Commutator-free Cayley methods

Technical Summary: Commutator-free Cayley Methods

Problem Statement
The paper addresses the numerical integration of non-autonomous linear ordinary and partial differential equations of the form $\dot{Y}(t) = A(t)Y(t)$, where the solution $Y(t)$ evolves on quadratic matrix Lie groups $G = \{X \in GL_n(\mathbb{C}) : XJX^* = J\}$. Such equations arise in classical mechanics and quantum dynamical systems (e.g., unitary groups $U_n(\mathbb{C})$ and symplectic groups). A critical requirement for these systems is the preservation of the underlying geometric structure (such as unitarity or symplecticity) to respect physical conservation laws like energy conservation and probability conservation.

Standard numerical schemes often fail to preserve these invariants. While Lie group integrators based on the Magnus expansion can preserve the structure, they face significant practical bottlenecks:

1. Nested Commutators: High-order Magnus expansions involve nested matrix commutators $[A(t_1), [A(t_2), \dots]]$, which are computationally expensive, reduce matrix sparsity, and complicate implementation.
2. Matrix Exponentials: Evaluating the matrix exponential $e^{\Omega}$ is costly. While rational approximations (like Padé approximants) preserve the Lie group structure, high-order methods require solving multiple linear systems, scaling the computational cost significantly.

Methodology
The authors propose a new class of high-order integrators that combine the Cayley–Magnus expansion with commutator-free techniques. The core methodology involves:

1. Cayley Transform as a Substitute: Instead of using the matrix exponential $e^{\Omega}$, the method utilizes the Cayley transform, defined as $\text{Cay}(\Omega) = (I - \frac{1}{2}\Omega)^{-1}(I + \frac{1}{2}\Omega)$. This transform maps the Lie algebra to the Lie group for quadratic Lie groups and is computationally cheaper than the exponential, requiring only the solution of a single linear system per step.
2. Cayley–Magnus Expansion: The authors utilize the expansion of the solution in terms of the Cayley transform, which avoids the direct computation of the exponential but traditionally still involves commutators.
3. Commutator-Free Derivation: To eliminate the nested commutators, the authors derive a Cayley–BCH (Baker–Campbell–Hausdorff) formula. They explicitly compute the expansion for the product of three Cayley transforms up to fourth-order accuracy. This allows the representation of the solution as a composition of three Cayley transforms:
   $$Y(t+\delta t) \approx \text{Cay}(\Omega_1) \text{Cay}(\Omega_2) \text{Cay}(\Omega_3) Y(t)$$
   where the arguments $\Omega_i$ are linear combinations of the matrix $A(t)$ evaluated at specific time points (derived via Gauss-Legendre quadrature).
4. Coefficient Determination: By equating the expanded product of Cayley transforms with the truncated Cayley–Magnus series (which includes terms up to order $\delta t^4$), the authors solve a system of nonlinear equations to determine the real-valued coefficients $\alpha_{i,k}$ required for the linear combinations. This ensures the resulting scheme is fourth-order accurate and maintains the Lie algebra structure (real coefficients for real Lie algebras).

Key Contributions

• New Integrator Scheme: The paper introduces the Commutator-Free Cayley Time-propagator (CFCT), a fourth-order Lie group integrator.
• Cayley–BCH Formula: A novel derivation of the Cayley-version of the BCH formula up to order four, enabling the decomposition of complex evolution operators into a product of simple Cayley transforms without nested commutators.
• Elimination of Expensive Operations: The method avoids both the computation of nested commutators and the evaluation of matrix exponentials, replacing them with Cayley transforms and linear combinations of $A(t)$.
• Structure Preservation: The scheme is designed to inherently respect the quadratic Lie group structure, ensuring the preservation of invariants like unitarity.

Results
The authors validate the method through two numerical experiments:

1. Driven Two-Level Quantum System: A spin-1/2 system in a magnetic field with an analytical solution. The CFCT method demonstrated fourth-order convergence and successfully preserved the norm (unitarity) and transition probabilities, unlike the classical RK45 method which showed norm drift.
2. Time-Dependent Schrödinger Equation: A 1D system with a time-dependent potential. The CFCT method was compared against the Commutator-Free Exponential Time-propagator (CFET4:2) and the Cayley–Magnus Time-propagator (CMT).
   • Accuracy: The CFCT method achieved accuracy comparable to the CMT method (though slightly less accurate than the reference CFET4:2 in this specific test, attributed to the choice of reference solution).
   • Efficiency: The CFCT method was reported to be more than six times faster than the CMT method due to the avoidance of commutator calculations and the lower cost of the Cayley transform compared to higher-order Padé approximations used in exponential methods.
   • Conservation: Both geometric integrators (CFCT and CMT) preserved the state norm, whereas the RK45 method failed to do so.

Significance and Claims
The paper claims that the proposed CFCT method offers a significant computational speed-up while maintaining the geometric properties essential for physical simulations, particularly in quantum optimal control. In such applications, optimization algorithms require repeated, fast, and accurate integrations of quantum systems. By replacing expensive matrix exponentials and commutator evaluations with efficient Cayley transforms, the method addresses the trade-off between computational cost and structural fidelity. The authors position this work as a step toward developing high-order methods tailored for quantum control problems, noting that the current work establishes a fourth-order scheme and that extending this to higher orders is a future objective.