Problem with non-linear fitting models and large numbers

I'm trying to fit some data and having trouble with a general model in the curve fitting toolbox. If I give it data with numbers greater than a million it doesn't even attempt to fit the data and says that the fit converged with 95% confidence between -Inf and Inf. Not very useful.

If I have data = 10, this works. If I have data = 1e9 it completely doesn't. If I have data = 1000, it gives me ~600 though I suspect it would get there eventually if I increased the number of iterations allowed. Interestingly with 1e9 it doesn't even do a single iteration.

This ('a+0*x') is a 'very' simplified version of the function I need to use. I know you would normally use a linear model for fitting a constant (which works fine) but this illustrates the problem. It doesn't work any better if I use a+b*x as the model. I've tried increasing DiffMaxChange with no effect.

Anyone have any ideas? Thanks.

x=(1:10)'; y=10*ones(10,1);

f = fittype('a+0*x'); [c, gof, output] = fit(x, y, f, 'Startpoint', [0], 'MaxIter', 600, 'MaxFunEvals', 2000);

c output

--- Result -------------------------- c =

     General model:
     c(x) = a+0*x
     Coefficients (with 95% confidence bounds):
       a =          10  (10, 10)

output =

           numobs: 10
         numparam: 1
        residuals: [10x1 double]
         Jacobian: [10x1 double]
         exitflag: 1
    firstorderopt: 0
       iterations: 1
        funcCount: 4
     cgiterations: 0
        algorithm: 'trust-region-reflective'
          message: 'Success. Fitting converged to a solution.'

---------------------------------------------

x=(1:10)'; y=1e9*ones(10,1);

f = fittype('a+0*x'); [c, gof, output] = fit(x, y, f, 'Startpoint', [0], 'MaxIter', 600, 'MaxFunEvals', 2000);

c output

--- Result -------------------------- c =

     General model:
     c(x) = a+0*x
     Coefficients (with 95% confidence bounds):
       a =           0  (-Inf, Inf)

output =

           numobs: 10
         numparam: 1
        residuals: [10x1 double]
         Jacobian: [10x1 double]
         exitflag: 1
    firstorderopt: 0
       iterations: 0
        funcCount: 2
     cgiterations: 0
        algorithm: 'trust-region-reflective'
          message: 'Success. Fitting converged to a solution.'