Recursive Green's Function Technique

• Recursive Green's Function (RGF) is a computational method using recursive block matrix inversions to construct Green's functions essential for quantum transport analysis.
• It employs a block-tridiagonal partitioning strategy that achieves near-linear scaling, enabling efficient simulation of mesoscopic and nanoscale quantum systems.
• RGF underpins advanced studies in quantum transport, localization, and non-equilibrium phenomena, and adapts to complex multi-terminal and inelastic scattering scenarios.

The Recursive Green's Function (RGF) technique is a foundational computational tool for determining Green's functions and associated transport properties in quantum systems featuring one or more spatially extended regions, leads, disorder, or geometric complexity. Using block-tridiagonal or hierarchical decompositions of the system Hamiltonian, RGF recasts the construction of the Green’s function—central to non-equilibrium and linear-response quantum transport—into a series of local matrix inversions and recursive updates. This results in algorithms with linear or near-linear scaling in system length, enabling the tractable simulation of meso- to macroscale systems at atomistic resolution. RGF methods underpin the simulation of quantum transport in nanoscale devices, Anderson localization, dissipative effects, quantum graphs, and are essential in non-equilibrium Green’s function (NEGF) formalisms across electronic, phononic, and photonic platforms.

1. Mathematical Foundations and Canonical Partitioning

RGF exploits the observation that, for systems with nearest-neighbor (or short-range) coupling only, the Hamiltonian $H$ can be expressed in a block-tridiagonal form: $$H_C = \begin{pmatrix} H_{1,1} & H_{1,2} & 0 & \cdots & 0 \ H_{2,1} & H_{2,2} & H_{2,3}& \ddots & \vdots \ 0 & H_{3,2} & \ddots & \ddots & 0 \ \vdots & \ddots & \ddots & H_{N_S-1,N_S-1} & H_{N_S-1,N_S} \ 0 & \cdots & 0 & H_{N_S,N_S-1} & H_{N_S,N_S} \end{pmatrix}$$ where $N_S$ is the number of slices/cells/layers along the direction of structural inhomogeneity or transport, and $H_{i,i+1}$ represents the coupling between adjacent slices (Nguyen et al., 2024).

Devices are coupled to external semi-infinite leads, whose effects are included via self-energy matrices $\Sigma_{L,R}(E)$. The retarded Green’s function for the finite region is

$$G_C(E) = [ E\,I - H_C - \Sigma_L(E) - \Sigma_R(E) ]^{-1}$$

This block structure is central to all RGF developments and is essential for achieving the favorable computational scaling (Teichert et al., 2017, Nguyen et al., 2024, Lewenkopf et al., 2013).

2. Recursive Algorithms: Forward and Backward Sweeps

The solution of $G_C(E)$ is accomplished not by direct inversion, but by recursive construction. The standard “left-to-right” algorithm follows:

1. Compute the Green’s function for the first slice (leftmost, with lead self-energy):

$G_{1,1} = [ E\,I - H_{1,1} - \Sigma_L ]^{-1}$

2. Recursively propagate to subsequent slices:

$\Sigma_{i+1}^L = H_{i+1,i}\,G_{i,i}\,H_{i,i+1}$

$$G_{i+1,i+1} = [ E\,I - H_{i+1,i+1} - \Sigma_{i+1}^L ]^{-1}$$

Backward recursion (right to left) is entirely analogous, enabling the efficient calculation of arbitrarily chosen block-diagonal or off-diagonal elements needed for local observables (Nguyen et al., 2024, Lewenkopf et al., 2013, Vaitkus et al., 2017). The recursive procedure sidesteps the $O(N^3)$ scaling of full-matrix inversion in favor of $O(N_S M^3)$ operations ($M$ the slice block size), allowing simulations of systems comprising $10^4$–$10^5$ atoms per slice (Teichert et al., 2017, Nguyen et al., 2024).

3. Applications in Quantum Transport and Localization

RGF is foundational in non-equilibrium Green’s function simulations of quantum transport in low-dimensional, disordered, and nanoscale systems. After obtaining the required block(s) of the Green’s function, physical observables are evaluated through Landauer-type formulas. For two-terminal configurations: $$T(E) = \mathrm{Tr}[\Gamma_R\,G_{N1}\,\Gamma_L\,G_{N1}^\dagger],\qquad \Gamma_{L,R} = i(\Sigma_{L,R} - \Sigma_{L,R}^\dagger)$$

$G = \frac{2e^2}{h} T(E_F)$

Disorder-induced localization properties can also be extracted. In defective carbon nanotubes, the exponential decay of the average conductance as a function of device length or defect count provides a direct estimate of the localization length (Teichert et al., 2017). For quasi-one-dimensional systems with sparse realistic defects, the improved RGF+RDA method further compresses the problem, yielding $O(N_{\rm def})$ computational scaling with the number of defects, independent of overall system size (Teichert et al., 2017).

4. Extensions: Dissipation, Multi-Terminal, and Large-Scale Devices

Advanced NEGF and atomistic simulations frequently require dissipation, dephasing, or inelastic scattering processes. RGF can be combined with Büttiker probes, introducing site- or slice-resolved fictitious leads whose associated self-energies enforce detailed balance and enable the simulation of inelastic or phase-breaking processes. The critical Jacobian for the Newton-Raphson update in the probe chemical potentials can also be recursively constructed, preserving favorable scaling (Vaitkus et al., 2017, Sadasivam et al., 2016). Multi-terminal and nontrivial geometries necessitate flexible block partitioning strategies (e.g., circular slicing for multi-terminal nanostructures (Thorgilsson et al., 2013)) and block-tridiagonal–arrowhead matrix forms (Maillou et al., 8 Jan 2026).

State-of-the-art distributed-memory parallel RGF solvers leverage Schur complement and selected inversion techniques to enable simulations of quantum devices with up to $10^6 \times 10^6$ degrees of freedom, far exceeding the limits of direct solvers (Maillou et al., 8 Jan 2026). Lead self-energy computation for extremely large supercells (e.g., Moiré superlattices) is optimized by further block decomposition and Schur reduction, allowing simulation of twisted bilayer graphene devices at the atomistic scale (Nguyen et al., 2024).

5. Special Cases: Quantum Graphs and Zero-Range Potentials

RGF is not limited to tight-binding or lattice Hamiltonians. In quantum graphs, the Green's function is recursively constructed by merging vertices or edge blocks, updating global reflection/transmission amplitudes via scattering matrices. The result is a fast algorithm for analytic or numeric evaluation of spectra, resonances, and transport in mesoscopic graph structures—cube graphs, binary trees, Sierpiński graphs—that avoids path enumeration (Andrade et al., 2016).

For systems with zero-range (delta) potentials in one or more dimensions, RGF yields exact analytical recursions for the Green’s function, exploiting the finite-rank nature of the perturbation. Extensions to manifold-supported delta functions and curved spaces are naturally accommodated; ultraviolet divergences in higher codimension are resolved by heat-kernel regularization and renormalization of couplings (Erman, 2016).

6. Numerical Stability, Implementation Details, and Limitations

RGF algorithms are numerically robust provided complex energy shifts ($i\eta$) are employed to regularize inverses near the spectrum. The block-tridiagonal structure is generally preserved for standard device-lead partitioning and can be flexibly adapted (adaptive cells, on-the-fly loading, block compression) for nonperiodic, defective, or multi-terminal configurations (Settnes et al., 2015).

Key computational advantages and trade-offs of RGF are summarized below:

Feature	RGF Scaling	Direct Inversion Scaling
Standard tight-binding device	$O(N_S M^3)$	$O(N_C^3)$
Sparse defects, improved RGF	$O(N_{\rm def} M^3)$	$O(N_S^3 M^3)$
Parallel selected inversion	$O(P \cdot N_p b^3)$ per partition	$O(N^3)$
Memory footprint (dense)	$O(N_S M^2)$	$O(N_C^2)$

Limitations arise for systems where the block structure is lost (e.g., long-range coupling), or in highly connected graphs where combinatorial complexity of scattering amplitudes grows. Still, in most physically relevant quasi-1D or layered structures, RGF remains the optimal approach.

7. Conceptual Generalizations and Theoretical Significance

The RGF formalism also serves as a mathematical lens for perturbation theory. Recursive Green's functions underpin the derivation of Feenberg perturbation theory, eliminating closed-loop (repetition) terms and providing determinant and continued-fraction representations of the exact propagator on finite Hilbert spaces (Ishida, 2019). This hierarchy unifies Rayleigh–Schrödinger, Brillouin–Wigner, and Feenberg expansions within a single recursive prescription, clarifying algebraic relationships between projection-operator, continued-fraction, and path-integral constructions.

In summary, the recursive Green’s function method is a universal and technically adaptable framework central to contemporary quantum simulation. It provides a unified path from strongly localized transport and disorder physics, multi-terminal device simulation, complex graph analysis, to formal developments in Green’s function theory, underpinned by algorithmic efficiency and analytic flexibility.