Exponential curve fitting

[jactex]

Can anyone help me to use the LINEST() function to fit the curve y = a *(1 - exp(-b*u)) to my dataset?

Would someone be able to explain what is the best way to fit.

Thanks,
Jactex

[MrShorty]

Are you certain that the function can be transformed into a linear function? Assuming a and b are the fitting parameters, I don't immediately see a way to transform that function into a linear function. I think that non-linear regression techniques (using Solver in Excel) may be necessary for this problem.

Can you show us a linear form for that function? Are you familiar with non-linear regression using Solver?

[jactex]

When running the "GRG Nonlinear" method of the Solver Add-In I frequently almost instantly get an error report with message "Solver encountered an error value...". I have tested all boundary conditions and can't find any variable values that produce errors.

[MrShorty]

Then it must have encountered an error value of some kind during the algorithm. The error value need not necessarily occur at the boundary conditions, but may occur elsewhere in the solution space. When it displayed the warning message, did you click on "keep current solution" or "Restore original values"? I would have expected that, if you had picked "keep current solution", the values that triggered the error would be retained, and you could see what error was returned, what values triggered the error, and be able to debug further.

I'm not sure how much we can help on this side of the internet based on the information given. If you can upload a sample file to the forum (go advanced -> manage attachments) that exhibits the error, then we can look at your file and see the Solver model and spreadsheet model that you are using in context.

[jactex]

Thanks for the instructions. Here you go.

[MrShorty]

I don't recreate the error. With no changes to either the spreadsheet or the Solver model, I click on Solve, and it converges to a believable solution (A~-3500, B~-0.39). I tried other starting values for A and B, and it sometimes converged on something unreasonable, but it never gave a "Solver encountered an error." I could get a #DIV/0 error if either A or B are exactly 0, but Solver did not seem to have a tendency to make either parameter exactly 0.

Have you got a different example that illustrates the problem?