Electron transport through a junction¶
vibeqc.transport computes coherent, elastic electron transport through a
molecule or interface held between two contacts. You get the transmission
function \(T(E)\), the zero-bias conductance and Landauer current-voltage
curves, all from the Hamiltonian and overlap of an ordinary SCF calculation.
It is a post-SCF property: the SCF itself is unchanged. The plan and its
tasks are in the electron-transport design document.
The tutorial works through one
junction.
Formalism¶
The AO basis of an extended system is split into a left contact region L, a device region D and a right contact region R. Each contact supplies the retarded Green’s function \(g_c(E)\) of its region. In a non-orthogonal basis the self-energy of contact \(c\), its broadening and the device Green’s function are (Xue, Yongqiang, Datta, Supriyo, and Ratner, Mark A. (2002). *First-principles based matrix Green’s function approach to molecular electronic devices: general formalism.* Chemical Physics, **281**(2-3), 151-170. doi:[10.1016/S0301-0104(02)00446-9](https://doi.org/10.1016/S0301-0104(02)00446-9), Eqs. 3.11 and 3.24):
The transmission per spin channel is \(T(E) = \mathrm{Tr}[\Gamma_L G \Gamma_R G^\dagger]\) (Caroli, C., Combescot, R., et al. (1971). *Direct calculation of the tunneling current.* Journal of Physics C: Solid State Physics, **4**(8), 916-929. doi:[10.1088/0022-3719/4/8/018](https://doi.org/10.1088/0022-3719/4/8/018)).
Conductance: the zero-bias conductance is \(G = G_0\,T(E_F)\) for a closed-shell junction, with \(G_0 = 2e^2/h = 77.48\ \mu\)S (Fisher, Daniel S. and Lee, Patrick A. (1981). *Relation between conductivity and transmission matrix.* Physical Review B, **23**(12), 6851-6854. doi:[10.1103/PhysRevB.23.6851](https://doi.org/10.1103/PhysRevB.23.6851)). An open-shell junction sums its two spin channels with weight \(e^2/h\) each.
Current: the Landauer expression (Meir, Yigal and Wingreen, Ned S. (1992). *Landauer formula for the current through an interacting electron region.* Physical Review Letters, **68**(16), 2512-2515. doi:[10.1103/PhysRevLett.68.2512](https://doi.org/10.1103/PhysRevLett.68.2512)) applies the bias symmetrically about \(E_F\):
\[ I(V) = \frac{e}{h}\, d \sum_\sigma \int T_\sigma(E)\,[f(E-\mu_L) - f(E-\mu_R)]\,dE, \qquad \mu_{L,R} = E_F \pm eV/2.\]Here \(d\) is the spin degeneracy and \(f\) the Fermi function at the electrode temperature.
Building a junction¶
Junction holds the partitioned matrices. Construct it from AO indices,
atom indices or a converged molecular SCF result:
import numpy as np
import vibeqc as vq
from vibeqc import transport as tr
chain = vq.Molecule([vq.Atom(1, [0.0, 0.0, 2.0 * i]) for i in range(10)]) # bohr
basis = vq.BasisSet(chain, "sto-3g")
opts = vq.RKSOptions()
opts.functional = "pbe"
scf = vq.run_rks(chain, basis, opts)
junction = tr.Junction.from_result(scf, basis, left_atoms=[0], right_atoms=[9])
print(junction.device.size, junction.discarded_coupling)
The device defaults to every AO that is not in a contact. The partition is
refused when the two contacts couple directly: dropping an \(H_{LR}\) or \(S_{LR}\)
element above coupling_tolerance (default \(10^{-8}\)) would change the
physics, so you must move atoms into the device. Hartree-Fock and hybrid
Fock matrices couple distant atoms through the exchange term: on the
hydrogen chain above, the RHF Fock element between the end atoms, 18 bohr
apart, is \(2.5\times10^{-3}\) E\(_h\), while the PBE element is
\(1.7\times10^{-12}\) E\(_h\). Expect to need longer devices for them than for
semilocal functionals.
Contacts¶
A contact is any object with an n_orbitals attribute and a
retarded(energy, spin=0) method returning its Green’s function
(tr.ContactGreensFunction). Two contacts are provided:
tr.DOSContact(dos): builds \(g = -i\pi\,\mathrm{diag}(\rho(E))\) from a density of states per contact orbital, in E\(_h^{-1}\).A constant array is the wide-band limit.
A callable
dos(energy, spin)gives an energy-dependent DOS.This is the contact form of the density-of-states schemes in the literature, with the retarded sign.
tr.SemiInfiniteChainContact(onsite, hopping, overlap): the end orbital of a semi-infinite nearest-neighbour chain, in closed form, optionally with an orbital overlap. Use it for model calculations and checks.
Contacts built from a periodic projected DOS (TR-02, #1130) and principal-layer electrodes (TR-04, #1132) are planned.
Transmission, conductance and current¶
energies = np.linspace(-0.6, 0.2, 801) # hartree
lead = tr.DOSContact([1.0]) # wide band, one orbital
t = tr.transmission(junction, lead, lead, energies)
e_f = 0.5 * (scf.mo_energies[4] + scf.mo_energies[5]) # mid gap, 5 occupied
print("G(E_F) =", tr.zero_bias_conductance(junction, lead, lead, e_f), "S")
iv = tr.iv_curve(t.energies, t, np.linspace(-1.0, 1.0, 21),
fermi_level=e_f, temperature=300.0)
transmission returns T(E) per spin channel (TransmissionResult). Pass
eigenchannels=True to also get the transmission eigenvalues, each between 0
and 1. Its conductance() method gives \(G/G_0\) on the grid.
landauer_current and iv_curve integrate the Fermi window on the energy
grid, interpolating \(T\) linearly between grid points:
At 0 K the window integral is exact for that interpolant.
At finite temperature the window is resolved to \(kT/8\), and the grid must extend \(40\,kT\) beyond the bias window.
Given one transmission table, the bias is a rigid shift of the
zero-bias Hamiltonian, which is an approximation. A callable bias -> table
supplies a transmission recomputed at every bias point instead; the published
schemes recompute the SCF in a field at each bias (see
the electric field).
IVCurve.bias_model records which of the two was used, as "rigid-shift" or
"per-bias".
Output files and QVF sections¶
paths = tr.write_transport_output("h10_junction", t, iv, fermi_level=e_f,
molecule=chain)
data = tr.read_transport_qvf(paths[".qvf"])
write_transport_output routes everything through the output layer:
.out: the report, with the T(E) table, \(G(E_F)\), the I-V table and the references..system: settings and the zero-bias conductance..bibtexand.references..qvf: with these noncritical vendor sections.
QVF section |
Content |
|---|---|
|
energies (hartree), \(T(E)\) per spin channel, spin degeneracy, device \(\eta\), optional eigenchannels, Fermi level |
|
biases (volt), current (ampere), \(dI/dV\) (siemens), Fermi level, temperature (kelvin), spin degeneracy, bias model |
They are vendor kinds (x_vibeqc.*) because vibe-qc is so far their only
producer. QVF promotes a kind to the canonical registry after independent
producers adopt it (QVF governance).
read_transport_qvf reads both back and checks each member’s SHA-256.
Plots¶
vibeqc.transport.plot draws with matplotlib, imported only when called:
import matplotlib
matplotlib.use("Agg")
from vibeqc.transport.plot import iv_figure, transmission_figure
transmission_figure(t, fermi_level=e_f, log=True).savefig("h10_transmission.png")
iv_figure(iv).savefig("h10_iv.png")
What the results mean¶
Coherent and elastic: inelastic scattering by vibrations, Coulomb blockade, Kondo physics, heating and structural changes under bias are outside the method.
Kohn-Sham levels are not quasiparticle levels: semilocal functionals place molecular levels too close to the Fermi level. Computed conductances of molecular junctions are commonly above measured ones, often by an order of magnitude. Compare with another implementation at the same level of theory, not with experiment.
No macroscopic contact resistance: a calculation on one junction ranks contact motifs. It does not give the resistance of a film of molecules on a rough metal.
Large biases compare models: several volts across one molecule is far outside the linear regime. Such curves compare models; they do not predict a device characteristic.
Validation¶
tests/test_transport_landauer.py checks the core against closed forms:
the perfect chain, including a non-orthogonal one;
impurity scattering, the Breit-Wigner level and a two-level bridge;
the invariants: \(\Gamma\) positive semidefinite, \(0 \le T \le\) open channels, left-right symmetry, and odd and ohmic currents;
the transport calculator of ASE, to \(5\times10^{-9}\).
tests/test_transport_output.py covers the output files, the round trip of
both QVF sections and the plots.