Electron transport through molecular junctions¶
Status: TR-01, the TR-03 field term and TR-05 implemented, the rest planned.
Tracker #1128. The core of
TR-01 (#1129) is in python/vibeqc/transport/: the partition, the
non-orthogonal self-energies, transmission, conductance and Landauer current,
with the contact protocol, the analytic chain contact and the
density-of-states contact. Its checks are in
tests/test_transport_landauer.py. The field term of TR-03 (#1131) is also
implemented: run_job(electric_field=...) and options.electric_field on the
RHF, RKS, UHF and UKS drivers (vibeqc.electric_field). Its bias points wait
for TR-02. TR-05 (#1133) is done: write_transport_output, the
x_vibeqc.transport.transmission and x_vibeqc.transport.iv QVF vendor
sections, vibeqc.transport.plot, the
user guide and the
tutorial. These tasks have no
assigned release or milestone and do not add a release gate.
This plan adds coherent, elastic electron transport through a molecule or interface held between two electrodes: the transmission function, the zero-bias conductance and current-voltage curves from a Kohn-Sham Hamiltonian in a Gaussian basis, combined with Green’s functions for the electrodes. It is a post-SCF property. It does not change the SCF, and it is separate from the Boltzmann band transport of PROP17, which treats diffusive transport in a bulk crystal.
Motivating use case¶
Nawarathne et al. (Nanoscale Adv. 2024, 6, 428, doi:10.1039/d3na00500c, section 2.2.5 and Fig. 5) compare the current through a carbon nanotube bonded to copper with and without an aminophenyl linker. Their workflow has three parts:
DFT-optimized structures, from which two fragments are cut and placed between two small gold electrodes.
A molecular LCAO DFT calculation on that extended molecule.
The density of states of bulk gold from a periodic calculation with CRYSTAL, which stands in for the electrode Green’s function. The Generalized Electron Nano-Interface Program (GENIP) combines 2 and 3 into I-V curves from -5 V to +5 V, and conductance and resistance follow by numerical differentiation.
vibe-qc already produces both ingredients, the molecular Kohn-Sham matrices and the periodic projected density of states, with one basis-set family and one program. What is missing is the transport step. Tier A below is the equivalent of the published workflow. Tiers B and C replace its approximations step by step.
Formalism¶
Partition the atomic-orbital basis into a left electrode L, a device region D and a right electrode R, with no direct L-R coupling. In a non-orthogonal basis, with Hamiltonian H and overlap S, the retarded device Green’s function is
G(E) = [E S_DD - H_DD - Sigma_L(E) - Sigma_R(E)]^-1
Sigma_c(E) = (E S_Dc - H_Dc) g_c(E) (E S_cD - H_cD), c = L, R
where g_c is the surface Green’s function of the isolated electrode. With the broadening matrices Gamma_c = i (Sigma_c - Sigma_c^dagger), the transmission function is
T(E, V) = Tr[Gamma_L G Gamma_R G^dagger]
and the current is the Landauer expression
I(V) = (2e/h) Int T(E, V) [f(E - mu_L) - f(E - mu_R)] dE,
mu_L - mu_R = eV.
The factor 2 holds for a closed-shell system; open-shell systems sum the two spin channels. The zero-bias conductance is G = G0 T(E_F) with G0 = 2e^2/h = 77.48 microsiemens. The Fermi functions carry the electronic temperature of the electrodes and nothing else: no vibrations, no dissipation inside the device.
Primary sources: Caroli et al., J. Phys. C 4, 916 (1971),
doi:10.1088/0022-3719/4/8/018; Fisher and Lee, Phys. Rev. B 23, 6851 (1981),
doi:10.1103/PhysRevB.23.6851; Meir and Wingreen, Phys. Rev. Lett. 68, 2512
(1992), doi:10.1103/PhysRevLett.68.2512; Taylor, Guo and Wang, Phys. Rev. B
63, 245407 (2001), doi:10.1103/PhysRevB.63.245407; Brandbyge et al., Phys.
Rev. B 65, 165401 (2002), doi:10.1103/PhysRevB.65.165401; Xue, Datta and
Ratner, Chem. Phys. 281, 151 (2002), doi:10.1016/S0301-0104(02)00446-9. For
non-orthogonal surface Green’s functions see Sanvito et al., Phys. Rev. B
59, 11936 (1999), doi:10.1103/PhysRevB.59.11936. These papers are filed in
the literature library under transport/, and the DOIs were checked
against its catalog.
Three tiers¶
Tier A: density-of-states contacts¶
This tier follows the GENIP scheme as published by Sotelo, Yan, Wang and Seminario (Phys. Rev. A 75, 022511 (2007), doi:10.1103/PhysRevA.75.022511, section II C, Eqs. 1 to 7) and by Yan and Seminario (Int. J. Quantum Chem. 107, 440 (2007), doi:10.1002/qua.21074). Both are in the literature library and were read for this plan.
An extended molecule, the molecule plus a few electrode atoms on each side, is computed with the ordinary molecular driver. Its matrices are then partitioned: the device region is the molecule M alone, and the electrode atoms of the extended molecule are the contact atoms X = L, R. The coupling blocks H_MX and S_MX come from the same calculation. The Green’s function of the contact atoms is diagonal and built from the bulk density of states,
g_X(E) = -i pi diag(g_X1, ..., g_XN),
g_XK(E) = diag(D_s, D_p, D_d(t2g), D_d(eg)) of contact atom K,
with one entry per basis function of the contact atom, taken from the projection of the bulk density of states onto s, p, d(t2g) and d(eg) functions. The device Green’s function, self-energies, transmission and current are those of the formalism section, with the bias applied symmetrically as f_L(E - eV/2) and f_R(E + eV/2).
Three points of the published procedure matter for an implementation:
Each bias point is a new SCF. The extended molecule is recomputed in a uniform external electric field for every voltage, optionally with a geometry optimization in the field, and the Hamiltonian and overlap of that calculation enter the transport step. A rigid shift of the chemical potentials at the zero-field Hamiltonian is our own cheaper approximation, not the published method. Reproducing a published curve therefore needs a uniform-field one-electron term in the molecular SCF (TR-03), which vibe-qc does not have. It is built from the existing dipole integrals.
Same method and basis set for both calculations. The papers compute the extended molecule and the bulk metal at one level (B3PW91 or B3LYP with LANL2DZ and its core potentials) and use experimental lattice constants for the bulk.
The Fermi level is adjusted. Yan and Seminario shift the calculated bulk Fermi level to the experimental work function before use. Sotelo et al. report the calculated values (Au -5.80 eV against -5.31 eV measured). The alignment is an input that must be stated with every result.
Departures and conventions of this plan:
Eq. 3 of Sotelo et al. prints the prefactor as pi i. We use the retarded sign, -i pi, so that the broadening matrices are positive semidefinite. The transmission is the same for either sign.
Yan and Seminario first form S^-1 H and take the couplings from that matrix; Sotelo et al. use (E S_MX - H_MX). We implement the second form as the default and the first as an option for reproducing older results.
The density of states must be per contact basis function and per spin, in reciprocal energy. Its normalization in a non-orthogonal basis (Mulliken or Loewdin projection) is fixed and tested in TR-02.
Still open, because the original paper of the method (Derosa and Seminario, J. Phys. Chem. B 2001, 105, 471, doi:10.1021/jp003033+) and two later ones (Jauregui and Seminario, IEEE Sens. J. 2008, 8, 803, doi:10.1109/JSEN.2008.923232; Benitez and Seminario, J. Phys. Chem. C 2016, 120, 17978, doi:10.1021/acs.jpcc.6b06446) could not be fetched automatically and await a manual download:
how an applied field is converted into a bias voltage (which length is used), and whether the contact atoms feel the field;
whether contact basis functions of the same angular type on one atom share one density-of-states value or divide it;
the dependence of the result on the number of contact atoms, which the papers acknowledge and which we report for every junction.
The bulk density of states comes from vibe-qc’s own periodic SCF with Fermi smearing and the existing projected density of states; the d projection needs the t2g and eg resolution.
Tier A is cheap and works with hybrid functionals. It neglects the real part and the off-diagonal structure of the electrode Green’s function, so absolute currents are qualitative. Sotelo et al. state the same purpose: one contact shape is used equally for the systems compared.
Tier B: electrodes from principal layers¶
The electrodes become semi-infinite periodic stacks. A periodic calculation
of the electrode gives the blocks H00, H01, S00 and S01 of one principal
layer and its coupling to the next. The surface Green’s function follows
from Lopez Sancho-Rubio decimation (J. Phys. F 15, 851 (1985),
doi:10.1088/0305-4608/15/4/009), which
vibeqc.periodic_embedding.decimation and
vibeqc.periodic_embedding.substrate_gf already implement for the
experimental surface-embedding route, including the Gaussian-basis layer
partition. The device region contains the molecule and at least one
principal layer of each electrode as a buffer, so that the device-electrode
coupling equals the bulk interlayer coupling and the potential at the
boundary is bulk-like.
Requirements beyond tier A:
transverse periodicity with k-point sampling of T(E), or a finite nanowire cross-section;
alignment of the electrode and device potentials through the buffer layers, with a recorded residual;
a converged metallic device calculation, which depends on Fermi smearing for slabs and on the experimental status of the embedding route;
semilocal functionals for the electrodes first. Exact exchange in a metallic lead is long-ranged and is not part of the first delivery.
Tier B delivers zero-bias transmission and conductance without the density-of-states approximation. A non-self-consistent bias window with rigidly shifted chemical potentials is available as a labelled approximation.
Tier C: self-consistent finite bias¶
A full non-equilibrium treatment builds the device density from the
Green’s function, with a complex-contour integral for the equilibrium part
(vibeqc.periodic_embedding.contour) and a real-axis integral over the
bias window for the non-equilibrium part, and solves the electrostatics
with the electrode potentials as boundary conditions. The second point is
a new boundary-value problem for a Gaussian-basis code. Tier C is therefore
a research item with a decision gate after tier B (TR-06), not a committed
delivery.
Decision gate for tier C (TR-06)¶
Recommendation: no-go for a committed tier C in the current density-fitting architecture, recorded 2026-11-13. The two published self-consistent finite-bias methods the gate asks to compare agree with the tier C plan on one ingredient and contradict vibe-qc on the other, and the contradicting one is the hard one.
Non-equilibrium density from the Green’s function. Both methods split the density into an equilibrium part, evaluated by a complex-contour integral over the Fermi poles (Taylor, Guo and Wang, PRB 63, 245407 (2001), “Calculation of charge density”; Brandbyge et al., PRB 65, 165401 (2002), section III.B, “Complex contour for the equilibrium density matrix”, Eqs. 15 to 25), and a non-equilibrium part, a real-axis integral over the bias window
[E_F - eV/2, E_F + eV/2]weighted byf_L - f_R(Taylor 2001, Eqs. 10 and 19; Brandbyge 2002, section III.C). The equilibrium contour is already available asEnergyContour(vibeqc.periodic_embedding.contour), and TR-01 supplies a rigid-shift bias-window integral for the current (_window_integralinvibeqc.transport.landauer). What is missing is the self-consistent non-equilibrium density, a new but small integral overG(E). This ingredient is not the blocker.The electrostatic boundary-value problem. This is the blocker. Both methods close the open-system KS potential with a screening approximation in which the device potential is matched to the bulk electrode potentials at two boundary planes: the effective potential equals the left bulk potential for
z < z_land the right bulk potential forz >= z_r(Taylor 2001, Eq. 1, and the section that solves the Poisson equation in three-dimensional real space; Brandbyge 2002, section IV), and they solve the Poisson equation in real space with those boundary conditions, re-solved at every self-consistent iteration under the bias. vibe-qc has none of this: its Coulomb is a global density fit (GDF/Bulirsch), not a real-space Poisson solve; it carries no electrode potential, no screening plane and no real-space Poisson operator. The transport core (TR-01), the tier A density-of-states contact and the rigid-shift bias-window current are invibeqc.transport/, but no self-consistent finite-bias density exists, and the prerequisite tier B (TR-04, principal-layer electrodes) is unstarted.
Cost and benefit on a junction:
Cost. A real-space Poisson boundary-value solver with electrode-potential boundary conditions (architecturally alien to the density-fitting Coulomb), the self-consistent bias-window density integral and a completed, validated tier B (TR-04, principal-layer electrodes). The transport core (TR-01) and the uniform field term (TR-03) are already in place. The remaining work is the electrostatics and the electrode model, a large and multi-component research effort.
No same-architecture reference. Neither required paper is a density-fitting LCAO code: Taylor 2001 uses norm-conserving pseudopotentials with a real-space potential and Brandbyge 2002 uses finite-range numerical atomic orbitals with a real-space Poisson solve. There is no same-level self-consistent finite-bias result to validate a vibe-qc tier C against; the comparison target would have to be a same-level independent implementation in a different architecture (for example a real-space or numerical-atomic-orbital NEGF code such as TranSIESTA), consistent with the “Validation” section, which already notes no ORCA or CRYSTAL reference exists for this property.
Limited payoff. As “What the results mean” states, semilocal functionals overestimate molecular-junction conductances by about an order of magnitude, so the target of a self-consistent tier C is another implementation at the same level of theory, not experiment. Tiers A and B already deliver qualitative and semi-quantitative
T(E),G0andI(V)with a labelled non-self-consistent bias; tier C’s marginal gain is self-consistency, not accuracy.
Recommendation. Do not commit tier C in the current architecture. Keep the non-self-consistent bias window as the labelled approximation in tiers A and B, and gate any future self-consistent finite-bias work behind (i) a real-space Poisson solver added to vibe-qc, (ii) a completed and validated tier B (TR-04), (iii) a same-level real-space or numerical-atomic-orbital NEGF reference for acceptance. The equilibrium contour and surface Green’s function ingredients that already exist should be kept, not removed. The quantitative junction-by-junction comparison against tiers A and B is deferred until TR-04 supplies a tier B junction to compare with; the architectural and literature findings above already make the no-go independent of that number.
Primary sources: Taylor, Guo and Wang, Phys. Rev. B 63, 245407 (2001),
doi:10.1103/PhysRevB.63.245407; Brandbyge et al., Phys. Rev. B 65, 165401
(2002), doi:10.1103/PhysRevB.65.165401. Both are filed in the literature
library under transport/.
What the results mean¶
Kohn-Sham eigenvalues are not quasiparticle levels. Semilocal functionals place molecular levels too close to the Fermi level, and computed conductances of molecular junctions are commonly above measured ones, often by an order of magnitude. The comparison target is another implementation at the same level of theory, not experiment.
The method describes coherent elastic transport. Inelastic scattering by vibrations, Coulomb blockade, Kondo physics, heating and any structural change under bias are outside it.
A bias of several volts across a single molecule is far outside the linear regime. Curves at such biases compare models; they are not predictions of a device characteristic.
A film of many nanotubes on a rough metal is not one molecular junction. The calculation ranks contact motifs. It does not give a contact resistance of the macroscopic interface.
Interface sketch¶
import vibeqc as vq
from vibeqc import transport as tr
em = vq.run_job(extended_molecule, method="rks", functional="pbe0", ...)
bulk = vq.run_periodic_job(gold_bulk, smearing=..., ...)
junction = tr.Junction.from_molecule(em, left=left_atoms, right=right_atoms)
contacts = tr.DOSContact.from_bulk(bulk) # tier A
# contacts = tr.LayerContact.from_periodic(lead) # tier B
t = tr.transmission(junction, contacts, energies=energies)
iv = tr.current(junction, contacts, bias=bias, temperature=300.0)
Names are provisional. The module lives in python/vibeqc/transport/ and
is NumPy-based: the matrices are small and dense, and a native kernel is
justified only if tier C goes ahead. Results are written through
vibeqc.output with the citations of the methods used. QVF gains the
vendor sections x_vibeqc.transport.transmission and x_vibeqc.transport.iv,
so that vibe-viewer can plot T(E), I(V) and dI/dV. QVF governance promotes a
vendor kind to the canonical transport.* names after independent producers
adopt it. The viewer request is vibe-viewer#212.
Validation¶
No ORCA or CRYSTAL reference exists for this property, so acceptance rests on analytic limits, an independent implementation and published values.
Check |
Expectation |
|---|---|
Uniform tight-binding chain |
T = 1 inside the band, 0 outside; reuses |
Single impurity and two-level bridge |
Closed-form T(E), including the Breit-Wigner limit |
Invariants |
Gamma positive semidefinite, 0 <= T(E) <= number of open channels, T identical for both directions, I(-V) = -I(V) for a symmetric junction |
Independent implementation |
Same H and S passed to the transport calculator of ASE ( |
Gold atomic chain (tier B) |
Conductance close to 1 G0 at the Fermi level |
Benzene-1,4-dithiol between gold (tier B) |
Agreement with published DFT-NEGF results at the same functional; the known gap to experiment is reported, not tuned away |
Linker comparison of the Cu paper (tier A) |
Same ordering of the two junctions and similar curve shape, from the authors’ structures; numerical agreement is not expected without their inputs |
Each calculation used as acceptance evidence is run through vq and
records geometry, functional, basis set, core potential, smearing, energy
grid and broadening.
Tasks¶
Task |
Content |
Depends on |
Priority |
|---|---|---|---|
TR-01 (#1129) |
Partition, self-energy algebra in a non-orthogonal basis, transmission, Landauer current, invariants, analytic models, optional ASE cross-check |
none |
P2 |
TR-02 (#1130) |
Tier A: density-of-states contacts from the periodic projected DOS with t2g and eg resolution, energy alignment, contact-size study, rigid-shift bias as a labelled approximation; reproduction of a published curve also needs TR-03 |
TR-01 |
P2 |
TR-03 (#1131) |
Uniform external field in the molecular SCF; one SCF per bias point as in the published scheme |
none for the field term; TR-02 for the bias points |
P2 |
TR-04 (#1132) |
Tier B: principal-layer electrodes from the existing decimation, buffer layers, k-point sampling, zero-bias conductance |
TR-01, metallic slab SCF |
P3 |
TR-05 (#1133) |
Output, QVF sections, plots, citations, user guide and tutorial |
TR-01 |
P2 |
TR-06 (#1134) |
Decision gate for tier C: boundary-value electrostatics, cost and benefit, written recommendation, decided no-go (2026-11-13; see the subsection) |
TR-04 |
P3 |
Out of scope¶
Inelastic and phonon-assisted transport, time-dependent transport, strong correlation, spin-orbit coupling and thermoelectric coefficients are not part of this plan. Boltzmann band transport remains PROP17.