Detecting Low Levels of Plutonium with HPGe Spectra, Compton Anticoincidencing, and Advanced Methods in Robust Fitting *



R. L. Coldwell, G. P. Lasche, and J. A. Nobel

Constellation Technology Corporation, 7887 Brian Dairy Road - Suite 100, Largo, FL 33777


The problem of accurately detecting extremely low levels of plutonium is rapidly gaining increasing importance in applications of nuclear counter-proliferation, verification, and environmental and waste management. If methods to confidently detect trace signatures of plutonium based on gamma-ray spectra alone could prove sufficiently effective, remote-monitoring systems could be greatly simplified and confidence levels and throughput of environmental and waste management systems could be greatly increased. Relative to signal from very low levels of 239Pu, the background from natural terrestrial sources is so high that, as in space applications, consideration of Compton anticoincidencing is compelling. However, its use in materials management applications requires demonstration that the increased sensitivity be sufficiently high to overcome added cost and weight. In this paper we report the results of a laboratory experiment to determine sensitivity that can be gained in detecting low levels of 239Pu from Compton anticoincidencing in combination with newly-developed spectral analysis methodology, in which we collected a logarithmically-varying time series of spectra both with and without Compton anticoincidencing from a 98-nCi laboratory source. It was found that by using Compton anticoincidencing the collection time needed to achieve the same level of confidence could be decreased by a factor of 1.8.




The experiment described in this paper is the difficult problem of directly detecting low levels of weapons-grade plutonium from gamma rays in a highly background-dominated natural radiation environment. The fits are a continuation of fits to the unsuppressed data begun in [1]. The code RobWin under development by Constellation Technology produced fits to supernova 1987A [2] that gave rise to the hope that this code might be able to accurately detect small amounts of radioactive material in the presence of relatively large amounts of natural background. Finding 239Pu is particularly challenging because, except for easily-attenuated x-rays at about 13.6 keV which accompany about 5% of the nuclear decays, only 5 gamma-rays at any energy above 13.6 keV are produced in every thousand decays of the 239Pu nucleus [5]. While all 5 gamma rays contribute to the analysis below, the two largest with yields of 27 and 6 photons per 100,000 decays at 51.6 keV and 129 keV will be shown in detail below. The data was taken with an intrinsic germanium detector both without suppression and also simultaneously in anti-coincidence with a BGO Compton shield. The anti-coincidence part is designed to reduce the background and hence to enhance detectability of the signals of interest.

In this paper the question of the effectiveness of suppression of background data is considered. A 100 keV part of the 3000-keV 32-hour suppressed and unsuppressed spectra is shown in Figure 1. The standard deviation in small peaks above the background is approximately equal to the square root of the background under the peaks. This background is down by a factor of two in the region of the 51.6 keV peak and by about 25% in the region of the 129 keV peak. This means that if the peak statistics were to remain background-dominated and if there were no loss from the peak in the suppression process, the

Figure 1. 32-hour suppressed and unsuppressed spectra in the region of the 51.6 and 129 keV Pu-239 lines


suppressed spectrum would have a 40% smaller standard deviation than the unsuppressed spectrum.

The key to making this methodology work is accurate tables for each nuclide present in a spectrum of all the peak energies, the relative peak strengths, and the uncertainties in these measurements, and the application of robustly-converging nonlinear minimization techniques derived from RobFit [3,4]. For the analyses in this paper, nuclide tables were drawn primarily from the Brookhaven Nuclear Data Library [5], with supplemental information from Shirley’s table's [6].



A 98 nano-curie 239Pu source was placed 10 cm from the face of a 100% n-type high-purity germanium detector along the axis of its cylindrical symmetry. The detector was surrounded by a bismuth-germanate anticoincidence shield, which served to reduce external background both passively and actively. Using a Constellation Technology multi-channel analyzer, two 16,384-channel spectra in the range from 0 to 3 MeV were collected simultaneously after exposures for each of 15 minutes, 30 minutes, 1 hour, 2 hours, 4 hours, 8 hours, 16 hours, and 32 hours.




Figure 2. The full 32-hour suppressed spectrum.

The only information needed for analysis with RobWin, other than the spectrum itself was a rough idea of the gain range (0-3 MeV) and the fact that plutonium was suspected to be present in a natural terrestrial environment – no independent information about energy or efficiency calibrations of the detector were required by the code or used. From the data in the 32-hour spectrum, rough starting values for energy calibration, an estimate of the FWHM and two peak shape templates – typical of energies below and above the pair production threshold – were generated by using facilities in the code. Next, the continuum was fit and naturally occurring nuclides were incrementally added to the analysis search list as suggested by residual peaks in the spectrum. Isotopes found to be significantly present were: 228Ac, 207Bi, 212Bi, 214Bi, 137Cs, 40K, 231Pa, 234Pa, 210Pb, 212Pb, 214Pb, 224Ra, 226Ra, 234Th, 208Tl, and 235U. In addition, significant elemental x-ray activity was observed from Au and Bi. During each iteration the detector energy calibration, detector efficiency response and FWHM as a function of energy were determined and refined iteratively by the code using non-linear minimization and internal tables of the known information about the energy, relative strength, and the uncertainties of each peak of each of the nuclides in the search list. The full fit for the spectrum thus achieved is shown in Figure 2.

A template was made of the fit to the 32-hour background for use in fitting the shorter accumulation time spectra. This background was actually fitted down to 20 keV while only peak energies above 35 keV were used.

Figure 3. The 32-hour spectrum in the low energy region.

The final fit to the 32-hour spectrum in the region including the 51.6 keV and 129 keV 239Pu peaks is shown in Figure 3.

With the background, energy calibration and efficiency determined by the other peaks and the entire range, it is relatively easy for RobWin to find the well isolated 51.6 keV 239Pu peak with 1252 counts shown in Figure 4.

Figure 4. The 51.6 keV region of the 32-hour spectrum.

The efficiency of the detector is much higher in the 129 keV region so that despite its factor of 4.3 drop in intensity, the 129 keV 239Pu line contains statistically significant 642 counts. The fact that the competing 228Ac line is determined by other 228Ac lines on either side of this region makes this peak meaningful for RobWin. This region is shown in Figure 5.

Figure 5. 129-keV region of the 32-hour spectrum.

The final result given in the table below includes the counts from all 239Pu peaks above 35 keV in energy and gives rise to



The background template determined in the previous section allows relatively short count spectra to be quickly fitted. The 15-minute spectrum is shown below in figures 6 and 7.

Figure 6. The region of the 51.6 keV 239Pu line in the 15-minute suppressed spectrum.

Figure 7. The region of the 129 keV 239Pu line in the 15 minute spectrum.


It can be seen in these two figures that there is a surplus of counts on the 51.6 keV line. This gives rise to which should be compared to the correct value of 1.11 ± 0.08 as can be found from the fit made below. The values stay high until the 4-hour run shown in Figures 8 and 9.

Figure 8. 50-keV region of the 4-hour spectrum.

Figure 9. The 129-keV region in 4-hr spectrum.

In the 4-hour spectrum the ratio of . This is to be compared with the more accurate value of 4.44 ± 0.34, from the fits to the full set of the data for all collection times.



Table 1. 239Pu counts vs Time

Time (hr)


Counts (BGO)


3977 ± 357

3854 ± 319


1281 ± 276

1813 ± 206


782 ± 195

902 ± 154


283 ± 135

350 ± 103


254 ± 93

371 ± 72


197 ± 66

209 ± 53


98 ± 47

115 ± 36


32 ± 32

59 ± 24


The value of minimized to a slightly high value of 11.9 for the 8 data points in determining the constant a and its standard deviation for the suppressed data and to 9.0 for the unsuppressed data. The result of this fit is

The value of minimized to10.0 and 8.7 in the process of yielding the standard deviations as

Figure 10 Counts upper line and their standard deviations lower line. The lines have been fitted to the values.


The standard deviation in the number of counts comes from the size of the background, which the suppression has lowered while the size of the counts in the ideal case is the same. The standard deviations are seen to have much less standard deviation than do the counts. This means that the significance or number of counts divided by the standard deviation in the number of counts as a function of time is given by

where c can be found by minimizing

with respect to c. The minimum c 2 is a slightly large 11.9 for the unsuppressed data and 12.9 for the suppressed data. The resulting fits to the significance are

Figure 11. Significance of the suppressed data. The line is the fit to the data.





The fits show that the square of the significance of the suppressed data divided by the unsuppressed data is 1.76± 0.32. This compares favorably with estimates of 1.25 to 2 from examination of Figure 1, from which it can be concluded that Compton suppression quite significantly increased detection confidence for plutonium. In this experiment the use of Compton suppression reduced background noise without significantly reducing signal from the peaks of interest as determined from the RobWin fit. With the assistance of RobWin Analysis software, a 98 nano-curie source of 239Pu placed 10 cm in front of a 100% n-type high-purity germanium detector can be detected with in 5 hours for BGO Compton-suppressed data and 9 hours for unsuppressed data. Owing to the fact that the background dominates the error, the significance at first is proportional to the size of the source rather than to its square root. This means that a 1 micro-curie source at this distance should be detectable with the use of Compton suppression at a significance level of 5 with a collection time on the order of 30 minutes.




[1] G.P.Lasche, R.L.Coldwell, and J.A.Nobel, "Detection of Low Levels of Plutonium

in Natural Environments from Gamma-Ray Spectra with Advanced Methods in Robust

Fitting," Symposium on Radiation Methods and Applications, Ann Arbor, MI, 1988.

[2] G.P. Lasche, R.L. Coldwell, J.A. Nobel, A.C. Rester, and J.I. Trombka, "Spectral Analysis in High Radiation Space Backgrounds with Robust Fitting," Conference on High Energy Radiation Background in Space, CHERBS 1997, Snowmass, CO, IEEE Operations Center, 1998.

[3] R.L. Coldwell, Nucl. Instr. and Meth. A242 (1986) 455.

[4] R.L. Coldwell and G.J. Bamford, The Theory and Operation of Spectral Analysis Using Robfit, AIP, New York (1991).

[5] PCNUDAT Nuclear Data master database used by permission of NNDC at B.N.L.

[6] E. Browne and Richard B. Firestone; V.S. Shirley, Editor, Table of Radioactive Isotopes, John Wiley, New York (1986) pp. C-19, C-24.