Midterm 2 - Math 5010 - Spring 2016 Name: Firas Rassoul-Agha Solve the following 5 problems. Note that there is also a 6th Extra Credit problem. You have to clearly explain your solution. The answer carries no points. Only the work does. CALCULATORS ARE NOT ALLOWED. Problem 1: Let X and Y be two continuous random variables with joint density given by −(x+y) 2e if 0 ≤ x ≤ y, f (x, y) = 0 otherwise. Find the marginal densities of X and Y . Solution: The marginal of X is given by Z ∞ Z Z ∞ −(x+y) −x 2e dy = 2e f (x, y) dy = fX (x) = −∞ x ∞ e−y dy = 2e−2x , x when x ≥ 0 and fX (x) = 0 for x < 0. The marginal of Y is given by Z ∞ Z y Z y −(x+y) −y fY (y) = f (x, y) dx = 2e dx = 2e e−x dx = 2e−y (1 − e−y ), −∞ 0 0 when y ≥ 0 and fY (y) = 0 for y < 0. 1 Problem 2: Let (X, Y ) be continuous random variables with joint density f (x, y) = (x + y)/8, 0 ≤ x ≤ 2, 0 ≤ y ≤ 2; f (x, y) = 0 elsewhere. Find the probability that X 2 + Y ≤ 1. Solution: The probability in question equals the integral of f (x, y) over the required region {(x, y) : x2 + y ≤ 1}. Since f (x, y) is not zero only when both 0 ≤ x ≤ 2 and 0 ≤ y ≤ 2, we see that we need to integrate (x + y)/8 below the parabola y = 1−x2 , but only in the first quadrant. In other words, the probability in question equals Z 0 1 Z 0 1−x2 Z 1 Z 1−x2 x y dy dx + dy dx 8 8 0 0 0 0 Z 1 Z 1 x(1 − x2 ) (1 − x2 )2 dx + dx = 8 16 0 0 Z 1 2x − 2x3 + 1 − 2x2 + x4 = dx 16 0 1 − 1/2 + 1 − 2/3 + 1/5 = 16 30 − 15 + 30 − 20 + 6 31 = = ≈ 6.5%. 16 × 2 × 3 × 5 480 x+y dy dx = 8 Z 1 Z 1−x2 2 Problem 3: Let Z be a random variable with standard normal distri2 bution, i.e. Z has pdf f (z) = √12π e−z /2 . Find the pdf of X = Z 2 . Do you recognize this distribution? (You can peak at the pdf in Problem 4.) Solution: First note that X ≥ 0 be because it is a square. So we know that fX (x) = 0 if x < 0. Assume now x > 0 (it does not really matter what the value of fX is at the one point x = 0). To find fX (x) we apply the change of variables formula: dz fX (x) = fZ (z) · dx √ √ 2 where z solves x = z . But this equation has two solutions x and − x, both of which are admissible (since Z can be negative or positive). Therefore, we apply the above to each of the solution and both formulas up. We thus have √ 2 √ 2 √ √ 1 1 1 fX (x) = √ e−( x) /2 |( x)0 | + √ e−(− x) /2 |(− x)0 | = √ x−1/2 e−x/2 . 2π 2π 2π The variable part x−1/2 e−x/2 is of the form xα−1 e−λx where α = 1/2 and λ = 1/2. So X is a Gamma(1/2,1/2) random variable. 3 Problem 4: Let α, β, λ be positive parameters. Consider two independent random variables X and Y with Gamma(α, λ) and Gamma(β, λ) distribution. This means the two have pdfs λα α−1 −λx x e when x > 0 and 0 otherwise, Γ(α) λβ β−1 −λy fY (y) = y e when y > 0 and 0 otherwise. Γ(β) fX (x) = Compute the pdf of V = X + Y . Do you recognize this distribution? At some point you will need to compute an integral of the form R v Hint: α−1 u (v − u)β−1 du. Change variables from u to s via u = vs. This will 0 lead you to showing that this integral in fact equals a constant times v α+β−1 . Remark: This can be done using moment generating functions. It is a good exercise to work it out that way. But at this point, we did not yet know these objects and so we do it using the transformation method. Solution: We need a two-to-two transformation with V = X + Y and such that inverting the transformation will not be too difficult. One easy way to achieve this is to set U = X. Then the inverse of the transformation is given by X=U and Y = V − U. We will use the formula fU,V (u, v) = fX,Y (x, y) × |detJ|. The Jacobian matrix J is given by " J= ∂u ∂x ∂v ∂x ∂u ∂y ∂v ∂y # 1 0 = 1 1 the determinant of which equals one. Since X and Y are independent we know that λα α−1 −λx λβ β−1 −λy fX,Y (x, y) = fX (x)fY (y) = x e × y e Γ(α) Γ(β) when x > 0 and y > 0 and zero otherwise. 4 Extra Page Then the joint pdf of (U, V ) then equals λα α−1 −λx λβ β−1 −λy x e × y e ×1 Γ(α) Γ(β) λα α−1 −λu λβ = u e × (v − u)β−1 e−λ(v−u Γ(α) Γ(β) λα+β = uα−1 (v − u)β−1 e−λv . Γ(α)Γ(β) fU,V (u, v) = This holds when u > 0 and v − u > 0. Otherwise, fU,V (u, v) = 0. Since we want the pdf of V we have to integrate the u out of the above joint pdf to get Z v Z ∞ λα+β −λv uα−1 (v − u)β−1 du. fV (v) = fU,V (u, v) du = e Γ(α)Γ(β) 0 −∞ To finish we need to compute the integral. The hint suggests the change of variables s = u/v. This turns the integral to Z v Z 1 α−1 β−1 u (v − u) du = v (sv)α−1 (v − sv)β−1 ds. 0 0 Note how the boundaries of the integral changed (since s = 0 when u = 0 and s = 1 when u = v). Also, the factor v outside the integral is there because u = sv means du = v ds. Factoring v out of (v − sv) and then factoring the v’s outside the integral we get Z Z 1 1 sα−1 (1 − s)β−1 ds. (sv)α−1 (v − sv)β−1 ds = v α+β−1 v 0 0 The remaining integral only depends on α and β, not on v. So we see that fV (v) = constant × v α+β−1 e−λv when v > 0 and fV (v) = 0 for v < 0. This says that V is a Gamma(α + β,λ) random variable. The constant equals λα+β Γ(α)Γ(β) R1 0 sα−1 (1 − s)β−1 ds and since this should equal λα+β−1 /Γ(α + β) we deduce the extra fact that Z 1 Γ(α + β) sα−1 (1 − s)β−1 ds = . Γ(α)Γ(β) 0 Problem 5: If Z1 and Z2 are two independent standard normal distributions, what is the distribution of Z12 + Z22 ? Hint: Put together your answers to Problems 3 and 4. Solution: From problem 3 we know that Z12 and Z22 are both Gamma(1/2,1/2) random variables. Since Z1 and Z2 are independent, so are Z12 and Z22 . But from Problem 4 we know that the sum of two independent Gamma(1/2,1/2) random variables gives a Gamma(1,1/2) random variable. Hence, Z12 + Z22 is a Gamma(1,1/2) random variable (which is an Exponential(1/2) random variable). 6 Extra Credit: (no partial credit) If Z1 , Z2 , . . . , Zn are independent standard normal distributions, what is the distribution of Z12 + Z22 + . . . + Zn2 ? This random variable appears frequently in statistics and is called a Chi Square random variable with n degrees of freedom. Solution: By the same reasoning as for Problem 5 we get that Z12 + Z22 + . . . + Zn2 is a Gamma(n/2,1/2) random variable. 7