Sub-Hamiltonian Techniques in Science

• Sub-Hamiltonian techniques are a set of methodologies that modify Hamiltonians to yield effective low-dimensional models while preserving crucial algebraic and geometric properties.
• They are applied in quantum many-body downfolding, symplectic model reduction, and nonconvex optimization to enhance computational efficiency and bypass local minima.
• These methods also support optimal control and sub-Riemannian geometry by constructing structured reduced models that capture essential dynamics under constraints.

A sub-Hamiltonian technique refers to a family of methodologies in mathematical physics, chemistry, control theory, and computational optimization that systematically reduce, restructure, or alternate Hamiltonians to obtain effective low-dimensional models, escape poor local minima, or generate geodesics under nonholonomic constraints—all while preserving crucial algebraic or geometric properties. These approaches include downfolding in many-body quantum systems, symplectic model reduction for large-scale Hamiltonian dynamics, metaheuristics for global optimization, and sub-Riemannian geometric constructions in control problems. The sub-Hamiltonian formalism is deployed across various disciplines, often under the ad hoc terminologies of downfolded/effective Hamiltonians, alternate Hamiltonians, or sub-Riemannian Hamiltonians.

1. Downfolded Sub-Hamiltonians in Quantum Many-Body Theory

In the context of electronic structure theory, sub-Hamiltonian techniques are employed to isolate a complete active space (CAS) where strong electron correlation dominates, while the external dynamical correlation is encapsulated into an effective, lower-dimensional Hamiltonian—often called a "downfolded" or "sub-Hamiltonian." The subsystem embedding subalgebras coupled-cluster (SES-CC) formalism segregates the full cluster operator $T$ into internal ($T_\mathrm{int}$) and external ($T_\mathrm{ext}$) excitations—the former acting within the CAS and the latter outside. The exact coupled cluster energy is then recovered from the eigenproblem

$$H_\mathrm{eff}\, |\Psi_{\mathrm{CAS}}\rangle = E\, |\Psi_{\mathrm{CAS}}\rangle, \quad |\Psi_{\mathrm{CAS}}\rangle= e^{T_\mathrm{int}}|\Phi\rangle,$$

where $H_\mathrm{eff}$ is obtained by similarity transforming the full Hamiltonian with respect to $T_\mathrm{ext}$ and projecting onto the CAS. This yields a generally non-Hermitian effective Hamiltonian.

The double unitary coupled-cluster (DUCC) formalism extends this by defining anti-Hermitian generators, $$\sigma_\mathrm{int} = T_\mathrm{int} - T_\mathrm{int}^\dagger$$ and $$\sigma_\mathrm{ext} = T_\mathrm{ext} - T_\mathrm{ext}^\dagger$$, constructing a Hermitian downfolded Hamiltonian:

$$H_\mathrm{eff}^{\mathrm{DUCC}} = (P+Q_\mathrm{int})\, e^{-\sigma_\mathrm{ext}}He^{\sigma_\mathrm{ext}}\, (P+Q_\mathrm{int}),$$

where $P$ and $T_\mathrm{int}$0 are projectors onto the reference and active excitations, respectively. The resulting CAS eigenproblem facilitates rigorous separation of internal and external correlation amplitudes. Perturbative expansions are derived systematically, truncating to low-body terms when computationally necessary. The Hermitian character enables compatibility with configuration interaction and quantum eigensolvers, yielding substantial resource reductions when $T_\mathrm{int}$1 (Bauman et al., 2019).

2. Symplectic Model Reduction via Sub-Hamiltonians

For large-scale Hamiltonian systems arising in physics or engineering, sub-Hamiltonian techniques underpin Proper Symplectic Decomposition (PSD) and symplectic model reduction. Letting $T_\mathrm{int}$2, the dynamics $T_\mathrm{int}$3 preserves the symplectic structure and energy. The goal is to find a symplectic subspace $T_\mathrm{int}$4, $T_\mathrm{int}$5 satisfying $T_\mathrm{int}$6, onto which the dynamics can be projected:

$T_\mathrm{int}$7

where $T_\mathrm{int}$8 is the symplectic inverse. Several algorithmic constructions for such $T_\mathrm{int}$9 include the cotangent lift, complex SVD, and constrained nonlinear programming. This projection guarantees conservation of symplecticity, energy, and stability—properties often lost in standard POD-Galerkin model reduction. Symplectic DEIM further reduces computational complexity for nonlinear Hamiltonians (Peng et al., 2014).

An extension, based on the covariant snapshot matrix and intertwining eigenproblems, constructs a sub-Hamiltonian directly from simulation data, even without explicit partitioning into potential and kinetic energy. By eigen-decomposition of the weighted Gramian of the snapshot matrix, one extracts canonical symplectic modes and constructs the reduced ("sub-") Hamiltonian whose flow inherits the structure of the true system and converges as the time discretization refines (Shirafkan et al., 2021).

3. Alternate Hamiltonians in Nonconvex Optimization

The alternate sub-Hamiltonian (or Hamiltonian alternation) metaheuristic targets global minimization in nonconvex landscapes. Given a cost function $T_\mathrm{ext}$0, an alternate Hamiltonian is constructed by composing $T_\mathrm{ext}$1 with a nonlinear map $T_\mathrm{ext}$2 to invert the energy landscape above a tunable cutoff $T_\mathrm{ext}$3:

$T_\mathrm{ext}$4

This procedure ensures that all local minima of $T_\mathrm{ext}$5 above $T_\mathrm{ext}$6 become local maxima of $T_\mathrm{ext}$7, while the global minimum (below $T_\mathrm{ext}$8) remains a shared minimum. Algorithmically, one alternates between local minimization of $T_\mathrm{ext}$9 and $$H_\mathrm{eff}\, |\Psi_{\mathrm{CAS}}\rangle = E\, |\Psi_{\mathrm{CAS}}\rangle, \quad |\Psi_{\mathrm{CAS}}\rangle= e^{T_\mathrm{int}}|\Phi\rangle,$$0, progressively lowering $$H_\mathrm{eff}\, |\Psi_{\mathrm{CAS}}\rangle = E\, |\Psi_{\mathrm{CAS}}\rangle, \quad |\Psi_{\mathrm{CAS}}\rangle= e^{T_\mathrm{int}}|\Phi\rangle,$$1 as better solutions are found. This effectively destabilizes spurious minima and guides the optimizer toward the global solution. Empirical results show significant improvements in problems like the Sherrington-Kirkpatrick spin glass, especially as the problem size increases (Apte et al., 2022).

4. Sub-Hamiltonian Formalism in Geometric Control and Sub-Riemannian Geometry

In geometric control and sub-Riemannian geometry, the sub-Hamiltonian framework arises from Pontryagin’s Maximum Principle (PMP), leading to Hamiltonian systems on constrained state manifolds. For configurations $$H_\mathrm{eff}\, |\Psi_{\mathrm{CAS}}\rangle = E\, |\Psi_{\mathrm{CAS}}\rangle, \quad |\Psi_{\mathrm{CAS}}\rangle= e^{T_\mathrm{int}}|\Phi\rangle,$$2 in $$H_\mathrm{eff}\, |\Psi_{\mathrm{CAS}}\rangle = E\, |\Psi_{\mathrm{CAS}}\rangle, \quad |\Psi_{\mathrm{CAS}}\rangle= e^{T_\mathrm{int}}|\Phi\rangle,$$3, typical of models like the visual cortex V1's border completion or the kinematics of a bicycle's rear wheel, the Hamiltonian is

$$H_\mathrm{eff}\, |\Psi_{\mathrm{CAS}}\rangle = E\, |\Psi_{\mathrm{CAS}}\rangle, \quad |\Psi_{\mathrm{CAS}}\rangle= e^{T_\mathrm{int}}|\Phi\rangle,$$4

Hamilton’s equations derived from this structure capture geodesics respecting nonholonomic constraints (i.e., motion restricted to horizontal distributions), with direct analogy to both neural and mechanical systems. These sub-Hamiltonians are used to parametrize optimal trajectories, describe association fields, and systematically encode control–cost relationships (Fioresi et al., 2023).

5. Algorithmic and Computational Workflows

Quantum Many-Body Downfolding

• Classical preprocessing: Evaluate external cluster amplitudes via CC or UCC (e.g., CCSD), assemble anti-Hermitian generators, expand and truncate the downfolded Hamiltonian using perturbative formulas and tensor contractions.
• Quantum/classical eigensolvers: Diagonalize the sub-Hamiltonian using FCI, DMRG, or quantum algorithms (QPE, VQE) after mapping to appropriate operator forms (e.g., Jordan–Wigner, Bravyi–Kitaev).
• Resource scaling: Downfolding leads to dramatically lower term counts and qubit requirements—$$H_\mathrm{eff}\, |\Psi_{\mathrm{CAS}}\rangle = E\, |\Psi_{\mathrm{CAS}}\rangle, \quad |\Psi_{\mathrm{CAS}}\rangle= e^{T_\mathrm{int}}|\Phi\rangle,$$5 for 2-body truncations, versus $$H_\mathrm{eff}\, |\Psi_{\mathrm{CAS}}\rangle = E\, |\Psi_{\mathrm{CAS}}\rangle, \quad |\Psi_{\mathrm{CAS}}\rangle= e^{T_\mathrm{int}}|\Phi\rangle,$$6 for the full system (Bauman et al., 2019).

Symplectic Model Reduction

• Offline: Assemble covariant snapshot matrices, compute SVDs or solve eigenproblems under symplectic constraints on $$H_\mathrm{eff}\, |\Psi_{\mathrm{CAS}}\rangle = E\, |\Psi_{\mathrm{CAS}}\rangle, \quad |\Psi_{\mathrm{CAS}}\rangle= e^{T_\mathrm{int}}|\Phi\rangle,$$7.
• Online: Project initial data, integrate the subspace-reduced Hamiltonian, evaluate nonlinear terms via SDEIM, guarantee preservation of energy and stability (Peng et al., 2014, Shirafkan et al., 2021).

Optimization via Hamiltonian Alternation

• Execution: For each alternation round, minimize $$H_\mathrm{eff}\, |\Psi_{\mathrm{CAS}}\rangle = E\, |\Psi_{\mathrm{CAS}}\rangle, \quad |\Psi_{\mathrm{CAS}}\rangle= e^{T_\mathrm{int}}|\Phi\rangle,$$8 and $$H_\mathrm{eff}\, |\Psi_{\mathrm{CAS}}\rangle = E\, |\Psi_{\mathrm{CAS}}\rangle, \quad |\Psi_{\mathrm{CAS}}\rangle= e^{T_\mathrm{int}}|\Phi\rangle,$$9 sequentially; update the cutoff $H_\mathrm{eff}$0 as new minima are found; break upon convergence to a shared (global) minimum.
• Theoretical guarantee: Hessian analysis shows that all minima of $H_\mathrm{eff}$1 above $H_\mathrm{eff}$2 are destabilized in $H_\mathrm{eff}$3, ensuring escape from local minima.
• Empirical metrics: Success rates and best-found energies are systematically improved over multi-start alone in several benchmarks (Apte et al., 2022).

6. Applicability, Limitations, and Comparative Features

The sub-Hamiltonian technique's applicability spans:

• Many-body electronic structure, where it enables resource-efficient, low-energy models compatible with quantum-classical hybrid architectures.
• Large-scale dynamical systems, offering model order reduction that exactly conserves geometric invariants—critical for long-time integration and physical fidelity.
• Nonconvex numerical optimization, as a metaheuristic layer compatible with any local search algorithm, especially valuable for high-dimensional and glassy energy landscapes.
• Control theory and geometric mechanics, through the explicit construction of optimal trajectories under nonholonomic constraints.

Limitations include the approximation errors stemming from truncation of external amplitudes or reduced modal representations, the challenge of symplectic optimization over matrix manifolds, and the lack of non-asymptotic guarantees when the true ground state is not bracketed in optimization contexts. Nevertheless, the technique systematically preserves or exploits key algebraic structures otherwise lost in generic model reduction or search heuristics.

7. Selected Summary Table

Context	Core Sub-Hamiltonian Construct	Main Advantage
Quantum many-body (SES-CC/DUCC)	Downfolded Hamiltonian on CAS: $H_\mathrm{eff}$4	Hermitian, resource-efficient, enables FCI/VQE
Symplectic model reduction (PSD)	Projected Hamiltonian $H_\mathrm{eff}$5	Preserves symplecticity/energy/stability
Nonconvex optimization (alternation)	Alternate Hamiltonian $H_\mathrm{eff}$6	Escapes spurious minima
Sub-Riemannian geometry (control, V1)	PMP-derived sub-Hamiltonian $H_\mathrm{eff}$7	Encodes constrained geodesics

The sub-Hamiltonian technique thus unites a broad spectrum of methodologies across quantum, classical, optimization, and geometric domains through the systematic construction, analysis, and exploitation of reduced or modified Hamiltonian structures that preserve essential algebraic or variational properties (Bauman et al., 2019, Peng et al., 2014, Shirafkan et al., 2021, Apte et al., 2022, Fioresi et al., 2023).