Product Formulas: An Overview
The family of splitting methods that approximate e^(-iHt) as ordered products of simple exponentials, from Trotter to randomized variants.
What unites product formulas
Product formulas approximate the propagator of H = sum_j H_j by chaining exponentials of individual terms. They share a common signature: no ancilla qubits, a circuit made of native few-body rotations, and error that scales polynomially in the target precision.
The main variants
- Lie-Trotter (first order): ordered product of e^(-i H_j dt).
- Strang / symmetric (second order): palindromic ordering cancels the leading error.
- Suzuki recursion (order 2k): nests symmetric formulas for higher accuracy.
- qDRIFT: samples terms at random with probability proportional to their weight.
- Random-permutation Trotter: shuffles term order each step to average out bias.
qDRIFT in brief
Instead of applying every term each step, qDRIFT (Campbell 2019) picks a term H_j at random with probability proportional to its coefficient norm, then applies a rotation of fixed angle. The gate count becomes independent of the number of terms L and scales instead with (sum of coefficients)^2 times t^2 / epsilon. This helps Hamiltonians with many small terms, such as molecular electronic structure.
When to choose which
If the Hamiltonian is local with a moderate number of terms, deterministic higher-order Trotter is usually best. If there are many terms of widely varying size, qDRIFT or importance-sampled splitting can win. If very high precision is required and ancillas are available, post-Trotter methods (LCU, qubitization) scale better in epsilon.
The enduring appeal
Product formulas dominate near-term experiments because they demand no extra qubits and map interaction terms onto physical gates directly. Their commutator-dependent error is often far below worst-case, so a second-order formula frequently suffices for dynamics studies of spin chains, Hubbard models, and small molecules. They form the baseline against which every fancier algorithm is measured.