+ Reply to Thread
Results 1 to 4 of 4

Thread: Discriminant Rule Proof

  1. #1

    Discriminant Rule Proof



    My HW question is:

    Show that...

    -\frac{1}{2}(x-\mu_1)^{T}\Sigma^{-1}(x-\mu_1)+\frac{1}{2}(x-\mu_2)^{T}\Sigma^{-1}(x-\mu_2)

    is equal to...

    (\mu_1 - \mu_2)^{T}\Sigma^{-1}x-\frac{1}{2}(\mu_1-\mu_2)^{T}\Sigma^{-1}(\mu_1+\mu_2)

    My efforts:

    Okay, so I started by expanding the first equation to this:

    -x^{2}2\Sigma^{-1}+x\mu_12\Sigma^{-1}+x\mu_12\Sigma^{-1}-\mu_1^{2}2\Sigma^{-1}+x^{2}2\Sigma^{-1}-x\mu_22\Sigma^{-1}-x\mu_22\Sigma^{-1}+\mu_2^{2}2\Sigma^{-1}

    After combining terms, I got this:

    2(x\mu_1)2\Sigma^{-1}-\mu_1^{2}2\Sigma^{-1}-2(x\mu_2)2\Sigma^{-1}+\mu_2^{2}2\Sigma^{-1}

    Then, I multiplied the entire equation by \frac{1}{2}:

    \mu_1\Sigma^{-1}x-\mu_2\Sigma^{-1}x-\frac{1}{2}\mu_1^{2}\Sigma^{-1}+\frac{1}{2}\mu_2^{2}\Sigma^{-1}

    From there, I've been unable to make any more forward progress. I'm pretty sure the first two terms can be rewritten like...

    (\mu_1-\mu_2)^{T}\Sigma^{-1}x

    But I'm not sure about how the second two terms could be factorized. I can only assume that I've made an error in the expansion and simplification steps, but I've been unable the locate the error in my work (if one exists).

    Please help!

  2. #2
    TS Contributor
    Points: 14,811, Level: 78
    Level completed: 91%, Points required for next Level: 39

    Posts
    2,295
    Thanks
    8
    Thanked 371 Times in 351 Posts

    Re: Discriminant Rule Proof

    When you expand the quadratic form, you cannot write something like x^2.

    Here x, \mu_1, \mu_2 are vectors, not a square matrix.

    So you should get something like

    \frac {1} {2} (-x^T\Sigma^{-1}x + ...)

  3. #3
    Points: 1,425, Level: 21
    Level completed: 25%, Points required for next Level: 75

    Posts
    7
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Discriminant Rule Proof

    I've been able to reduce the initial expression to this, after following BGM's advice:

    \frac{1}{2}(\mu_1-\mu_2)\Sigma^{-1}x^{T}+\frac{1}{2}(\mu_1-\mu_2)^{T}\Sigma^{-1}x+\frac{1}{2}(\mu_1+\mu_2)^{T}\Sigma^{-1}(\mu_1+\mu_2)

    I am unsure of how to eliminate or combine the first two terms such that they are equal to the equation in the original post, though. Any suggestions would be welcome.

  4. #4
    TS Contributor
    Points: 14,811, Level: 78
    Level completed: 91%, Points required for next Level: 39

    Posts
    2,295
    Thanks
    8
    Thanked 371 Times in 351 Posts

    Re: Discriminant Rule Proof


    The first two terms are equal,
    because this quadratic term is just a real number (or a 1 by 1 matrix if you like)

    So you can always take transpose on it.

+ Reply to Thread

Similar Threads

  1. discriminant functions analysis
    By Poppy in forum Statistics
    Replies: 1
    Last Post: 03-17-2011, 04:29 AM
  2. Discriminant Analysis Issue
    By cinta19 in forum Regression Analysis
    Replies: 0
    Last Post: 02-27-2011, 06:26 PM
  3. [SYSTAT] - help - discriminant analysis
    By krisfire in forum Other Software
    Replies: 4
    Last Post: 01-27-2011, 05:56 AM
  4. Need help with the discriminant analysis in SPSS
    By neonlight in forum Statistics
    Replies: 0
    Last Post: 05-20-2009, 05:56 AM
  5. Discriminant function analysis
    By jacks in forum Statistics
    Replies: 1
    Last Post: 10-03-2008, 07:08 AM

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts








Advertise on Talk Stats