Rigorous criteria for operator-entanglement boundedness and truncation-error control in Heisenberg-picture DMRG simulation of one-dimensional quantum dynamics

The density matrix renormalization group in the Heisenberg picture (H-DMRG) evolves the observable of interest, rather than the quantum state, as a matrix product operator. Hartmann, Prior, Clark and Plenio [Phys. Rev. Lett. 102, 057202 (2009)] showed that this converges dramatically faster than Schrödinger-picture DMRG for dissipative one-dimensional spin chains, with the cumulative truncation error appearing to saturate with chain length - an indication that the operator entanglement of Heisenberg-evolved local operators stays bounded - and they explicitly posed as future work the analytical explanation of this saturation and a rigorous demonstration of the Heisenberg-picture advantage for mixed-state dynamics, while leaving the relation between the discarded-weight truncation diagnostic and the true error in expectation values uncharacterized. This record asks for a rigorous theory delivering: (i) provable criteria, stated in terms of the Hamiltonian and Lindblad operators, for when the operator space entanglement entropy of Heisenberg-evolved local operators remains bounded with system size, with a bound independent of time, or grows at most logarithmically in time, with bounded and at-most-logarithmic growth counted together as the controlled class (dissipation coexisting with conservation laws being an instance of its logarithmic case), and delimiting also the complementary regimes (chaotic dynamics, and any other dynamics whose operator entanglement grows faster than logarithmically in time) where it grows without bound; (ii) under such criteria, a comparison of the resources (bond dimension at fixed target accuracy) needed to compute the expectation value of a local observable by Heisenberg-picture evolution of the observable and by Schrödinger-picture evolution of the state; and (iii) bounds relating the cumulative discarded weight accumulated during matrix-product-operator truncation to the error in computed expectation values. Later literature resolves special cases - exact finite-bond-dimension solutions for quadratic open chains, logarithmic bounds for integrable systems, conditional entanglement-scaling-to-simulability theorems, and the discovery that U(1)-conserving dissipation induces persistent logarithmic operator-entanglement growth - but general analytical criteria and cumulative truncation-error control remain open.

Background

The numerical simulation of quantum many-body systems is limited by the exponential growth of the Hilbert space with system size. The density matrix renormalization group (DMRG) circumvents this in many one-dimensional settings by working with matrix product states (MPS): wavefunctions written as a contracted chain of small tensors, one per lattice site, whose internal ('bond') dimension controls both cost and accuracy. The success of DMRG is tied to entanglement: ground states of gapped one-dimensional Hamiltonians obey an area law - the entanglement entropy across a cut between two halves of the chain is bounded independently of chain length - so a moderate bond dimension suffices. Out of equilibrium, by contrast, the entanglement entropy after a global quench (a sudden change of a Hamiltonian parameter) typically grows linearly in time, so the required bond dimension grows exponentially and long-time real-time simulations become intractable.

Real-time evolution with MPS is usually performed with the time-evolving block decimation (TEBD) algorithm of Vidal, which applies local time-evolution gates and truncates the wavefunction bond by bond, keeping the largest Schmidt coefficients (the singular values of the wavefunction viewed as a matrix across the cut) of each bipartition. The sum of the squares of the discarded coefficients, accumulated over all truncations, is the discarded weight, the standard heuristic error diagnostic of these methods.

Open systems coupled to an environment evolve into mixed states described by a Lindblad master equation, $\dot{\rho} = -i[H,\rho] + \sum_j \gamma_j (L_j \rho L_j^\dagger - \tfrac{1}{2}\{L_j^\dagger L_j, \rho\})$, with local Lindblad operators $L_j$; the density matrix can then be represented as a matrix product operator (MPO), the operator analogue of an MPS in which each site carries a two-sided physical index. Zwolak and Vidal extended DMRG to this setting in the Schrödinger picture.

Hartmann, Prior, Clark and Plenio proposed the complementary Heisenberg-picture strategy: instead of evolving the (mixed) state, evolve the observable. For a chain of $N$ spins with collective unitary evolution $U(t)$, the time-evolved observable $X_m(t) = U(t) X_m U(t)^\dagger$, where $X_m$ is a local operator acting on site $m$ - or, with dissipation, the solution of the adjoint (Heisenberg-picture) Lindblad equation - is represented as an MPO and propagated by TEBD in operator space, after which its expectation value can be computed for any initial state. The cost is governed by the operator's block entanglement, quantified by the operator space entanglement entropy (OSEE): identifying operators with state vectors via the Hilbert-Schmidt (Frobenius) inner product - so-called vectorization - the OSEE is the von Neumann entropy of the reduced vectorized operator across a bipartition. Low OSEE is equivalent to a compressible MPO, exactly as low entanglement is equivalent to a compressible MPS.

The paper established two things. First, exactness: for the anisotropic XY Hamiltonian ($J_z = 0$ in the spin-chain Hamiltonian $H = \sum_j B^z \sigma^z_j + \sum_{j,\alpha} J_\alpha \sigma^\alpha_j \sigma^\alpha_{j+1}$), operators that remain local under the Jordan-Wigner mapping (the standard exact rewriting of a spin chain as a chain of spinless fermions) - for example $\sigma^z_m$ - are exact MPOs of bond dimension at most 4 for all times, even though the corresponding state evolution is provably inefficient. Second, numerics: for dissipative XXZ-type chains with local pumping and damping Lindblad terms $\sigma^\pm_j$, Heisenberg-picture simulations converge much faster with bond dimension than Schrödinger-picture ones, and the cumulative truncation error saturates as the chain length grows at fixed bond dimension, hinting at OSEE saturation. Contemporaneously, Prosen and Pižorn introduced the OSEE as a complexity measure and computed it analytically for quadratic chains, finding saturation or slow logarithmic growth for local operators.

What the paper itself flagged as unresolved is the target of this problem. The saturation of the truncation error is a purely numerical observation with no analytical explanation. Because truncation errors in the state and operator representations are measured in different representations, the authors note that it is 'not obvious that lower truncation errors in one representation imply a better approximation for the expectation value of an observable' - indeed, on short chains they found comparable truncation errors in the two pictures while the observable error differed greatly. The Frobenius normalization that makes the eigenvalues of the reduced operator interpretable as Schmidt coefficients is not preserved by the master-equation propagators and must be re-imposed at every time step, with unquantified consequences. And failure modes must exist: if Heisenberg-picture simulation were always efficient, arbitrary quantum circuits could be simulated efficiently, since their readout consists of single-spin measurements.

Problem Statement

Convert the empirical observation that Heisenberg-picture DMRG compresses one-dimensional quantum dynamics into rigorous statements. For one-dimensional quantum lattice systems with local Hamiltonians and/or local Lindblad dissipators, the problem asks for:

1. Boundedness criteria. Derive analytical criteria - stated in terms of the Hamiltonian, the Lindblad operators, and their conservation laws - under which the operator space entanglement entropy of the Heisenberg-evolved local operator $X_m(t)$ (with $X_m$ a local operator acting on site $m$) across any bipartition remains bounded as the chain length $N \to \infty$ at fixed time $t$ with a bound independent of $t$, or grows at most logarithmically in $t$. Both branches concern the entanglement of the same operator, and the complementary regimes are the complement of this controlled class, namely growth faster than logarithmic in $t$; bounding the entanglement at fixed $t$ alone therefore does not suffice for membership in the controlled class. Growth at most logarithmic in $t$, as for dissipation coexisting with conserved quantities, belongs to this controlled class rather than to a complementary regime. The criteria should cover interacting, non-quadratic dynamics beyond the exactly solvable cases, and should delimit the complementary regimes (for example chaotic unitary dynamics, or any dynamics in which the operator entanglement grows faster than logarithmically in time) in which the operator entanglement provably grows without bound.

2. Efficiency separation. Under such criteria, prove statements comparing the resources (bond dimension at fixed target accuracy) needed to compute the expectation value $\langle X_m(t) \rangle$ by Heisenberg-picture MPO evolution of the observable versus by Schrödinger-picture evolution of the generally mixed state.

3. Error control. Prove a bound relating the cumulative discarded weight $\epsilon$ accumulated during MPO truncation to the error in the computed expectation values, making explicit in which norm and under which conditions a small $\epsilon$ guarantees a small observable error, and clarifying when truncation errors measured in the operator and state representations can be compared meaningfully.

These deliverables jointly support one overarching objective: a rigorous theory of when and why evolving the observable rather than the state is the computationally efficient route, subsuming the known partial results for quadratic, integrable, and conditionally-solvable classes. A theory restricted to free-fermion dynamics, or a simulability statement conditional on an assumed entanglement scaling rather than derived from the dynamics, does not by itself complete the objective.

References

1. Michael J. Hartmann, Javier Prior, Stephen R. Clark, and Martin B. Plenio, 'Density Matrix Renormalization Group in the Heisenberg Picture', Physical Review Letters 102, 057202 (2009), DOI: 10.1103/PhysRevLett.102.057202, arXiv:0808.0666, https://arxiv.org/abs/0808.0666
2. Guifré Vidal, 'Efficient Simulation of One-Dimensional Quantum Many-Body Systems', Physical Review Letters 93, 040502 (2004), DOI: 10.1103/PhysRevLett.93.040502, https://doi.org/10.1103/PhysRevLett.93.040502
3. Michael Zwolak and Guifré Vidal, 'Mixed-State Dynamics in One-Dimensional Quantum Lattice Systems: A Time-Dependent Superoperator Renormalization Algorithm', Physical Review Letters 93, 207205 (2004), DOI: 10.1103/PhysRevLett.93.207205, https://doi.org/10.1103/PhysRevLett.93.207205
4. Tomaž Prosen and Iztok Pižorn, 'Operator space entanglement entropy in a transverse Ising chain', Physical Review A 76, 032316 (2007), DOI: 10.1103/PhysRevA.76.032316, arXiv:0706.2480, https://arxiv.org/abs/0706.2480
5. Iztok Pižorn and Tomaž Prosen, 'Operator space entanglement entropy in XY spin chains', Physical Review B 79, 184416 (2009), DOI: 10.1103/PhysRevB.79.184416, arXiv:0903.2432, https://arxiv.org/abs/0903.2432
6. Stephen R. Clark, Javier Prior, Michael J. Hartmann, Dieter Jaksch, and Martin B. Plenio, 'Exact matrix product solutions in the Heisenberg picture of an open quantum spin chain', New Journal of Physics 12, 025005 (2010), DOI: 10.1088/1367-2630/12/2/025005, arXiv:0907.5582, https://arxiv.org/abs/0907.5582
7. Thomas Barthel and Martin Kliesch, 'Quasilocality and efficient simulation of Markovian quantum dynamics', Physical Review Letters 108, 230504 (2012), DOI: 10.1103/PhysRevLett.108.230504, arXiv:1111.4210, https://arxiv.org/abs/1111.4210
8. Jérôme Dubail, 'Entanglement scaling of operators: a conformal field theory approach, with a glimpse of simulability of long-time dynamics in 1+1d', Journal of Physics A: Mathematical and Theoretical 50, 234001 (2017), DOI: 10.1088/1751-8121/aa6f38, arXiv:1612.08630, https://arxiv.org/abs/1612.08630
9. Kévin Hémery, Frank Pollmann, and David J. Luitz, 'Matrix product states approaches to operator spreading in ergodic quantum systems', Physical Review B 100, 104303 (2019), DOI: 10.1103/PhysRevB.100.104303, arXiv:1901.05793, https://arxiv.org/abs/1901.05793
10. Vincenzo Alba, Jérôme Dubail, and Marko Medenjak, 'Operator Entanglement in Interacting Integrable Quantum Systems: the Case of the Rule 54 Chain', Physical Review Letters 122, 250603 (2019), DOI: 10.1103/PhysRevLett.122.250603, arXiv:1901.04521, https://arxiv.org/abs/1901.04521
11. Sebastian Paeckel, Thomas Köhler, Andreas Swoboda, Salvatore R. Manmana, Ulrich Schollwöck, and Claudius Hubig, 'Time-evolution methods for matrix-product states', Annals of Physics 411, 167998 (2019), DOI: 10.1016/j.aop.2019.167998, arXiv:1901.05824, https://arxiv.org/abs/1901.05824
12. Bruno Bertini, Pavel Kos, and Tomaž Prosen, 'Operator Entanglement in Local Quantum Circuits I: Chaotic Dual-Unitary Circuits', SciPost Physics 8, 067 (2020), DOI: 10.21468/SciPostPhys.8.4.067, arXiv:1909.07407, https://arxiv.org/abs/1909.07407
13. Vincenzo Alba, 'Diffusion and operator entanglement spreading', Physical Review B 104, 094410 (2021), DOI: 10.1103/PhysRevB.104.094410, arXiv:2006.02788, https://arxiv.org/abs/2006.02788
14. Tibor Rakovszky, C. W. von Keyserlingk, and Frank Pollmann, 'Dissipation-assisted operator evolution method for capturing hydrodynamic transport', Physical Review B 105, 075131 (2022), DOI: 10.1103/PhysRevB.105.075131, arXiv:2004.05177, https://arxiv.org/abs/2004.05177
15. David Wellnitz, Guillermo Preisser, Vincenzo Alba, Jérôme Dubail, and Johannes Schachenmayer, 'Rise and fall, and slow rise again, of operator entanglement under dephasing', Physical Review Letters 129, 170401 (2022), DOI: 10.1103/PhysRevLett.129.170401, arXiv:2201.05099, https://arxiv.org/abs/2201.05099
16. Guillermo Preisser, David Wellnitz, Thomas Botzung, and Johannes Schachenmayer, 'Comparing bipartite entropy growth in open-system matrix-product simulation methods', Physical Review A 108, 012616 (2023), DOI: 10.1103/PhysRevA.108.012616, arXiv:2303.09426, https://arxiv.org/abs/2303.09426
17. Stefano Carignano, Carlos Ramos Marimón, and Luca Tagliacozzo, 'Temporal entropy and the complexity of computing the expectation value of local operators after a quench', Physical Review Research 6, 033021 (2024), DOI: 10.1103/PhysRevResearch.6.033021, arXiv:2307.11649, https://arxiv.org/abs/2307.11649
18. Lin Zhang, 'Operator entanglement in SU(2)-symmetric dissipative quantum many-body dynamics', Physical Review B 112, 144303 (2025), DOI: 10.1103/d8kg-h1t7, arXiv:2410.18468, https://arxiv.org/abs/2410.18468
19. Lucia Vilchez-Estevez, Alexander Yosifov, and Jinzhao Sun, 'Direct probing of the simulation complexity of open quantum many-body dynamics', arXiv:2508.19959 (2025), https://arxiv.org/abs/2508.19959
20. Neil Dowling, 'Classical Simulability from Operator Entanglement Scaling', arXiv:2603.05656 (2026), https://arxiv.org/abs/2603.05656