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

Global optimization of MPS in quantum-inspired numerical analysis

Paula García-Molina, Luca Tagliacozzo, Juan José García‐Ripoll

Published Mar 17, 2023• 0 citations
Publisher

Abstract

This work discusses the solution of partial differential equations (PDEs) using matrix product states (MPS). The study focuses on the search for the lowest eigenstates of a Hamiltonian equation, for which five algorithms are introduced: imaginary-time evolution, steepest gradient descent, an improved gradient descent, an implicitly restarted Arnoldi method, and density matrix renormalization group (DMRG) optimization. The first four methods are engineered using a framework of limited-precision linear algebra, where operations between MPS and matrix product operators (MPOs) are implemented with finite resources. All methods are benchmarked using the PDE for a quantum harmonic oscillator in up to two dimensions, over a regular grid with up to 2^{28} points. Our study reveals that all MPSbased techniques outperform exact diagonalization techniques based on vectors, with respect to memory usage. Imaginary-time algorithms are shown to underperform any type of gradient descent, both in terms of calibration needs and costs. Finally, Arnoldilike methods and DMRG asymptotically outperform all other methods, including exact diagonalization, as problem size increases, with an exponential advantage in memory and time usage.

DOI
10.5281/zenodo.8133831

Related papers

Comparative study of matrix product state/quantized tensor-train algorithms for solving time-independent partial differential equations

Paula García-Molina, Luca Tagliacozzo, Juan José García‐Ripoll

This work presents a comparative study of new and existing optimization and diagonalization methods for solving time-independent partial differential equations (PDEs) using matrix product states (MPS) in the quantized tensor-train formalism (QTT). This study focuses on Hamiltonian equations, for which five algorithms are introduced: explicit imaginary-time evolution methods, steepest gradient descent in conventional and optimized forms, a power method, and an explicitly restarted Arnoldi method. The first five methods are engineered using a framework of limited-precision linear algebra, in which operators -- i.e., the equation itself -- and vectors are represented using matrix product operator (MPO) and matrix product state (MPS) formalisms, and where operator-vector multiplication and vector addition are approximated with limited resources. All methods are benchmarked using an exactly solvable PDE for a quantum harmonic oscillator in one and two dimensions over a regular grid with up to $2^{30}$ points and compared with the density matrix renormalization group (DMRG) method. Our study reveals that all MPS-based techniques exponentially outperform exact diagonalization techniques based on vectors regarding memory usage. Imaginary-time algorithms are shown to underperform any gradient descent in terms of calibration needs and costs. Finally, MPS DMRG and interpolated Arnoldi-like asymptotically outperform all other methods, including state-of-the-art vector-based exact diagonalization, with significant advantages in time and memory use.

arXiv:2303.09430•Mar 16, 2023•1 citation

Who can compete with quantum computers? Lecture notes on quantum inspired tensor networks computational techniques

Xavier Waintal, Chen-How Huang, Christoph W. Groth

This is a set of lectures on tensor networks with a strong emphasis on the core algorithms involving Matrix Product States (MPS) and Matrix Product Operators (MPO). Compared to other presentations, particular care has been given to disentangle aspects of tensor networks from the quantum many-body problem: MPO/MPS algorithms are presented as a way to deal with linear algebra on extremely (exponentially) large matrices and vectors, regardless of any particular application. The lectures include well-known algorithms to find eigenvectors of MPOs (the celebrated DMRG), solve linear problems, and recent learning algorithms that allow one to map a known function into an MPS (the Tensor Cross Interpolation, or TCI, algorithm). The lectures end with a discussion of how to represent functions and perform calculus with tensor networks using the "quantics" representation. They include the detailed analytical construction of important MPOs such as those for differentiation, indefinite integration, convolution, and the quantum Fourier transform. Three concrete applications are discussed in detail: the simulation of a quantum computer (either exactly or with compression), the simulation of a quantum annealer, and techniques to solve partial differential equations (e.g. Poisson, diffusion, or Gross-Pitaevskii) within the "quantics" representation. The lectures have been designed to be accessible to a first-year PhD student and include detailed proofs of all statements.

quant-pharXiv:2601.03035•SciPost Physics Lecture Notes•Jan 7, 2026•0 citations

Numerical methods for variational quantum and quantum-inspired computing.

Jorge Abraham Gidi Chomalí

Quantum computation is a promising approach to solving problems in physics, optimization, and numerical simulation. Quantum algorithms leverage superposition and entanglement in an exponentially large Hilbert space to encode and solve problems that are intractable for classical computers. However, these algorithms require error-corrected quantum hardware, which is not yet available. This thesis explores two alternative quantum-based approaches that promise to achieve better scalability than traditional algorithms in the near-term: Hybrid variational quantum algorithms (VQAs) and fully classical, quantum-inspired algorithms. In the first part, we review and benchmark optimization methods well suited for VQAs, specifically the Simultaneous Perturbation Stochastic Approximation (SPSA). We introduce complex-variable extensions and improvements to the second-order and quantum natural implementations. Finally, we assess their performance across quantum control, quantum tomography, and the variational quantum eigensolver, providing practical guidelines for common VQA scenarios. In the second part, we apply Matrix Product States (MPS) to solve time-dependent partial differential equations (PDEs). By integrating Hermite Distributed Approximating Functionals (HDAF) into the MPS framework, we develop a novel encoding to efficiently approximate differential operators with exponential precision. We benchmark this approach against traditional schemes for simulating the expansion of a levitated nanoparticle. The quantum-inspired numerical integration schemes deliver large memory savings and adequate runtimes even in presence of moderate chirping.

•Jun 5, 2025•0 citations

Matrix product states algorithms and continuous systems

S. Iblisdir, Román Orús, José Ignacio Latorre

A generic method to investigate many-body continuous-variable systems is pedagogically presented. It is based on the notion of matrix product states (so-called MPS's) and the algorithms thereof. The method is quite versatile and can be applied to a wide variety of situations. As a first test, we show how it provides reliable results in the computation of fundamental properties of a chain of quantum harmonic oscillators achieving off-critical and critical relative errors of the order of ${10}^{\ensuremath{-}8}$ and ${10}^{\ensuremath{-}4}$, respectively. Next, we use it to study the ground-state properties of the quantum rotor model in one spatial dimension, a model that can be mapped to the Mott insulator limit of the one-dimensional Bose-Hubbard model. At the quantum critical point, the central charge associated with the underlying conformal field theory can be computed with good accuracy by measuring the finite-size corrections of the ground-state energy. Examples of MPS computations both in the finite-size regime and in the thermodynamic limit are given. The precision of our results is found to be comparable to that previously encountered in the MPS studies of, for instance, quantum spin chains. Finally, we present a spin-off application: an iterative technique to efficiently get numerical solutions of partial differential equations of many variables. We illustrate this technique by solving Poisson-like equations with precisions of the order of ${10}^{\ensuremath{-}7}$.

arXiv:cond-mat/0610530•Physical Review B•Mar 23, 2007•32 citations

Quantum Tensor Network Learning with DMRG

Gustav J L Jäger, Martin B Plenio, Hans-Martin Rieser

Tensor Networks are a relatively new machine learning approach. The architectures proposed initially are inspired by approaches from quantum many-body physics simulations. One common layout is the matrix product state (MPS) also known as a tensor train optimized with gradient descent techniques. We introduce a global normalization condition, so that the MPS represents a quantum state. We investigate two optimization methods that find the locally optimal tensors and compare them regarding their effectiveness. One is based on gradient descent and the other on an adaptation of DMRG.

quant-pharXiv:2608.18901•European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN)•Aug 20, 2026•1 citation

Quantum-Based Methods for Solving Partial Differential Equations in Electromagnetics Engineering

Emanuel Colella, Luca Bastianelli, Valter Mariani Primiani +2 more

Quantum methods for solving partial differential equations (PDEs) generally fall into a few general categories, each leveraging different quantum mechanisms to achieve potential speedups. One well-known approach is the quantum linear system solver, epitomized by the Harrow-Hassidim-Lloyd (HHL) algorithm, which uses phase estimation and eigenvalue inversion to solve large, sparse linear systems arising from discretized linear PDEs [1]. This method can yield exponential improvements over classical solvers under certain conditions (mostly when the matrix is well-conditioned and sparse). More recent developments, such as Quantum Singular Value Transformation (QSVT), provide a framework for matrix inversion in a more flexible way, often improving on the cost and error dependence of earlier solvers [2]. Time-dependent PDEs can be tackled via Hamiltonian simulations. The Trotter-Suzuki method decomposes the exponential of a Hamiltonian into products of exponentials of simpler sub-Hamiltonians, allowing a piecewise approximation of continuous dynamics [3]. Continuous-time quantum walks, which treat a PDE operator as a Hamiltonian that governs the walk, similarly simulate processes like diffusion or wave propagation on discretized grids or graphs [4]. Beyond direct simulation, variational quantum algorithms have emerged as a promising route in the noisy intermediate-scale quantum era. These approaches encode the PDE solution as a parameterized quantum state and optimize the parameters via classical feedback, thereby minimizing a cost function tied to the PDE residual or energy functional [5]. Quantum annealing or adiabatic methods, meanwhile, can solve optimization problems derived from PDE constraints by adiabatically evolving a simple initial Hamiltonian to the final problem Hamiltonian and staying in the system's ground state [3]. In our study, we implement a new parameterized quantum model, specifically designed to solve large linear systems on quantum hardware. This model is less deep and computationally faster than the proposed one in [5], reducing the complexity and speeding up convergence times. Such improvements are particularly valuable in cases where classical algorithms may fail due to the problem's size or complexity. Ongoing research is focusing on generalized quantum models that can be adapted to practical PDE problems for solving large electromagnetic problems where classical computers can fail.

USNC-URSI Radio Science Meeting•Jul 13, 2025•0 citations
This work discusses the solution of partial differential equations (PDEs) using matrix product states (MPS). The study focuses on the search for the lowest eigenstates of a Hamiltonian equation, for which five algorithms are introduced: imaginary-time evolution, steepest gradient descent, an improved gradient descent, an implicitly restarted Arnoldi method, and density matrix renormalization group (DMRG) optimization. The first four methods are engineered using a framework of limited-precision linear algebra, where operations between MPS and matrix product operators (MPOs) are implemented with finite resources. All methods are benchmarked using the PDE for a quantum harmonic oscillator in up to two dimensions, over a regular grid with up to 2^{28} points. Our study reveals that all MPSbased techniques outperform exact diagonalization techniques based on vectors, with respect to memory usage. Imaginary-time algorithms are shown to underperform any type of gradient descent, both in terms of calibration needs and costs. Finally, Arnoldilike methods and DMRG asymptotically outperform all other methods, including exact diagonalization, as problem size increases, with an exponential advantage in memory and time usage.
This work presents a comparative study of new and existing optimization and diagonalization methods for solving time-independent partial differential equations (PDEs) using matrix product states (MPS) in the quantized tensor-train formalism (QTT). This study focuses on Hamiltonian equations, for which five algorithms are introduced: explicit imaginary-time evolution methods, steepest gradient descent in conventional and optimized forms, a power method, and an explicitly restarted Arnoldi method. The first five methods are engineered using a framework of limited-precision linear algebra, in which operators -- i.e., the equation itself -- and vectors are represented using matrix product operator (MPO) and matrix product state (MPS) formalisms, and where operator-vector multiplication and vector addition are approximated with limited resources. All methods are benchmarked using an exactly solvable PDE for a quantum harmonic oscillator in one and two dimensions over a regular grid with up to 2302^{30}230 points and compared with the density matrix renormalization group (DMRG) method. Our study reveals that all MPS-based techniques exponentially outperform exact diagonalization techniques based on vectors regarding memory usage. Imaginary-time algorithms are shown to underperform any gradient descent in terms of calibration needs and costs. Finally, MPS DMRG and interpolated Arnoldi-like asymptotically outperform all other methods, including state-of-the-art vector-based exact diagonalization, with significant advantages in time and memory use.
This is a set of lectures on tensor networks with a strong emphasis on the core algorithms involving Matrix Product States (MPS) and Matrix Product Operators (MPO). Compared to other presentations, particular care has been given to disentangle aspects of tensor networks from the quantum many-body problem: MPO/MPS algorithms are presented as a way to deal with linear algebra on extremely (exponentially) large matrices and vectors, regardless of any particular application. The lectures include well-known algorithms to find eigenvectors of MPOs (the celebrated DMRG), solve linear problems, and recent learning algorithms that allow one to map a known function into an MPS (the Tensor Cross Interpolation, or TCI, algorithm). The lectures end with a discussion of how to represent functions and perform calculus with tensor networks using the "quantics" representation. They include the detailed analytical construction of important MPOs such as those for differentiation, indefinite integration, convolution, and the quantum Fourier transform. Three concrete applications are discussed in detail: the simulation of a quantum computer (either exactly or with compression), the simulation of a quantum annealer, and techniques to solve partial differential equations (e.g. Poisson, diffusion, or Gross-Pitaevskii) within the "quantics" representation. The lectures have been designed to be accessible to a first-year PhD student and include detailed proofs of all statements.
Quantum computation is a promising approach to solving problems in physics, optimization, and numerical simulation. Quantum algorithms leverage superposition and entanglement in an exponentially large Hilbert space to encode and solve problems that are intractable for classical computers. However, these algorithms require error-corrected quantum hardware, which is not yet available. This thesis explores two alternative quantum-based approaches that promise to achieve better scalability than traditional algorithms in the near-term: Hybrid variational quantum algorithms (VQAs) and fully classical, quantum-inspired algorithms. In the first part, we review and benchmark optimization methods well suited for VQAs, specifically the Simultaneous Perturbation Stochastic Approximation (SPSA). We introduce complex-variable extensions and improvements to the second-order and quantum natural implementations. Finally, we assess their performance across quantum control, quantum tomography, and the variational quantum eigensolver, providing practical guidelines for common VQA scenarios. In the second part, we apply Matrix Product States (MPS) to solve time-dependent partial differential equations (PDEs). By integrating Hermite Distributed Approximating Functionals (HDAF) into the MPS framework, we develop a novel encoding to efficiently approximate differential operators with exponential precision. We benchmark this approach against traditional schemes for simulating the expansion of a levitated nanoparticle. The quantum-inspired numerical integration schemes deliver large memory savings and adequate runtimes even in presence of moderate chirping.
A generic method to investigate many-body continuous-variable systems is pedagogically presented. It is based on the notion of matrix product states (so-called MPS's) and the algorithms thereof. The method is quite versatile and can be applied to a wide variety of situations. As a first test, we show how it provides reliable results in the computation of fundamental properties of a chain of quantum harmonic oscillators achieving off-critical and critical relative errors of the order of ${10}^{\ensuremath{-}8}$ and ${10}^{\ensuremath{-}4}$, respectively. Next, we use it to study the ground-state properties of the quantum rotor model in one spatial dimension, a model that can be mapped to the Mott insulator limit of the one-dimensional Bose-Hubbard model. At the quantum critical point, the central charge associated with the underlying conformal field theory can be computed with good accuracy by measuring the finite-size corrections of the ground-state energy. Examples of MPS computations both in the finite-size regime and in the thermodynamic limit are given. The precision of our results is found to be comparable to that previously encountered in the MPS studies of, for instance, quantum spin chains. Finally, we present a spin-off application: an iterative technique to efficiently get numerical solutions of partial differential equations of many variables. We illustrate this technique by solving Poisson-like equations with precisions of the order of ${10}^{\ensuremath{-}7}$.
Tensor Networks are a relatively new machine learning approach. The architectures proposed initially are inspired by approaches from quantum many-body physics simulations. One common layout is the matrix product state (MPS) also known as a tensor train optimized with gradient descent techniques. We introduce a global normalization condition, so that the MPS represents a quantum state. We investigate two optimization methods that find the locally optimal tensors and compare them regarding their effectiveness. One is based on gradient descent and the other on an adaptation of DMRG.
Quantum methods for solving partial differential equations (PDEs) generally fall into a few general categories, each leveraging different quantum mechanisms to achieve potential speedups. One well-known approach is the quantum linear system solver, epitomized by the Harrow-Hassidim-Lloyd (HHL) algorithm, which uses phase estimation and eigenvalue inversion to solve large, sparse linear systems arising from discretized linear PDEs [1]. This method can yield exponential improvements over classical solvers under certain conditions (mostly when the matrix is well-conditioned and sparse). More recent developments, such as Quantum Singular Value Transformation (QSVT), provide a framework for matrix inversion in a more flexible way, often improving on the cost and error dependence of earlier solvers [2]. Time-dependent PDEs can be tackled via Hamiltonian simulations. The Trotter-Suzuki method decomposes the exponential of a Hamiltonian into products of exponentials of simpler sub-Hamiltonians, allowing a piecewise approximation of continuous dynamics [3]. Continuous-time quantum walks, which treat a PDE operator as a Hamiltonian that governs the walk, similarly simulate processes like diffusion or wave propagation on discretized grids or graphs [4]. Beyond direct simulation, variational quantum algorithms have emerged as a promising route in the noisy intermediate-scale quantum era. These approaches encode the PDE solution as a parameterized quantum state and optimize the parameters via classical feedback, thereby minimizing a cost function tied to the PDE residual or energy functional [5]. Quantum annealing or adiabatic methods, meanwhile, can solve optimization problems derived from PDE constraints by adiabatically evolving a simple initial Hamiltonian to the final problem Hamiltonian and staying in the system's ground state [3]. In our study, we implement a new parameterized quantum model, specifically designed to solve large linear systems on quantum hardware. This model is less deep and computationally faster than the proposed one in [5], reducing the complexity and speeding up convergence times. Such improvements are particularly valuable in cases where classical algorithms may fail due to the problem's size or complexity. Ongoing research is focusing on generalized quantum models that can be adapted to practical PDE problems for solving large electromagnetic problems where classical computers can fail.