MTH366 20130403 week 1 Wednesday April 3, 2013 page 1 Last time… If we know X̲̅ and fX̲̅(x) and Y̲̅ = U(X̲̅ ), sometimes we can find fY̲̅(y). new HW posted, due Wednesday of week 2 quiz this week: read syllabus and email teacher about it: email the components of your grade based on the syllabus. due Friday of week 1 at 5:00 PM PDT fX̲̅(x) is the p.d.f. for continuous variable From the p.d.f. for Y̲̅ we can compute: ∞ the mean μX̅ ̲ =E(Y̲̅ )= ∫ yfY̲̅ (y)dy -∞ the variance 2 2 σ2Y̲̅ = E ((Y̲̅ − μY̲̅ ) ) = E (Y̲̅ ) − μ2Y̲̅ σY̲̅ = √σ2Y̲̅ the standard deviation ∞ since E(U(x))= ∫ U(x)fX̲̅ (x)dx -∞ Summary of what we’ve done so far. Let Y̲̅ be a function of continuous r.v.’s X̲̅1, X̲̅2, …X̲̅n. Then we: 1. Find the region Y̲̅ = y in the (x1, x2, … xn) space. 2. Find the region Y̲̅ ≤ y. 3. Find the distribution FY̲̅(y) = G(y) = P(Y̲̅≤y) by integrating “fX̲̅(x)” over the region Y̲̅≤y. 4. Find the density function fY̲̅(y) = G(y) by differentiating FY̲̅(y) = G(y). If we have h(x,y,z) find ∂ d not ∂z dz Let’s generalize the approach. The method of transformations (change of variables) through the distribution function approach we can arrive at a single method of finding the p.d.f. for Y̲̅ = U(X̲̅ ) provided U(X̲̅ ) is either increasing or decreasing. Increasing? From calculus class we say h(x) is increasing if h’(x)>0. We say h(x) is decreasing if h’(x)<0. Invertible? From MTH111, we had: one to one: one output for each input onto: everything in the range comes from something in the domain horizontal line test for one to one vertical line test for function If y=x 2 then ±√y=x (not technically invertible) Suppose U(X̲̅ ) is increasing and that Y̲̅ = U(X̲̅ ) where X̲̅ has p.d.f. fX̲̅(x). If y = U(x) then U-1(y)=x. Apply U-1 to both sides so U-1(y)=U-1(U(x)) and U-1(y)=x. So in theory FY̲̅(y) = P(Y̲̅≤y) by definition of the distribution function and also = P(U(X̲̅ )). Aside: Start with U(x)≤y so x≤U-1(y) and FY̲̅(y) = P(X̲̅≤U-1(y)). We’re changing an unknown integration into one we have. FY̲̅(y)=FX̲̅ (U-1(y)) where FX̲̅ is the distribution function for X̲̅ . To find the density function (p.d.f.) of Y̲̅ we must differentiate FY̲̅(y). fY̲̅ (y)= d d FY̲̅ (y)= (F (U -1 (y))) dy dy X̲̅ d (U -1 (y)) d U(t) d(eU ) dU d d dx dx U(t) ' (t)= -1 (y))) = (FX̲̅ (U = (FX̲̅ (x)) = fX̲̅ (U -1 (y)) (e )= e U dt dU dt dx dy dx dy dy Note that dx 1 = dy dy ( ) dx ex) 2x 0≤x≤1 fX (x)= { 0 elsewhere Find the p.d.f. for Y̲̅ = 3X̲̅-1 1. What is U? Is it “nice?” U(x) = 3x-1 or y=3x-1 x= 1 (y+1)= U -1 (y) 3 dx d 1 1 = ( y+1) = dy dy 3 3 Then fY̲̅ (y)= fX̲̅ (U -1 (y)) 1 -1 dx = {2U (y) 3 -1≤y≤2 dy 0 elsewhere 2 = {9 (y+1) -1≤y≤2 0 elsewhere We really need U(X̲̅ ) to be invertible if we write things like U-1(Y̲̅) so 1-1 and onto or injective and surjective, or bijective. Then we can do the following: FY̲̅(y) = FX̲̅(U-1(y)) and differentiating fY̲̅ (y)= d d dx (FY̲̅ (y)) = (FX̲̅ (x)) | | dy dx dy