Timeline for FindRoot in FindFit
Current License: CC BY-SA 4.0
13 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| May 11 at 9:40 | vote | accept | mathemania | ||
| May 6 at 19:31 | comment | added | Michael E2 | I wouldn't be concerned about errors for parameter values for a and b that are large negative numbers, and therefore irrelevant with regard to your reasonable range, not to mention that FindFit[] does not return those values for a and b. FWIW, here are two examples giving the same error, one having a root and the other not: FindRoot[CubeRoot[x^2 - 2] == 0, {x, 1}] and FindRoot[CubeRoot[1*^16 x^2 - 2*^16]^2 + 10 == 0, {x, 1}]. You might examine your equation and see if something similar is going on, if you are interested in parameter values like {a -> -76.9678, b -> -257.561}. | |
| May 6 at 17:14 | answer | added | Daniel Huber | timeline score: 4 | |
| May 6 at 16:03 | comment | added | mathemania | @MichaelE2 On the other hand, it still does not explain why FindRoot is giving errors. It is complaining about the tolerance specified by AccuracyGoal and PrecisionGoal. | |
| May 6 at 16:02 | comment | added | mathemania | @MichaelE2 In my experimental setup, there is no universal value for $\{a,b\}$, although they should be both non-negative. In addition, they cannot be both too big. Let's just say roughly that a reasonable range is $0 \leq a \leq 5$ and $0 \leq b \leq 15$. In addition, please see my update. I think one cause for the large deviation of the fit is my choice of points when calculating the slope. I think it is better to choose points 5 & 7 instead of 6 & 7. | |
| May 6 at 15:59 | history | edited | mathemania | CC BY-SA 4.0 | Added some information |
| May 6 at 14:40 | comment | added | Michael E2 | I wonder what the range for a and b should be, since this does not come near to bracketing the data near T=9: Table[fitfunc[9, a, b], {a, -50, 1050, 50}, {b, 1, 2021, 200}] // MinMax | |
| May 6 at 14:38 | comment | added | Michael E2 | The three-dot button will show the stack trace. This reveals the values of a and b being tested. Are they reasonable, or should you restrict the search range in FindFit? | |
| May 6 at 14:20 | comment | added | mathemania | @Bill Good point about the code depending strongly on the slope! As for the fitting function, that is the required function to be used. On the other hand, I just wanted to know why the errors are coming out, and whether there are fixes needed to be done. Or the errors are just complaints by MMA such that even if the errors are fixed it will give almost the same parameters? | |
| May 6 at 13:54 | comment | added | mathemania | @SjoerdSmit I have updated the post including the fitting curve. As you see, the fitting curve only covered two points. I'm not saying this is wrong or correct. It's just that I received errors when calculating the fitting parameters, so I'm unsure whether the result is correct or not. | |
| May 6 at 13:46 | history | edited | mathemania | CC BY-SA 4.0 | Added small information |
| May 6 at 13:30 | comment | added | Sjoerd Smit | Why not try plotting the fit result in the same plot as the data to see if the fit is good? | |
| May 6 at 13:24 | history | asked | mathemania | CC BY-SA 4.0 |