Re: [Phys-l] quadratic uncertainty

[John Denker <jsd@av8n.com>]

On 08/26/2010 04:22 AM, Brian Blais wrote:

> In my little error propagation (monte carlo) routines, I get this  
> nice little bit:
>
> a=Quantity('1 +- .0001')
> b=Quantity('-2.08 +- .01')
> c=Quantity('1.08 +- .01')
>
>
> In [3]:evaluate('b**2-4*a*c')
> Out[3]:0.00699313256158 +- 0.0579303054605
>
> which gives rise to complex solutions if I try:
>
> In [4]:evaluate('-b+sqrt(b**2-4*a*c)/(2*a)')
> Out[4]:nan +- nan
>
> Not sure what the best way to handle that is!

If the tool cannot directly represent complex numbers,
the trick is to ask it to solve for Re(x) and Im(x) 
separately.

This trick requires two or three times more work than 
would be required if the tool supported complex numbers
(or if the task didn't require complex numbers).  A
factor of two or three is annoying but not fatal.