betaFeedback

WorksPhysics and Astronomy

Sign-Problem-Free Nuclear Quantum Monte Carlo Simulation

Abstract

Quantum Monte Carlo (QMC) methods offer exact solutions for quantum many-body systems but face severe limitations in fermionic systems like atomic nuclei due to the sign problem. While sign-problem-free QMC algorithms exist and provide valuable insights across disciplines, they have been restricted to simple models with limited quantitative predictive power. Here we overcome this barrier by developing a novel lattice nuclear force that is rigorously sign-problem-free for even-even nuclei. This interaction achieves a standard deviation of σ=2.932σ= 2.932 MeV from experimental binding energies for 76 even-even nuclei (N,Z≤28N,Z \leq 28), matching state-of-the-art phenomenological mean-field models. Key innovations include the first sign-problem-free implementation of spin-orbit coupling for shell evolutions and an efficient QMC-optimized framework for global parameter fitting. Using this approach, we compute binding energies from 4^4He to 132^{132}Sn with unprecedented one-thousandth level numerical precision, reproduce symmetric nuclear matter saturation, and reveal novel spin-orbit-driven clustering in light nuclei. This work transforms sign-problem-free QMC into a scalable and predictive nuclear structure tool, while establishing a high-fidelity, non-perturbative foundation for ab initio calculations of heavy nuclei.

Introduction

Understanding the emergence of the nuclear structure from fundamental nucleon-nucleon interactions is a central challenge in modern nuclear physics. While exact diagonalization approaches are limited by the exponential growth of Hilbert space dimensionality, quantum Monte Carlo (QMC) methods offer a promising solution by stochastically sampling the vast Hilbert space, reducing computational complexity to polynomial scaling with nucleon number. Recent years have seen numerous QMC applications to nuclear many-body problems (see [1, 2, 3, 4, 5, 6, 7] for reviews).

Despite significant successes, most QMC applications are fundamentally constrained by the fermionic sign problem [8]. This arises when integrating highly oscillatory functions, where cancellations between positive and negative amplitudes lead to exponentially large statistical noise scaling with particle number. Although mitigation techniques exist, e.g., fixed-node [9, 10, 11]/constrained path approximations [12, 13, 14], the Lefschetz thimble method [15, 16], and perturbative expansions [17, 18, 19, 20], their implementations present significant numerical challenges and often introduce systematic biases.

Notably, for certain symmetry-protected fermionic systems, QMC calculations are immune to the sign problem. In these rare cases, QMC delivers unbiased results with high numerical precision and exceptional scalability. For example, in lattice quantum chromodynamics (QCD), the isospin symmetry between the up and down quarks implies that the fermionic determinant factorizes into a perfect square, guaranteeing positive amplitudes for any gauge-field configuration [21]. It was further shown that a broader class of interactions without a factorizable determinant can also be sign-problem-free, where positivity is protected by the time-reversal symmetry of the fermion matrix [22, 23]. See Ref. [24] and references therein for sign-problem-free QMC applied to phenomenological condensed matter models.

In nuclear physics, sign-problem-free QMC algorithms are highly desirable but remain underutilized, particularly for nuclear ab initio calculations. Such calculations typically employ complex, high-fidelity nuclear forces derived from fundamental symmetries, which inevitably induce sign problems [25, 26] [27, 28, 29]. Crucially, any non-zero sign problem, even negligible in light nuclei, grows exponentially with particle number, preventing direct QMC simulations of heavy nuclei. Remedies like perturbation theory exist [17, 18] but still require sign-problem-free interactions as the non-perturbative foundation. Consequently, developing sign-problem-free QMC methods would profoundly deepen our first-principles understanding of heavy nuclei. Currently, a prominent sign-problem-free nuclear force is the spin-isospin independent Wigner-SU(4) interaction [30–33], which successfully reproduces light-nuclei binding energies and charge radii, describes emergent geometry in 12C{}^{12}\mathrm{C} [34], clustering in hot nuclear matter [35, 36], Beryllium isotope spectra [37], and α\alpha-particle monopole transitions [38]. These significant achievements strongly motivate the search for more accurate sign-problem-free interactions. The core challenge, however, is that the time-reversal principle enabling sign-problem-free simulations inherently forbids essential interactions such as the tensor force and most spin/isospin-dependent terms [39–42], severely constraining the accessible model space.

In this Letter, we tackle this challenge by constructing a novel minimal nuclear force that is simple enough to avoid the Monte Carlo sign problem, yet sufficiently sophisticated to reproduce key experimental data. To satisfy these competing requirements, we introduce a sign-problem-free spin-orbit interaction in addition to the Wigner-SU(4) force to capture shell evolution, and optimize its parameters against selected nuclear binding energies using an efficient derivative-based method. While conceptually inspired by phenomenological mean-field force construction [43–45], crucially, our approach yields exact solutions to the Hamiltonian.

Method We formulate the QMC algorithm within the framework of the ab initio approach of nuclear lattice effective field theory (NLEFT), where the nuclear force models are discretized and numerically solved on a three-dimensional cubic lattice using auxiliary field Monte Carlo methods. Recent advances in NLEFT have yielded significant achievements across multiple domains, e.g., nuclear ground states [46–49] and excited states [50], intrinsic density distributions and clustering phenomena [51–55], nucleus-nucleus scattering dynamics [56, 57], and hypernucleus [58–61].

The minimal nuclear force employed here comprises both two- and three-body interactions supplemented by a spin-orbit coupling term. The Hamiltonian is defined on a L3L^3 cubic lattice with integer coordinates n=(nx,ny,nz)\boldsymbol{n}=(n_x,n_y,n_z),

H=∑n[−Ψ†∇2Ψ2M+:C22ρˉ 2+C36ρˉ 3+Csρˉρˉs:],(1)H=\sum_{\boldsymbol{n}}\left[-\frac{\Psi^\dagger\nabla^2\Psi}{2M}+:\frac{C_2}{2}\bar{\rho}^{\,2}+\frac{C_3}{6}\bar{\rho}^{\,3}+C_s\bar{\rho}\bar{\rho}_s:\right], \tag*{(1)}

where the summation spans all lattice sites and the colons mean the normal ordering. Here Ψ(n)\Psi(\boldsymbol{n}) and Ψ†(n)\Psi^\dagger(\boldsymbol{n}) denote annihilation and creation operators, M=938.92M=938.92 MeV is the nucleon mass, ∇2\nabla^2 is the Laplace operator implemented via fast Fourier transform (FFT), and C2C_2, C3C_3, CsC_s are coupling constants. We fix the lattice spacing to a=1.32a=1.32 fm and use natural units ℏ=c=1\hbar=c=1 throughout. We use symbols with overline to denote smeared quantities,

σˉ(n)=σ(n)+sL∑∣n′−n∣=1σ(n′),σ=ρ,ρs,(2)\bar{\sigma}(\boldsymbol{n})=\sigma(\boldsymbol{n})+s_{\mathrm{L}}\sum_{|\boldsymbol{n}'-\boldsymbol{n}|=1}\sigma(\boldsymbol{n}'),\qquad\sigma=\rho,\rho_s, \tag*{(2)}

with sLs_{\mathrm{L}} a real number controlling local smearing. The densities ρ\rho and ρs\rho_s are defined as

ρ=Ψˉ†Ψˉ,\rho=\bar{\Psi}^{\dagger}\bar{\Psi},
ρs=i4∑ijkϵijk∇i[Ψˉ†(∇→j−∇←j)σkΨˉ],(3)\rho_s=\frac{i}{4}\sum_{ijk}\epsilon_{ijk}\nabla_i\left[\bar{\Psi}^{\dagger}(\overrightarrow{\nabla}_j-\overleftarrow{\nabla}_j)\sigma_k\bar{\Psi}\right], \tag*{(3)}

where ϵijk\epsilon_{ijk} is the Levi-Civita symbol, ∇\nabla represents the spatial gradients implemented with the FFT, and σk\sigma_k is the spin Pauli matrices. The smeared operators Ψˉ\bar{\Psi} and Ψˉ†\bar{\Psi}^{\dagger} are defined as,

σˉ(n)=σ(n)+sNL∑∣n′−n∣=1σ(n′),σ=Ψ,Ψ†,(4)\bar{\sigma}(\boldsymbol{n})=\sigma(\boldsymbol{n})+s_{\mathrm{NL}}\sum_{|\boldsymbol{n}'-\boldsymbol{n}|=1}\sigma(\boldsymbol{n}'),\qquad\sigma=\Psi,\Psi^\dagger, \tag*{(4)}

with sNLs_{\mathrm{NL}} a real number controlling non-local smearing. Implicit spin/isospin indices in Ψ(n)\Psi(\boldsymbol{n}), Ψ†(n)\Psi^\dagger(\boldsymbol{n}), and σk\sigma_k are summed. Both local (Eq. (2)) and nonlocal (Eq. (4)) smearing govern the effective interaction range. The relative strength between sLs_{\mathrm{L}} and sNLs_{\mathrm{NL}} determines the locality, an essential element for nuclear binding. In particular, the heavy nuclei dissociate into small clusters (A≤4A\leq4) at the non-local limit (sL=0s_{\mathrm{L}}=0) and exhibit severe overbinding at the local limit (sNL=0s_{\mathrm{NL}}=0) [32]. The three-body force (C3C_3) significantly improves the light nuclei structure [30]. The newly introduced spin-orbit term (CsC_s) generates requisite shell structure evolution. For Cs=0C_s=0 we return to the Wigner-SU(4) interaction thoroughly studied in Ref. [30]. We also incorporate a perturbative Coulomb force for protons [62].

We solve the Hamiltonian Eq. (1) using the imaginary time projection method,

∣Φ⟩∝lim⁡τ→∞exp⁡(−τH)∣ΦT⟩,(5)|\Phi\rangle\propto\lim_{\tau\to\infty}\exp(-\tau H)|\Phi_T\rangle, \tag*{(5)}

where ∣Φ⟩|\Phi\rangle is the ground state and ∣Φ⟩T|\Phi\rangle_T is a trial wave function with non-zero overlap. For sufficiently large τ\tau, excited state contributions are exponentially suppressed. Expectation values are computed through numerical extrapolation

⟨O⟩=lim⁡τ→∞⟨ΦT∣e−τH/2Oe−τH/2∣ΦT⟩⟨ΦT∣e−τH∣ΦT⟩.(6)\langle O\rangle=\lim_{\tau\to\infty}\frac{\langle\Phi_T|e^{-\tau H/2}Oe^{-\tau H/2}|\Phi_T\rangle}{\langle\Phi_T|e^{-\tau H}|\Phi_T\rangle}. \tag*{(6)}

In NLEFT we implement Eq. (6) via auxiliary field transformation. For the two-body forces,

e−atH=∫Dcexp⁡[−∑nc(n)22+at2MΨ†∇2Ψ+−atC2∑nc(n)(ρˉ(n)+Cs2C2ρˉs(n))].(7)e^{-a_tH}=\int\mathcal{D}c\exp\left[-\sum_{\boldsymbol{n}}\frac{c(\boldsymbol{n})^2}{2}+\frac{a_t}{2M}\Psi^\dagger\nabla^2\Psi+\sqrt{-a_tC_2}\sum_{\boldsymbol{n}}c(\boldsymbol{n})\left(\bar{\rho}(\boldsymbol{n})+\frac{C_s}{2C_2}\bar{\rho}_s(\boldsymbol{n})\right)\right]. \tag*{(7)}

where ata_t is the temporal step and c(n)c(\boldsymbol{n}) is a real auxiliary field. The terms of order O(C22)O(C_2^2) generated from this transformation are negligible (Transformation involving 3NF follows [62]). The cc fields at each imaginary time step are collectively integrated using the importance sampling Monte Carlo. Without loss of generality, we consider even-even nuclei and take the trial wave function ∣ΦT⟩|\Phi_T\rangle as a Slater determinant of nucleon wave functions paired by time-reversal symmetry T=iσyK\mathcal{T}=i\sigma_y\mathcal{K}, where K\mathcal{K} is the complex conjugate. These nucleons interact with the cc-field and propagate independently under Eq. (7). As the single-particle Hamiltonian in Eq. (7) commute with T\mathcal{T}, the nucleons remain paired during the imaginary-time evolution. The resulting determinant of the fermionic correlation matrix is positively definite, ensuring completely sign-problem-free simulations (See [62] for proof). For odd-even and odd-odd nuclei, the nucleons can not be completely paired up and a mild sign problem remains.

The interaction in Eq. (1) contains five parameters C2C_2, C3C_3, CsC_s, sLs_{\mathrm{L}}, and sNLs_{\mathrm{NL}}. We found that the mass prediction is not sensitive to sNLs_{\mathrm{NL}} and fix sNL=0.45s_{\mathrm{NL}}=0.45 throughout. The remaining four parameters are optimized against experimental binding energies of 4He{}^{4}\mathrm{He}, 16O{}^{16}\mathrm{O}, 24Mg{}^{24}\mathrm{Mg}, 28Si{}^{28}\mathrm{Si}, 32S{}^{32}\mathrm{S} and 40Ca{}^{40}\mathrm{Ca}. Here we select three doubly-magic nuclei and three open-shell nuclei to constrain both the bulk properties and the shell evolution. The optimization minimize the loss function

χ2=∑A[E(A)−Eexp(A)Δ(A)]2,(8)\chi^2=\sum_A\left[\frac{E(A)-E_{\mathrm{exp}}(A)}{\Delta(A)}\right]^2, \tag*{(8)}

with weights Δ(A)=1\Delta(A)=1 MeV for 4He{}^{4}\mathrm{He} and 3 MeV for other nuclei. To apply the steepest descent method, we need to calculate the partial derivatives of χ2\chi^2 against the fitting parameters, which in turn depend on the derivatives of the binding energies E(A)E(A). The latter was computed using the Feynman-Hellmann theorem,

∂E(A)∂x=⟨Φ∣H(x+δ)−H(x)∣Φ⟩/δ,(9)\frac{\partial E(A)}{\partial x}=\langle\Phi|H(x+\delta)-H(x)|\Phi\rangle/\delta, \tag*{(9)}

where xx is one of the fitting parameters and δ\delta is a small number chosen for convenience. The expectation values in Eq. (9) are measured concurrently with other observables in Monte Carlo simulations. Crucially, our sign-problem-free QMC enables precise, unbiased estimation of all derivatives simultaneously, facilitating rapid convergence with adaptive learning rate. The optimization with a reasonable initial guess typically converges within ten iterations. Among multiple local minima, we identify the globally optimal values C2=−4.410×10−7 MeV−2C_2=-4.410\times10^{-7}\ \mathrm{MeV}^{-2}, C3=1.561×10−15 MeV−5C_3=1.561\times10^{-15}\ \mathrm{MeV}^{-5}, Cs=8.590×10−12 MeV−4C_s=8.590\times10^{-12}\ \mathrm{MeV}^{-4} and sL=8.082×10−2s_L=8.082\times10^{-2}. In what follows we denote this group of parameters as LAT-OPT1. Our strategy, distinct from derivative-free methods used in recent chiral force optimizations against finite nuclei [63, 64], exhibits an efficiency independent of parameter space dimensionality due to simultaneous derivative measurement.

Image not extracted

Figure 1. Ground state energies from LAT-OPT1 (Errors smaller than the symbols). Red circles enclose the target nuclei for calibrating the interaction. Crosses denote experimental values [65]. Isotopic chains are connected to guide the eyes. See [62] for numerical results.

Results

Next we present numerical results calculated using the parameter set LAT-OPT1. Unless otherwise specified, all calculations employ a L=11L=11 cubic lattice with periodic boundary conditions. The temporal step is set to at=(1000 MeV)−1a_t=(1000\ \mathrm{MeV})^{-1}. We perform QMC calculations using 50, 100, ⋯\cdots, 350 temporal slices and extrapolate the results to infinite imaginary time. The reported errors are combined QMC statistical uncertainties and extrapolation errors.

In Fig. 1 we show binding energies for 76 even-even nuclei with N,Z≤28N,Z\le28. The results show excellent agreement with experimental values, yielding a standard error of σ=2.932 MeV\sigma=2.932\ \mathrm{MeV}. For comparison, state-of-the-art density functional theories typically give σ≈2\sigma\approx2-3 MeV in this mass region [66–68]. Note that our model achieves this level of accuracy with only four parameters trained on six nuclei. This remarkable generalization capability can be attributed to the incorporation of full quantum correlations, with shell effect, pairing, clustering, deformation, collective rotation, and shape fluctuation all treated on an equal footing. Systematic overbinding at large neutron numbers indicates missing components in our nuclear force, such as the pion-exchange potentials dictated by the chiral dynamics [69–72].

While existing nuclear QMC approaches using high-fidelity forces remain confined to light/medium-mass systems and require error mitigation, our sign-problem-free method can be directly used to investigate heavy nuclei. In Figure 2 we plot the QMC results for 100Sn^{100}\mathrm{Sn}, extrapolated to infinite imaginary time. We find a combined QMC statistical and extrapolation error below 1 MeV. We also probe dependencies on temporal step size and box volume, finding only negligible variations. Moreover, The extrapolated binding energy reproduces the experiment within 1 MeV. These unprecedented one-thousandth level numerical precision and remarkable predictive capability demonstrate sign-problem-free QMC as an accurate and scalable approach for heavy nuclei.

Image not extracted

Figure 2. Circles and diamonds denote the 100Sn^{100}\mathrm{Sn} binding energies calculated with the temporal step at=(1000 MeV)−1a_t=(1000\ \mathrm{MeV})^{-1} and (1200 MeV)−1(1200\ \mathrm{MeV})^{-1}, respectively. Lines represent the exponential fits E(τ)=E(∞)+Ce−τΔE(\tau)=E(\infty)+Ce^{-\tau\Delta}. Rightmost symbols show extrapolated values. Dotted line denotes experimental value [65].

Table I lists calculated binding energies for representative nuclei. The relative errors against the experiments are below 3% for 4He^{4}\mathrm{He} and 132Sn^{132}\mathrm{Sn}, and below 1% for other nuclei, demonstrating high accuracy across the nuclear chart. Furthermore, to quantify the impact of the spin-orbit coupling term proportional to CsC_s, we also list its partial contributions EslE_{\mathrm{sl}}. This term accounts for 5-10% of the binding energies on average and systematically fluctuates with the number of nucleons. Notably, the fraction Esl/EbindE_{\mathrm{sl}}/E_{\mathrm{bind}} is always below 5% for nuclei with traditional harmonic oscillator magic numbers 2, 8, 20, 40 such as 4He^{4}\mathrm{He}, 16O^{16}\mathrm{O}, 40Ca^{40}\mathrm{Ca} and 80Zr^{80}\mathrm{Zr}, but significantly enhanced for nuclei with new magic numbers 28, 50, 82 such as 56Ni^{56}\mathrm{Ni}, 100Sn^{100}\mathrm{Sn} and 208Pb^{208}\mathrm{Pb}. This extra binding reflects the enhanced stability of these nuclei obtained from shell gaps generated by spin-orbit splitting, directly validating the Mayer-Jensen shell model paradigm within our framework without employing a single-particle mean field.

NucleusEbindE_{\mathrm{bind}} (MeV)EslE_{\mathrm{sl}} (MeV)Esl/EbindE_{\mathrm{sl}}/E_{\mathrm{bind}}EXP (MeV)
4He{}^{4}\mathrm{He}−28.8-28.8−0.3-0.30.0100.010−28.3-28.3
12C{}^{12}\mathrm{C}−91.3(1)-91.3(1)−13.3-13.30.1460.146−92.2-92.2
14C{}^{14}\mathrm{C}−104.6(1)-104.6(1)−12.7-12.70.1210.121−105.3-105.3
16O{}^{16}\mathrm{O}−126.9(2)-126.9(2)−5.6-5.60.0440.044−127.6-127.6
40Ca{}^{40}\mathrm{Ca}−343.0(2)-343.0(2)−13.6-13.60.0400.040−342.1-342.1
48Ca{}^{48}\mathrm{Ca}−414.5(3)-414.5(3)−42.3-42.30.1020.102−416.0-416.0
56Ni{}^{56}\mathrm{Ni}−479.3(6)-479.3(6)−74.6-74.60.1560.156−484.0-484.0
80Zr{}^{80}\mathrm{Zr}−672.1(8)-672.1(8)−23.3-23.30.0350.035−669.2-669.2
90Zr{}^{90}\mathrm{Zr}−782.1(5)-782.1(5)−64.8-64.80.0830.083−783.9-783.9
100Sn{}^{100}\mathrm{Sn}−824.7(8)-824.7(8)−103.0-103.00.1250.125−825.2-825.2
132Sn{}^{132}\mathrm{Sn}−1134.2(27)-1134.2(27)−110.9-110.90.0980.098−1102.8-1102.8

Table I. Calculated binding energies EbindE_{\mathrm{bind}}, expectation values of the spin-orbit coupling term EslE_{\mathrm{sl}}, their ratio and experimental binding energies [65] (EXP) for selected nuclei.

Our approach solve the quantum many-body problem non-perturbatively, enabling studies of strong-correlation phenomena like nuclear clustering that emerges from the intricate interplay among shell structure, quantum fluctuations and concontinuum effects [73, 74]. Figure 3 (a) shows the loss function χ2\chi^2 in Eq. (8) as a function of the spin-orbit coupling constant CsC_s, with other parameters fixed to LAT-OPT1 values. A pronounced minimum occurs precisely at the LAT-OPT1 value CsoptC_s^{\mathrm{opt}} (vertical line), confirming that our optimization robustly identifies the physical spin-orbit strength stringently constrained by target nuclei. Figure 3 (b) displays binding energies for 4He{}^{4}\mathrm{He} and light α\alpha-conjugate nuclei (8Be{ }^{8}\mathrm{Be}, 12C{}^{12}\mathrm{C} and 16O{}^{16}\mathrm{O}) relative to nn-α\alpha breakup thresholds. The 4He{}^{4}\mathrm{He} binding is independent of CsC_s, while 8Be{}^{8}\mathrm{Be} persists unbound against two-α\alpha decay. The 3-α\alpha (12C{ }^{12}\mathrm{C}) and 4-α\alpha (16O{ }^{16}\mathrm{O}) systems are more sensitive to spin-orbit coupling. In the SU(4)-symmetric limit (Cs→0C_s \to0), 16O{}^{16}\mathrm{O} is stable, but 12C{}^{12}\mathrm{C} is unbound against 3α\alpha dissociation. Conversely, at the other extreme of strong spin-orbit coupling (Cs≫CsoptC_s \gg C_s^{\mathrm{opt}}), the 3-α\alpha and 4-α\alpha binding energies coincide, indicating that 12C{}^{12}\mathrm{C} becomes deeply bound, while 16O{}^{16}\mathrm{O} turns unstable against 12C+α{}^{12}\mathrm{C}+\alpha breakup. Remarkably, only near the χ2\chi^2-optimized value CsoptC_s^{\mathrm{opt}} do we reproduce the experimental binding hierarchy. While numerical calculations demonstrate that inter-cluster effective interactions are sensitive to the strength and locality of the underlying fundamental interactions [32, 75, 76], here we further establish a direct link between spin-orbit coupling and effective α\alpha-cluster interactions. Likewise, other key components for clusterization like the tensor force can be studied similarly [77–80]. Such investigations pave the way towards a unified mechanism governing both shell and clustering phenomena in light nuclei.

Image not extracted

Figure 3. (a) Loss function (Eq. (8)) versus spin-orbit coupling CsC_s. (b) Binding energies relative to nn-α\alpha thresholds: 8Be{}^{8}\mathrm{Be} (squares), 12C{}^{12}\mathrm{C} (circles), 16O{}^{16}\mathrm{O} (diamonds). Full symbols: experimental values [65]; solid/open triangles: calculated/experimental 4He{}^{4}\mathrm{He} energies. Vertical line: optimal CsoptC_s^{\mathrm{opt}} minimizing χ2\chi^2.

Lastly, Figure 4 shows the predicted equation of state for symmetric nuclear matter. Comparison to the Wigner-SU(4) interaction [30] reveals significant improvement. We identify a saturation point at (ρ0,E/A)=(0.175 fm−3,−15.4 MeV)(\rho_0, E/A) = (0.175\ \mathrm{fm}^{-3}, -15.4\ \mathrm{MeV}), consistent with empirical values.

Image not extracted

Figure 4. Equation of state for symmetric nuclear matter calculated with LAT-OPT1 (circles) and Wigner-SU(4) [30] (diamonds) interactions. Full symbols denote corresponding saturation points. Green rectangle mark the empirical saturation point (0.16±0.01 fm−3,−16.0±1.0 MeV)(0.16 \pm0.01\ \mathrm{fm}^{-3}, -16.0 \pm1.0\ \mathrm{MeV}) [81].

Summary and perspective

This work presents the first exactly solvable quantum many-body model for atomic nuclei that quantitatively reproduces experimental data across light to heavy nuclei. Exactly solvable quantum many-body problems are rare benchmarks essential for evaluating advanced algorithms, including those based on artificial neural networks [7] or quantum computing [82, 83]. Here, exact solvability implies achieving the desired precision within polynomial time. For modeling nuclear structure, the proposed sign-problem-free algorithm offers orders of magnitude acceleration compared to calculations using realistic interactions for heavy nuclei. The approach extends directly to investigating excited-state spectra [51], density correlations [54], and finite-temperature nuclear properties [35] using advanced lattice QMC algorithms. Furthermore, it establishes a robust non-perturbative foundation for abab initio calculations with complex chiral nuclear forces, enabling perturbative incorporation of missing physics order-by-order [17] [28].

Finally, this work establishes a scalable, systematically improvable computational paradigm that bridges complex ab initio frameworks and simple phenomenological models. Recent ab initio studies have uncovered inconsistencies among few-body systems, many-body nuclei, and nuclear matter predictions [84–87], prompting the development of minimal interactions to resolve these discrepancies [30, 88, 89]. Our framework represents such a minimal interaction approach, potentially enhancing the explainability of modern ab initio nuclear theory by distilling the essential elements of nuclear binding. This advances the quest for fundamental nuclear interactions and will likely inspire future research across nuclear physics, quantum many-body methods, and related disciplines.

Acknowledgement We thank members of the Nuclear Lattice Effective Field Theory Collaboration for insightful discussions, and Dean Lee for critical reading of the manuscript. This work has been supported by NSAF No. U2330401 and National Natural Science Foundation of China with Grant No. 12275259.

Supplemental Material

This supplement presents the implementation details of the three-body force, the proof of positivity for our approach, and the treatment of Coulomb interactions. While these elements are documented elsewhere in the literature, they are included here for completeness and self-sufficiency. We further elaborate on: (i) the optimization algorithm employed, (ii) computational complexity analysis, and (iii) finite volume effects. Numerical results and original data are provided in comprehensive tables and composite figures, including nucleon-nucleon phase shifts, odd-even staggering in oxygen isotopes, the pure neutron matter equation of state, and charge density distributions.

Auxiliary field transformation for three-body force

For non-perturbative simulations of many-body forces, we employ a discrete auxiliary field that simultaneously simulates two-, three-, and four-body forces without sign oscillations. This transformation has been used in Ref. [1]:

:exp⁡(−12C2atρ2−16C3atρ3−124C4atρ4):=∑k=1Nωk:exp⁡(−C2atϕkρ):(10):\exp\left(-\frac{1}{2}C_{2}a_{t}\rho^{2}-\frac{1}{6}C_{3}a_{t}\rho^{3}-\frac{1}{24}C_{4}a_{t}\rho^{4}\right):=\sum_{k=1}^{N}\omega_{k}:\exp\left(\sqrt{-C_{2}a_{t}}\phi_{k}\rho\right): \tag*{(10)}

where ρ\rho is a general one-body density operator, and C2C_{2}, C3C_{3}, C4C_{4} are the two-, three-, and four-body coefficients, respectively. We solve for the real numbers ωk\omega_{k} and ϕk\phi_{k}. In this work, we consider only attractive two-body interactions with C2<0C_{2}<0. To avoid the sign problem, we further require ωk>0\omega_{k}>0 for all kk.

To determine ϕk\phi_{k} and ωk\omega_{k}, we expand Eq. (10) up to O(ρ4)\mathcal{O}\left(\rho^{4}\right) and compare both sides order by order. For nuclear forces where three- and four-body interactions are typically much weaker than the two-body interaction, we use the ansatz with N=3N=3:

ω1=1ϕ1(ϕ1−ϕ3),ω2=1+1ϕ1ϕ3,ω3=1ϕ3(ϕ3−ϕ1),(11)\omega_{1}=\frac{1}{\phi_{1}\left(\phi_{1}-\phi_{3}\right)},\qquad\omega_{2}=1+\frac{1}{\phi_{1}\phi_{3}},\qquad\omega_{3}=\frac{1}{\phi_{3}\left(\phi_{3}-\phi_{1}\right)}, \tag*{(11)}

where ϕ2=0\phi_{2}=0, and ϕ1\phi_{1} and ϕ3\phi_{3} are roots of the quadratic equation:

ϕ2+C3−C23atϕ−C32C23at+C4C22at−3=0.(12)\phi^{2}+\frac{C_{3}}{\sqrt{-C_{2}^{3}a_{t}}}\phi-\frac{C_{3}^{2}}{C_{2}^{3}a_{t}}+\frac{C_{4}}{C_{2}^{2}a_{t}}-3=0. \tag*{(12)}

Using Vieta’s formulas relating polynomial coefficients to sums and products of roots, we verify that Eq. (11) satisfies Eq. (10) up to O(ρ4)\mathcal{O}\left(\rho^{4}\right). For a pure two-body interaction (C3,4=0C_{3,4}=0), the solution simplifies to ϕ1=−ϕ3=3\phi_{1}=-\phi_{3}=\sqrt{3}, ϕ2=0\phi_{2}=0, ω1=ω3=1/6\omega_{1}=\omega_{3}=1/6, ω2=2/3\omega_{2}=2/3. The formalism in Eq. (11) efficiently simulates many-body forces.

In practical calculations, we take C4=0C_{4}=0 and substitute

ρ=ρˉ+Cs2C2ρˉs(13)\rho=\bar{\rho}+\frac{C_{s}}{2C_{2}}\bar{\rho}_{s} \tag*{(13)}

into Eq. (10). The coefficients ωk\omega_{k} and ϕk\phi_{k} are determined by C2C_{2}, C3C_{3} and ata_{t} at the outset. Rather than updating continuous auxiliary fields, we randomly select indices 1≤k≤31\leq k\leq3 per lattice site and apply the Metropolis algorithm to accept or reject the new index field. After sufficient warm-up, thermal equilibrium is achieved, and observables are measured by inserting operators at the central time step. Energy is obtained by directly measuring the Hamiltonian expectation value. Note that Eq. (10) generates additional interactions of order O(Cs2/C2)\mathcal{O}\left(C_{s}^{2}/C_{2}\right) or O(C3Cs/C2)\mathcal{O}\left(C_{3}C_{s}/C_{2}\right), which are highly suppressed and negligible.

Proof for the positivity of the amplitude

The sufficient condition for a positive definite correlation matrix was given in Ref. [2]. Here we present a step-by-step derivation specific to NLEFT calculations. The expectation value of any operator OO is expressed as

⟨O⟩=lim⁡τ→∞⟨ΦT∣e−τH/2Oe−τH/2∣ΦT⟩⟨ΦT∣e−τH∣ΦT⟩=lim⁡Lt→∞⟨ΦT∣MLt/2OMLt/2∣ΦT⟩⟨ΦT∣MLt∣ΦT⟩=lim⁡Lt→∞∫Dcexp⁡(−∑i=1Ltci2/2)⟨ΦT∣M(cLt)⋯M(cLt/2+1)OM(cLt/2)⋯M(c1)∣ΦT⟩∫Dcexp⁡(−∑i=1Ltci2/2)⟨ΦT∣M(cLt)⋯M(c1)∣ΦT⟩.(14)\begin{aligned} \langle O\rangle=\lim_{\tau\to\infty}\frac{\langle\Phi_{T}|e^{-\tau H/2}Oe^{-\tau H/2}|\Phi_{T}\rangle}{\langle\Phi_{T}|e^{-\tau H}|\Phi_{T}\rangle}=\lim_{L_{t}\to\infty}\frac{\langle\Phi_{T}|M^{L_{t}/2}OM^{L_{t}/2}|\Phi_{T}\rangle}{\langle\Phi_{T}|M^{L_{t}}|\Phi_{T}\rangle} \\ &=\lim_{L_{t}\to\infty}\frac{\int\mathcal{D}c\exp\left(-\sum_{i=1}^{L_{t}}c_{i}^{2}/2\right)\langle\Phi_{T}|M(c_{L_{t}})\cdots M(c_{L_{t}/2+1})OM(c_{L_{t}/2})\cdots M(c_{1})|\Phi_{T}\rangle}{\int\mathcal{D}c\exp\left(-\sum_{i=1}^{L_{t}}c_{i}^{2}/2\right)\langle\Phi_{T}|M(c_{L_{t}})\cdots M(c_{1})|\Phi_{T}\rangle}. \tag*{(14)} \end{aligned}

where we have neglected the three-body force and applied the traditional Hubbard-Stratonovich auxiliary field transformation and τ=Ltat\tau= L_t a_t. The proof remains essentially unchanged if we include the three-body force using the discrete transformation. Here M=:exp⁡(−atH):M = :\exp(-a_t H): is the transfer matrix, ci(n)c_i(n) is the auxiliary field at the ii-th time slice, and

M(c)=exp⁡[∑n,n′atΨ†(n)∇nn′2Ψ(n′)2M+∑n−atC2c(n)(ρ‾(n)+Cs2C2ρ‾s(n))](15)M(c) = \exp\left[\sum_{n,n'} a_t \frac{\Psi^\dagger(n)\nabla^2_{nn'}\Psi(n')}{2M}+\sum_n \sqrt{-a_t C_2}c(n)\left(\overline{\rho}(n)+\frac{C_s}{2C_2}\overline{\rho}_s(n)\right)\right] \tag*{(15)}

is the transformed transfer matrix. The summations over n,n′n,n' run over all L3L^3 lattice sites, and ∇nn′2\nabla^2_{nn'} represents the discretized Laplacian operator implemented using FFT.

The transfer matrix in Eq. (15) is bilinear in the fermion creation and annihilation operators. When acting on a single-particle wave function, it corresponds to a one-body imaginary-time evolution operator of dimension L3×L3L^3 \times L^3. For products of single-particle wave functions, it induces no many-body correlations and evolves each particle independently:

M(c)(∣ψ1⟩∧∣ψ2⟩∧⋯∧∣ψA⟩)=M(c)∣ψ1⟩∧M(c)∣ψ2⟩∧⋯∧M(c)∣ψA⟩,(16)M(c)\left(\lvert\psi_1\rangle\wedge\lvert\psi_2\rangle\wedge\cdots\wedge\lvert\psi_A\rangle\right)=M(c)\lvert\psi_1\rangle\wedge M(c)\lvert\psi_2\rangle\wedge\cdots\wedge M(c)\lvert\psi_A\rangle, \tag*{(16)}

where ψ1,…,A\psi_1,\ldots,A are a set of single-particle wave functions, and ∧\wedge denotes the antisymmetrized product, i.e., the Slater determinant.

Without loss of generality, we assume that ∣ΦT⟩\lvert\Phi_T\rangle is a Slater determinant composed of single-particle wave functions ϕ1,ϕ2,…,ϕA\phi_1,\phi_2,\ldots,\phi_A. Repeated application of Eq. (16) allows us to rewrite the denominator of Eq. (14) as

Z=∫Dcexp⁡(−∑i=1Ltci2/2)[⋀i=1A⟨ϕi∣][⋀i=1AM‾(c)∣ϕi⟩]=∫Dcexp⁡(−∑i=1Ltci2/2)det⁡(Z(c)),(17)\mathcal{Z}=\int\mathcal{D}c\exp\left(-\sum_{i=1}^{L_t}c_i^2/2\right)\left[\bigwedge_{i=1}^{A}\langle\phi_i\lvert\right]\left[\bigwedge_{i=1}^{A}\overline{M}(c)\lvert\phi_i\rangle\right]=\int\mathcal{D}c\exp\left(-\sum_{i=1}^{L_t}c_i^2/2\right)\det(Z(c)), \tag*{(17)}

where M‾(c)=M(cLt)⋯M(c1)\overline{M}(c)=M(c_{L_t})\cdots M(c_1) is the product of single-particle evolution operators, and ZZ is an A×AA\times A fermionic correlation matrix:

Z(c)=[⟨ϕi∣M‾(c)∣ϕj⟩]i,j=1A=(⟨ϕ1∣M‾(c)∣ϕ1⟩⟨ϕ1∣M‾(c)∣ϕ2⟩⋯⟨ϕ1∣M‾(c)∣ϕA⟩⟨ϕ2∣M‾(c)∣ϕ1⟩⟨ϕ2∣M‾(c)∣ϕ2⟩⋯⟨ϕ2∣M‾(c)∣ϕA⟩⋮⋮⋱⋮⟨ϕA∣M‾(c)∣ϕ1⟩⟨ϕA∣M‾(c)∣ϕ2⟩⋯⟨ϕA∣M‾(c)∣ϕA⟩).(18)Z(c)=\left[\langle\phi_i\lvert\overline{M}(c)\lvert\phi_j\rangle\right]_{i,j=1}^{A} =\begin{pmatrix} \langle\phi_1\lvert\overline{M}(c)\lvert\phi_1\rangle& \langle\phi_1\lvert\overline{M}(c)\lvert\phi_2\rangle& \cdots& \langle\phi_1\lvert\overline{M}(c)\lvert\phi_A\rangle\\ \langle\phi_2\lvert\overline{M}(c)\lvert\phi_1\rangle& \langle\phi_2\lvert\overline{M}(c)\lvert\phi_2\rangle& \cdots& \langle\phi_2\lvert\overline{M}(c)\lvert\phi_A\rangle\\ \vdots& \vdots& \ddots& \vdots\\ \langle\phi_A\lvert\overline{M}(c)\lvert\phi_1\rangle& \langle\phi_A\lvert\overline{M}(c)\lvert\phi_2\rangle& \cdots& \langle\phi_A\lvert\overline{M}(c)\lvert\phi_A\rangle \end{pmatrix}. \tag*{(18)}

In NLEFT calculations, we generate an ensemble of auxiliary-field configurations {c}\{c\} with the probability distribution

P(c)∝exp⁡(−∑i=1Ltci2/2)∣det⁡(Z(c))∣(19)P(c)\propto\exp\left(-\sum_{i=1}^{L_t}c_i^2/2\right)\lvert\det(Z(c))\rvert \tag*{(19)}

and measure observables as arithmetic averages. The sign problem occurs when the determinant in Eq. (19) is not positive definite. In this case, the averaged sign

⟨eiθ⟩=∫DcP(c)det⁡(Z(c))∣det⁡(Z(c))∣(20)\langle e^{i\theta}\rangle=\int\mathcal{D}cP(c)\frac{\det(Z(c))}{\lvert\det(Z(c))\rvert} \tag*{(20)}

quantifies the severity of the sign problem. For ⟨eiθ⟩=1\langle e^{i\theta}\rangle=1, there is no sign problem, while ⟨eiθ⟩≈0\langle e^{i\theta}\rangle\approx0 indicates a severe sign problem.

The single nucleon wave functions include spin and isospin indices and can be written as complex vectors of dimension 4×L34\times L^3. Correspondingly, the transfer matrix in Eq. (15) becomes a 4L3×4L34L^3\times4L^3 matrix for any given cc-field configuration. Since the interaction used in this work has no isospin-mixing term, the determinant factorizes:

det⁡(Z(c))=det⁡(ZP(c))det⁡(ZN(c)),(21)\det(Z(c))=\det(Z_P(c))\det(Z_N(c)), \tag*{(21)}

where ZPZ_P and ZNZ_N are the correlation matrices for protons and neutrons, respectively. In what follows, we omit the isospin degrees of freedom; all conclusions hold for Z(c)Z(c), ZP(c)Z_P(c), and ZN(c)Z_N(c).

For even-even nuclei, we prepare the single nucleon wave functions as paired by the time-reversal operation:

∣ϕA/2+k⟩=T∣ϕk⟩,k=1,2,…,A/2,(22)\lvert\phi_{A/2+k}\rangle=\mathcal{T}\lvert\phi_k\rangle,\qquad k=1,2,\ldots,A/2, \tag*{(22)}

where T=iσyK\mathcal{T}=i\sigma_y\mathcal{K} is the time-reversal operator and K\mathcal{K} denotes complex conjugation. The matrix M‾(c)\overline{M}(c) commutes with T\mathcal{T}, as the spin Pauli matrices and the imaginary factor ii in ρs\rho_s both contribute sign changes that cancel under time reversal. Note that we focus on the T\mathcal{T}-symmetry of the decomposed transfer matrix in Eq. (15), not the original interactions, which correspond to squares of the decomposed interactions and are always time-reversal even. For example, while the Coulomb force is time-reversal invariant, its auxiliary field transformation explicitly involves an imaginary phase and cannot be simulated without a sign problem.

Using the properties of T\mathcal{T} and Eq. (22), we obtain the relations:

⟨ϕA/2+i∣M‾(c)∣ϕA/2+j⟩=⟨Tϕi∣M‾(c)∣Tϕj⟩=⟨ϕi∣M‾(c)∣T†Tϕj⟩∗=⟨ϕi∣M‾(c)∣ϕj⟩∗,(23)\langle\phi_{A/2+i}|\overline{M}(c)|\phi_{A/2+j}\rangle= \langle\mathcal{T}\phi_i|\overline{M}(c)|\mathcal{T}\phi_j\rangle= \langle\phi_i|\overline{M}(c)|\mathcal{T}^{\dagger}\mathcal{T}\phi_j\rangle^{*} = \langle\phi_i|\overline{M}(c)|\phi_j\rangle^{*}, \tag*{(23)}
⟨ϕA/2+i∣M‾(c)∣ϕj⟩=⟨Tϕi∣M‾(c)∣ϕj⟩=⟨ϕi∣M‾(c)∣T†ϕj⟩∗=−⟨ϕi∣M‾(c)∣ϕA/2+j⟩∗.\langle\phi_{A/2+i}|\overline{M}(c)|\phi_j\rangle= \langle\mathcal{T}\phi_i|\overline{M}(c)|\phi_j\rangle= \langle\phi_i|\overline{M}(c)|\mathcal{T}^{\dagger}\phi_j\rangle^{*} = -\langle\phi_i|\overline{M}(c)|\phi_{A/2+j}\rangle^{*}.

The correlation matrix then has the structure:

Z=(U−V∗VU∗),(24)Z = \begin{pmatrix} U & -V^{*} \\ V & U^{*} \end{pmatrix}, \tag*{(24)}

where UU and VV are A/2×A/2A/2 \times A/2 complex matrices.

We define a spin-flipping matrix:

Σ=iσy⊗IA/2=(0IA/2−IA/20),(25)\Sigma= i\sigma_y \otimes I_{A/2} = \begin{pmatrix} 0 & I_{A/2} \\ -I_{A/2} & 0 \end{pmatrix}, \tag*{(25)}

where IA/2I_{A/2} is the A/2×A/2A/2 \times A/2 identity matrix. Direct verification shows that

ZΣ=ΣZ∗.(26)Z\Sigma= \Sigma Z^{*}. \tag*{(26)}

For any eigenvalue λ∈C\lambda\in\mathbb{C} with corresponding eigenvector vv satisfying Zv=λvZv = \lambda v, we have

Z(Σv∗)=ΣZ∗v∗=Σ(λ∗v∗)=λ∗(Σv∗),(27)Z(\Sigma v^{*}) = \Sigma Z^{*}v^{*} = \Sigma(\lambda^{*}v^{*}) = \lambda^{*}(\Sigma v^{*}), \tag*{(27)}

implying that λ∗\lambda^{*} is also an eigenvalue with eigenvector w=Σv∗w = \Sigma v^{*}. Thus, complex eigenvalues of ZZ always appear in conjugate pairs. If λ\lambda is real, we find

⟨v∣w⟩=v†Σv∗=(vTΣv)∗=0,(28)\langle v \mid w\rangle= v^{\dagger}\Sigma v^{*} = (v^{T}\Sigma v)^{*} = 0, \tag*{(28)}

indicating that vv and ww are orthogonal eigenvectors corresponding to the same real eigenvalue. In this case, the real eigenvalue appears twice in the diagonalized form of ZZ. Consequently, the determinant det⁡(Z)\det(Z), being the product of all eigenvalues, is always nonnegative. We note that for odd-even or odd-odd nuclei the correlation matrix can not be written in the form of Eq. (24) and the resulting determinant could be negative or complex.

Dependence on the lattice spacing

In this work, we adopt a fixed lattice spacing of a=1.32a = 1.32 fm, corresponding to a momentum cutoff Λ=π/a≈471\Lambda= \pi/a \approx471 MeV. This value is prevalent in recent NLEFT investigations [1, 3, 4, 5, 6]. We note that most continuum chiral EFT calculations employ a smooth cutoff Λ\Lambda between 450 and 550 MeV [7, 8], yielding similar cutoff dependencies. This cutoff scale also exposes a hidden spin-isospin exchange symmetry originating from QCD [9].

So far, most NLEFT calculations have been performed at a single lattice spacing. Systematic multi-cutoff studies have been limited to nucleon-nucleon scattering [10, 11, 12] and few-body systems (A≤4A \le4) [13, 14]. In this work, we expect that our optimization algorithm also works for other lattice spacings in a reasonable interval and the optimized nuclear force exhibits similar performance, with residual cutoff dependencies that can be estimated within the EFT framework. In typical nuclear EFTs, cutoff-dependencies scale as O((Q/Λ)ν+1)O((Q/\Lambda)^{\nu+1}), where QQ is the characteristic momentum and ν\nu is the order of the EFT interaction [15]. Our interaction incorporates an improved leading-order (O(Q0)O(Q^{0})) term with partial absorption of next-to-leading-order (O(Q2)O(Q^{2})) effects, including finite-range smearing and spin-orbit terms, and next-to-next-to-leading-order (O(Q3)O(Q^{3})) three-body forces. Thus we make a rough estimate ν≈2\nu\approx2. In nuclei, Q∼ME∼100Q \sim\sqrt{ME} \sim100 MeV, with MM the nucleon mass and EE the typical single-nucleon separation energy. For sufficiently large Λ\Lambda (e.g., Λ≥300\Lambda\ge300 MeV, corresponding to a≤2a \le2 fm), cutoff dependencies are suppressed to the order of O((100/300)3)≈O(0.037)O((100/300)^3) \approx O(0.037). These residual lattice effects can be numerically verified and further reduced by including higher-order interactions, which we leave for future work.

Galilean invariance

The interactions introduced in this work explicitly break Galilean invariance, which originates from the essential nonlocality required to reproduce experimental data. In the center-of-mass frame, our local and non-local smearing effectively mimic the q2q^{2} and k2k^{2} terms at next-to-leading order in the chiral expansion, while the spin-orbit term reduces to the standard contact coupling i2(q×k)⋅(σ1+σ2)\frac{i}{2}(\mathbf{q} \times\mathbf{k}) \cdot(\boldsymbol{\sigma}_{1}+\boldsymbol{\sigma}_{2}). To our knowledge, no existing nuclear force simultaneously satisfies Galilean invariance, avoids the QMC sign problem, and reproduces experimental binding energies.

Coulomb force

A static Coulomb force VcouV_{\mathrm{cou}} is included. The form is the same as that used in Ref. [16] with a different cutoff. We compute the first-order perturbative energy of VcouV_{\mathrm{cou}}. In momentum space, we have

Vcou=αq2exp⁡(−q22Λcou2),(29)V_{\mathrm{cou}} = \frac{\alpha}{q^{2}} \exp\left(-\frac{q^{2}}{2\Lambda_{\mathrm{cou}}^{2}}\right), \tag*{(29)}

where α=1/137\alpha= 1/137 is the fine structure constant, Λcou=180 MeV\Lambda_{\mathrm{cou}} = 180\ \mathrm{MeV} is a momentum cutoff that removes the singularity at r=0r = 0, and q=p′−p\mathbf{q} = \mathbf{p}^{\prime} - \mathbf{p} is the momentum transfer. In coordinate space, the potential is expressed using the error function:

Vcou(r)=αerf⁡(Λcour/2)r.(30)V_{\mathrm{cou}}(r) = \alpha\frac{\operatorname{erf}(\Lambda_{\mathrm{cou}} r/2)}{r}. \tag*{(30)}

The asymptotic behavior satisfies

Vcou(0)=αΛcouπandVcou(r)→αras Λcou→∞.V_{\mathrm{cou}}(0) = \frac{\alpha\Lambda_{\mathrm{cou}}}{\sqrt{\pi}} \quad\text{and} \quad V_{\mathrm{cou}}(r) \to\frac{\alpha}{r} \quad\text{as } \Lambda_{\mathrm{cou}} \to\infty.

The potential VcouV_{\mathrm{cou}} is smooth for all rr and can be implemented on the lattice using FFT.

Gradient Descent Algorithm for parameter optimization

Gradient descent is an optimization algorithm that leverages gradient information to iteratively update parameters. The method used in this work is summarized as follows:

  1. Initialization: Select a reasonable initial parameter set θ0\theta_{0}.

  1. Parameter rescaling: Rescale the parameters to approximately normalize the partial derivatives. During early iterations:

θi→θi∣∂χ2∂θi∣t=0,\theta_{i} \to\theta_{i}\left|\frac{\partial\chi^{2}}{\partial\theta_{i}}\right|_{t=0},

yielding partial derivatives of magnitude ∼1\sim1. Near convergence when iterations slow, apply fixed scaling:

θi→θi×ki,\theta_{i} \to\theta_{i} \times k_{i},

with scaling factors: ksL=1.853×105k_{sL} = 1.853 \times10^{5}, kC2=3.946×1011k_{C2} = 3.946 \times10^{11}, kC3=1.004×1018k_{C3} = 1.004 \times10^{18}, and kCs=1.260×1015k_{Cs} = 1.260 \times10^{15}. This technique significantly accelerates convergence.

  1. Gradient computation: For each iteration, the partial derivatives of the loss function ∂χ2/∂θi\partial\chi^{2}/\partial\theta_{i} must be computed. Figure 5 compares two methods for computing these partial derivatives.

Image not extracted

Figure 5. Comparison of methods for computing derivatives. (Upper panel) Ground state energies of 16O{}^{16}\mathrm{O} calculated with sL=0.0808s_{\mathrm{L}}=0.0808 (blue circles) and 0.08180.0818 (red diamonds), respectively. Other parameters take the LAT–OPT1 values. Errorbars denote independent statistical errors in each calculation. (Lower panel) Blue circles represent the difference between the two groups of points in the upper panel, with combined errors estimated with the error propagation formula e=e12+e22e=\sqrt{e_{1}^{2}+e_{2}^{2}}. Red squares denote the results from the Feynman–Hellmann theorem, with the statistical errors smaller than the symbols.

One straightforward approach is to directly calculate the energies using two sets of parameters differing by small variations θi\theta_{i} and θi+δθi\theta_{i}+\delta\theta_{i}, and then approximate the derivative via finite difference. However, these two calculations are independent and their combined statistical errors are typically similar in magnitude to the difference being measured. Consequently, the resulting derivative estimate often suffers from large statistical uncertainty.

Alternatively, we can apply the Feynman-Hellmann theorem (FHT). This allows us to directly measure the derivatives as observables within a single simulation run. Crucially, this approach significantly reduces the statistical errors. The key reason for this improvement is that the FHT method inherently captures the correlation between the energies at the slightly different parameter values within the same ensemble. Furthermore, this method enables the simultaneous measurement of derivatives with respect to any number of parameters.

This precise and efficient estimation of the derivatives forms the essential foundation of our optimization algorithm. The algorithm’s capability to handle derivatives effectively makes it suitable for application to a wider range of target nuclei and larger sets of adjustable parameters.

  1. Parameter update:

θt+1=θt−α∂χ2(θt)∂θ\theta_{t+1}=\theta_t-\alpha\frac{\partial\chi^2(\theta_t)}{\partial\theta}

with adaptive learning rate:

α={α×2if χ2 decreasesα×0.2otherwise\alpha= \begin{cases} \alpha\times2 & \text{if }\chi^2\text{ decreases}\\ \alpha\times0.2 & \text{otherwise} \end{cases}

This strategy increases step size in smooth regions and decreases it to prevent instability or overshooting.

  1. Convergence check: Terminate when:

∣χ2(θt+1)−χ2(θt)∣<ϵ.\left|\chi^2(\theta_{t+1})-\chi^2(\theta_t)\right|<\epsilon.

Figure 6 (Upper) demonstrates the monotonic decrease of χ2\chi^2 during optimization, confirming the method’s efficacy. Figure 6 (Lower) schematically illustrates parameter convergence to final values.

Image not extracted

Figure 6. optimization of loss function. (Upper) Loss function χ2\chi^2 versus iteration steps. (Lower) Dimensionless parameters normalized to converged values. Deviations of scaled sLs_L and C2C_2 are exaggerated for visibility. We divide each parameter by its value at the last iteration in order to present the convergence of all parameters within the same figure in an illustrative manner.

Computational complexity for heavy nuclei

The sign-problem-free QMC method is particularly suited for simulating heavy nuclei, as its computational time scales polynomially with nucleon number, whereas the sign problem inevitably introduces exponential scaling. We numerically demonstrate this advantage below.

Figure 7(a) depicts the average CPU hours per auxiliary-field configuration, measured on a consistent supercomputing platform. The dashed parabolic curve serves as a visual guide. Using fixed parameters (L=11L = 11, Lt=200L_t = 200), we confirm polynomial time complexity and observe approximately linear scaling for A≥40A \ge40.

Image not extracted

Figure 7. Computational performance. (a) Average CPU hours per configuration; (b) Absolute binding energy statistical uncertainties (16,800 configurations); (c) Average sign ⟨eiθ⟩\langle e^{i\theta}\rangle for LAT–OPT1 (orange squares) versus N2^{2}LO chiral force (blue circles, sign-problematic); (d) Relative binding energy statistical uncertainties.

Figure 7(b) displays statistical uncertainties in binding energies across 16,800 configurations. Uncertainties increase gradually with mass number but remain below 1 MeV for nuclei up to A∼100A \sim100. Relative QMC statistical errors (Figure 7(d)) exhibit rapid convergence in heavy nuclei, stabilizing near one-thousandth for A≥40A \ge40.

Combining these observations, we achieve this accuracy level with linearly increasing computational resources. This scalability is highly encouraging for QMC studies of heavy nuclei and warrants further investigation. Although larger simulation boxes are required for A>100A > 100 to suppress finite-volume effects, high-precision calculations remain feasible.

Finally, Figure 7(c) compares the average sign ⟨eiθ⟩\langle e^{i\theta}\rangle between our sign-problem-free interaction and a non-perturbative N2^2LO chiral force. The chiral interaction, featuring complex spin/isospin dependencies, exhibits a sign problem where ⟨eiθ⟩\langle e^{i\theta}\rangle decreases rapidly with mass number. Conversely, our approach maintains ⟨eiθ⟩=1\langle e^{i\theta}\rangle= 1 for all systems.

Finite Volume Effects

All calculations in this work employ a lattice box of size L=11L = 11, which yields negligible finite-volume effects for binding-energy computations of most nuclei. This is illustrated in Figure 8, showing convergence of binding energies (excluding Coulomb interactions) for 40^{40}Ca (Lt=400L_t = 400) and 100^{100}Sn (Lt=200L_t = 200) as box size increases. The distinct plateau beyond L=10L = 10 confirms effective convergence of finite-volume effects.

Image not extracted

Figure 8. Finite-volume convergence of binding energies. Binding energies (Coulomb excluded) for 40Ca^{40}\mathrm{Ca} (Lt=400L_t = 400, left) and 100Sn^{100}\mathrm{Sn} (Lt=200L_t = 200, right) versus lattice size LL. Blue symbols: Monte Carlo results with statistical uncertainties (error bars); red curves: exponential fits E(∞)+Ce−kLE(\infty) + Ce^{-kL}. Plateau formation at L≳10L \gtrsim10 confirms negligible finite-volume effects.

It is interesting to discuss the possible application of our method to even heavier nuclei, such as 208^{208}Pb or actinides. Based on the approximately linear scaling observed in Figure 7, we estimate that computations for nuclei with A≈200A \approx200 will require computational time roughly twice that for 100Sn^{100}\mathrm{Sn}. This remains computationally feasible with current supercomputing resources.

However, heavier nuclei exhibit larger radii, necessitating a larger simulation box size to mitigate finite-volume effects. The required box size should scale proportionally to the nuclear radius RR, which itself scales as A1/3A^{1/3}. Given that our algorithm scales as O(L3)O(L^3) with box size, this volume increase introduces an additional factor of AA into the overall scaling. Consequently, we estimate the scaling for precision calculations of heavy nuclei to be O(A2)O(A^2). This implies that simulating 208Pb^{208}\mathrm{Pb} would require approximately four times the computational resources as 100Sn^{100}\mathrm{Sn}.

For such simulations, addressing challenges like efficient memory management and re-optimization of the effective interaction will be essential prerequisites, along with GPU acceleration to facilitate the calculations.

Nucleon-nucleon scattering phase shifts

Nuclear ab initio calculations aim to solve nuclear structure from bare nuclear forces constrained solely by nucleon-nucleon phase shifts and few-body data. However, reconciling theoretical predictions with experimental data remains challenging. Consequently, modern ab initio approaches often employ interactions constrained by both NN phase shifts and nuclear properties.

We therefore investigate the inverse problem: whether nuclear forces determined solely by finite-nuclei observables can reproduce NN phase shifts. This approach is theoretically feasible since nuclear masses and charge radii can be measured with higher precision than scattering cross sections. Conversely, these observables should encode complete information about nuclear forces. Solving such inverse many-body problems requires high-precision nuclear solvers. While numerous mean-field and density functional interactions are calibrated to finite nuclei, these methods are based on variational principles and not exact. The sign-problem-free QMC method provides the essential foundation for this problem, offering both an unbiased high-precision solver and an efficient optimizer for nuclear masses.

Figure 9 shows that our LAT-OPT1 predictions for NN S-wave phase shifts fall between empirical 1S0{}^{1}S_{0} and 3S1{}^{3}S_{1} values across the momentum range up to 250 MeV, demonstrating reasonable reproduction of S-wave interaction strength. Fitting the phase shifts to the effective-range expansion,

Image not extracted

Figure 9. S-wave nucleon-nucleon phase shifts. Empirical phase shifts δ\delta for 1S0{}^{1}S_{0} (blue circles) and 3S1{}^{3}S_{1} (red squares) channels versus relative momentum PrelP_{\mathrm{rel}} [17]. Black dash-dot curve: LAT-OPT1 predictions. Inset: Extracted scattering length a0a_{0} and effective range r0r_{0}. LAT-OPT1 accurately captures the empirical trend up to Prel≈250P_{\mathrm{rel}} \approx250 MeV.

kcot⁡δ=−1a0+12r0k2+⋯k \cot\delta= -\frac{1}{a_{0}} + \frac{1}{2}r_{0}k^{2} + \cdots

yields a scattering length a0=6.86a_{0} = 6.86 fm and effective range r0=2.10r_{0} = 2.10 fm, both within the range of 1S0{}^{1}S_{0} and 3S1{}^{3}S_{1} experimental values. This represents a leading-order phase shift prediction from nuclear structure data.

While this simple interaction cannot reproduce S-wave splitting, S-D mixing, or P-wave phase shifts, including additional spin/isospin-dependent operators and pion-exchange potentials should improve description accuracy. Such extensions would however require sign-problem mitigation techniques.

Oxygen isotopes and pure neutron matter

Figure 10 presents binding energies per nucleon across the oxygen isotopic chain for both even-AA and odd-AA nuclei. The inset displays corresponding average phases. Calculations for odd nucleon numbers introduce a mild sign oscillation due to slight time-reversal symmetry breaking, yet yield statistical uncertainties comparable to even-AA nuclei, indicating no significant efficiency degradation. The computed energies reproduce odd-even staggering with slightly overestimated pairing gaps. Discrepancies at large neutron numbers may indicate missing interaction components (e.g., tensor forces).

Image not extracted

Figure 10. Binding energies in oxygen isotopes. Binding energy per nucleon E/AE/A versus mass number AA. Blue circles: even-AA; red circles: odd-AA; black crosses: experimental values. Inset: Average phase ⟨eiθ⟩\langle e^{i\theta}\rangle (= 1: sign-problem-free).

Figure 11 compares pure neutron matter equations of state predicted by different interaction models. The LAT-OPT1 parameterization shows excellent agreement with benchmark results up to saturation density. As demonstrated in the main text, this model approximately reproduces empirical saturation properties of symmetric nuclear matter. These results confirm that infinite nuclear matter properties can be reliably extracted from light nuclei fitting (A≤40A \le40), suggesting the essential role of many-body correlations.

Image not extracted

Figure 11. Pure neutron matter equation of state. Energy per particle versus neutron density for LAT-OPT1 interaction (red squares). Comparisons include: N3LON^3\mathrm{LO} chiral interactions (EM 500 MeV, EGM 450/500 MeV, EGM 450/700 MeV) [19]; variational APR [20]; Auxiliary Field Diffusion MC (GCR) [21].

Sign problem for odd-even and odd-odd nuclei

We have examined the performance of LAT-OPT1 for various odd-even and odd-odd heavy nuclei. Table II presents the average sign ⟨eiθ⟩\langle e^{i\theta}\rangle for representative nuclei computed with Lt=350L_t = 350 using identical computational settings as for even-even nuclei. The selected nuclei include three odd-even systems (41Ca{}^{41}\mathrm{Ca}, 57Ni{}^{57}\mathrm{Ni}, 85Zr{}^{85}\mathrm{Zr}) and one odd-odd nucleus (82Nb{}^{82}\mathrm{Nb}). For all odd-even nuclei, the average sign exceeds 0.9, indicating that the QMC statistical errors are comparable to those of neighboring even-even nuclei. For 82Nb{}^{82}\mathrm{Nb}, where ⟨eiθ⟩≈0.75\langle e^{i\theta}\rangle\approx0.75, the statistical uncertainty is merely ∼0.1\sim0.1 MeV larger than in adjacent even-even nuclei. We therefore conclude that our method enables precise extraction of ground-state energies for all nuclei with atomic mass numbers A≤100A \le100.

41Ca{}^{41}\mathrm{Ca}57Ni{}^{57}\mathrm{Ni}82Nb{}^{82}\mathrm{Nb}85Zr{}^{85}\mathrm{Zr}
⟨eiθ⟩\langle e^{i\theta}\rangle0.9150.9200.7500.927

Table 2. Average sign ⟨eiθ⟩\langle e^{i\theta}\rangle for selected odd-even and odd-odd nuclei.

Charge density distributions and charge radii

Charge density distributions (Figure 12) and corresponding charge radii (Table III) for the six target nuclei were calculated using the pinhole algorithm [18]. This method induces sign problems and is typically restricted to light nuclei (A≤40A \le40). Calculations incorporated the proton size Rp=0.84R_p = 0.84 fm via convolution.

Image not extracted

Figure 12. Charge density distributions. Calculated charge density distributions for selected nuclei without (blue circles) and with (red squares) perturbative Coulomb corrections compared with the empirical results [25].

Nucleus4He^{4}\mathrm{He}16O^{16}\mathrm{O}24Mg^{24}\mathrm{Mg}28Si^{28}\mathrm{Si}32S^{32}\mathrm{S}40Ca^{40}\mathrm{Ca}
LAT-OPT11.797(3)2.872(6)3.235(24)3.319(25)3.409(30)3.537(23)
Experiment1.6755(28)2.6991(52)3.0570(16)3.1224(24)3.2611(18)3.4776(19)
Relative error [%]7.36.45.86.34.51.7

Table III. Charge radii. RMS charge radii RR (fm) for selected nuclei computed with LAT-OPT1 versus experimental values [26].

Results with and without perturbative Coulomb inclusion demonstrate only minor corrections for these light nuclei. The computed charge densities approximately reproduce empirical profiles, though central regions exhibit large statistical uncertainties and slight underestimation compared to experimental data. Charge radii are overestimated by ∼5\sim5-7%. Inclusion of charge radii as optimization constraints should improve these results.

Comparison with phenomenological models

When evaluated as a nuclear mass model, LAT-OPT1 predictions compare favorably with established approaches: macroscopic-microscopic (mac-mic) models, relativistic mean field (RMF) theory, and Skyrme density functional theory. These models demonstrate strong performance in describing nuclear structures, with parameters constrained by global fitting to binding energies, charge radii, and nuclear matter saturation properties across training sets typically spanning 16O{}^{16}\mathrm{O} to 208Pb{}^{208}\mathrm{Pb}.

We compare three published parameterizations: the FRDM [22] (mac-mic), PC-PK1 [23] (RMF), and UNEDF1 [24] (Skyrme). Standard deviations from experimental masses are computed for even-even nuclei with N,Z≤28N, Z \le28 (Table IV). Phenomenological models show ∼\sim 2-3 MeV RMS deviations in this region, improving significantly for heavier nuclei where mean-field contributions dominate. LAT-OPT1 achieves comparable precision (∼\sim 3 MeV RMS) despite fewer parameters, while mac-mic models yield smaller deviations with more adjustable parameters.

ModelParameterizationParametersσ\sigma (MeV)Nucleiσall\sigma_{\mathrm{all}} (MeV)
Macroscopic-microscopicFRDM [22]>30> 301.15690.56
Relativistic mean fieldPC-PK1 [23]112.25601.52
Skyrme DFTUNEDF1 [24]123.43751.91
Lattice EFTWigner-SU(4) [1]410.2155—
Lattice EFTLAT-OPT152.9376—

Table IV. Standard deviations σ\sigma of ground-state binding energies for even-even nuclei (N,Z≤28N, Z \le28) across mass models. Here, Parameters is adjustable parameter count, Nuclei is nuclei count for σ\sigma, and σall\sigma_{\mathrm{all}} is standard deviation over full nuclide chart.

Table IV also includes results from the Wigner-SU(4) interaction in Ref. [1], demonstrating significant improvement through: (1) systematic finite-volume effect elimination via larger boxes; (2) comprehensive imaginary-time extrapolation replacing fixed-τ\tau calculations. The inclusion of spin-orbit coupling and efficient optimization further reduced deviations from ∼\sim 10 MeV to 3 MeV. Future extensions to larger training sets may further improve accuracy.

Complete numerical results (Table V) compare LAT-OPT1 with mean-field calculations, listing Coulomb contributions separately. We also present the original data as separate figures for each nuclei. We show the results without the Coulomb energy, which can be restored by looking up Table V.

AANNUNEDF1PC-PK1LAT-OPT1EXPCOU (MeV)
Z=2Z = 2 (He)
42−18.70−28.81−28.300.543
64−23.24−33.17(6)−29.270.527
86−24.61−37.68(10)−31.390.523
Z=4Z = 4 (Be)
84−43.42−56.98(20)−56.502.450
106−58.52−66.96(16)−64.972.553
128−64.89−72.29(30)−68.642.560
1410−77.28(29)−69.9162.482
Z=6Z = 6 (C)
AANNUNEDF1PC-PK1LAT–OPT1EXPCOU (MeV)
104−54.07-54.07−63.45(16)-63.45(16)−60.32-60.326.056
126−83.11-83.11−91.33(12)-91.33(12)−92.16-92.166.180
148−99.80-99.80−104.60(10)-104.60(10)−105.28-105.286.297
1610−105.23-105.23−112.88(30)-112.88(30)−110.75-110.756.175
1812−109.27-109.27−119.75(51)-119.75(51)−115.67-115.676.038
2014−112.72-112.72−125.61(47)-125.61(47)−119.22-119.226.047
2216−114.57-114.57−131.86(55)-131.86(55)−119.26-119.266.001
Z=8Z = 8 (O)
124−55.10-55.10−59.70-59.70−63.70(12)-63.70(12)−58.58-58.5810.875
146−93.98-93.98−101.16-101.16−99.62(10)-99.62(10)−98.73-98.7311.278
168−121.71-121.71−128.03-128.03−126.94(15)-126.94(15)−127.62-127.6211.318
1810−136.12-136.12−141.84-141.84−139.33(22)-139.33(22)−139.81-139.8111.221
2012−147.62-147.62−152.98-152.98−152.22(21)-152.22(21)−151.37-151.3711.142
2214−157.22-157.22−161.39-161.39−163.63(26)-163.63(26)−162.03-162.0311.099
2416−164.00-164.00−165.98-165.98−174.87(30)-174.87(30)−168.95-168.9511.051
2618−165.45-165.45−167.41-167.41−179.02(34)-179.02(34)−168.93-168.9311.873
Z=10Z = 10 (Ne)
166−100.15-100.15−101.86(16)-101.86(16)−97.33-97.3317.054
188−129.33-129.33−134.37-134.37−133.48(40)-133.48(40)−132.14-132.1417.190
2010−152.93-152.93−160.81-160.81−158.90(52)-158.90(52)−160.64-160.6417.123
2212−172.32-172.32−177.97-177.97−176.33(25)-176.33(25)−177.77-177.7717.011
2414−186.86-186.86−192.61-192.61−190.22(23)-190.22(23)−191.84-191.8417.112
2616−198.32-198.32−202.25-202.25−203.86(16)-203.86(16)−201.55-201.5517.052
2818−204.30-204.30−208.55-208.55−211.98(36)-211.98(36)−206.87-206.8716.867
3020−207.95-207.95−212.44-212.44−217.41(29)-217.41(29)−211.04-211.0416.653
Z=12Z = 12 (Mg)
208−132.57-132.57−135.54-135.54−138.08(20)-138.08(20)−134.56-134.5624.448
2210−163.81-163.81−169.09-169.09−168.72(22)-168.72(22)−168.58-168.5824.334
2412−190.53-190.53−206.16-206.16−196.03(32)-196.03(32)−198.26-198.2624.424
2614−211.19-211.19−220.52-220.52−213.13(32)-213.13(32)−216.68-216.6824.378
2816−227.37-227.37−231.59-231.59−229.96(27)-229.96(27)−231.63-231.6324.412
3018−238.48-238.48−243.12-243.12−242.85(18)-242.85(18)−241.63-241.6324.193
3220−246.65-246.65−252.21-252.21−250.49(21)-250.49(21)−249.72-249.7223.985
3422−251.27-251.27−260.61-260.61−259.95(36)-259.95(36)−256.71-256.7123.616
3624−255.81-255.81−266.85-266.85−267.76(33)-267.76(33)−260.80-260.8023.399
Z=14Z = 14 (Si)
2410−168.73-168.73−172.61-172.61−174.45(41)-174.45(41)−172.01-172.0132.929
2612−201.15-201.15−207.64-207.64−204.98(50)-204.98(50)−206.04-206.0432.941

Table V. Ground-state binding energies of even-even nuclei calculated with UNEDF1 [24], PC-PK1 [23] and LAT-OPT1, alongside experimental values (EXP) and perturbative Coulomb corrections in LAT-OPT1 calculations (COU). The LAT-OPT1 errors are combined statistical and imaginary-time-extrapolation errors. All energies are in MeV.

AANNUNEDF1PC-PK1LAT-OPT1EXPCOU (MeV)
2814−229.40-229.40−239.37-239.37−233.18(35)-233.18(35)−236.54-236.5433.227
3016−251.31-251.31−253.02-253.02−252.60(21)-252.60(21)−255.62-255.6233.015
3218−267.55-267.55−271.61-271.61−268.49(36)-268.49(36)−271.41-271.4132.872
3420−280.56-280.56−284.74-284.74−281.45(27)-281.45(27)−283.46-283.4632.687
3622−288.46-288.46−292.93-292.93−292.97(39)-292.97(39)−292.05-292.0532.318
3824−295.95-295.95−303.93-303.93−302.81(39)-302.81(39)−299.93-299.9331.932
4026−301.71-301.71−311.49-311.49−311.39(76)-311.39(76)−306.23-306.2331.677
Z=16Z = 16 (S)
2812−206.45-206.45−209.32-209.32−212.12(28)-212.12(28)−209.41-209.4142.472
3014−240.24-240.24−241.38-241.38−242.86(21)-242.86(21)−243.68-243.6842.741
3216−267.57-267.57−269.35-269.35−271.34(32)-271.34(32)−271.78-271.7842.930
3418−288.88-288.88−289.29-289.29−291.99(11)-291.99(11)−291.84-291.8442.443
3620−306.68-306.68−308.37-308.37−308.54(20)-308.54(20)−308.71-308.7142.294
3822−318.83-318.83−322.56-322.56−321.67(26)-321.67(26)−321.05-321.0541.941
4024−330.10-330.10−335.55-335.55−334.04(77)-334.04(77)−333.17-333.1741.665
4226−339.61-339.61−346.39-346.39−345.48(43)-345.48(43)−344.12-344.1241.420
4428−347.49-347.49−354.35-354.35−354.82(54)-354.82(54)−351.82-351.8241.284
Z=18Z = 18 (Ar)
3214−244.70-244.70−246.95-246.95−248.05(29)-248.05(29)−246.40-246.4053.414
3416−276.88-276.88−276.40-276.40−281.05(16)-281.05(16)−278.72-278.7253.397
3618−303.51-303.51−306.28-306.28−306.11(40)-306.11(40)−306.72-306.7253.459
3820−326.49-326.49−327.13-327.13−328.09(14)-328.09(14)−327.34-327.3453.073
4022−343.13-343.13−343.28-343.28−344.37(28)-344.37(28)−343.81-343.8152.717
4224−357.79-357.79−360.56-360.56−360.18(32)-360.18(32)−359.34-359.3452.351
4426−371.15-371.15−374.98-374.98−373.95(46)-373.95(46)−373.73-373.7352.170
4628−382.74-382.74−387.54-387.54−387.62(32)-387.62(32)−386.97-386.9752.050
Z=20Z = 20 (Ca)
3616−281.41-281.41−281.45-281.45−285.52(20)-285.52(20)−281.37-281.3765.306
3818−313.11-313.11−313.03-313.03−315.74(10)-315.74(10)−313.12-313.1265.166
4020−341.11-341.11−343.06-343.06−343.00(18)-343.00(18)−342.05-342.0565.436
4222−362.24-362.24−363.51-363.51−362.55(21)-362.55(21)−361.90-361.9064.697
4424−381.13-381.13−382.12-382.12−380.36(33)-380.36(33)−380.96-380.9664.434
4626−398.49-398.49−399.38-399.38−397.78(12)-397.78(12)−398.77-398.7764.181
4828−413.75-413.75−415.50-415.50−414.49(29)-414.49(29)−416.00-416.0064.110
Z=22Z = 22 (Ti)
4018−315.78-315.78−314.46-314.46−319.86(38)-319.86(38)−314.63-314.6377.408
4220−348.27-348.27−348.76-348.76−349.70(17)-349.70(17)−346.89-346.8977.381
4422−373.46-373.46−373.09-373.09−376.20(45)-376.20(45)−375.47-375.4777.218
AANNUNEDF1PC-PK1LAT-OPT1EXPCOU (MeV)
4624−396.43-396.43−398.46-398.46−396.45(44)-396.45(44)−398.20-398.2076.715
4826−417.47-417.47−419.46-419.46−416.18(28)-416.18(28)−418.70-418.7076.576
5028−436.29-436.29−436.72-436.72−434.70(46)-434.70(46)−437.79-437.7976.446
Z=24Z = 24 (Cr)
4420−351.82-351.82−351.51-351.51−354.10(46)-354.10(46)−349.78-349.7890.962
4622−381.05-381.05−382.51-382.51−383.20(76)-383.20(76)−381.98-381.9890.616
4824−409.68-409.68−412.29-412.29−411.70(59)-411.70(59)−411.47-411.4790.573
5026−434.00-434.00−435.67-435.67−431.81(20)-431.81(20)−435.05-435.0590.115
5228−455.12-455.12−454.84-454.84−451.75(32)-451.75(32)−456.35-456.3590.032
Z=26Z = 26 (Fe)
4822−385.80-385.80−386.47-386.47−386.93(27)-386.93(27)−385.09-385.09105.36
5024−417.55-417.55−418.55-418.55−417.94(71)-417.94(71)−417.70-417.70105.05
5226−445.60-445.60−447.92-447.92−447.08(78)-447.08(78)−447.70-447.70105.25
5428−470.35-470.35−470.35-470.35−467.81(40)-467.81(40)−471.76-471.76104.86
Z=28Z = 28 (Ni)
5224−421.10-421.10−419.22-419.22−421.19(45)-421.19(45)−420.36-420.36120.98
5426−452.63-452.63−451.87-451.87−450.88(32)-450.88(32)−453.22-453.22120.80
5628−481.59-481.59−483.65-483.65−479.33(60)-479.33(60)−484.00-484.00121.35
Z=40Z = 40 (Zr)
8040−669.13-669.13−666.35-666.35−672.1(8)-672.1(8)−669.20-669.20233.90
9050−786.02-786.02−783.47-783.47−782.1(5)-782.1(5)−783.90-783.90230.52
Z=50Z = 50 (Sn)
10050−828.83-828.83−827.97-827.97−824.7(8)-824.7(8)−825.16-825.16348.76
12070−1020.24-1020.24−1019.77-1019.77−1033.65(11)-1033.65(11)−1020.48-1020.48339.35
13282−1101.53-1101.53−1104.47-1104.47−1134.2(27)-1134.2(27)−1102.84-1102.84333.14

Table 1.

Image not extracted

Figure 13 (PDF p. 22). 6^{6}He: Energy without COU (MeV) versus Lt; values −30.35, −32.33, −33.01, −33.47, −33.46, −33.73, −33.70; Extrapolated (Lt→∞Lt \to\infty): -33.70 MeV; Uncertainty in Extrapolation: 0.06. 8^{8}He: Energy without COU (MeV) versus Lt; values −31.13, −34.86, −36.90, −37.63, −37.81, −37.95, −38.23; Extrapolated (Lt→∞Lt \to\infty): -38.20 MeV; Uncertainty in Extrapolation: 0.10. 8^{8}Be: Energy without COU (MeV) versus Lt; values −48.77, −54.94, −57.04, −58.24, −58.88, −59.42, −59.55; Extrapolated (Lt→∞Lt \to\infty): -59.43 MeV; Uncertainty in Extrapolation: 0.20. 10^{10}Be: Energy without COU (MeV) versus Lt; values −60.14, −66.06, −68.54, −68.93, −69.21, −69.20, −69.86; Extrapolated (Lt→∞Lt \to\infty): -69.51 MeV; Uncertainty in Extrapolation: 0.16. 12^{12}Be: Energy without COU (MeV) versus Lt; values −65.99, −70.96, −72.55, −73.48, −73.97, −74.90, −74.92; Extrapolated (Lt→∞Lt \to\infty): -74.85 MeV; Uncertainty in Extrapolation: 0.30. 14^{14}Be: Energy without COU (MeV) versus Lt; values −59.93, −69.36, −74.11, −77.14, −77.98, −78.51, −79.64; Extrapolated (Lt→∞Lt \to\infty): -79.76 MeV; Uncertainty in Extrapolation: 0.29. 10^{10}C: Energy without COU (MeV) versus Lt; values −60.14, −66.06, −68.48, −68.93, −69.21, −69.20, −69.86; Extrapolated (Lt→∞Lt \to\infty): -69.51 MeV; Uncertainty in Extrapolation: 0.16. 12^{12}C: Energy without COU (MeV) versus Lt; values −84.61, −92.88, −95.95, −96.44, −97.16, −97.21, −97.34; Extrapolated (Lt→∞Lt \to\infty): -97.26 MeV; Uncertainty in Extrapolation: 0.12.

Image not extracted

Figure 14.

Image not extracted

Figure 15 (PDF p. 24).

Image not extracted

Figure 16 (PDF p. 25).

Image not extracted

Figure 17 (PDF p. 26).

Image not extracted

Figure 18 (PDF p. 27).

Image not extracted

Figure 19.

Image not extracted

Figure 20. 32^{32}Ar: Energy without COU (MeV); Lt; Extrapolated (Lt→∞Lt \to\infty): -301.46 MeV; Uncertainty in Extrapolation: 0.29; plotted values -286.49, -296.24, -299.17, -300.52, -300.67, -301.48, -302.18. 34^{34}Ar: Energy without COU (MeV); Lt; Extrapolated (Lt→∞Lt \to\infty): -334.45 MeV; Uncertainty in Extrapolation: 0.16; plotted values -318.54, -329.23, -332.97, -333.50, -334.28, -334.25, -334.75. 36^{36}Ar: Energy without COU (MeV); Lt; Extrapolated (Lt→∞Lt \to\infty): -359.57 MeV; Uncertainty in Extrapolation: 0.40; plotted values -343.02, -352.64, -356.39, -359.17, -358.25, -359.63, -359.44. 38^{38}Ar: Energy without COU (MeV); Lt; Extrapolated (Lt→∞Lt \to\infty): -381.17 MeV; Uncertainty in Extrapolation: 0.14; plotted values -365.94, -375.21, -378.53, -379.87, -380.75, -381.20, -381.14. 40^{40}Ar: Energy without COU (MeV); Lt; Extrapolated (Lt→∞Lt \to\infty): -397.09 MeV; Uncertainty in Extrapolation: 0.28; plotted values -378.69, -389.64, -394.37, -395.51, -396.82, -396.30, -397.43. 42^{42}Ar: Energy without COU (MeV); Lt; Extrapolated (Lt→∞Lt \to\infty): -412.53 MeV; Uncertainty in Extrapolation: 0.32; plotted values -392.24, -404.07, -409.15, -411.03, -411.14, -412.12, -413.03. 44^{44}Ar: Energy without COU (MeV); Lt; Extrapolated (Lt→∞Lt \to\infty): -426.12 MeV; Uncertainty in Extrapolation: 0.46; plotted values -405.92, -418.18, -422.27, -424.01, -425.45, -425.36, -427.00. 46^{46}Ar: Energy without COU (MeV); Lt; Extrapolated (Lt→∞Lt \to\infty): -439.67 MeV; Uncertainty in Extrapolation: 0.32; plotted values -420.45, -430.57, -435.18, -437.33, -438.41, -438.75, -440.07.

Figure 21 (PDF p. 30). 36{}^{36}Ca: Energy without COU (MeV) versus LtL t; data labels −337.09-337.09, −345.06-345.06, −348.34-348.34, −350.20-350.20, −350.71-350.71, −350.54-350.54, −350.40-350.40; Extrapolated (Lt→∞L t \to\infty): −350.83-350.83 MeV; Uncertainty in Extrapolation: 0.200.20. 38{}^{38}Ca: Energy without COU (MeV) versus LtL t; data labels −365.60-365.60, −375.22-375.22, −379.01-379.01, −380.35-380.35, −380.44-380.44, −380.71-380.71, −380.93-380.93; Extrapolated (Lt→∞L t \to\infty): −380.91-380.91 MeV; Uncertainty in Extrapolation: 0.100.10. 40{}^{40}Ca: Energy without COU (MeV) versus LtL t; data labels −393.37-393.37, −403.12-403.12, −406.61-406.61, −407.42-407.42, −408.26-408.26, −408.09-408.09, −408.86-408.86; Extrapolated (Lt→∞L t \to\infty): −408.44-408.44 MeV; Uncertainty in Extrapolation: 0.180.18. 42{}^{42}Ca: Energy without COU (MeV) versus LtL t; data labels −410.82-410.82, −420.59-420.59, −424.94-424.94, −426.16-426.16, −426.33-426.33, −427.03-427.03, −427.54-427.54; Extrapolated (Lt→∞L t \to\infty): −427.25-427.25 MeV; Uncertainty in Extrapolation: 0.210.21. 44{}^{44}Ca: Energy without COU (MeV) versus LtL t; data labels −427.55-427.55, −439.10-439.10, −442.71-442.71, −443.30-443.30, −445.16-445.16, −444.58-444.58, −444.99-444.99; Extrapolated (Lt→∞L t \to\infty): −444.79-444.79 MeV; Uncertainty in Extrapolation: 0.330.33. 46{}^{46}Ca: Energy without COU (MeV) versus LtL t; data labels −445.28-445.28, −455.15-455.15, −459.05-459.05, −460.76-460.76, −461.25-461.25, −461.75-461.75, −462.13-462.13; Extrapolated (Lt→∞L t \to\infty): −461.96-461.96 MeV; Uncertainty in Extrapolation: 0.120.12. 48{}^{48}Ca: Energy without COU (MeV) versus LtL t; data labels −462.84-462.84, −471.71-471.71, −476.33-476.33, −477.06-477.06, −477.65-477.65, −478.68-478.68, −478.65-478.65; Extrapolated (Lt→∞L t \to\infty): −478.60-478.60 MeV; Uncertainty in Extrapolation: 0.290.29. 40{}^{40}Ti: Energy without COU (MeV) versus LtL t; data labels −378.81-378.81, −389.15-389.15, −394.42-394.42, −395.65-395.65, −395.99-395.99, −396.99-396.99, −397.82-397.82; Extrapolated (Lt→∞L t \to\infty): −397.27-397.27 MeV; Uncertainty in Extrapolation: 0.380.38.

Figure 1 (PDF p. 31).

Figure 22.

Image not extracted

Figure 23 (PDF p. 33).

References

  1. [1]B.-N. Lu, N. Li, S. Elhatisari, D. Lee, E. Epelbaum and U.-G. Meißner. Essential elements for nuclear binding. Phys. Lett. B 797, 134863 (2019).DOI
  2. [2]C.-J. Wu and S.-C. Zhang. Sufficient condition for absence of the sign problem in the fermionic quantum Monte Carlo algorithm. Phys. Rev. B 71, 155115 (2005).DOI
  3. [3]B.-N. Lu, N. Li, S. Elhatisari, Y.-Z. Ma, D. Lee and U.-G. Meißner. Perturbative Quantum Monte Carlo Method for Nuclear Physics. Phys. Rev. Lett. 128, 242501 (2022).
  4. [4]S. Elhatisari, L. Bovermann, Y.-Z. Ma, E. Epelbaum, D. Frame, F. Hildenbrand, M. Kim, Y. Kim, H. Krebs, T. A. Lähde, D. Lee, N. Li, B.-N. Lu, U.-G. Meißner, G. Rupak, S. Shen, Y.-H. Song and G. Stellin. Wavefunction matching for solving quantum many-body problems. Nature 630, 59 (2024).
  5. [5]Y.-Z. Ma, Z. Lin, B.-N. Lu, S. Elhatisari, D. Lee, N. Li, U.-G. Meißner, A. W. Steiner and Q. Wang. Structure Factors for Hot Neutron Matter from Ab Initio Lattice Simulations with High-Fidelity Chiral Interactions. Phys. Rev. Lett. 132, 232502 (2024).
  6. [6]S. Shen, S. Elhatisari, D. Lee, U.-G. Meißner and Z. Ren. Ab Initio Study of the Beryllium Isotopes ⁷Be to ¹²Be. Phys. Rev. Lett. 134, 162503 (2025).
  7. [7]E. Epelbaum, W. Glöckle and U.-G. Meißner. The two-nucleon system at next-to-next-to-next-to-leading order. Nucl. Phys. A 747, 362–424 (2005).DOI
  8. [8]D. R. Entem and R. Machleidt. Accurate charge-dependent nucleon-nucleon potential at fourth order of chiral perturbation theory. Phys. Rev. C 68, 041001 (2003).DOI
  9. [9]D. Lee, S. Bogner, B. A. Brown, S. Elhatisari, E. Epelbaum, H. Hergert, M. Hjorth-Jensen, H. Krebs, N. Li, B.-N. Lu and U.-G. Meißner. Hidden Spin-Isospin Exchange Symmetry. Phys. Rev. Lett. 127, 062501 (2021).DOI
  10. [10]J. M. Alarcón, D. Du, N. Klein, T. A. Lähde, D. Lee, N. Li, B.-N. Lu, T. Luu and U.-G. Meißner. Neutron-proton scattering at next-to-next-to-leading order in Nuclear Lattice Effective Field Theory. Eur. Phys. J. A 53, 83 (2017).
  11. [11]N. Li, S. Elhatisari, E. Epelbaum, D. Lee, B.-N. Lu and U.-G. Meißner. Neutron-proton scattering with lattice chiral effective field theory at next-to-next-to-next-to-leading order. Phys. Rev. C 98, 044002 (2018).DOI
  12. [12]C. Wu, T. Wang, B.-N. Lu and N. Li. Charge-dependent nucleon-nucleon interaction at N³LO in nuclear lattice effective field theory. Phys. Rev. C 112, 014009 (2025).DOI
  13. [13]N. Klein, D. Lee, W. Liu and U.-G. Meißner. Regularization methods for Nuclear Lattice Effective Field Theory. Phys. Lett. B 747, 511–516 (2015).DOI
  14. [14]J.-A. Shi, C.-W. Wang and B.-N. Lu. Observation of renormalization group invariance in symmetry-restored nuclear lattice effective field theory. arXiv:2509.02953 [nucl-th] (2025).arxiv.org/abs/2509.02953
  15. [15]H.-W. Hammer, S. König and U. van Kolck. Nuclear effective field theory: Status and perspectives. Rev. Mod. Phys. 92, 025004 (2020).DOI
  16. [16]J. Liu, T. Wang and B.-N. Lu. Perturbative quantum Monte Carlo calculation with high-fidelity nuclear forces. Eur. Phys. J. A 61, 85 (2025).DOI
  17. [17]V. G. J. Stoks, R. A. M. Klomp, M. C. M. Rentmeester, and J. J. de Swart. Partial-wave analysis of all nucleon–nucleon scattering data below 350 MeV. Phys. Rev. C 48, 792 (1993).pubmed.ncbi.nlm.nih.gov/9968891
  18. [18]S. Elhatisari, E. Epelbaum, H. Krebs, T. A. Lähde, D. Lee, N. Li, B.-N. Lu, U.-G. Meißner and G. Rupak. Ab initio Calculations of the Isotopic Dependence of Nuclear Clustering. Phys. Rev. Lett. 119, 222505 (2017).DOI
  19. [19]I. Tews, T. Krüger, K. Hebeler and A. Schwenk. Neutron Matter at Next-to-Next-to-Next-to-Leading Order in Chiral Effective Field Theory. Phys. Rev. Lett. 110, 032504 (2013).pubmed.ncbi.nlm.nih.gov/23373917
  20. [20]A. Akmal, V. R. Pandharipande and D. G. Ravenhall. Equation of state of nucleon matter and neutron star structure. Phys. Rev. C 58, 1804 (1998).DOI
  21. [21]S. Gandolfi, J. Carlson and S. Reddy. Maximum mass and radius of neutron stars, and the nuclear symmetry energy. Phys. Rev. C 85, 032801 (2012).DOI
  22. [22]P. Möller, A. J. Sierk, T. Ichikawa and H. Sagawa. Nuclear ground-state masses and deformations: FRDM(2012). At. Data Nucl. Data Tables 109–110, 1 (2016).
  23. [23]K. Zhang et al. Nuclear mass table in deformed relativistic Hartree-Bogoliubov theory in continuum, I: Even-even nuclei. At. Data Nucl. Data Tables 144, 101488 (2022).arxiv.org/abs/2201.03216
  24. [24]M. Kortelainen, J. McDonnell, W. Nazarewicz, P.-G. Reinhard, J. Sarich, N. Schunck, M. V. Stoitsov and S. M. Wild. Nuclear energy density optimization: Large deformations. Phys. Rev. C 85, 024304 (2012).DOI
  25. [25]H. De Vries, C. W. De Jager and C. De Vries. Nuclear charge-density-distribution parameters from elastic electron scattering. At. Data Nucl. Data Tables 36, 495 (1987).DOI
  26. [26]M. Wang, G. Audi, A. H. Wapstra, F. G. Kondev, M. MacCormick, X. Xu and B. Pfeiffer. The Ame2012 atomic mass evaluation. Chin. Phys. C 36, 1603 (2012).

Paper details

vlnc:W_dsq32hn7
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents