Theory and Shared Conventions

This document describes concepts shared across Quantum Lattice Models. Concise Hamiltonians, variables, structures, package examples, and validation notes for individual models live in the model reference.

Represented systems

The package uses several distinct matrix representations:

• Spin-$\tfrac12$ models use the computational basis $|s_0s_1\ldots s_{N-1}\rangle$ and have dimension $2^N$.
• Single-particle tight-binding models use one basis state $|i\rangle$ per site or orbital; their dimension is the number of sites or orbitals.
• Bose-Hubbard models use truncated local occupations $|0\rangle,\ldots,|n_{\max}\rangle$ and dimension $(n_{\max}+1)^N$.
• Spinful Fermi-Hubbard models use binary occupation of $2N$ ordered spin-orbitals and dimension $2^{2N}$.
• Bogoliubov-de Gennes models use a doubled Nambu single-particle basis. They are not many-body occupation-basis Hamiltonians.

These representations are not interchangeable. Model reference pages state the basis and dimension explicitly.

Operators and normalization

Spin builders use Pauli matrices $X$, $Y$, and $Z$ directly. Consequently, $X_iX_j+Y_iY_j+Z_iZ_j$ is four times the conventional $\mathbf S_i\cdot\mathbf S_j$ expression for spin operators $\mathbf S=\mathbf P/2$. Couplings in the package multiply Pauli products as written in each model page.

Site index zero is the leftmost tensor factor in Kronecker-product spin operators. Fermi-Hubbard orbitals are ordered

$$ (0\uparrow,0\downarrow,1\uparrow,1\downarrow,\ldots). $$

Tight-binding sign convention

A generic Hermitian hopping term is represented as

$$ H_{ij}=v,\qquad H_{ji}=v^*. $$

Named tight-binding builders conventionally use $v=-t$ for a real hopping parameter $t$. Custom three-item bond records use the supplied matrix element directly; two-item records use -hopping.

Complex hoppings therefore encode phases without an additional implicit sign. The relevant gauge convention is documented on each topological-model page.

Boundary conditions

Open boundaries include only bonds that remain inside the finite geometry. Periodic boundaries reconnect opposite ends or edges. Chain models use periodic; two-dimensional finite lattices use independent periodic_x and periodic_y flags.

Periodic finite real-space matrices are distinct from Bloch Hamiltonians $H(\mathbf k)$. Bloch and reciprocal-space support remains roadmap work.

Dense and sparse matrices

Dense matrices are NumPy arrays or metadata-bearing NumPy subclasses. Selected lattice and Hubbard models also provide SciPy CSR builders. Matching dense and sparse lattice builders share construction paths and are tested for numerical equivalence.

Sparse storage reduces memory when the matrix contains few nonzero entries, but it does not change exponential Hilbert-space growth in many-body models. Use estimate_model_dimension, estimate_dense_memory, and diagnose_matrix before constructing larger systems.

For XXZ and Heisenberg chains with $J_x=J_y$, fixed-magnetization builders restrict the basis to a total Pauli-$Z$ eigenvalue

$$ M=\sum_i Z_i=N-2n_1. $$

The reduced dimension is $\binom{N}{(N-M)/2}$. Reduced basis states are stored as their integer indices in the full computational basis, allowing explicit projection, embedding, and full-space block validation.

Exact diagonalization

Exact diagonalization solves

$$ H|\psi_n\rangle=E_n|\psi_n\rangle. $$

Hermitian matrices use Hermitian eigensolvers. Full dense diagonalization gives all eigenpairs; sparse routines can request only low-energy eigenvalues or the ground state. Iterative sparse results should not be interpreted as complete spectra.

The package is intended for transparent small-system calculations and reference workflows, not large-scale many-body simulation.

Observables and states

State vectors are expected to use the same basis ordering as their Hamiltonian. expectation computes $\langle\psi|O|\psi\rangle; spin helpers provide site-resolved and total $Z$ magnetization, same-axis correlation matrices, connected correlations, and static structure factors. These routines accept either the full computational basis or a FixedMagnetizationBasis`.

Reduced density matrices group amplitudes by subsystem and environment bit patterns. Sector states are handled directly from their full-basis integer labels rather than first allocating a vector of length $2^N$. Bipartite entanglement uses the von Neumann entropy

$$ S_A=-\operatorname{Tr}(\rho_A\log \rho_A). $$

The default logarithm base is two, so Bell-state entropy is one bit. Localization helpers provide inverse participation ratios or SSH edge weights.

Probability plots normalize $|\psi_i|^2$ by default. Phase-resolved lattice plots use the complex argument of each amplitude.

Symmetries and diagnostics

Hermiticity is checked through $H=H^\dagger$. BdG particle-hole diagnostics currently test spectral pairing $E\leftrightarrow-E$ rather than constructing the full antiunitary symmetry operator. Numerical tolerances must be chosen for the problem scale.

Further parity, translation-sector, commutator, degeneracy, and topological diagnostics are listed in ROADMAP.md.

Portable model specifications

ModelSpec records a registered model family, parameters, basis, requested dense or sparse representation, and optional LatticeSpec. JSON encodes complex values explicitly. Loading a specification validates it against the registry before reconstruction.

HamiltonianResult preserves matrix construction metadata, including reduced spin-sector basis information where applicable.

Quantum-algorithm testbeds

Small lattice Hamiltonians provide classically verifiable inputs for VQE, QPE, Hamiltonian simulation, QSVT, and quantum-walk prototypes. The package supplies reference matrices and conventions; it does not claim quantum advantage.

Model reference

Browse the model index for per-model equations, variables, structures, examples, limitations, and validation status.