Partial solutions to assignment 3 Math 5080-1, Spring 2011 p. 283, #5. We know that T1 , T2 , · · · ∼ EXP(100) are independent. Recall that EXP(100) is the same distribution as GAMMA(100 , 1). P10 (a) i=1 Ti ∼ GAMMA(100 , 10). P10 (b) “Successful operation for at least 1.5 years” is the same sentence as “ i=1 Ti > 1.5 × 365 = 547,” assuming that : (1) There are 365 days per year; and (2) The Ti ’s are measured in days. Therefore we are interested in the total area, under the GAMMA(100 , 10) density curve, to the right of 547. Since ) ( 10 ) ( 10 X 2 X Ti > 10.74 , P Ti > 547 = P 100 i=1 i=1 Theorem 8.3.3 [page 269] tells us that we want the probability that if X ∼ χ2 (20), then ( 10 ) X P Ti > 547 = P {X > 10.74}. i=1 According to Table 4 [page 604] of your text, P {X ≤ 9.59} = 0.025} and P {X ≤ 10.85} = 0.05. Therefore, ( 10 ) X 0.95 = P {X > 10.85} ≤ P Ti > 547 ≤ P {X > 9.59} = 0.975. i=1 Therefore, the probability of interest is somewhere between 0.95 and 0.975. [Note that the problem is stated incompletely and has other interpretations. ForP instance, if the Ti ’s are measured in years, then we should seek to compute 10 P { i=1 Ti > 1.5} = P {X > 0.03} ≈ 1. (c) If we continue to assume that the Ti ’s are measured in days and there are 365 days per year, then 2 years is 2 × 365 = 730 days. And the question is stating the following: Find the smallest integer N such that (N ) X P Ti > 730 ≈ 0.95. i=1 The number of N . PN of needed spared is simply N − 1, for this valueP N Now i=1 Ti ∼ GAMMA(100 , N ); therefore, XN := 0.02 I=1 Ti ∼ χ2 (2N ). In other words, we need to find N such that P {XN ≤ 0.02 × 730} = P {XN ≤ 0.02 × 14.6} ≈ 0.05. Use Table 4 on page 604 to see that P {X25 ≤ 14.61} ≈ 0.05. Therefore, we can choose N ≤ 25/2 = 12.5. But we cannot choose N ≤ 24/2 = 12 because 1 P {X24 ≤ 14.6} > P {X24 ≤ 13.85} ≈ 0.05 Therefore, the answer is N = 13; i.e., 13 − 1 = 12 spares are needed. p. 283, #8. We consider MGF’s: MY −X (t) = MY (t)M−X (t) = MY (t)MX (−t). This holds simply because MX (−t) = Ee−tX = M−X (t). Now, we use the MGF table [behind the front cover of the text, for example] to find that if |t| < 1/2, then n/2 m/2 1 1 , MX (−t) = . MY (t) = 1 − 2t 1 + 2t [Otherwise, one the preceding two is infinity.] Therefore, 1 1 1 (1 − 2t)n/2 (1 + 2t)m/2 if − 2 < t < 2 , MY −X (t) = ∞, otherwise. And this is not equal to the MGF of a χ2 (n − m), which is (n−m)/2 1 if −∞ < t < 12 , 1 − 2t ∞ otherwise. p. 283, #9. We compute the MGF once again: Because X and Y are independent, and since S − X = Y , MS (t) = MX (t)MS−X (t). Now, MS (t) = And MX (t) = 1 1 − 2t ∞ MY (t) = 1 1 − 2t m/2 ∞ if t < 12 , otherwise. It follows that (n+m)/2 if t < 21 , otherwise. 1 1 − 2t nm/2 ∞ if t < 21 , otherwise. 2 This is the MGF of a χ2 (n). p. 283, #10. Let X1 , . . . , X15 be i.i.d. EXP(θ)-distributed random variables. We are asked to find c such that ( 15 ) X 15θ P Xi < = 0.95. c i=1 P15 P15 Because i=1 Xi ∼ GAMMA(θ , 15), we know that 2 i=1 Xi /θ ∼ χ2 (30) [Theorem 8.3.3, page 269]. Therefore, we can write our questions as follows: Find c such that ( 15 ) 2X 30 P Xi < = 0.95. θ i=1 c Table 4 [page 604] tells us that ) ( 15 2X Xi < 43.77 = 0.95. P θ i=1 Therefore, we set (30/c) = 43.77 and solve to obtain c = 30/43.77 ≈ 0.685401. p. 283, #14. We can realize T as Z , T =p V /ν where Z ∼ N (0 , 1) and V ∼ χ2 (ν) are independent. Therefore, T2 = Z2 V /ν has the form {χ21 (1)/1}/{χ22 (ν)/ν} for independent χ2 variables χ21 and χ22 . It follows that T 2 ∼ F1,ν . 3