Trotter(m,n) Adaptive-Order Protocol

• The paper introduces an adaptive Trotterization protocol that dynamically compares lower- and higher-order product formulas to select the largest error-compliant simulation time step.
• The method drastically reduces circuit depth and gate count by enabling larger time steps while maintaining a preset error tolerance via real-time empirical feedback.
• It generalizes adaptive Trotterization to both time-independent and time-dependent Hamiltonians, ensuring bounded global error and leveraging conservation laws for stability.

The Trotter$(m,n)$ adaptive-order protocol is an approach for precision-guaranteed quantum simulation of Hamiltonian dynamics in which the Trotter time-evolution is adaptively controlled by dynamically measuring empirical error between two product formula orders, $m<n$, at each time step. By leveraging real-time estimations of the error—using higher-order Trotterization as an on-device gauge—this protocol enables the largest simulation time steps consistent with a preset error budget, thereby minimizing circuit depth and gate count for noisy intermediate-scale quantum (NISQ) devices and beyond. The protocol generalizes and unifies adaptive Trotterization methods, providing systematic error control for both time-independent and time-dependent Hamiltonians (Ikeda et al., 2023), and has been rigorously benchmarked for quantum spin chain simulation where it demonstrates an order-of-magnitude reduction in the number of required steps and significant depth savings over traditional methods (Zhao et al., 2022).

1. Suzuki–Trotter Product Formulas and Adaptive-Order Concept

Let $H = \sum_{j=1}^N H_j$ represent a target many-body Hamiltonian. The standard Suzuki–Trotter product formula of order $k$ for approximating evolution $U(t) = e^{-iHt}$ on a time window of length $\Delta t$ is $U_T^{(k)}(\Delta t) = S_k(\Delta t)$, a specifically ordered sequence of exponentials $\exp(-i a_\ell \Delta t H_{j_\ell})$ engineered to match the Taylor expansion of $e^{-iH\Delta t}$ up to $O(\Delta t^{k+1})$ (Bachmann et al., 2021). Examples include the first-order splitting $m<n$0, second-order symmetric (Strang) splitting $m<n$1, and higher recursive Suzuki formulas.

The core of the adaptive-order protocol is to:

• Use two product formulas of orders $m<n$2 (e.g., $m<n$3, $m<n$4) for the same time step $m<n$5
• Empirically estimate the simulation error on the quantum device by comparing outputs of $m<n$6 and $m<n$7
• Systematically select $m<n$8 to maximize efficiency while guaranteeing error per time step remains below a user-specified threshold $m<n$9 (Ikeda et al., 2023)

This strategy is fundamentally different from mathematical error-bound-based Trotterization, instead being data-driven and device-native.

2. On-Device Trotter Error Estimation

For a given quantum state $H = \sum_{j=1}^N H_j$0, the action of two product formulas yields: $H = \sum_{j=1}^N H_j$1 with local errors $H = \sum_{j=1}^N H_j$2 and $H = \sum_{j=1}^N H_j$3, respectively. The key observation is that the difference $H = \sum_{j=1}^N H_j$4 (in operator norm, fidelity, or observable expectation) approximates the true error of the lower-order formula up to $H = \sum_{j=1}^N H_j$5: $H = \sum_{j=1}^N H_j$6 Additional estimators include the state fidelity, $H = \sum_{j=1}^N H_j$7, and local observable error, $H = \sum_{j=1}^N H_j$8 (Ikeda et al., 2023). These quantities are efficiently accessible on quantum hardware, requiring no ancillary qubits.

3. Adaptive-Step Selection Algorithm

At each time step, the protocol performs a binary or continuous search to find the largest $H = \sum_{j=1}^N H_j$9 such that the estimated error $k$0 is below preset tolerance $k$1. The update rule in the regime $k$2 is

$k$3

This process is iterated until $k$4 is achieved, at which point the step is accepted and the simulation proceeds. Pseudocode for the protocol—both in fidelity-based and observable-based variants—is specified in (Ikeda et al., 2023).

In the alternative adaptive Trotterization approach (ADA-Trotter), the adaptive feedback loop is based on conservation-law constraints: post-step measurements on candidate states are used to ensure energy, variance, or other gauge quantities remain within user-specified drifts. Step sizes are modified by bisection or sequential back-off accordingly, with empirical evidence showing the required number of trial steps per time slice (typically $k$5–$k$6) does not increase with system size (Zhao et al., 2022).

4. Error Control, Stability, and Conservation Laws

Global error in this protocol grows only linearly in the number of time steps $k$7: $k$8 Thus, for a fixed total evolution time $k$9 and dynamically chosen $U(t) = e^{-iHt}$0, overall error can be tightly bounded, and accumulation of Trotter errors can be systematically suppressed (Ikeda et al., 2023).

For ADA-type approaches, a crucial feature is the enforcement of conservation laws. At each adaptive step $U(t) = e^{-iHt}$1, quantities such as mean energy and energy variance, or other conserved operators $U(t) = e^{-iHt}$2, are measured and required to stay within preset tolerances $U(t) = e^{-iHt}$3, $U(t) = e^{-iHt}$4. This enforcement achieves a uniform-in-time $U(t) = e^{-iHt}$5 error ceiling for local observables, in contrast to secular error growth ($U(t) = e^{-iHt}$6) in fixed-step protocols. The long-time bounds rely on the eigenstate-thermalization hypothesis (ETH), ensuring the error in diagonal-ensemble averages is $U(t) = e^{-iHt}$7, with $U(t) = e^{-iHt}$8 the system size, and both terms remain controlled independent of simulation time (Zhao et al., 2022).

5. Circuit Depth, Gate Count, and Efficiency

Adaptive-order Trotterization directly translates into substantial reductions of circuit depth. In numerical quantum spin-chain benchmarks, observed optimal step sizes $U(t) = e^{-iHt}$9 are up to $\Delta t$0 times larger than those conservative upper bounds from commutator-norm analysis would allow. For per-step errors $\Delta t$1 or $\Delta t$2 and safety factors $\Delta t$3, simulations with system size $\Delta t$4 maintain global errors within prescribed limits using orders of magnitude fewer Trotter steps (Ikeda et al., 2023).

A representative table based on (Ikeda et al., 2023), with $\Delta t$5, is:

Protocol	Typical $\Delta t$6	Error Control	Circuit Depth (relative)
Fixed-$\Delta t$7 (norm)	$\Delta t$8	Operator norm	$\Delta t$9
Trotter$U_T^{(k)}(\Delta t) = S_k(\Delta t)$0 adaptive	$U_T^{(k)}(\Delta t) = S_k(\Delta t)$1	On-device error	$U_T^{(k)}(\Delta t) = S_k(\Delta t)$2

For ADA-Trotter, comparative data on a nonintegrable Ising chain ($U_T^{(k)}(\Delta t) = S_k(\Delta t)$3) show $U_T^{(k)}(\Delta t) = S_k(\Delta t)$4 circuit-depth savings versus fixed-step protocols at identical local-observable error (Zhao et al., 2022). Overall measurement overhead is $U_T^{(k)}(\Delta t) = S_k(\Delta t)$5 per step (energy and variance), reducible to $U_T^{(k)}(\Delta t) = S_k(\Delta t)$6 with classical shadows, and the number of trial steps per slice is consistently small and independent of system size.

6. Extension to Time-Dependent Hamiltonians and Generalizations

The Trotter$U_T^{(k)}(\Delta t) = S_k(\Delta t)$7 protocol is directly extensible to time-dependent Hamiltonians $U_T^{(k)}(\Delta t) = S_k(\Delta t)$8. The same adaptive-step logic and error estimation apply by employing time-dependent Trotter–Suzuki formulas, such as the second-order midpoint rule and higher-order seven- or fifteen-exponential schemes. No reliance on energy conservation is necessary, and rigorous error and efficiency advantages are maintained for both static and driven systems (Ikeda et al., 2023).

The original Bachmann–Lange formalism provides a mathematically rigorous foundation for Trotter$U_T^{(k)}(\Delta t) = S_k(\Delta t)$9 product formulae in quantum lattice systems, establishing $\exp(-i a_\ell \Delta t H_{j_\ell})$0 pointwise norm bounds for local observables and explicit recursion formulas for arbitrary order $\exp(-i a_\ell \Delta t H_{j_\ell})$1 (Bachmann et al., 2021). While adaptive choice of $\exp(-i a_\ell \Delta t H_{j_\ell})$2 on a per-slice basis or fully variable order across $\exp(-i a_\ell \Delta t H_{j_\ell})$3 steps is not formalized in the convergence proof, the conceptual structure supports such generalizations for practical simulation cost optimization.

7. Benchmarks, Practical Relevance, and Cost-Benefit Analysis

In detailed quantum spin-chain benchmarks, adaptive-order protocols ensure simulation fidelity within target tolerances while allowing for significantly larger steps and lower total gate counts. For example, with fixed error per step $\exp(-i a_\ell \Delta t H_{j_\ell})$4 and safety $\exp(-i a_\ell \Delta t H_{j_\ell})$5, average $\exp(-i a_\ell \Delta t H_{j_\ell})$6 is $\exp(-i a_\ell \Delta t H_{j_\ell})$7, whereas naive rigorous upper bounds would limit one to $\exp(-i a_\ell \Delta t H_{j_\ell})$8. The total number of required quantum measurements (for observable-based error estimation) remains within the reach of NISQ-era devices, e.g., $\exp(-i a_\ell \Delta t H_{j_\ell})$9–$e^{-iH\Delta t}$0 for short total times and observables (Ikeda et al., 2023).

These results establish Trotter$e^{-iH\Delta t}$1 as a method that enables precision guarantee with bounded global error, minimizes simulation cost, and provides a systematic advantage over fixed-step Trotterization—crucial for practical quantum simulation efforts, especially under resource constraints intrinsic to near-term quantum hardware (Zhao et al., 2022, Ikeda et al., 2023, Bachmann et al., 2021).