Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

Constructing networks of quantum channels for state preparation

Dmytro Bondarenko

Published Jan 1, 2021• 1 citation
Publisher

Abstract

Entangled possibly mixed states are an essential resource for quantum computation, communication, metrology, and the simulation of many-body systems. It is important to develop and improve preparation protocols for such states. One possible way to prepare states of interest is to design an open system that evolves only towards the desired states. A Markovian evolution of a quantum system can be generally described by a Lindbladian. Tensor networks provide a framework to construct physically relevant entangled states. In particular, matrix product density operators (MPDOs) form an important variational class of states. MPDOs generalize matrix product states to mixed states, can represent thermal states of local one-dimensional Hamiltonians at sufficiently large temperatures, describe systems that satisfy the area law of entanglement, and form the basis of powerful numerical methods. In this work we develop an algorithm that determines for a given linear subspace of MPDOs whether this subspace can be the stable space of some frustration free k-local Lindbladian and, if so, outputs an appropriate Lindbladian. We proceed by using machine learning with networks of quantum channels, also known as quantum neural networks (QNNs), to train denoising post-processing devices for quantum sources. First, we show that QNNs can be trained on imperfect devices even when part of the training data is corrupted. Second, we show that QNNs can be trained to extrapolate quantum states to, e.g., lower temperatures. Third, we show how to denoise quantum states in an unsupervised manner. We develop a novel quantum autoencoder that successfully denoises Greenberger-Horne-Zeilinger, W, Dicke, and cluster states subject to spin-flip, dephasing errors, and random unitary noise. Finally, we develop recurrent QNNs (RQNNs) for denoising that requires memory, such as combating drifts. RQNNs can be thought of as matrix product quantum channels with a quantum algorithm for training and are closely related to MPDOs. The proposed preparation and denoising protocols can be beneficial for various emergent quantum technologies and are within reach of present-day experiments.

DOI
10.15488/11050

Related papers

Quantum Autoencoders to Denoise Quantum Data

Dmytro Bondarenko, Polina Feldmann

Entangled states are an important resource for quantum computation, communication, metrology, and the simulation of many-body systems. However, noise limits the experimental preparation of such states. Classical data can be efficiently denoised by autoencoders-neural networks trained in unsupervised manner. We develop a novel quantum autoencoder that successfully denoises Greenberger-Horne-Zeilinger, W, Dicke, and cluster states subject to spin-flip errors and random unitary noise. Various emergent quantum technologies could benefit from the proposed unsupervised quantum neural networks.

arXiv:1910.09169•Physical Review Letters•Mar 31, 2020•124 citations

One-dimensional many-body entangled open quantum systems with tensor network methods

Daniel Jaschke, Simone Montangero, Lincoln D. Carr

Abstract We present a collection of methods to simulate entangled dynamics of open quantum systems governed by the Lindblad master equation with tensor network methods. Tensor network methods using matrix product states (MPSs) have been proven very useful to simulate many-body quantum systems and have driven many innovations in research. Since the MPS design is tailored for closed one-dimensional systems governed by the Schrödinger equation, the next step for many-body quantum dynamics is the simulation of one-dimensional open quantum systems. We review the three dominant approaches to the simulation of open quantum systems via the Lindblad master equation: quantum trajectories, matrix product density operators (MPDOs), and locally purified tensor networks. Selected examples guide possible applications of the methods and serve moreover as a benchmark between the techniques. These examples include the finite temperature states of the transverse quantum Ising model, the dynamics of an exciton traveling under the influence of spontaneous emission and dephasing, and a double-well potential simulated with the Bose–Hubbard model including dephasing. We analyze which approach is favorable leading to the conclusion that a complete set of all three methods is most beneficial, pushing the limits of different scenarios. The convergence studies using analytical results for macroscopic variables and exact diagonalization methods as comparison, show, for example, that MPDOs are favorable for the exciton problem in our study. All three methods access the same library, i.e., the software package Open Source Matrix Product States , allowing us to have a meaningful comparison between the different approaches based on the selected examples. For example, tensor operations are accessed from the same subroutines and with the same optimization eliminating one possible bias in a comparison of such numerical methods.

arXiv:1804.09796•Quantum Science and Technology•Oct 10, 2018•92 citations

Tensor-network-based machine learning of non-Markovian quantum processes

Chu Guo, Kavan Kishore Modi, Dario Poletti

We show how a tensor-network-based machine learning algorithm can learn the structures of generic, non-Markovian, quantum stochastic processes. First, a process is represented as a matrix product operator (MPO) and trained with a database of local input states at different times and the corresponding time-nonlocal output state. We then apply the algorithm to predict the output state of a process at different times for a system that is coupled to a spin-chain environment. We then reconstruct the full process, and we quantify the non-Markovian memory by means of the bond dimension of the MPO for various properties of the system, of the environment, and of their interaction. Our study paves the way for a possible experimental investigation into the process tensor and its properties, and an effective characterization of noise in quantum devices.

arXiv:2004.11038•Physical Review A•Dec 17, 2020•59 citations

Enabling large-depth simulation of noisy quantum circuits with positive tensor networks

Ambroise Müller, Thomas Ayral, Corentin Bertrand

Matrix product density operators (MPDOs) are tensor network representations of locally purified density matrices where each physical degree of freedom is associated to an environment degree of freedom. MPDOs have interesting properties for mixed state representations: guaranteed positivity by construction, efficient conservation of the trace and computation of local observables. However, they have been challenging to use for noisy quantum circuit simulation, as the application of noise increases the dimension of the environment Hilbert space, leading to an exponential growth of bond dimensions. MPDOs also lack a unique canonical form, due to the freedom in the choice of basis for the environment Hilbert space, which leads to a vast variation of bond dimensions. In this work, we present a systematic way to reduce the bond dimensions of MPDOs by disentangling the purified state. We optimize the basis for the environment Hilbert space by performing density matrix renormalization group (DMRG)-like sweeps of local 2-qubit basis optimization. Interestingly, we find that targeting only the disentanglement of the purified state leads to a reduction of the environment dimension. In other words, a compact MPDO representation requires a low-entanglement purified state. We apply our compression method to the emulation of noisy random quantum circuits. Our technique allows us to keep bounded bond dimensions, and thus bounded memory, contrary to previous works on MPDOs, while keeping reasonable truncation fidelities.

arXiv:2403.00152•Feb 29, 2024•9 citations

Tensor network approaches to open quantum systems, higher dimensional lattices and random quantum circuits

James Allen

In this thesis, we will start by introducing some tensor network methods used to simulate complex physical system. First, we will simulate a neutral atom open quantum system, to develop a relationship between the noise, system size, and performance of a quantum approximate optimization algorithm (QAOA) iteration on this system. In order to help maintain the physicality of the density matrix representing the system, we develop a purity-preserving truncation technique that performs better than a naïve matrix product operator representation while being far more efficient than approaches based on matrix product density operators. We find that for most noise sources, the effect of noise on a QAOA iteration does not scale with system size, although the failure rate of the circuit increases. Next, we will find the ground state of Kagome lattices over different spin values, and identify different properties of the phases found there. In particular, we look at the Jz-J2 phase diagram of the XXZ model on a Kagome lattice for spins 1/2 to 2. For each spin value above 1, we identify a narrow intermediate phase between the q=0 and root(3)-by-root(3) phases that has nematic ordering in their next-nearest neighbour bonds. For spin 1, we also find a wide intermediate phase around the Heisenberg point that has no consistent order parameter. In addition, we will develop a new method of time evolving two-dimensional tensor networks in the form of projected entangled pair states (PEPS), based on converting the PEPS to an isometric tensor network state for each update. This method may be more efficient than a full update based on the boundary MPS method, but the numerical data is inconclusive on whether it has advantages over a simple update. In addition, we will also demonstrate some theoretical and numerical results on random quantum circuits (RQC). We first develop a scheme for numerically computing the anticoncentration of two-dimensional brickwork circuits for a large system size limit and depths up to 15. This approach is based on taking the statistical mechanics model of RQCs and interpreting a depth-wise slice of the system as the matrix product operator in an infinite DMRG algorithm. We find that for these depths, the two-dimensional brickwork maintains the log-depth anticoncentration relation found in one-dimensional brickworks by previous works. We also find that the statistical mechanics model of two-dimensional brickworks naturally hosts a quasiparticle model at very low local Hilbert space dimensions. Next, we show a generalization of results on the epsilon-approximate t-design depths of RQCs to a much wider class of architectures. Previous results were successful at proving that the approximate t-design depth is linear in system size for one-dimensional brickwork architectures, but were unable to generalize these results to arbitrary architectures. We develop an approach based on bounding subleading singular values for each layer in a particular connected block of gates, which naturally translates to a "cluster-merging picture" in which a hierarchy of graphs is connected over time. With this, we were able to translate established one-dimensional brickwork bounds to any possible architecture, preserving system size linearity in the vast majority of them. Finally, we will develop results on bounding the spectral gap of these circuits, which helps lower the t dependence of the approximate t-design depths. In particular, we show that the spectral gap of a one-dimensional brickwork is independent of t for all t <= q, where q is the local Hilbert space dimension of the original circuit. Thus, the epsilon-approximate t-design depth in the small epsilon limit has no t dependence at all. We accomplish this by reducing the spectral gap of a one-dimensional brickwork of any size to a spectral gap of an operator on an effective three-site system. Then, we bound this operator by simplifying its domain according to irreducible representations of the symmetric group, and obtain bounds on the resulting matrix product.

•Aug 27, 2024•0 citations

Disentangling quantum autoencoder

Adithya Sireesh, Abdulla Alhajri, Myungshik Kim +1 more

Abstract Entangled quantum states are highly sensitive to noise, which makes it difficult to transfer them over noisy quantum channels or to store them in quantum memory. Here, we propose the disentangling quantum autoencoder (DQAE) to encode entangled states into single-qubit product states. The DQAE provides an exponential improvement in the number of copies needed to transport entangled states across qubit-loss or leakage channels compared to unencoded states. The DQAE can be trained in an unsupervised manner from entangled quantum data. For general states, we train via variational quantum algorithms based on gradient descent with purity-based cost functions, while stabilizer states can be trained via a Metropolis algorithm. For particular classes of states, the number of training data needed to generalize is surprisingly low: for stabilizer states, DQAE generalizes by learning from a number of training data that scales linearly with the number of qubits, while only 1 training sample is sufficient for states evolved with the transverse-field Ising Hamiltonian. Our work provides practical applications for enhancing near-term quantum computers.

arXiv:2502.18580•Quantum Science and Technology•Aug 15, 2025•1 citation
Entangled possibly mixed states are an essential resource for quantum computation, communication, metrology, and the simulation of many-body systems. It is important to develop and improve preparation protocols for such states. One possible way to prepare states of interest is to design an open system that evolves only towards the desired states. A Markovian evolution of a quantum system can be generally described by a Lindbladian. Tensor networks provide a framework to construct physically relevant entangled states. In particular, matrix product density operators (MPDOs) form an important variational class of states. MPDOs generalize matrix product states to mixed states, can represent thermal states of local one-dimensional Hamiltonians at sufficiently large temperatures, describe systems that satisfy the area law of entanglement, and form the basis of powerful numerical methods. In this work we develop an algorithm that determines for a given linear subspace of MPDOs whether this subspace can be the stable space of some frustration free k-local Lindbladian and, if so, outputs an appropriate Lindbladian. We proceed by using machine learning with networks of quantum channels, also known as quantum neural networks (QNNs), to train denoising post-processing devices for quantum sources. First, we show that QNNs can be trained on imperfect devices even when part of the training data is corrupted. Second, we show that QNNs can be trained to extrapolate quantum states to, e.g., lower temperatures. Third, we show how to denoise quantum states in an unsupervised manner. We develop a novel quantum autoencoder that successfully denoises Greenberger-Horne-Zeilinger, W, Dicke, and cluster states subject to spin-flip, dephasing errors, and random unitary noise. Finally, we develop recurrent QNNs (RQNNs) for denoising that requires memory, such as combating drifts. RQNNs can be thought of as matrix product quantum channels with a quantum algorithm for training and are closely related to MPDOs. The proposed preparation and denoising protocols can be beneficial for various emergent quantum technologies and are within reach of present-day experiments.
Entangled states are an important resource for quantum computation, communication, metrology, and the simulation of many-body systems. However, noise limits the experimental preparation of such states. Classical data can be efficiently denoised by autoencoders-neural networks trained in unsupervised manner. We develop a novel quantum autoencoder that successfully denoises Greenberger-Horne-Zeilinger, W, Dicke, and cluster states subject to spin-flip errors and random unitary noise. Various emergent quantum technologies could benefit from the proposed unsupervised quantum neural networks.
Abstract We present a collection of methods to simulate entangled dynamics of open quantum systems governed by the Lindblad master equation with tensor network methods. Tensor network methods using matrix product states (MPSs) have been proven very useful to simulate many-body quantum systems and have driven many innovations in research. Since the MPS design is tailored for closed one-dimensional systems governed by the Schrödinger equation, the next step for many-body quantum dynamics is the simulation of one-dimensional open quantum systems. We review the three dominant approaches to the simulation of open quantum systems via the Lindblad master equation: quantum trajectories, matrix product density operators (MPDOs), and locally purified tensor networks. Selected examples guide possible applications of the methods and serve moreover as a benchmark between the techniques. These examples include the finite temperature states of the transverse quantum Ising model, the dynamics of an exciton traveling under the influence of spontaneous emission and dephasing, and a double-well potential simulated with the Bose–Hubbard model including dephasing. We analyze which approach is favorable leading to the conclusion that a complete set of all three methods is most beneficial, pushing the limits of different scenarios. The convergence studies using analytical results for macroscopic variables and exact diagonalization methods as comparison, show, for example, that MPDOs are favorable for the exciton problem in our study. All three methods access the same library, i.e., the software package Open Source Matrix Product States , allowing us to have a meaningful comparison between the different approaches based on the selected examples. For example, tensor operations are accessed from the same subroutines and with the same optimization eliminating one possible bias in a comparison of such numerical methods.
We show how a tensor-network-based machine learning algorithm can learn the structures of generic, non-Markovian, quantum stochastic processes. First, a process is represented as a matrix product operator (MPO) and trained with a database of local input states at different times and the corresponding time-nonlocal output state. We then apply the algorithm to predict the output state of a process at different times for a system that is coupled to a spin-chain environment. We then reconstruct the full process, and we quantify the non-Markovian memory by means of the bond dimension of the MPO for various properties of the system, of the environment, and of their interaction. Our study paves the way for a possible experimental investigation into the process tensor and its properties, and an effective characterization of noise in quantum devices.
Matrix product density operators (MPDOs) are tensor network representations of locally purified density matrices where each physical degree of freedom is associated to an environment degree of freedom. MPDOs have interesting properties for mixed state representations: guaranteed positivity by construction, efficient conservation of the trace and computation of local observables. However, they have been challenging to use for noisy quantum circuit simulation, as the application of noise increases the dimension of the environment Hilbert space, leading to an exponential growth of bond dimensions. MPDOs also lack a unique canonical form, due to the freedom in the choice of basis for the environment Hilbert space, which leads to a vast variation of bond dimensions. In this work, we present a systematic way to reduce the bond dimensions of MPDOs by disentangling the purified state. We optimize the basis for the environment Hilbert space by performing density matrix renormalization group (DMRG)-like sweeps of local 2-qubit basis optimization. Interestingly, we find that targeting only the disentanglement of the purified state leads to a reduction of the environment dimension. In other words, a compact MPDO representation requires a low-entanglement purified state. We apply our compression method to the emulation of noisy random quantum circuits. Our technique allows us to keep bounded bond dimensions, and thus bounded memory, contrary to previous works on MPDOs, while keeping reasonable truncation fidelities.
In this thesis, we will start by introducing some tensor network methods used to simulate complex physical system. First, we will simulate a neutral atom open quantum system, to develop a relationship between the noise, system size, and performance of a quantum approximate optimization algorithm (QAOA) iteration on this system. In order to help maintain the physicality of the density matrix representing the system, we develop a purity-preserving truncation technique that performs better than a naïve matrix product operator representation while being far more efficient than approaches based on matrix product density operators. We find that for most noise sources, the effect of noise on a QAOA iteration does not scale with system size, although the failure rate of the circuit increases. Next, we will find the ground state of Kagome lattices over different spin values, and identify different properties of the phases found there. In particular, we look at the Jz-J2 phase diagram of the XXZ model on a Kagome lattice for spins 1/2 to 2. For each spin value above 1, we identify a narrow intermediate phase between the q=0 and root(3)-by-root(3) phases that has nematic ordering in their next-nearest neighbour bonds. For spin 1, we also find a wide intermediate phase around the Heisenberg point that has no consistent order parameter. In addition, we will develop a new method of time evolving two-dimensional tensor networks in the form of projected entangled pair states (PEPS), based on converting the PEPS to an isometric tensor network state for each update. This method may be more efficient than a full update based on the boundary MPS method, but the numerical data is inconclusive on whether it has advantages over a simple update. In addition, we will also demonstrate some theoretical and numerical results on random quantum circuits (RQC). We first develop a scheme for numerically computing the anticoncentration of two-dimensional brickwork circuits for a large system size limit and depths up to 15. This approach is based on taking the statistical mechanics model of RQCs and interpreting a depth-wise slice of the system as the matrix product operator in an infinite DMRG algorithm. We find that for these depths, the two-dimensional brickwork maintains the log-depth anticoncentration relation found in one-dimensional brickworks by previous works. We also find that the statistical mechanics model of two-dimensional brickworks naturally hosts a quasiparticle model at very low local Hilbert space dimensions. Next, we show a generalization of results on the epsilon-approximate t-design depths of RQCs to a much wider class of architectures. Previous results were successful at proving that the approximate t-design depth is linear in system size for one-dimensional brickwork architectures, but were unable to generalize these results to arbitrary architectures. We develop an approach based on bounding subleading singular values for each layer in a particular connected block of gates, which naturally translates to a "cluster-merging picture" in which a hierarchy of graphs is connected over time. With this, we were able to translate established one-dimensional brickwork bounds to any possible architecture, preserving system size linearity in the vast majority of them. Finally, we will develop results on bounding the spectral gap of these circuits, which helps lower the t dependence of the approximate t-design depths. In particular, we show that the spectral gap of a one-dimensional brickwork is independent of t for all t <= q, where q is the local Hilbert space dimension of the original circuit. Thus, the epsilon-approximate t-design depth in the small epsilon limit has no t dependence at all. We accomplish this by reducing the spectral gap of a one-dimensional brickwork of any size to a spectral gap of an operator on an effective three-site system. Then, we bound this operator by simplifying its domain according to irreducible representations of the symmetric group, and obtain bounds on the resulting matrix product.
Abstract Entangled quantum states are highly sensitive to noise, which makes it difficult to transfer them over noisy quantum channels or to store them in quantum memory. Here, we propose the disentangling quantum autoencoder (DQAE) to encode entangled states into single-qubit product states. The DQAE provides an exponential improvement in the number of copies needed to transport entangled states across qubit-loss or leakage channels compared to unencoded states. The DQAE can be trained in an unsupervised manner from entangled quantum data. For general states, we train via variational quantum algorithms based on gradient descent with purity-based cost functions, while stabilizer states can be trained via a Metropolis algorithm. For particular classes of states, the number of training data needed to generalize is surprisingly low: for stabilizer states, DQAE generalizes by learning from a number of training data that scales linearly with the number of qubits, while only 1 training sample is sufficient for states evolved with the transverse-field Ising Hamiltonian. Our work provides practical applications for enhancing near-term quantum computers.