pith. sign in

arxiv: 2504.21298 · v3 · submitted 2025-04-30 · 🪐 quant-ph

Preparation Circuits for Matrix Product States by Classical Variational Disentanglement

Refik Mansuroglu , Norbert Schuch This is my paper

Pith reviewed 2026-05-22 17:53 UTC · model grok-4.3

classification 🪐 quant-ph
keywords matrix product statesvariational disentanglementquantum circuit compilationlow-entanglement statesclassical optimizationpreparation circuits
0
0 comments X

The pith

A classical layer-by-layer variational method finds disentangling gates that compile efficient quantum circuits for preparing matrix product states.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents an algorithm that compiles quantum preparation circuits for matrix product states by repeatedly applying layers of parameterized disentangling gates and optimizing them to minimize bipartite entanglement. A successful disentangler reduces entanglement on average, which keeps each optimization step efficient on a classical computer even when the total circuit depth grows large. Because the Schmidt coefficients at every bond are directly available from the standard canonical form of the state, the optimization for different bonds can run in parallel. The authors test the approach on ground states of one-dimensional local Hamiltonians and on states whose entanglement has been artificially spread by error-correcting codes.

Core claim

By reversing the action of a disentangler that is variationally optimized to reduce bipartite entanglement after each layer of gates, one obtains a quantum circuit whose bond dimensions remain manageable throughout the compilation; the resulting circuit therefore prepares the target matrix product state with resources that scale with the original low-entanglement description rather than with system size.

What carries the argument

Reverse application of a disentangler obtained by minimizing bipartite entanglement measures on layers of parameterized two-qubit gates, using the canonical Gamma-Lambda form of the MPS to access all Schmidt coefficients locally.

If this is right

  • Layer-by-layer optimization remains classically efficient for deep circuits whenever the disentangler succeeds in lowering bond dimension on average.
  • The procedure can be heavily parallelized because every bond's Schmidt spectrum is locally accessible in the canonical MPS form.
  • Numerical tests confirm that the compiled circuits prepare both ground states of one-dimensional local Hamiltonians and states with artificially delocalized entanglement from error-correcting codes.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same reverse-disentanglement strategy might be applied to other tensor-network states whose canonical forms also give direct access to entanglement spectra.
  • If the method scales, it supplies a practical route for near-term devices to load classically simulable states without requiring full variational quantum optimization loops.
  • One could test whether the compiled circuits remain short when the target state is taken from a two-dimensional tensor network rather than a one-dimensional MPS.

Load-bearing premise

A successful disentangler is expected to decrease the bond dimension on average so that each successive optimization layer stays classically tractable even for deep circuits.

What would settle it

Observation that, for a family of low-bond-dimension MPS, the variational optimization of disentangling layers fails to reduce average bond dimension or produces circuits whose total gate count grows exponentially with system size would falsify the claimed classical efficiency.

Figures

Figures reproduced from arXiv: 2504.21298 by Norbert Schuch, Refik Mansuroglu.

Figure 1
Figure 1. Figure 1: Graphical representation of a unitary disentangler [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Bond dimensions for disentanglement of the ground state of a 1D Ising Model with skew magnetic field of [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Disentanglement of the ground states of the Ising model with skew magnetic field [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Disentanglement of an MPS approximation to the ground state of the Fermi-Hubbard model with hopping [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Disentanglement of logical Bell pair in a [5,1,3] and [11,1,5] stabilizer code in both the separated and the [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Disentanglement of the GHZ, Cluster and AKLT state on 50 sites. (a) A plot of the tail weights reveals that [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
read the original abstract

We study the classical compilation of quantum circuits for the preparation of matrix product states (MPS), which are quantum states of low entanglement with an efficient classical description. Our algorithm represents a near-term alternative to previous sequential approaches by reverse application of a disentangler, which can be found by minimizing bipartite entanglement measures after the application of a layer of parameterized disentangling gates. Since a successful disentangler is expected to decrease the bond dimension on average, such a layer-by-layer optimization remains classically efficient even for deep circuits. Additionally, as the Schmidt coefficients of all bonds are locally accessible through the canonical $\Gamma$-$\Lambda$ form of an MPS, the optimization algorithm can be heavily parallelized. We discuss guarantees and limitations to trainability and show numerical results for ground states of one-dimensional, local Hamiltonians as well as artificially spread out entanglement among multiple qubits using error correcting codes.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript presents a classical variational algorithm for compiling quantum circuits that prepare matrix product states (MPS) by iteratively applying layers of parameterized disentangling gates in reverse and optimizing them to minimize bipartite entanglement measures. The central claim is that successful disentanglement reduces the average bond dimension, keeping the procedure classically efficient even for deep circuits, with additional benefits from parallelization via the canonical MPS form; numerical demonstrations are given for ground states of 1D local Hamiltonians and states with delocalized entanglement constructed via error-correcting codes, along with a discussion of trainability guarantees and limitations.

Significance. If the average bond-dimension reduction holds reliably, the approach would offer a practical classical method for finding shallow preparation circuits for low-entanglement states, serving as a near-term alternative to sequential compilation techniques with potential utility in quantum simulation and state preparation on NISQ hardware. The emphasis on local accessibility of Schmidt coefficients for parallel optimization is a concrete implementation strength.

major comments (2)
  1. [Abstract] Abstract: The efficiency claim for deep circuits rests on the statement that 'a successful disentangler is expected to decrease the bond dimension on average,' but this expectation is not accompanied by proven bounds, convergence guarantees, or scaling analysis; the numerical examples on 1D Hamiltonians and error-correcting-code states do not include system-size scaling or explicit verification that the reduction persists when entanglement is delocalized.
  2. [Numerical results] Numerical results (as summarized in abstract): The reported tests lack quantitative error bars, detailed convergence metrics, or direct comparison baselines against sequential methods, leaving the practical advantage over prior approaches unquantified and the central efficiency assertion dependent on unverified average reduction.
minor comments (1)
  1. [Abstract] The discussion of 'guarantees and limitations to trainability' is mentioned but would benefit from explicit statements on optimization landscape properties or failure modes to clarify the scope of applicability.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for their careful reading and constructive feedback on our manuscript. We address each major comment below, clarifying the scope of our claims and indicating revisions where appropriate.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The efficiency claim for deep circuits rests on the statement that 'a successful disentangler is expected to decrease the bond dimension on average,' but this expectation is not accompanied by proven bounds, convergence guarantees, or scaling analysis; the numerical examples on 1D Hamiltonians and error-correcting-code states do not include system-size scaling or explicit verification that the reduction persists when entanglement is delocalized.

    Authors: We agree that the efficiency claim for deep circuits is based on the heuristic expectation that a successful disentangler decreases the average bond dimension, rather than on rigorous mathematical bounds or convergence guarantees. This expectation follows from the variational minimization of bipartite entanglement measures, which targets the Schmidt coefficients directly accessible in the canonical MPS form. The manuscript already includes a discussion of trainability guarantees and limitations. The numerical demonstrations cover both 1D local Hamiltonians and delocalized entanglement constructed via error-correcting codes; we have added explicit verification that the bond-dimension reduction persists in the latter case, along with system-size scaling results up to moderate sizes in the revised supplementary material. revision: partial

  2. Referee: [Numerical results] Numerical results (as summarized in abstract): The reported tests lack quantitative error bars, detailed convergence metrics, or direct comparison baselines against sequential methods, leaving the practical advantage over prior approaches unquantified and the central efficiency assertion dependent on unverified average reduction.

    Authors: We acknowledge that the original numerical presentation could be strengthened with additional quantitative details. In the revised manuscript we have added error bars obtained from multiple independent optimization runs, included plots of the entanglement measure convergence during each layer optimization, and provided direct comparisons of circuit depth and preparation fidelity against sequential compilation baselines for the same test states. These additions quantify the practical advantage for the considered cases. revision: yes

standing simulated objections not resolved
  • Rigorous proven bounds or convergence guarantees for the average bond-dimension reduction achieved by the variational disentanglement layers.

Circularity Check

0 steps flagged

No circularity: algorithm derives from direct MPS entanglement minimization without self-referential reductions

full rationale

The paper's core procedure optimizes parameterized disentangling gates by minimizing bipartite entanglement measures computed directly from the canonical Γ-Λ form of the input MPS. Efficiency for deep circuits is framed as an expectation that successful layers reduce average bond dimension, supported by the local accessibility of Schmidt coefficients rather than any fitted parameter or self-citation chain. No step equates a prediction to its own inputs by construction, renames a known result, or imports uniqueness from prior author work; numerical examples on 1D Hamiltonians and error-correcting codes provide independent validation outside the optimization loop itself. The derivation remains self-contained against standard MPS properties.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The method assumes that variational minimization of local entanglement measures will on average reduce bond dimension sufficiently to keep classical simulation tractable; it further relies on the canonical MPS form providing immediate access to all Schmidt values.

free parameters (1)
  • disentangling gate parameters
    Variational parameters of the two-qubit gates in each layer are optimized to minimize the chosen entanglement measure.
axioms (1)
  • domain assumption MPS admits a canonical Gamma-Lambda representation in which Schmidt coefficients at every bond are locally accessible
    This property is invoked to justify parallel optimization across layers and bonds.

pith-pipeline@v0.9.0 · 5675 in / 1163 out tokens · 37719 ms · 2026-05-22T17:53:54.814900+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Disentangling strategies and entanglement transitions in unitary circuit games with matchgates

    quant-ph 2025-07 unverdicted novelty 7.0

    Introduces a minimal matchgate circuit representation for fermionic Gaussian states together with a Yang-Baxter update algorithm, then maps out entanglement transitions in unitary circuit games under braiding and gene...

  2. Near-Optimal Quantum Time Evolution Circuits via Provably Convergent Compression

    quant-ph 2026-05 unverdicted novelty 6.0

    A recipe for initial points in variational compression of quantum time-evolution operators that provably converges to near-optimal O(N t polylog(N t/ε)) gate complexity for local translationally invariant Hamiltonians.

  3. Tensor-based phase difference estimation on time series analysis

    quant-ph 2026-01 unverdicted novelty 6.0

    Tensor-network compression of nearest-neighbor circuits plus four-type measurements yields 0.4-4.7% error on 8-qubit Hubbard energy gaps and enables QPE-type runs on IBM devices up to 52 qubits with over 4000 two-qubit gates.

Reference graph

Works this paper leans on

50 extracted references · 50 canonical work pages · cited by 3 Pith papers · 6 internal anchors

  1. [1]

    Preparation Circuits for Matrix Product States by Classical Variational Disentanglement

    · · ·A bn [n]|b1 . . . bn⟩ = D1 D2 A[1] A[2] ... A[n] ,(1) as a product of matricesA [k] ∈Mat(D k−1 ×D k,C) with maximal bond dimensionD= max k Dk, which is assumed to be constant in system size. Exploiting iso- metrical properties of the matricesA [k], quantum cir- cuits for exact state preparation have been formulated early on [8]. These circuits sequen...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  2. [2]

    (5) The most general SU(4) operator from [26] also in- cludes single-qubit gates at the end

    on two qubits, that is Gi,j(θ) =e −i(θi,1XiXj+θi,2YiYj+θi,3ZiZj)× × e−i(θi,4Xj+θi,5Yj+θi,6Zj) ⊗e −i(θi,7Xi+θi,8Yi+θi,9Zi) . (5) The most general SU(4) operator from [26] also in- cludes single-qubit gates at the end. These do not need to be part ofG i,i+1, as they do not remove any entanglement. With 9 parameters per gate, the total number of parameters i...

    work page

  3. [3]

    Q-M&S”, I5868-N/FOR5249 “QUAST

    and allowing for higher maximal bond dimensions along the way. The tail weights shown in Fig. 3 (a) dis- play a gradual concentration towards a product state, asS ∞ decreases starting from the boundary towards the center bonds. The maximal bond dimension of D= 100 was never reached during CVD. Using the MPS representation of the ground state of a local, g...

    work page doi:10.55776/coe1 2020

  4. [4]

    Marti, R

    L. Marti, R. Mansuroglu, and M. J. Hartmann, Quan- tum9, 1635 (2025)

    work page 2025

  5. [5]

    J. R. McClean, S. Boixo, V. N. Smelyanskiy, R. Bab- bush, and H. Neven, Nature Communications9, 10.1038/s41467-018-07090-4 (2018)

    work page doi:10.1038/s41467-018-07090-4 2018

  6. [6]

    Verstraete and J

    F. Verstraete and J. I. Cirac, Physical Review B73, 10.1103/physrevb.73.094423 (2006)

    work page doi:10.1103/physrevb.73.094423 2006

  7. [7]

    M. B. Hastings, Journal of Statistical Mechan- ics: Theory and Experiment2007, P08024–P08024 (2007)

    work page 2007

  8. [8]

    Schuch, M

    N. Schuch, M. M. Wolf, K. G. H. Vollbrecht, and J. I. Cirac, New Journal of Physics10, 033032 (2008)

    work page 2008

  9. [9]

    M. Will, T. A. Cochran, E. Rosenberg, B. Jobst, N. M. Eassa, P. Roushan, M. Knap, A. Gammon- Smith, and F. Pollmann, Probing non-equilibrium topological order on a quantum processor (2025), arXiv:2501.18461 [quant-ph]

    work page arXiv 2025

  10. [10]

    S. J. Evered, M. Kalinowski, A. A. Geim, T. Manovitz, D. Bluvstein, S. H. Li, N. Maskara, H. Zhou, S. Ebadi, M. Xu, J. Campo, M. Cain, 11 S. Ostermann, S. F. Yelin, S. Sachdev, M. Greiner, V. Vuleti´ c, and M. D. Lukin, Probing topological mat- ter and fermion dynamics on a neutral-atom quantum computer (2025), arXiv:2501.18554 [quant-ph]

    work page arXiv 2025

  11. [11]

    Sch¨ on, E

    C. Sch¨ on, E. Solano, F. Verstraete, J. I. Cirac, and M. M. Wolf, Physical Review Letters95, 10.1103/physrevlett.95.110503 (2005)

    work page doi:10.1103/physrevlett.95.110503 2005

  12. [13]

    State preparation with parallel-sequential circuits

    Z.-Y. Wei and D. Malz, State preparation with parallel-sequential circuits (2025), arXiv:2503.14645 [quant-ph]

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  13. [14]

    K. C. Smith, A. Khan, B. K. Clark, S. Girvin, and T.-C. Wei, PRX Quantum5, 030344 (2024)

    work page 2024

  14. [16]

    M. S. Rudolph, J. Chen, J. Miller, A. Acharya, and A. Perdomo-Ortiz, Quantum Science and Technology 9, 015012 (2023)

    work page 2023

  15. [17]

    Jobst, K

    B. Jobst, K. Shen, C. A. Riofr´ ıo, E. Shishenina, and F. Pollmann, Quantum8, 1544 (2024)

    work page 2024

  16. [18]

    Murota, Adiabatic encoding of pre-trained mps classifiers into quantum circuits (2025), arXiv:2504.09250 [quant-ph]

    K. Murota, Adiabatic encoding of pre-trained mps classifiers into quantum circuits (2025), arXiv:2504.09250 [quant-ph]

    work page arXiv 2025

  17. [19]

    Jaderberg, G

    B. Jaderberg, G. Pennington, K. V. Marshall, L. W. Anderson, A. Agarwal, L. P. Lindoy, I. Rungger, S. Mensa, and J. Crain, Variational preparation of normal matrix product states on quantum computers (2025), arXiv:2503.09683 [quant-ph]

    work page arXiv 2025

  18. [20]

    N. F. Robertson, A. Akhriev, J. Vala, and S. Zhuk, Approximate quantum compiling for quantum sim- ulation: A tensor network based approach (2024), arXiv:2301.08609 [quant-ph]

    work page arXiv 2024

  19. [22]

    Mc Keever and M

    C. Mc Keever and M. Lubasch, PRX Quantum5, 10.1103/prxquantum.5.020362 (2024)

    work page doi:10.1103/prxquantum.5.020362 2024

  20. [24]

    Ben-Dov, D

    M. Ben-Dov, D. Shnaiderov, A. Makmal, and E. G. Dalla Torre, npj Quantum Information10, 10.1038/s41534-024-00858-1 (2024)

    work page doi:10.1038/s41534-024-00858-1 2024

  21. [25]

    A. Khan, B. K. Clark, and N. M. Tubman, Pre- optimizing variational quantum eigensolvers with ten- sor networks (2023), arXiv:2310.12965 [quant-ph]

    work page arXiv 2023

  22. [26]

    A. A. Melnikov, A. A. Termanova, S. V. Dolgov, F. Neukart, and M. R. Perelshtein, Quantum Science and Technology8, 035027 (2023)

    work page 2023

  23. [28]

    Schollw¨ ock, Annals of Physics326, 96–192 (2011)

    U. Schollw¨ ock, Annals of Physics326, 96–192 (2011)

    work page 2011

  24. [29]

    Kraus and J

    B. Kraus and J. I. Cirac, Physical Review A63, 10.1103/physreva.63.062309 (2001)

    work page doi:10.1103/physreva.63.062309 2001

  25. [30]

    Cerezo, M

    M. Cerezo, M. Larocca, D. Garc´ ıa-Mart´ ın, N. L. Diaz, P. Braccia, E. Fontana, M. S. Rudolph, P. Bermejo, A. Ijaz, S. Thanasilp, E. R. Anschuetz, and Z. Holmes, Does provable absence of barren plateaus imply clas- sical simulability? or, why we need to rethink varia- tional quantum computing (2024), arXiv:2312.09121 [quant-ph]

    work page arXiv 2024

  26. [31]

    Fishman, S

    M. Fishman, S. R. White, and E. M. Stoudenmire, SciPost Phys. Codebases , 4 (2022)

    work page 2022

  27. [32]

    Huennefeld, W

    R. Mansuroglu and N. Schuch, Classical variational disentanglement, Software on Zenodo (2025), avail- able online at:https://doi.org/10.5281/zenodo. 15058443, last accessed on 24.04.2025

    work page doi:10.5281/zenodo 2025

  28. [33]

    Perfect Quantum Error Correction Code

    R. Laflamme, C. Miquel, J. P. Paz, and W. H. Zurek, Perfect quantum error correction code (1996), arXiv:quant-ph/9602019 [quant-ph]

    work page internal anchor Pith review Pith/arXiv arXiv 1996

  29. [34]

    A. R. Calderbank, E. M. Rains, P. W. Shor, and N. J. A. Sloane, Quantum error correction via codes over gf(4) (1997), arXiv:quant-ph/9608006 [quant- ph]

    work page internal anchor Pith review Pith/arXiv arXiv 1997

  30. [35]

    Schuch and F

    N. Schuch and F. Verstraete, Nature Physics5, 732–735 (2009)

    work page 2009

  31. [36]

    S. R. White and E. M. Stoudenmire, Physical Review B99, 10.1103/physrevb.99.081110 (2019)

    work page doi:10.1103/physrevb.99.081110 2019

  32. [37]

    Gonz´ alez-Garc´ ıa, R

    G. Gonz´ alez-Garc´ ıa, R. Trivedi, and J. I. Cirac, PRX Quantum3, 10.1103/prxquantum.3.040326 (2022)

    work page doi:10.1103/prxquantum.3.040326 2022

  33. [38]

    Gharibian and F

    S. Gharibian and F. Le Gall, SIAM Journal on Com- puting52, 1009–1038 (2023)

    work page 2023

  34. [39]

    C. Cade, L. Mineh, A. Montanaro, and S. Stanisic, Phys. Rev. B102, 235122 (2020)

    work page 2020

  35. [40]

    Eckstein, R

    T. Eckstein, R. Mansuroglu, S. Wolf, L. N¨ utzel, S. Tasler, M. Kliesch, and M. J. Hartmann, Shot- noise reduction for lattice hamiltonians (2024), arXiv:2410.21251 [quant-ph]

    work page arXiv 2024

  36. [41]

    Huang, Y

    H.-Y. Huang, Y. Liu, M. Broughton, I. Kim, A. An- shu, Z. Landau, and J. R. McClean, inProceedings of the 56th Annual ACM Symposium on Theory of Computing, STOC ’24 (ACM, 2024) p. 1343–1351

    work page 2024

  37. [42]

    Mansuroglu, A

    R. Mansuroglu, A. Adil, M. J. Hartmann, Z. Holmes, and A. T. Sornborger, PRX Quan- tum5, 10.1103/prxquantum.5.030306 (2024)

    work page doi:10.1103/prxquantum.5.030306 2024

  38. [43]

    C. M. Keever and M. Lubasch, Physical Review Re- search5, 10.1103/physrevresearch.5.023146 (2023)

    work page doi:10.1103/physrevresearch.5.023146 2023

  39. [44]

    Mansuroglu, T

    R. Mansuroglu, T. Eckstein, L. N¨ utzel, S. A. Wilkin- son, and M. J. Hartmann, Quantum Science and Tech- nology8, 025006 (2023)

    work page 2023

  40. [45]

    Mansuroglu, F

    R. Mansuroglu, F. Fischer, and M. J. Hartmann, Physical Review Research5, 10.1103/physrevre- search.5.043035 (2023)

    work page doi:10.1103/physrevre- 2023

  41. [46]

    Nibbi and C

    M. Nibbi and C. B. Mendl, Physical Review A110, 10.1103/physreva.110.042427 (2024)

    work page doi:10.1103/physreva.110.042427 2024

  42. [47]

    J. I. Latorre, Image compression and entanglement (2005), arXiv:quant-ph/0510031 [quant-ph]

    work page internal anchor Pith review Pith/arXiv arXiv 2005

  43. [48]

    E. M. Stoudenmire and D. J. Schwab, Super- vised learning with quantum-inspired tensor networks (2017), arXiv:1605.05775 [stat.ML]

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  44. [49]

    Chatziioannou, K

    N. Schuch, M. M. Wolf, F. Verstraete, and J. I. Cirac, Physical Review Letters100, 10.1103/phys- revlett.100.030504 (2008)

    work page doi:10.1103/phys- 2008

  45. [50]

    T. Chen, R. Shen, C. H. Lee, and B. Yang, SciPost Physics15, 10.21468/scipostphys.15.4.170 (2023)

    work page doi:10.21468/scipostphys.15.4.170 2023

  46. [51]

    Chen and T

    T. Chen and T. Byrnes, Quantum8, 1557 (2024). 12 Appendix A: Classical Efficiency of CVD In this section, we provide technical details for the proofs of classical efficiency of CVD including a bound on the bond dimension for a given entanglement entropy value, an error bound for the approximation of the prepared state by truncation of MPS and finally a pr...

    work page 2024

  47. [52]

    We derive a relation for the minimal entropy defined by the squared Schmidt coefficientsλ 2 i =:p i, which also satisfy the normalization constraint PD i=1 pi = 1−p

    Bounds on Bond Dimension – Proof of Lemma 1 Just as the entanglement entropy is bounded by the bond dimensionD, vice versa for a given entropy value SA,α, the bond dimension is bounded. We derive a relation for the minimal entropy defined by the squared Schmidt coefficientsλ 2 i =:p i, which also satisfy the normalization constraint PD i=1 pi = 1−p. The e...

    work page

  48. [53]

    The first part is the accuracy of the optimizerε= |ψ⟩ − Q1 i=L TD ◦G † i |0⟩ , which can be efficiently calculated classically

    Approximation Error Bound – Proof of Lemma 2 We show the error bound |ψ⟩ −U †|0⟩ ≤ε+L p 2(n−1)pby separating the error into two parts. The first part is the accuracy of the optimizerε= |ψ⟩ − Q1 i=L TD ◦G † i |0⟩ , which can be efficiently calculated classically. The second part is the truncation error arising for the sake of a bounded bond dimension. By t...

    work page

  49. [54]

    Absence of Barren Plateaus in CVD – Proof of Lemma 3 Atθ= 0, the gradient from Eq. (10) has a relatively simple form and can be calculated in graphical notation to be −iTr A TrAc (|ψ⟩⟨ψ|) TrAc h σ(l) j ⊗σ (r) j ,|ψ⟩⟨ψ| i = σr σl Γl Γr Λ0Λl Λr ¯Γl ¯Γr Λ0Λl Λr Λ0Λl Λr Γl Γr Λ0Λl Λr ¯Γl ¯Γr − σr Γl Γr Λ0Λl Λr ¯Γl ¯Γr Λ0Λl Λr Λ0Λl Λr Γl Γr σl Λ0Λl Λr ¯Γl ¯Γr ...

    work page

  50. [55]

    The second moment is calculated using Eq

    = 0, (A13) and similarly the second term vanishes. The second moment is calculated using Eq. (A10), whereas two of the four contractions vanish as they separate into terms of the form of Eq. (A13). The remaining terms are integrals over squares of σrσl A BΛ0 ¯A ¯BΛ30 , once with Λ 3 0 on the upper contraction and once on the lower contraction, as well as ...

    work page