differintP 0.0.3
0.0.3 (7/4/2025)
- Redesigned
GLpoint:
Replaced the previous GLpoint implementation with a highly efficient, single-pass C++-style recurrence. The new method uses a Numba JIT-compiled kernel for substantial speed gains, directly computing the Grünwald-Letnikov fractional derivative at the endpoint in one loop over the function values.
Legacy and alternative endpoint methods (GLpoint_direct,GLpoint_via_GL) have been moved tospecial.pyfor reference and comparison.
Benchmark Result for GLpoint:
| count | differintP | differint | Speedup Factor |
|---|---|---|---|
| 1+e2 | 0.0148 ms | 0.0416 ms | x2.81 |
| 1+e3 | 0.0276 ms | 0.4213 ms | x15.26 |
| 1+e4 | 0.0397 ms | 3.8376 ms | x96.66 |
| 1+e5 | 0.3285 ms | 36.303 ms | x110.5 |
| 1+e6 | 5.267 ms | 385.9205 ms | x73.27 |
| 1+e7 | 51.2645 ms | 3674.3541 ms | x71.67 |
-
Optimized
GLIimplementation:- Rewrote core algorithm to use a Numba-accelerated helper for the main rolling-window convolution, matching the original literature.
- Ensured correctness by flipping GL coefficients within the moving-window convolution.
- Replaced class-based coefficient calculation with direct variable computation for improved clarity and efficiency.
- All critical loops are now compiled with Numba for major speed improvements, especially at high grid resolutions.
- Modernized function interface and added robust array conversion and shape checks.
Benchmark Result for GLI:
| count | differintP | differint | Speedup Factor |
|---|---|---|---|
| 1+e2 | 0.0760 ms | 0.7582 ms | x6.06 |
| 1+e3 | 2.0336 ms | 7.391 ms | x3.63 |
| 1+e4 | 155.7649 ms | 128.3709 ms | x0.8 |
| 5+e4 | 3895.176 ms | 15181.82 ms | x3.89 |
-
Wiki Pages for
CaputoL1pointCaputoL2pointCaputoL2CpointCaputoFromRLpointCRONEGLGLpointGLIGL_gpuRLRLpointPCsolverfunctions Namespaces
-
Aditional test and test restructuring
-
Restructuring the Namespaces