Convergence not achieved using ml maximize

[Jasmine Xu]

Hi,

I am writing my own maximum likelihood program to estimate parameters in a structural model.
I have three parameters to estimate:

clear
set more off
program drop _all

program define bl
    args logl lamda phi alp
    ...

end

mat start = (0.65,0.38, 1.3)

ml model lf bl /lamda /phi /alp
ml init start, copy
ml max, difficult search(norescale) trace

initial:       log likelihood = -2.7290672
-------------------------------------------------------------------------------------
Iteration 0:
Parameter vector:
        /:     /:     /:
    lamda    phi    alp
r1    .65    .38    1.3

r(430);

I have tried to use different starting values, but it does not help with the convergence.
I also have tried to use different maximization methods such as tech(bfgs 1 nr 5), this does not help either. I guess this is because the iteration stop at the starting values.
The convergence does achieve if I only estimate lamda and phi. And the starting values in the above code of lamda and phi are taken from the model without alpha, while the starting value of alpha is taken from the literature (which varies with papers).

I wonder if there is any other things that I can try to achieve convergence.
Thanks.