specific model and fit

[pmartin-upd7]

Hi
I would like to use a specific function to fit my data.
This function is coming from the Simmons tunnel transport equation.
I tried to use lmfit to fit my data and to obtain values of barrier height (phi) and thickness (d)
I tried several ways to fit but never work
I defined function as follow:

def phi(x, phio, gamma):
    return -gamma*1.602e-19*x + phio

def Simmons (x, d, phio, gamma):
    
    I = (1.602e-19/(2*3.14159*6.626e-34*d**2))*(
        (
            (phi(x, phio, gamma) - 1.602e-19*x/2)*
            np.exp(-(4*3.14159*d/6.626e-34)*
                   np.sqrt(2*9.109e-31*(phi(x, phio, gamma) - 1.602e-19*x/2))
                   )
            )-
        (
            (phi(x, phio, gamma) + 1.602e-19*x/2)*
            np.exp(-(4*3.14159*d/6.626e-34)*
                   np.sqrt(2*9.109e-31*(phi(x, phio, gamma) + 1.602e-19*x/2))
                   )
            )
        )

    return I

def residual(pars, x, original):
    model = Simmons(x, pars['d'], pars['phio'], pars['gamma'])
    return model - original

I extract my data from a txt file

x = chunk[:,0] y = chunk[:,1]/400e-12

then define parameters and fit
params = lmfit.Parameters() params.add_many(('d', 20e-10, True, 5e-10, 50e-10, None, None), ('phio', 1.5e-10, True, 0.1e-10, 4e-10, None, None), ('gamma', 0.2, True, 0.1, 0.4, None, None) ) gmodel = Model(Simmons) result = gmodel.fit(data=y, **params, x=x, nan_policy='omit') #dely = result.eval_uncertainty(x=x, sigma = 3)

at the beginning I got a NaN error thus I added omit param
then I tried to change the model also but still same error
`
File C:\ProgramData\miniconda3\Lib\site-packages\scipy\optimize_minpack_py.py:430, in leastsq(func, x0, args, Dfun, full_output, col_deriv, ftol, xtol, gtol, maxfev, epsfcn, factor, diag)
427 m = shape[0]
429 if n > m:
--> 430 raise TypeError(f"Improper input: func input vector length N={n} must"
431 f" not exceed func output vector length M={m}")
433 if epsfcn is None:
434 epsfcn = finfo(dtype).eps

TypeError: Improper input: func input vector length N=3 must not exceed func output vector length M=0`

if so have an idea
thanks

[pmartin-upd7]

finally I found a solution

[newville]

@pmartin-upd7
a) omit_nans really only solves NaNs in the data. If your function generates a nan (or inf) you have to catch and fix that yourself. You function includes np.exp and has values that appear to have a wide range of magnitude. Hint: try print(np.exp(710)). We can not solve that for you, your model function must handle that.

b) If you mean π, do not use 3.14159. Use np.pi. Yes, the numerical difference is small (well 1.e-6). You are writing Python code to communicate with other people. Similarly, there are common numerical constants in scipy.constants. Use them.

c) however, and "also" to both a) and b): when fitting data, scale and order of magnitude matter, and can matter significantly. 64-bit floats only tolerate difference in scale of about 2e-16. Variables used in any fitting method should be on the order of 1e-5 to 1e5 or so. if you have variables that you expect to be on the order of 1.e-10, change units. This is especially true if some variables are at the other extreme of magnitude.