betaFeedback

WorksPhysics and Astronomy

Neutron drip line in calcium isotopes from a chiral interaction

Abstract

Interactions derived from effective field theories of quantum chromodynamics have thus far failed to bind calcium nuclei beyond neutron number N=40N=40, while nuclear density functionals typically place the neutron dripline near 70^{70}Ca, at N=50N=50. We present the chiral interaction N3^3LOTexas_{\rm Texas}, a combination of two- and three-nucleon potentials at fourth and third chiral order, respectively, with low-energy constants optimized using emulator-accelerated fits to few- and many-body data. This interaction accurately reproduces binding energies and charge radii of key nuclei with mass number A=3A=3 to 208208, important excited states, and nuclear matter near saturation. Using ab-initio methods, coupled cluster and valence-space in-medium similarity renormalization group, we find that the calcium two-neutron dripline extends to 71^{71}Ca.

Introduction

What are the limits of nuclear binding? This question lies at the heart of our understanding of how visible matter is formed in the universe. It drives experimental and theoretical research in nuclear physics and astrophysics alike [2, 3, 4, 5] [1]. The answer requires determining the limits of the nuclear chart, defined by the proton and neutron driplines—the boundaries beyond which an additional proton or neutron cannot be bound to the nucleus. Experimentally, the proton dripline is known up to neptunium (proton number Z=93Z = 93) [6], while the neutron dripline has only been mapped up to neon (Z=10Z = 10) [7]. Very neutron-rich nuclei are difficult to synthesize in the laboratory and locating the neutron dripline for heavier elements will be extremely challenging [8, 9]. Nevertheless, knowing how many neutrons can be bound in a given isotopic chain is essential for modeling nucleosynthesis, particularly the rapid neutron-capture process that produces the heavy elements in neutron-star mergers and core-collapse supernovae [10, 11]. Reliable theoretical predictions beyond experimentally accessible regions are needed.

The calcium isotopes (Z=20Z = 20) are especially interesting due to their closed proton shell. Energy density functional calculations predict their neutron dripline to extend to 70Ca^{70}\mathrm{Ca} [1], or even 76Ca^{76}\mathrm{Ca} [14], while relativistic mean-field calculation place it between N=42N = 42 and 56. Nuclear shell model calculation, based on realistic Hamiltonian derived from linked-diagram perturbation theory, shows no sign of two-neutron emission until 70Ca^{70}\mathrm{Ca} [15]. These predictions, which all put the dripline beyond N=40N = 40, are consistent with the experimental discovery of 60Ca^{60}\mathrm{Ca} [16], and measurements of the spectra of 56,58Ca^{56,58}\mathrm{Ca} and 62Ti^{62}\mathrm{Ti} [17, 18]. In contrast, existing ab initio computations based on interactions from chiral effective field theory [19–21] place the neutron dripline closer to stability, around 60Ca^{60}\mathrm{Ca} [5] [22, 23, 24, 25, 26], finding that 61Ca^{61}\mathrm{Ca} is unbound with respect to 60Ca^{60}\mathrm{Ca} and further decreasing binding energy as the neutron number increases beyond N=40N = 40. 70Ca^{70}\mathrm{Ca} only appears as a probable dripline nucleus when

considerable theoretical uncertainties are taken into account [5]. Clearly, present interactions from chiral effective field theory used in abab initioinitio dripline predictions lack precision at best; at worst, they may exhibit a significant flaw.

In this Letter we address this shortcoming and our main results are shown in Figure 1. These abab initioinitio results were obtained by optimizing a new chiral interaction, named N3^{3}LOTexas_{\text{Texas}}, and using many-body methods including coupled-cluster [27–31] and valence-space in-medium similarity renormalization group (VS-IMSRG) [32–35]. This interaction includes higher order terms in the chiral (Weinberg) power counting compared to the previous chiral interactions [36, 37] and has almost twice the number of low-energy constants (LECs), making the optimization much more challenging. We overcome this challenge by employing high-fidelity emulators for all observables in the objective function. We see that the predicted dripline with N3^{3}LOTexas_{\text{Texas}} extends up to at least 70^{70}Ca. We note that the region of neutron-rich nuclei around N=50N=50 is fascinating for nuclear structure because of the interplay of weak binding, deformation, and strong inter-nucleon correlations [22, 26, 38–40]. Furthermore, one expects that subtle details of the two-nucleon (2N) and three-nucleon (3N) interactions become amplified near the dripline, and all these effects may impact its precise location [41, 42].

Figure 1. Ground-state energies (EgsE_{\mathrm{gs}}}), one-neutron (S1nS_{1n}) and two-neutron (S2nS_{2n}) separation energies for calcium isotopes computed with the interaction N3LOTexas\mathrm{N}^3\mathrm{LO}_{\mathrm{Texas}} using coupled-cluster (CC) and valence-space in-medium similarity renormalization group (VS-IMSRG) methods, compared with 1.8/2.0(EM) results [5] and data [12] [13]. Stars (open) are masses (extrapolations) from AME2020 [13]. The red and gray bands indicate many-body uncertainty and different decoupled valence spaces, respectively. The CC computations are performed only for even-even nuclei.

This Letter is organized as follows. We discuss the interaction, the optimization protocol, model-space parameters, and employed many-body methods leading to the results shown in Figure 1. For validation we present results for the dripline in oxygen isotopes and compare with available data, we also compute binding energies and charge radii of closed-shell nuclei ranging from 4^{4}He all the way up to the heavy nucleus 208^{208}Pb, and saturation properties of infinite nuclear matter. Additional information regarding the optimization, excitation spectra in selected calcium isotopes, and the high-fidelity emulators developed in this work are presented in the End Matter and in the Supplemental Material.

Methods

Methods.— We start from the intrinsic Hamiltonian

H^=T^−T^CoM+V^2N+V^3N,(1)\hat{H} = \hat{T} - \hat{T}_{\mathrm{CoM}} + \hat{V}_{2N} + \hat{V}_{3N}, \tag*{(1)}

where T^\hat{T} and T^CoM\hat{T}_{\mathrm{CoM}} are the total kinetic energy and the center-of-mass kinetic energy, respectively. For the 2N potential V^2N\hat{V}_{2N} and the 3N potential V^3N\hat{V}_{3N}, we use the chiral N3^{3}LOTexas_{\text{Texas}} potential developed in this work. It consists of a 2N interaction up to next-to-next-to-next-to-leading order (N3^{3}LO), based on the formalism presented in Refs. [43, 44], and the 3N interaction up to next-to-next-to-leading order (NNLO) from [45], with nonlocal regulator functions f(p)=exp⁡[−(p/Λ)2n]f(p)=\exp[-(p/\Lambda)^{2n}] and cutoff Λ=394\Lambda=394 MeV with power n=4n=4. Note that the 2N and 3N sector interactions come from different orders. We use a multi-start strategy to optimize the values of the 28 low-energy constants (LECs) in the 2N and 3N sectors to

neutron-proton scattering phase shifts, nucleon-nucleon 1S0{}^{1}S_{0} effective range parameters, and bound-state observables in 2^{2}H, 4^{4}He, and 16^{16}O. This latter nucleus is important because (i) it is sensitive to partial waves in the interaction beyond those probed in the lightest nuclei; (ii) we can build high-fidelity emulators for this nucleus; and (iii) experience shows that nuclear saturation is hard to reproduce without this anchor [36, 37, 46, 47]. Importantly, the use of emulators reduces the computational cost of the optimization by several orders of magnitude and facilitates exploration across a broad LEC parameter domain. For nucleon-nucleon scattering we employed the eigenvector continuation [48] formulation presented in [49]; for 2^{2}H and 4^{4}He, we used the eigenvector continuation framework tailored to full configuration interaction (FCI) and Jacobi no-core shell model (J-NCSM) [50, 51] calculations, respectively; for the 16^{16}O ground-state energy and charge radius, we used the sub-space projected coupled-cluster (SPCC) method [52]. In a final step of the optimization protocol we computed the binding energies of 22,24^{22,24}O and 40,48^{40,48}Ca for a few candidate interactions to identify the optimal one. We note that there was a tension between the reproduction of the scattering phase shifts in the partial-wave 3P2^{3}P_{2} channel and the charge radius of 16^{16}O, consistent with the global sensitivity analysis in [52].

We express the Hamiltonian given in Eq. (1) in a spherical harmonic oscillator basis with spacing ℏω\hbar\omega and single-particle energies up to (emax⁡+3/2)ℏω(e_{\max}+3/2)\hbar\omega, as defined below. The 3N interaction has an additional energy cut on the allowed excitation of three nucleons given by E3max⁡ℏωE_{3\max}\hbar\omega. With the Hamiltonian expressed in the harmonic oscillator basis we perform a Hartree-Fock calculation which serves as a starting point for our many-body computations. The Hamiltonian is normal-ordered with respect to the Hartree-Fock basis and the 3N interaction is approximated at the normal-ordered two-body level [53–59].

In our computations of the calcium and oxygen isotopes we employ the multishell VS-IMSRG [34, 60], and for the calcium isotopes we also perform coupled-cluster computations that start from a reference state that breaks rotational symmetry [31, 40, 61, 62]. Our VS-IMSRG calculations start from a model-space within emax⁡=14e_{\max}=14, E3max⁡=28E_{3\max}=28, and ℏω=12\hbar\omega=12 MeV. We use the VS-IMSRG [60] to decouple the full AA-body Hamiltonian within a spherical Hartree-Fock basis into a multishell valence-space Hamiltonian. Specifically we employ a neutron {1p3/2,1p1/2,0f5/2,0g9/2,1d5/2,2s1/2}\{1p_{3/2},1p_{1/2},0f_{5/2},0g_{9/2},1d_{5/2},2s_{1/2}\} valence space with a 48^{48}Ca core for 49−78^{49-78}Ca, a neutron pp-shell for 13−15^{13-15}O, and a neutron {sd-shell+0f7/2,1p3/2}\{sd\text{-shell}+0f_{7/2},1p_{3/2}\} valence space with 16^{16}O core for 17−32^{17-32}O. We then perform exact diagonalizations with the KSHELL code [63]. Our coupled-cluster calculations start from a product state built from axially symmetric natural orbitals [64–66] computed for emax⁡=14e_{\max}=14, E3max⁡=28E_{3\max}=28, and ℏω=12\hbar\omega=12 MeV. To facilitate the coupled-cluster computations, the deformed natural orbital basis are subsequently truncated to an emax⁡=12e_{\max}=12 space following Ref. [67]. We use the coupled-cluster singles-and-doubles approximation (CCSD), and we add 10% of the CCSD correlation energy as an estimate of neglected triples excitations [31, 68], and a small correction from angular momentum projected Hartree-Fock [61, 65].

For closed-shell nuclei, we use IMSRG(2) and the approximate IMSRG(3) [69, 70] scheme known as IMSRG(3f2) [71], as well as coupled-cluster computations with leading-order iterative triples corrections, CCSDT-3 [72]. For the energy computations shown in the upper panel of Figure 3 we performed extrapolations to the infinite basis limit following Ref. [73]. Energies computed by IMSRG(2) and IMSRG(3f2) were extrapolated using emax⁡=8,10,12,14e_{\max}=8, 10, 12, 14 with ℏω=16\hbar\omega=16 MeV (except for 208^{208}Pb, where we used ℏω=12\hbar\omega=12 MeV) and fixed E3max⁡=28E_{3\max}=28. For 208^{208}Pb, we additionally included extrapolated energies based on E3max⁡=18,20,22,24,26E_{3\max}=18, 20, 22, 24, 26 and 28, as well as the difference between emax⁡=14e_{\max}=14 and emax⁡=12e_{\max}=12 calculations. CCSDT-3 results of 40^{40}Ca and 48^{48}Ca were extrapolated from emax⁡=8,10,12e_{\max}=8, 10, 12 with fixed E3max⁡=16E_{3\max}=16, and those for 4^{4}He and 16,22,24^{16,22,24}O were extrapolated from emax⁡=8,10,12,14e_{\max}=8, 10, 12, 14 with the same E3max⁡=16E_{3\max}=16. All VS-IMSRG charge radii were calculated using emax⁡=14e_{\max}=14 and E3max⁡=28E_{3\max}=28, and CCSDT-3 used emax⁡=12e_{\max}=12 and E3max⁡=16E_{3\max}=16. We have verified that E3max⁡=16E_{3\max}=16 is sufficient to achieve convergence for light and medium-mass nuclei. For infinite nuclear matter, we use periodic boundary conditions and perform coupled-cluster computations with the Taube–Bartlett Λ\Lambda-CCSD(T) approximation [74], on a discrete lattice in momentum space [75, 76]. For few-nucleon systems (A≤4A\le4), we employed the J-NCSM to obtain virtually exact results [51].

Figure 3. Binding energies per nucleon and charge radii for doubly closed-shell nuclei from 4He{}^{4}\mathrm{He} to 208Pb{}^{208}\mathrm{Pb}, computed with the J-NCSM, IMSRG(2), IMSRG(3f2) and CCSDT-3 methods using N3^{3}LOTexas_{\mathrm{Texas}}, and compared to 1.8/2.0(EM) and experiment [12] [83, 84]. Error bars on the IMSRG(3f2) results represent the uncertainty from extrapolation to the infinite model space. In the legend, “Triple” indicates the inclusion of perturbative triples [71].

Results

Results.—Figure 1 shows the calculated ground-state energy (EgsE_{\rm gs}), and separation energies (S1nS_{1n} and S2nS_{2n}) of neutron-rich calcium isotopes, compared with the 1.8/2.0(EM) results from Stroberg et al. [5]. In our coupled-cluster calculations, 56^{56}Ca and 58^{58}Ca are nearly spherical, and we do not include the estimated energy contribution from angular momentum projection. This omission is one reason for the discrepancy observed between the coupled cluster and VS-IMSRG results. The predicted S1nS_{1n} and S2nS_{2n} values for calcium isotopes agree with experiment [12]. The N3^{3}LOTexas_{\text{Texas}} interaction predicts that 61^{61}Ca is unbound with respect to 1nn emission, while the 2nn dripline extends to 71^{71}Ca. These results are consistent with the Bayesian posterior probabilities for neutron binding in a Bayesian model average of global density functional theory models [14]. In comparison, the energies computed by Stroberg et al. [5] place the dripline around 60^{60}Ca. However, when those results are combined with experimental data in a Bayesian linear regression model then there is an approximate 50% probability that calcium isotopes heavier than 60^{60}Ca are bound [5]. With the N3^{3}LOTexas_{\text{Texas}} interaction we found that the N=40N=40 sub-shell gap in heavy-mass calcium nuclei is smaller, and this pushes the 2nn dripline toward 70^{70}Ca. This is consistent with expectations from the discovery of 60^{60}Ca [16] and measured spectra of 56,58^{56,58}Ca and 62^{62}Ti [17, 18]. The computed spectra of 50,52,54,56,58^{50,52,54,56,58}Ca agree with large-scale shell model calculations [77] and available experimental data. The positive parity states are of similar quality as that obtained with the 1.8/2.0(EM) interaction, whereas the negative-parity states lie significantly lower, reflecting the smaller N=40N=40 gap (see Supplemental Material).

We now turn to neutron-rich oxygen isotopes as they provide a stringent test for chiral interactions and quantum many-body methods [42]. The neutron dripline in oxygen isotopes is at 24^{24}O [41, 78, 79], with the ground states of 25^{25}O and 26^{26}O unbound by approximately 830 keV [80] and 18 keV [81, 82], respectively, relative to the 24^{24}O ground state. Recently, Kondo et al. [42] observed that both 27^{27}O and 28^{28}O exist as narrow unbound resonances.

Figure 2 compares the VS-IMSRG results with the N3^{3}LOTexas_{\text{Texas}} interaction to data and to those from Stroberg et al. [5]. The top panel shows that ground-state energies are accurately reproduced while the bottom panel shows accurate results for the S1nS_{1n} and S2nS_{2n}. The thin red band for VS-IMSRG with N3^{3}LOTexas_{\text{Texas}} indicates the estimated many-body uncertainty, quantified as the EgsE_{\rm gs} difference between “IMSRG(3f2)+Triple” and “IMSRG(2)” for oxygen and calcium isotopes listed in Figure 3, which is approximately 1%. The gray band for the 1.8/2.0(EM) shows the uncertainty arising from different decoupled valence spaces. We note that the results by Stroberg et al. [5] were based on ℏω=16\hbar\omega=16 MeV, whereas we use a smaller ℏω=12\hbar\omega=12 MeV for both oxygen and calcium isotopes to better capture weak binding effects near the dripline region. We checked that our dripline predictions are independent of ℏω\hbar\omega, and we also found accurate results for fluorine isotopes (see Supplemental Material).

Figure 2. Similar to Figure 1 but for oxygen isotopes. Experimental data taken from Refs. [12] [42].

The N3^{3}LOTexas_{\text{Texas}} interaction also reproduces low-lying spectra like the first 2+2^{+} excitation energies in doubly-magic oxygen isotopes, computed using the equation-of-motion method EOM-CCSDT-3 with emax⁡=8e_{\max}=8 and 10, see Tab. I in the End Matter.

Finally we turn to bulk properties of key closed-shell nuclei ranging from 4^{4}He to 208^{208}Pb and saturation properties of infinite nuclear matter. Figure 3 presents the calculated binding energies and charge radii using the VS-IMSRG and coupled-cluster methods and compares them with results from the 1.8/2.0(EM) interaction. The N3^{3}LOTexas_{\text{Texas}} interaction accurately describes these bulk properties up to 208^{208}Pb with deviations below 1.5%. We note that the N=ZN=Z doubly-magic nuclei 40^{40}Ca, 56^{56}Ni, and 100^{100}Sn deviate a bit more from data than the other nuclei. This could be because 16^{16}O was optimized using the computationally expensive CCSDT-3 method, while the heavier nuclei were computed using the more affordable IMSRG(3f2) method.

Figure 4 shows our predictions for the equation of state of nuclear matter, compared with the results from the Δ\DeltaNNLOGO_{\mathrm{GO}}(394) [37, 76], Δ\DeltaNNLO(450) [76, 87], and NNLOsat [36] interactions. Although nuclear-matter properties were not included in the optimization, the N3^{3}LOTexas_{\mathrm{Texas}} prediction is located within the empirically constrained regions of nuclear saturation [85, 86]. In this work we employed the Λ\Lambda-CCD(T) method [74], which includes more accurate triples corrections compared to the CCD(T) approximation [88] used in Ref. [76] (see Supplemental Material).

Figure 4. Energy per nucleon for symmetric nuclear matter (top) and pure neutron matter (bottom). The gray rectangle marks the empirical saturation region from Ref. [85]. The pink ellipse shows a 95% confidence region for the saturation point inferred from Bayesian posterior distributions based on a set of density functional theory predictions [86].

Summary and outlook.— By developing a novel chiral interaction, N3^{3}LOTexas_{\mathrm{Texas}}, we resolve the long-standing tension over the calcium neutron dripline, reconciling ab initio predictions with density functional theory and existing experimental results. This interaction also reproduces nucleon-nucleon scattering phase shifts, charge radii, spectra, and nuclear-matter properties. The key ingredients were (i) employing a regulator-consistent chiral 2N potential at N3^{3}LO and a 3N potential at NNLO, (ii) jointly calibrating 2N and 3N LECs to selected few-nucleon data and the binding energy and charge radius of 16O{}^{16}\mathrm{O}, while validating against heavier-mass calcium and neutron-rich oxygen isotopes, and (iii) using emulator-accelerated multi-start optimization algorithms to navigate the high-dimensional parameter space spanned by the LECs. Our results suggest that N3^{3}LOTexas_{\mathrm{Texas}} yields energy results that are comparable to the chiral interaction 1.8/2.0(EM), while improving radii systematics and dripline physics. Notably, N3^{3}LOTexas_{\mathrm{Texas}} predicts that the two-neutron calcium dripline is located at 71Ca{}^{71}\mathrm{Ca}, a significant extension from previous limits set by ab initio calculations based on chiral interactions, but in line

with recent experimental trends. A next step is to use N3^{3}LOTexas_{\mathrm{Texas}} as the starting point of a Bayesian analysis that accounts for effective field theory and ab initio many-body uncertainties [47, 89–91], thereby enabling posterior predictive distributions and a systematic assessment of the predictive power of consistently regulated higher-order 3N contributions [92–94]. The 2N and 3N matrix elements for the chiral N3^{3}LOTexas_{\mathrm{Texas}} interaction are available upon request and are included in the NuHamil code [95], as well as provided in Ref. [96].

Acknowledgments

We thank Ragnar Stroberg and Bingcheng He for the imsrg++ code [97] used to perform the IMSRG(2) and IMSRG(3f2) calculations. This work was supported by the U.S. Department of Energy, Office of Science, under grants Nos. DE-FG02-97ER41014, DE-FG02-96ER40963, DE-SC0024465, and DE-SC0026198 (STREAMLINE collaboration), by the Swedish Research Council (Grants No. 2020-05127, No. 2021-04507, and No. 2024-04681), by the Alexander von Humboldt Foundation, by the Cyclotron Institute at Texas A&M University, by JST ER-ATO Grant No. JPMJER2304, Japan, and by JSPS KAKENHI Grant Numbers 25K07294, 25K00995, and 25K07330. This research used resources from the Oak Ridge Leadership Computing Facility located at Oak Ridge National Laboratory, which is supported by the Office of Science of the U.S. Department of Energy under contract No. DE-AC05-00OR22725. Advanced computing resources are also provided by Texas A&M High Performance Research Computing.

† [email protected]

References

End Matter

The N3^{3}LOTexas_{\mathrm{Texas}} interaction comprises a chiral 2N potential up to N3^{3}LO and a 3N potential up to NNLO. For the 2N interactions we follow the formalism presented in Refs. [43, 44]. In this approach, the two-pion exchange bubble diagram proportional to ci2c_i^2 that appears at N3^{3}LO is very attractive, while the N4^{4}LO two-pion exchange correction proportional to ci/MNc_i/M_N, is large and repulsive. Therefore, as in [43, 44], we promote these corrections to N3^{3}LO and arrive at a more realistic intermediate attraction at this order. For the subleading two-pion exchange we also use spectral function regularization [107] with a cutoff 700 MeV. We employ a standard nonlocal regulator function f(p)=exp⁡(−(p/Λ)2n)f(p)=\exp(-(p/\Lambda)^{2n}) with cutoff Λ=394\Lambda=394 MeV and power n=4n=4 consistently for both the 2N and 3N potentials. This low-momentum cutoff provides favorable convergence with respect to model space size, enabling many-body calculations of heavy-mass nuclei without further softening the interaction via similarity renormalization group methods. The momentum cutoff and regulator power nn are the same as those used for the 3N potential in the 1.8/2.0(EM) interaction. However, unlike the 2N potential in the 1.8/2.0(EM), we employ n=4n=4 consistently across all partial waves.

For the pion-nucleon LECs we employ the central values obtained in the Roy-Steiner sub-threshold analysis of pion-nucleon scattering [108, 109], see also Table S1. The statistical uncertainties of these LEC values are exceedingly small, typically at the percent level or below, with the exception of dˉ5\bar{d}_5, which is also the numerically smallest LEC. We therefore treat the pion-nucleon LECs as fixed in the optimization of the contact LECs in the 2N and 3N potentials.

LECValueLECValue
π\piN
c1c_1−1.07-1.07c2c_23.203.20
c3c_3−5.32-5.32c4c_43.563.56
dˉ1+dˉ2\bar{d}_1+\bar{d}_21.041.04dˉ3\bar{d}_3−0.48-0.48
dˉ5\bar{d}_50.140.14dˉ14−dˉ15\bar{d}_{14}-\bar{d}_{15}−1.90-1.90
2N (LO)
C~1S0pp\widetilde{C}^{pp}_{1S_0}−0.154291-0.154291C~1S0np\widetilde{C}^{np}_{1S_0}−0.155631-0.155631
C~1S0nn\widetilde{C}^{nn}_{1S_0}−0.155010-0.155010C~3S1\widetilde{C}_{3S_1}−0.202464-0.202464
2N (NLO)
C1S0C_{1S_0}2.5142732.514273C3P0C_{3P_0}1.0254131.025413
C1P1C_{1P_1}0.0827960.082796C3P1C_{3P_1}−0.977112-0.977112
C3S1C_{3S_1}1.0755981.075598C3S1−3D1C_{3S_1-3D_1}0.5284900.528490
C3P2C_{3P_2}−0.960907-0.960907
2N (N3^3LO)
D^1S0\widehat{D}_{1S_0}−1.643852-1.643852D1S0D_{1S_0}−19.286790-19.286790
D3P0D_{3P_0}5.6076925.607692D1P1D_{1P_1}10.52934410.529344
D3P1D_{3P_1}5.4273485.427348D^3S1\widehat{D}_{3S_1}5.7151135.715113
D3S1D_{3S_1}−40.129215-40.129215D3D1D_{3D_1}−4.622881-4.622881
D^3S1−3D1\widehat{D}_{3S_1-3D_1}2.8521852.852185D3S1−3D1D_{3S_1-3D_1}0.5059380.505938
D1D2D_{1D_2}−2.369999-2.369999D3D2D_{3D_2}−4.984577-4.984577
D3P2D_{3P_2}6.76355406.7635540D3P2−3F2D_{3P_2-3F_2}0.1880850.188085
D3D3D_{3D_3}−1.196276-1.196276
3N (N2^2LO)
cDc_D−5.060563-5.060563cEc_E−1.034629-1.034629

Table S1. Low-energy constants (LECs) for the N3^3LOTexas_{\mathrm{Texas}} interaction. The constants cic_i, dˉi\bar{d}_i, C~i\widetilde{C}_i, CiC_i, and DiD_i (also D^i\widehat{D}_i) are in units of GeV−1^{-1}, GeV−2^{-2}, 10410^4 GeV−2^{-2}, 10410^4 GeV−4^{-4}, and 10410^4 GeV−6^{-6}, respectively. The 3N LECs cDc_D and cEc_E are dimensionless.

The 2N contact potential at N3^{3}LO is parameterized by 26 LECs, and the 3N contacts at NNLO are parameterized by the LECs cDc_D and cEc_E. In this work, we simultaneously optimize the values of these 28 LECs using the following data: low-energy neutron-proton scattering phase shifts from the Granada analysis up to 200 MeV scattering energy in the laboratory system [98]; the 1S0^1S_0 scattering lengths aa and effective ranges rr in the neutron-neutron (nnnn), neutron-proton (npnp) and proton-proton (pppp) sectors; the deuteron quadrupole moment Q(2H)Q(^2\mathrm{H}); the ground-state energies (EgsE_{\mathrm{gs}}) and charge radii (RchR_{\mathrm{ch}}) of 2H^2\mathrm{H}, 4He^4\mathrm{He}, and 16O^{16}\mathrm{O}. We employed reduced-basis methods [48] to develop high-fidelity emulators for all these observables, see the Supplemental Material for further details. For the bound-state observables we use an L1 objective function, while for the scattering phase shifts we use an L2 objective function. The former is more tolerant to outliers. We assign relative weights to the calibration data in the objective function to balance impact of the different kinds of data and to ensure that the calibration is not dominated by the nucleon–nucleon

Figure 5. Neutron-proton scattering phase shifts and mixing angles as functions of relative momenta, compared with the Granada phase shift analysis [98] and results from the 1.8/2.0(EM) [99, 100], Δ\DeltaNNLOGO(394) [37] and N3^{3}LOLEFT [101] interactions.

scattering phase shifts. These weights were selected as follows: Weight 50 for the deuteron energy, radius, and quadrupole moment; 15 (45) for the 4^{4}He energy (radius); 10 (40) for the 16^{16}O energy (radius); 60 for the nnnn and npnp effective range parameters; and 40 for the pppp parameters. For the phase-shift data, we use weights of 100 for the 3S1^{3}S_{1} and 3D1^{3}D_{1} channels, 100/140 for ϵ1\epsilon_{1}; 9 and 0.2 for the 3P2^{3}P_{2} and ϵ2\epsilon_{2}; and 0.4 for the 3D3^{3}D_{3} channel. A greater number implies heavier weight in the objective function.

We face a high-dimensional and non-convex optimization [110] and therefore employ a multi-start optimization approach with 2000 initial values for the LECs randomly selected using a space-filling Latin hypercube design within conservative intervals for the LEC values, see Tab. S2. For each starting point, we find an optimum using a constrained Sequential Least Squares Programming algorithm with approximate derivatives [111, 112]. The optimization also incorporates constraints that allow for up to 1% charge symmetry breaking (CSB) or charge independence breaking in the leading-order LECs C~1S0np\widetilde{C}^{np}_{^{1}S_{0}}, C~1S0nn\widetilde{C}^{nn}_{^{1}S_{0}}, C~1S0pp\widetilde{C}^{pp}_{^{1}S_{0}}. The latter LEC is tuned separately after the optimization to empirical pppp effective range parameters, including Coulomb effects.

We retain the 20 candidate interactions with the lowest objective values, all of which provide an excellent description of all calibration data (see Figure S5 of the Supplemental Material). Finally, we perform a validation step using ab initio many-body calculations for the ground-state energies of 22,24^{22,24}O and 40,48^{40,48}Ca. The interaction providing the best overall agreement with these validation observables is denoted N3^{3}LOTexas.

Figure S5. Ground-state energies of 4He^{4}\mathrm{He}, 16,22,24O^{16,22,24}\mathrm{O} and 40,48Ca^{40,48}\mathrm{Ca} computed by CCSDT-3 with 20 candidate interactions. The top three performers are N3LOTexasN^{3}\mathrm{LO}_{\mathrm{Texas}}, and samples number 494 and 523.

Figure 5 shows the npnp phase shifts computed with the N3^{3}LOTexas compared with the Granada partial wave analysis [98]. We also display the results from the 1.8/2.0(EM) [99, 100], Δ\DeltaNNLOGO [37] and N3^{3}LOLEFT [101] interactions. Note that the 2N sector of 1.8/2.0(EM) interaction produces the same phase shifts as the N3^{3}LOEM [113] 2N interaction because it is obtained by SRG evolution that leaves 2N observables invariant. As shown in Fig. 5, the N3^{3}LOEM gives the best description of phase shifts, but this interaction systematically underestimates charge radii. The N3^{3}LOTexas and N3^{3}LOLEFT interactions exhibit a similar level of accuracy in describing the phase shifts and can also describe charge radii well.

Table 1 lists the observables used during the optimization and validation steps as well as selected predictions.

N3^{3}LOTexas_{\mathrm{Texas}}Exp.
appCa_{pp}^{C}−7.820−7.8196(26)a^{a}
rppCr_{pp}^{C}2.7572.790(14)a^{a}
anna_{nn}−18.950−18.95(40)
rnnr_{nn}2.7942.75(11)
anpa_{np}−23.741−23.740(20)
rnpr_{np}2.6812.77(5)
Egs(2H)E_{\mathrm{gs}}(^2\mathrm{H})−2.225−2.224575(9)
Rch(2H)R_{\mathrm{ch}}(^2\mathrm{H})2.1312.1421(88)
Q(2H)Q(^2\mathrm{H})0.270.27b^{b}
PD(2H)P_D(^2\mathrm{H})2.68%–
Egs(4He)E_{\mathrm{gs}}(^4\mathrm{He})−28.61−28.2957
Rch(4He)R_{\mathrm{ch}}(^4\mathrm{He})1.6751.6775(28)
Egs(16O)E_{\mathrm{gs}}(^{16}\mathrm{O})−126.54−127.6193
Rch(16O)R_{\mathrm{ch}}(^{16}\mathrm{O})2.7072.6991(52)
Egs(22O)E_{\mathrm{gs}}(^{22}\mathrm{O})−162.06−162.0272(572)
Egs(24O)E_{\mathrm{gs}}(^{24}\mathrm{O})−168.77−168.9525(1680)
Egs(40Ca)E_{\mathrm{gs}}(^{40}\mathrm{Ca})−339.15−342.0522
Egs(48Ca)E_{\mathrm{gs}}(^{48}\mathrm{Ca})−417.63−416.0012
Egs(3H)E_{\mathrm{gs}}(^3\mathrm{H})−8.50−8.4818
Rch(3H)R_{\mathrm{ch}}(^3\mathrm{H})1.7691.7591(363)
Egs(3He)E_{\mathrm{gs}}(^3\mathrm{He})−7.76−7.7180
Rch(3He)R_{\mathrm{ch}}(^3\mathrm{He})1.9671.9661(30)
E2+(22O)E_2^+(^{22}\mathrm{O})3.28(7)3.199
E2+(24O)E_2^+(^{24}\mathrm{O})4.05(11)4.760
Egs(78Ni)E_{\mathrm{gs}}(^{78}\mathrm{Ni})−647(2)−642.5640(3900)c^{c}
Rch(78Ni)R_{\mathrm{ch}}(^{78}\mathrm{Ni})3.961–
Egs(132Sn)E_{\mathrm{gs}}(^{132}\mathrm{Sn})−1102(5)−1102.8432(19)
Rch(132Sn)R_{\mathrm{ch}}(^{132}\mathrm{Sn})4.7234.7093(76)
Egs(208Pb)E_{\mathrm{gs}}(^{208}\mathrm{Pb})−1613(21)−1636.4302(12)
Rch(208Pb)R_{\mathrm{ch}}(^{208}\mathrm{Pb})5.5305.5012(13)

Table 1. 1S0^1S_0 effective range parameters (a,ra,r) in fm, ground-state energies (EgsE_{\mathrm{gs}}) in MeV, charge radii (RchR_{\mathrm{ch}}) in fm, and quadrupole moments (QQ) in fm2^2 used to optimize the LECs. Results based on J-NCSM for A=2,3,4A=2,3,4, EOM-CCSDT-3 for 22,24^{22,24}O, and IMSRG(3f2) for 78^{78}Ni, 132^{132}Sn and 208^{208}Pb. Results from the validation step are computed with CCSDT-3. The uncertainties in the EOM-CCSDT-3 computations stem from the truncated model space. Experiment from [12] [13, 44, 83, 102, 103, 104, 105].

Supplemental Material

The dripline in fluorine isotopes

Figure S1 shows that the N3^{3}LOTexas_{\mathrm{Texas}} interaction accurately reproduces the dripline in fluorine isotopes. The top panel shows ground-state energies while the bottom panel shows results for the S1nS_{1n} and S2nS_{2n}. The thin red band for VS-IMSRG with N3^{3}LOTexas_{\mathrm{Texas}} indicates the estimated many-body uncertainty, quantified as the EgsE_{\mathrm{gs}} difference between “IMSRG(3f2)+Triple” and “IMSRG(2).” The gray band for the 1.8/2.0(EM) shows the uncertainty given in Ref. [5] arising from different decoupled valence spaces.

Figure S1. Ground-state energies (EgsE_{\mathrm{gs}}), one-neutron (S1nS_{1n}) and two-neutron (S2nS_{2n}) separation energies for fluorine isotopes computed with the interaction N3^{3}LOTexas_{\mathrm{Texas}} and the valence-space in-medium similarity renormalization group (VS-IMSRG) method, compared with 1.8/2.0(EM) results [5] and data [12] [13]. Stars (open) are masses (extrapolations) from AME2020 [13]. The red and gray bands indicate many-body uncertainty and different decoupled valence spaces, respectively.

Dripline sensitivity to the oscillator spacing ℏω\hbar\omega

Figure S2 shows the predicted dripline in calcium isotopes for two different values of ℏω\hbar\omega. The 2n dripline is robustly placed at 70^{70}Ca for both values.

Figure S2. One-neutron (S1nS_{1n}) and two-neutron (S2nS_{2n}) separation energies for calcium isotopes computed with different ℏω\hbar\omega.

Excited states in 50,52,54,56,58^{50,52,54,56,58}Ca

The N3^{3}LOTexas_{\mathrm{Texas}} interaction not only accurately reproduces bulk properties of the calcium isotopes but also yields realistic spectra. Figure S3 shows the low-lying excitation energies of 50,52,54,56,58^{50,52,54,56,58}Ca computed with the VS-IMSRG. As in the main text, we decouple the multishell valence-space Hamiltonian within the 48^{48}Ca core and a neutron 1p3/21p_{3/2}, 1p1/21p_{1/2}, 0f5/20f_{5/2}, 0g9/20g_{9/2}, 1d5/21d_{5/2}, 2s1/22s_{1/2} valence space from an initial model space within emax⁡=14e_{\max}=14 and ℏω=12\hbar\omega=12 MeV. We see that N3^{3}LOTexas_{\mathrm{Texas}} achieves accuracy comparable to widely used 1.8/2.0(EM) interaction for these low-lying states.

Figure S3. Excitation energies of 50,52,54,56,58Ca^{50,52,54,56,58}\mathrm{Ca} computed by VS-IMSRG and large-scale shell model (LSSM) with GXPF1A interaction [77]. Experimental data are from Ref. [12].

Nuclear matter

Figure S4 presents the comparison of different coupled-cluster approximations for nuclear matter. The results reported in the present work, shown by the red curves, are obtained with the more accurate Λ\Lambda-CCD(T) method [74], which provides an improved treatment of triple correlations relative to CCD(T) [88]. The figure is included to clarify the quantitative differences among the various coupled-cluster approximations for both symmetric nuclear matter and pure neutron matter.

Figure S4. Energy per nucleon of symmetric nuclear matter (top panel) and pure neutron matter (bottom panel) calculated using different coupled-cluster approximations.

Validation step

Figure S5 shows ground-state energies for various nuclei computed with the CCSDT-3 method and the 20 candidate interactions that survived the first optimization sweep. We highlight the top three performers.

Cross validation of emulators used in this work

We constructed a fast and accurate emulator for scattering phase shifts by combining Newton’s variational method with ideas from eigenvector continuation [49]. The absolute residuals relative to the exact solutions are less than 10−810^{-8} degrees across all partial waves, demonstrating excellent agreement with exact solutions. Using these emulators, we also computed the scattering length and effective range via effective range expansion.

For the observables in 2H{}^{2}\mathrm{H} and 4He{}^{4}\mathrm{He}, we developed emulators using the eigenvector continuation framework as presented in [50], and employed 45 and 50 eigenvectors, respectively, obtained from large-space full configuration interaction calculations in a Jacobi basis. As expected, these model-driven emulators are very accurate and precise. Using 200 randomly sampled sets of LEC values, we find at most a 1% absolute discrepancy with respect to exact calculations of the 4He{}^{4}\mathrm{He} ground-state energy and radius in the relevant ranges [−40,−20][-40,-20] MeV and [1.6,3.3][1.6,3.3] fm. For 2H{}^{2}\mathrm{H} the absolute discrepancies are below 10−6%10^{-6}\%.

We also constructed a high-precision emulator for the 16O{}^{16}\mathrm{O} ground-state energy and charge radius using the sub-space projected coupled-cluster (SPCC) method [52]. To achieve higher precision in the emulator, we extended the SP-CCSD approximation used in Ref. [52] by including leading-order triples excitations via the CCSDT-3 method [72]. Our training vectors were generated from full CCSDT-3 calculations in a model space of 13 major harmonic oscillator shells with oscillator frequency ℏω=16\hbar\omega= 16 MeV and three-body truncation of E3max⁡=16ℏωE_{3\max} = 16\hbar\omega. We selected 84 different training points in the 28-dimensional space of LECs using a space-filling Latin hypercube design with a 20% variation around a roughly optimized LEC set. Recognizing the large uncertainties associated with the three-body cDc_D and cEc_E, we generated samples within cD∈[−5,5]c_D \in[-5,5] and cE∈[−3,3]c_E \in[-3,3]. For each training point, we performed a full CCSDT-3 calculation to obtain the training vectors, for which we then constructed the sub-space projected norm and Hamiltonian matrices. Once these matrices are constructed, we can obtain the ground-state energy and charge radii for any target values of the LECs by diagonalizing a 84×84 generalized eigenvalue problem. We quantified the accuracy of the emulator by cross-validation against 200 exact CCSDT-3 calculations of the ground-state energy and point-proton radius, and found the relative error is smaller than 1% in both observables, as shown also in Figure S6. We also checked that the SPCC emulator extrapolates well outside the training region, and at the N3LOTexasN^3\mathrm{LO}_{\mathrm{Texas}} point it deviates by 0.7 MeV in binding energy and 0.005 fm for the charge radius.

Figure S6. Cross-validation of emulator for the ground-state energy (left) and proton-point radius (right) using 200 exact CCSDT-3 calculations. Only energies between −170-170 MeV and −90-90 MeV are shown.

LECMinMaxLECMinMax
C~1S0pp\widetilde{C}^{pp}_{1S_0}−0.30-0.30−0.10-0.10C~1S0np\widetilde{C}^{np}_{1S_0}−0.30-0.30−0.10-0.10
C~1S0nn\widetilde{C}^{nn}_{1S_0}−0.30-0.30−0.10-0.10C~3S1\widetilde{C}_{3S_1}−0.30-0.30−0.10-0.10
C1S0C_{1S_0}2.302.302.902.90C3P0C_{3P_0}0.800.801.401.40
C1P1C_{1P_1}−1.00-1.001.001.00C3P1C_{3P_1}−1.40-1.401.501.50
C3S1C_{3S_1}0.100.101.701.70C3S1−3D1C_{3S_1-3D_1}0.000.001.201.20
C3P2C_{3P_2}−2.10-2.102.402.40D^1S0\widehat{D}_{1S_0}−2.20-2.204.504.50
D1S0D_{1S_0}−28.00-28.00−18.00-18.00D3P0D_{3P_0}3.903.906.406.40
D1P1D_{1P_1}8.508.5018.0018.00D3P1D_{3P_1}2.002.0010.5010.50
D^3S1\widehat{D}_{3S_1}−9.70-9.709.309.30D3S1D_{3S_1}−40.60-40.6016.7016.70
D3D1D_{3D_1}−8.20-8.204.504.50D^3S1−3D1\widehat{D}_{3S_1-3D_1}−10.00-10.009.309.30
D3S1−3D1D_{3S_1-3D_1}−7.40-7.4010.0010.00D1D2D_{1D_2}−2.60-2.601.101.10
D3D2D_{3D_2}−8.00-8.00−2.00-2.00D3P2D_{3P_2}−3.00-3.0015.3015.30
D3P2−3F2D_{3P_2-3F_2}−2.60-2.602.902.90D3D3D_{3D_3}−3.00-3.001.001.00
cDc_D−8.00-8.008.008.00cEc_E−8.00-8.008.008.00

Table S2. Initial search domains for the LEC values, where C~i\widetilde{C}_i, CiC_i, and DiD_i (also D^i\widehat{D}_i) are in units of 10410^4 GeV−2^{-2}, 10410^4 GeV−4^{-4}, and 10410^4 GeV−6^{-6}, respectively. The 3N LECs cDc_D and cEc_E are dimensionless.

References

  1. [2]Nuclear landscape in covariant density functional the-ory, Physics Lett. B 726, 680 (2013).
  2. [3]F.-K. Thielemann, M. Eichler, I. Panov, and B. Wehmeyer, Neutron star mergers and nucleosynthe-sis of heavy elements, Ann. Rev. Nucl. Part. Sci. 67, 253 (2017).DOI
  3. [4]L. Neufcourt, Y. Cao, S. A. Giuliani, W. Nazarewicz, E. Olsen, and O. B. Tarasov, Quantified limits of the nuclear landscape, Phys. Rev. C 101, 044307 (2020).DOI
  4. [5]S. R. Stroberg, J. D. Holt, A. Schwenk, and J. Simonis, Ab Initio Limits of Atomic Nuclei, Phys. Rev. Lett. 126, 022501 (2021).DOI
  5. [6]Z. Y. Zhang, Z. G. Gan, H. B. Yang, L. Ma, M. H. Huang, C. L. Yang, M. M. Zhang, Y. L. Tian, Y. S. Wang, M. D. Sun, H. Y. Lu, W. Q. Zhang, H. B. Zhou, X. Wang, C. G. Wu, L. M. Duan, W. X. Huang, Z. Liu, Z. Z. Ren, S. G. Zhou, X. H. Zhou, H. S. Xu, Y. S. Tsyganov, A. A. Voinov, and A. N. Polyakov, New isotope 220Np: Probing the robustness of the n = 126 shell closure in neptunium, Phys. Rev. Lett. 122, 192503 (2019).
  6. [7]D. S. Ahn, N. Fukuda, H. Geissel, N. Inabe, N. Iwasa, T. Kubo, K. Kusaka, D. J. Morrissey, D. Murai, T. Nakamura, M. Ohtake, H. Otsu, H. Sato, B. M. Sherrill, Y. Shimizu, H. Suzuki, H. Takeda, O. B. Tarasov, H. Ueno, Y. Yanagisawa, and K. Yoshida, Location of the neutron dripline at fluorine and neon, Phys. Rev. Lett. 123, 212501 (2019).DOI
  7. [8]H. L. Crawford, V. Tripathi, J. M. Allmond, B. P. Crider, R. Grzywacz, S. N. Liddick, A. Andalib, E. Argo, C. Benetti, S. Bhattacharya, C. M. Campbell, M. P. Carpenter, J. Chan, A. Chester, J. Christie, B. R. Clark, I. Cox, A. A. Doetsch, J. G. Dopfer, J. G. Duarte, P. Fallon, A. F. Forster, T. Gaballah, F. J. Gray, J. T. Harke, J. Heideman, H. Heugen, R. Jain, T. T. King, N. Kitamura, K. Kolos, F. G. Kondev, A. Lamaninack, B. Longfellow, R. S. Lubna, S. Luitel, M. Madurga, R. Mahajan, M. J. Mogannam, C. Morse, S. Neupane, A. Nowicki, T. H. Ogunbeku, W.-J. Ong, C. Porzio, C. J. Prokop, B. C. Rasco, E. K. Ronning, E. Rubino, T. J. Ruland, K. P. Ryczkawski, L. Schaedig, D. Seweryniak, K. Siegl, M. Singh, S. L. Tabor, T. L. Tang, T. Wheeler, J. A. Winger, and Z. Xu, Crossing n = 28 toward the neutron drip line: First measurement of half-lives at frib, Phys. Rev. Lett. 129, 212501 (2022).DOI
  8. [9]O. B. Tarasov, A. Gade, K. Fukushima, M. Hausmann, E. Kwan, M. Portillo, M. Smith, D. S. Ahn, D. Bazin, R. Chyzh, S. Giraud, K. Haak, T. Kubo, D. J. Morrissey, P. N. Ostroumov, I. Richardson, B. M. Sherrill, A. Stolz, S. Watters, D. Weisshaar, and T. Zhang, Observation of new isotopes in the fragmentation of 198Pt at frib, Phys. Rev. Lett. 132, 072501 (2024).DOI
  9. [10]M. R. Mumpower, R. Surman, G. C. McLaughlin, and A. Aprahamian, The impact of individual nuclear properties on r-process nucleosynthesis, Prog. Part. Nucl. Phys. 86, 86 (2016).DOI
  10. [11]I. U. Roederer, N. Vassh, E. M. Holmbeck, M. R. Mumpower, R. Surman, J. J. Cowan, T. C. Beers, R. Ezzeddine, A. Frebel, T. T. Hansen, V. M. Placco, and C. M. Sakari, Element abundance patterns in stars indicate fission of nuclei heavier than uranium, Science 382, 1177 (2023), https://www.science.org/doi/pdf/10.1126/science.adf1341.DOI
  11. [12]Evaluated Nuclear Structure Data File (ENSDF) (2023).

Paper details

vlnc:W_b7p7rtaz
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents