Nonadiabatic Renormalization Group for Strongly Coupled Multiscale Quantum Systems

This paper introduces the nonadiabatic renormalization group, a novel non-perturbative framework that iteratively suppresses high-energy degrees of freedom to generate a nested fiber bundle geometric structure and new tensor network states for modeling strongly coupled multiscale quantum systems, with demonstrated applications in interacting boson models and ab initio quantum chemistry.

The Gist

Imagine you are trying to understand a massive, chaotic orchestra where every instrument is playing at a different speed and volume, and they are all tightly connected. Some instruments (like the fast, high-pitched violins) are playing so quickly that they seem to blur, while others (like the slow, heavy tubas) move at a glacial pace. In physics and chemistry, these "instruments" are particles like electrons and atoms. The problem is that when they interact strongly, trying to calculate how they move together is like trying to solve a puzzle where the number of pieces grows exponentially, making it impossible for even the fastest supercomputers to handle.

This paper introduces a new way to solve this puzzle, called the Nonadiabatic Renormalization Group (NARG). Here is how it works, broken down into simple concepts:

1. The Old Way vs. The New Way

Traditionally, when scientists try to simplify these complex systems, they use a method called "tracing out." Imagine you have a noisy room with a fast-talking person and a slow-talking person. The old method says, "Let's just ignore the fast talker completely and pretend they don't exist, so we can focus on the slow talker." This works okay if the fast talker is quiet, but if they are shouting and shaking the slow talker's chair, ignoring them gives you a wrong answer.

NARG does something different. Instead of ignoring the fast talker, it suppresses them. It keeps the fast talker in the room but organizes the information so that the fast talker's influence is neatly folded into the description of the slow talker. It doesn't throw the fast information away; it tucks it away in a way that preserves the connection between the two.

2. The "Russian Doll" of Geometry

The paper describes a beautiful geometric structure that emerges from this method. Imagine a set of Russian nesting dolls.

• The outer doll represents the slowest, heaviest part of the system (like the nuclei of an atom).
• Inside that doll is another doll representing the faster parts (like electrons).
• But here is the twist: The "skin" of the outer doll isn't just a simple shell; it is itself made of a complex, layered structure that holds the inner doll.

The authors call this a nested fiber bundle. Think of it like a library where every book (a specific state of the fast particles) is organized on a shelf (the slow particles). But the shelf itself is a library containing even smaller books. This structure allows the math to handle the "strong coupling" (the shouting and shaking) without the numbers exploding into infinity. It captures the "shape" of how the fast particles react to the slow ones, including tricky geometric effects that usually break other math methods.

3. The "Leg-Tied" Tensor Network (LETTA)

To make this calculation work on a computer, the authors created a new type of digital building block called LETTA (Leg-Tied Tensor Ansatz).

• The Old Building Block (MPS): Imagine a standard chain of paperclips. Each paperclip (representing a part of the system) is connected to only its immediate neighbor. It's a simple, one-dimensional line.
• The New Building Block (LETTA): Imagine a chain where the paperclips are tied together in a more complex web. In this new method, a single "leg" (a connection point) is shared among three or more paperclips at once, not just two.

This is like moving from a simple necklace to a complex, multi-layered net. By sharing these "legs," the new method can hold onto much more information about how different parts of the system are "entangled" (connected) with each other. It breaks the limits of the old paperclip chains, allowing scientists to model systems that were previously too messy to calculate.

4. Real-World Tests

The authors didn't just dream this up; they tested it on two real problems:

• Interacting Bosons (Vibrating Atoms): They modeled a system of 20 vibrating atoms that were strongly coupled. The old methods would have taken forever or failed, but NARG found the answers in less than 20 seconds with high accuracy.
• Quantum Chemistry (Electrons in a Hydrogen Chain): They applied it to a chain of hydrogen atoms to see how electrons interact. By keeping a moderate number of "retained states" (the folded-up fast information), they managed to capture over 80% of the complex electron correlation energy. This is a huge deal because calculating electron interactions is one of the hardest problems in chemistry.

Summary

In short, this paper proposes a new mathematical "lens" for looking at complex quantum systems. Instead of throwing away the fast-moving parts of a system, it folds them into the slow-moving parts using a clever geometric structure. This leads to a new way of building computer models (LETTA) that can handle much more complexity than before, offering a faster and more accurate way to understand everything from vibrating molecules to the behavior of electrons in new materials.

Technical Summary

1. Problem Statement

Complex quantum systems in physics, chemistry, and materials science often exhibit multiscale nature with strong coupling between degrees of freedom (DOFs) spanning different energy (or time) scales.

• Examples: Electron-nuclear dynamics (electrons vs. nuclei), core-valence electron interactions, and vibrational-rotational coupling.
• Challenge: Standard computational methods struggle due to the exponential scaling of resources with system size.
• Limitations of Existing Methods:
  • Born-Oppenheimer (BO) Approximation: Assumes adiabatic separation. It fails in strongly coupled regimes (e.g., conical intersections) where non-adiabatic couplings diverge due to state degeneracy.
  • Standard Renormalization Group (RG): Often involves "tracing out" high-energy DOFs to create an effective action. This discards information about the high-energy states, making it difficult to recover their observables or correlation functions later.
  • Tensor Network States (MPS): Conventional Matrix Product States (MPS) typically encode entanglement via virtual legs, but may not fully capture the specific geometric structures arising from strong non-adiabatic couplings.

2. Methodology: Nonadiabatic Renormalization Group (NARG)

The author proposes a novel framework called Nonadiabatic Renormalization Group (NARG). Instead of tracing out high-energy DOFs, NARG iteratively suppresses them while retaining them in the basis set to preserve observables.

A. Strong Coupling Representation (Two-Scale)

For a bipartitioned system with fast ($x_f$) and slow ($x_s$) DOFs:

• Basis Construction: Instead of a direct product basis, NARG uses a local trivialization of the fiber bundle. It employs the eigenstates of the slow coordinate operator ($|x_s^n\rangle$) as the parameter space.
• Conditional Adiabatic States: For each configuration of the slow DOF ($x_s^n$), the fast DOFs are solved to obtain conditional eigenstates $|\phi_\alpha(x_s^n)\rangle$.
• Global Overlap Matrix: The expansion coefficients involve a global overlap matrix $A_{m\beta, n\alpha} = \langle \phi_\beta(x_s^m) | \phi_\alpha(x_s^n) \rangle$.
  • This matrix encodes the quantum geometry (Riemannian metric and Berry curvature) of the system.
  • It provides a gauge-invariant description, avoiding the singularities associated with conical intersections in traditional adiabatic representations.
• Hamiltonian Formulation: The total Hamiltonian is constructed using these overlap matrices and kinetic energy terms, effectively treating the system as a discretized fiber bundle.

B. Multiscale Extension (Nested Fiber Bundles)

The method is generalized to $L$ energy scales ($x_0, x_1, \dots, x_{L-1}$) ordered by descending energy:

1. Step 1: Solve for the fastest scale ($x_0$) conditioned on the next scale ($x_1$).
2. Step 2: Truncate the adiabatic eigenstates of the composite system ($x_0, x_1$) to a manageable dimension $D$ (coarse-graining).
3. Iteration: Treat the truncated composite state as the new "fast" block and couple it to the next scale ($x_2$).
4. Structure: This creates a nested (or continued) fiber bundle structure, where the fiber of one layer is itself a fiber bundle of the next layer.

3. Key Contributions

A. Theoretical Framework: NARG

• Non-Tracing Out: Unlike effective action approaches, NARG keeps high-energy DOFs in the basis, allowing for the calculation of their correlation functions and observables.
• Geometric Encoding: The strong coupling between scales is encoded in the quantum geometric structure of the nested fiber bundle, specifically via the non-Abelian quantum geometric tensor (metric and curvature).
• Handling Singularities: By using a global overlap matrix in a discrete variable representation, the method bypasses the gauge-fixing problems and singularities found at conical intersections.

B. New Tensor Network: LETTA

NARG leads to a new class of tensor network states called Leg-Tied Tensor Ansatz (LETTA).

• Structure: In conventional Matrix Product States (MPS), tensors share virtual legs (one physical leg per site). In LETTA, multiple tensors share physical legs.
• Entanglement: LETTA encodes entanglement not just in the coefficients but also in the basis set itself. It represents a generalization of MPS that can handle systems where "sites" are not equivalent (e.g., different energy scales) or where long-range leg sharing occurs.
• Flexibility: LETTA can be applied to standard spin models (where sites are equivalent) and multiscale problems, potentially breaking the fundamental entanglement limitations of standard MPS.

4. Results and Applications

A. Interacting Boson Model

• System: A model with 20 interacting bosonic modes with strong anharmonicity and mode coupling.
• Performance: NARG demonstrated rapid convergence for the lowest 16 eigenstates with respect to the number of retained states ($D$).
• Efficiency: Calculations were completed in under 20 seconds using a pure Python implementation, whereas exact diagonalization would be computationally infeasible.

B. Ab Initio Quantum Chemistry

• System: Electron correlation in a hydrogen chain ($H_8$) using the STO-6G basis set.
• Methodology: Molecular orbitals were ordered by energy (core to valence). The method iteratively added orbitals, diagonalizing the block Hamiltonian and retaining $D$ adiabatic states.
• Performance:
  • With a moderate bond dimension ($D=200$), NARG captured >80% of the correlation energy.
  • The method successfully incorporated electron correlation gradually, bridging the gap between Hartree-Fock and Full Configuration Interaction (FCI).

5. Significance and Future Outlook

• New Paradigm: NARG offers a powerful alternative to standard RG and tensor network methods for strongly coupled multiscale systems, particularly where non-adiabatic effects are dominant.
• Broad Applicability: The framework is applicable to diverse fields including:
  • Quantum chemistry (electron-nuclear dynamics).
  • Condensed matter physics (lattice field theory, quantum impurities).
  • Quantum optics (strong light-matter interaction, Floquet engineering).
  • Open quantum systems.
• Future Directions:
  • Development of optimization algorithms (analogous to DMRG) specifically for LETTA.
  • Application to non-equilibrium real-time quantum dynamics.
  • Exploration of mixed tensor networks combining MPS and LETTA.

In summary, this paper introduces a mathematically rigorous and computationally efficient method (NARG) that leverages the geometry of fiber bundles to solve strongly coupled quantum problems, resulting in a novel tensor network architecture (LETTA) capable of capturing complex entanglement structures beyond current state-of-the-art methods.