Three degrees of freedom and cavity QED (Pauli–Fierz)

solve_tise works on one-, two- and three-dimensional meshes, and each mesh coordinate may belong to a different particle with its own mass. This makes it possible to solve, without any approximation beyond the discretization, problems with three one-dimensional degrees of freedom, e.g. a nucleus, an electron and a cavity photon mode. A worked example is the tutorial Nucleus, electron and photon.

Masses

The kinetic energy is $-\sum_k \frac{1}{2M_k}\partial_k^2$, with masses in units of the electron mass:

masses = (1836.15, 1.0, 1.0)     # proton (R), electron (z), photon mode (q)
sol = solve_tise(model, V; masses = masses, nev = 10)

In the input-file driver, the same is set with the optional line masses = 1836.15 1.0 1.0 (see Input file reference).

For a diatomic molecule, the nuclear coordinate $R$ uses the reduced mass $\mu = M_AM_B/(M_A+M_B)$.

The Pauli–Fierz Hamiltonian

In the length gauge and the dipole approximation, matter with dipole moment $d(\mathbf r)$ coupled to one cavity mode of frequency $\omega$ is described by (atomic units)

\[\hat H = \hat H_{\rm matter}(\mathbf r) - \frac{1}{2}\frac{\partial^2}{\partial q^2} + \frac{1}{2}\omega^2 q^2 + g\sqrt{2\omega}\; q\, d(\mathbf r) + \frac{g^2}{\omega}\, d(\mathbf r)^2 .\]

  • $q$ is the displacement coordinate of the photon mode, with unit mass. Its harmonic levels $\omega(n+\tfrac12)$ are the photon-number states.
  • $g$ is the light–matter coupling strength (it grows as the cavity volume decreases).
  • The last term, the dipole self-energy, keeps the Hamiltonian bounded from below. The three cavity terms combine into $\frac12\omega^2\big(q + g\sqrt{2\omega}\,d/\omega^2\big)^2$.

pauli_fierz_potential builds the potential of this Hamiltonian from the matter potential and the dipole function. The last mesh coordinate is $q$:

# nucleus R and electron z, charge Z = 1: dipole d = R - z
V_matter(r) = my_potential(r[1], r[2])          # r = (R, z)
dipole(r) = r[1] - r[2]
V = pauli_fierz_potential(V_matter, dipole; frequency = 0.1, coupling = 0.01)

model = CartesianDiscreteModel((0.5, 8.0, -10.0, 10.0, -20.0, 20.0), (24, 32, 32))
sol = solve_tise(model, V; masses = (918.0, 1.0, 1.0), nev = 8, sigma = -1.0)

The eigenstates are polaritons: mixed light–matter states. Since $g\sqrt{2\omega}\,q = g(\hat a+\hat a^\dagger)$, when a matter excitation of energy $\omega_{01}$ is resonant with the cavity ($\omega=\omega_{01}$) the two uncoupled states "excited matter, no photon" and "ground-state matter, one photon" split into a lower and an upper polariton separated by the Rabi splitting $\Omega_R \approx 2g\,|d_{01}|$, with $d_{01}$ the transition dipole (for weak coupling).

Validation

The test suite and the tutorial check the 3D solver against a quadratic model with a harmonic nucleus and electron and a linear dipole. Its potential is a quadratic form, so the exact levels are $\sum_k\Omega_k(n_k+\tfrac12)$, with $\Omega_k^2$ the eigenvalues of the mass-weighted Hessian. With $16^3$ quadratic elements, the lowest levels agree to about $10^{-3}$ and are upper bounds.

Cost and practical advice

Three-dimensional meshes grow quickly: a mesh of $n^3$ cells with quadratic elements has about $8n^3$ unknowns, and the sparse factorization used by the eigensolver costs much more than in 1D or 2D. For reference, $16^3$ cells ($3\times10^4$ unknowns) take about 16 s on a laptop; memory and time grow quickly with $n$. Recommendations:

  • Choose the box for each coordinate from the extent of the states of interest; the photon coordinate needs about $\pm\sqrt{2n_{\max}+1}/\sqrt{\omega}$.
  • Heavy coordinates (nuclei) have much narrower wave functions than light ones and need finer meshes in their direction. Adjust $n_x, n_y, n_z$ independently.
  • Check convergence with two meshes, as described in Accuracy and best practices.
  • For realistic molecules, a common strategy is to treat the nucleus adiabatically (2D electron–photon problems along a grid of $R$, a parameter sweep), and to solve the full 3D problem only to check the adiabatic approximation.