Quantum states mimic classical polynomials with new approximation method

Sviatoslav V. Dzhenzher has demonstrated a quantum approximation framework where the resulting measurement probability precisely reproduces the multidimensional Bernstein polynomial of a target function. This work establishes a quantum analogue of the classical Kolmogorov-Arnold representation theorem, addressing limitations in Quantum Neural Network expressivity. Originating from Hilbert’s 13th problem, which asked if multivariate continuous functions can be expressed as superpositions, the research links classical superposition theory with probabilistic and coherent quantum circuits. The approach factorises a multidimensional operator into 2d + 1 commuting universal inner observables, where ‘d’ represents the dimensionality of the input.

Quantum Bernstein-Kolmogorov-Arnold Approximation for Function Representation

Sviatoslav V. Dzhenzher demonstrated this within a constructive approximation framework using fixed input coherent states, embedding function arguments in coordinate qubits and dictating a single outer unitary evolution. This approach uses a Sprecher-type representation applied to the rotation angle of the quantum state, achieving the described factorization and potentially simplifying the computational pathway.

A surprising connection between quantum mechanics and classical mathematics emerges from the precise reproduction of a multidimensional Bernstein polynomial by the resulting measurement probability. This is not merely a functional equivalence; the quantum measurement outcomes directly mirror the values generated by a well-established classical polynomial, linking the probabilistic nature of quantum states to deterministic mathematical constructs.

Kolmogorov and Arnold later extended this to univariate functions and addition, and Dzhenzher now establishes a quantum analogue of their representation theorem. Dzhenzher writes that “Theorem 2.2 is equipped with all the delightful properties of the classical Bernstein approximation,” noting that for Lipschitz functions, the convergence of the quantum state to the target output can be estimated using the function’s Lipschitz constant. The research establishes a quantum Bernstein-Kolmogorov-Arnold approximation theorem for functions on the unit cube, offering a new tool for exploring function representation in quantum systems.

Kolmogorov-Arnold Theorem: Classical Foundations and Quantum Analogues

Researchers focused on bridging a gap between theoretical quantum results and the classical Kolmogorov-Arnold representation theorem, a challenge highlighted by existing limitations in representation theory. Their approach uses a fixed-input-state model where the function’s argument is encoded within coordinate qubits, while a unitary operator, dependent on the target function, governs the evolution of the quantum state.

This structure allows for a specific organization with universal inner observables and a single outer function, applicable across different approximation levels, mirroring the classical theorem’s construction. Further research directions include exploring quantum counterparts of embedding techniques and investigating whether Sternfeld arrays, mathematical structures used to analyze function approximation, can reveal limitations or universality criteria in quantum systems, potentially complementing constructive results with impossibility proofs.

Sprecher Representation Factorizes Unitary Evolution into 2d+1 Observables

The approach embeds the function’s argument entirely within coordinate qubits, leaving a single outer unitary evolution dictated by the target function itself. This correspondence is notable because it connects the abstract realm of quantum mechanics to a concrete, well-defined polynomial representation used extensively in classical approximation theory. The researchers’ framework operates on fixed input coherent states, a deliberate choice to isolate the function’s dependence on the input coordinates and simplify the overall quantum circuit design.

Originating from a challenge posed over a century ago, Hilbert’s 13th problem, this research builds upon the work of Kolmogorov and Arnold, who previously addressed the problem for univariate functions and addition. The current study extends this classical foundation to the quantum domain, proving uniform convergence of the joint quantum state to the desired target output register. The findings offer a constructive approximation framework with implications for the development of Quantum Neural Networks and address the expressivity and representation gaps within the field.

Coherent State Encoding Embeds Function Arguments into Coordinate Qubits

The approach embeds a function’s arguments directly into coordinate qubits, a strategy that simplifies the quantum representation of complex mathematical operations. Dzhenzher’s approach uses fixed input coherent states, deliberately isolating the function’s dependence within the coordinate qubits to achieve this.

Bernstein Polynomial Convergence Achieves Uniform Approximation of Target Functions

The researchers proved that this factorization holds true for any dimension ‘d’, offering a potentially scalable approach to quantum function approximation. The team’s approach does not require the target function to be continuous to establish the relationship, though convergence estimates are stronger for Lipschitz functions. The method was successfully demonstrated for functions defined on the unit cube, opening possibilities for applications in areas requiring accurate function representation.

👉 More information

🗞 Quantum Bernstein–Kolmogorov–Arnold approximation

✍️ Sviatoslav V. Dzhenzher

🧠 ArXiv: https://arxiv.org/abs/2610.41608