Introduction

Sometimes, algebraic manipulations with quantum-mechanical operators can be an important part of a numerical algorithm in their own right. It is, however, much more common to use the resulting algebraic expressions as operators acting in a space of quantum states (Hilbert space). This scenario is of particular importance when implementing algorithms for partial or full diagonalization of quantum Hamiltonians – so called exact diagonalization (ED) methods.

libcommute gives user a handful of tools to minimize overhead when writing ED codes. The following few steps describe a typical workflow of a libcommute-based ED code.

• Define Hamiltonian of a quantum system in a form of a polynomial expression.

• Make a finite-dimensional Hilbert space. This can be done either automatically for a given Hamiltonian, or – in more tricky cases – by explicitly constructing a product of elementary spaces.

• Use the polynomial expression and the Hilbert space to make a linear operator object representing the Hamiltonian. This object can be either a simple linear operator or a parametric linear operator.

• Act with the constructed linear operator on state vectors to obtain a matrix representation of the Hamiltonian. Instead of computing all matrix elements at once and resorting to a dense diagonalization method such as QR algorithm, one could opt to power iteration or Lanczos algorithm for sparse eigenproblems. In the latter case, one must act with the operator on a state at each step of the iterative diagonalization scheme.

There is also an option to find invariant subspaces of the Hamiltonian to reduce the numerical cost of the diagonalization procedure. The mapped_basis_view state vector adapter can be acted upon by linear operators defined in the full Hilbert space, while saving memory by storing only the part of a state vector belonging to a certain invariant subspace.