Mastering Random Variables: A Math154 Tutorial on Gamma, Maxwell, Benford, Cauchy, and Support

Introduction to Random Variables in Math154

Random variables are the backbone of probability theory and statistics. In this tutorial, we tackle the key problems from Math154 Homework 3, covering the Gamma, Maxwell, Benford, and Cauchy distributions, as well as the concept of support. Whether you're studying econometrics, physics, or data science, these distributions appear everywhere—from modeling waiting times to analyzing first digits in financial data. Let's dive in with clear derivations and real-world analogies.

The Gamma Distribution: Shape and Rate Parameters

The Gamma distribution with shape \(\alpha > 0\) and rate \(\lambda > 0\) has PDF \(f(x) = \frac{\lambda^\alpha}{\Gamma(\alpha)} x^{\alpha-1} e^{-\lambda x}\) for \(x \geq 0\). It's widely used in econometrics to model waiting times or insurance claims. For example, think of the time until a stock market crash—Gamma distributions can capture such risks.

a) Special Case: \(\alpha = 1\)

When \(\alpha = 1\), the Gamma PDF reduces to \(f(x) = \lambda e^{-\lambda x}\), which is the exponential distribution. This makes sense: the exponential is a special case of the Gamma with shape 1.

b) Verifying the PDF Properties

To be a valid PDF, \(f(x)\) must be non-negative and integrate to 1 over \([0,\infty)\). Non-negativity is clear. The integral: \(\int_0^\infty \frac{\lambda^\alpha}{\Gamma(\alpha)} x^{\alpha-1} e^{-\lambda x} dx = 1\) by definition of the Gamma function.

c) Expectation and Variance

Using the moment generating function or direct integration: \(E[X] = \frac{\alpha}{\lambda}\) and \(Var[X] = \frac{\alpha}{\lambda^2}\). For example, if \(\alpha = 2, \lambda = 0.5\), the mean is 4 and variance is 8.

d) Moment Generating Function

\(M_X(t) = E[e^{tX}] = \left(1 - \frac{t}{\lambda}\right)^{-\alpha}\) for \(t < \lambda\). This is derived by completing the exponential integral.

e) Gamma in \(L^p\) for All \(p\)

Since the Gamma distribution has finite moments of all orders (its MGF exists in a neighborhood of 0), it belongs to \(L^p\) for any \(p > 0\). This is crucial for convergence theorems.

The Maxwell Distribution: Modeling Molecular Speeds

The Maxwell distribution for speed \(v \geq 0\) has PDF \(f(v) = \sqrt{\frac{2}{\pi}} \frac{v^2}{\theta^3} e^{-v^2/(2\theta^2)}\) with scale parameter \(\theta > 0\). It's used in physics to model the speed of gas molecules. Think of the latest AI-driven climate models that simulate particle speeds—Maxwell is key.

To verify it's a PDF, integrate from 0 to \(\infty\): substitute \(u = v^2/(2\theta^2)\) to get \(\int_0^\infty f(v) dv = 1\). The expectation: \(E[V] = \int_0^\infty v f(v) dv = 2\theta \sqrt{\frac{2}{\pi}}\).

Benford's Law: First Digit Distribution

Benford's law states that the first significant digit \(d\) (1-9) appears with probability \(P(d) = \log_{10}(1 + 1/d)\). For example, digit 1 appears about 30.1% of the time. This law is observed in many real datasets: stock prices, physical constants, and even the lengths of rivers. A viral TikTok trend showed how Benford's law can detect election fraud—amazing!

a) Expectation and Variance

Expectation: \(E[D] = \sum_{d=1}^9 d \log_{10}(1+1/d) \approx 3.44\). Variance: \(E[D^2] - (E[D])^2\) with \(E[D^2] = \sum d^2 \log_{10}(1+1/d) \approx 11.03\), so variance ≈ 0.86.

b) Sequence \(2^n\) and Benford

The sequence \(2^n\) (for \(n=1,2,\ldots\)) follows Benford's law. For instance, \(2^1=2\), \(2^2=4\), \(2^3=8\), \(2^4=16\) (first digit 1), etc. A histogram of first digits from \(2^n\) for \(n=1\) to 10000 shows the Benford distribution.

The Cauchy Distribution: When Means Fail

The centered Cauchy distribution has PDF \(f(x) = \frac{1}{\pi(1+x^2)}\). It's famous for having no finite mean or variance. In finance, Cauchy distributions model extreme market moves—like the 2020 crash or meme stock surges.

a) Not in \(L^1\)

The expectation \(E[|X|] = \int_{-\infty}^\infty \frac{|x|}{\pi(1+x^2)} dx\) diverges (like \(\ln(1+x^2)\) at infinity), so \(X \notin L^1\).

b) Convergence in the Cauchy Sense

The Cauchy principal value of the expectation is 0: \(\lim_{a\to\infty} \int_{-a}^a x f(x) dx = 0\). This generalized expectation exists.

c) Variance and Higher Moments

Variance is infinite; all moments beyond the first (in the principal value sense) are undefined. The MGF does not exist because the tails are too heavy.

d) Generating Cauchy with Cot[Pi Random[]]

If \(U \sim \text{Uniform}(0,1)\), then \(\cot(\pi U)\) is Cauchy distributed. This is because the CDF of Cauchy is \(F(x) = \frac{1}{\pi}\arctan(x) + 1/2\), and the inverse transform yields \(\tan(\pi(U-1/2)) = \cot(\pi U)\).

Support of a Distribution: Constructing Measures

The support \(K\) of a measure \(\mu\) is the smallest closed set with \(\mu(K^c)=0\). Here we construct measures with various supports.

a) Absolutely Continuous with Cantor-like Support

Take a modified Cantor set of positive measure (fat Cantor set). For example, remove middle thirds of length \(1/3^n\) but keep total length >0. Define PDF as the indicator of this set normalized by its measure. The support is the fat Cantor set.

b) Singular Continuous with Support [0,1]

The Cantor distribution (standard Cantor set) is singular continuous with support the Cantor set (measure zero). To get support [0,1], take a mixture: e.g., the Cantor distribution convolved with a uniform? Actually, the standard Cantor distribution has support the Cantor set, not [0,1]. To get support [0,1], we need a singular continuous measure that assigns positive measure to every interval. One example: the distribution of \(\sum_{n=1}^\infty X_n/2^n\) where \(X_n\) are i.i.d. with \(P(X_n=0)=P(X_n=1)=1/2\), but this is uniform on [0,1] (absolutely continuous). For singular continuous, use base 3 digits with unequal probabilities: \(P(0)=1/2, P(2)=1/2\). This yields a singular continuous measure with support the Cantor set. To get support [0,1], we need a dense set of points. One construction: take a random variable \(X = \sum_{n=1}^\infty Y_n/2^n\) where \(Y_n\) are i.i.d. with distribution that is singular continuous? This is tricky. Alternatively, use the distribution of a random variable with CDF that is the Cantor function but with a dense set of jumps? Actually, the Cantor function is continuous and singular, and its support is the Cantor set (not [0,1]). To have support [0,1], we need a measure that is singular but its support is [0,1]. For example, take a dense set of points \(\{q_n\}\) in [0,1] and assign masses \(2^{-n}\) to each. That's pure point, not singular continuous. For singular continuous, one can use a random series with digits that produce a dense set. The standard construction: let \(X = \sum_{n=1}^\infty a_n/2^n\) where \(a_n\) are i.i.d. with \(P(a_n=0)=1/2, P(a_n=1)=1/2\). Then \(X\) is uniform on [0,1] (absolutely continuous). To make it singular, use base 3 digits with probabilities \(P(0)=1/2, P(2)=1/2\) and then map to [0,1] via a bijection? The support of that is the Cantor set. So to get support [0,1], we need a bijection from the Cantor set to [0,1] that pushes the measure forward. For example, let \(\phi\) be the Cantor function (which maps Cantor set onto [0,1] continuously). Then the pushforward of the Cantor measure under \(\phi\) is the uniform distribution on [0,1] (absolutely continuous). That's not singular. Actually, the Cantor function is monotone and continuous, and it maps the Cantor set (measure zero) to [0,1], but the pushforward of the Cantor measure is the Lebesgue measure? Wait: The Cantor function is constant on intervals removed, so it maps the Cantor set onto [0,1] in a one-to-one way (except endpoints). The Cantor measure is supported on the Cantor set, and its pushforward under the Cantor function is the Lebesgue measure on [0,1] (since the Cantor function is the CDF of the Cantor distribution). So the pushforward is absolutely continuous. To get a singular continuous measure with support [0,1], one can take the Cantor measure and then apply a homeomorphism that maps the Cantor set onto [0,1] but is not absolutely continuous. For instance, take a strictly increasing continuous function \(g\) that maps the Cantor set onto [0,1] but has derivative zero almost everywhere (like the Cantor function but with a different scaling). Then the pushforward of the Cantor measure under \(g\) is singular continuous with support [0,1].

c) Pure Point with Support [0,1]

Enumerate all rationals in [0,1] as \(q_1, q_2, \ldots\) and assign mass \(2^{-n}\) to each. This discrete measure has support [0,1] because rationals are dense.

d) Any Closed Set \(K \subset [0,1]\)

For any closed \(K\), take the uniform distribution on \(K\) if \(K\) has positive Lebesgue measure; otherwise, take a weighted sum of Dirac masses at a dense countable subset of \(K\). The measure's support is exactly \(K\).

Conclusion

This tutorial covered the Gamma, Maxwell, Benford, and Cauchy distributions, plus the concept of support. These ideas are not just academic—they appear in econometrics, physics, data science, and even viral internet trends. Keep practicing with real datasets, and you'll master random variables in no time!