🔢 mathematics

Products of unitaries as continuous evolutions at finite precision

This paper presents a bidirectional correspondence between continuous unitary evolutions and products of unitaries at finite precision, enabling the reduction of discrete adiabatic theorems to continuous ones with explicit $O(1/T)$ error bounds and eliminating the need for separate discrete adiabatic analysis in applications like quantum linear systems.

Original authors: Michael Jarret

Published 2026-10-06

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Michael Jarret

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the quantum world, time does not always flow smoothly in the way we experience it. To simulate how a quantum system changes over time, scientists often have to break that flow into tiny, frozen snapshots. Imagine trying to film a river with a camera that can only take a single picture every second; you would miss the current's fluid motion and only see a series of still images. In quantum computing, this is a fundamental challenge. Researchers must translate the continuous, smooth evolution of a system, driven by a changing energy field, into a sequence of discrete steps that a computer can actually execute. This process is essential for solving complex problems, from designing new materials to simulating chemical reactions, but it introduces errors. The faster the system changes, or the more precise the simulation needs to be, the harder it becomes to keep the digital steps aligned with the continuous reality they are meant to represent.

Michael Jarret's work addresses this gap by showing that, under the right conditions, these two ways of looking at time—the smooth flow and the stepped sequence—are effectively the same thing. He demonstrates a precise mathematical bridge that allows scientists to move back and forth between continuous theories and discrete computer instructions without losing the essential guarantees of the original theory. Instead of treating the discrete steps as a rough approximation that requires its own separate, complicated rules to prove it works, Jarret shows that if you take a continuous path and simply round off the numbers to a finite precision, you get a valid sequence of steps. Conversely, if you have a sequence of steps taken at regular intervals, you can reconstruct the continuous path they represent, provided the steps are small enough and the system's energy levels are spaced in a specific way.

The core of this discovery lies in how the errors behave. When a system evolves continuously, its state changes gradually. When we force it into discrete steps, we introduce a small mismatch at every turn. Jarret found that by carefully correcting for the first few moments of change within each step, these mismatches cancel each other out almost perfectly. The remaining error shrinks predictably as the steps get smaller. This means that if a continuous process is known to be stable and accurate over a long period, a corresponding sequence of discrete steps will also be stable and accurate, with only a tiny, calculable amount of extra error added. This is a significant shift in perspective because it allows researchers to use powerful, well-understood theorems about continuous motion to prove that their discrete computer algorithms will work, rather than having to reinvent the wheel for every new digital simulation.

One of the most practical applications of this finding appears in the field of quantum linear systems, which involves solving complex equations to find specific states of matter. Previous methods for solving these problems required a number of computational steps that grew rapidly with the difficulty of the problem. Jarret's approach confirms that a specific type of digital walk, composed of a sequence of unitary steps, can solve these problems with a number of steps that scales efficiently with the system's condition number and the desired precision. This matches the best-known performance of previous methods but arrives there through a simpler, more direct route that relies on continuous theory rather than complex discrete proofs. The result is a confirmation that the digital approximation is not just a clumsy substitute for the continuous reality, but a faithful partner that can inherit the reliability of the original theory.

The paper also clarifies what happens when the system's energy levels are very close together or when the path of change is irregular. Jarret shows that as long as the energy levels do not hit certain specific, problematic values relative to the step size, the method remains robust. If the steps are too large or the energy levels are too crowded, the cancellation of errors fails, and the simulation breaks down. However, within the safe zone, the method holds firm. This provides a clear set of rules for engineers and scientists: they can design their quantum circuits by first thinking about the smooth, continuous flow of the system, knowing that they can translate this into a finite sequence of operations with a guaranteed level of accuracy.

By establishing this two-way correspondence, the work removes a layer of uncertainty from quantum simulation. It suggests that the distinction between continuous evolution and discrete computation is less of a barrier and more of a translation layer. For the curious observer, this means that the complex machinery of quantum computing is not built on a foundation of fragile approximations, but on a solid connection between the smooth laws of physics and the step-by-step logic of the machine. The paper does not claim to have solved every problem in quantum simulation, nor does it suggest that all continuous theories can be instantly digitized without cost. Instead, it offers a rigorous, proven method to ensure that when we do digitize, we do not lose the very properties that make the system interesting in the first place. The result is a clearer path forward for building quantum computers that can reliably tackle the hardest problems in science.

Technical Summary: Products of Unitaries as Continuous Evolutions at Finite Precision

Problem Statement
The paper addresses the fundamental challenge of discretizing continuous-time unitary evolution for simulation within the quantum circuit model. Typically, simulating a continuous evolution generated by a time-dependent Hamiltonian $H(t)$ requires approximating it with a finite product of unitaries or a piecewise-constant Hamiltonian. Existing approaches often rely on specific discrete adiabatic theorems or impose strict conditions on the discretization steps. The author seeks to establish a rigorous, bidirectional correspondence between continuous unitary evolutions and products of unitaries at finite precision, allowing theorems and bounds derived in the continuous setting to be transferred directly to the discrete setting without re-proving them from scratch.

Methodology
The paper employs two primary directions of comparison, utilizing the Duhamel formula for evolution error bounds and the Davis–Kahan $\sin \Theta$ theorem for spectral subspace (leakage) bounds.

Continuous to Discrete (Truncation):
- Bit Truncation: The author defines a method where the real and imaginary parts of the Hamiltonian entries are truncated to $k$ -bit precision. If the truncated Hamiltonian $H^{(k)}(t)$ is piecewise constant, it induces a product of unitaries $W$ .
- Midpoint Sampling: For regulated families (those with one-sided limits) where simple truncation may not be piecewise constant, the author proposes holding "truncated midpoint samples" constant on intervals defined by "equal-variation checkpoints." This controls the variation within each interval.
- Error Analysis: Using the Duhamel estimate, the paper bounds the operator-norm distance between the continuous evolution $U(T)$ and the discrete product $W$ . The error is shown to be proportional to the truncation error and the total evolution time.
- Adiabatic Transfer: By combining the evolution error with the Davis–Kahan theorem, the paper demonstrates that leakage bounds (probability of leaving a spectral subspace) for the continuous evolution transfer to the discrete product, provided the spectral gaps are maintained.
Discrete to Continuous (Rounding the Clock):
- Unit-Step Sampling: The paper considers a product of unitaries $W_T$ formed by sampling a smooth Hamiltonian path at integer time steps (unit steps).
- Frame Correction: Direct comparison via Duhamel yields only an $O(1)$ bound. To improve this, the author introduces a "frame" operator (a Hermitian correction) that cancels first-order changes within each time step. This relies on the assumption that the eigenvalue differences of the sampled Hamiltonians are bounded away from nonzero multiples of $2\pi$ (separation condition).
- Resulting Bound: With this correction, the product $W_T$ (up to two half-step phase factors) approximates the continuous evolution with an error of $O(1/T)$ . This allows continuous adiabatic bounds to transfer to the discrete product.

Key Contributions and Results

Bidirectional Correspondence: The paper establishes that continuous evolutions and products of unitaries are effectively the same object under finite precision, provided specific regularity and spectral conditions are met.
"Theorem" A & B (Fine-Step Discretization):
- For a $\sigma$ -sparse, regulated Hamiltonian, bit truncation yields a product $W$ such that $\|W - U(T)\| \le \sqrt{2}\sigma 2^{-k} T$ .
- Using truncated midpoint samples with equal-variation checkpoints, a product within error $\epsilon$ can be constructed with $O(T\ell/\epsilon)$ factors, where $\ell$ is the total variation of the Hamiltonian.
"Theorem" C (Unit-Step Comparison):
- For a twice-differentiable Hamiltonian with bounded derivatives and spectral separation, the phase-corrected product of unit-step samples approximates the continuous evolution with error $O(1/T)$ .
- Crucially, this transfer preserves the hypotheses of the continuous adiabatic theorem. If a continuous theorem guarantees leakage $C/T$ , the discrete product achieves leakage $O(1/T)$ with the same path-dependent constants.
Application to Quantum Linear Systems:
- The paper applies these results to the problem of solving linear systems $A^{-1}b$ using adiabatic quantum computing.
- By utilizing An and Lin's path with an $AQC(3/2)$ schedule, the author shows that the continuous adiabatic bound (leakage $O(\kappa/T)$ ) transfers to the discrete product.
- This yields a rescaled qubitized walk requiring $O(\kappa/\delta)$ steps for leakage $\delta$ , matching the scaling previously derived via discrete adiabatic theorems but without needing a separate discrete proof.
Walks and Logarithms: The methodology extends to "walks" (ordered sequences of unitaries) by constructing a Hermitian logarithm path within a spectral arc free of eigenvalues, allowing the transfer of continuous adiabatic bounds to the walk.

Significance and Claims
The paper claims that its primary significance lies in the transfer of theorems. Instead of proving discrete adiabatic theorems from scratch for every new discretization scheme or path, researchers can derive discrete results directly from established continuous theorems.

Modesty on Precision: The author notes that the correspondence is "first order." While continuous theorems can offer exponential decay for Gevrey or analytic families, the current discretization methods (bit truncation and unit-step rounding) introduce errors that scale polynomially ( $O(1/T)$ or $O(\epsilon)$ ). Consequently, transferring exponential decay bounds would require precision that grows linearly with $T$ , potentially leading to an exponential number of factors, which the paper identifies as an open question.
Implementation: The approach is practical for both digital quantum circuits (via qubitization) and analog hardware (where finite-resolution controls naturally truncate Hamiltonian entries).
Verification: The author notes that the numbered results and constants have been verified using the Lean 4 proof assistant, ensuring the rigorous correctness of the bounds and constants derived.

In summary, the paper provides a unified framework for discretizing unitary evolution, demonstrating that under finite precision and specific spectral conditions, discrete products faithfully replicate continuous dynamics and their adiabatic properties, thereby simplifying the theoretical analysis of quantum simulation algorithms.

More like this