Optimal Lowrank Quantum Dynamics Compression

---
name: optimal-lowrank-quantum-dynamics-compression
version: 1.0.0
description: "Use when budgeting MPO rank in quantum simulation."
tags: [quantum, tensor-networks, mpo, simulation, compression, entanglement, renormalization]
source: arXiv:2609.27497
---

# Optimal Low-Rank Compression of Quantum Dynamics

Based on arXiv:2609.27497 "Optimal low-rank compression of quantum dynamics".

## Core Insight

Local quantum interactions generate dynamics in an exponentially large Hilbert space, yet **locality and entanglement growth bounds** restrict the information that actually spreads. This work determines, up to logarithmic factors, the **irreducible rank** any low-rank representation (MPO/tensor network) must retain for short-range interactions — with matching lower bounds, making the bounds essentially tight.

## The Tight Rank Bounds

- **Time-independent evolution**: log D = Õ(t + √log(1/ε)) — the accuracy exponent 1/2 is provably necessary (matching lower bound)
- **Arbitrary (driven) evolution**: fixed-time matching bounds give accuracy exponent 2/3
- **Dynamical entanglement spectra**: distinct small-α Rényi laws — α⁻¹ (static) and α⁻² (driven)
- **1D static case**: constructively attained by an explicit MPO algorithm; analogous extension to Liouvillian (open-system) dynamics

Interpretation: the ε-dependence enters only through √log(1/ε) — polynomial-in-log accuracy cost — while the linear-in-t term reflects the Lieb-Robinson velocity of information spread. These are the fundamental limits of how much quantum dynamics can be compressed.

## Methodology Pipeline

1. **Upper bound construction (1D static)**: Iteratively factorize the evolution operator into a sum of MPOs whose bond dimension grows as t + √log(1/ε); use Lieb-Robinson locality to truncate operator tails.
2. **Lower bound strategy**: Information-theoretic argument — construct states whose evolution generates Rényi-entropic structure that any rank-D representation must preserve; the α⁻¹ / α⁻² spectral laws pin down the required rank.
3. **Driven systems**: Fixed-time analysis with time-dependent Hamiltonians yields the weaker 2/3 exponent via coherent alternation arguments.
4. **Liouvillian extension**: Apply the compression to vectorized density-matrix evolution (open systems), replacing operator norm with trace-norm accuracy.

## Practical Guidance (for simulation engineers)

- Budget MPO bond dimension as D ≈ exp(c·t) with only polylog dependence on target accuracy — deep-time simulation cost is dominated by time, not precision.
- Use the small-α Rényi spectral law (α⁻¹ static / α⁻² driven) as a diagnostic: measure the entanglement spectrum of the propagated operator to verify whether a simulation is in the compressible regime.
- For Liouvillian (dissipative) evolution, the same rank budgeting applies — dissipation often *reduces* effective entanglement growth.

## Activation Keywords

- optimal low-rank quantum dynamics
- MPO bond dimension scaling
- Lieb-Robinson compression bound
- entanglement spectrum Renyi law
- tensor network simulation limits
- Liouvillian MPO simulation
- quantum dynamics information-theoretic lower bound
- 量子动力学最优低秩压缩

## Cross-Domain Applications

- Neural ODE / operator learning: information-spread bounds as model-capacity budgeting for temporal evolution operators
- Systems engineering: channel capacity limits for spatially local propagation
- Data compression: principled truncation levels when compressing spatiotemporally local kernels