Questions: Distribution Functions and Moments

Question 1

Why do physicists prefer cumulants over raw moments for characterizing fluctuations in statistical mechanical systems?

• A. Cumulants are always smaller in magnitude than the corresponding raw moments, making calculations more tractable
• B. Cumulants are additive for statistically independent subsystems, while raw moments are not — so they directly reflect the independent contributions of uncorrelated parts
• C. Cumulants are defined only for Gaussian distributions, which is the primary distribution in statistical mechanics
• D. Cumulants measure the average value of an observable, while moments measure fluctuations

Answer: (B)

The key advantage is additivity: if A = A₁ + A₂ for independent subsystems, then κₙ(A) = κₙ(A₁) + κₙ(A₂) for all n. Raw moments do not have this property. This reflects the physics: uncorrelated subsystems contribute independently to fluctuations, so a measure that is additive over independent contributions is the natural language for statistical mechanics. The additivity also underlies the cluster expansion, where the nth cumulant corresponds to an irreducible n-body correlation.

Question 2

A distribution has mean μ, variance σ², zero skewness, and zero excess kurtosis. What can you conclude?

• A. The distribution is uniform on an interval
• B. The distribution is a Poisson distribution
• C. The distribution is Gaussian — all cumulants beyond the second are zero, which uniquely characterizes a Gaussian
• D. The distribution is exponential, since skewness and kurtosis vanish only for symmetric distributions

Answer: (C)

The defining property of the Gaussian distribution is that ALL cumulants beyond κ₂ (the variance) are identically zero. Zero skewness means κ₃ = 0; zero excess kurtosis means κ₄ = 0. If both hold (and by extension all higher cumulants vanish), the distribution is Gaussian. This is the precise mathematical statement of why the Gaussian is 'the simplest' distribution — it is fully characterized by just two parameters. Any nonzero higher cumulant is a direct signature of non-Gaussianity.

Question 3

The variance of a distribution and its second cumulant κ₂ are the same quantity.

• T. True
• F. False

Answer: True

Correct. The second cumulant κ₂ = ⟨(ΔA)²⟩ = ⟨A²⟩ − ⟨A⟩², which is precisely the variance σ². The first cumulant κ₁ = ⟨A⟩ is the mean. Higher cumulants differ from the corresponding central moments: for example, κ₄ = ⟨(ΔA)⁴⟩ − 3σ⁴, which subtracts the 'trivial' contribution of the Gaussian part. This correction is exactly what makes higher cumulants measure *non-Gaussianity* rather than raw moment size.

Question 4

For a Gaussian distribution, most cumulants are zero — including the mean and variance.

• T. True
• F. False

Answer: False

Only cumulants *beyond* the second are zero for a Gaussian. The first cumulant κ₁ = ⟨A⟩ (the mean) is nonzero unless the distribution is centered at zero. The second cumulant κ₂ = σ² (the variance) is nonzero for any non-degenerate Gaussian. The defining property is that κₙ = 0 for all n ≥ 3. Saying 'all cumulants are zero' would describe a degenerate point mass distribution, not a Gaussian.

Question 5

Why does a nonzero fourth cumulant (excess kurtosis) near a phase transition indicate something physically significant?

Model answer: Near a critical point, fluctuations in extensive quantities (like energy or order parameter) become correlated across the entire system, breaking the statistical independence that underlies Gaussian behavior. The fourth cumulant κ₄ measures departures from Gaussianity caused by these long-range correlations — it can diverge at criticality while the mean and variance vary smoothly. This makes higher cumulants sensitive probes of phase structure; for example, the kurtosis of baryon number distributions is predicted to change sign and diverge near the QCD critical point, providing a measurable experimental signature.

The deeper reason is that additivity of cumulants holds only for *independent* subsystems. At a critical point, correlations span the entire system, destroying the independence assumption. The cumulant generating function encodes these correlations through its higher derivatives, so the nth cumulant reflects irreducible n-body correlations. The Gaussian approximation (keeping only κ₁ and κ₂) misses the critical structure entirely — this is why thermodynamics, which operates with means and variances, cannot detect a critical point as sharply as higher cumulant measurements can.