1. PF Contest - Win "Conquering the Physics GRE" book! Click Here to Enter
    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!

Algebra: proving an inequality

  1. Jul 14, 2010 #1
    1. The problem statement, all variables and given/known data
    f(x) = (1/2+x) log [(1+x)/x] + (3/2-x) log [(1-x)/(2-x)]
    where log is natural logarithm and 0 < x <= 1/2. Show that f(x) >= 0 for all x.

    2. Relevant equations
    The only inequality that I can think of is the log sum inequality:

    3. The attempt at a solution
    I try to plot f (see figure), and it seems that the statement is correct. In fact, the figure suggests that f is decreasing and equals 0 at x = 1/2. However, I can't figure out an analytical proof of this.

    Attached Files:

    • f_x.jpg
      File size:
      11 KB
  2. jcsd
  3. Jul 14, 2010 #2
    I haven't done this out, but, if you find f(1/2) = 0 and that f(x) is decreasing on (0,1/2], what does that tell you about f(x) on (0,1/2]? How would you show that f(1/2) = 0 and f(x) is decreasing on (0,1/2]?

    EDIT: On second thought, it may not be completely straightforward to show that f(x) is decreasing on (0,1/2]. If so, let us know what you've tried.
  4. Jul 14, 2010 #3
    Certainly if I can show that f'(x) <=0 then the inequality is proved. However, it seems to me that showing f' being negative is even more difficult than showing f being nonnegative directly.

  5. Jul 15, 2010 #4
    It turns out that the function f can be simplified as
    f(x) = d\left(\frac{1+x}{2} \parallel \frac{x}{2}\right) - d\left(\frac{x}{2} \parallel \frac{1+x}{2}\right)
    d(p \parallel q) = p \log \frac{p}{q} + (1 - p) \log \frac{1-p}{1-q}
    is the KL divergence between two Bernoulli random variables with respective expectations p and q. So all I need to do now is to show that the difference of these two KL divergences is nonnegative when x <= 1/2. Any idea how to do that?
Know someone interested in this topic? Share this thread via Reddit, Google+, Twitter, or Facebook