The Gaussian Distribution and Its Ubiquity
~6 min read · Tier toggle changes depth (T1 plain · T2 standard · T3 rigour)
You should be comfortable with the definition of a probability density function (PDF) and the concept of the Central Limit Theorem (CLT) as established in the previous lecture.
Why this matters
The Gaussian distribution is the single most important function in experimental physics.
In the previous lecture, we established that the sum of many independent random variables tends toward a normal distribution, regardless of the underlying distributions of those variables. This lecture formalises the properties of that limiting distribution, the Gaussian (or normal) distribution. We will derive its key statistical moments, demonstrate why it is the maximum entropy distribution for a given variance, and show how it dictates the propagation of uncertainty in linear measurements. Mastery of this section is essential because it provides the mathematical foundation for the error bars you will calculate in every subsequent experimental unit.
The Shape of Uncertainty
Imagine you are measuring the position of a particle in a detector. The detector is not perfect; it has noise, electronic fluctuations, and slight mechanical vibrations. If you repeat the measurement many times, you will not get the same number every time. Instead, you will get a spread of values. The Gaussian distribution describes the most natural way this spread occurs when many small, independent sources of error combine.
Unlike a uniform distribution, where every value in a range is equally likely, the Gaussian is peaked. It tells us that values close to the mean are much more probable than values far away. Crucially, it is symmetric: there is no inherent bias towards higher or lower values. This symmetry and the rapid decay of the probability as you move away from the centre make it the ideal model for "random noise" in physical systems. It is the mathematical embodiment of the idea that while individual fluctuations are unpredictable, the overall pattern of fluctuations is perfectly predictable.
Mathematical Definition and Normalisation
The probability density function (PDF) of a Gaussian distribution with mean \(\mu\) and standard deviation \(\sigma\) is given by:
Here, \(\mu\) is the location of the peak (the mean), and \(\sigma\) controls the width of the distribution. The constant \(1/(\sigma \sqrt{2\pi})\) is the normalisation factor. It ensures that the total probability integrates to 1 over the entire real line. This is a direct consequence of the Gaussian integral:
By substituting \(a = 1/(2\sigma^2)\) and shifting the variable by \(\mu\), we confirm that \(\int_{-\infty}^{\infty} f(x) \, dx = 1\). Note that the Gaussian is defined for all real numbers, meaning there is a non-zero, albeit infinitesimally small, probability of measuring a value arbitrarily far from the mean. In practice, we often truncate the distribution, but mathematically, the tails extend to infinity.
Moments and Variance
The first moment of the distribution is its mean. By symmetry, the mean is exactly \(\mu\). The second central moment is the variance, \(\sigma^2\). We can calculate the variance \(\text{Var}(X) = E[(X - \mu)^2]\) using integration by parts or by differentiating the characteristic function. The result is:
This is a defining property: the parameter \(\sigma\) in the PDF is literally the standard deviation of the distribution. This makes the Gaussian unique among common distributions in that its shape is fully determined by its first two moments. Higher-order moments are also fixed by \(\sigma\). For example, the fourth central moment (related to kurtosis) is \(3\sigma^4\). This specific relationship between moments is a hallmark of the Gaussian and is used in tests for normality in data analysis.
Connection to the Central Limit Theorem
Recall the Central Limit Theorem (CLT) from the previous lecture. If \(X_1, X_2, \dots, X_N\) are independent and identically distributed (i.i.d.) random variables with finite mean \(\mu\) and finite variance \(\sigma^2\), then the standardised sum:
converges in distribution to a standard Gaussian \(\mathcal{N}(0,1)\) as \(N \to \infty\). This theorem explains the ubiquity of the Gaussian in physics. Most physical measurements are sums of many microscopic interactions (e.g., thermal motion of atoms, photon counting, electronic noise). Even if the individual interactions are not Gaussian, their sum will be approximately Gaussian if \(N\) is large. This is why we use Gaussian error bars: they are the natural statistical description of the aggregate effect of many small, independent uncertainties.
The derivation
The Central Limit Theorem
Let X1, X2, …, XN be independent, identically distributed random variables with finite mean μ = E[Xi] and finite, nonzero variance σ2 = Var(Xi). Form the sum SN = ∑i=1N Xi and standardise it: ZN = (SN − Nμ)/(σ√N). Then as N → ∞, ZN converges in distribution to a standard Gaussian: ZN →d 𝒩(0,1), meaning P(ZN ≤ z) → Φ(z) = ∫−∞z (1/√(2π)) e−u2/2 du for every z. The limiting shape is Gaussian regardless of the shape of the parent distribution, provided only that its variance is finite.
A common misconception is that the Gaussian distribution implies that all errors are independent. While the CLT requires independence for the strict convergence, the Gaussian shape can also arise from correlated errors if the correlation structure is specific. However, in standard experimental physics, we assume independence to justify the simple addition of variances. If your errors are strongly correlated, the simple Gaussian model may underestimate the true uncertainty. C
Worked example
Problem. A physicist measures the decay time of a particle. The detector has a Gaussian resolution with a standard deviation \(\sigma = 2.5\ \mathrm{ns}\). The true mean decay time is \(\mu = 100\ \mathrm{ns}\). What is the probability that a single measurement falls within the range \(95\ \mathrm{ns}\) to \(105\ \mathrm{ns}\)?
- Identify the parameters: \(\mu = 100\ \mathrm{ns}\), \(\sigma = 2.5\ \mathrm{ns}\).
- Define the range in terms of standard deviations (z-scores). The lower bound is \(95\ \mathrm{ns}\), which is \(100 - 5\ \mathrm{ns}\). The upper bound is \(105\ \mathrm{ns}\), which is \(100 + 5\ \mathrm{ns}\).
- Calculate the z-scores: \(z_{\text{lower}} = \frac{95 - 100}{2.5} = -2\) and \(z_{\text{upper}} = \frac{105 - 100}{2.5} = 2\).
- We need the probability \(P(-2 \le Z \le 2)\) for a standard Gaussian \(Z \sim \mathcal{N}(0,1)\).
- Using the empirical rule (or standard normal tables), the probability within \(\pm 2\sigma\) is approximately \(95.45\% \).
Units check. The z-score calculation is dimensionless, as required. The final probability is a dimensionless number between 0 and 1, which is correct.
Three quick tests
- What is the value of the PDF at the mean \(\mu\) for a Gaussian with standard deviation \(\sigma\)?
Answer
\( \frac{1}{\sigma \sqrt{2\pi}} \) - If you double the standard deviation \(\sigma\), how does the peak height of the PDF change?
Answer
It halves, because the area under the curve must remain 1. - Why is the Gaussian distribution considered "maximum entropy" for a given variance?
Answer
It represents the most "uncertain" or "random" distribution consistent with the known mean and variance, implying no additional structure or bias.
Problem set for PU-106 →
Each derivation in this unit ends with five graded problems with full solutions.
Lecture notes drafted with a local model and checked line by line by the editor; every number in the worked example was recomputed.
Derivations cited: D-118 · Threads: chance