8. Many-body solver

In the previous pages, we introduced non-linear Poisson and effective Schrödinger solvers for electrons and holes. In QTCAD, these solvers may account for arbitrary 1D, 2D, or 3D geometries. However, even though Coulomb interactions within classical electron or hole gases or between quantum-mechanically confined carriers and classical gases may be accounted for within these approaches, interactions between confined electrons or holes are not modeled at a quantum-mechanical level in these solvers. Even though quantum-confined carriers do modify the confinement potential within the self-consistent Schrödinger–Poisson solver, this approach does not exactly treat exchange and correlation effects arising from a rigorous quantum-mechanical treatment of the Coulomb interaction.

To remedy this, QTCAD features a many-body solver based on the exact-diagonalization approach [GuccluSGH02]. Within a truncated basis set, the exact-diagonalization approach includes exchange and correlation effects exactly, in contrast with approaches like the Hartree–Fock method [BSA00], in which a mean-field approximation is implied. Despite exponential scaling of the required computational resources with the number of electrons included in the calculation, QTCAD is designed for few-electron systems that are relevant to qubit implementations and for which meaningful results may still be obtained within minutes on a laptop or a modest workstation.

Many-body Hamiltonian

In second quantization and within the effective-mass approach presented in Effective Schrödinger equation for electrons, the Hamiltonian for \(N\) conduction electrons that are coupled through the Coulomb interaction is [BF04]

(8.1)\[\begin{split}H = &-\frac{\hbar^2}2\sum_\sigma \int d\mathbf r \hat \Psi^\dagger_\sigma (\mathbf r) \nabla\cdot\left[\mathbf M_e^{-1}\cdot\nabla\hat\Psi_\sigma(\mathbf r)\right] + \int d\mathbf r V_\mathrm{conf}(\mathbf r) \hat n(\mathbf r)\\ &\qquad+ \frac12\int d\mathbf r_1 d\mathbf r_2 \sum_{\sigma_1\sigma_2} V_\mathrm{Coul.}(\mathbf r_1, \mathbf r_2) \Psi^\dagger_{\sigma_1}(\mathbf r_1)\Psi^\dagger_{\sigma_2}(\mathbf r_2) \Psi_{\sigma_2}(\mathbf r_2)\Psi_{\sigma_1}(\mathbf r_1),\end{split}\]

where the sum over \(\sigma\) runs over degenerate states (e.g. \(\uparrow\) and \(\downarrow\) spin states, or valley states), and where we have introduced the Coulomb interaction potential

(8.2)\[V_\mathrm{Coul.}(\mathbf r_1, \mathbf r_2) = \frac{e^2}{4\pi\varepsilon(\mathbf r_2)|\mathbf r_1-\mathbf r_2|},\]

and the electron particle-density operator in position representation

(8.3)\[\hat n(\mathbf r) = \sum_\sigma \hat\Psi^\dagger_\sigma(\mathbf r) \hat\Psi_\sigma(\mathbf r).\]

In the above equations, we have also introduced the quantum field operator \(\hat \Psi_\sigma(\mathbf r)\), which annihilates and electron of spin \(\sigma\) at position \(\mathbf r\).

Introducing an othornormal basis of single-electron states \(\left\{F_i(\mathbf r)\right\}\), we recall the relation between the electron quantum field operator and the fermionic operator \(\hat c_{i\sigma}\) which annihilates an electron of spin \(\sigma\) in the \(i\)-th basis state:

(8.4)\[\hat \Psi_\sigma(\mathbf r) = \sum_i F_i(\mathbf r)\hat c_{i\sigma}.\]

In terms of fermionic operators, the many-body Hamiltonian written in (8.1) is

(8.5)\[H = \sum_{i\sigma} \epsilon_{i\sigma}\hat c^\dagger_{i\sigma}c_{i\sigma} + \frac 12\sum\limits_{ijkl,\sigma \sigma '} {{V_{ijlk}}} \hat c_{i\sigma }^\dagger \hat c_{j\sigma '}^\dagger {\hat c_{k\sigma '}}{\hat c_{l\sigma }},\]

where \(\epsilon_{i\sigma}\) is the energy of the single-particle eigenstate with orbital index \(i\) and spin index \(\sigma\), and \(V_{ijkl}\) represents the Coulomb integrals of the basis set:

(8.6)\[V_{ijkl} = \int d\mathbf r_1 d\mathbf r_2\, F_i^\ast(\mathbf r_1)F_j^\ast(\mathbf r_2) \frac{e^2}{4\pi\varepsilon(\mathbf r_2)|\mathbf r_1-\mathbf r_2|} F_k(\mathbf r_1)F_l(\mathbf r_2).\]

Note

The spatial dependence of the dielectric permittivity \(\varepsilon\) implemented here is only accurate in situations where \(\varepsilon\) varies negligibly over the region in which the electron wave functions are significant. For example, for a quantum dot defined by a semiconductor surrounded by an oxide, the wave function should not penetrate significantly into this oxide.

Exact diagonalization

The Hamiltonian given in (8.5) is written on the basis spanned by single-particle envelope functions (orbitals) \(\{F_i(\mathbf r)\},\;i\;\in\{1,...,n_\mathrm{states}\}\). Of course, an exact description of a many-body problem may require a very large or even infinite number of such basis states. Here, we assume that truncating the basis to a finite \(n_\mathrm{states}\) does not significantly influence the solution of the many-body problem. This is often the case for few-electron systems; it can be easily verified by checking that the solution of a given problem converges as \(n_\mathrm{states}\) is increased.

For each orbital, we assume a degeneracy factor \(n_\mathrm{degen}\)—for example, for spin degeneracy, \(n_\mathrm{degen}=2\). Because of fermionic statistics, each orbital can be occupied by a maximum of \(n_\mathrm{degen}\) electrons, for a total single-particle basis set of dimension \(n_\mathrm{states}\times n_\mathrm{degen}\).

Since the Hamiltonian given in (8.5) preserves the number \(N\) of conduction electrons, we may break it down into \(N_\mathrm{sub}\) many-body subspaces containing \(N\;\in\;\{0,1,...,N_\mathrm{sub}-1\}\) electrons.

For each \(N\)-electron subspace, we define a many-body basis set in the particle-number representation. In this representation, a many-body basis state is a binary string of length \(n_\mathrm{states}\times n_\mathrm{degen}\) whose \(j\)-th digit is 0 (1) for empty (occupied) single-particle basis states, under the constraint that the total number of occupied single-particle basis states sums to \(N\). For example, for \(n_\mathrm{states}=n_\mathrm{degen}=2\), the \(N=2\) many-body basis set is

(8.7)\[\left\{|1100\rangle, |1010\rangle, |1001\rangle, |0110\rangle, |0101\rangle, |0011\rangle\right\},\]

where

(8.8)\[|1100\rangle \equiv \hat c^\dagger_1\hat c^\dagger_2|0\rangle,\qquad |1010\rangle \equiv \hat c^\dagger_1\hat c^\dagger_3|0\rangle,\qquad...,\]

with \(\hat c_j\) being the fermionic operator that annihilates an electron in the \(j\)-th single-particle basis state and \(|0\rangle\) being the vacuum state, in which no conduction electron is present.

Importantly, since each single-particle basis state can host a maximum of one electron, the total number of many-body subspaces for a truncated basis set is \(n_\mathrm{states}\times n_\mathrm{degen}+1\). Equivalently, the maximum number of electrons that can be modeled is \(N_\mathrm{max}=n_\mathrm{states}\times n_\mathrm{degen}\).

When provided with \(n_\mathrm{states}\), single-body envelope functions \(\left\{F_i(\mathbf r)\right\}\), and a degeneracy factor \(n_\mathrm{degen}\), QTCAD’s many-body solver decomposes the Hamiltonian of Eq. (8.5) into \(N_\mathrm{max}+1\) many-body subspaces in which the Hamiltonian is first calculated by evaluating Coulomb integrals and then diagonalized. As a result, many-body eigensolutions are found within each subspace, providing a numerically tractable solution to the many-body problem when \(N_\mathrm{max}\) is sufficiently small (typically, less than or on the order of 10 electrons for laptops and modest workstations).

Inputs and outputs of the many-body solver

Like all other Solver classes in QTCAD, the many-body Solver is instantiated with a mandatory Device argument. If the input Device object already contains single-particle envelope functions in its eigenfunctions attribute, the first \(n_\mathrm{states}\) of these envelope functions are used to form the single-particle basis set. If no eigenfunction is stored in the Device object, the single-electron Schrödinger equation is solved over the device for the first \(n_\mathrm{states}\) envelope functions, which are then used to form the basis set.

A typical instantiation of the many-body Solver thus looks like this:

solver_params = ManyBodySolverParams()
solver_params.num_states = 3
solver_params.n_degen = 2
slv = ManyBodySolver(dvc, solver_params=solver_params)

As for other solvers, the many-body Solver object is instantiated using the Device object and a SolverParams object containing its parameters. The many-body Solver is then run using the solve method. This produces three main results that are stored in the input Device object:

many_body_subspaces: information about the many-body subspaces and the corresponding many-body eigensolution.
chem_potentials: the chemical potentials of the device structure.
coulomb_peak_pos: the Coulomb peaks of the device.

In addition, once the solution of the many-body problem is found, it becomes possible to calculate the electron density for each many-body eigenstate using the get_many_body_density method of the Device object.

These outputs are further described below.

Many-body subspaces

The many_body_subspaces attribute is a list of Subspace objects for \(N = 0, 1, ..., N_\mathrm{max}\).

The Subspace objects are described in the API Reference, in the device.quantum module.

The most useful physical properties encapsulated in the Subspace objects are:

The many-body basis set, which can be accessed with the get_bas_set method.
The many-body Hamiltonian, which can be accessed with the get_many_body_ham method.
The many-body eigenvalues, which are stored in the eigval attribute of the Subspace object.
The many-body eigenstates, which are stored in the eigvec attribute of the Subspace object.

For each \(N\)-electron subspace, there are as many eigenvalues and eigenvectors as there are many-body basis states.

The many-body eigenstates for a given subspace can also be output using the get_eig_state method of the Subspace class.

The output of this method is an MbState object (see API Reference). The most important attributes of this object are the energy of the many-body state (energy) and the coefficients that define the expansion of the many-body eigenstate over the many-body basis set of the subspace (coeff). For example, if the many-body basis set is the one given in Eq. (8.7), the many-body eigenstates will take the form

where \(\{a_n\}\) are the coefficients stored in the coeff attribute.

The data representation, storage, and ordering conventions for the many-body states and many-body basis sets are specified in the API references of the MbState and Subspace classes, respectively.

Chemical potentials

As was already seen in Eq. (6.2) in Lever arm theory, the chemical potential of a system containing \(N\) electrons is defined by

(8.10)\[\mu(N) \equiv E_\mathrm{tot}(N) - E_\mathrm{tot}(N-1),\]

where \(E_\mathrm{tot}(N)\) is the total energy of the system in the \(N\)-electron ground state.

After calling the solve method of the many-body Solver object, the chemical potentials of the device in its current voltage configuration are calculated and stored in the chem_potentials attribute of the Device class.

Coulomb peak positions

Frequently, we are interested in the possibility of letting a current flow through the device. Under strong Coulomb interactions, such as the ones typically studied through QTCAD’s many-body solver, current can be blocked due to the phenomenon of Coulomb blockade.

The Coulomb blockade phenomenon is illustrated in Fig. 8.1. The horizontal lines within the potential well illustrate the chemical potentials of the quantum dot, i.e. the energy required to add an electron to this system. The orange regions on the side represent electron reservoirs at a source and a drain, with the highest occupied state of these reservoirs corresponding to the Fermi energy at zero temperature. In the sequential tunneling regime, for current to flow through the quantum dot, an electron from the source must first tunnel into the dot and then tunnel into the drain. At sufficiently low temperature, this can only be achieved if a chemical potential of the dot lies within the energy bias window between the source and drain chemical potentials. In all other cases, current will be blocked because there is no available quantum-dot state that allows tunneling to happen while conserving energy.

The energy-level structure of a quantum dot can often be shifted by applying a potential on some gate. In Lever arm theory, we have seen that in the linear regime, the gate potential and quantum-dot energy levels are related by the lever arm \(\alpha\). Such a lever arm may be entered into the many-body Solver through the optional argument alpha of the class constructor. A gate bias \(\varphi_\mathrm{bias}\) with respect to a reference configuration in which the many-body solver is run (and initial chemical potentials calculated) then shifts the chemical potentials by \(-e\alpha \varphi_\mathrm{bias}\), as seen in Eq. (6.1). Compared with the reference configuration (in which \(\varphi_\mathrm{bias}\equiv 0\)), current is expected to flow through the device at weak source–drain bias when

(8.11)\[\varphi_\mathrm{bias} = \mu(N)/e\alpha.\]

We thus expect to see a peak in current associated with every distinct electron number \(N\) as \(\varphi_\mathrm{bias}\) is swept. These peaks are called Coulomb peaks; their positions are stored in the coulomb_peak_pos attribute of the device object after running the solve method of the many-body Solver class.

Particle density

The particle density associated with each many-body eigenstate may be obtained using the get_many_body_density method of the Device class (see API Reference).

The electron density in a many-body eigenstate \(|\psi\rangle\) is given by

(8.12)\[\langle\rho(\mathbf r)\rangle_\psi =\sum_{j,j'}F_j^\ast(\mathbf r)F_{j'}(\mathbf r) \langle \hat c^\dagger_j \hat c_{j'}\rangle_\psi,\]

where \(j\) runs over all single-particle basis states (including the degeneracy index) and \(\langle \hat O\rangle_\psi\equiv \langle\psi|\hat O|\psi\rangle\) is the expectation value of operator \(\hat O\) in state \(|\psi\rangle\).

A simplified many-body Hamiltonian: The Fermi–Hubbard model

For weakly overlapping basis states, Eq. (8.5) reduces to

(8.13)\[H_D \approx \sum_{i\sigma}\epsilon_{i\sigma}n_{i\sigma} + \sum_{i} \frac{U_i}2 n_i(n_i-1) + \frac 12\sum_{j,j\neq i}V_{ij}n_i n_j,\]

where we have introduced the particle number operators

(8.14)\[n_i = \sum_{\sigma} n_{i\sigma},\qquad n_{i\sigma} = c^{\dagger}_{i\sigma} c_{i\sigma}.\]

In addition, in Eq. (8.13), we have introduced the Coulomb interaction energies \(U_i\) and \(V_{ij}\), which are related to the Coulomb integrals defined in Eq. (8.6) by

(8.15)\[U_i = V_{iiii}, \qquad V_{ij} = V_{ijij}.\]

The main advantage of using Eq. (8.13) instead of Eq. (8.5) to model Coulomb interactions is that for a basis set containing \(n_\mathrm{states}\) single-particle orbitals, the number of Coulomb integrals to evaluate scales like \(\mathcal O(n_\mathrm{states}^2)\) instead of the more expensive \(\mathcal O (n_\mathrm{states}^4)\) scaling achieved when including overlap terms.

Relevant device attributes

Parameter	Symbol	QTCAD name	Unit	Default	Setter
Confinement potential	\(V\)	`V`	J	0	`set_V_from_phi`, `set_V`
External confinement potential	\(V_\mathrm{ext}\)	`Vext`	J	0	`set_Vext`
Inverse effective mass tensor	\(\mathbf M^{-1}_e\)	`Me_inv`	1/kg	See note	Through `materials`