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

Abstract: 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.

• The paper demonstrates that tensor network methods such as MPS/MPO and TCI enable efficient, high-fidelity quantum circuit emulation.
• It rigorously analyzes compression techniques like SVD to quantify entanglement and optimize simulation fidelity based on bond dimension.
• The work introduces the quantics representation for solving large-scale PDEs on massive grids, challenging claims of exclusive quantum advantage.

Quantum-Inspired Tensor Network Computational Techniques: Scope, Algorithms, and Applications

Introduction

This work presents a rigorous, methodically structured survey of tensor network algorithms, emphasizing Matrix Product States (MPS), Matrix Product Operators (MPO), and Tensor Cross Interpolation (TCI), with an explicit focus on computational tasks that are typically associated with quantum computers. The exposition is distinctly pedagogical yet technically comprehensive, bridging foundational linear algebra, entanglement theory, tensor network compression strategies, and their intersection with quantum circuit simulation and the numerical solution of high-dimensional problems. The formalism maintains clear separation between tensor networks as abstract representations for large-scale linear algebra and their traditional quantum many-body physics heritage.

Tensor Networks and Quantum Computation

The manuscript initiates with a critical analysis of the dimensionality challenge in quantum computing. It highlights that while the state space for $N$-qubit systems nominally grows as $2^N$, prohibiting naïve classical simulation for $N \gtrsim 50$, structured approaches based on tensor networks can effectively circumvent apparent infeasibility for a significant subset of computational states. It rigorously traces quantum circuits to tensor network contractions, establishing that unitary gate application, sampling, and amplitude estimation can be made formally equivalent to specific contraction sequences.

Figure 1: GHZ state generation runtime scaling demonstrating the superiority of the MPS approach over the state-vector simulator for systems with growing numbers of qubits.

Crucially, the work identifies and discusses the "I/O bottleneck" intrinsic to quantum computers, noting that only marginal statistical output (typically $N$ bits per run) is available, despite the exponentially large computational Hilbert space. The authors emphasize that the operationally reachable quantum state manifold is dramatically smaller than the full Hilbert space, given hardware constraints on circuit depth and gate parameterization.

Canonical MPS/MPO Algorithms and Compression

The manuscript systematically develops the standard computational toolkit for MPS and MPO representations, treating MPS/MPOs as generic representations of exponentially large vectors/matrices rather than quantum states/observables per se. Extensive discussion is provided on tensor contractions, index fusing/flattening, and the associated factorizations (QR, SVD, LU/prrLU). The authors stress the dual role of SVD: as a vehicle for optimal compression (in the Frobenius norm) and as a quantifier of bipartite entanglement, making rigorous connections to area- and volume-law entanglement scaling.

Subsequently, the document details state-of-the-art algorithms for both exact and approximate quantum circuit emulation:

• Exact simulation: Full state-vector contraction and exact MPS simulation, scalable in practice only for shallow or low-entanglement circuits.
• Approximate simulation: Time-Evolving Block Decimation (TEBD) and DMRG algorithms, with fidelity analysis demonstrating exponential error scaling tied to discarded singular values.

  Figure 2: Fidelity decay in MPS-based random quantum circuit simulation, displaying exponential scaling in bond dimension, in agreement with theoretical prediction.

The accuracy-resource trade-off, governed by the truncated bond dimension $\chi$, is quantified, and the importance of operating in canonical gauge for optimal SVD-based compression is underlined.

Tensor Cross Interpolation and Active Learning

A focal point of the lecture is the detailed derivation and application of Tensor Cross Interpolation (TCI), which enables construction of low-rank tensor network surrogates of functions or datasets by sparse sampling—thus circumventing the exponential memory bottleneck. The exposition includes:

• Cross interpolation for matrices, based on partial rank-revealing LU and Schur complements, operationalized with maxvol (maximum-volume) pivot selection.
• Extension to tensors, which constructs MPS representations by iteratively adding pivot elements, with explicit multi-index bookkeeping.

The TCI method is notably efficient when the target function is of low entanglement or can be well-approximated by polynomials, as evidenced by sharp convergence in high-dimensional integration benchmarks.

Figure 4: Convergence plot of TCI for a highly oscillatory 10D integral, showcasing rapid precision gains versus number of function evaluations, unattainable by Monte Carlo methods in presence of a sign problem.

Quantics: High-Dimensional Function Representation and PDE Solvers

A salient contribution of the work is the exploration of the "quantics" representation, where a function on a hypercubic grid is mapped to an MPS in its binary index decomposition. This representation is shown to foster exponentially efficient encoding for wide classes of problems—especially those with multi-scale structure and limited inter-bit entanglement. The quantics machinery supports construction of low-rank MPOs for operations such as differentiation, integration, and linear transforms, including the Quantum Fourier Transform (QFT).

The quantics approach is leveraged to design algorithms for direct and efficient solution of PDEs (e.g., Poisson, heat, Schrödinger, Gross–Pitaevskii), with grid sizes reaching $2^{20}$ and beyond in each dimension, vastly exceeding the feasible capacity of direct grid-based solvers.

Figure 6: Simulation of the Gross–Pitaevskii equation using quantics, exploiting trillions of grid points on a classical workstation, demonstrating the scalability of tensor network solvers for nonlinear PDEs.

Figure 8: Utilization of the low-rank QFT-MPO to solve the heat equation, enabling treatment of non-smooth, highly oscillatory solutions on massive grids.

Discussion and Implications

The authors rigorously critique both the scope and limitations of tensor network methods for quantum circuit and many-body simulation, highlighting that their efficacy is ultimately restricted by the entanglement structure of the target problem (volume-law states remain generically hard). The formalism introduced—with TCI as a generic interface—extends the range of applications to domains classically considered exclusive to quantum advantage, questioning hard claims of quantum supremacy for circuits or PDE instances with structured/low-entanglement solutions.

On the practical side, the modularity of MPS/MPO representations and their operations (e.g., direct sampling, MPO-MPS multiplication, linear system solvers, eigenproblems) enables construction of scalable algorithms for high-dimensional, dense linear algebra, nonlinear dynamics, and spectral transformations. The formal equivalence of the QFT to a low-rank MPO directly underscores that certain "quantum speedup" results (e.g., for QFT-based algorithms) may not be as exclusive to quantum hardware as traditionally posited.

Future Prospects

Technologically, the stratified tensor network framework—augmented with adaptive, learning-based factorization (TCI)—opens the door to classical emulation of a broader class of quantum algorithms and efficient numerical solutions in high-dimensional computational science. Open research fronts include the systematic characterization of function classes admitting low-rank quantics representations, extension to non-MPS network geometries (e.g., tree tensor networks), efficient boundary condition encoding in PDE contexts, and bridging TCI with scalable neural or kernel-based surrogates for high-entropy datasets.

Conclusion

This lecture series provides a technically robust yet highly accessible guide to the tensor network apparatus, emphasizing its universal linear algebraic scope beyond quantum many-body systems and quantum simulation. With core algorithms for MPS/MPO manipulation, TCI-based active learning, and quantics PDE solution, the work substantiates the claim that—when leveraged judiciously—classical tensor network methods can rival or surpass quantum computers for a wide domain of feasible, structured computational problems.