🔢 mathematics

Improved convergence radius of the Fer expansion for Hermitian generators

This paper improves the known convergence radius of the Fer expansion for Hermitian generators from 2 to approximately 2.6058, thereby extending the range of validity for this unitary, doubly-exponentially convergent approximation of time-dependent quantum propagators.

Original authors: Lorenzo Bagnasacco, Vittorio Giovannetti

Published 2026-09-24

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Lorenzo Bagnasacco, Vittorio Giovannetti

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

The story of how a quantum system changes over time is a central puzzle in modern physics. When scientists describe a quantum system, they use a mathematical object called a Hamiltonian, which acts like a set of instructions telling the system how to evolve. If these instructions change as time passes, the math becomes incredibly difficult. To solve it, researchers often rely on a method called the Dyson series, which breaks the problem down into a long list of smaller pieces. However, this approach has a fatal flaw: if you stop the list at any point to get a practical answer, the result loses a fundamental property called unitarity. In the quantum world, losing this property is like losing the ability to conserve probability; the math stops making physical sense.

To fix this, scientists have developed alternative ways to organize the calculation that keep the math physically valid at every step. One famous method, known as the Magnus expansion, bundles everything into a single exponential formula. Another approach, the Fer expansion, does something different: it builds the solution as a chain of separate steps, where each step is a perfect, self-contained unit. This method is prized because it converges very quickly, meaning fewer steps are needed to get a highly accurate answer. But for a long time, this speed came with a strict limitation. The method was only guaranteed to work if the total "strength" of the changing instructions stayed below a specific threshold. If the instructions became too strong or lasted too long, the chain would break, and the math would fail.

In a recent study, researchers Lorenzo Bagnasacco and Vittorio Giovannetti have pushed this limit significantly further. They discovered a way to make the Fer expansion work for systems that are roughly thirty percent stronger or longer-lasting than previously thought possible. The key to their success was not a new formula or a complex computer simulation, but a simple geometric insight about how numbers behave on a circle. By rethinking how they measured the distance between mathematical steps, they found a way to squeeze more efficiency out of the method without changing its fundamental structure.

The Fer expansion works by peeling away the main part of the problem, leaving behind a smaller, simpler remainder to be solved in the next step. Imagine trying to walk across a room by taking steps that get progressively smaller. The method ensures that each step is a perfect, valid move. The problem arises when the initial instructions are so intense that the first step is too large to be handled safely. Previous research had proven that as long as the total intensity of the instructions stayed below a value of two, the method would work. This value of two was considered the best possible limit known to science.

The authors of this new paper realized that the standard way of measuring this intensity was too conservative. The old method measured the distance of a mathematical step from a specific reference point, much like measuring how far a point on a clock face is from the number twelve. However, they noticed that in the mathematics of quantum mechanics, any point on the clock face could serve as a reference, provided it was chosen carefully. The researchers asked a simple question: what if we chose the reference point that was closest to the actual path the step was taking?

This question turned out to be a problem of basic geometry. The path of a quantum step traces an arc along the edge of a circle. The old method measured the length of the chord connecting the start and end of this arc, which is a straight line. The new method looked for the smallest possible circle that could completely contain that entire arc. By finding the center of this smallest circle, they could measure the distance more accurately. This geometric optimization allowed them to show that the step could be larger than previously allowed before the method broke down.

Using this refined measurement, the researchers calculated a new limit. They found that the Fer expansion is guaranteed to converge as long as the total intensity of the instructions stays below approximately 2.6058. This is a significant improvement over the previous limit of two. The researchers proved this result mathematically, showing that the method remains stable and accurate within this new, wider range. They also demonstrated that the speed at which the method converges remains just as fast as before; the improvement lies entirely in the size of the problems it can now solve, not in how quickly it solves them.

This finding matters because it expands the range of physical systems that can be modeled with high precision using this efficient technique. In fields like nuclear magnetic resonance, where scientists manipulate quantum spins to create images or study materials, having a method that works for stronger signals or longer durations is invaluable. The researchers did not invent a new way to calculate quantum evolution; they simply found a way to make an existing, powerful tool work in more situations. By optimizing a single geometric choice, they removed a barrier that had stood for decades, allowing the Fer expansion to handle a broader class of quantum problems with the same reliability and speed it has always been known for.

Technical Summary: Improved Convergence Radius of the Fer Expansion for Hermitian Generators

Problem Statement
The time evolution of a quantum system governed by a time-dependent Hamiltonian $H(t)$ is formally described by the time-ordered exponential. While the Dyson series provides a power-series expansion, its finite-order truncations generally fail to preserve unitarity. To address this, the Fer expansion represents the propagator as an infinite product of matrix exponentials, $U(t) = \prod e^{F_j(t)}$ , where each factor is unitary by construction. A significant advantage of the Fer expansion over other unitary-preserving methods (such as the Magnus expansion or Trotter-Suzuki formulas) is its doubly exponential convergence rate once the series converges.

However, the convergence of the Fer expansion is guaranteed only within a finite radius. Previous work established that for a general generator, the time integral of the norm must be small ( $\int \|A(s)\| ds \lesssim 0.8604$ ). Specifically, for Hermitian Hamiltonians (where the generators are anti-Hermitian), the known sufficient convergence radius was $\rho_{\text{old}} = 2$ . This value, derived in Ref. [14], represents the condition $\int_0^t \|H(s)\| ds < 2$ . The authors note that this radius is smaller than that of the Magnus expansion ( $\pi$ ), despite the Fer expansion's superior convergence speed.

Methodology
The paper employs a classical fixed-point analysis strategy to derive a tighter sufficient condition for convergence. The core of the method involves bounding the size of the effective Hamiltonian at each step of the Fer recursion.

Recursive Structure: The Fer expansion is defined by a recursion where the remainder Hamiltonian $H_{j+1}(t)$ is generated from the previous step via a linear map $R_{F_j}$ acting on $H_j(t)$ . This map involves unitary conjugations and integrals of commutators.
Finite-Angle Estimate: The authors focus on bounding the norm of the commutator $[e^{-\mu X}, Y]$ , where $X$ is anti-Hermitian. The standard elementary estimate replaces the identity matrix with itself ( $c=1$ ), leading to a bound proportional to the arc length $\mu\|X\|$ .
Geometric Optimization: The key methodological innovation is the observation that the identity matrix in the commutator bound can be replaced by any scalar multiple $c\mathbb{1}$ $c 1$ . Since $c\mathbb{1}$ $c 1$ commutes with all operators, the commutator remains unchanged. The authors optimize the choice of $c$ $c$ to minimize the distance between the spectrum of the unitary operator $e^{-\mu X}$ $e^{- μ X}$ (which lies on an arc of the unit circle) and the point $c$ $c$ .
- This is formulated as a plane geometry problem: finding the radius of the smallest disc centered at $c$ that contains the arc $|\theta| \le \mu\|X\|$ .
- For $\mu\|X\| \le \pi/2$ , the optimal center is $c = \cos(\mu\|X\|)$ , yielding a bound proportional to $\sin(\mu\|X\|)$ .
- For $\mu\|X\| > \pi/2$ , the arc contains antipodal points, and the optimal radius is 1 (centered at the origin).
Scalar Recursion: This geometric optimization yields a new scalar function $\psi(\|X\|)$ that bounds the growth of the Hamiltonian norms. By integrating this function, the authors define a map $\Psi(x)$ governing the evolution of the cumulative norm $K_j(t) = \int_0^t \|H_j(s)\| ds$ . The convergence radius is determined by the first non-trivial positive fixed point $\rho$ where $\Psi(\rho) = \rho$ .

Key Contributions and Results

Improved Convergence Radius: The authors derive a new sufficient convergence radius for the Fer expansion with Hermitian generators. By solving the fixed-point equation numerically, they find:
$\rho_{\text{new}} \approx 2.6058$
This represents an improvement of approximately 30% over the previously known value of $\rho_{\text{old}} = 2$ .
Preservation of Convergence Rate: The analysis shows that near the origin (small norms), the new map $\Psi(x)$ agrees with the old map $M(x)$ to leading order. Specifically, $\Psi(x) = M(x) - x^4/48 + O(x^6)$ . Consequently, the asymptotic doubly exponential decay rate of the error remains unchanged. The improvement solely extends the domain of initial data (Hamiltonians) for which this rapid convergence is guaranteed.
Explicit Error Bound: The paper provides an explicit error bound for the truncated product. If the initial cumulative norm $K_1(t) < \rho_{\text{new}}$ , the error after $N$ factors is bounded by the $N$ -fold iterate of $\Psi$ applied to $K_1(t)$ , which decays doubly exponentially.

Significance
The paper claims that the improvement stems from a single geometric observation regarding the spectrum of unitary conjugations. By optimizing the reference point for the commutator bound, the authors replace the linear arc-length estimate with a tighter geometric bound (the radius of the minimal enclosing disc).

The significance lies in expanding the applicability of the Fer expansion. While the rate of convergence is already known to be superior to other methods, the strict radius condition ( $\rho=2$ ) had limited its use for stronger Hamiltonians or longer time intervals. The new bound ( $\rho \approx 2.6058$ ) allows the Fer expansion to be rigorously applied to a broader class of time-dependent Hamiltonians while maintaining its unitary structure and fast convergence properties. The authors note that this result improves upon the best-known value to date without altering the fundamental structure of the expansion or its asymptotic behavior.

Technical Summary: Improved Convergence Radius of the Fer Expansion for Hermitian Generators

More like this