X-ray physics · Inverse problems · Scientific software
M.S. student in Physics, Sogang University
Not measured data: a simulated diffraction intensity reconstructed with QoraPhase. Reproduce the animation.
X-ray measurements record intensities, not structures directly. I develop numerical models and software to extract structural information from diffraction, reflectivity and coherent imaging data.
My work spans inverse problems, time-dependent scattering and beamline data processing, primarily in Python (NumPy, JAX and PyTorch).
- TraTT — transient acoustic diffraction. Simulates Bragg satellites during acoustic-wave propagation using time-dependent Takagi–Taupin equations. Calculated findings · KPS 2026 Fall poster.
- ReflectoLearn — XRR inverse problems. Combines neural-network initialization with physics-based fitting and investigates parameter ambiguity and identifiability. Findings and limits.
- QoraPhase — phase retrieval. A JAX package for iterative reconstruction of objects from diffraction intensities with missing phase information. Synthetic demonstration · Measured-data limitations.
- QoraFlow — XFEL data processing. Groups pump–probe detector frames by laser state and delay, applies corrections and extracts time traces from diffraction peaks. Processing steps · Simulated example.
- QoraPhaseCU explores C++20/CUDA implementations of phase retrieval, alongside the JAX work.
- xray.ooguy provides interactive calculators for X-ray and beamline work.
- qoraplot provides Matplotlib layouts for publication figures; cvdkit combines color-vision-aware palettes with marker and line styles.
- codigest packages selected code, diffs and terminal output into context for coding agents.
- herdr-courier is designed to prevent agent prompts from interrupting human typing.
- hanfix repairs terminal commands entered with the Korean keyboard layout.
Technical notes · Repository index (2026-10-23 snapshot)
Results and limitations are recorded in the evidence and reproduction guide.
Recorded results and limitations
| Work | Recorded result | What this does not prove |
|---|---|---|
| TraTT | Calculated acoustic-satellite overshoot mapped over 11,211 parameter pairs; selected full Takagi–Taupin checks agree with the reduced model. | No comparison to a measured transient dataset. |
| ReflectoLearn | 935 bounded local-fit refinements. Six incorrect parameter regions require at least 10⁴ counts to distinguish from the reference in the studied conditions (5 in test group R2a, 1 in X2e). | Does not search every possible solution or establish accuracy on a measured film. |
| Public XRR counts | 100 repeated public scans: median dispersion index 1.34 versus a Poisson benchmark near 1 (draft; not independently verified). | One experimental setup; the supported verdict requires a post-pre-check amendment. Under the original exclusion rule there is no verdict; the excess-noise source is unknown. |
| QoraPhase | Synthetic images were reconstructed; on two public coherent-diffraction datasets (CXIDB), the published images were not reproduced. | Matching diffraction intensity does not prove the reconstructed image is correct. |
Recent research and upstream contributions
NaNIsaac-Research/ReflectoLearnrefactor: take the forward physics from reflectosim mergedNaNIsaac-Research/TraTTdocs: drop private paths and internal workspace references from tracked text mergedNaNIsaac-Research/TraTTdocs: paper-form summary (background, method, results, limits) mergedNaNIsaac-Research/ReflectoLearndocs(findings): add structured Hypothesis fields for ICC research tab mergedNaNIsaac-Research/ReflectoLearndocs: index research findings and evidence mergedNaNIsaac-Research/TraTTdocs: add a findings page and a Findings section to the README mergedNaNIsaac-Research/ReflectoLearndocs(identifiability): redraw the experiment 001 figure from the saved results mergedNaNIsaac-Research/ReflectoLearndocs: paper draft on identifiability and search reach in XRR fits open
clatlan/cdiutilsFix int32 overflow in SpaceConverter.get_transformation_matrix mergedpydata/xarrayDOC: Link windowed operations tutorial from computation guide merged
Pepper (후추), my black-and-tan Shiba Inu, takes care of the contribution garden.
Previously SJB7777 (renamed October 2026).