I'm studying a decay process as shown for a number of samples below. The higher the starting point, the faster the decay. The idea is to predict the value at any timeopint from measured Initial value. I would like to 'lock' the model so that at timepoint zero, the predicted value is exactly the measured Initial value. If not, predictions as short time times can be very off for some samples, as in graph below.
Any help on how to do this is very welcome!
Hi @Ake ,
What platform are you using to fit the data? There should be a way to constrain the y-intercept of your fit to the first data point of the curve, but this might depend on which platform you're using.
DS
Hi and thanks for the fast response! I will practice putting my questions more clearly! 
I'm using Fit Model (jmp 16), putting just Initial Value and Time as factors into a response surface.
I would want only one predictor function for all samples, with only the initial value as input. My (little) under standing is that Nonlinear fit gives you one predictor function per sample?
Best, Åke
UPDATE:
Hi SDF1,
You are right, I try to predict the value V at any time t from just the initial value of V. It seems your suggestion requires me to know the constants for each sample separately. I need to assume all samples have a common decay curvature, so that I only need to know initV to do the prediction for every new sample. I'm no completely sure the curvature is completely exponential, so fit model and transformations seems an ideal tool to find the best curvature (the measurement is physical, so no best theoretical curvature), except that I don't know how to get rid of the difference between V and Pred V at time zero.
I included my dataset, with a prediction Pred V (fit model) from using fit model, and a prediction formula Prev V (solver) constructed by hand and constants found by using solver in Excel.
The latter fulfills my desire to give V = initV at time zero, but using Excel feels like a step backward. I am hoping to do something similar or better in jmp.
