curve fitting toolbox for y=y0.exp((x0/x)^v) where data is known, but y0,x0 and v are unknown constants

Hi everyone

I have some scatter data that should take the form of y=y0.exp((x0/x)^v) (where x is the independent variable).

I have tried using the cftool gui but one of 2 things happens; either no model appears on the plot and I get values that can't possibly be right, or I get a totally straight line appearing across the plot.

I should say that y0,x0 and v are unknown constants in a range that I have a rough idea about. I was wondering if a different MATLAB tool would be better for this even. I can't see whats wrong with my approach.

Thanks for your time.

[Image Analyst]

Make it easy for people to help you -- attach your data.

[Em]

Thanks for the advice, done :)

[darova]

See my recommendation

Answer 1

[Alex Sha]

Hi, Em, if fit as the function: y=y0.exp((x0/x)^v), then the result will be about：

Root of Mean Square Error (RMSE): 9.39279339716383E-8
Sum of Squared Residual: 2.20561419504511E-13
Correlation Coef. (R): 0.82258630535955
R-Square: 0.676648229765075

Parameter	Best Estimate
----------	-------------
y0	5.04851528699078E-118
x0	8.3186904534098E-210
v	-0.0113812101308709

While, if change your fit function from " y=y0.exp((x0/x)^v)" to " y=y0.exp(x0^v/x^v)", take x0^v as x1, the function become: y=y0.exp(x1/x^v), the result will be more better and stable as below, the biggest difference is: x0 in original function will never be negative, but x1 in new function can be any value:

Root of Mean Square Error (RMSE): 7.55037962863218E-8
Sum of Squared Residual: 1.42520581341159E-13
Correlation Coef. (R): 0.886423351349966
R-Square: 0.785746357818505

Parameter	Best Estimate
----------	-------------
y0	3.84276923298628E-7
x1	-2.23082444717647E24
v	10.629499809373