1. Limited time only! Sign up for a free 30min personal tutor trial with Chegg Tutors
    Dismiss Notice
Dismiss Notice
Join Physics Forums Today!
The friendliest, high quality science and math community on the planet! Everyone who loves science is here!

Homework Help: Convex functions

  1. Nov 6, 2009 #1
    1. The problem statement, all variables and given/known data
    How do I show that if [tex]f\in C^2 \text{(}\mathbb{R}\text{)}[/tex] is convex then the function [tex]yf(y^{-1}\textbf{x})[/tex] is convex on (x,y):y>0?

    2. Relevant equations

    I know the standard definitions and whatnot about convexity, but I tried chugging through the algebra and didn't have any luck, can anyone show me a nice way to solve this?


  2. jcsd
  3. Nov 6, 2009 #2


    Staff: Mentor

    Show us what you tried...
  4. Nov 7, 2009 #3
    I figured [tex]yf(y^{-1}\textbf{x})=yf(\frac{x}{y},1)[/tex], so set [tex]f_x=\frac{\partial}{\partial{x}}f(\frac{x}{y},1), f_{yy}=\frac{\partial^2}{\partial{y^2}}f(\frac{x}{y},1)[/tex] and so on:

    then we get

    [tex]\frac{\partial{}}{\partial{x}}(yf(\frac{x}{y},1))=yf_x \Rightarrow \frac{\partial{}^2}{\partial{x^2}}=yf_{xx}[/tex]


    [tex]\frac{\partial{}^2}{\partial{x}\partial{y}} (yf(\frac{x}{y},1))=f_x+yf_{xy}[/tex]

    Also, W.R.T y:

    [tex] \frac{\partial{}}{\partial{y}}=yf_y+f(\frac{x}{y},1) \Rightarrow \frac{\partial{}^2}{\partial{y}^2} = yf_{yy}+2f_y[/tex] - right?

    Then looking at the Hessian, we know the first principal minor (=yfxx) is >=0 if y is, because f is convex so its corresponding first principal minor must also be >=0. With regards to the second principal minor though, i.e. the determinant of the Hessian, we get

    [tex] \frac{\partial{}^2}{\partial{x}^2} \frac{\partial{}^2}{\partial{y}^2} - (\frac{\partial{}^2}{\partial{x}\partial{y}})^2=y^2(f_{xx}f_{yy}-f_{xy}^2)+2y(f_{xx}f_y-f_{xy}f_x)-f_x^2[/tex]

    if my algebra is correct. We want to show this >=0 - the first term (the thing in the brackets multiplied by y2) is >=0 because it corresponds to the determinant of the Hessian for f - however, my concern is that I've gone wrong because something in the derivative of f(x/y,1) means this wouldn't work the same as for f(x,y) and so I'm not sure how to proceed... thanks :)
Share this great discussion with others via Reddit, Google+, Twitter, or Facebook