Stochastic Similarity Renormalization Group

Technical Summary: Stochastic Similarity Renormalization Group (SRGQMC)

Problem Statement
The In-Medium Similarity Renormalization Group (IMSRG) is a powerful ab initio framework for nuclear structure that evolves the many-body Hamiltonian to decouple a target state from excited configurations. However, its predictive power is fundamentally limited by the necessity of truncating the induced many-body operators. While the normal-ordered two-body approximation, IMSRG(2), is widely successful, it systematically omits higher-order correlations essential for describing phenomena such as nuclear collectivity and clustering. Extending these calculations to IMSRG(3) or IMSRG(4) is theoretically desirable but computationally prohibitive for deterministic implementations due to the combinatorial explosion of the operator space and nested tensor contractions. Consequently, high-order IMSRG(4) calculations have remained unfeasible, restricting the approach to the full configuration-interaction (FCI) limit.

Methodology: The SRGQMC Framework
To circumvent the "curse of dimensionality" in high-order tensor spaces, the authors introduce the Stochastic Similarity Renormalization Group (SRGQMC). This approach integrates Quantum Monte Carlo (QMC) techniques into the SRG flow equations.

• Stochastic Mapping: Instead of deterministically evolving operator matrices, the continuous unitary flow is mapped onto an ensemble of signed "random walkers." These walkers reside on the matrix elements (or multi-index tensor configurations) of the Hamiltonian.
• Walker Dynamics: The evolution is driven by sampling the commutator structures of the flow equation, $dH/ds = [\eta, H]$. During a time step $\Delta s$, walkers at a specific site $(i, j)$ stochastically spawn child walkers at new sites based on the generator $\eta$. The spawning amplitudes are proportional to the commutator terms, and the signs of the walkers are determined by the algebraic signs of the interaction terms.
• Annihilation and Symmetrization: Walkers arriving at the same site with opposite signs undergo algebraic summation (annihilation), which controls the population size and reduces variance. Hermiticity is explicitly enforced at each step by symmetrizing the walker matrix.
• Scalability: The method replaces exhaustive algebraic summations over intermediate states with statistically controlled estimates. The scaling factor $N$ is dynamically adjusted to maintain a target total walker population $N_w$, ensuring the physical Hamiltonian is reconstructed from the net walker populations $C_{ij}$ via $H_{ij} \approx N C_{ij}$.

Key Contributions and Results

1. Free-Space Benchmarking:
   The authors first validated SRGQMC against deterministic free-space SRG evolutions of realistic nucleon-nucleon (NN) interactions (specifically the $N^3LO_{EMN}(500)$ potential).

   • Results: Using $N_w = 10^4$ walkers and $N_{loop} = 10$ independent loops, the stochastic evolution reproduced the deterministic matrix elements and scattering phase shifts with high fidelity.
   • Accuracy: The relative root-mean-square (RMS) deviations between stochastic and deterministic trajectories were within 1.5%–5.2% for matrix elements and below 0.60° for phase shifts. These deviations were largely encompassed by the quantified statistical error bars, demonstrating that the method preserves the unitary decoupling principle without introducing significant systematic bias.

2. Many-Body Calculations (Richardson Pairing Model):
   The framework was tested on the $A=4$ Richardson pairing model, a solvable many-body system used to benchmark truncation schemes.

   • Validation: SRGQMC(2) and SRGQMC(3) results matched their deterministic IMSRG(2) and IMSRG(3) counterparts within statistical uncertainties (RMS deviations of 0.35% and 0.51%, respectively).
   • IMSRG(4) Breakthrough: The authors performed the first-ever IMSRG(4) calculation using the stochastic technique. While deterministic IMSRG(4) is unfeasible due to computational costs, SRGQMC(4) (with $N_w = 10^6$) successfully reproduced the exact FCI results for the $A=4$ system.
   • Correlation Recovery: The inclusion of four-body operators via SRGQMC(4) demonstrated a substantial restoration of missing higher-order correlations, particularly in the strongly correlated regime ($|g| \gtrsim 1.0$ MeV), where lower-order truncations deviate significantly from the exact solution.

3. Computational Efficiency:
   A comparison of computational costs revealed a favorable scaling advantage for the stochastic approach. When moving from two-body to three-body truncations, the computational cost for SRGQMC increased by a factor of only 23, compared to a factor of 132 for the deterministic IMSRG. This suggests that SRGQMC offers a more practical pathway for high-order calculations.

Significance and Claims
The paper claims that SRGQMC provides a "practical pathway" to higher-order IMSRG calculations that were previously inaccessible due to combinatorial tensor-space explosions. By replacing deterministic matrix evolution with a stochastic ensemble of walkers, the method:

• Overcomes the four-body operator barrier that has traditionally restricted deterministic IMSRG calculations.
• Preserves the decoupling behavior of SRG while providing fully quantified, controllable statistical uncertainties.
• Enables the first IMSRG(4) calculation, showing that stochastic sampling can faithfully recover the full configuration-interaction limit in strongly correlated systems.

The authors emphasize that because the algorithm samples the algebraic structure of continuous unitary transformations rather than model-specific details, the strategy is adaptable to broader applications in quantum chemistry, condensed matter physics, and quantum information science where Hamiltonian flows are limited by the rapid growth of operator space. The work does not claim to solve all many-body problems but specifically addresses the bottleneck of high-order truncations in similarity renormalization group methods.