Error bounds for the truncated Baker--Campbell--Hausdorff and Zassenhaus formulas in unitary problems

Technical Summary: Error Bounds for Truncated Baker–Campbell–Hausdorff and Zassenhaus Formulas in Unitary Problems

Problem Statement
The Baker–Campbell–Hausdorff (BCH) formula and the Zassenhaus formula are fundamental tools in Lie theory, expressing the logarithm of a product of exponentials and the exponential of a sum of non-commuting operators, respectively, as infinite series of nested commutators. In practical applications, particularly in the numerical analysis of differential equations and quantum simulation, these infinite series must be truncated. While the convergence domains of these series are well-studied, rigorous error bounds for the resulting truncated approximations—specifically in terms of the norms of nested commutators—have been largely absent from the literature, with the exception of the trivial Lie–Trotter case. The lack of such bounds hinders the precise quantification of gate complexity and error scaling in quantum algorithms.

Methodology
The authors propose a general strategy to derive rigorous error bounds for both the BCH and Zassenhaus formulas when the operators involved are skew-adjoint (a condition satisfied in unitary quantum evolution problems). The methodology relies on the following steps:

1. Differential Equation Formulation: The problem is recast by defining two time-dependent unitary operators, $F_1(t)$ and $F_2(t)$, where $F_1(t)$ represents the exact product (or sum) of exponentials and $F_2(t)$ represents the truncated approximation. Both satisfy initial value problems of the form $\frac{dF}{dt} = M(t)F$.
2. Generator Analysis: The authors analyze the difference between the generator operators $M_1(t)$ and $M_2(t)$. For the BCH case, $M_1(t)$ is derived from the product $e^{tA}e^{tB}$, while $M_2(t)$ is derived from the exponential of the truncated series $e^{Z^{[m]}(tA, tB)}$.
3. Integral Remainder Estimates: Using Lemma 2.1, which bounds the difference of two unitary operators by the integral of the difference of their generators, the authors reduce the error estimation to bounding $\|M_1(t) - M_2(t)\|$.
4. Taylor Expansion with Remainder: The generators are expanded in powers of the parameter $t$ up to the truncation order $m$, utilizing integral forms of the remainder (Lemma 2.2). This allows the authors to isolate the leading error terms, which are expressed as linear combinations of nested commutators.
5. Recursive Extension: The approach is generalized to handle $N$ operators (Section 4) and the symmetric BCH formula (Section 5) by defining recursive sequences for the generator terms. For the Zassenhaus formula (Section 6), a similar recursive procedure is applied to the product of exponentials.

Key Contributions and Results

• Explicit Error Bounds for BCH: The paper derives explicit error bounds for the truncated BCH series $Z^{[m]}$. For the case of truncating after two terms ($m=2$), the authors provide the bound:
  $$\|e^{tA}e^{tB} - e^{Z^{[2]}(tA, tB)}\| \leq \frac{t^3}{4}\|[A, [A, B]]\| + \frac{t^3}{12}\|[B, [A, B]]\|$$
  This result recovers the known Lie–Trotter bound for $m=1$ and provides a new, explicit estimate for the second-order approximation.
• Explicit Error Bounds for Zassenhaus: The authors present, for the first time to their knowledge, rigorous error bounds for the truncated Zassenhaus formula. For the truncation after the first correction term ($q=2$), the bound is:
  $$\|e^{t(A+B)} - e^{tA}e^{tB}e^{t^2 C_2}\| \leq \frac{t^3}{3}\|[B, [A, B]]\| + \frac{t^3}{6}\|[A, [A, B]]\|$$
  The authors note that these coefficients are optimal.
• Generalization to $N$ Operators and Symmetric Cases: The strategy is extended to the BCH formula involving $N$ skew-adjoint operators and the symmetric BCH formula (relevant for palindromic splitting methods). Theorem 4.1 and Theorem 5.1 provide general recursive formulas for the error bounds in these cases.
• Commutator Scaling Property: A central finding is that the derived error bounds are polynomials in the parameter $t$ whose coefficients are linear combinations of norms of nested commutators. This exhibits the "commutator scaling property," which is advantageous when operators nearly commute.
• Computational Implementation: The authors provide explicit expressions for the error constants up to $m=3$ (and $m=4$ via a provided website) using the classical Hall basis of the free Lie algebra. They demonstrate via numerical experiments with random $20 \times 20$ matrices that the derived bounds are sharp, overestimating the actual error by a factor of less than 3 even for relatively large $t$.

Significance and Claims
The paper claims to address a gap in the literature regarding rigorous error analysis for truncated BCH and Zassenhaus expansions beyond the trivial first-order case. The significance of these results lies in their applicability to:

• Quantum Simulation: Providing theoretical bounds that directly impact the analysis of gate complexity and depth optimization for product formulas in quantum algorithms.
• Error Quantification: Offering bounds that depend on nested commutators, which allows for tighter estimates when the operators involved are nearly commuting.
• Rigorous Analysis: Establishing a systematic framework to generate these bounds for arbitrary truncation orders $m$ and $N$ operators, implementable via computer algebra systems.

The authors maintain a modest tone, noting that while the bounds are rigorous and explicit, their computational complexity grows exponentially with the truncation order $m$. They emphasize that the results are specific to the finite-dimensional unitary case (skew-adjoint operators) but suggest the treatment could be extended to other variants of the formulas.