Calibrating a gamma-ray detector with natural rock standards
A 2025 study calibrates a germanium detector with cheap reference powders of potassium, uranium and thorium instead of a costly mixed source. We refit its efficiency curve and check the source's decay with fibub's tools.
Labs that measure radioactivity in soil, building stone or food use germanium detectors, and every detector has to be calibrated: for each gamma-ray energy, what share of the gamma rays leaving the sample does it actually count? The usual answer is a certified mixed source with ten or so man-made nuclides, which is expensive and decays within a year or two. A 2025 study shows that three cheap, long-lived reference powders of potassium, uranium and thorium can do most of the job. This article refits the paper’s efficiency curve with the curve fitter and checks its reference source with the decay calculator; for counting uncertainties see the counting statistics helper tool.
The paper in brief
A high-purity germanium (HPGe) detector turns each gamma ray it fully absorbs into a sharp peak in a spectrum. The efficiency at an energy is the number of counts in that peak divided by the number of gamma rays of that energy the sample gave off. It falls with energy, because high-energy gamma rays pass straight through the crystal more often.
The study filled the same 121 cm³ aluminium can with three IAEA reference materials: RGK-1 (potassium, for the 1461 keV line of potassium-40), RGU-1 (uranium ore, with lines of lead-214, bismuth-214 and protactinium-234m) and RGTh-1 (thorium ore, for lead-212). The trick is in the choice of lines. Many nuclides send out two gamma rays at once, and when both hit the detector they add up into one wrong peak. This is called true coincidence summing. The author picked six lines, from 239 to 2204 keV, that it barely affects, so no correction is needed. A curve through those six points gave efficiencies within about 4% of a commercial multinuclide source for every energy above 200 keV. That range covers most routine environmental work.
Checked with the tools
We took the paper’s six efficiencies (Table 4) and the multinuclide source’s numbers (Tables 2, 5 and 6) and ran them through the tools’ maths.
Step 1: refit the efficiency curve. The paper fits a parabola to ln(efficiency) against ln(energy). The curve fitter‘s Quadratic model on those logarithms gives ln ε = 5.2464 − 0.59875 ln E − 0.01346 (ln E)², with ε in % and E in keV, and R² = 0.9996. The six points sit within 1.9% of the curve. Our leftovers, −1.1, +1.8, −0.9, −1.3, +1.5 and −0.1%, match the residuals the paper prints, to rounding. So we have the same curve the author used.
Step 2: compare with the multinuclide source. Reading our curve at the six check energies, from 392 keV (tin-113) to 1333 keV (cobalt-60), and comparing with the efficiencies the paper measured with the commercial source gives differences of −4.1, −0.9, +0.0, +1.1, +1.2 and +1.5%. The paper’s Table 6 has −4.1, −0.9, 0.0, 1.0, 1.2 and 1.4%. The average size of the difference is 1.46%, against the paper’s 1.5%. The worst point, at 392 keV, is the one nearest the low-energy end, where the curve rests on only two reference lines.
Step 3: how old was the reference source? The paper gives the multinuclide source’s activities on its certificate date and on the measurement date, but not the dates. The Decay calculator’s “when will it fall to” maths, run for each nuclide, answers the question four times over: tin-113 243.4 days, manganese-54 243.2, zinc-65 243.3 and cobalt-60 243.1. Four nuclides with half-lives from 115 days to 5.3 years agree to within a third of a day, which is a good sign the table is consistent (Figure 2). Caesium-137 gives 250 days, but its 30-year half-life makes it a poor clock: after 243.2 days it should have 815.4 Bq left, which rounds to the 815 Bq the paper lists.
Step 4: why some lines were left out. At 243 days, the same maths leaves chromium-51 (half-life 27.7 days) with 9.7 Bq of its original 4,275 Bq, and strontium-85 with 43 Bq of 583. Too weak to count well in a reasonable time, which is the practical weakness of mixed sources the paper sets out to avoid. The natural materials, with half-lives of hundreds of millions of years or more, never need replacing.
| Check | Paper | Our tool result |
|---|---|---|
| Fit residuals at the six reference lines | −1.1, 1.8, −0.9, −1.2, 1.6, −0.1% | −1.1, 1.8, −0.9, −1.3, 1.5, −0.1% |
| Difference from the multinuclide source | −4.1 to +1.4%, average 1.5% | −4.1 to +1.5%, average 1.46% |
| Time from certificate to measurement | not stated | 243 days (4 nuclides agree within 0.3 days) |
| Cs-137 on the measurement date | 815 Bq | 815.4 Bq at 243.2 days |
| Cr-51 left at measurement | (line not used) | 9.7 Bq of 4,275 Bq |
How we made the graphs
The charts and every number above come from scripts/blog-figures/hpge-detector-efficiency-natural-radionuclide-reference-materials.mjs in fibub’s public code. It calls the tools’ own functions:
fitCurvefromsrc/lib/sci/phcurvefit.ts, the least-squares engine of the curve fitter, with the formulaa + b*x + c*x^2on ln E and ln ε. The fit is unweighted; that is our choice, and it reproduces the paper’s residuals. A fit weighted by the paper’s uncertainties gives a slightly different curve.timeToanddecayfromsrc/lib/sci/decay.ts, the maths of the decay calculator.
Inputs, all from the paper: energies, efficiencies and uncertainties (Table 4); half-lives and certificate activities (Table 2); activities on the measurement date and the multinuclide efficiencies for the silica matrix (Table 5); the bias column (Table 6). We converted half-lives in years to days with 365.25 days per year. Run the script and it prints every number in this article and redraws both charts.
How we test the maths
On every change to fibub, the test suite checks these functions against known answers:
- In
tests/sci-tools/curve-fitter.mjs, a weighted exponential fit must match SciPy’scurve_fitto one part in a million (decay time 4.5554733), and the line, quadratic and cubic chi-square values must match SciPy too (quadratic p-value 0.19860555). The polynomial starting values must be exact on a made-up parabola. - In
tests/sci.mjs, the Decay calculator must turn 370 MBq of fluorine-18 into 173.43 MBq after 2 hours, back-correct it to 370 MBq, and give 207.19 minutes to fall to 100 MBq. - The same block checks that one half-life of iodine-131 halves the activity exactly, and that carbon-14’s molar activity is 62.72 mCi/mmol.
Check it yourself
The efficiency curve, on the curve fitter:
-
In Columns: x, y and (optional) the uncertainty in y, replace the example with these six lines (ln of energy in keV, then ln of efficiency in %):
5.4748 1.5539 5.8633 1.2912 6.9088 0.4581 7.2867 0.1561 7.4756 0.0334 7.6981 -0.1613 -
Leave Uncertainty in y on None (unweighted) and choose Quadratic in Model.
-
The fit reads a ≈ 5.245, b ≈ −0.598, c ≈ −0.0135, and R² 0.9996.
-
To get the efficiency at 661.7 keV, put x = ln 661.7 = 6.4948 into the formula: e^(5.245 − 0.598 × 6.4948 − 0.0135 × 6.4948²) ≈ 2.20%. The paper measured 2.182% with the commercial source.
The age of the reference source, on the decay calculator:
- In Nuclide, choose Your own half-life…, type 312.19 in Half-life and choose d as the Unit.
- Type 1316 in Activity at the reference time and choose Bq.
- Type 767 in When will it fall to (same unit, optional).
- The result reads Falls to 767 Bq, about 243.2 days after the reference time. Repeat with tin-113 (115.09 d, 849 to 196 Bq) or zinc-65 (244.01 d, 2381 to 1193 Bq) and you get the same 243 days. For cobalt-60, choose Co-60 from the list; its built-in half-life (5.2713 years) differs a little from the paper’s 5.2711, which moves the answer by less than an hour.
What researchers can use it for
- Refit an efficiency curve from your own calibration lines with the curve fitter, and see the residuals and parameter uncertainties at once.
- Check a calibration source’s certificate against your counting date with the decay calculator, and spot lines that have decayed too far to use.
- Sanity-check a published table: if several nuclides in one source give different ages, something in the table is off.
- Plan counting times with the counting statistics helper tool before a long run on a weak line.
- Teach why efficiency falls with energy, and why short-lived calibration sources need replacing.
The fine print
- All numbers come from the paper’s tables as cited above; our fit and decay ages are recalculations, not new measurements.
- The paper’s method holds only above about 200 keV. Below that, self-absorption in the sample and summing effects grow, and a mixed source is still needed.
- The efficiencies apply to one detector, one can shape and a silica matrix of 1.505 g/cm³. A different detector, sample height or density needs its own calibration.
- We fitted without weights. The paper does not say in the parts we read whether its fit was weighted; the unweighted fit reproduces its residuals, so we used that.
- During the 220-hour count of the multinuclide source, tin-113 loses about 5% of its activity from start to end, so it matters which moment “activity on the measurement date” refers to.
Sources
- Jodłowski, P. (2025), A Simplified Method for HPGe Detector Efficiency Calibration Using Certified Reference Materials Containing Natural Radionuclides, Applied Sciences 15(17), 9774, doi:10.3390/app15179774. Licence CC BY 4.0 (stated in the article footer; checked 2026-11-16). Values used here: Table 4 (six calibration energies and efficiencies with uncertainties, fit residuals), Table 2 (half-lives and certificate activities of the multinuclide source), Table 5 (activities on the measurement date and efficiencies from the multinuclide source), Table 6 (bias between the two methods), the detector (n-type coaxial, 42% relative efficiency), the 121 cm³ sample can and the counting times (about 20, 100 and 220 hours).
- The fitted form, a second-degree polynomial in ln(efficiency) against ln(energy), is the one the paper names; the fit itself is ours, made with the Curve fitter's maths.
If you are an author or publisher and want a credit or a passage changed or removed, tell us and we will act quickly.