| Author |
 Topic  |
|
|
binowizu
16 Posts |
 Posted - 01/26/2026 : 1:16:54 PM
|
Hi everyone,
I am using the NLFit tool in OriginPro and I would like to better understand the underlying mechanics of the Levenberg-Marquardt algorithm implementation, specifically regarding the Jacobian matrix.
I have two questions concerning the calculation of derivatives:
- Built-in Functions: How does Origin handle the Jacobian for its library of built-in functions? Does it use analytical Jacobians (exact formulas) or is it based on a numerical approximation approach?
- User-Defined Functions (UDF): When a user creates a new fitting function, how is the Jacobian determined by the engine? I am interested to know if there is a default method (such as finite differences) and if there are options for the user to specify or influence how these derivatives are calculated to ensure better convergence. |
Edited by - binowizu on 01/26/2026 1:23:52 PM |
|
|
Sam Fang
294 Posts |
Posted - 01/27/2026 : 12:27:11 AM
|
Hi binowizu,
1) For most built-in functions, Origin provides analytical Jacobians. For a few complex built-in functions, it uses a numerical approximation method.
2) For user-defined functions, if you want to provide analytical Jacobians, in Fitting Function Organizer dialog, you can choose Origin C Function Type, check Derivatives check box, in Function edit box, click Edit button on the right to open Code Builder, it will have more variables to define, e.g. dy_A0, dy_xc,... where dy_A0, dy_xc, ... are derivatives of y with respect to parameters A0 and xc. If you don't check Derivatives check box, Origin will use finite differences to calculate derivatives.
Thanks.
Sam
Sam
OriginLab Technical Services |
Edited by - Sam Fang on 01/27/2026 12:30:00 AM |
 |
|
| |
Topic |
|
|
|
NLFit algorithm