A random variable X being proper means that it is well-defined over its probability space and does not take on infinite values with any positive probability. If E[X], the expectation, is finite, then X must be proper because a finite expectation implies that the probability mass of X does not concentrate at infinity. Thus, the finite expectation ensures that X maintains finite valuedness and does not diverge, affirming its propriety as a random variable .

To demonstrate E[X] = Σ xiP({xi}) for a nonnegative, discrete random variable X, consider each point xi in its sample space with probability P({xi}). The expectation E[X] is the sum of the product of each possible value xi and its corresponding probability P({xi}), mathematically represented as Σ xiP({xi}). This expands the definition of expectation which accounts for the values X can take weighted by their probabilities. The finiteness of E[X] relies on the convergence of this series, assuming that it is absolutely convergent .

Consider a random variable X such that P(X = 2^n) = 1/(n^2) for n = 1, 2, 3,... This setup ensures that E[X] = Σ (2^n/n^2) is finite because this series converges. However, E[X^2] = Σ (4^n/n^2) diverges, as it resembles a form of exponential growth with a polynomially decreasing probability which does not sufficiently counterbalance the increase, leading to an infinite sum for E[X^2]. This example demonstrates how careful choice of probabilities and values can control finite expectation yet result in infinite second moments .

The almost sure versions of convergence theorems vary in conditions and applications. MCT applies when {Xn} is an increasing sequence and concludes that E[Xn] converges to E[X] if Xn → X almost surely. DCT requires that Xn is dominated by an integrable function, M, meaning |Xn| ≤ M for all n, allowing for pointwise convergence to imply expected convergence. BCT handles convergence of {Xn}, assuming each Xn is bounded by a fixed constant, enabling E[Xn] to converge to E[X]. While similar in enforcing pointwise conditions translate to expectations, they differ in criteria: monotonicity (MCT), domination (DCT), and boundedness (BCT).

The variance of a linear transformation aX + b is derived using the properties of variance. Since variance, Var, measures spread around the mean, adding a constant b does not alter variability, so Var(aX + b) = Var(aX). Further, scaling X by a factor of 'a' scales its variance by a^2, due to the properties of scaling in variance: Var(aX) = a^2Var(X). Thus, Var(aX + b) = a^2Var(X) given that Var measures deviation around the mean, and scaling affects this deviation by the square of the scalar .

The problem set highlights the dual approach of collaboration and independent practice as essential in mastering stochastic processes. Collaboration fosters exchange of ideas, clarifies concepts through discussion, and exposes learners to diverse problem-solving techniques, enhancing understanding. Independent practice, however, is crucial for internalizing knowledge, testing personal comprehension, and developing individual problem-solving abilities without external aid or resources. This combination ensures a deep and thorough grasp, reinforcing learning through collective and personal exploration of complex stochastic concepts .

When Xn is such that Xn ≥ Y for all n and Xn ↑ X, by the Monotone Convergence Theorem (MCT), the expectation of the sequence E[Xn] will also converge to E[X]. This relies on the properties of monotonicity and boundedness: since Xn is increasing and bounded, and the sequence of expectations is also non-decreasing, MCT ensures E[Xn] → E[X].

Fatou's Lemma in the context of almost sure convergence provides a relation for expectations of almost surely convergent random variables. It states that for a sequence of nonnegative random variables {Xn} converging almost surely to X, the expectation satisfies lim inf E[Xn] ≥ E[lim inf Xn]. This implies that the lower bound of the expectations is preserved in the limit and provides a way to bridge pointwise convergence with mean convergence, supporting the handling of convergent series even when the individual expectations are not fully known .

To construct the piece-wise linear approximation of the semicircle X(ω) = √(0.52 - (ω-0.5)^2), begin by dividing the interval [0,1] into n sub-intervals. Within each sub-interval, approximate the arc of the semicircle using linear segments that connect the calculated endpoints of the circle arc over these sub-intervals. As n increases, these linear segments will more precisely approximate the semicircle curve, and the function Xn(ω) representing the nth approximation will approach X(ω). This sequence of random variables, Xn, will converge to X, and the expectations will satisfy E[X] = lim E[Xn] as the linear approximations become more precise, reflecting convergence in mean .

For the sequence Xn = nI(0, 1/n)(ω), lim Xn as n approaches infinity is 0 almost everywhere since the intervals (0, 1/n) shrink to the null set. Regarding the expectations, E[Xn] = n * P(0 < ω < 1/n) = n * 1/n = 1. Therefore, without considering bounds at infinity, E[X] = lim E[Xn] does not hold as lim E[Xn] = 1 while E[X] = 0, highlighting that expectations may not align with limits in pointwise convergence when limits are influenced by shrinking intervals .