Simplify cold cavity RK4 refinement. The simplification may impact the precision of cold cavity solution.
Fix a bug in cr_find_cold_resonance() introduced from v1.8.0. Sanitizer could not find this bug, as all memories are valid.
Strength cold cavity functions by adding more asserts, which leads to an enhancement of the internal function make_omega_range().
v1.9.0 2025-11-14
Breaking change: Separate center radius and span of the bcc grid out of beam_init array, in order to support variable initial radius between iterations.
Fix a bug in passing span of bcc into cr_shoot_blt(), which was mistakenly treated as single-side spread instead of double-side span. The previous guiding centers were "squeezed", i.e. behaved more like classical codes without calibration of guiding center. After this bug fix, the error between Cartesian pusher and cyclotron pusher are further reduced. However, the overall impact on the typical simulation scenarios is tiny.
Fix an issue of invalid resonances with negative frequencies in the cr_find_all_cold_resonances(). (Negative frequencies were used to mark discarded data, however, they were not deleted as expected.)
Improve angle wrapping: faster and more precise. The improvement of precision can be observed e.g., in the electron position by using 32-bit floating point type on the duality (self-referred) test.
Improve accuracy of field solution as a compile-time option. Improving accuracy costs typically 8× time (at least 2×, max 128×), depending on the convergence.
Fix a bug introduced by v1.7.4 when minimum harmonic is not zero.
v1.7.4 2025-08-30
Perf: Improve performance of Lorentz force evaluation up to 80% in multimode simulations by transforming the code to assist the vectorizing or eliminate some elementary math functions.
Build: Remove CUDA device-link and device-LTO for performance.
Build: Overhaul the build system for preparing OpenCL-backend GPU build by chipStar.
v1.7.3 2025-08-16
CROCK: Clean the API macros without affecting the API functions. The CROCK library remains binary compatible.
Performance tuning
v1.7.2 2025-08-12
Remove a dummy memory barrier for synchronizing CPU caches.
On a Sandy-Bridge desktop CPU, adding the barrier improves performance by 2%.
On a modern heterogeneous CPU with P/E-cores, skipping the barrier improves performance by 2× on Linux, while nearly 20x on Windows.
v1.7.1 (Hotfix)
Fix a bug due to incorrect sorting of cold resonances
v1.7.0 2025-08-11
Use the concept span rather than spread, see the updated API documentation.
Previously only alpha was in Gaussian distribution, since this version both alpha and gamma are in Gaussian distribution.
build: EL8 docker build switch from conda to uv (license issue)
build: Support numpy>=2.0
v1.6.3 2025-04-22
Fix cr_free() inconsistency among compilers on Windows
v1.6.2 2025-04-21 (Easter)
Avoid NaN in mvn_src at zero'th harmonic interaction introduced in v1.6.1
Fix a compatibility issue of Cython function return type
Free caches for sanitizer check
Add another test for second harmonic interaction
Improve the organization of installed header files
v1.6.1 2025-01-20
Add a high-order perturbation to the calculation of mvn_src.
Generate documentations of the data rows from the deployed rows.h.
v1.6.0 2024-12-17
Correct the equation for calculating the excitation (mvn_src). This correction should fix the magnitude of mvn_src in harmonic oscillation. However, since this formulation only considers the first low-order term in the expansion of eigenvector derivatives, the harmonics below the true harmonic number in the calculation of mvn_src may cause instability, therefore, they are omitted. If instability still appears, try to enable ROCK_MASK_BLT_SRC_MONOHARM.
Improve cold resonance calculation of cavities with up-taper. The position of cavity center can be optionally specified in cr_search_cold_resonance().
v1.5.0 2024-11-05
Support Python 3.13
Introduce option ROCK_MASK_BLT_SRC_MONOHARM, when enabled only a single harmonic contribution is taken into account. This is less accurate, but does not introduce other aliased frequencies. Note that the Lorentz force still includes all harmonics.
Introduce cr_make_df() and deprecate the delta-frequency calculation in cr_step_field(). Instead of calculating frequency shift in a single cr_step_field() step, the average (but aliased) frequency shift, typically after every 100 steps, should be calculated. This avoids the drifting of carrier frequency by correlated noises.
Introduce cr_update_bc_simple() for a fake broadband Boundary Condition (BC): the matched frequency of a narrowband simple BC can be updated so often that the current transient field frequency of each mode is matched in the next step. If doing this with sufficient temporal resolution, the bandwidth of BC is effectively increased.
Cold resonance: Stay with the first axial mode when the cold cavity resonance spectrum is dense.
v1.4.0 2024-09-26
Support longitudinally corrugated hollow waveguides, however, only as impedance model.
Disable the fast atan2() variant, since it causes wrong azimuthal bunching of guiding centers.
Add beamlet options:
MASK_BLT_CALI_CYC_PHASE: Consider calibration of cyclotron phase, reduce performance due to atan2().
MASK_BLT_GC_AZI_RF : Consider azimuthal motion (drift, oscillation etc.) of guiding center caused by RF field, reduce performance due to atan2().
MASK_BLT_ALL_FEATURES: All features except DEBUG. Slowest option.
v1.1.1 2024-08-16
Hotfix: Fix type of arguments in cr_numa_bless() in tests and PROCK. Depending on ABI, some systems may not be affected. (x64 Linux is fine, but x64 Windows ABI encounters problem)
v1.1.0 2024-08-15
Breaking change: Remove the deprecated mem_pool, use cr_cache instead.
Breaking change: The order vnb is changed to bvn. (e.g. in the array of electron position)
Significantly increase the OpenMP scalability
Reduce beamlet corrector step to 2.
Fix potential bug/mistake of data race
Warn if mesh discretization bigger than 4 times Larmor radius.
Deprecate GPU-offloading by OpenMP and OpenACC pragmas. With pragmas, the code is unclear covering all cases, while neither CPU nor GPU have the best performance.
v1.0.1 2024-08-02
Switch build system due to the difficulty in maintaining combinations of PIC+LTO+Offload+OffloadLib+OffloadLTO+StaticLib+__builtin_math portable among compilers.
v1.0.0 2024-07-19
Take into account multiple harmonics of the same mode
1.15.0