Cache the Ramanathan PMF and vectorize its temperature interpolation

[roytsmart]

Summary

Follow-up to the profiling in #196: probability_of_n_pairs took ~1.6 s per call, a fixed overhead paid by every electrons_measured invocation regardless of photon count. For photon-light calls (unit tests, small ROIs, single-wavelength VMR checks) this intercept exceeded the entire Monte Carlo — a zero-photon electrons_measured call took 1.71 s. Together with #196 (which attacks the per-electron slope), this removes the intercept, and directly speeds up the #195 test loop.

Where the 1.6 s went

step	cost
parsing the three .dat tables, every call	0.25 s
na.interp over the temperature axis	1.35 s
na.interp over the energy axis	0.002 s

The temperature interpolation is slow because na.interp loops over the ~20,000 elements of the non-interpolated (wavelength, num_electron) axes in Python — to interpolate at a single temperature.

Changes

• _probability_of_n_pairs_ramanathan() is now functools.cached. It takes no arguments and returns constant tabulated data, so the parse happens once per process. quantum_yield_ideal and fano_factor call it too and benefit equally. The docstring notes the returned data is shared and read-only.
• The temperature interpolation is a vectorized gather-and-blend: locate the bracketing samples with a broadcast comparison, gather both by named-array indexing, blend linearly, clamping outside the tabulated range exactly as np.interp does. The tabulated axis is renamed internally (_temperature_interp) so a temperature argument carrying an axis literally named temperature still works, as it did with na.interp.
• The (cheap) energy interpolation is left as na.interp.

No public API changes, no new files.

Results

• probability_of_n_pairs: 1.60 s → 2.6 ms warm (~600×)
• zero-photon electrons_measured: 1.71 s → 0.11 s
• quantum_yield_ideal + fano_factor (cold interp, warm cache): 90 ms

Validation

• New implementation matches the previous na.interp chain to 6×10⁻¹⁷ (one ulp) across scalar and array wavelengths and temperatures, including values outside the tabulated range and an argument axis named temperature.
• New test_probability_of_n_pairs (8 parametrized cases) pins the result to an na.interp reference and checks the PMF sums to 1.
• All 46 tests in _ramanathan_2020_test.py and the 48 downstream _materials_test.py tests touching signal/electrons_measured/uncertainty/quantum_yield/fano pass locally.

Merges independently of #196 (non-overlapping regions of the same files); whichever lands second rebases trivially.

The residual 0.11 s intercept is absorption_effective/multilayer_efficiency on the _materials.py side — left alone here, but worth a look if the fixed cost still matters after this.

🤖 Generated with Claude Code

https://claude.ai/code/session_011q4461XE8hCcZCKsViMQC1

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.57%. Comparing base (64e05e6) to head (4fa15fa).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #197   +/-   ##
=======================================
  Coverage   99.57%   99.57%           
=======================================
  Files         118      118           
  Lines        6829     6852   +23     
=======================================
+ Hits         6800     6823   +23     
  Misses         29       29           

Flag	Coverage Δ
unittests	99.57% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.