betaFeedback

WorksPhysics and Astronomy

Stochastic Similarity Renormalization Group

Abstract

By integrating the quantum Monte Carlo technique into the similarity renormalization group (SRG), we have developed a stochastic SRG framework (SRGQMC) capable of both free-space two-body and in-medium many-body evolutions. This approach circumvents the combinatorial tensor-space explosion of many-body flow equations by mapping continuous unitary transformations onto an ensemble of signed random walkers. We benchmark the SRGQMC against deterministic free-space SRG evolutions of realistic nucleon-nucleon (NN) interactions, as well as against in-medium SRG (IMSRG) many-body calculations with the Richardson pairing model at two- and three-body levels [IMSRG(2)/(3)]. While a deterministic extension to the four-body level [IMSRG(4)] remains unfeasible due to prohibitive computational costs, we have achieved the first IMSRG(4) calculation by using the stochastic technique, demonstrating a substantial improvement toward the full configuration-interaction limit. This stochastic framework provides a practical pathway to higher-order IMSRG calculations.

main.pdf

Stochastic Similarity Renormalization Group

Introduction

The renormalization group (RG) [1, 2] provides a powerful reorganizing framework for multiscale quantum systems, decoupling low-energy physics from high-energy or short-range correlations. The similarity renormalization group (SRG) [3, 4] realizes this through a continuous unitary evolution of the Hamiltonian toward a band- or block-diagonal structure while preserving physical observables. The free-space SRG transformation softens nuclear forces through non-perturbative decoupling of high- and low-momentum components, thereby greatly accelerating the convergence of subsequent many-body calculations [5]. The in-medium SRG (IMSRG) [6–8] generalizes the evolution strategy of the flow equation, directly evolving the many-body Hamiltonian to decouple the target state of the finite nucleus or infinite nuclear matter.

However, the predictive power of IMSRG remains fundamentally constrained by mandatory truncations of induced high-rank terms in Hamiltonian and other many-body operators [7]. While the normal-ordered two-body approximation, IMSRG(2), has achieved widespread success [9–11], it systematically omits higher-order correlations that are essential for describing, e.g., nuclear collectivities [12–15] and clustering [16–18]. The truncation to higher-order IMSRG(3) or even IMSRG(4) can capture more many-body correlations. However, traditional deterministic implementations trigger a combinatorial explosion of the operator space and nested tensor contractions, making high-order evolution computationally infeasible [19–21].

In this Letter, we circumvent this bottleneck by introducing the stochastic similarity renormalization group with quantum Monte Carlo (SRGQMC), in which the continuous unitary flow is represented by a dynamic ensemble of random walkers rather than by the deterministic evolution of operator matrices. The walkers sample the commutator structures of the flow equation and faithfully preserve the renormalization trajectory with controllable statistical uncertainties. In this way, SRGQMC preserves the decoupling principle of SRG while offering a scalable Monte Carlo route toward higher-order IMSRG calculations. The resulting energies calculated at the four-body level show a substantial restoration of missing higher-order correlations toward the full configuration-interaction (FCI) limit, demonstrating that walker sampling can overcome the four-body operator barrier that has traditionally restricted deterministic IMSRG calculations.

For evolution of NNNN interaction

We first formulate the stochastic free-space SRG evolution for a two-nucleon Hamiltonian. For a given partial wave, the intrinsic two-nucleon Hamiltonian is partitioned as H(s)=T+V(s)H(s)=T+V(s), where ss is the continuous flow parameter, and TT is the relative kinetic energy which is ss-independent. Employing the standard Wegner generator [2], η(s)=[T,H(s)]\eta(s)=[T,H(s)], the unitary evolution is governed by the operator flow equation [3]

dH(s)ds=[η(s),H(s)]=[[T,H(s)],H(s)],(1)\frac{\mathrm{d}H(s)}{\mathrm{d}s}=[\eta(s),H(s)]=[[T,H(s)],H(s)], \tag*{(1)}

which is written into a set of coupled nonlinear matrix equations on a discrete momentum mesh [3],

dHijds=∑k(ηikHkj−Hikηkj),(2)\frac{\mathrm{d}H_{ij}}{\mathrm{d}s}=\sum_k\left(\eta_{ik}H_{kj}-H_{ik}\eta_{kj}\right), \tag*{(2)}

with

ηij=(ϵi−ϵj)Hij,(3)\eta_{ij}=(\epsilon_i-\epsilon_j)H_{ij}, \tag*{(3)}

where ϵi\epsilon_i is the discrete kinetic energy.

Figure 1. SRG- and SRGQMC-evolved neutron-proton interactions at λ=2.0 fm−1\lambda= 2.0\,\mathrm{fm}^{-1} for N3^{3}LOEMN_{\mathrm{EMN}}(500). Columns from left to right denote the 1S0^{1}S_{0}, 3P0^{3}P_{0}, 3S1^{3}S_{1}, 3D1^{3}D_{1}, and ϵ1\epsilon_{1} channels. Row 2 presents the SRGQMC-evolved matrix elements V(p′,p)V(p',p) using the stochastic loop-averaged sampling with Nw=104N_w = 10^4 and Nloop=10N_{\mathrm{loop}} = 10. Row 3 shows the scattering phase shifts (including the mixing angle ϵ1\epsilon_{1}) against the laboratory energy TlabT_{\mathrm{lab}}, derived from SRG (blue solid lines) and SRGQMC (red dots with error bars), compared with the empirical Nijmegen analysis (open diamonds) [22] as a reference.

In the SRGQMC framework, the deterministic matrix HijH_{ij} is mapped onto an ensemble of signed discrete walkers residing on all possible matrix sites (i,j)(i,j). The stochastic estimator of the Hamiltonian is given by Hij≃NCijH_{ij} \simeq\mathcal{N} C_{ij}, where CijC_{ij} represents the net walker population on site (i,j)(i,j). To control the simulation size, the scaling factor N\mathcal{N} is uniquely defined by the target total walker population NwN_w at s=0s=0 via

N=1Nw∑ij∣Hij(0)∣.(4)\mathcal{N} = \frac{1}{N_w} \sum_{ij} \lvert H_{ij}(0)\rvert. \tag*{(4)}

The walker distribution is initialized, proportional to the unevolved Hamiltonian, Cij(0)∝Hij(0)C_{ij}(0) \propto H_{ij}(0).

The stochastic flow is driven by sampling the commutator in Eq. (2). During a “time” step Δs\Delta s, each walker at site (i,j)(i,j) can stochastically spawn child walkers by updating either its left index, (i,j)→(k,j)(i,j) \rightarrow(k,j), with an amplitude proportional to ηki\eta_{ki}, or its right index, (i,j)→(i,l)(i,j) \rightarrow(i,l), with an amplitude proportional to −ηjl-\eta_{jl}. More precisely, given the proposal probabilities PL(k∣i)P_L(k|i) and PR(l∣j)P_R(l|j) for selecting the target indices, the expected

population changes are

ΔCkj=ΔsPL(k∣i)ηkiCij.(5)\Delta C_{kj} = \frac{\Delta s}{P_L(k|i)} \eta_{ki} C_{ij}. \tag*{(5)}
ΔCil=−ΔsPR(l∣j)ηjlCij.(6)\Delta C_{il} = -\frac{\Delta s}{P_R(l|j)} \eta_{jl} C_{ij}. \tag*{(6)}

In practice, the target indices kk and ll are sampled uniformly from the available off-diagonal mesh sites, and the signs of their corresponding spawned walkers are determined by the signs of ηkiCij\eta_{ki}C_{ij} and −ηjlCij-\eta_{jl}C_{ij}, respectively. Each spawning sweep is immediately followed by an annihilation step, where walkers arriving at the same site cancel out via algebraic summation of their signed populations. To guarantee physical consistency, Hermiticity is explicitly enforced at each step by symmetrizing the walker matrix, after which the generator η\eta is dynamically updated using the new stochastic Hamiltonian. This sequence delivers a faithful stochastic realization of the continuous SRG flow while completely bypassing full deterministic matrix-matrix multiplications.

A full stochastic trajectory from s=0s=0 to the target resolution scale λ=s−1/4\lambda= s^{-1/4} defines a single loop. To eliminate bias and compute robust errors, the evolution is repeated across NloopN_{\mathrm{loop}} independent loops with distinct random seed series, with statistical uncertainties quoted as standard errors of the mean. Details of uncertainty analysis are given in the Supplemental Material [23].

Figure 2. Correlation energy of the A=4A = 4 Richardson pairing Hamiltonian as a function of the pairing strength gg. The deterministic IMSRG(2), IMSRG(3) and FCI results are compared with SRGQMC(2), SRGQMC(3) and SRGQMC(4) calculations. Error bars denote the statistical uncertainties from independent stochastic loops. The FCI result is obtained by directly diagonalizing the Hamiltonian in the specified model space.

To illustrate the practical capability of this stochastic framework, we apply SRGQMC to the free-space evolution of the N3^{3}LOEMN_{\mathrm{EMN}}(500) interaction [24] down to a resolution scale of λ=2.0 fm−1\lambda= 2.0\ \mathrm{fm}^{-1}. Figure 1 presents the evolved interaction matrix elements V(p′,p)V(p',p) and the resulting phase shifts, with a walker population of Nw=104N_{\mathrm{w}} = 10^{4} and Nloop=10N_{\mathrm{loop}} = 10 independent loops. The first two rows compare the evolved interaction matrix elements computed by the deterministic SRG and by the loop-averaged SRGQMC for the 1S0{}^{1}S_{0}, 3P0{}^{3}P_{0}, and coupled 3S1{}^{3}S_{1}-3D1{}^{3}D_{1} channels. The excellent agreement between the interaction matrix elements evolved via SRG and SRGQMC clearly demonstrates that the stochastic dynamics faithfully achieves the objective of the SRG evolution, which is to suppress off-diagonal contributions.

A more sensitive validation is provided by the phase shifts shown in the third row of Fig. 1, where our results are also compared with the empirical Nijmegen partial-wave analysis [22] as a reference. Since the continuous flow is strictly unitary, these scattering observables serve

as a precise diagnosis to verify that the stochastic integration preserves the underlying physics. Quantitatively, the relative root-mean-square (RMS) deviations of the matrix elements between the stochastic and deterministic trajectories remain within 1.5%1.5\%–5.2%5.2\% across all the five channels. Crucially, both the matrix element variances and the minor phase-shift deviations (with the RMS differences below 0.60∘0.60^\circ) are largely encompassed by the quantified statistical error bars. Within the tested walker populations, no statistically significant finite-walker bias is observed, and residual stochastic fluctuations can be systematically suppressed by increasing NwN_{\mathrm{w}} or NloopN_{\mathrm{loop}}.

For many-body calculations.— We have tested many-body calculations using the Richardson pairing model [25–27].

H^=δ∑p=1pmax⁡∑σ=↑,↓(p−1)apσ†apσ−g2∑p,q=1pmax⁡ap↑†ap↓†aq↑aq↓(7)\hat{H} = \delta\sum_{p=1}^{p_{\max}} \sum_{\sigma=\uparrow,\downarrow} (p-1)a^\dagger_{p\sigma}a_{p\sigma} - \frac{g}{2}\sum_{p,q=1}^{p_{\max}}a^\dagger_{p\uparrow}a^\dagger_{p\downarrow}a_{q\uparrow}a_{q\downarrow} \tag*{(7)}

where δ\delta is the single-particle level spacing (set to 1.0 MeV) and gg represents the pairing strength. Although simplified, this model allows us to analyze how the calculation responds to the change in interaction strength. We evaluated a half-filled system with A=4A = 4 nucleons distributed across pmax⁡=4p_{\max} = 4 doubly-degenerate levels, varying gg from −2.0-2.0 to 2.02.0 MeV. Results are reported in terms of the correlation energy, EcorrE_{\mathrm{corr}}, which is obtained by subtracting the reference state energy from the total energy of the system.

In IMSRG framework, the many-body Hamiltonian is normal-ordered with respect to a reference state, and truncated to control the operator hierarchy [7].

H=E0+∑ijfij{ai†aj}+1(2!)2∑ijklΓijkl{ai†aj†alak}+1(3!)2∑ijklmnWijklmn{ai†aj†ak†anamal}+1(4!)2∑ijklmnopXijklmnop{ai†aj†ak†al†apaoanam}+⋯ .(8)\begin{aligned} H ={}& E_{0} + \sum_{ij} f_{ij}\{a^\dagger_i a_j\} + \frac{1}{(2!)^2}\sum_{ijkl}\Gamma_{ijkl}\{a^\dagger_i a^\dagger_j a_l a_k\} \\ &+ \frac{1}{(3!)^2}\sum_{ijklmn}W_{ijklmn}\{a^\dagger_i a^\dagger_j a^\dagger_k a_n a_m a_l\} \\ &+ \frac{1}{(4!)^2}\sum_{ijklmnop}X_{ijklmnop}\{a^\dagger_i a^\dagger_j a^\dagger_k a^\dagger_l a_p a_o a_n a_m\} \\ &+ \cdots. \tag*{(8)} \end{aligned}

The standard IMSRG(2) truncation restricts the hierarchy to the normal-ordered two-body operator Γ\Gamma, while IMSRG(3) and IMSRG(4) further retain the three-body (WW) and four-body (XX) tensors, respectively. The flow equation dH/ds=[η,H]\mathrm{d}H/\mathrm{d}s = [\eta,H] then yields a set of coupled equations of tensor contractions for these operators [23].

The core strategy of SRGQMC scales naturally to this many-body tensor space. Instead of tracking full matrices, walker sites are extended to multi-index configurations that identify both the operator rank and tensor indices (E0E_{0}, fijf_{ij}, Γijkl\Gamma_{ijkl}, WijklmnW_{ijklmn} or XijklmnopX_{ijklmnop}). Rather than calculating deterministic multi-loop sums over intermediate states, each diagrammatic contraction in the flow equation is converted into a corresponding stochastic sampling algorithm. All details of the deterministic flow equations and corresponding sampling algorithms can be found in the Supplemental Material [23]. The walker annihilation and loop-averaging algorithms are the same as those in the free-space case.

Figure 2 presents the calculated correlation energies EcorrE_{\mathrm{corr}} as a function of the coupling strength gg. SRGQMC(nn) stands for the calculation carried out at the level of nn-body operators. Using a population of Nw=104N_{\mathrm{w}} = 10^{4} walkers, SRGQMC(2) and SRGQMC(3) calculations reproduce their corresponding deterministic IMSRG results within the quantified statistical uncertainties with relative RMS deviations of 0.35% and 0.51% for the two- and three-body truncations. Figure 2 also illustrates the growing importance of higher-order correlations, omitted in standard IMSRG truncations, within the strongly correlated regime at larger coupling strengths. For ∣g∣≳1.0\lvert g\rvert\gtrsim1.0 MeV, IMSRG(2) rapidly deviates from the exact FCI curve, while the truncation at three-body level provides a remarkable improvement toward the exact solution. The inclusion of four-body operators through SRGQMC(4) (with Nw=106N_{\mathrm{w}} = 10^{6}) gives further improvement, well reproducing the FCI result. The comparisons indicate that three- and four-body correlations would play non-negligible roles in the descriptions of strongly correlated systems.

To provide an intuitive and dynamical picture of the stochastic behavior, Fig. 3 details the evolution of the correlation energy as a function of the flow parameter ss at g=1.0g = 1.0 MeV with different walker populations. We see that the loop-averaged stochastic trajectories at two- and three-body levels smoothly track toward their corresponding exact deterministic calculations as the flow progresses. Deterministic IMSRG(4) has not been available. We benchmark SRGQMC(4) against the exact FCI result which can be obtained for the A=4A = 4 system by directly diagonalizing the Hamiltonian. As expected, increasing the walker population remarkably reduces statistical fluctuations. The deterministic results remain within the SRGQMC uncertainty bands regardless of the walker population, suggesting the unbiased nature of the SRGQMC framework.

A rough CPU-time comparison at g=1.0g = 1.0 MeV further illustrates the practical advantage of SRGQMC. When going from two- to three-body truncations, the computational cost increases by a factor of only 23 for the stochastic SRGQMC (Nw=103N_{\mathrm{w}} = 10^{3} in Fig. 3), compared to 132 for the deterministic IMSRG. This suggests the more favorable practical scaling of the SRGQMC framework.

Summary and outlook.— In this Letter, we introduce the SRGQMC as a stochastic realization of continuous similarity renormalization group flows, replacing the explicit deterministic evolution of operator matrices with a walker sampling of the same commutator structures. This method preserves the decoupling behavior of SRG while providing fully quantified and controllable statistical uncertainties for the evolved operators and observables. We first validated this idea for the free-space SRG evolution of a realistic nucleon-nucleon interaction, where

Figure 3. SRGQMC flow of the correlation energy for the A=4A = 4 Richardson pairing Hamiltonian at g=1.0g = 1.0 MeV. Shaded bands indicate statistical uncertainties. Deterministic IMSRG(2), IMSRG(3) and FCI results are shown on the right axis by black dots.

SRGQMC faithfully reproduces evolved matrix elements and scattering phase shifts. We then benchmarked it against IMSRG(2), IMSRG(3), and full configuration-interaction calculations of the A=4A = 4 Richardson pairing model, demonstrating that stochastic trajectories exhibit excellent agreement with the deterministic many-body calculations across the entire range of coupling strengths.

The central implication of this work is that SRGQMC provides a highly practical and scalable route for pushing ab initio IMSRG calculations toward higher truncation orders. Deterministic IMSRG(4) calculations are prohibitive due to the combinatorial growth of induced higher-order operators and tensor contractions—precisely the regime where stochastic sampling can replace exhausting algebraic summations with statistically controlled estimates [27–31]. In this work, we performed the first IMSRG(4) calculation using SRGQMC. Accessing the higher-order correlations is crucial for high-density nuclear matter [32] and nuclei where collective excitations, deformation, and multi-nucleon clustering challenge current ab initio many-body calculations. Furthermore, because the algorithm samples the algebraic structure of continuous unitary transformations rather than model-specific details, the same strategy is readily adaptable to broader aspects across quantum chemistry, condensed matter physics and quantum information science where Hamiltonian flows are limited by rapid growth of operator space.

Acknowledgments.— This work has been supported by the National Key R&D Program of China under Grants Nos. 2024YFA1610900 and 2023YFA1606400; the National Natural Science Foundation of China under Grants Nos. 12335007, 12535008 and 2475118; and the High-Performance Computing Platform of Peking University.

Data availability.— The data that support the findings of this article are openly available [33].

Supplemental Material for: Stochastic Similarity Renormalization Group

R. Z. Hu; X. Zhen; F. R. Xu; J. C. Pei

SRGQMC Algorithm Details

In this section, we provide a more detailed description of the SRGQMC algorithm. Direct sampling in the full configuration space becomes prohibitive for large systems such as nuclear matter. We therefore place walkers on the matrix elements themselves, treating the evolved Hamiltonian as the sampled object in operator space. In this representation, the spawning and annihilation processes are designed to satisfy the SRG flow equations, yielding an unbiased stochastic realization of the evolution at each step. The commutator of normal-ordered operators generally induces operators of higher particle rank. For two normal-ordered operators containing up to mm- and nn-body terms, respectively, their commutator generates contributions ranging from ∣m−n∣\lvert m-n\rvert-body to (m+n−1)(m+n-1)-body operators,

[A(m),B(n)]=∑k=∣m−n∣m+n−1C(k).(9)\left[A^{(m)}, B^{(n)}\right] = \sum_{k=\lvert m-n\rvert}^{m+n-1} C^{(k)}. \tag*{(9)}

In IMSRG(nn), all operators and commutators are truncated at the normal-ordered nn-body level. Once the relevant commutators have been evaluated, they can be used to construct the flow equations for both deterministic IMSRG and stochastic SRGQMC, differing only in whether the resulting evolution is represented and solved deterministically or stochastically. It is worth noting that, unlike the situation encountered in most applications to finite nuclei, the momentum-space matrix elements in nuclear matter are generally complex-valued. As a result, several simplifications that rely on real-valued matrix elements are no longer applicable, and the resulting flow equations differ slightly from their more familiar forms used in finite nuclei calculations.

In the following expressions, the occupation number npn_p is defined in the usual way as

np={1,for hole states,0,for particle states.(10)n_p = \begin{cases} 1, & \text{for hole states},\\ 0, & \text{for particle states}. \end{cases} \tag*{(10)}

We also define nˉp≡1−np\bar{n}_p \equiv1-n_p. The exchange operators are defined by

P(pq⋯/rs⋯ )≡1−Ppr−Pps−⋯−Pqr−Pqs−⋯ ,(11)P(pq\cdots/rs\cdots) \equiv1-P_{pr}-P_{ps}-\cdots-P_{qr}-P_{qs}-\cdots, \tag*{(11)}

where the elementary exchange operator PabP_{ab} interchanges the indices aa and bb in the expression on which it acts. All matrix elements are assumed to be fully antisymmetrized.

A final algorithmic detail concerns the control of the total walker population. At s=0s=0, the walker representation is normalized through

N(0)=1Nw∑ij∣Hij(0)∣,(12)\mathcal{N}(0)=\frac{1}{N_{\mathrm{w}}}\sum_{ij}\lvert H_{ij}(0)\rvert, \tag*{(12)}

so that Hij(0)=N(0)Cij(0)H_{ij}(0)=\mathcal{N}(0)C_{ij}(0) and the initial total population is fixed by the target value NwN_{\mathrm{w}}. During the subsequent evolution, spawning and annihilation generally change the instantaneous population N~=∑ij∣C~ij∣\widetilde{\mathcal{N}}=\sum_{ij}\lvert\widetilde{C}_{ij}\rvert. After each step, we therefore apply a global rescaling Cij←(Nw/N~)C~ijC_{ij}\leftarrow(N_{\mathrm{w}}/\widetilde{\mathcal{N}})\widetilde{C}_{ij} and update the physical scale factor as N←NN~/Nw\mathcal{N}\leftarrow\mathcal{N}\widetilde{\mathcal{N}}/N_{\mathrm{w}}. The accumulated product of these scale factors is used when reconstructing the physical Hamiltonian and observables, so this population control changes only the stochastic representation, not the underlying SRGQMC estimate.

SRGQMC(1)

There are two commutators included at this level:

[A(1),B(1)](0)=∑ijAijBji(ni−nj),(13)[A^{(1)}, B^{(1)}]^{(0)} = \sum_{ij} A_{ij}B_{ji}(n_i-n_j), \tag*{(13)}
[A(1),B(1)](1)=∑ij∑aai†aj(AiaBaj−BiaAaj).(14)[A^{(1)}, B^{(1)}]^{(1)} = \sum_{ij}\sum_a a_i^\dagger a_j (A_{ia}B_{aj}-B_{ia}A_{aj}). \tag*{(14)}

The corresponding flow equations of IMSRG(1) are

dEds=∑ij(ni−nj)ηijfij,(15)\frac{\mathrm{d}E}{\mathrm{d}s} = \sum_{ij}(n_i-n_j)\eta_{ij}f_{ij}, \tag*{(15)}
dfijds=∑a(ηiafaj−fiaηaj).(16)\frac{\mathrm{d}f_{ij}}{\mathrm{d}s} = \sum_a(\eta_{ia}f_{aj}-f_{ia}\eta_{aj}). \tag*{(16)}

These terms contribute for a general Hamiltonian. In nuclear matter, however, translational invariance ensures that the one-body operator ff remains diagonal throughout the evolution. Consequently, the one-body part of the generator, η(1)\eta^{(1)}, vanishes identically, and these terms do not contribute in the present case. We nevertheless include them here for completeness.

The situation is slightly different in SRGQMC. In this formulation, walkers reside on the matrix elements of HH and stochastically spawn so as to reproduce the corresponding flow equations. Thus, the same flow equations should be interpreted in the opposite direction: instead of evaluating a given matrix element from all contributing terms in the deterministic sense, each occupied matrix element acts as a source that spawns walkers to the matrix elements to which it contributes. After relabeling the indices accordingly, the flow equations can be read directly as the spawning rules of walkers:

ΔE=−Δs∑ab(na−nb)ηbafab,(17)\Delta E=-\Delta s\sum_{ab}(n_a-n_b)\eta_{ba}f_{ab}, \tag*{(17)}
Δfib=Δs∑aηiafab,(18)\Delta f_{ib}=\Delta s\sum_a\eta_{ia}f_{ab}, \tag*{(18)}
Δfai=−Δs∑bηbifab.(19)\Delta f_{ai}=-\Delta s\sum_b\eta_{bi}f_{ab}. \tag*{(19)}

We now use SRGQMC(1) as a simple example to illustrate the trajectory of a walker. Consider a walker located on the one-body matrix element fabf_{ab}, which carries two indices aa and bb. To satisfy Eq. (17), this walker must identify the corresponding generator matrix element, ηba=−ηab∗\eta_{ba}=-\eta_{ab}^{*}, and contribute to the increment of the normal-ordered zero-body energy through a simple product. For Eqs. (18) and (19), however, the walker must keep one of its original indices fixed and randomly sample an additional index ii. Once ii is selected, the walker contributes simultaneously to the matrix elements faif_{ai} and fibf_{ib}, with weights determined by the corresponding terms in the flow equation. Equivalently, the two spawning channels can be combined into a single heat-bath distribution. Defining

Zab=∑i∣ηia∣+∑i∣ηbi∣,(20)Z_{ab}=\sum_i\lvert\eta_{ia}\rvert+\sum_i\lvert\eta_{bi}\rvert, \tag*{(20)}

we choose the left channel fab→fibf_{ab}\to f_{ib} with probability

PL(i∣a,b)=∣ηia∣Zab,(21)P_L(i\mid a,b)=\frac{\lvert\eta_{ia}\rvert}{Z_{ab}}, \tag*{(21)}

and the right channel fab→faif_{ab}\to f_{ai} with probability

PR(i∣a,b)=∣ηbi∣Zab.(22)P_R(i\mid a,b)=\frac{\lvert\eta_{bi}\rvert}{Z_{ab}}. \tag*{(22)}

The corresponding spawned weights are then

wibspawn=Δs wabηiaPL(i∣a,b),(23)w_{ib}^{\mathrm{spawn}}=\Delta s\,w_{ab}\frac{\eta_{ia}}{P_L(i\mid a,b)}, \tag*{(23)}
waispawn=−Δs wabηbiPR(i∣a,b)(24)w_{ai}^{\mathrm{spawn}}=-\Delta s\,w_{ab}\frac{\eta_{bi}}{P_R(i\mid a,b)} \tag*{(24)}

With this heat-bath sampling strategy, each individual term in the flow equations can be estimated without bias at a given step. However, since the generator η\eta depends on the stochastically sampled Hamiltonian HH, the full flow equation is globally nonlinear. In general,

⟨[η(H),H]⟩≠[η(⟨H⟩),⟨H⟩],\langle[\eta(H),H]\rangle\ne[\eta(\langle H\rangle),\langle H\rangle],

so the stochastic evolution possibly would accumulate a finite-population sampling bias relative to the exact deterministic flow. However as discussed in the following section, this possible bias can be efficiently eliminated by increasing the walker population and/or averaging over more independent loops. After the spawning step, all walkers accumulated on Δf\Delta f are collected and annihilated (two walkers with different signs on the same site cancel with each other), yielding the total stochastic increment used for the next evolution step.

In the following discussion of SRGQMC(2) and SRGQMC(3), we no longer explain each term individually. Instead, we present the corresponding commutators and the flow equations in both the deterministic IMSRG and stochastic SRGQMC forms. Because the occupation-number structure becomes increasingly involved, we first introduce several compact notations. We define

OabB≡na−nb,OabC=1−na−nb,(25)O^B_{ab}\equiv n_a-n_b,\qquad O^C_{ab}=1-n_a-n_b, \tag*{(25)}

and also the symmetric and antisymmetric occupation-number combinations

Oab…/cd…S≡nanb…nˉcnˉd…+nˉanˉb…ncnd…,Oab…/cd…A≡nanb…nˉcnˉd…−nˉanˉb…ncnd….(26)O^S_{ab\ldots/cd\ldots}\equiv n_an_b\ldots\bar n_c\bar n_d\ldots+\bar n_a\bar n_b\ldots n_cn_d\ldots,\qquad O^A_{ab\ldots/cd\ldots}\equiv n_an_b\ldots\bar n_c\bar n_d\ldots-\bar n_a\bar n_b\ldots n_cn_d\ldots. \tag*{(26)}
SRGQMC(2)

There are five more commutators included at this level:

[A(1),B(2)](1)=∑ij∑ab:ai†aj:OabBAabBbiaj,(27)[A^{(1)},B^{(2)}]^{(1)}=\sum_{ij}\sum_{ab}:a_i^\dagger a_j:O^B_{ab}A_{ab}B_{biaj}, \tag*{(27)}
[A(1),B(2)](2)=14∑ijkl∑a:ai†aj†alak:{P(i/j)AiaBajkl−P(k/l)AakBijal},(28)[A^{(1)},B^{(2)}]^{(2)}=\frac{1}{4}\sum_{ijkl}\sum_a:a_i^\dagger a_j^\dagger a_l a_k:\left\{P(i/j)A_{ia}B_{ajkl}-P(k/l)A_{ak}B_{ijal}\right\}, \tag*{(28)}
[A(2),B(2)](0)=14∑ijklOij/klAAijklBklij,(29)[A^{(2)},B^{(2)}]^{(0)}=\frac{1}{4}\sum_{ijkl}O^A_{ij/kl}A_{ijkl}B_{klij}, \tag*{(29)}
[A(2),B(2)](1)=12∑ij∑abc:ai†aj:Oab/cS(AciabBabcj−BciabAabcj),(30)[A^{(2)},B^{(2)}]^{(1)}=\frac{1}{2}\sum_{ij}\sum_{abc}:a_i^\dagger a_j:O^S_{ab/c}\left(A_{ciab}B_{abcj}-B_{ciab}A_{abcj}\right), \tag*{(30)}
[A(2),B(2)](2)=14∑ijkl∑ab:ai†aj†alak:{12OabC(AijabBabkl−BijabAabkl)+OabBP(i/j)P(k/l)AaibkBbjal}.(31)[A^{(2)},B^{(2)}]^{(2)}=\frac{1}{4}\sum_{ijkl}\sum_{ab}:a_i^\dagger a_j^\dagger a_l a_k:\left\{\frac{1}{2}O^C_{ab}\left(A_{ijab}B_{abkl}-B_{ijab}A_{abkl}\right)+O^B_{ab}P(i/j)P(k/l)A_{aibk}B_{bjal}\right\}. \tag*{(31)}

The corresponding IMSRG(2) flow equations are

dEds=∑ijOijBηijfji+14∑ijklOij/klAηijklΓklij,(32)\frac{\mathrm{d}E}{\mathrm{d}s}=\sum_{ij}O^B_{ij}\eta_{ij}f_{ji}+\frac{1}{4}\sum_{ijkl}O^A_{ij/kl}\eta_{ijkl}\Gamma_{klij}, \tag*{(32)}
dfijds=∑a(ηiafaj−fiaηaj)+∑abOabB(ηabΓbiaj−fabηbiaj)+12∑abcOab/cS(ηciabΓabcj−Γciabηabcj),(33)\frac{\mathrm{d}f_{ij}}{\mathrm{d}s}=\sum_a(\eta_{ia}f_{aj}-f_{ia}\eta_{aj})+\sum_{ab}O^B_{ab}(\eta_{ab}\Gamma_{biaj}-f_{ab}\eta_{biaj})+\frac{1}{2}\sum_{abc}O^S_{ab/c}(\eta_{ciab}\Gamma_{abcj}-\Gamma_{ciab}\eta_{abcj}), \tag*{(33)}
dΓijklds=∑a[P(i/j)(ηiaΓajkl−fiaηijkl)−P(k/l)(ηakΓijal−fakηijal)]+∑ab[12OabC(ηijabΓabkl−Γijabηabkl)+OabBP(i/j)P(k/l)ηaibkΓbjal].(34)\begin{aligned} \frac{\mathrm{d}\Gamma_{ijkl}}{\mathrm{d}s}={}&\sum_a\left[P(i/j)(\eta_{ia}\Gamma_{ajkl}-f_{ia}\eta_{ijkl})-P(k/l)(\eta_{ak}\Gamma_{ijal}-f_{ak}\eta_{ijal})\right]\\ &+\sum_{ab}\left[\frac{1}{2}O^C_{ab}(\eta_{ijab}\Gamma_{abkl}-\Gamma_{ijab}\eta_{abkl})+O^B_{ab}P(i/j)P(k/l)\eta_{aibk}\Gamma_{bjal}\right]. \tag*{(34)} \end{aligned}

The standard flow equation above can also be found in Ref. [1]. The spawning rules of SRGQMC(2) are:

ΔE=−Δs∑abOabBηbafab+14Δs∑abcdOab/cdAηabcdΓcdab,(35)\Delta E=-\Delta s\sum_{ab}O^B_{ab}\eta_{ba}f_{ab}+\frac{1}{4}\Delta s\sum_{abcd}O^A_{ab/cd}\eta_{abcd}\Gamma_{cdab}, \tag*{(35)}
Δfib=Δs∑aηiafab,Δfai=−Δs∑bηbifab,(36)\Delta f_{ib}=\Delta s\sum_a\eta_{ia}f_{ab},\quad\Delta f_{ai}=-\Delta s\sum_b\eta_{bi}f_{ab}, \tag*{(36)}
Δfij=−Δs∑abOabBηbiajfab,(37)\Delta f_{ij}=-\Delta s\sum_{ab}O^{B}_{ab}\eta_{bi}a_{j}f_{ab}, \tag*{(37)}
ΔΓajkl−ΔΓjakl=−Δs∑bηbjklfab,ΔΓijbl−ΔΓijlb=Δs∑aηijalfab,(38)\Delta\Gamma_{ajkl}-\Delta\Gamma_{jakl}=-\Delta s\sum_b\eta_{bjkl}f_{ab},\quad\Delta\Gamma_{ijbl}-\Delta\Gamma_{ijlb}=\Delta s\sum_a\eta_{ijal}f_{ab}, \tag*{(38)}
Δfbd=Δs∑acOcaBηcaΓabcd,(39)\Delta f_{bd}=\Delta s\sum_{ac}O^{B}_{ca}\eta_{ca}\Gamma_{abcd}, \tag*{(39)}
ΔΓibcd−ΔΓbicd=Δs∑aηiaΓabcd,ΔΓabid−ΔΓabdi=−Δs∑cηciΓabcd,(40)\Delta\Gamma_{ibcd}-\Delta\Gamma_{bicd}=\Delta s\sum_a\eta_{ia}\Gamma_{abcd},\quad\Delta\Gamma_{abid}-\Delta\Gamma_{abdi}=-\Delta s\sum_c\eta_{ci}\Gamma_{abcd}, \tag*{(40)}
Δfid=12Δs∑abcOab/cSηciabΓabcd,Δfbi=−12Δs∑acdOcd/aSηcdaiΓabcd,(41)\Delta f_{id}=\frac{1}{2}\Delta s\sum_{abc}O^{S}_{ab/c}\eta_{ciab}\Gamma_{abcd},\quad\Delta f_{bi}=-\frac{1}{2}\Delta s\sum_{acd}O^{S}_{cd/a}\eta_{cdai}\Gamma_{abcd}, \tag*{(41)}
ΔΓijcd=12Δs∑abOabCηijabΓabcd,ΔΓabkl=−12Δs∑cdOcdCηcdklΓabcd,(42)\Delta\Gamma_{ijcd}=\frac{1}{2}\Delta s\sum_{ab}O^{C}_{ab}\eta_{ijab}\Gamma_{abcd},\quad\Delta\Gamma_{abkl}=-\frac{1}{2}\Delta s\sum_{cd}O^{C}_{cd}\eta_{cdkl}\Gamma_{abcd}, \tag*{(42)}
ΔΓibjd−ΔΓbijd−ΔΓibdj+ΔΓbidj=Δs∑acOcaBηciajΓabcd.(43)\Delta\Gamma_{ibjd}-\Delta\Gamma_{bijd}-\Delta\Gamma_{ibdj}+\Delta\Gamma_{bidj}=\Delta s\sum_{ac}O^{B}_{ca}\eta_{ciaj}\Gamma_{abcd}. \tag*{(43)}
SRGQMC(3)

The IMSRG(3) flow equations have been presented in several previous works [1, 2, 3]. Since these expressions are algebraically involved and minor typographical inconsistencies can easily arise, we have independently re-derived the following equations in order to ensure consistency and accuracy.

There are ten more commutators appearing at this level:

[A(1),B(3)](2)=14∑ijkl∑ab:ai†aj†alak:OabBAabBbijakl(44)[A^{(1)},B^{(3)}]^{(2)}=\frac{1}{4}\sum_{ijkl}\sum_{ab}:a_i^\dagger a_j^\dagger a_l a_k: O^{B}_{ab}A_{ab}B_{b i j a k l} \tag*{(44)}
[A(1),B(3)](3)=136∑ijklmn∑a:ai†aj†ak†anamal:{P(i/jk)AiaBajklmn−P(l/mn)AalBijkamn}(45)[A^{(1)},B^{(3)}]^{(3)}=\frac{1}{36}\sum_{ijklmn}\sum_{a}:a_i^\dagger a_j^\dagger a_k^\dagger a_n a_m a_l: \left\{P(i/jk)A_{ia}B_{ajklmn}-P(l/mn)A_{al}B_{ijkamn}\right\} \tag*{(45)}
[A(2),B(2)](3)=136∑ijklmn∑a:ai†aj†ak†anamal:{P(ij/k)P(l/mn)(AijlaBakmn−BijlaAakmn)}(46)[A^{(2)},B^{(2)}]^{(3)}=\frac{1}{36}\sum_{ijklmn}\sum_a:a_i^\dagger a_j^\dagger a_k^\dagger a_n a_m a_l:\left\{P(ij/k)P(l/mn)\left(A_{ijla}B_{akmn}-B_{ijla}A_{akmn}\right)\right\} \tag*{(46)}
[A(2),B(3)](1)=−14∑ij∑abcd:ai†aj:Oab/cdAAcdabBabijcd(47)[A^{(2)},B^{(3)}]^{(1)}=-\frac{1}{4}\sum_{ij}\sum_{abcd}:a_i^\dagger a_j:O^{A}_{ab/cd}A_{cdab}B_{abijcd} \tag*{(47)}
[A(2),B(3)](2)=−18∑ijkl∑abc:ai†aj†alak:Oab/cS{P(i/j)AicabBabjklc−P(k/l)AabkcBijcabl}(48)[A^{(2)},B^{(3)}]^{(2)}=-\frac{1}{8}\sum_{ijkl}\sum_{abc}:a_i^\dagger a_j^\dagger a_l a_k:O^{S}_{ab/c}\left\{P(i/j)A_{icab}B_{abjklc}-P(k/l)A_{abkc}B_{ijcabl}\right\} \tag*{(48)}
[A(2),B(3)](3)=136∑ijklmn∑ab:ai†aj†ak†anamal:×{12OabC[P(ij/k)AijabBabklmn−P(lm/n)AablmBijkabn]−OabBP(ij/k)P(lm/n)AbkanBijalmb}(49)[A^{(2)},B^{(3)}]^{(3)}=\frac{1}{36}\sum_{ijklmn}\sum_{ab}:a_i^\dagger a_j^\dagger a_k^\dagger a_n a_m a_l: \times\left\{\frac{1}{2}O^{C}_{ab}\left[P(ij/k)A_{ijab}B_{abklmn}-P(lm/n)A_{ablm}B_{ijkabn}\right]-O^{B}_{ab}P(ij/k)P(lm/n)A_{bkan}B_{ijalm b}\right\} \tag*{(49)}
[A(3),B(3)](0)=136∑ijklmnOijk/lmnAAijklmnBlmnijk(50)[A^{(3)},B^{(3)}]^{(0)}=\frac{1}{36}\sum_{ijklmn}O^{A}_{ijk/lmn}A_{ijklmn}B_{lmnijk} \tag*{(50)}
[A(3),B(3)](1)=112∑ij∑abcde:ai†aj:Oab/cdeS(AabicdeBcdeabj−BabicdeAcdeabj)(51)[A^{(3)},B^{(3)}]^{(1)}=\frac{1}{12}\sum_{ij}\sum_{abcde}:a_i^\dagger a_j:O^{S}_{ab/cde}\left(A_{abicde}B_{cdeabj}-B_{abicde}A_{cdeabj}\right) \tag*{(51)}
[A(3),B(3)](2)=14∑ijkl∑abcd:ai†aj†alak:×{16Oa/bcdA(AaijbcdBbcdakl−AbcdaklBaijbcd)−14Oab/cdA(1−Pij)(1−Pkl)AabicdlBcdjabk}(52)[A^{(3)},B^{(3)}]^{(2)}=\frac{1}{4}\sum_{ijkl}\sum_{abcd}:a_i^\dagger a_j^\dagger a_l a_k: \times\left\{\frac{1}{6}O^{A}_{a/bcd}\left(A_{aijbcd}B_{bcdakl}-A_{bcdakl}B_{aijbcd}\right)-\frac{1}{4}O^{A}_{ab/cd}(1-P_{ij})(1-P_{kl})A_{abicdl}B_{cdjabk}\right\} \tag*{(52)}
[A(3),B(3)](3)=136∑ijklmn∑abc:ai†aj†ak†anamal:×{16OabcB(AijkabcBabclmn−BijkabcAabclmn)−12Oab/cSP(ij/k)P(lm/n)(AijcabnBabklmc−AabklmcBijcabn)}(53)[A^{(3)},B^{(3)}]^{(3)}=\frac{1}{36}\sum_{ijklmn}\sum_{abc}:a_i^\dagger a_j^\dagger a_k^\dagger a_n a_m a_l: \times\left\{\frac{1}{6}O^B_{abc}(A_{ijkabc}B_{abclmn}-B_{ijkabc}A_{abclmn})-\frac{1}{2}O^S_{ab/c}P(ij/k)P(lm/n)(A_{ijcabn}B_{abklmc}-A_{abklmc}B_{ijcabn})\right\} \tag*{(53)}

The corresponding IMSRG(3) flow equations are

dEds=∑ijOijBηijfij+14∑ijklOij/klAηijklΓklij+136∑ijklmnOijk/lmnAηijklmnWlmnijk,(54)\frac{dE}{ds}=\sum_{ij}O^B_{ij}\eta_{ij}f_{ij}+\frac{1}{4}\sum_{ijkl}O^A_{ij/kl}\eta_{ijkl}\Gamma_{klij}+\frac{1}{36}\sum_{ijklmn}O^A_{ijk/lmn}\eta_{ijklmn}W_{lmnijk}, \tag*{(54)}
dfijds=∑a(ηiafaj−fiaηaj)+∑abOabB(ηabΓbiaj−fabηbiaj)+12∑abcOab/cS(ηciabΓabcj−Γciabηabcj)−14∑abcdOab/cdA(ηcdabWabijcd−Γcdabηabijcd)+112∑abcdeOab/cdeS(ηabicdeWcdeabj−Wabicdeηcdeabj),(55)\frac{df_{ij}}{ds}=\sum_a(\eta_{ia}f_{aj}-f_{ia}\eta_{aj})+\sum_{ab}O^B_{ab}(\eta_{ab}\Gamma_{biaj}-f_{ab}\eta_{biaj})+\frac{1}{2}\sum_{abc}O^S_{ab/c}(\eta_{ciab}\Gamma_{abcj}-\Gamma_{ciab}\eta_{abcj}) -\frac{1}{4}\sum_{abcd}O^A_{ab/cd}(\eta_{cdab}W_{abijcd}-\Gamma_{cdab}\eta_{abijcd})+\frac{1}{12}\sum_{abcde}O^S_{ab/cde}(\eta_{abicde}W_{cdeabj}-W_{abicde}\eta_{cdeabj}), \tag*{(55)}
dΓijklds=∑a[P(i/j)(ηiaΓajkl−fiaηajkl)−P(k/l)(ηakΓijal−fakηijal)]+∑abOabB(ηabWbijakl−fabηbijakl)+∑ab[12OabC(ηijabΓabkl−Γijabηabkl)+OabBP(i/j)P(k/l)ηaibkΓbjal]−12∑abcOab/cS{P(i/j)(ηicabWabjklc−Γicabηabjklc)−P(k/l)(ηabkcWijcabl−Γabkcηijcabl)}+∑abcd{16Oab/cdA(ηaijbcdWbcdakl−Waijbcdηbcdakl)−14Oab/cdA(1−Pij)(1−Pkl)ηabicdlWcdjabk},(56)\begin{aligned} \frac{d\Gamma_{ijkl}}{ds}={}&\sum_a\left[P(i/j)(\eta_{ia}\Gamma_{ajkl}-f_{ia}\eta_{ajkl})-P(k/l)(\eta_{ak}\Gamma_{ijal}-f_{ak}\eta_{ijal})\right]\\ &+\sum_{ab}O^B_{ab}(\eta_{ab}W_{bijakl}-f_{ab}\eta_{bijakl})\\ &+\sum_{ab}\left[\frac{1}{2}O^C_{ab}(\eta_{ijab}\Gamma_{abkl}-\Gamma_{ijab}\eta_{abkl})+O^B_{ab}P(i/j)P(k/l)\eta_{aibk}\Gamma_{bjal}\right]\\ &-\frac{1}{2}\sum_{abc}O^S_{ab/c}\left\{P(i/j)(\eta_{icab}W_{abjklc}-\Gamma_{icab}\eta_{abjklc})-P(k/l)(\eta_{abkc}W_{ijcabl}-\Gamma_{abkc}\eta_{ijcabl})\right\}\\ &+\sum_{abcd}\left\{\frac{1}{6}O^A_{ab/cd}(\eta_{aijbcd}W_{bcdakl}-W_{aijbcd}\eta_{bcdakl})-\frac{1}{4}O^A_{ab/cd}(1-P_{ij})(1-P_{kl})\eta_{abicdl}W_{cdjabk}\right\}, \tag*{(56)} \end{aligned}
dWijklmnds=∑a{P(i/jk)(ηiaWajklmn−fiaηajklmn)−P(l/mn)(ηalWijkamn−falηijkamn)}+∑a{P(ij/k)P(l/mn)(ηijlaΓakmn−Γijlaηakmn)}−∑abOabBP(ij/k)P(lm/n)(ηbkanWijalmb−Γbkanηijalmb)+∑ab{12OabC[P(ij/k)(ηijabWabklmn−Γijabηabklmn)−P(lm/n)(ηablmWijkabn−Γablmηijkabn)]}+∑abc{16OabcS(ηijkabcWabclmn−Wijkabcηabclmn)−12Oab/cSP(ij/k)P(lm/n)(ηijcabnWabklmc−Wijcabnηabklmc)}.(57)\begin{aligned} \frac{dW_{ijklmn}}{ds}={}&\sum_a\left\{P(i/jk)(\eta_{ia}W_{ajklmn}-f_{ia}\eta_{ajklmn})-P(l/mn)(\eta_{al}W_{ijkamn}-f_{al}\eta_{ijkamn})\right\}\\ &+\sum_a\left\{P(ij/k)P(l/mn)(\eta_{ijla}\Gamma_{akmn}-\Gamma_{ijla}\eta_{akmn})\right\}-\sum_{ab}O^B_{ab}P(ij/k)P(lm/n)(\eta_{bkan}W_{ijalmb}-\Gamma_{bkan}\eta_{ijalmb})\\ &+\sum_{ab}\left\{\frac{1}{2}O^C_{ab}\left[P(ij/k)(\eta_{ijab}W_{abklmn}-\Gamma_{ijab}\eta_{abklmn})-P(lm/n)(\eta_{ablm}W_{ijkabn}-\Gamma_{ablm}\eta_{ijkabn})\right]\right\}\\ &+\sum_{abc}\left\{\frac{1}{6}O^S_{abc}(\eta_{ijkabc}W_{abclmn}-W_{ijkabc}\eta_{abclmn})-\frac{1}{2}O^S_{ab/c}P(ij/k)P(lm/n)(\eta_{ijcabn}W_{abklmc}-W_{ijcabn}\eta_{abklmc})\right\}. \tag*{(57)} \end{aligned}

The spawning rules of SRGQMC(3) are:

ΔE=−Δs∑abOabBηbafab+14Δs∑abcdOab/cdAηabcdΓcdab+136Δs∑abcdefOabc/defAηabcdefWdefabc,(58)\Delta E=-\Delta s\sum_{ab}O^B_{ab}\eta_{ba}f_{ab}+\frac{1}{4}\Delta s\sum_{abcd}O^A_{ab/cd}\eta_{abcd}\Gamma_{cdab}+\frac{1}{36}\Delta s\sum_{abcdef}O^A_{abc/def}\eta_{abcdef}W_{defabc}, \tag*{(58)}
Δfib=Δs∑aηiafab,Δfai=−Δs∑bηbifab,(59)\Delta f_{ib}=\Delta s\sum_a\eta_{ia}f_{ab},\quad\Delta f_{ai}=-\Delta s\sum_b\eta_{bi}f_{ab}, \tag*{(59)}
Δfij=−Δs∑abOabBηbiajfab,(60)\Delta f_{ij}=-\Delta s\sum_{ab}O^B_{ab}\eta_{biaj}f_{ab}, \tag*{(60)}
ΔΓajkl,−ΔΓjakl=−Δs∑bηbjklfab,ΔΓijbl,−ΔΓijlb=Δs∑aηijalfab,(61)\Delta\Gamma_{ajkl},-\Delta\Gamma_{jakl}=-\Delta s\sum_b\eta_{bjkl}f_{ab},\quad\Delta\Gamma_{ijbl},-\Delta\Gamma_{ijlb}=\Delta s\sum_a\eta_{ijal}f_{ab}, \tag*{(61)}
ΔΓijkl=Δs∑abObaBηbijaklfab,(62)\Delta\Gamma_{ijkl}=\Delta s\sum_{ab}O^B_{ba}\eta_{bijakl}f_{ab}, \tag*{(62)}
ΔWajklmn,−ΔWjaklmn,−ΔWkjalmn=−Δs∑bηbjklmnfab,(63)\Delta W_{ajklmn},-\Delta W_{jaklmn},-\Delta W_{kjalmn}=-\Delta s\sum_b\eta_{bjklmn}f_{ab}, \tag*{(63)}
ΔWijkbmn,−ΔWijkmbn,−ΔWijknmb=Δs∑aηijkamnfab,(64)\Delta W_{ijkbmn},-\Delta W_{ijkmbn},-\Delta W_{ijknmb}=\Delta s\sum_a\eta_{ijkamn}f_{ab}, \tag*{(64)}
Δfbd=Δs∑acOcaBηcaΓabcd,(65)\Delta f_{bd}=\Delta s\sum_{ac}O^B_{ca}\eta_{ca}\Gamma_{abcd}, \tag*{(65)}
ΔΓibcd,−ΔΓbicd=Δs∑aηiaΓabcd,ΔΓabid,−ΔΓabdi=−Δs∑cηciΓabcd,(66)\Delta\Gamma_{ibcd},-\Delta\Gamma_{bicd}=\Delta s\sum_a\eta_{ia}\Gamma_{abcd},\quad\Delta\Gamma_{abid},-\Delta\Gamma_{abdi}=-\Delta s\sum_c\eta_{ci}\Gamma_{abcd}, \tag*{(66)}
Δfid=12Δs∑abcOab/cSηciabΓabcd,Δfbi=−12Δs∑acdOcd/aSηcdaiΓabcd,(67)\Delta f_{id} = \frac{1}{2}\Delta s \sum_{abc} O^{S}_{ab/c}\eta_{ciab}\Gamma_{abcd},\quad\Delta f_{bi} = -\frac{1}{2}\Delta s \sum_{acd} O^{S}_{cd/a}\eta_{cdai}\Gamma_{abcd}, \tag*{(67)}
ΔΓijcd=12Δs∑abOabCηijabΓabcd,ΔΓabkl=−12Δs∑cdOcdCηcdklΓabcd,(68)\Delta\Gamma_{ijcd} = \frac{1}{2}\Delta s \sum_{ab} O^{C}_{ab}\eta_{ijab}\Gamma_{abcd},\quad\Delta\Gamma_{abkl} = -\frac{1}{2}\Delta s \sum_{cd} O^{C}_{cd}\eta_{cdkl}\Gamma_{abcd}, \tag*{(68)}
ΔΓibjd,−ΔΓbijd,−ΔΓibdj,ΔΓbidj=Δs∑acOcaBηciajΓabcd,(69)\Delta\Gamma_{ibjd},-\Delta\Gamma_{bijd},-\Delta\Gamma_{ibdj},\Delta\Gamma_{bidj} = \Delta s \sum_{ac} O^{B}_{ca}\eta_{ciaj}\Gamma_{abcd}, \tag*{(69)}
P(ij/b)P(l/cd)ΔWijblcd=Δs∑aηijlaΓabcd,P(ab/k)P(c/mn)ΔWabkcmn=−Δs∑dηdkmnΓabcd,(70)P(ij/b)P(l/cd)\Delta W_{ijblcd} = \Delta s \sum_{a}\eta_{ijla}\Gamma_{abcd},\quad P(ab/k)P(c/mn)\Delta W_{abkcmn} = -\Delta s \sum_{d}\eta_{dkmn}\Gamma_{abcd}, \tag*{(70)}
Δfij=14Δs∑abcdOcd/abAηcdijabΓabcd,(71)\Delta f_{ij} = \frac{1}{4}\Delta s \sum_{abcd} O^{A}_{cd/ab}\eta_{cdijab}\Gamma_{abcd}, \tag*{(71)}
ΔΓajkl,−ΔΓjakl=12Δs∑bcdOcd/bSηcdjklbΓabcd,ΔΓijcl,−ΔΓijlc=−12Δs∑abdOab/dSηijdablΓabcd,(72)\Delta\Gamma_{ajkl},-\Delta\Gamma_{jakl} = \frac{1}{2}\Delta s \sum_{bcd} O^{S}_{cd/b}\eta_{cdjklb}\Gamma_{abcd},\quad\Delta\Gamma_{ijcl},-\Delta\Gamma_{ijlc} = -\frac{1}{2}\Delta s \sum_{abd} O^{S}_{ab/d}\eta_{ijdabl}\Gamma_{abcd}, \tag*{(72)}
P(ij/b)P(lm/d)ΔWijblmd=Δs∑acOcaBηijclmaΓabcd,(73)P(ij/b)P(lm/d)\Delta W_{ijblmd} = \Delta s \sum_{ac} O^{B}_{ca}\eta_{ijclma}\Gamma_{abcd}, \tag*{(73)}
P(ab/k)ΔWabklmn=−12Δs∑cdOcdCηcdklmnΓabcd,P(cd/n)ΔWijkcdn=12Δs∑abOabCηijkabnΓabcd,(74)P(ab/k)\Delta W_{abklmn} = -\frac{1}{2}\Delta s \sum_{cd} O^{C}_{cd}\eta_{cdklmn}\Gamma_{abcd},\quad P(cd/n)\Delta W_{ijkcdn} = \frac{1}{2}\Delta s \sum_{ab} O^{C}_{ab}\eta_{ijkabn}\Gamma_{abcd}, \tag*{(74)}
ΔΓbcef=Δs∑adOdaBηdaWabcdef,(75)\Delta\Gamma_{bcef} = \Delta s \sum_{ad} O^{B}_{da}\eta_{da}W_{abcdef}, \tag*{(75)}
P(i/bc)ΔWibcdef=Δs∑aηiaWabcdef,P(l/ef)ΔWabclef=−Δs∑dηdlWabcdef,(76)P(i/bc)\Delta W_{ibcdef} = \Delta s \sum_{a}\eta_{ia}W_{abcdef},\quad P(l/ef)\Delta W_{abclef} = -\Delta s \sum_{d}\eta_{dl}W_{abcdef}, \tag*{(76)}
Δfcd=−14Δs∑abefOab/efAηefabWabcdef,(77)\Delta f_{cd} = -\frac{1}{4}\Delta s \sum_{abef} O^{A}_{ab/ef}\eta_{efab}W_{abcdef}, \tag*{(77)}
P(i/c)ΔΓicde=−12Δs∑abfOab/fSηifabWabcdef,P(k/f)ΔΓabkf=12Δs∑cdeOde/cSηdekcWabcdef,(78)P(i/c)\Delta\Gamma_{icde} = -\frac{1}{2}\Delta s \sum_{abf} O^{S}_{ab/f}\eta_{ifab}W_{abcdef},\quad P(k/f)\Delta\Gamma_{abkf} = \frac{1}{2}\Delta s \sum_{cde} O^{S}_{de/c}\eta_{dekc}W_{abcdef}, \tag*{(78)}
P(ij/c)ΔWijcdef=12Δs∑abOabCηijabWabcdef,P(lm/f)ΔWabclmf=−12Δs∑deOdeCηdelmWabcdef,(79)P(ij/c)\Delta W_{ijcdef} = \frac{1}{2}\Delta s \sum_{ab} O^{C}_{ab}\eta_{ijab}W_{abcdef},\quad P(lm/f)\Delta W_{abclmf} = -\frac{1}{2}\Delta s \sum_{de} O^{C}_{de}\eta_{delm}W_{abcdef}, \tag*{(79)}
P(ab/k)P(de/n)ΔWabkden=−Δs∑cfOcfBηfknWabcdef,(80)P(ab/k)P(de/n)\Delta W_{abkden} = -\Delta s \sum_{cf} O^{B}_{cf}\eta_{fkn}W_{abcdef}, \tag*{(80)}
Δfif=112Δs∑abcdeOde/abcSηdeiabcWabcdef,Δfcj=−112Δs∑abdefOab/defSηdefabjWabcdef,(81)\Delta f_{if} = \frac{1}{12}\Delta s \sum_{abcde} O^{S}_{de/abc}\eta_{deiabc}W_{abcdef},\quad\Delta f_{cj} = -\frac{1}{12}\Delta s \sum_{abdef} O^{S}_{ab/def}\eta_{defabj}W_{abcdef}, \tag*{(81)}
ΔΓijef=16Δs∑abcdOd/abcAηdijabcWabcdef,ΔΓbckl=−16Δs∑adefOa/defAηdefaklWabcdef,(82)\Delta\Gamma_{ijef} = \frac{1}{6}\Delta s \sum_{abcd} O^{A}_{d/abc}\eta_{dijabc}W_{abcdef},\quad\Delta\Gamma_{bckl} = -\frac{1}{6}\Delta s \sum_{adef} O^{A}_{a/def}\eta_{defakl}W_{abcdef}, \tag*{(82)}
P(i/c)P(f/l)ΔΓicfl=14Δs∑abdeOab/deAηdeiablWabcdef,(83)P(i/c)P(f/l)\Delta\Gamma_{icfl} = \frac{1}{4}\Delta s \sum_{abde} O^{A}_{ab/de}\eta_{deiabl}W_{abcdef}, \tag*{(83)}
ΔWijkdef=16Δs∑abcOabcSηijkabcWabcdef,ΔWabclmn=−16Δs∑defOdefSηdeflmnWabcdef,(84)\Delta W_{ijkdef} = \frac{1}{6}\Delta s \sum_{abc} O^{S}_{abc}\eta_{ijkabc}W_{abcdef},\quad\Delta W_{abclmn} = -\frac{1}{6}\Delta s \sum_{def} O^{S}_{def}\eta_{deflmn}W_{abcdef}, \tag*{(84)}
P(ij/c)P(de/n)ΔWijcden=−12Δs∑abfOab/fSηijfabnWabcdef,(85)P(ij/c)P(de/n)\Delta W_{ijcden} = -\frac{1}{2}\Delta s \sum_{abf} O^{S}_{ab/f}\eta_{ijfabn}W_{abcdef}, \tag*{(85)}
P(ab/k)P(lm/f)ΔWabklmf=12Δs∑cdeOde/cSηdeklmcWabcdef.(86)P(ab/k)P(lm/f)\Delta W_{abklmf} = \frac{1}{2}\Delta s \sum_{cde} O^{S}_{de/c}\eta_{deklmc}W_{abcdef}. \tag*{(86)}

IMSRG(4) Flow Equation

To the best of our knowledge, no full implementation (even the formalism) of the IMSRG truncated at the normal-ordered four-body level [IMSRG(4)] has been reported to date. Here, we present the IMSRG(4) flow equations, from which the corresponding SRGQMC(4) formalism was derived using a procedure similar to that in the derivations of SRGQMC(2)/(3). While the tedious formulas are omitted here, the SRGQMC(4) framework follows those of SRGQMC(2)/(3) but explicitly includes four-body contributions in the Hamiltonian and other operators. It is worth noting that recent automated diagram-generation tools [4, 5] provide valuable means of cross-checking lengthy many-body commutator formulas. All matrix elements are assumed to be fully antisymmetrized. For compactness, we denote an nn-body operator O(n)O^{(n)} simply by nn, and suppress the explicit normal-ordered strings in the resulting commutators. No Hermiticity assumption is imposed on the operators. At the IMSRG(4) level, the commutator [A,B][A,B] contains the following 16 additional terms beyond IMSRG(3).

[1,]
[1,4]→3=136∑ijklmn∑ab{(na−nb)AabBijkblmna}(87)[1,4]\to3=\frac{1}{36}\sum_{ijklmn}\sum_{ab}\left\{(n_a-n_b)A_{ab}B_{ijkblmna}\right\} \tag*{(87)}
[1,4]→4=1576∑ijklmnpq∑a{P(i/jkl)AiaBajklmnpq−P(m/npq)AamBijklanpq}(88)[1,4]\to4=\frac{1}{576}\sum_{ijklmnpq}\sum_a\left\{P(i/jkl)A_{ia}B_{ajklmnpq}-P(m/npq)A_{am}B_{ijkl anpq}\right\} \tag*{(88)}
[2,]
[2,3]→4=1576∑ijklmnpq∑a{(1−Pik−Pil−Pjk−Pjl+PikPjl)P(m/npq)AijmaBklanpq−(1−Pmp−Pmq−Pnp−Pnq+PmpPnq)P(i/jkl)AiamnBjklpqa}(89)[2,3]\to4=\frac{1}{576}\sum_{ijklmnpq}\sum_a\left\{(1-P_{ik}-P_{il}-P_{jk}-P_{jl}+P_{ik}P_{jl})P(m/npq)A_{ijma}B_{klanpq} -(1-P_{mp}-P_{mq}-P_{np}-P_{nq}+P_{mp}P_{nq})P(i/jkl)A_{iamn}B_{jklpqa}\right\} \tag*{(89)}
[2,4]→2=116∑ijkl∑abcd(nˉanˉbncnd−nanbnˉcnˉd)AcdabBabijklcd(90)[2,4]\to2=\frac{1}{16}\sum_{ijkl}\sum_{abcd}(\bar{n}_a\bar{n}_b n_c n_d-n_a n_b\bar{n}_c\bar{n}_d)A_{cdab}B_{abijklcd} \tag*{(90)}
[2,4]→3=172∑ijklmn∑abc(nanbnˉc+nˉanˉbnc){P(i/jk)AicabBjkablmnc−P(l/mn)AablcBijkcmnab}(91)[2,4]\to3=\frac{1}{72}\sum_{ijklmn}\sum_{abc}(n_a n_b\bar{n}_c+\bar{n}_a\bar{n}_b n_c)\left\{P(i/jk)A_{icab}B_{jkab l m n c}-P(l/mn)A_{ablc}B_{ijkcmnab}\right\} \tag*{(91)}
[2,4]→4=1576∑ijklmnpq∑ab{12(1−na−nb)(1−Pik−Pil−Pjk−Pjl+PikPjl)AijabBabklmnpq−(1−Pmp−Pmq−Pnp−Pnq+PmpPnq)AabmnBijklpqab+(na−nb)P(i/jkl)P(m/npq)AiambBjklbnpqa}(92)[2,4]\to4=\frac{1}{576}\sum_{ijklmnpq}\sum_{ab}\left\{\frac{1}{2}(1-n_a-n_b)(1-P_{ik}-P_{il}-P_{jk}-P_{jl}+P_{ik}P_{jl})A_{ijab}B_{abklmnpq} -(1-P_{mp}-P_{mq}-P_{np}-P_{nq}+P_{mp}P_{nq})A_{abmn}B_{ijklpqab} +(n_a-n_b)P(i/jkl)P(m/npq)A_{iamb}B_{jklbnpqa}\right\} \tag*{(92)}
[3,]
[3,3]→4=1576∑ijklmnpq∑ab{12(1−na−nb)[P(ijk/l)P(m/npq)AijkmabBablnpq−P(i/jkl)P(mnp/q)AabimnpBjklqabi]−(na−nb)(1−Pik−Pil−Pjk−Pjl+PikPjl)(1−Pmp−Pmq−Pnp−Pnq+PmpPnq)AijbmnaBklapqb}(93)[3,3]\to4=\frac{1}{576}\sum_{ijklmnpq}\sum_{ab}\left\{\frac{1}{2}(1-n_a-n_b)\left[P(ijk/l)P(m/npq)A_{ijkmab}B_{ab lnpq}-P(i/jkl)P(mnp/q)A_{abimnp}B_{jklqabi}\right] -(n_a-n_b)(1-P_{ik}-P_{il}-P_{jk}-P_{jl}+P_{ik}P_{jl})(1-P_{mp}-P_{mq}-P_{np}-P_{nq}+P_{mp}P_{nq})A_{ijbmna}B_{klapqb}\right\} \tag*{(93)}
[3,4]→1=136∑ij∑abcdef(nanbncnˉdnˉenˉf−nˉanˉbnˉcndnenf)AabcdefBidefjabc(94)[3,4]\to1=\frac{1}{36}\sum_{ij}\sum_{abcdef}(n_a n_b n_c\bar{n}_d\bar{n}_e\bar{n}_f-\bar{n}_a\bar{n}_b\bar{n}_c n_d n_e n_f)A_{abcdef}B_{idefjabc} \tag*{(94)}
[3,4]→2=−148∑ijkl∑abcde(nanbncnˉdnˉe+nˉanˉbnˉcndne)[P(i/j)AideabcBjabcklde−P(k/l)AabckdeBijdelabc](95)[3,4]\to2=-\frac{1}{48}\sum_{ijkl}\sum_{abcde}(n_a n_b n_c\bar{n}_d\bar{n}_e+\bar{n}_a\bar{n}_b\bar{n}_c n_d n_e)\left[P(i/j)A_{ideabc}B_{jabcklde}-P(k/l)A_{abckde}B_{ijdelabc}\right] \tag*{(95)}
[3,4]→3=136∑ijklmn∑abcd{16(nˉanˉbnˉcnˉd−nˉanbncnd)[P(ij/k)AaijbcdBkbcdlmna−P(lm/n)AbcdalmBijkanbcd]+14(nˉanˉbncnd−nanbnˉcnˉd)P(i/jk)P(l/mn)AcdiablBabjkcdmn},(96)[3,4]\to3 = \frac{1}{36}\sum_{ijklmn}\sum_{abcd}\left\{\frac{1}{6}(\bar{n}_{a}\bar{n}_{b}\bar{n}_{c}\bar{n}_{d}-\bar{n}_{a}n_{b}n_{c}n_{d})\left[P(ij/k)A_{aijbcd}B_{kbcdlmna}-P(lm/n)A_{bcdalm}B_{ijkanbcd}\right] +\frac{1}{4}(\bar{n}_{a}\bar{n}_{b}n_{c}n_{d}-n_{a}n_{b}\bar{n}_{c}\bar{n}_{d})P(i/jk)P(l/mn)A_{cdiabl}B_{abjkcdmn}\right\}, \tag*{(96)}
[3,4]→4=1576∑ijklmnpq∑abcd{16(nanbnc+nˉanˉbnˉc)[P(ijk/l)AijkabcBabclmnpq−P(mnp/q)AabcmnpBijklabcq]+12(nanˉbnˉc+nˉanbnc)[(1−Pik−Pil−Pjk−Pjl+PikPjl)P(m/npq)AijambcBklbcnpqa−(1−Pmp−Pmq−Pnp−Pnq+PmpPnq)P(i/jkl)AibcmnaBjklapqbc]}.(97)[3,4]\to4 = \frac{1}{576}\sum_{ijklmnpq}\sum_{abcd}\left\{\frac{1}{6}(n_{a}n_{b}n_{c}+\bar{n}_{a}\bar{n}_{b}\bar{n}_{c})\left[P(ijk/l)A_{ijkabc}B_{abclmnpq}-P(mnp/q)A_{abcmnp}B_{ijklabcq}\right] +\frac{1}{2}(n_{a}\bar{n}_{b}\bar{n}_{c}+\bar{n}_{a}n_{b}n_{c})\left[(1-P_{ik}-P_{il}-P_{jk}-P_{jl}+P_{ik}P_{jl})P(m/npq)A_{ijambc}B_{klbcnpqa} -(1-P_{mp}-P_{mq}-P_{np}-P_{nq}+P_{mp}P_{nq})P(i/jkl)A_{ibcmna}B_{jklapqbc}\right]\right\}. \tag*{(97)}
[4,]
[4,4]→0=1576∑ijkl∑abcd(ninjnknlnˉanˉbnˉcnˉd−nˉinˉjnˉknˉlnanbncnd)AijklabcdBabcdijkl,(98)[4,4]\to0 = \frac{1}{576}\sum_{ijkl}\sum_{abcd}(n_{i}n_{j}n_{k}n_{l}\bar{n}_{a}\bar{n}_{b}\bar{n}_{c}\bar{n}_{d}-\bar{n}_{i}\bar{n}_{j}\bar{n}_{k}\bar{n}_{l}n_{a}n_{b}n_{c}n_{d})A_{ijkl abcd}B_{abcdijkl}, \tag*{(98)}
[4,4]→1=1144∑ij∑abcdefg(nanbncndnˉenˉfnˉg+nˉanˉbnˉcnˉdnenfng)(AiefgabcdBabcdjefg−AabcdjefgBiefgabcd),(99)[4,4]\to1 = \frac{1}{144}\sum_{ij}\sum_{abcdefg}(n_{a}n_{b}n_{c}n_{d}\bar{n}_{e}\bar{n}_{f}\bar{n}_{g}+\bar{n}_{a}\bar{n}_{b}\bar{n}_{c}\bar{n}_{d}n_{e}n_{f}n_{g})(A_{iefg abcd}B_{abcdjefg}-A_{abcdjefg}B_{iefg abcd}), \tag*{(99)}
[4,4]→2=14∑ijkl∑abcdef{148(nanbncnˉdnˉenˉf−nˉanˉbnˉcndnenf)(AabijcdefBcdefklab−AcdefklabBabijcdef)+136(nanbncnˉdnˉenˉf−nˉanˉbnˉcndnenf)(1−Pij)(1−Pkl)AiabckdefBjdeflabc},(100)[4,4]\to2 = \frac{1}{4}\sum_{ijkl}\sum_{abcdef}\left\{\frac{1}{48}(n_{a}n_{b}n_{c}\bar{n}_{d}\bar{n}_{e}\bar{n}_{f}-\bar{n}_{a}\bar{n}_{b}\bar{n}_{c}n_{d}n_{e}n_{f})(A_{abijcdef}B_{cdefklab}-A_{cdefklab}B_{abijcdef}) +\frac{1}{36}(n_{a}n_{b}n_{c}\bar{n}_{d}\bar{n}_{e}\bar{n}_{f}-\bar{n}_{a}\bar{n}_{b}\bar{n}_{c}n_{d}n_{e}n_{f})(1-P_{ij})(1-P_{kl})A_{iabckdef}B_{jdeflab c}\right\}, \tag*{(100)}
[4,4]→3=136∑ijklmn∑abcde{124(nanbncndnˉe+nˉanˉbnˉcnˉdne)(AijkedabcBdabclmne−AcdablmneBijkecdab)+112(nanbncnˉdnˉe+nˉanˉbnˉcndne)[P(i/jk)P(lm/n)AiabclmdeBjkdenabc−P(ij/k)P(l/mn)AijdelabcBkabcmnde]},(101)[4,4]\to3 = \frac{1}{36}\sum_{ijklmn}\sum_{abcde}\left\{\frac{1}{24}(n_{a}n_{b}n_{c}n_{d}\bar{n}_{e}+\bar{n}_{a}\bar{n}_{b}\bar{n}_{c}\bar{n}_{d}n_{e})(A_{ijkedabc}B_{dabclmne}-A_{cdablmne}B_{ijkecdab}) +\frac{1}{12}(n_{a}n_{b}n_{c}\bar{n}_{d}\bar{n}_{e}+\bar{n}_{a}\bar{n}_{b}\bar{n}_{c}n_{d}n_{e})\left[P(i/jk)P(lm/n)A_{iabclmde}B_{jkdenabc}-P(ij/k)P(l/mn)A_{ijdelabc}B_{kabcmnde}\right]\right\}, \tag*{(101)}
[4,4]→4=1576∑ijklmnpq∑abcd{124(nˉanˉbnˉcnˉd−nanbncnd)(AijklabcdBabcdmnpq−BijklabcdAabcdmnpq)+16(nanˉbnˉcnˉd−nˉanbncnd)[P(ijk/l)P(m/npq)AijkambcdBlbcdnpqa−P(i/jkl)P(mnp/q)AibcdmnpaBjklaqbcd]+14(nanbnˉcnˉd−nˉanˉbncnd)(1−Pik−Pil−Pjk−Pjl+PikPjl)(1−Pmp−Pmq−Pnp−Pnq+PmpPnq)AijabmncdBklcdpqab}.(102)[4,4]\to4 = \frac{1}{576}\sum_{ijklmnpq}\sum_{abcd}\left\{\frac{1}{24}(\bar{n}_{a}\bar{n}_{b}\bar{n}_{c}\bar{n}_{d}-n_{a}n_{b}n_{c}n_{d})(A_{ijkl abcd}B_{abcdmnpq}-B_{ijkl abcd}A_{abcdmnpq}) +\frac{1}{6}(n_{a}\bar{n}_{b}\bar{n}_{c}\bar{n}_{d}-\bar{n}_{a}n_{b}n_{c}n_{d})\left[P(ijk/l)P(m/npq)A_{ijkambcd}B_{lbcdnpqa}-P(i/jkl)P(mnp/q)A_{ibcdmnpa}B_{jklaqbcd}\right] +\frac{1}{4}(n_{a}n_{b}\bar{n}_{c}\bar{n}_{d}-\bar{n}_{a}\bar{n}_{b}n_{c}n_{d})(1-P_{ik}-P_{il}-P_{jk}-P_{jl}+P_{ik}P_{jl})(1-P_{mp}-P_{mq}-P_{np}-P_{nq}+P_{mp}P_{nq})A_{ijabmncd}B_{klcdpqab}\right\}. \tag*{(102)}

While a deterministic IMSRG(4) calculation remains unimplemented due to prohibitive computational costs, the SRGQMC(4) delivers an IMSRG(4) computation by using the stochastic quantum Monte Carlo technique.

Uncertainty Analysis

In this section we provide additional convergence tests for the statistical uncertainty of SRGQMC. For an observable OO, each independent stochastic evolution gives one estimate O(α)O^{(\alpha)}. The quoted central value and uncertainty are taken as

Figure 1 (PDF p. 14). Convergence of the calculated phase shifts with respect to the walker population NwN_{\mathrm{w}}. Three rows correspond to Nw=102N_{\mathrm{w}} = 10^{2}, 10310^{3}, and 10410^{4}, respectively, with Nloop=10N_{\mathrm{loop}} = 10 fixed. Columns show the 1S0{}^{1}S_{0}, 3P0{}^{3}P_{0}, 3S1{}^{3}S_{1}, 3D1{}^{3}D_{1}, and ϵ1\epsilon_{1} channels for the N3LOEMN(500){\mathrm{N}}^{3}\mathrm{LO}_{\mathrm{EMN}}(500) interaction evolved to λ=2.0 fm−1\lambda= 2.0\,\mathrm{fm}^{-1}. Solid curves denote deterministic SRG results, red points with error bars denote SRGQMC results, and open diamonds show the Nijmegen phase-shift analysis as an empirical reference.

O‾=1Nloop∑α=1NloopO(α),ΔO=1Nloop(Nloop−1)∑α=1Nloop(O(α)−O‾)2.(103)\overline{O} = \frac{1}{N_{\mathrm{loop}}}\sum_{\alpha=1}^{N_{\mathrm{loop}}} O^{(\alpha)}, \qquad\Delta O = \sqrt{\frac{1}{N_{\mathrm{loop}}(N_{\mathrm{loop}}-1)}\sum_{\alpha=1}^{N_{\mathrm{loop}}}\left(O^{(\alpha)}-\overline{O}\right)^{2}}. \tag*{(103)}

Here ΔO\Delta O is the standard error of the mean over independent loops. The two numerical setups in the present implementation therefore play complementary roles: increasing the walker population NwN_{\mathrm{w}} improves the stochastic resolution of each individual SRGQMC trajectory, while increasing NloopN_{\mathrm{loop}} reduces the sampling error of the final averaged estimator.

Figure 1 examines the NwN_{\mathrm{w}} convergence of the free-space NN calculation through scattering phase shifts. This is a stringent test because the phase shifts are not sampled directly; they are obtained after solving the NN scattering problem with the stochastically evolved interaction. Already for Nw=102N_{\mathrm{w}} = 10^{2}, the SRGQMC phase shifts follow the deterministic SRG curves in all channels, with rather large stochastic fluctuations. Increasing the population to Nw=103N_{\mathrm{w}} = 10^{3} and then to 10410^{4} systematically reduces the error bars and brings the central values into closer agreement with the deterministic SRG results over the full range of laboratory energies shown. Besides, no resolvable finite-walker systematic bias is observed in these NN phase-shift calculations.

Similar behavior is found in in-medium calculations shown in Fig. 2. We consider the Richardson pairing model at five representative coupling strengths, g=0.2g = 0.2, 0.40.4, 0.60.6, 0.80.8, and 1.0 MeV1.0\,\mathrm{MeV}, and compare SRGQMC with their deterministic counterparts. For SRGQMC(2) and SRGQMC(3), walker populations Nw=102N_{\mathrm{w}} = 10^{2}, 10310^{3}, and 10410^{4} are compared with deterministic IMSRG(2) and IMSRG(3), respectively. For SRGQMC(4), the larger operator space requires larger populations, and we show Nw=104N_{\mathrm{w}} = 10^{4}, 10510^{5}, and 10610^{6}, compared with FCI. Since the model contains A=4A = 4 particles, retaining normal-ordered four-body operators is sufficient to recover the exact many-body results.

Across all coupling strengths and all three truncation orders, the SRGQMC central values remain statistically compatible with the corresponding deterministic reference values. As NwN_{\mathrm{w}} is increased, the statistical error decreases and the stochastic estimates approach the reference lines. This trend persists from the weak-coupling regime to the strongest coupling shown. These observations support the interpretation that the residual discrepancies are dominated by stochastic sampling noise.

Finally, Fig. 3 shows the role of the number of independent stochastic loops. Here we fix the walker population to the deliberately small value Nw=100N_{\mathrm{w}} = 100 and focus on SRGQMC(2), comparing with deterministic IMSRG(2) at the same five pairing strengths. The loop number is varied from Nloop=10N_{\mathrm{loop}} = 10 to 10001000. The central values remain centered around the deterministic IMSRG(2) results, while the uncertainty decreases as more independent loops are included, as expected for an average over statistically independent stochastic trajectories. This shows that increasing NloopN_{\mathrm{loop}} provides a direct and controllable way to reduce the uncertainty of the final estimator even when each individual trajectory is generated with a small walker population.

Figure 2 (PDF p. 15). Convergence of the Richardson pairing-model correlation energy with respect to NwN_{\mathrm{w}}. Columns correspond to g=0.2g = 0.2, 0.40.4, 0.60.6, 0.80.8, and 1.0 MeV1.0\,\mathrm{MeV}. The three rows show SRGQMC(2), SRGQMC(3), and SRGQMC(4), respectively. Dashed horizontal lines denote the corresponding deterministic IMSRG(2), IMSRG(3) and IMSRG(4) values, the last of which is equivalent to the exact FCI for the present A=4A = 4 system.

Figure 3 (PDF p. 15). Convergence of SRGQMC(2) correlation energies with respect to the number of independent loops NloopN_{\mathrm{loop}} at fixed walker population Nw=100N_{\mathrm{w}} = 100. Panels correspond to g=0.2g = 0.2, 0.40.4, 0.60.6, 0.80.8, and 1.0 MeV1.0\,\mathrm{MeV}. Points with error bars are SRGQMC(2) estimates obtained from independent-loop samples, while dashed horizontal lines denote deterministic IMSRG(2) results.

Taken together, Figs. 1–3 show that the SRGQMC uncertainty behaves as expected for a stochastic sampling of the underlying flow equations. We do not observe a statistically significant systematic bias associated with finite NwN_{\mathrm{w}} in the tested NN SRG or pairing-model IMSRG calculations. Instead, the deviations from deterministic SRG or deterministic IMSRG are generally consistent with statistical fluctuations that can be reduced systematically by increasing either the walker population NwN_{\mathrm{w}} or the number of independent loops NloopN_{\mathrm{loop}}.

References

References

  1. [1]H. Hergert, S. Bogner, T. Morris, A. Schwenk, and K. Tsukiyama, Phys. Rep. 621, 165–222 (2016).
  2. [2]M. Heinz, A. Tichai, J. Hoppe, K. Hebeler, and A. Schwenk, Phys. Rev. C 103, 044318 (2021).
  3. [3]S. R. Stroberg, T. D. Morris, and B. C. He, Phys. Rev. C 110, 044316 (2024).
  4. [4]L. H. Chen, Y. Li, H. Hergert, and J. M. Yao, ComCombo: A python package for automated commutator calculations of quantum many-body operators (2026), arXiv:2603.24399 [nucl-th].DOI
  5. [5]A. Tichai, P. Arthuis, H. Hergert, and T. Duguet, The European Physical Journal A 58, 2 (2022).

Paper details

vlnc:W_pmktbyjn
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