WorksPhysics and Astronomy
Radiopurity measurements of liquid scintillator for the COSINE-100 Upgrade
Abstract
A new 2,400 L liquid scintillator has been produced for the COSINE-100 Upgrade, which is under construction at Yemilab for the next COSINE dark matter experiment phase. The linear-alkyl-benzene-based scintillator is designed to serve as a veto for NaI(Tl) crystal targets and a separate platform for rare event searches. We measured using a sample consisting of a custom-made 445 mL cylindrical Teflon container equipped with two 3-inch photomultiplier tubes. Analyses show activity levels of mBq/kg for U and mBq/kg for Th.
E-mail: [email protected], [email protected]
Introduction
Observational evidence for the existence of dark matter has been well-documented [1, 2, 3], leading to a surge in direct detection experiments [4]. Among these, the DAMA experiment has claimed the observation of an annual modulation signal that aligns with the presence of weakly interacting massive particles (WIMPs), a leading dark matter candidate [5, 6]. To independently test the DAMA experiment, the COSINE experiment was launched, using the same target material, NaI(Tl) crystals [7]. After 6.5 years of operation at the Yangyang Underground Laboratory, the COSINE-100 experiment has produced results challenging DAMA’s claims with a significance of more than 3 standard deviations [8]. Preparations are now underway for the next phase, the COSINE-100 Upgrade (COSINE-100U) [9], at a new underground facility, Yemilab, in Jeongseon, Korea [10].
A liquid scintillator (LS) system will be used in the upcoming COSINE-100U experiment to help characterize signals and backgrounds in the crystal detectors. A similar system was used in the COSINE-100 experiment where the NaI(Tl) crystal assemblies are positioned at the center of an acrylic box, which is filled with LS, ensuring a minimum veto thickness of 40 cm surrounding the crystal assembly at all points. Eighteen 5-inch photomultiplier tubes (PMTs) attached to the box record light signals produced in the LS, and the data acquisition system simultaneously combines these signals with those from the crystals. Thus, the LS functions as a tagger for gamma rays originating both from outside the box and from the crystals themselves. The LS veto system has demonstrated tagging efficiencies of 15–30% in the 2–6 keV range, with 65–75% efficiency for the detection of originating from the crystals [11]. Tagged events can also serve as an independent channel for understanding the background [12]. Additionally, the LS is utilized for exotic dark matter searches when the coincidence interactions simultaneously with crystals occur [13, 14].
A total of 2,400 liters of the linear-alkyl-benzene (LAB) based LS has been newly produced and is awaiting use in the COSINE-100U experiment. Three grams of 2,5-diphenyloxazole (PPO) were added per liter of LAB to act as a fluorescent material. Since the emission spectrum of PPO does not match the spectral response of the PMT, 30 mg/L of 1,4-bis[2-methylstyryl]benzene (bis-MSB) was added as a wavelength shifter for efficient light collection [15]. The radiopurity of the LAB-LS was measured prior to installation in the main detector to minimize the risk of introducing significant background radioactivity to the crystals. Data were collected for one month using a small-scale detector at a ground laboratory. Using the data, we developed pulse shape discrimination (PSD) and coincidence analysis to determine the intrinsic background level of the liquid. Furthermore, radioactivity measurements of the LAB-LS sample were performed using high purity germanium detectors [16] for comparison.
Experimental setup for radiopurity measurements
To quickly screen the LAB-LS for its radiopurity, a mini-detector designed to contain a 445 mL sample of the liquid sample was built. As shown in Figure 1 (a), two cylindrical-hollow polytetrafluoroethylene (PTFE) cases with embedded Hamamatsu R12669SEL{footnote}1 PMT are designed so that the space between them can be filled with the liquid. The inner radius of the case matches the radius of the PMT to mechanically stabilize the PMTs and prevent leakage through the gap between the case and PMT. The outer radius of the case is 7.6 cm, and there are six 3 mm radius holes that go through the cases as shown in {xref}Fig. 1 (a) and (b). When the two PTFE cases are joined, an O-ring is placed at the matching area, and long bolts are inserted into the six holes, with nuts tightened on the outside of the cases. After filling the LAB-LS through the chimney, the chimney is sealed with a PTFE stopper, as shown in {xref}Fig. 1` (c), to prevent contamination by external radioactivities, most notably radons.

Figure 1. (a) A schematic diagram of the detector. The yellow region denotes the PTFE container and the gray region denotes the PMTs on both sides. (b) A cross-sectional view of the actual detector, showing the teflon housing and PMT window. (c) The detector assembly inside the lead castle. (d) An external view of the lead shielding before installation of polyethylene blocks. (e) An external view of the polyethylene block shielding structure.
The assembled detector was shielded with lead bricks to reduce external background, as shown in Figure 1 (c) and (d). An additional layer of polyethylene blocks was added for environmental neutron mitigation (see Figure 1 (e)). The analog signal pulses were digitized, and a full waveform is recorded using analog-to-digital converter (ADC) modules with a 2 ns sampling precision. An event was triggered only when both PMT signals exceeded 6 mV discrimination threshold. With a trigger rate of 6 Hz, a total of 522 hours of data was collected over the course of a month.
Detector response
Particle identification
Pulse shape discrimination (PSD), which distinguishes between and based on the decay time of the liquid, is a popular technique for particle identification (PID). To identify the difference in decay time from a waveform, a PSD parameter is defined as the ratio of to ,
where is the range of summation for , while and represents the end time bin and the charge of bin of the waveform, respectively. Figure 2 (a) shows the distribution of PSD parameter when is 30 ns. One can see the events are distributed from 0.07 to 0.12 while events are distributed from 0.14 to 0.18. The decay time of events is relatively short, and thus appear in the lower range of PSD parameter, while is distributed in the higher range. The data regions are fitted with models based on Gaussian functions for the and separately. Using the resulting fit parameters, the discriminative power of PSD is quantified by the figure of merit (FoM), which is defined as follows,
Figure 2. optimization. As varying the from 10 ns to 60 ns, draw the PSD parameter distribution as (a) and fit it with Gaussian function (blue and red curve). The FoM representing the separation between the and was derived from these fits, as shown on (b). The FoM peaked at 3.9 when is 30 ns.
where and are the mean and standard deviation of the best-fit, respectively. Since the FoM depends on the , we optimize the FoM by varying the from 10 to 60 ns. Figure 2 (b) shows the scan results of the FoM as a function of . We determined to be 30 ns with the largest FoM value.
The charge-dependent distribution of the PSD parameter with the optimized can be seen in Figure 3. One can see a clear distinction between the and bands separated by the red line. The three islands in the band represent events and a continuous underlying distribution for environmental . To determine the criteria for selection, the data was divided into nine zones based on a two-PMT total charge, from 6,000 ADC to 24,000 ADC. For each zone, the data on the PSD parameter was fitted with two separate Gaussian functions, and then we found a cut value that rejects 99.87 % of events. These cut values are plotted in Figure 3 as pink dots, and a linear function was obtained by fitting them to use as a selection criterion. Here, the PSD cut shows an efficiency of over 94.51 % for events over 10,000 ADCs.
Figure 3. The PSD result for distinguishing and events. The clear band below the red line shows the distribution of events, while the band with three islands above the curve represent the distribution of events. The red line was used to separate events from events.
Energy calibration
The three islands identified in Figure 3 appear as three distinct peaks in the charge distribution of the events selected via PID in § 3.1 as shown in Figure 4 (a). They appear to come from peaks caused by contaminated during LS production and detector assembly, and the charge ratios between them are in good agreement with the ratios between the decay energies of , , and which allows for an energy calibration.
Figure 4. Energy Calibration. (a) Three Gaussian functions with an exponential and constant term for background were used to fit the , and peaks. The red solid line indicates the fitted model, and the three Gaussian peaks are represented by the dashed lines. Total three means and three widths of each peaks were obtained and these are used for the energy calibration function (b) and the resolution function (c). The black data points with error bars were obtained from the fitting results with known energies of three peaks.
The charge distribution of the events was modeled with three Gaussian functions to describe the three peaks and one exponential function with a constant term to account for the background consisting mainly of fast neutrons. The means and standard deviations of the Gaussian functions were used to determine the calibration and resolution functions for energy. The Gaussian mean values of the charge sum distributions were fitted by a linear function and used as a calibration function for energy, while the resolutions were fitted by,
where is the energy resolution, while and are used as free parameters. The fitted functions for energy calibration and resolution are shown as red solid lines in Figure 4 (b) and (c), respectively.
Coincidence analysis
from chain
To measure the activity of the LAB-LS -contaminated series, the decays of and in the decay chain were investigated. The isotope decays to via -decay with a Q-value of 3.27 MeV. subsequently undergoes -decay with a Q-value of 7.83 MeV and a half-life of 164.3 s to produce . Thanks to the short decay time of , these sequential decays can appear as a time-coincidence event in the current DAQ setup.
Utilizing the PID from § 3.1, the time interval between and over a range of 2 to 1000 s is shown in Figure 5 (a). When time difference is within this time span, the energy distribution of those coincident events is represented by the tagged events in Figure 5 (b). The coincidence time between and events can be determined by fitting the data with an exponential function. Since random coincidences between and events may occur, we used an exponential function with an additional constant term,
Figure 5. Sequential decays of and identified via time coincidence. (a) Time difference between and events is fitted with an exponential function (red curve) to tag the sequential decays of - (black dots with error bars). (b) The energy distribution of identified by the PSD method in § 3.1. The black dots denotes total events, while the red-filled histogram represents the events from decay.
where the variable is coincident time difference. The three fitting parameters are , , and . Here, represents the activity of , is its half-life, and accounts for the amount of random coincidence. We obtained the best-fit half-life of , which is consistent with the known value.
A significant portion of the remaining time-coincidence events is attributed to contamination from (with a half-life of 3.8 days). Consequently, it is necessary to analyze these events over an extended period. The collected data over a month were divided into appropriate time intervals for this analysis. The distribution in each interval was fitted by Eq. (4) with the half-life fixed at , while keeping and as free parameters.
The activity , determined from the fitting, is plotted in Figure 6. The observed decrease in activity appears to be due to contamination from prior to detector assembly. To model this decrease, we used the same function as in Eq. (4). The exponential term represents the contribution from , while the constant term accounts for internal contamination from the series. The best-fit model aligns well with the observed decrease in activity. The internal contamination from , as determined from the constant term, is mBq/kg.
Figure 6. Time evolution of the measured activity. The black dots with error bars represent the activity (in mBq/kg) at each time interval. They were fitted with a function represented by the red solid line, of which the constant term is depicted by the red dashed line.
from chain
The sequential decays of and in the decay chain can be identified by time-coincidence condition, thanks to the short ’s half-life ( [] = 140 ms). A dataset consisting of -only events identified via PID was prepared, and the coincidence events were selected using two criteria: We require the time interval between two consecutive events to be less than 1 second, with energy selection within 5 standard deviations of the measured energies. The standard deviations are based on detector resolutions modeled using Gaussian functions, with the known energies as the means.
Similarly, through coincidence analysis, we obtained an activity of 0.012 0.007 mBq/kg. In Figure 7, the time difference of selected coincidence events are represented as data points with error bars. The coincidence time distribution was fitted by Eq. (4), with the half-life fixed at 140 ms. The contamination by is derived from the amplitude term of the best-fit model.
Figure 7. The coincidence events by a sequential decays of and . The black dots with error bars denote the measured time intervals between coincident events, while the red line represents the exponential fit based on Eq. (4).
Summary and discussion
In the COSINE dark matter search project, the unique LS veto system plays a crucial role in background reduction and detector understanding. The next phase of COSINE-100, named as COSINE-100U, is under preparation, and the total of 2,400 L of LAB-based LS has been produced for use in the veto system. As for the performance testing of the newly produced LAB-LS, the activities of and are measured.
A small-sized radiopurity screening detector is created and one month of data was collected for the test. With the PSD technique, energy calibration of events is performed by separating events from the sample. The coincidence analysis of sequential decays of and in the decay chain provides measurements of internal contamination by and additional contamination by , resulting in a measured activity of 0.091 0.042 mBq/kg. Additionally, the sequential decays of and in the decay chain were identified, determining the activity to be 0.012 0.007 mBq/kg. These results are consistent with those of the liquid scintillator used in the COSINE-100 experiment, which reported activities of 0.087 mBq/kg and 0.016 mBq/kg for and , respectively [11].
In order to check the influence of the measured activities on the background of the NaI(Tl) crystal, we compared those with the background model of COSINE-100 [12]. In Figure 8, one can see that the overall contributions inferred from the measured results are negligible for the dark matter search region and, in particular, the contribution at the region of interest for WIMP signals (- keV of single-sited interaction region) is lower than 0.02 %. Hence, we confirmed that the LAB-LS sample is sufficiently pure to be used as the veto system in the COSINE-100U experiment.
Figure 8. Simulated background model. The red histogram indicates the total COSINE-100 background model and the cyan and blue histograms indicate the contributions from and contamination in LS, respectively. The single-hit energy spectrum (a) shows contributions of 0.07 %, while the multiple-hit spectrum (b) indicates contributions of 3.14 %. These low levels illustrate the minimal impact of LS contamination compared to the total COSINE-100 background model.
Activity measurements using HPGe detector For cross-validation, the radioactivities of the LAB-LS sample were also measured using a High Purity Germanium (HPGe) detector [17]. The HPGe detector collected data for 17.25 days in Yemilab. The results for four radionuclides are presented in Table 1, showing that all measurements were within the upper limits, indicating that the LAB-LS is sufficiently clean.
| Nucleus | ||||
| Activity [mBq/kg] | < 0.90 | < 7.39 | < 2.01 | < 0.74 |
Table 1. Measured radioactivity levels of the LAB-LS sample using an HPGe detector.
Acknowledgments
This research was supported by: the Institute for Basic Science under project code IBS-R016-A1 Republic of Korea, the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (NRF-2021R1A2C1013761), and the Chung-Ang University Research Scholarship Grants in 2023.
References
References
- [1]Rubin, V. C., & Ford, W. K. (1970). Rotation of the andromeda nebula from a spectroscopic survey of emission regions. The Astrophysical Journal, 159, 379.
- [2]Clowe, D., Bradač, M., Gonzalez, A. H., Markevitch, M., Randall, S. W., Jones, C., & al., . et . (2006). A direct empirical proof of the existence of dark matter*. The Astrophysical Journal, 648, L109.
- [3]collaboration, P. (2014). Planck 2013 results. XVII. Gravitational lensing by large-scale structure. Astron. Astrophys., 571, A17.
- [4]LUX-ZEPLIN collaboration, First dark matter search results from the lux-zeplin (lz) experiment, Phys. Rev. Lett. 131 (2023) 041002.DOI
- [5]collaboration, D. (2008). First results from DAMA/LIBRA and the combined results with DAMA/NaI. Eur. Phys. J. C, 56, 333.
- [6]Bernabei, R., & al., . et . (2013). Final model independent result of DAMA/LIBRA-phase1. Eur. Phys. J. C, 73, 2648.
- [7]Adhikari, G., & al., . et . (2018). Initial Performance of the COSINE-100 Experiment. Eur. Phys. J. C, 78, 107.
- [8]Carlin, N., & al., . et . (2024). COSINE-100 Full Dataset Challenges the Annual Modulation Signal of DAMA/LIBRA.arXiv
- [9]Lee, D., & al., . et . (2025). Upgrading the COSINE-100 experiment for enhanced sensitivity to low-mass dark matter detection. Commun. Phys., 8, 135.
- [10]Park, K. S., Kim, Y. D., Bang, K. M., Park, H. K., Lee, M. H., So, J., & al., . et . (2024). Construction of Yemilab. Front. In Phys., 12, 1323991.
- [11]Adhikari, G., & al., . et . (2021). The COSINE-100 liquid scintillator veto system. Nucl. Instrum. Meth. A, 1006, 165431.
- [12]collaboration, C.-. 100 . (2025). Improved background modeling for dark matter search with COSINE-100. Eur. Phys. J. C, 85, 32.
- [13]Borexino collaboration, Search for Solar Axions Produced in p(d, ³He)A Reaction with Borexino Detector, Phys. Rev. D 85 (2012) 092003.DOI
- [14]COSINE-100 collaboration, First Direct Search for Inelastic Boosted Dark Matter with COSINE-100, Phys. Rev. Lett. 122 (2019) 131802.DOI
- [15]Park, J., & al., . et . (2013). Production and optical properties of gd-loaded liquid scintillator for the reno neutrino detector. Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, 707, 45.
- [16]Sala, E., Hahn, I. S., Kang, W. G., Kim, G. W., Kim, Y. D., Lee, M. H., & al., . et . (2016). Development of an underground low background instrument for high sensitivity measurements. J. Phys. Conf. Ser., 718, 062050.
- [17]Lee, E. K., & al., . et . (2019). Measurements of detector material samples with two HPGe detectors at the Yangyang Underground Lab. PoS ICHEP2018, 809.






