Jump to content


Condition number

  • Please log in to reply
2 replies to this topic

#1 Shu-Pei Wu

Shu-Pei Wu


  • Members
  • PipPip
  • 26 posts

Posted 06 August 2014 - 07:58 PM

Hi Pharsight Modellers,


I have question regarding to the calculation of condition number in phoenix.

In the user guide, it states that when the condition number exceeds 1E+6, then the model is probably overparameterized.


However, I recalled that when using NONMEM, the criteria is <1000, (it happens to be square root of 1E+6).


I wonder how the difference just because the calculation?




#2 Simon Davis

Simon Davis

    Advanced Member

  • Administrators
  • 1,178 posts

Posted 07 August 2014 - 11:40 AM

Shu-Pei, as I understand the condition number is a property of the actual problem so as a general rule of thumb, if the condition number is less than = 10^number of parameters then it is probably well-conditioned.


  However I've seen it exceed this and still been a good fit by other metrics so I wouldn't look at it in isolation.



#3 Shu-Pei Wu

Shu-Pei Wu


  • Members
  • PipPip
  • 26 posts

Posted 07 August 2014 - 02:54 PM

Hi, Simon,


Thanks for your response for sharing the general rule of thumb of looking at condition number.

I also agree that it is just one of many diagnostic indicators for accessing the model.\


Thanks again,




0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users