Helmut Jarausch
1/31/2008 3:52:00 PM
Mikael Olofsson wrote:
> Helmut Jarausch wrote:
>> Your model is A*sin(omega*t+alpha) where A and alpha are sought.
>> Let T=(t_1,...,t_N)' and Y=(y_1,..,y_N)' your measurements (t_i,y_i)
>> ( ' denotes transposition )
>>
>> First, A*sin(omega*t+alpha) =
>> A*cos(alpha)*sin(omega*t) + A*sin(alpha)*cos(omega*t) =
>> B*sin(omega*t) + D*cos(omega*t)
>>
>> by setting B=A*cos(alpha) and D=A*sin(alpha)
>>
>> Once, you have B and D, tan(alpha)= D/B A=sqrt(B^2+D^2)
>
> This is all very true, but the equation tan(alpha)=D/B may fool you.
You're right: standard Python's math library missing the function arctan2.
But you can get it from numpy or numarray.
In case of doubt just compute the residuum with different possibilities .
> This may lead you to believe that alpha=arctan(D/B) is a solution, which
> is not always the case. The point (B,D) may be in any of the four
> quadrants of the plane. Assuming B!=0, the solutions to this equation
> fall into the two classes
>
> alpha = arctan(D/B) + 2*k*pi
>
> and
>
> alpha = arctan(D/B) + (2*k+1)*pi,
>
> where k is an integer. The sign of B tells you which class gives you the
> solution. If B is positive, the solutions are those in the first class.
> If B is negative, the solutions are instead those in the second class.
> Whithin the correct class, you may of course choose any alternative.
>
> Then we have the case B=0. Then the sign of D determines alpha. If D is
> positive, we have alpha=pi/2, and if D is negative, we have alpha=-pi/2.
>
> Last if both B and D are zero, any alpha will do.
>
> /MiO
--
Helmut Jarausch
Lehrstuhl fuer Numerische Mathematik
RWTH - Aachen University
D 52056 Aachen, Germany