Research figures

Replotted from vector coordinates in the user-selected source figures. Values are digitized approximations, not original numerical tables. Source URLs, panel selections, data and extrapolation flags are in research-figure-data.json.

• STC: seven selected molecular benchmarks. Data extracted from the last seven rows of Fig. 7, arXiv:2602.17158v1, PDF page 9. The error axis ends at 3 kcal/mol; longer bars have jagged truncation marks and their full values annotated. Labels use N_bf for the number of basis functions. Runtime remains logarithmic; the exact-CCSD(T) series is a timing reference only. Source error values are mean absolute errors over 25 independent STC calculations.
• Quantum: Fig. 3(a), arXiv:2610.02916v1, PDF page 12. The plot retains T-optimal Quantum BSE, Classical BSE, Full Hamiltonian (1st quantized), and Full Hamiltonian (2nd quantized), in that legend order. Fitted lines and individual open/filled marker statuses are preserved. The qubit-optimized BSE series remains in the source data but is not plotted. Classical FLOPs and quantum Toffoli counts are distinct quantities. The panel is for diamond, m=3, 0.1 eV QPE tolerance, 0.1 mEh MP2 fitting error per cell, and no space-group adaptation.
• MOB-ML: Fig. 3, arXiv:2207.08317v1, PDF page 4. The plotted curves are MOB-ML (KA-GPR), OrbNet-Equi, and QML, labeled MOB-ML, OrbNet-Equi, and QML. GMM/GPR remains in the source data but is not plotted. Axis calibration uses the consistent 1, 0.3, 0.1, and 0.03 ticks. The source’s top tick is printed as 3 at the coordinate corresponding to 2; the replot generates correctly positioned logarithmic tick labels. Targets: MP2/cc-pVTZ for MOB-ML versus B3LYP/6-31G(2df,p) for OrbNet-Equi and QML. Feature-generation methods are described in the page caption.

Regenerate SVG figures and PNG previews from the repository root:

python3 scripts/generate_research_figures.py images/research/research-figure-data.json /tmp/research-figures

Dependencies: NumPy and Matplotlib. The script does not require network access. PDF vector extraction was performed using PyMuPDF; extraction code is retained alongside the generation script. For extraction, download the three PDFs as quantum.pdf, mobml.pdf, and stc.pdf next to the extraction script and run it there. The downloaded PDFs and intermediate screenshots are not part of the website.