled by $\langle N\rangle^2/V$ . Physically the connection is intuitive: the compressibility measures how easily you can change the fluid's density by squeezing it, and spontaneous density fluctuations are just the fluid changing its own local density without being squeezed. A fluid that is easy to compress is also one whose density wanders freely on its own — softness and fluctuation are the same property viewed from two sides. This is the number-fluctuation instance of the general fluctuation–response (fluctuation–dissipation) theorem, exactly parallel to $\langle\Delta E^2\rangle=k_BT^2C_V$ for energy.

If the box is open, isn't the particle number just whatever the reservoir dictates — why does it fluctuate at all?

The reservoir fixes the chemical potential, not the particle number. Just as a heat bath fixes the temperature and lets the energy fluctuate around a mean set by that temperature, a particle reservoir fixes $\mu$ and lets $N$ fluctuate around a mean set by $\mu$ . Particles are free to cross the boundary in both directions, and at any instant slightly more may have wandered in than out, or vice versa. What the grand canonical ensemble computes is the full probability distribution $P(N)\propto z^N Q_N$ , whose mean is $\langle N\rangle$ and whose spread is $\langle\Delta N^2\rangle$ . For a classical ideal gas that distribution is exactly Poisson, and its width is $\sqrt{\langle N\rangle}$ — small relative to the mean for macroscopic systems, but never zero. The fluctuation is real; it is only negligible in relative terms because $1/\sqrt{\langle N\rangle}$ is tiny when $\langle N\rangle\sim10^{23}$ .

The absolute fluctuation for a mole is 10¹¹ particles — that sounds enormous. Should I worry about it?

Not for bulk thermodynamics, but the number is genuinely large in absolute terms — that is not a mistake. For $\langle N\rangle=6\times10^{23}$ the RMS fluctuation is $\sqrt{\langle N\rangle}\approx7.8\times10^{11}$ particles, so at any instant roughly a trillion molecules more or fewer are in your box than the average. What makes this harmless is that it is only about one part in $10^{12}$ of the total, far below the sensitivity of any macroscopic density measurement. The lesson is to always ask about the relative fluctuation, $\sigma_N/\langle N\rangle=1/\sqrt{\langle N\rangle}$ , not the absolute one. The absolute fluctuation grows with system size (as $\sqrt N$ ), but the relative fluctuation shrinks (as $1/\sqrt N$ ), and it is the relative fluctuation that decides whether $N$ can be treated as effectively fixed and whether the ensemble choice matters.

What exactly is bunching and anti-bunching, and where do the ±⟨n⟩² terms come from?

For an ideal quantum gas each single-particle mode is an independent open subsystem, and its occupation variance works out to $\langle n_k\rangle(1\pm\langle n_k\rangle)$ : the $+$ sign for bosons, the $-$ sign for fermions. The classical part is the $\langle n_k\rangle$ , which alone would give Poisson statistics. The extra $\pm\langle n_k\rangle^2$ is a purely quantum-statistical correlation. For bosons it is positive — identical bosons "prefer" to occupy the same mode, so finding one particle in a mode makes it more likely to find another, and fluctuations are enhanced (bunching). For fermions it is negative — Pauli exclusion forbids two fermions in the same mode, so occupations are anti-correlated and fluctuations are suppressed (anti-bunching); when a mode is fully occupied ( $\langle n_k\rangle\to1$ ) its variance vanishes entirely, since a filled level cannot fluctuate. Bunching is the matter-wave version of the Hanbury Brown–Twiss photon-bunching effect; anti-bunching is directly imaged as sub-Poissonian density noise in degenerate Fermi gases.

Why do fluctuations blow up at a critical point but not elsewhere?

Because the compressibility $\kappa_T$ diverges there, and $\langle\Delta N^2\rangle\propto\kappa_T$ . Away from any phase transition $\kappa_T$ is finite, so the variance is extensive ( $\propto\langle N\rangle$ ) and the relative fluctuation is the usual tiny $1/\sqrt{\langle N\rangle}$ . But at the liquid–gas critical point the distinction between liquid and gas disappears, and the free-energy curvature that normally penalizes density changes flattens out — it costs almost nothing to change the local density, which is exactly what " $\kappa_T\to\infty$ " means. With the restoring cost gone, density fluctuations grow without bound, become correlated over long distances (the correlation length diverges too), and reach the scale of visible light. The fluid then scatters light strongly in all directions and turns from transparent to milky — critical opalescence. This same divergence is why ensemble equivalence breaks down at criticality: the grand canonical $N$ is no longer sharp, so it no longer matches the fixed- $N$ canonical description.

How is any of this actually measured?

Through scattering and through direct particle counting. In light or neutron scattering, the intensity scattered in the near-forward direction is proportional to $\langle\Delta N^2\rangle$ , so measuring how much a fluid scatters is measuring its number fluctuations — and hence, via the fluctuation–compressibility relation, its compressibility. Near a critical point the surge in small-angle scattering (and the visible opalescence) is the divergence of $\kappa_T$ made observable. In cold-atom laboratories the fluctuations are counted directly: quantum-gas microscopes image individual atoms in a lattice or trap, so one can literally histogram the number of atoms in a chosen subvolume over many experimental runs and read off the Fano factor $\langle\Delta N^2\rangle/\langle N\rangle$ . Fermi gases show Fano factors below $1$ (Pauli suppression), thermal Bose gases above $1$ (bunching), and engineered number-squeezed states are pushed deliberately below the Poisson limit for use in precision atom interferometry.

Does the relation ⟨ΔN²⟩ = k_BT(∂N/∂μ) hold at fixed T and V, or fixed T and p?

Fixed $T$ and $V$ — this matters, and getting it wrong is a common error. The grand canonical ensemble holds $T$ , $V$ , and $\mu$ as its control variables, and the fluctuation formula is the second derivative of $\ln\Xi(T,V,\mu)$ with respect to $\beta\mu$ at fixed $T,V$ . So the susceptibility that appears is $(\partial\langle N\rangle/\partial\mu)_{T,V}$ , not $(\partial\langle N\rangle/\partial\mu)_{T,p}$ , which is a genuinely different derivative (and in fact ill-defined for a single-component fluid, where $T$ and $p$ fix $\mu$ completely by Gibbs–Duhem). When you convert the fixed- $(T,V)$ susceptibility into the compressibility, you must carry the constraints correctly through the Maxwell and Gibbs–Duhem relations; the bridge derivation above does exactly that and lands on $\kappa_T=-V^{-1}(\partial V/\partial p)_{T,N}$ . Keeping the held-fixed variables straight is the single most common slip in this calculation.