Thermodynamic Limit and Fluctuation Suppression

A cup of water at room temperature does not spontaneously freeze on one side and boil on the other, even though nothing in the microscopic laws forbids it. The reason is statistical: such a large fluctuation is astronomically improbable. But small fluctuations happen constantly — the energy of a system in a heat bath, or the number of molecules in a region, is always jittering around its mean. The central fact of the thermodynamic limit is that while these fluctuations never vanish, their relative size shrinks as the system grows, scaling as $1/\sqrt{N}$ . This single scaling law is the quiet foundation of all of thermodynamics. It is why "temperature," "pressure," and "energy density" are sharp, reproducible numbers for a macroscopic sample rather than fuzzy, ill-defined averages. It is why the different statistical ensembles agree. And it is why thermodynamics — a theory with no fluctuations at all — is an exact description of macroscopic matter: the fluctuations are simply too small to see. For a mole of gas the relative energy fluctuation is about one part in $10^{12}$ , far below any measurement. The scaling follows from a fluctuation–response relation. The energy variance in the canonical ensemble is $\langle\Delta E^2\rangle=k_BT^2C_V$ , and both the heat capacity $C_V$ and the mean energy $U$ are extensive — proportional to $N$ . Therefore the standard deviation $\sigma_E=\sqrt{k_BT^2C_V}$ grows only as $\sqrt{N}$ , while the mean $U$ grows as $N$ , and their ratio falls as $1/\sqrt{N}$ . For the monatomic ideal gas this becomes the exact, memorable result $\sigma_E/\langle E\rangle=\sqrt{2/(3N)}$ .

⚡ Key Takeaways

Fluctuations of extensive quantities never disappear, but their relative size shrinks: $\sigma_X/\langle X\rangle\sim1/\sqrt{N}$ .
The energy fluctuation obeys the fluctuation–response relation $\langle\Delta E^2\rangle=k_BT^2C_V$ ; both $C_V$ and $U$ are extensive.
For the monatomic ideal gas this gives the exact result $\sigma_E/\langle E\rangle=\sqrt{2/(3N)}$ .
Numerically that is $\approx8\%$ for $N=100$ but $\approx10^{-12}$ for a mole — which is why macroscopic thermodynamics has no visible fluctuations.
The same $1/\sqrt{N}$ suppression is what makes intensive variables sharp and the statistical ensembles equivalent.
Key idea: the standard deviation of an extensive quantity grows as $\sqrt{N}$ while its mean grows as $N$ , so the relative fluctuation vanishes as $1/\sqrt{N}$ in the thermodynamic limit.

💡 The Core Insight

Extensive quantities (energy, particle number, volume) fluctuate with a standard deviation $\propto\sqrt{N}$ but a mean $\propto N$ , so their relative fluctuation dies as $1/\sqrt{N}$ . This vanishing relative width is why macroscopic thermodynamic variables are sharp and reproducible, and why the different ensembles give the same answers. For the monatomic ideal gas, $\sigma_E/\langle E\rangle=\sqrt{2/(3N)}$ .

The physical picture

Consider a system in a heat bath. Energy flows back and forth, so at any instant the system's energy differs from its long-time average by some $\Delta E$ . The size of this jitter is set by the heat capacity: a system that can soak up a lot of energy for a small temperature change (large $C_V$ ) fluctuates more. But here is the crucial point about scaling. Double the system — put two identical copies side by side. The energies of the two halves fluctuate independently, so their fluctuations partly cancel: the variance of the total is the sum of the variances (grows as $N$ ), so the standard deviation grows only as $\sqrt{N}$ . Meanwhile the mean energy simply doubles (grows as $N$ ). Relative jitter therefore falls. This is the same law of large numbers that makes a poll of ten thousand people far more reliable than a poll of ten. Independent contributions add in variance, not in standard deviation, so averaging suppresses relative noise as $1/\sqrt{N}$ . A macroscopic sample is a poll of $10^{23}$ molecules; its "opinion" — the energy density — is sharp to a part in $10^{12}$ .

Figure

Independent subsystems: fluctuations add in variance, so the standard deviation of the total energy grows as √N while the mean grows as N — the relative jitter falls as 1/√N.

The punchline plot shows the relative energy fluctuation of an ideal gas versus $N$ on log axes: a clean straight line of slope $-1/2$ , dropping from a visible $8\%$ at $N=100$ to an invisible $10^{-12}$ at a mole. The dashed line marks the resolution of any real experiment — everything to its right is thermodynamics with no measurable noise.

Figure

Punchline: the exact ideal-gas relative energy fluctuation √(2/(3N)) versus N on log axes — a straight line of slope −1/2 falling from 8% at N=100 to 10⁻¹² at N_A, crossing below any experimental resolution.

The core equations

The size of energy fluctuations in the canonical ensemble is fixed, not free — it is a second derivative of the log partition function, and it equals a thermodynamic response function. This is the fluctuation–response relation for energy.

\langle\Delta E^2\rangle \equiv \langle E^2\rangle-\langle E\rangle^2 = \frac{\partial^2\ln Z}{\partial\beta^2} = k_BT^2\,C_V \qquad (1)

where: $Z$ — canonical partition function; $\beta=1/k_BT$ ; $C_V=(\partial U/\partial T)_V$ — heat capacity at constant volume. The variance of the energy is $k_BT^2$ times the heat capacity: fluctuation on the left, response on the right. Since $C_V$ and $U=\langle E\rangle$ are both extensive, dividing the standard deviation by the mean produces the $1/\sqrt{N}$ scaling explicitly.

\frac{\sigma_E}{\langle E\rangle}=\frac{\sqrt{k_BT^2C_V}}{U}\;\propto\;\frac{\sqrt{N}}{N}=\frac{1}{\sqrt{N}} \qquad (2)

where: $C_V=Nc$ and $U=Nu$ with intensive $c,u$ , so $\sigma_E/\langle E\rangle=\sqrt{k_BT^2c/u^2}\,/\sqrt{N}$ . The prefactor is intensive; only the $1/\sqrt{N}$ carries the size dependence. For the monatomic ideal gas everything is explicit: $U=\tfrac32 Nk_BT$ and $C_V=\tfrac32 Nk_B$ , giving an exact closed form with no unknown constants.

\boxed{\;\frac{\sigma_E}{\langle E\rangle}=\sqrt{\frac{2}{3N}}\;} \qquad (\text{monatomic ideal gas}) \qquad (3)

where: the number $2/3$ comes from $k_BT^2C_V/U^2=(k_BT^2\cdot\tfrac32 Nk_B)/(\tfrac32 Nk_BT)^2=2/(3N)$ . This is the cleanest possible illustration of the thermodynamic-limit suppression. Because the variance is $k_BT^2C_V\ge0$ , the relation also forces a stability condition: a positive variance requires a positive heat capacity in the canonical ensemble.

\langle\Delta E^2\rangle=k_BT^2C_V\ge0\;\Longrightarrow\;C_V\ge0 \qquad (4)

where: the left side is manifestly a variance (non-negative), so $C_V$ cannot be negative in a canonical (fixed- $T$ ) description. A negative $C_V$ can appear only in the microcanonical ensemble for non-extensive systems — a signal of ensemble inequivalence.

Derivation AThe energy fluctuation equals kT²C_V

The goal is to prove the fluctuation–response relation $\langle\Delta E^2\rangle=k_BT^2C_V$ directly from the canonical partition function, showing that the spontaneous energy jitter of a system in a heat bath is completely determined by its heat capacity. The strategy is to recognize that successive $\beta$ -derivatives of $\ln Z$ generate the cumulants of the energy: the first gives the mean, the second gives the variance.

Step 1. In the canonical ensemble $\langle E\rangle=-\partial\ln Z/\partial\beta$ , because each Boltzmann weight $e^{-\beta E}$ brings down a factor $-E$ upon differentiation.

Step 2. Differentiate again: $\partial^2\ln Z/\partial\beta^2=-\partial\langle E\rangle/\partial\beta=\langle E^2\rangle-\langle E\rangle^2=\langle\Delta E^2\rangle$ . The second logarithmic derivative is always the variance (the second cumulant).

Step 3. Convert the $\beta$ -derivative to a $T$ -derivative: $\partial/\partial\beta=-k_BT^2\,\partial/\partial T$ . Hence $\langle\Delta E^2\rangle=-\partial\langle E\rangle/\partial\beta=k_BT^2\,\partial U/\partial T=k_BT^2C_V$ . Why the logarithm. Derivatives of $Z$ itself give raw moments; derivatives of $\ln Z$ give cumulants — mean, variance, skewness — which is why $\ln Z$ (not $Z$ ) is the generating function of choice and why the second derivative is precisely the variance.

Step 4. Test on the ideal gas: $Z\propto\beta^{-3N/2}$ , so $\ln Z=-\tfrac{3N}{2}\ln\beta+\text{const}$ . Then $\langle E\rangle=-\partial_\beta\ln Z=\tfrac{3N}{2\beta}$ and $\langle\Delta E^2\rangle=\partial^2_\beta\ln Z=\tfrac{3N}{2\beta^2}$ . Both match $k_BT^2C_V$ with $C_V=\tfrac32 Nk_B$ . Verified: symbolic $\partial^2_\beta\ln Z=\tfrac{3N}{2\beta^2}$ equals $k_BT^2C_V$ after substituting $\beta=1/k_BT$ , $C_V=\tfrac32 Nk_B$ — verify_aq-sm-thermodynamic-limit-fluctuations.py.

Derivation BThe relative fluctuation scales as one over root N

The goal is to turn the fluctuation–response relation into the central scaling law of the thermodynamic limit, $\sigma_E/\langle E\rangle\sim1/\sqrt{N}$ , and to nail the exact ideal-gas coefficient. The key is that $C_V$ and $U$ are both extensive, so their combination in the ratio leaves a pure $1/\sqrt{N}$ dependence — the statistical shadow of the law of large numbers.

Step 1. Write the extensive quantities as $C_V=Nc$ and $U=Nu$ with $c,u$ intensive (per-particle). Then $\sigma_E=\sqrt{k_BT^2C_V}=\sqrt{k_BT^2 Nc}$ grows as $\sqrt{N}$ .

Step 2. The mean $\langle E\rangle=U=Nu$ grows as $N$ . Their ratio is $\sigma_E/\langle E\rangle=\sqrt{k_BT^2 Nc}/(Nu)=\sqrt{k_BT^2c/u^2}\cdot N^{-1/2}$ .

Step 3. The prefactor $\sqrt{k_BT^2c/u^2}$ is intensive (no $N$ ), so the entire size dependence is $N^{-1/2}$ . Multiply by $\sqrt{N}$ and the result is $N$ -independent — the defining signature of $1/\sqrt{N}$ scaling. Why it must be $1/\sqrt N$ . Fluctuations of independent subsystems add in variance ( $\propto N$ ), so the deviation grows as $\sqrt N$ ; the mean grows as $N$ ; the ratio therefore falls as $1/\sqrt N$ . Any extensive observable behaves the same way.

Step 4. For the monatomic ideal gas, $c=\tfrac32 k_B$ and $u=\tfrac32 k_BT$ , so $k_BT^2c/u^2=k_BT^2\cdot\tfrac32 k_B/(\tfrac32 k_BT)^2=2/3$ . Hence $\sigma_E/\langle E\rangle=\sqrt{2/(3N)}$ — equation (3), exact. Verified: $\sigma_E/\langle E\rangle=\sqrt{2/(3N)}$ symbolically; multiplying by $\sqrt N$ gives an $N$ -independent constant; numeric $1.05\times10^{-12}$ at $N=N_A$ and $8.16\%$ at $N=100$ — verify script.

Limiting cases

The scaling law bridges the noisy world of small systems and the deterministic world of thermodynamics.

Case	Result	Meaning
$N\to\infty$	$\sigma_E/\langle E\rangle\to0$	Thermodynamic limit; intensive variables are sharp, ensembles equivalent.
Monatomic ideal gas	$\sigma_E/\langle E\rangle=\sqrt{2/(3N)}$	Exact; the reference case.
$N\sim10^2$ (nanocluster)	$\sim10\%$	Fluctuations visible; ensemble choice and finite-size effects matter.
$N=1$ (single molecule)	$O(1)$	"Thermodynamics" of one particle is meaningless; only distributions exist.
Near a critical point	$C_V\to\infty$ , ratio grows	Fluctuations become anomalously large even at macroscopic $N$ ; the $1/\sqrt{N}$ argument breaks near $T_c$ .

⚠️ Common pitfall

The $1/\sqrt{N}$ law suppresses relative fluctuations, not absolute ones. The absolute variance $\langle\Delta E^2\rangle=k_BT^2C_V$ actually grows with system size (it is extensive). What vanishes is the ratio to the mean. A second trap: the scaling assumes a normal (short-range) system where $C_V$ and $U$ are extensive and $C_V$ stays finite. Near a critical point $C_V$ diverges, and for long-range forces $U$ is super-extensive — in both cases the naive $1/\sqrt{N}$ argument fails and fluctuations can be macroscopically large. The law is the generic behavior, not a universal guarantee.

Worked examples

Example 1 — Relative energy fluctuation of a mole of gas (enumerate, apply, boundary number). Take one mole of a monatomic ideal gas, $N=N_A=6.022\times10^{23}$ , at any temperature.

Step 1. Use the exact formula $\sigma_E/\langle E\rangle=\sqrt{2/(3N)}$ .

Step 2. Substitute: $\sigma_E/\langle E\rangle=\sqrt{2/(3\times6.022\times10^{23})}=\sqrt{1.107\times10^{-24}}=\mathbf{1.05\times10^{-12}}$ .

Step 3. Interpret: the energy of a mole of gas is defined to a part in $10^{12}$ purely by statistics. No thermometer or calorimeter comes within many orders of magnitude of resolving this jitter, which is why thermodynamics — a fluctuation-free theory — describes macroscopic gases exactly.

Step 4. Boundary number: for the fluctuation to reach a detectable $1\%$ you would need $\sqrt{2/(3N)}=0.01$ , i.e. $N=2/(3\times10^{-4})=\mathbf{6.7\times10^{3}}$ atoms — a cluster a few nanometers across. Below that, energy fluctuations are experimentally real.

Example 2 — Absolute versus relative for a nanocluster. Take $N=100$ argon atoms at $T=300\,\mathrm{K}$ ; compare the absolute and relative energy fluctuations.

Step 1. Mean energy $U=\tfrac32 Nk_BT=\tfrac32(100)(1.380649\times10^{-23})(300)=\mathbf{6.21\times10^{-19}\,J}$ .

Step 2. Variance $\langle\Delta E^2\rangle=k_BT^2C_V=k_BT^2\cdot\tfrac32 Nk_B=(1.380649\times10^{-23})(300)^2(\tfrac32)(100)(1.380649\times10^{-23})=2.57\times10^{-39}\,\mathrm{J^2}$ , so $\sigma_E=\mathbf{5.07\times10^{-20}\,J}$ .

Step 3. Relative fluctuation $\sigma_E/U=5.07\times10^{-20}/6.21\times10^{-19}=\mathbf{0.082}$ , i.e. $8.2\%$ — exactly $\sqrt{2/300}$ . For a 100-atom cluster the energy genuinely jitters by nearly a tenth of its mean, and the concept of a sharp "temperature" is already fraying.

central limit theorem

probability theory · de Moivre 1733 · Lyapunov 1901 · the mathematics behind 1/√N

The central limit theorem is the mathematical engine behind the vanishing of relative fluctuations. It states that the sum of a large number $N$ of independent random contributions, each with finite variance, has a distribution approaching a Gaussian whose absolute width grows only as $\sqrt{N}$ while its mean grows as $N$ — so the relative width shrinks as $1/\sqrt{N}$ . The total energy of a system with short-range interactions is exactly such a sum: it is built from $\sim N$ roughly independent local contributions. That is why the relative energy fluctuation $\sigma_E/\langle E\rangle$ falls as $1/\sqrt{N}$ and the canonical energy distribution collapses onto a sharp value in the thermodynamic limit. First glimpsed by de Moivre in 1733 and developed by Laplace, then put on a rigorous footing by Lyapunov in 1901, the theorem is the reason macroscopic thermodynamic quantities are sharp even though every microscopic degree of freedom jitters — the statistical-mechanical property of "self-averaging."

⏳ History

1827

— Robert Brown observes the erratic motion of pollen grains, later named Brownian motion.

1905

— Einstein's fluctuation theory of Brownian motion links diffusion to molecular agitation.

1906

— Smoluchowski independently derives the statistical theory of Brownian fluctuations.

1908

— Smoluchowski and Einstein explain critical opalescence via diverging density fluctuations.

1908

— Perrin's experiments on Brownian motion confirm the fluctuation predictions and pin down N_A.

Where this lives today

The $1/\sqrt{N}$ suppression of relative fluctuations is the license under which every thermodynamic measurement is interpreted, but its breakdown is where much modern physics lives. Nanoscience routinely operates where $N$ is small enough that fluctuations are large: single-molecule experiments, nanoscale calorimetry, and quantum dots all show energy and number fluctuations that would be invisible in bulk. Stochastic thermodynamics has turned these fluctuations into a resource, with fluctuation theorems (Jarzynski, Crooks) extracting equilibrium free energies from the statistics of small-system trajectories. And near continuous phase transitions the heat capacity diverges, so fluctuations grow anomalously large even in macroscopic samples — the regime where renormalization-group theory replaces the naive law of large numbers.

Going further

References. Pathria & Beale, Statistical Mechanics, 4th ed., §3.6–3.7 and §4.5 · Reif, Fundamentals of Statistical and Thermal Physics, §6.5, §10 · Landau & Lifshitz, Statistical Physics Part 1, §112 (fluctuations) · Kardar, Statistical Physics of Particles, §4 · Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity, Ch. 3.

Provenance. The fluctuation–response relation $\langle\Delta E^2\rangle=k_BT^2C_V$ and the $1/\sqrt{N}$ scaling are standard (Pathria §3.6, Reif §6.5, Landau–Lifshitz §112). A verifier should confirm: (i) $\partial^2_\beta\ln Z=\tfrac{3N}{2\beta^2}$ equals $k_BT^2C_V$ for the ideal gas; (ii) $\sigma_E/\langle E\rangle=\sqrt{2/(3N)}$ exactly; (iii) $\sigma_E/\langle E\rangle\cdot\sqrt N$ is $N$ -independent (the $1/\sqrt N$ law); (iv) numeric $1.05\times10^{-12}$ at $N=N_A$ and $8.16\%$ at $N=100$ ; (v) $\langle\Delta E^2\rangle\ge0\Rightarrow C_V\ge0$ . All in verify_aq-sm-thermodynamic-limit-fluctuations.py. Constants: $k_B=1.380649\times10^{-23}\,\mathrm{J/K}$ , $N_A=6.02214076\times10^{23}$ .

Builds on: Canonical Ensemble · Energy Fluctuations in the Canonical Ensemble · Heat Capacity and Fluctuation-Response Relation · Factorization of the Partition Function Leads to: Derivation of the Canonical Distribution from the Microcanonical Ensemble · Helmholtz Free Energy from the Partition Function · Grand Canonical Ensemble

If fluctuations shrink as 1/√N, do they ever really matter?

Yes, in two important regimes. First, for small systems: at $N\sim100$ the relative energy fluctuation is already $\sim8\%$ , and single-molecule and nanoscale experiments live entirely in this regime. Second, near critical points: there the heat capacity diverges, so $\sigma_E/\langle E\rangle$ can be large even for macroscopic $N$ , and the whole $1/\sqrt{N}$ argument breaks down. Fluctuations are negligible only for ordinary macroscopic systems away from transitions — which happens to cover most of everyday thermodynamics, and is precisely why thermodynamics works.

Why do variances add but not standard deviations?

Because variance is the mean-square deviation, and for independent contributions the cross terms average to zero: $\mathrm{Var}(X_1+X_2)=\mathrm{Var}(X_1)+\mathrm{Var}(X_2)$ . Standard deviation is the square root of variance, so it adds "in quadrature," not linearly. When you combine $N$ independent subsystems, the total variance is $N$ times a single one, so the total standard deviation is $\sqrt{N}$ times a single one — while the mean is $N$ times a single one. The ratio therefore falls as $1/\sqrt{N}$ . This is the law of large numbers.

Is the ⟨ΔE²⟩=kT²C_V relation exact or an approximation?

It is exact in the canonical ensemble — it follows directly from $\langle\Delta E^2\rangle=\partial^2\ln Z/\partial\beta^2$ and the definition $C_V=\partial U/\partial T$ , with no approximation. What is approximate (to $O(\ln N/N)$ ) is the statement that this canonical variance also describes an isolated (microcanonical) system, since strictly the microcanonical energy variance is zero. But as a canonical identity, $\langle\Delta E^2\rangle=k_BT^2C_V$ holds for any system, small or large, interacting or not.

Does this mean a negative heat capacity is impossible?

In the canonical ensemble, yes: $\langle\Delta E^2\rangle=k_BT^2C_V$ is a variance, which cannot be negative, so $C_V\ge0$ whenever temperature is the fixed control variable. But a negative heat capacity is genuinely real in the microcanonical ensemble for non-extensive systems — most famously self-gravitating star clusters, which get hotter as they lose energy. This is not a contradiction; it is a signature of ensemble inequivalence, and it is exactly the regime where the $1/\sqrt{N}$ argument (which assumes short-range, extensive behavior) does not apply.

How is this connected to the equivalence of ensembles?

It is the mechanism behind it. Ensemble equivalence holds because the canonical energy distribution is a sharp spike around its mean, with relative width $\sigma_E/\langle E\rangle\sim1/\sqrt{N}$ . As $N\to\infty$ that spike becomes a delta function at the microcanonical energy, so fixing the temperature (canonical) becomes indistinguishable from fixing the energy (microcanonical). The vanishing relative fluctuation is the reason different ensembles give the same thermodynamics; this article supplies the quantitative scaling, and the equivalence-of-ensembles result draws the conclusion.

What sets the numerical prefactor 2/3 for the ideal gas?

It comes entirely from the ideal gas having $\tfrac32$ degrees of freedom's worth of energy per particle. With $U=\tfrac32 Nk_BT$ and $C_V=\tfrac32 Nk_B$ , the combination $k_BT^2C_V/U^2=(k_BT^2)(\tfrac32 Nk_B)/(\tfrac32 Nk_BT)^2=2/(3N)$ . The $\tfrac32$ in $C_V$ and the $(\tfrac32)^2$ in $U^2$ combine to give $2/3$ . A diatomic gas with $C_V=\tfrac52 Nk_B$ and $U=\tfrac52 Nk_BT$ would give $2/(5N)$ instead — the prefactor tracks the number of active quadratic degrees of freedom.