Eve library

phase-118.md

Tracker

Revision 1

On this page

Historical task source. The Eve task board holds current task state.

Phase 118: Kalika Many-Body Electronic Structure — GW Approximation, Bethe-Salpeter Equation, DMFT, Quantum Monte Carlo & Coupled Cluster for Solids#

Beyond-DFT methods essential for accurate band gaps, optical spectra, and strongly correlated materials. Covers the full hierarchy from perturbative GW through BSE excitonics, DMFT for strong correlations, QMC stochastic methods, and wavefunction-based coupled cluster for solids. Every library targets parity with the leading codes in each domain: BerkeleyGW and VASP for GW/BSE, ABINIT GW for space-time methods, w2dynamics and TRIQS for DMFT, QMCPACK for QMC, and PySCF/CC4S for periodic coupled cluster. Rust (~55%) handles compute-heavy many-body kernels (matrix operations, Monte Carlo sampling, tensor contractions), TypeScript (~35%) provides orchestration, I/O, and user-facing APIs, Python (~10%) bridges to existing electronic structure ecosystem (PySCF, ASE, GPAW).


118.1 Scaffolding & Core Types (@kalika/many-body)#

118.1.1 Green's Function Infrastructure#

  • 118.1.1.1 Implement GreenFunction type: frequency-dependent single-particle Green's function G(k,ω) stored on either a Matsubara frequency grid (iω_n = i(2n+1)πT for fermions) or real-frequency grid, with spin indices (σ,σ'), band indices (n,m), and k-point labels — supporting both retarded G^R(ω) and Matsubara G(iω_n) representations with conversion between them
  • 118.1.1.2 Implement SelfEnergy type: Σ(k,ω) with the same index structure as GreenFunction, plus methods for extracting quasiparticle properties — Z-factor Znk = (1 - ∂ReΣ/∂ω|{ε_nk})⁻¹, quasiparticle energy E_nk = ε_nk + Z_nk · ReΣ(k,ε_nk), and quasiparticle lifetime τ_nk = ℏ/(2|ImΣ(k,E_nk)|)
  • 118.1.1.3 Implement Polarizability type: irreducible polarizability χ₀(G,G',q,ω) as a matrix in reciprocal lattice vectors G,G' for each (q,ω) point, with methods for computing from KS orbitals via the Adler-Wiser formula: χ₀(G,G',q,ω) = (2/Ω) Σ*{v,c,k} ⟨v,k|e^{-i(q+G)·r}|c,k+q⟩ ⟨c,k+q|e^{i(q+G')·r}|v,k⟩ × [1/(ω - ε*{ck+q} + ε*{vk} + iη) - 1/(ω + ε*{ck+q} - ε_{vk} - iη)]
  • 118.1.1.4 Implement DielectricFunction type: ε(G,G',q,ω) = δ*{GG'} - v(q+G) χ₀(G,G',q,ω) where v(q+G) = 4πe²/|q+G|² is the bare Coulomb interaction, with methods for computing the macroscopic dielectric function ε_M(q,ω) = 1/ε⁻¹*{G=0,G'=0}(q,ω) including local field effects
  • 118.1.1.5 Implement ScreenedInteraction type: W(G,G',q,ω) = ε⁻¹(G,G',q,ω) · v(q+G'), computed by inverting the dielectric matrix at each (q,ω) point, with methods for extracting the screened Coulomb interaction W_c = W - v (correlation part only)
  • 118.1.1.6 Implement SpectralFunction type: A(k,ω) = -(1/π) Im G^R(k,ω) for single-particle spectral functions, with methods for computing momentum-resolved A(k,ω) and momentum-integrated DOS N(ω) = Σ_k A(k,ω), plus Fermi surface extraction at ω=0
  • 118.1.1.7 Implement Vertex type: three-point vertex function Γ(k,k',q;ω,ω',ν) in the particle-hole channel, with the bare vertex Γ₀ = 1 and the dressed vertex satisfying the Bethe-Salpeter equation in the vertex channel — stored as a rank-3 tensor over (k,k',q) and three frequency arguments

118.1.2 Frequency & Momentum Grids#

  • 118.1.2.1 Implement Matsubara frequency grid: uniform grid iω_n = i(2n+1)πT for fermions and iν_m = i·2mπT for bosons at temperature T, with high-frequency tail fitting G(iω_n) ~ 1/iω_n + c₁/(iω_n)² + c₂/(iω_n)³ for accurate Fourier transforms to imaginary time
  • 118.1.2.2 Implement real-frequency grid: non-uniform grid with dense sampling near the Fermi level (Gauss-Legendre or tangent-mesh), broadening parameter η management, and Hilbert transform for converting between spectral function A(ω) and retarded Green's function G^R(ω)
  • 118.1.2.3 Implement imaginary-time grid: τ ∈ [0,β] with optimized non-uniform mesh (IR basis or DLR — discrete Lehmann representation), supporting fast Fourier transforms between τ and iω_n via the intermediate representation (IR) basis of Shinaoka et al.
  • 118.1.2.4 Implement k-point grid infrastructure: Monkhorst-Pack grids with time-reversal symmetry reduction, symmetry-inequivalent k-point selection, tetrahedron method for BZ integration (Bloechl improved tetrahedron), and interpolation between k-grids via Wannier functions
  • 118.1.2.5 Implement frequency-momentum interpolation: Wannier interpolation for band energies ε_nk and matrix elements ⟨nk|O|mk+q⟩ onto arbitrary fine k-grids from a coarse DFT grid, using maximally localized Wannier functions (MLWF) from Wannier90-compatible data

118.1.3 I/O & Interoperability#

  • 118.1.3.1 Implement DFT input readers: parse KS eigenvalues, orbitals, and wavefunctions from VASP (WAVECAR, EIGENVAL, vasprun.xml), Quantum ESPRESSO (data-file-schema.xml, UPF pseudopotentials), ABINIT (WFK files), and GPAW (gpw files) — extracting band energies, occupation numbers, k-point weights, and plane-wave coefficients
  • 118.1.3.2 Implement Wannier90 interface: read/write wannier90.win, wannier90.amn (projection matrix), wannier90.mmn (overlap matrix), wannier90.eig (eigenvalues), and the resulting _hr.dat (Hamiltonian in Wannier basis) — enabling Wannier interpolation of all many-body quantities
  • 118.1.3.3 Implement HDF5-based checkpoint format: save and restore intermediate quantities (χ₀, ε, W, Σ) to HDF5 files with standardized group structure, enabling restart of interrupted calculations and reuse of expensive intermediates (e.g., reuse W for multiple BSE calculations)
  • 118.1.3.4 Implement BerkeleyGW compatibility layer: read/write BerkeleyGW binary formats (epsmat, eps0mat for dielectric matrices; sigma for self-energy; bsedmat/bsexmat for BSE kernel) for cross-validation

118.1.4 Parallelism & Memory Management#

  • 118.1.4.1 Implement distributed-memory data layout for large matrices: block-cyclic distribution of ε(G,G',q,ω) across MPI ranks using ScaLAPACK conventions, with 2D process grid mapping and efficient redistribution between different parallel decompositions (q-parallelism for χ₀ vs. G-parallelism for matrix inversion)
  • 118.1.4.2 Implement memory-efficient frequency treatment: out-of-core storage for full-frequency quantities, with on-demand loading of frequency slices, LRU caching of recently accessed (q,ω) blocks, and streaming computation patterns that process one frequency at a time
  • 118.1.4.3 Implement GPU acceleration stubs with WASM/napi-rs bridge: define GPU kernel interfaces for matrix-matrix multiplication (ZGEMM), batched matrix inversion, and FFT — with Rust implementations via wgpu/cublas bindings and fallback to CPU BLAS (OpenBLAS/MKL via ndarray-linalg)

118.1.5 Symmetry Infrastructure#

  • 118.1.5.1 Implement crystal symmetry operations for many-body quantities: apply space group symmetries to reduce BZ sums in χ₀, Σ, and W — for each symmetry operation S in the little group of q, relate matrix elements at Sk to those at k via rotation of plane-wave coefficients, reducing the number of independent k-points by a factor of |G_q| (little group order)
  • 118.1.5.2 Implement time-reversal symmetry exploitation: use G(k,ω) = G*(-k,ω) and W(q,ω) = W(-q,ω) to halve the BZ sampling, with proper handling of systems where time-reversal is broken (magnetic systems, systems with SOC and broken inversion)
  • 118.1.5.3 Implement symmetry-reduced k-point unfolding: given quantities computed on the irreducible BZ, unfold to the full BZ using symmetry operations for band-structure plotting and spectral function visualization along arbitrary k-paths
  • 118.1.5.4 Implement q-point symmetry for the dielectric matrix: use the relation ε(Sq,ω) = S · ε(q,ω) · S⁻¹ to reduce the number of independent q-points for W and χ₀ computations, with proper treatment of the q→0 limit under symmetry operations

118.1.6 Unit & Conversion Infrastructure#

  • 118.1.6.1 Implement energy unit system: automatic conversion between Hartree (atomic units), eV, Ry, cm⁻¹, K, and THz — with all internal calculations in atomic units and user-facing output in configurable units, including physical constants (Bohr radius, fine structure constant, etc.) at CODATA 2018 precision
  • 118.1.6.2 Implement Fourier transform utilities: FFT between real-space and reciprocal-space representations for wavefunctions, densities, and potentials, with proper normalization conventions (1/√Ω for plane waves), zero-padding for interpolation, and support for non-cubic cells via the cell-adapted FFT grid
  • 118.1.6.3 Implement Coulomb matrix element utilities: compute bare Coulomb matrix elements ⟨nk|v|mk+q⟩ in the plane-wave basis, with proper treatment of the G=0 divergence (auxiliary function method or Gygi-Baldereschi method) and head/wing/body decomposition for the q→0 limit

118.2 GW Approximation (@kalika/gw)#

118.2.1 RPA Polarizability & Dielectric Function#

  • 118.2.1.1 Implement Adler-Wiser sum-over-states for χ₀(G,G',q,ω): compute the irreducible polarizability from KS orbitals and eigenvalues, with proper treatment of the q→0 limit (head, wing, body elements of χ₀ matrix), occupation factors for metallic systems, and symmetry reduction of the BZ sum using crystal symmetries that leave q invariant (little group)
  • 118.2.1.2 Implement static COHSEX approximation: Coulomb-hole plus screened-exchange (COHSEX) as a computationally cheap starting point — Σ_SEX(r,r') = -Σ_n f_n ψ_n(r) ψ*_n(r') W(r,r',ω=0) and Σ_COH(r,r') = ½ δ(r-r') [W(r,r',ω=0) - v(r,r')] — useful for generating better starting wavefunctions for subsequent G₀W₀
  • 118.2.1.3 Implement dielectric matrix construction and inversion: ε(G,G',q,ω) = δ_{GG'} - v(q+G) χ₀(G,G',q,ω), then compute ε⁻¹ via dense matrix inversion (ZGETRF/ZGETRI) for each (q,ω) point, with proper treatment of the G=0, G'=0 element (head of ε) which determines the macroscopic dielectric constant
  • 118.2.1.4 Implement screened Coulomb interaction W: W(G,G',q,ω) = Σ_{G''} ε⁻¹(G,G'',q,ω) v(q+G''), and the correlation part W_c = W - v, with symmetry W(G,G',q,ω) = W*(G',G,q,-ω) for Hermiticity verification

118.2.2 Plasmon-Pole Models#

  • 118.2.2.1 Implement Hybertsen-Louie plasmon-pole model (PPM): fit ε⁻¹(G,G',q,ω) = δ*{GG'} + Ω²*{GG'}(q) / (ω² - ω̃²_{GG'}(q)) using two constraints — static ε⁻¹(ω=0) from RPA and the f-sum rule ∫₀^∞ ω Im[ε⁻¹(G,G',q,ω)] dω = -π/2 · ω²_p · (q+G)·(q+G')/|q+G|² — to determine Ω² and ω̃² parameters
  • 118.2.2.2 Implement Godby-Needs plasmon-pole model: fit ε⁻¹ at two frequencies (ω=0 and an imaginary frequency iω_1, typically iω_1 ≈ 1 Ry) to the single-pole form, which avoids the f-sum rule and is better suited for systems where the sum rule is poorly satisfied
  • 118.2.2.3 Implement Engel-Farid generalized plasmon-pole model: use multiple poles to better represent the frequency dependence of ε⁻¹, with constraints from higher-order sum rules — bridging the gap between single-pole PPMs and full-frequency integration

118.2.3 Full-Frequency GW#

  • 118.2.3.1 Implement contour deformation (CD) method for Σ(ω): deform the frequency integral in Σ = iGW from the real axis to the imaginary axis plus residue contributions: Σ(ω) = (i/2π) ∫_{-∞}^{∞} G(ω+iω')W_c(iω')dω' + Σ_j R_j(ω), where R_j are residues from poles of G enclosed by the contour deformation, with adaptive frequency quadrature on the imaginary axis
  • 118.2.3.2 Implement analytic continuation (AC) method: compute Σ on the imaginary frequency axis Σ(iω_n) (where quantities are smooth), then analytically continue to real frequencies using Padé approximants (Thiele's continued fraction) or multipole fitting — with validation of pole structure and comparison against CD results
  • 118.2.3.3 Implement full-frequency integration on the real axis: direct numerical integration of Σ(ω) = (i/2π) ∫ G(ω+ω')W_c(ω')dω' on a dense real-frequency grid with broadening η, using adaptive Gauss-Kronrod quadrature — as reference for validating CD and AC methods

118.2.4 One-Shot G₀W₀#

  • 118.2.4.1 Implement G₀W₀ self-energy calculation: compute Σ^{G₀W₀}(k,ω) from DFT orbitals and eigenvalues, constructing G₀ from KS eigenvalues ε_nk and W₀ from the RPA dielectric function built with KS orbitals — the standard perturbative GW correction on top of DFT
  • 118.2.4.2 Implement quasiparticle equation solver: solve E_nk = ε_nk + Z_nk · [ReΣ(k,ε_nk) - V_xc(k)] where V_xc is the DFT exchange-correlation potential, using either the linearized solution (Z-factor approximation) or the graphical/iterative solution of the full nonlinear equation ω = ε_nk + ReΣ(k,ω) - V_xc(k)
  • 118.2.4.3 Implement spectral function output: compute A(k,ω) = -(1/π) ImΣ(k,ω) / [(ω - ε_nk - ReΣ(k,ω) + V_xc)² + (ImΣ(k,ω))²] on a real-frequency grid, showing satellite structures beyond the quasiparticle peak (plasmon satellites at ω ≈ E_nk ± ω_p)
  • 118.2.4.4 Implement starting-point dependence analysis: run G₀W₀ from different DFT functionals (LDA, PBE, HSE, PBE0) and quantify the spread in predicted band gaps, identifying the optimal starting point via comparison with experiment or higher-level theory

118.2.5 Partial Self-Consistency#

  • 118.2.5.1 Implement GW₀ (eigenvalue self-consistency in G only): iterate G with updated quasiparticle energies while keeping W₀ fixed at the RPA level — update ε_nk → E_nk in G, recompute Σ = iGW₀, solve QP equation, repeat until quasiparticle energies converge (typically 3-5 iterations), with convergence threshold |ΔE_nk| < 1 meV
  • 118.2.5.2 Implement scGW (full eigenvalue self-consistency in both G and W): iterate both G and W — update ε_nk → E_nk in both G and χ₀ (recalculate polarizability with GW energies), recompute W, recompute Σ, until convergence — with damping/mixing to stabilize the self-consistency loop (Pulay DIIS or linear mixing)
  • 118.2.5.3 Implement scGW with off-diagonal self-energy: go beyond diagonal approximation by including off-diagonal matrix elements Σ*{nm}(k,ω) with n≠m, requiring rotation of the single-particle basis at each iteration to diagonalize the QP Hamiltonian H^QP*{nm}(k) = εnk δ{nm} + ⟨nk|Σ(E_nk) - V_xc|mk⟩

118.2.6 Quasiparticle Self-Consistent GW (QSGW)#

  • 118.2.6.1 Implement QSGW self-consistency loop: construct an optimal non-interacting Hamiltonian H₀ = T + Vext + V_H + Σ_xc^{QSGW} where Σ_xc^{QSGW} is a static, Hermitian approximation to the GW self-energy: Σ_xc^{QSGW}{nm}(k) = ½ Re[Σ_{nm}(k,E_nk) + Σ_{nm}(k,E_mk)], then solve H₀ for new orbitals and eigenvalues, and iterate to self-consistency
  • 118.2.6.2 Implement QSGW orbital update: diagonalize the QSGW Hamiltonian to obtain new orbitals {ψ_nk^{QSGW}} that differ from KS orbitals, requiring plane-wave coefficient updates and re-evaluation of all matrix elements at each iteration
  • 118.2.6.3 Implement QSGW+BSE workflow: use converged QSGW orbitals and eigenvalues as input for BSE optical spectra calculations, with proper computation of the screened interaction W within the QSGW framework

118.2.7 Space-Time Method & Scaling Reduction#

  • 118.2.7.1 Implement GW space-time method (Rojas-Godby-Needs): compute the GW self-energy in imaginary time and real space: Σ(r,r',iτ) = -G(r,r',iτ) · W_c(r,r',iτ), avoiding the expensive frequency convolution — transform quantities between (r,τ) ↔ (G,iω) representations using FFT, achieving O(N³) scaling instead of O(N⁴)
  • 118.2.7.2 Implement minimax time-frequency grids: use the minimax grids of Kaltak-Klimes-Kresse for optimal quadrature on the imaginary time/frequency axes with minimal number of grid points (typically 16-32 points suffice for meV accuracy), including the transformation matrices between τ and iω representations
  • 118.2.7.3 Implement low-rank decomposition of dielectric matrix: interpolative decomposition (ID) or randomized SVD of ε(G,G',q,ω) to reduce the auxiliary basis size from N_G to N_aux ≪ N_G, using the fact that ε is often numerically low-rank — reducing memory from O(N_G²) to O(N_G · N_aux) per (q,ω) point

118.2.8 Coulomb Truncation & Reduced Dimensionality#

  • 118.2.8.1 Implement slab Coulomb truncation for 2D systems: replace the 3D Coulomb kernel v(q+G) = 4π/|q+G|² with a truncated kernel that vanishes beyond a cutoff length Lz/2 in the non-periodic direction, v_trunc(q∥,Gz) = (4π/|q+G|²)[1 - e^{-|q+G∥|L_z/2} cos(G_z L_z/2)], eliminating spurious interlayer interactions in slab supercells
  • 118.2.8.2 Implement wire Coulomb truncation for 1D systems: cylindrical truncation with radius R_c, and spherical truncation for 0D (molecular) systems with radius R_s, both derived from the Beigi-Ismail-Louie truncation scheme
  • 118.2.8.3 Implement Wigner-Seitz truncation: general truncation scheme that restricts the Coulomb interaction to the Wigner-Seitz cell of the superlattice, applicable to any dimensionality — computed via auxiliary real-space integration with FFT acceleration

118.2.9 Spin-Orbit & Vertex Corrections#

  • 118.2.9.1 Implement GW+SOC: include spin-orbit coupling in the GW calculation by working with spinor wavefunctions (two-component), requiring 2×2 spin blocks in G, Σ, χ₀, and W — with proper treatment of time-reversal symmetry breaking in the spin-orbit case and Kramers degeneracy at time-reversal invariant momenta
  • 118.2.9.2 Implement vertex corrections (GWΓ): compute the first-order vertex correction Γ = 1 + δΣ/δG · G · G · Γ in the local vertex approximation (Ward identity preserving), improving the self-energy beyond the GW level — particularly important for d- and f-electron systems where GW alone underestimates correlations
  • 118.2.9.3 Implement GW total energy via RPA correlation energy: E_c^{RPA} = (1/2π) ∫₀^∞ Tr[ln(1 - vχ₀(iω)) + vχ₀(iω)] dω + exact exchange E_x, providing a total energy consistent with the GW approximation for equation-of-state and structural optimization

118.2.10 GW Convergence & Workflow#

  • 118.2.10.1 Implement band-count convergence analysis: systematically increase the number of unoccupied bands N_bands in χ₀ and Σ summations, track quasiparticle energies vs. N_bands, and apply the extrapolar correction (Bruneval-Gonze) to accelerate convergence by approximating high-lying states with free-electron-like energies
  • 118.2.10.2 Implement G-vector cutoff convergence: vary the reciprocal lattice vector cutoff |G|_max for the dielectric matrix dimension, tracking convergence of band gaps — with the Hybertsen-Louie static remainder technique to reduce the required cutoff by analytically summing contributions from G-vectors beyond the cutoff
  • 118.2.10.3 Implement k-point convergence analysis: run GW at increasing k-grid densities (4×4×4, 6×6×6, 8×8×8, ...) and extrapolate band gaps to the infinite k-grid limit using 1/N_k fits, identifying materials where dense k-grids are essential (small-gap semiconductors, 2D materials)
  • 118.2.10.4 Implement GW workflow orchestrator: automated pipeline that chains DFT → Wannier90 → χ₀ → ε → W → Σ → QP energies, with checkpoint/restart at each stage, convergence parameter validation, and automatic fallback from full-frequency to PPM when memory is insufficient
  • 118.2.10.5 Implement GW band structure interpolation: interpolate quasiparticle corrections Σ_nk - V_xc(k) from the coarse GW k-grid onto a fine k-path using Wannier interpolation, enabling smooth GW band structure plots along high-symmetry paths without running GW at every k-point
  • 118.2.10.6 Implement GW density of states: compute the quasiparticle DOS from the spectral function A(k,ω) using tetrahedron integration on the fine k-grid, with proper treatment of the energy-dependent broadening from Im[Σ(k,ω)] — comparing QP-DOS with the DFT DOS to visualize the GW correction effect
  • 118.2.10.7 Implement core-level GW: extend GW to compute core-level binding energies (XPS) by including core states in the self-energy evaluation, with proper treatment of the deep core-hole screening and the core-valence exchange interaction — targeting agreement with experimental XPS binding energies within 0.3 eV

118.3 Bethe-Salpeter Equation (@kalika/bse)#

118.3.1 BSE Kernel Construction#

  • 118.3.1.1 Implement the electron-hole interaction kernel: construct the BSE Hamiltonian matrix H^{BSE}{(vck),(v'c'k')} = (E_ck - E_vk)δ{vv'}δ*{cc'}δ*{kk'} + (2V^x*{(vck),(v'c'k')} - W^d*{(vck),(v'c'k')}) where V^x is the bare exchange (repulsive electron-hole exchange) and W^d is the screened direct (attractive electron-hole Coulomb) interaction
  • 118.3.1.2 Implement bare exchange kernel V^x: V^x*{(vck),(v'c'k')} = ∫∫ ψ*_ck(r) ψ_c'k'(r) v(r,r') ψ*{vk}(r') ψ*_{v'k'}(r') dr dr', computed in reciprocal space as a sum over G vectors using plane-wave matrix elements ⟨ck|e^{iG·r}|c'k'⟩
  • 118.3.1.3 Implement screened direct kernel W^d: W^d*{(vck),(v'c'k')} = ∫∫ ψ*_ck(r) ψ*{v'k'}(r) W(r,r',ω=0) ψ*{vk}(r') ψ**{c'k'}(r') dr dr', using the statically screened interaction W(ω=0) from the GW calculation, with proper interpolation of W from the coarse q-grid to the fine k-grid via Wannier or Fourier interpolation
  • 118.3.1.4 Implement kernel symmetries and storage optimization: exploit time-reversal symmetry H^{BSE}{(vck),(v'c'k')} = [H^{BSE}{(v'c'k'),(vck)}]* for Hermiticity, crystal symmetry to relate kernel elements at symmetry-equivalent k-points, and band truncation (include only valence/conduction bands within an energy window of the gap)

118.3.2 Tamm-Dancoff Approximation (TDA)#

  • 118.3.2.1 Implement TDA-BSE eigenvalue problem: solve H^{res} AS = Ω_S A_S where H^{res} is the resonant (positive-frequency) block of the BSE Hamiltonian, yielding exciton energies Ω_S and excitonic wavefunctions Ψ_S(r_e,r_h) = Σ{vck} A^S*{vck} ψ_ck(r_e) ψ**{vk}(r_h) — the standard approximation used in most BSE codes
  • 118.3.2.2 Implement direct diagonalization for small systems: full ZHEEV/ZHEEVR diagonalization of H^{res} when the matrix dimension (N_v × N_c × N_k) is manageable (< 10,000), yielding all exciton states
  • 118.3.2.3 Implement iterative Lanczos solver: Lanczos recursion for computing the optical absorption spectrum Im[ε(ω)] = Σ_S |d_S|² δ(ω-Ω_S) without full diagonalization — only the Lanczos coefficients (α_n, β_n) are needed, and the spectrum is obtained from the continued fraction ε(ω) ~ d† (ω - H^{res})⁻¹ d where d is the dipole vector
  • 118.3.2.4 Implement Haydock recursion method: three-term recursion for the resolvent (ω - H)⁻¹ projected onto the dipole direction, generating continued-fraction coefficients with typically 500-2000 iterations for converged spectra, plus kernel polynomial method (KPM) as alternative with Chebyshev expansion and Jackson damping

118.3.3 Full BSE (Beyond TDA)#

  • 118.3.3.1 Implement full BSE Hamiltonian including coupling: construct the 2×2 block structure [[H^{res}, K^{coup}], [-K^{coup*}, -H^{res*}]] coupling resonant and anti-resonant excitations, where K^{coup}_{(vck),(v'c'k')} has the same kernel elements as H^{res} but with different band index ordering
  • 118.3.3.2 Implement full BSE solver: solve the non-Hermitian eigenvalue problem via Casida-style transformation to a Hermitian problem (H^{res} - K^{coup})(H^{res} + K^{coup}) F = Ω² F, or via direct diagonalization of the full 2N×2N matrix — with proper normalization of eigenvectors under the bosonic metric
  • 118.3.3.3 Implement beyond-TDA corrections to spectra: compute the difference between full BSE and TDA spectra, quantifying the effect of coupling (typically small for absorption but significant for CD spectra, second-harmonic generation, and near band-edge features in low-dimensional systems)

118.3.4 Optical Properties from BSE#

  • 118.3.4.1 Implement optical absorption spectrum: compute Im[ε_M(ω)] = (8π²e²)/(Ωω²) Σ_S |⟨0|v̂|S⟩|² δ(ω-Ω_S) where v̂ is the velocity operator and |S⟩ are BSE eigenstates, with Lorentzian broadening δ(ω) → η/π/(ω²+η²) and proper treatment of the q→0 limit using the velocity gauge (avoiding the dipole-matrix-element divergence)
  • 118.3.4.2 Implement exciton binding energies: E_b = E_gap^{QP} - Ω₁ where Ω₁ is the lowest bright exciton energy, with convergence analysis w.r.t. k-grid density (exciton binding energies converge slowly with k-points, requiring interpolation schemes or extrapolation)
  • 118.3.4.3 Implement exciton wavefunctions and spatial analysis: compute the exciton envelope function |Ψ_S(r_e - r_h)|² by fixing the hole position r_h and plotting the electron probability distribution, extract the exciton radius r_exc = √(⟨|r_e-r_h|²⟩), and classify excitons as Wannier-Mott (r_exc ≫ a₀), Frenkel (r_exc ~ a₀), or charge-transfer type
  • 118.3.4.4 Implement oscillator strengths: f_S = (2m_e Ω_S)/(ℏ²) · |⟨0|r̂|S⟩|² for each exciton state, summing over polarization directions for cubic systems or reporting tensor components for anisotropic crystals
  • 118.3.4.5 Implement radiative lifetimes: τ_S = (3πε₀ℏ⁴c³)/(e²Ω_S³ n_r f_S) where n_r is the refractive index — incorporating both the oscillator strength and the photonic density of states, with temperature-dependent averaging over thermally populated exciton states
  • 118.3.4.6 Implement exciton band structure: solve BSE at finite momentum transfer q to obtain the exciton dispersion Ω_S(q), requiring the q-dependent BSE kernel W^d(q) and proper treatment of the exchange kernel at finite q (which splits longitudinal and transverse excitons at q→0)

118.3.5 Advanced BSE Features#

  • 118.3.5.1 Implement dynamical BSE: replace the static screened interaction W(ω=0) with the frequency-dependent W(ω) in the BSE kernel, requiring self-consistent solution of the BSE with energy-dependent screening — important for systems where dynamical screening significantly renormalizes excitonic effects (e.g., metals, doped semiconductors)
  • 118.3.5.2 Implement electron energy loss spectrum (EELS) from BSE: compute the loss function -Im[1/ε_M(q,ω)] at finite q from the BSE, capturing plasmon excitations and their coupling to interband transitions, with proper inclusion of local field effects for the loss function
  • 118.3.5.3 Implement BSE for 2D materials: combine Coulomb truncation (118.2.8.1) with the BSE kernel, using the truncated W for the direct kernel, with proper treatment of the non-hydrogenic Rydberg series of excitons in 2D (Keldysh potential screening) and convergence acceleration via the Rohlfing-Louie interpolation scheme
  • 118.3.5.4 Implement spin-triplet excitons: solve the BSE with only the direct kernel (no exchange) to obtain triplet exciton energies, compute the singlet-triplet splitting ΔE_{ST} = 2V^x, and analyze the implications for phosphorescence lifetimes and intersystem crossing rates
  • 118.3.5.5 Implement BSE with spin-orbit coupling: extend the BSE kernel to spinor wavefunctions, mixing singlet and triplet exciton channels via SOC, enabling computation of fine structure of excitons in heavy-element semiconductors and topological materials

118.3.6 BSE Interpolation & Convergence#

  • 118.3.6.1 Implement BSE kernel interpolation (Rohlfing-Louie): compute the BSE kernel W^d and V^x on a coarse k-grid, then interpolate to a fine k-grid using Fourier interpolation in real space — W^d(k,k') = Σ_R e^{ik·R} W^d(R,R') e^{-ik'·R'} — reducing the cost of the expensive dielectric matrix calculation while achieving converged exciton energies
  • 118.3.6.2 Implement Wannier-based BSE interpolation: express the BSE Hamiltonian in a Wannier basis, interpolate to arbitrary k-grids via Wannier90-style interpolation, enabling ultra-dense k-grid BSE calculations (100×100×100) from coarse-grid kernel data — following the methodology of Marsili et al.
  • 118.3.6.3 Implement BSE k-point convergence extrapolation: track exciton binding energies and peak positions as a function of k-grid density, fit convergence curves, and extrapolate to the infinite-k limit — particularly critical for 2D materials where convergence is extremely slow (requiring ~100×100 k-grids for meV accuracy)
  • 118.3.6.4 Implement BSE with velocity operator: use the commutator v̂ = i[H,r̂]/ℏ to compute optical matrix elements in the velocity gauge, which is numerically more stable than the length gauge for periodic systems and avoids the position-operator ambiguity in periodic boundary conditions
  • 118.3.6.5 Implement BSE circular dichroism spectrum: compute the rotatory strength R_S = Im[⟨0|μ̂|S⟩ · ⟨S|m̂|0⟩] where μ̂ is the electric dipole and m̂ is the magnetic dipole operator, enabling prediction of circular dichroism (CD) spectra for chiral materials and molecules
  • 118.3.6.6 Implement BSE nonlinear optical response: compute second-order susceptibility χ^(2)(ω₁,ω₂) from BSE eigenstates using the sum-over-states expression, enabling prediction of second-harmonic generation (SHG) coefficients for nonlinear optical materials

118.4 Dynamical Mean-Field Theory (@kalika/dmft)#

118.4.1 DMFT Self-Consistency Loop#

  • 118.4.1.1 Implement the DMFT self-consistency cycle: (1) start with initial self-energy Σ(iω_n), (2) compute local Green's function G_loc(iω_n) = (1/N_k) Σ_k [iω_n + μ - H_k - Σ(iω_n)]⁻¹ via BZ integration, (3) extract Weiss field G₀⁻¹(iω_n) = G_loc⁻¹(iω_n) + Σ(iω_n), (4) solve the impurity problem to get new Σ_imp(iω_n), (5) update Σ(iω_n) = Σ_imp(iω_n) and repeat until ||Σ^{new} - Σ^{old}|| < tolerance
  • 118.4.1.2 Implement chemical potential adjustment: at each DMFT iteration, adjust μ to satisfy the electron count constraint n = -2T Σn G_loc(iω_n) e^{iω_n 0⁺} + n∞ (where n_∞ accounts for the high-frequency tail), using bisection or Brent's method with the density as a function of μ
  • 118.4.1.3 Implement convergence acceleration: Broyden mixing of self-energy between iterations Σ^{new} = αΣ^{out} + (1-α)Σ^{in} with adaptive α, and DIIS (direct inversion of iterative subspace) for faster convergence — typically 10-30 DMFT iterations for convergence
  • 118.4.1.4 Implement multi-orbital DMFT: extend the self-consistency loop to multi-orbital impurity problems with d-orbital subspace (5 orbitals) or f-orbital subspace (7 orbitals), including crystal-field splitting, and the rotationally invariant Slater-Kanamori interaction parametrized by U (Hubbard), J (Hund's coupling), and optionally U' = U-2J

118.4.2 Impurity Solvers — Continuous-Time QMC#

  • 118.4.2.1 Implement CT-HYB (hybridization expansion) solver: expand the partition function in powers of the hybridization function Δ(τ), with Monte Carlo sampling of configurations {(τ₁,a₁), ..., (τk,a_k)} where τ_i are imaginary time points and a_i are orbital indices — the weight of each configuration involves a determinant of the hybridization matrix det[Δ(τ_i - τ'_j)] and the local trace Tr[e^{-βH_loc} d{a₁}(τ₁) d†_{a'₁}(τ'₁)...], using segment representation for density-density interactions or matrix representation for general interactions
  • 118.4.2.2 Implement CT-HYB Monte Carlo moves: insertion/removal of hybridization events (pairs of creation/annihilation operators), segment moves (shift, split, merge) for density-density interactions, global moves (spin flip, orbital swap) for ergodicity in multi-orbital problems, with Metropolis-Hastings acceptance ratios computed from determinant ratios (Sherman-Morrison-Woodbury formula for rank-1 updates)
  • 118.4.2.3 Implement CT-HYB measurement of Green's function: accumulate G_imp(τ) = -⟨Tτ d(τ) d†(0)⟩ from the Monte Carlo sampling by removing one hybridization line (improved estimator) or by direct measurement on the segment picture, then Fourier transform to G_imp(iω_n) with proper high-frequency tail treatment
  • 118.4.2.4 Implement CT-INT (interaction expansion) solver: expand in powers of the interaction U, sampling configurations of interaction vertices at random imaginary times — the weight involves a determinant of the non-interacting Green's function matrix, advantageous for weakly correlated multi-orbital problems where the hybridization expansion converges slowly
  • 118.4.2.5 Implement CT-AUX (auxiliary field) solver: Hirsch-Fye-like decomposition of the interaction via discrete Hubbard-Stratonovich transformation, with continuous-time formulation — sampling auxiliary Ising spin configurations with determinantal weights, offering an alternative to CT-HYB that avoids the local trace calculation

118.4.3 Impurity Solvers — Non-QMC#

  • 118.4.3.1 Implement exact diagonalization (ED) solver: represent the impurity problem with a finite bath of Nb sites (typically 4-8), discretize the hybridization function Δ(iω_n) ≈ Σ{l=1}^{N_b} |V_l|²/(iω_n - ε_l) by fitting bath parameters {V_l, ε_l}, exactly diagonalize the resulting (N_b+1)-site Hamiltonian using Lanczos or Davidson, and compute G_imp from the Lehmann representation
  • 118.4.3.2 Implement NRG (numerical renormalization group) solver: Wilson's NRG with logarithmic discretization of the hybridization function, iterative diagonalization of the Wilson chain, truncation to lowest N_s states at each step, and Anders-Schiller basis for computing spectral functions — particularly accurate for low-energy properties and Kondo physics
  • 118.4.3.3 Implement bath fitting for ED/NRG: optimize bath parameters {V_l, ε_l} to minimize the distance between the model hybridization Δ_model(iω_n) and the DMFT hybridization Δ(iω_n) = iω_n + μ - ε_d - Σ(iω_n) - G_loc⁻¹(iω_n), using conjugate gradient or simulated annealing optimization

118.4.4 DFT+DMFT#

  • 118.4.4.1 Implement DFT+DMFT charge self-consistency: full charge loop where the DMFT density matrix ρ_DMFT is fed back into the DFT Kohn-Sham potential, requiring: (1) compute ρ_DMFT from the DMFT Green's function via Matsubara sum, (2) recompute V_H[ρ_DMFT] + V_xc[ρ_DMFT], (3) re-solve KS equations, (4) update projectors and hybridization function, iterate until both DFT and DMFT are self-consistent
  • 118.4.4.2 Implement projector construction: compute projectors P*{mσ}(k) from Bloch states |ψ*{nk}⟩ to correlated orbitals |φm⟩ (Wannier functions or atomic-like orbitals) — using projective Wannier functions, maximally localized Wannier functions, or PAW projectors — defining the correlated subspace Hamiltonian H^{corr}{mm'}(k) = Σ*n P*{mn}(k) εnk P*{m'n}(k)
  • 118.4.4.3 Implement double-counting corrections: (1) fully localized limit (FLL): E_dc = U·n(n-1)/2 - J·[n↑(n↑-1) + n↓(n↓-1)]/2, (2) around-mean-field (AMF): E_dc = U·n²/4 - J·(n↑² + n↓²)/2, (3) exact double counting (Haule): self-consistently determined from the requirement that the DFT+DMFT total energy is stationary — with analysis of how each choice affects band gaps and spectral weight transfer
  • 118.4.4.4 Implement DFT+DMFT total energy: E_tot = E_DFT[ρ_DMFT] + ⟨H_int⟩_DMFT - E_dc, where ⟨H_int⟩_DMFT is the interaction energy from the impurity solver (computed via Galitskii-Migdal formula or directly from the impurity partition function) — enabling structural optimization and equation-of-state calculations with DMFT

118.4.5 Analytic Continuation#

  • 118.4.5.1 Implement Maximum Entropy (MaxEnt) analytic continuation: continue G(iω_n) or Σ(iω_n) from Matsubara to real frequencies using Bayesian inference — maximize the posterior P[A|G] ∝ P[G|A] · P_prior[A] where the entropy S[A] = -∫ A(ω) ln[A(ω)/D(ω)] dω serves as the prior (D(ω) is the default model, typically flat), with Bryan's algorithm for the optimal regularization parameter α
  • 118.4.5.2 Implement Padé analytic continuation: fit G(iω_n) to a ratio of polynomials P_M(z)/Q_N(z) using Thiele's continued fraction algorithm, then evaluate on the real axis G^R(ω) = G(iω_n → ω+iη) — fast but numerically unstable for noisy QMC data, requiring careful selection of Matsubara points (Beach-Gooding-Marsiglio prescription)
  • 118.4.5.3 Implement Nevanlinna analytic continuation: the Nevanlinna interpolation method that guarantees the analytically continued function satisfies the Herglotz property (positive spectral weight), providing more stable results than Padé for noisy data while being computationally simpler than MaxEnt
  • 118.4.5.4 Implement stochastic analytic continuation (SAC): sample spectral functions A(ω) from the posterior distribution P[A|G] using Monte Carlo, averaging over many samples to reduce noise — particularly useful for resolving sharp features (e.g., quasiparticle peaks) that MaxEnt tends to broaden

118.4.6 Extended DMFT#

  • 118.4.6.1 Implement DMFT+GW (GW+DMFT): combine local correlations from DMFT with nonlocal correlations from GW — use DMFT self-energy Σ_DMFT(ω) for the local part and GW self-energy Σ_GW(k,ω) - Σ_GW^{loc}(ω) for the nonlocal part, with proper subtraction to avoid double counting of local correlations: Σ_tot(k,ω) = Σ_DMFT(ω) + [Σ_GW(k,ω) - Σ_GW^{loc}(ω)]
  • 118.4.6.2 Implement DCA (dynamical cluster approximation): embed a cluster of Nc sites (typically 2×2, 4×4 in 2D) in a self-consistent bath, preserving translational invariance within the cluster — the coarse-grained Green's function G(K,iω_n) = (N_c/N_k) Σ{k∈patch(K)} [iω_n + μ - ε_k - Σ(K,iω_n)]⁻¹ with K = cluster momenta, enabling momentum-dependent self-energy
  • 118.4.6.3 Implement CDMFT (cellular DMFT): embed a real-space cluster in a self-consistent bath, breaking translational invariance (unlike DCA), with the cluster self-energy Σ_{ij}(iω_n) depending on intra-cluster site indices i,j — better for capturing short-range spatial correlations and symmetry-broken phases
  • 118.4.6.4 Implement non-equilibrium DMFT basics: Keldysh contour formulation with lesser/greater/retarded Green's functions on the Kadanoff-Baym contour, time-stepping propagation for quench dynamics and pump-probe responses, with a simple non-equilibrium impurity solver (non-crossing approximation or iterated perturbation theory)

118.4.7 DMFT Physical Observables#

  • 118.4.7.1 Implement DMFT spectral functions: compute A(k,ω) = -(1/π) Im[(ω + μ - ε_k - Σ(k,ω) + iη)⁻¹] on a fine k-path through the BZ, using the analytically continued self-energy Σ(ω), and the momentum-integrated DOS N(ω) = Σ_k A(k,ω)
  • 118.4.7.2 Implement DMFT phase diagram computation: track the quasiparticle weight Z = (1 - ∂ImΣ/∂ωn|{ω_n→0})⁻¹ as a function of U/W (interaction/bandwidth ratio) and temperature, identifying the Mott metal-insulator transition (Z→0), with hysteresis detection for the first-order transition
  • 118.4.7.3 Implement orbital-selective Mott transition (OSMT) detection: for multi-orbital systems, track Z_m per orbital m, identifying regimes where some orbitals are metallic (Z_m > 0) while others are insulating (Z_m → 0) — characterizing the orbital-selective Mott phase driven by crystal-field splitting and Hund's coupling
  • 118.4.7.4 Implement DMFT susceptibilities: compute spin susceptibility χ_spin(q,ω), charge susceptibility χ_charge(q,ω), and orbital susceptibility from the two-particle Green's function using the Bethe-Salpeter equation in the particle-hole channel with the DMFT vertex — detecting magnetic ordering tendencies, charge density wave instabilities, and orbital ordering
  • 118.4.7.5 Implement DMFT optical conductivity: compute σ(ω) from the current-current correlation function using the DMFT self-energy and vertex corrections (important for sum-rule satisfaction), with Drude weight extraction D = πe²n/m* for metals and optical gap identification for insulators

118.4.8 DMFT Interaction Parameters & Workflow#

  • 118.4.8.1 Implement constrained RPA (cRPA) for U and J: compute the partially screened Coulomb interaction U(ω) = W_r(ω) by excluding transitions within the correlated subspace from the polarizability — χ₀^r = χ₀ - χ₀^{dd} where χ₀^{dd} includes only d-d (or f-f) transitions — yielding first-principles values of U, U', and J without adjustable parameters
  • 118.4.8.2 Implement constrained LDA (cLDA) for U: compute U from the total energy difference E(n+1) + E(n-1) - 2E(n) of the correlated subspace under constrained occupations, using the linear-response approach of Cococcioni-de Gironcoli — simpler than cRPA but less accurate for frequency-dependent interactions
  • 118.4.8.3 Implement Hubbard-Kanamori interaction Hamiltonian: construct the full rotationally invariant interaction in the d-orbital subspace Hint = U Σ_m n{m↑}n*{m↓} + U' Σ*{m≠m'} n*{m↑}n*{m'↓} + (U'-J) Σ*{m<m',σ} n*{mσ}n*{m'σ} - J Σ*{m≠m'} d†{m↑}d{m↓}d†{m'↓}d{m'↑} + J Σ*{m≠m'} d†*{m↑}d†{m↓}d{m'↓}d_{m'↑} with spin-flip and pair-hopping terms, parametrized by Slater integrals F⁰, F², F⁴ (for d-electrons)
  • 118.4.8.4 Implement DMFT workflow orchestrator: automated pipeline DFT → Wannier/projectors → hybridization → impurity solver → self-energy → convergence check → analytic continuation → observables, with checkpoint/restart, convergence monitoring, and parameter validation
  • 118.4.8.5 Implement DMFT entropy and free energy: compute the impurity entropy S_imp from the partition function Z_imp or via thermodynamic integration, and the DMFT free energy F = E - TS using the Galitskii-Migdal formula for the internal energy — enabling thermodynamic analysis of phase transitions at finite temperature
  • 118.4.8.6 Implement QMC sign problem diagnostic: monitor the average sign ⟨sign⟩ in CT-QMC calculations as a function of temperature and interaction strength, warn when ⟨sign⟩ < 0.1 (results unreliable), and suggest mitigation strategies (basis rotation, interaction truncation, alternative solver)

118.5 Quantum Monte Carlo (@kalika/qmc)#

118.5.1 Variational Monte Carlo (VMC)#

  • 118.5.1.1 Implement Slater-Jastrow trial wavefunction: Ψ_T(R) = J(R) · det[φ_i(r_j↑)] · det[φ_i(r_j↓)] where det[φ_i(r_j)] is a Slater determinant of single-particle orbitals (from DFT or HF) and J(R) is the Jastrow correlation factor — with efficient evaluation using Sherman-Morrison updates for single-electron moves in the determinant
  • 118.5.1.2 Implement Jastrow factor: J(R) = exp[Σ_{i<j} u(r_ij) + Σ_{iI} χ(r_iI) + Σ_{i<j,I} f(r_iI, r_jI, r_ij)] with electron-electron u(r), electron-nucleus χ(r), and three-body f(r₁,r₂,r₃) terms, parametrized using Drummond-Towler-Needs functions with cusp conditions u'(0) = -1/(2(1+br)) for like-spin and u'(0) = -1/((1+br)) for unlike-spin (satisfying the Kato cusp condition)
  • 118.5.1.3 Implement VMC energy evaluation: EVMC = ⟨Ψ_T|H|Ψ_T⟩/⟨Ψ_T|Ψ_T⟩ = (1/N_s) Σ{i=1}^{N_s} E_L(R_i) where E_L(R) = Ψ_T(R)⁻¹ H Ψ_T(R) is the local energy and R_i are sampled from |Ψ_T|² using Metropolis-Hastings with electron-by-electron moves, drift-diffusion proposals, and acceptance ratio targeting ~50%
  • 118.5.1.4 Implement wavefunction optimization — linear method (Umrigar): minimize EVMC w.r.t. Jastrow parameters {p} using the linearized wavefunction |Ψ(p+δp)⟩ ≈ (1 + Σ_i δp_i O_i)|Ψ(p)⟩ where O_i = ∂ln Ψ/∂p_i, leading to a generalized eigenvalue problem H{ij} cj = E S{ij} cj with H{ij} = ⟨Oi H O_j⟩ and S{ij} = ⟨O_i O_j⟩ — with regularization of S for numerical stability
  • 118.5.1.5 Implement stochastic reconfiguration (SR) optimization: δpi = -Δτ Σ_j S⁻¹{ij} ⟨Oj (E_L - ⟨E_L⟩)⟩ where S{ij} is the overlap matrix of log-derivatives, equivalent to imaginary-time evolution in the variational subspace — with diagonal shift regularization of S and step-size control via energy variance monitoring

118.5.2 Diffusion Monte Carlo (DMC)#

  • 118.5.2.1 Implement DMC algorithm with importance sampling: evolve walkers R(τ) according to the drift-diffusion equation R(τ+δτ) = R(τ) + δτ·v_D(R) + ξ√δτ where v_D = ∇ln|Ψ_T|² is the drift velocity and ξ is a Gaussian random vector, with branching weights w = exp[-δτ(E_L(R) + E_L(R'))/2 + δτ·E_ref] controlling population growth/death
  • 118.5.2.2 Implement fixed-node approximation: constrain walkers to remain within the nodal region defined by Ψ_T(R) > 0 (or < 0), enforcing the node by rejecting moves that cross the nodal surface — the resulting DMC energy E_DMC is a variational upper bound to the true ground state energy, with the fixed-node error depending on the quality of the trial wavefunction nodes
  • 118.5.2.3 Implement population control: maintain a target walker population N_w through branching (integer replication based on weight w), with feedback in E_ref to stabilize the population — E_ref = E_est - γ ln(N_w/N_target) where γ is a damping parameter, plus comb/pair algorithm for weight redistribution to reduce population fluctuations
  • 118.5.2.4 Implement time-step extrapolation: run DMC at multiple time steps δτ (e.g., 0.01, 0.005, 0.002 Ha⁻¹) and extrapolate to δτ→0 using linear or quadratic fitting to remove the O(δτ) time-step bias arising from the short-time approximation to the Green's function
  • 118.5.2.5 Implement size-consistent corrections: compute the size-consistency error of fixed-node DMC for supercells and apply corrections based on the multi-reference character of the wavefunction, using Zen-Sorella-Alfè prescription or the reptation QMC approach for improved estimators
  • 118.5.2.6 Implement pseudopotential handling: locality approximation for non-local pseudopotentials in DMC (Mitáš-Shirley-Ceperley), and the improved T-moves scheme (Casula) that restores variationality by treating the non-local pseudopotential via a stochastic projection — with size-consistent T-moves (CASINO-style) for accurate molecular and solid-state calculations

118.5.3 Auxiliary-Field QMC (AFQMC)#

  • 118.5.3.1 Implement Hubbard-Stratonovich (HS) transformation: decouple the two-body interaction V = Σ*{pqrs} V*{pqrs} a†_p a†_r a_s a_q into a sum over one-body operators coupled to auxiliary fields σ via e^{-δτ V} = ∫ dσ P(σ) e^{σ·B(σ)} where B(σ) is a one-body operator and P(σ) is a Gaussian weight — using either discrete (Hirsch) or continuous (continuous HS) decomposition
  • 118.5.3.2 Implement phaseless AFQMC (Zhang-Krakauer): constrain the auxiliary-field random walk to eliminate the sign/phase problem by imposing the phaseless constraint — walkers with Re[⟨Ψ_T|φ_i⟩ · w_i] < 0 are killed, where w_i is the walker weight, introducing a controlled approximation that depends on the trial wavefunction quality
  • 118.5.3.3 Implement AFQMC propagation: propagate Slater determinant walkers |φ(τ+δτ)⟩ = e^{-δτ(H₁+σ·B)} |φ(τ)⟩ where H₁ is the one-body part, using Trotterization and stochastic sampling of σ — with orthogonalization (modified Gram-Schmidt) every N_ortho steps to maintain numerical stability of the determinant representation
  • 118.5.3.4 Implement back-propagation for observables: compute expectation values ⟨O⟩ = ⟨Ψ_T|e^{-τ_BP H} O e^{-τ_BP H}|φ⟩ / ⟨Ψ_T|e^{-2τ_BP H}|φ⟩ by propagating backward from the measurement point, providing improved estimators for non-diagonal observables (pair correlation function, structure factor, momentum distribution)
  • 118.5.3.5 Implement AFQMC for periodic systems: use plane-wave basis with Bloch orbitals, Ewald summation for the Coulomb interaction with proper handling of the G=0 divergence via Madelung constant subtraction, and twist-averaged boundary conditions for finite-size error reduction

118.5.4 Finite-Size Corrections#

  • 118.5.4.1 Implement twist-averaged boundary conditions (TABC): average QMC energies over a grid of twist angles θ applied to boundary conditions ψ(r+L) = e^{iθ·L} ψ(r), reducing finite-size shell-filling errors by effectively averaging over k-point grids — using either uniform twist grid or optimal twist selection from Dagrada et al.
  • 118.5.4.2 Implement KZK (Kwee-Zhang-Krakauer) correction: estimate the finite-size error in the kinetic energy from the DFT kinetic energy difference between the supercell and the thermodynamic limit, and in the exchange-correlation energy from the structure factor, with the total correction ΔE_KZK = (E_DFT^{∞} - E_DFT^{N}) applied to QMC
  • 118.5.4.3 Implement MPC (model periodic Coulomb) correction: replace the Ewald Coulomb interaction with the MPC interaction that removes the spurious image interaction between the exchange-correlation hole and its periodic images, using the model exchange-correlation hole from the DFT charge density — ΔE_MPC = (1/2)∫∫ ρ_xc(r,r')[v_MPC(r-r') - v_Ewald(r-r')] dr dr'
  • 118.5.4.4 Implement supercell extrapolation: run QMC for multiple supercell sizes (e.g., N, 2N, 4N atoms) and extrapolate to the thermodynamic limit using E(N) = E(∞) + a/N + b/N^{5/3} (for 3D) or E(N) = E(∞) + a/N + b/N^{3/2} (for 2D), with proper error propagation from the stochastic QMC uncertainties

118.5.5 QMC for Solids & Advanced Features#

  • 118.5.5.1 Implement Bloch orbital input processing: read DFT orbitals in plane-wave or Gaussian basis, convert to real-space B-spline representation on a uniform grid for efficient evaluation during QMC random walks — using the einspline library conventions (cubic B-splines with periodic boundary conditions)
  • 118.5.5.2 Implement QMC forces: compute Hellmann-Feynman forces F_I = -∂E/∂R_I plus Pulay corrections arising from the basis-set dependence on atomic positions, using the space-warp transformation (Filippi-Umrigar) for variance reduction of force estimators
  • 118.5.5.3 Implement excited-state QMC: state-specific VMC/DMC for excited states using penalty-function method En = ⟨H⟩ + Σ{i<n} λ_i |⟨Ψ_i|Ψ_n⟩|², or fixed-node DMC with excited-state nodal surfaces from CI or TDDFT calculations, with proper treatment of orthogonality constraints
  • 118.5.5.4 Implement multi-determinant trial wavefunctions: Ψ_T = J(R) · Σ_I c_I D^↑_I D^↓_I where the sum runs over selected determinants from CASSCF, CIPSI, or selected-CI — with efficient evaluation using the table method of Clark-Morales-McMinis for O(N_det · N²) scaling instead of O(N_det · N³)

118.5.6 QMC Statistical Analysis & Workflow#

  • 118.5.6.1 Implement QMC error analysis: compute statistical error bars from correlated Monte Carlo samples using the blocking method (Flyvbjerg- Petersen) — divide the sample into blocks of increasing size, compute the standard error of the block means, and identify the block size where the error plateaus (indicating decorrelation), reporting the autocorrelation time τ_corr
  • 118.5.6.2 Implement reblocking analysis: automated determination of optimal block size via the reblocking procedure, with convergence diagnostic plots showing error vs. block size, and flagging when insufficient samples prevent reliable error estimation (total samples < 10 × τ_corr)
  • 118.5.6.3 Implement QMC observable accumulation: accumulate not only the total energy but also kinetic energy, potential energy (electron-electron, electron-ion, ion-ion), pair correlation function g(r), structure factor S(q), one-body density matrix n(r,r'), and momentum distribution n(k) — with proper error propagation for derived quantities
  • 118.5.6.4 Implement QMC workflow orchestrator: automated pipeline from DFT orbitals → Jastrow optimization (VMC) → DMC production runs → finite-size corrections → time-step extrapolation → final energy with error budget, with checkpoint/restart and parameter validation at each stage
  • 118.5.6.5 Implement DMC mixed estimator correction: for observables that don't commute with H, correct the mixed estimator ⟨Ψ_T|O|Ψ_DMC⟩ using the extrapolated estimator 2⟨O⟩_DMC - ⟨O⟩_VMC, or the more accurate forward-walking/reptation approach for pure estimators
  • 118.5.6.6 Implement QMC pseudopotential testing: validate pseudopotentials by comparing all-electron and pseudopotential QMC results for atoms and small molecules, quantifying the transferability error and identifying pseudopotentials with large localization errors in DMC
  • 118.5.6.7 Implement backflow transformation: enhance the trial wavefunction with backflow coordinates xi = r_i + Σ{j≠i} η(r_ij)(r_i-r_j) where η(r) is an optimizable function, improving the nodal surface beyond the Slater-Jastrow level — particularly important for reducing fixed-node error in strongly correlated systems and achieving ~90% of correlation energy

118.6 Coupled Cluster for Solids (@kalika/cc-solids)#

118.6.1 MP2 for Periodic Systems#

  • 118.6.1.1 Implement canonical MP2 for periodic systems: EMP2 = Σ{ijab,k_i,k_j,k_a} |⟨ij||ab⟩|² / (ε_i + ε_j - ε_a - ε_b) where i,j are occupied Bloch orbitals and a,b are virtual Bloch orbitals, with crystal momentum conservation k_i + k_j = k_a + k_b + G, using the resolution of identity (RI/DF) approximation for two-electron integrals to reduce scaling
  • 118.6.1.2 Implement Laplace-transform MP2 (LT-MP2): transform the energy denominator to imaginary time via 1/(εa+ε_b-ε_i-ε_j) = ∫₀^∞ e^{-(ε_a+ε_b-ε_i-ε_j)τ} dτ, discretized with Nτ quadrature points, enabling O(N⁴) scaling through factorization of the occupied and virtual orbital sums — with minimax quadrature grids for optimal τ-point selection
  • 118.6.1.3 Implement MP2 natural orbitals: diagonalize the MP2 density matrix to obtain natural orbitals and occupation numbers, truncating the virtual space to natural orbitals with occupation > threshold — providing a compact basis for subsequent CCSD calculations and reducing the virtual-space dimension by 50-70%

118.6.2 CCSD for Periodic Systems#

  • 118.6.2.1 Implement periodic CCSD amplitude equations: solve the coupled T₁ and T₂ amplitude equations with crystal momentum conservation, 0 = ⟨Φ^ai|H̄|Φ₀⟩ (singles) and 0 = ⟨Φ^{ab}{ij}|H̄|Φ₀⟩ (doubles) where H̄ = e^{-T}He^T is the similarity-transformed Hamiltonian, using the spin-adapted formulation with restricted HF reference
  • 118.6.2.2 Implement CCSD tensor contraction engine: efficient evaluation of the 8 terms in the doubles equation and 4 terms in the singles equation, with proper index routing for momentum conservation — intermediates W*{mnij}, F*{ae}, F*{mi}, W*{abei}, W_{mbej} following Stanton-Bartlett notation, stored with 3-index k-point dependence
  • 118.6.2.3 Implement CCSD convergence acceleration: DIIS (direct inversion of iterative subspace) extrapolation for T₁ and T₂ amplitudes, with separate DIIS for each k-point block, level shifting for difficult convergence cases, and amplitude damping in early iterations
  • 118.6.2.4 Implement CCSD correlation energy: ECCSD = Σ{ia,k} f*{ia}(k) t^a_i(k) + (1/4) Σ*{ijab,ki,k_j,k_a} ⟨ij||ab⟩ τ^{ab}{ij} where τ^{ab}{ij} = t^{ab}{ij} + t^a_i t^b_j - t^b_i t^a_j, with proper k-point summation weights

118.6.3 CCSD(T) for Periodic Systems#

  • 118.6.3.1 Implement perturbative triples correction (T): E*(T) = (1/36) Σ*{ijkabc} |W*{ijk}^{abc}|² / D*{ijk}^{abc} where W*{ijk}^{abc} = P(ijk)P(abc)[t^{bc}*{jk} ⟨ia||f⟩ + ...] and D_{ijk}^{abc} = ε_i+ε_j+ε_k-ε_a-ε_b-ε_c, with the O(N⁷) scaling handled via batched evaluation over k-point sextuplets (k_i,k_j,k_k,k_a,k_b,k_c) satisfying momentum conservation
  • 118.6.3.2 Implement (T) memory management: the triples amplitudes t^{abc}_{ijk} are never fully stored; instead, compute and contract batches of triples amplitudes on-the-fly, requiring O(N_occ³ · N_virt³) memory per k-point batch rather than the full O(N_occ³ · N_virt³ · N_k⁶) storage
  • 118.6.3.3 Implement distinguishable cluster approximation (DCSD): a simplified variant of CCSD that breaks the antisymmetry of the T₂ amplitudes, achieving near-CCSD(T) accuracy at CCSD cost for many systems — particularly useful for periodic calculations where (T) is prohibitive

118.6.4 EOM-CCSD for Excited States#

  • 118.6.4.1 Implement EOM-CCSD eigenvalue problem: solve H̄R|Φ₀⟩ = ωR|Φ₀⟩ where R = r₀ + Σ*{ia} r^a_i a†_a a_i + (1/4)Σ*{ijab} r^{ab}_{ij} a†_a a†_b a_j a_i is the excitation operator, using Davidson iterative diagonalization for the lowest N_roots eigenvalues of the similarity-transformed Hamiltonian H̄
  • 118.6.4.2 Implement EOM-CCSD sigma vector: efficient computation of σ = H̄R for each trial vector R, reusing the CCSD intermediates (W*{mnij}, F*{ae}, etc.) and implementing the 12+ terms in the sigma equations with proper k-point routing
  • 118.6.4.3 Implement EOM-CCSD properties: transition dipole moments ⟨Ψ₀|μ̂|Ψ_n⟩ = ⟨Φ₀|L_n μ̄ R_n|Φ₀⟩ where L_n is the left eigenvector, oscillator strengths, and excited-state densities for characterization of excitation character (valence, charge-transfer, Rydberg)

118.6.5 Local Correlation & Basis Set Methods#

  • 118.6.5.1 Implement pair natural orbital (PNO) framework for periodic systems: for each occupied pair (i,j), construct the PNO space from the MP2 pair density matrix D^{ij}{ab} = t^{ab}{ij}/(ε_a+ε_b-ε_i-ε_j), truncate to PNOs with occupation > threshold (typically 10⁻⁷), and solve the CC equations in the truncated PNO space — achieving near-linear scaling with system size
  • 118.6.5.2 Implement domain-based local PNO-CCSD(T) (DLPNO-CCSD(T)): assign each occupied orbital i to a domain of atoms based on Mulliken or Löwdin population analysis, restrict pair interactions to spatially close domains (distance-based prescreening), and solve CCSD(T) in the local PNO basis — with three accuracy levels: LoosePNO, NormalPNO, TightPNO corresponding to different truncation thresholds
  • 118.6.5.3 Implement basis set extrapolation to CBS limit: two-point extrapolation E_CBS = (E_X · X³ - E_Y · Y³)/(X³ - Y³) using cc-pVXZ/cc-pVYZ (X<Y) correlation-consistent basis sets, or separate extrapolation of HF (exponential) and correlation (X⁻³) contributions
  • 118.6.5.4 Implement focal-point analysis: combine large-basis lower-level theory with small-basis higher-level theory: E_best ≈ E_MP2/CBS + (E_CCSD/small - E_MP2/small) + (E_CCSD(T)/smaller - E_CCSD/smaller), exploiting the faster basis set convergence of the post-MP2 correlation corrections
  • 118.6.5.5 Implement finite-size corrections for periodic CC: correct for the slow convergence of CC energies with k-point grid density using the structure factor interpolation technique or the transition structure factor approach, analogous to QMC finite-size corrections but adapted to the deterministic CC framework

118.7 Time-Dependent DFT (@kalika/tddft)#

118.7.1 Linear-Response TDDFT#

  • 118.7.1.1 Implement Casida equation: solve the eigenvalue problem [[A, B], [-B*, -A*]] [X, Y]^T = ω [X, Y]^T where A*{ia,jb} = δ*{ij}δ*{ab}(ε_a-ε_i) + (ia|f_Hxc|jb) and B*{ia,jb} = (ia|f_Hxc|bj), with the Hartree-exchange-correlation kernel f_Hxc(r,r',ω) = v(r,r') + f_xc(r,r',ω) — in the Tamm-Dancoff approximation (B=0) and full form
  • 118.7.1.2 Implement adiabatic LDA kernel: fxc^{ALDA}(r,r') = δ(r-r') · d²e_xc^{LDA}/dn² |{n=n(r)} where e_xc^{LDA} is the LDA exchange-correlation energy density — the simplest frequency-independent kernel, exact for the uniform electron gas
  • 118.7.1.3 Implement adiabatic GGA kernels: f_xc^{AGGA}(r,r') derived from PBE, BLYP, or other GGA functionals, including gradient terms ∂²e_xc/∂n², ∂²e_xc/∂n∂|∇n|, ∂²e_xc/∂|∇n|² with proper treatment of the second functional derivative of the gradient-dependent terms
  • 118.7.1.4 Implement hybrid TDDFT kernels: for hybrid functionals (B3LYP, PBE0, HSE), include the fraction of exact exchange in the kernel f_Hxc = (1-α)f_xc^{semi-local} + α·v(r,r') where α is the exact exchange fraction — critical for charge-transfer excitations where semi-local kernels fail qualitatively

118.7.2 Long-Range & Improved Kernels#

  • 118.7.2.1 Implement bootstrap kernel (Sharma et al.): a self-consistent xc kernel derived from the condition that the macroscopic dielectric function εM(ω=0) from TDDFT equals the RPA value — iterate f_xc(G,G') = ε⁻¹{G=0,G'=0} · v(G) · [1 - ε⁻¹(G,G')]/[χ₀(G,G') · v(G')] until convergence, producing a long-range kernel that captures excitonic effects without BSE
  • 118.7.2.2 Implement long-range corrected (LRC) kernel: f_xc(q) = -α/q² where α is a material-dependent parameter fitted to reproduce BSE optical spectra or exciton binding energies — the simplest kernel that captures bound excitons in TDDFT, following Reining et al.
  • 118.7.2.3 Implement RPA+ kernel: f_xc = f_xc^{ALDA} + Δf_xc^{GW} combining the local ALDA kernel with a nonlocal correction derived from the GW self-energy via the Bethe-Salpeter equation, providing a systematically improvable TDDFT kernel hierarchy

118.7.3 Real-Time TDDFT#

  • 118.7.3.1 Implement real-time TDDFT propagation (RT-TDDFT): solve iℏ ∂ψ_n(r,t)/∂t = H_KS[n(r,t)] ψ_n(r,t) using time propagators — enforced time-reversal symmetry (ETRS): ψ(t+δt) = e^{-iH(t+δt/2)δt/ℏ} ψ(t) with predictor-corrector for the midpoint Hamiltonian, or Crank-Nicolson: ψ(t+δt) = (1 + iHδt/2ℏ)⁻¹(1 - iHδt/2ℏ) ψ(t)
  • 118.7.3.2 Implement delta-kick excitation: apply an instantaneous perturbation ψ_n(r,t=0⁺) = e^{iκ·r} ψ_n(r,t=0⁻) with small κ, then propagate and extract the optical absorption spectrum from the Fourier transform of the time-dependent dipole moment σ(ω) ∝ FT[d(t)]/κ where d(t) = ∫ r · n(r,t) dr — valid in the linear-response regime for small κ
  • 118.7.3.3 Implement RT-TDDFT for periodic systems: propagate Bloch orbitals ψ*{nk}(r,t) on a k-point grid, computing the time-dependent current j(t) = (e/m) Σ*{nk} f*{nk} ⟨ψ*{nk}(t)|p̂+ℏk|ψ_{nk}(t)⟩ for optical response, with absorbing boundary conditions or mask functions for finite systems
  • 118.7.3.4 Implement Ehrenfest molecular dynamics from RT-TDDFT: couple the electronic RT-TDDFT propagation to classical nuclear dynamics via F_I(t) = -∂E[n(r,t)]/∂R_I, propagating ions with velocity Verlet and electrons with ETRS on the same time grid — for modeling photo-excited dynamics, charge transfer, and stopping power

118.7.4 TDDFT vs. BSE Comparison Tools#

  • 118.7.4.1 Implement automated TDDFT-BSE comparison pipeline: run both TDDFT (with various kernels: ALDA, bootstrap, LRC) and BSE on the same system, compare optical spectra (peak positions, oscillator strengths, lineshapes), and compute deviation metrics (peak position error, spectral overlap integral, exciton binding energy error)
  • 118.7.4.2 Implement kernel assessment tool: for a given material, evaluate which TDDFT kernel best reproduces BSE results, outputting a ranking of kernels by accuracy with statistical uncertainty — guiding kernel selection for large-scale TDDFT calculations where BSE is too expensive

118.8 Embedding Methods (@kalika/embedding)#

118.8.1 DFT Embedding#

  • 118.8.1.1 Implement frozen density embedding (FDE): partition the total system into subsystems A and B, solve KS equations for subsystem A with an embedding potential v_emb(r) = v_Hn_B + v_xcn_A+n_B - v_xcn_A + v_T^{nadd}n_A,n_B where v_T^{nadd} is the non-additive kinetic potential — iterate to self-consistency between subsystems using freeze-and-thaw cycles
  • 118.8.1.2 Implement kinetic energy functionals for embedding: Thomas-Fermi T_TF[n] = C_F ∫ n^{5/3} dr, von Weizsäcker T_vW[n] = (1/8)∫ |∇n|²/n dr, and GGA kinetic functionals (LC94, revAPBEk) for the non-additive kinetic potential — with assessment of their accuracy for different embedding scenarios (covalent vs. non-covalent subsystem boundaries)
  • 118.8.1.3 Implement subsystem DFT for large systems: partition a periodic solid into molecular or atomic fragments, solve each fragment self-consistently with embedding potential from the frozen densities of all other fragments, achieving linear-scaling DFT for weakly interacting subsystems (molecular crystals, surfaces with adsorbates)

118.8.2 QM/MM for Materials#

  • 118.8.2.1 Implement QM/MM interface: define QM and MM regions in a periodic solid, compute forces from DFT in the QM region and from classical force fields (Lennard-Jones, Buckingham, Coulomb with Ewald) in the MM region, with electrostatic embedding (MM point charges enter the QM Hamiltonian) and mechanical embedding (QM-MM interaction via force field only)
  • 118.8.2.2 Implement QM/MM boundary treatment: hydrogen link atoms or generalized hybrid orbitals (GHO) for covalent bonds crossing the QM/MM boundary, charge shifting/redistribution to avoid overpolarization near the boundary, and adaptive QM/MM with dynamic resizing of the QM region
  • 118.8.2.3 Implement ONIOM-style multi-layer embedding: three-layer (high:medium:low) partitioning with DFT:DFTB:FF or CCSD(T):DFT:FF combinations, subtractive scheme E_ONIOM = E_high(model) + E_low(real) - E_low(model) for two layers, with proper link-atom corrections at each boundary

118.8.3 Many-Body Embedding#

  • 118.8.3.1 Implement GW embedding: solve the GW equations for a subsystem (defect site, surface adsorbate) with the screened Coulomb interaction W computed from the full periodic system — the defect self-energy Σ_GW includes screening from the bulk environment while the expensive GW calculation is restricted to a small active space
  • 118.8.3.2 Implement DMFT embedding: define a correlated impurity region embedded in a DFT bath, computing the hybridization function from the full periodic DFT band structure and solving the impurity problem only for the correlated atoms — standard DFT+DMFT is already a form of embedding, but here extended to cluster impurity problems for multi-site correlations
  • 118.8.3.3 Implement projection-based embedding (Manby-Miller): define the embedding by a projector P onto the active subsystem, with the embedding potential μ^{proj}(r,r') constructed to enforce orthogonality between subsystem orbitals — enabling WFT-in-DFT (e.g., CCSD(T)-in-DFT) where the active region is treated with coupled cluster and the environment with DFT, with exact embedding in the complete basis limit

118.8.4 Defect Embedding#

  • 118.8.4.1 Implement quantum defect embedding: embed a defect cluster (defect + nearest neighbors) in the periodic DFT host crystal, using Green's function embedding G_defect(ω) = [G₀⁻¹(ω) - V_pert]⁻¹ where G₀ is the bulk Green's function and V_pert is the defect perturbation potential — avoiding supercell artifacts for dilute defects
  • 118.8.4.2 Implement charged defect corrections: Freysoldt-Neugebauer-Van de Walle (FNV) correction for charged defects in periodic supercells, accounting for the spurious electrostatic interaction between periodic images: ΔE_corr = E_isolated^{Madelung} - E_periodic^{Madelung} + alignment term — with the Kumagai-Oba extension for anisotropic dielectric screening
  • 118.8.4.3 Implement defect formation energy workflow: E_f[X^q] = E_tot[X^q] - E_tot[bulk] - Σ_i n_i μ_i + q(ε_VBM + E_F) + E_corr, computing formation energies as a function of Fermi level and chemical potentials, with thermodynamic transition levels ε(q/q') where E_f[X^q] = E_f[X^{q'}]

118.9 Response Functions & Dielectric Properties (@kalika/response)#

118.9.1 Density-Density Response#

  • 118.9.1.1 Implement density-density response function χ(q,ω): the full (interacting) response function satisfying the Dyson equation χ = χ₀ + χ₀ · (v + f_xc) · χ, where χ₀ is the non-interacting KS response (Adler-Wiser), v is the Coulomb kernel, and f_xc is the exchange-correlation kernel — computed by matrix inversion at each (q,ω)
  • 118.9.1.2 Implement static response χ(q,ω=0): the static limit relevant for screening of point charges, with proper treatment of metallic systems (Lindhard function) including the Kohn anomaly at q = 2k_F, and insulating systems (finite dielectric constant)
  • 118.9.1.3 Implement charge susceptibility from response: compute the static charge susceptibility χ_c = ∂n/∂μ from the q→0 limit of χ(q,ω=0), identifying metallic (χ_c > 0) vs. insulating (χ_c = 0) behavior, and compressibility κ = (1/n²)χ_c

118.9.2 Dielectric Properties#

  • 118.9.2.1 Implement macroscopic dielectric function: εM(q→0,ω) = 1/ε⁻¹{G=0,G'=0}(q→0,ω) including local field effects (off-diagonal elements of the dielectric matrix), with distinction between optical (electronic only, clamped ions) and static (electronic + ionic) dielectric constants
  • 118.9.2.2 Implement local field effects analysis: compute the difference between ε with and without local field effects (LFE), quantifying the impact of microscopic charge redistribution — particularly important for systems with inhomogeneous electron density (surfaces, nanostructures, layered materials)
  • 118.9.2.3 Implement dielectric tensor for anisotropic systems: compute the full 3×3 dielectric tensor ε_{αβ}(ω) for non-cubic crystals, identifying the ordinary and extraordinary dielectric constants, optical axes, and birefringence Δn = n_e - n_o

118.9.3 Sum Rules & Consistency Checks#

  • 118.9.3.1 Implement f-sum rule verification: check ∫₀^∞ ω Im[ε(ω)] dω = (π/2) ω_p² where ω_p = √(4πne²/m) is the plasma frequency — deviation from the exact sum rule quantifies basis set incompleteness, truncation errors, and numerical inaccuracies
  • 118.9.3.2 Implement Kramers-Kronig consistency check: verify that Re[ε(ω)] and Im[ε(ω)] satisfy the Kramers-Kronig relations Re[ε(ω)] - 1 = (2/π) P∫₀^∞ ω'Im[ε(ω')]/(ω'²-ω²) dω', with the deviation serving as a measure of frequency-grid completeness
  • 118.9.3.3 Implement conductivity sum rule: verify ∫₀^∞ Re[σ(ω)] dω = (π/2)(ne²/m) for the optical conductivity, and extract the Drude weight D = πe²n/m* for metals from the zero-frequency peak

118.9.4 Derived Optical Properties#

  • 118.9.4.1 Implement electron energy loss function: compute -Im[1/ε_M(q,ω)] from the macroscopic dielectric function, identifying plasmon resonances (peaks in the loss function), their dispersion with q, and damping (width of plasmon peaks from Landau damping at large q)
  • 118.9.4.2 Implement reflectivity and optical constants: compute the complex refractive index ñ(ω) = n(ω) + ik(ω) = √ε(ω), reflectivity R(ω) = |(ñ-1)/(ñ+1)|², absorption coefficient α(ω) = 2ωk(ω)/c, and the optical conductivity σ(ω) = -iω[ε(ω)-1]/(4π) — all derived from the computed ε(ω)
  • 118.9.4.3 Implement joint density of states (JDOS): compute J(ω) = Σ*{v,c,k} δ(ε*{ck} - ε_{vk} - ω) as the baseline for optical absorption (without matrix elements), identifying van Hove singularities and critical points in the band structure that produce features in optical spectra

118.10 Benchmarking & Validation (@kalika/many-body-benchmarks)#

118.10.1 GW Benchmarks#

  • 118.10.1.1 Implement GW100 molecular benchmark: ionization potentials of 100 molecules (HOMO energies) from G₀W₀@PBE, compared against CCSD(T) reference values from the van Setten et al. GW100 dataset — compute mean absolute error (MAE), max error, and correlation plot, targeting MAE < 0.3 eV matching BerkeleyGW/VASP results
  • 118.10.1.2 Implement semiconductor band gap benchmarks: compute quasiparticle band gaps for Si (1.17 eV), Ge (0.74 eV), GaAs (1.52 eV), GaN (3.50 eV), ZnO (3.44 eV), MgO (7.83 eV), diamond (5.48 eV), BN (6.36 eV), and LiF (14.2 eV) at the G₀W₀@PBE level, comparing with experimental values and published GW results — with convergence analysis w.r.t. number of bands, energy cutoff, and k-point grid
  • 118.10.1.3 Implement GW self-consistency benchmark: compare G₀W₀, GW₀, scGW, and QSGW band gaps for the same set of semiconductors, quantifying the effect of self-consistency and identifying the optimal level for each material class (sp semiconductors, d-electron compounds, f-electron systems)
  • 118.10.1.4 Implement strongly correlated system GW benchmark: test GW for transition-metal oxides NiO, MnO, FeO, CoO (antiferromagnetic insulators where DFT fails), comparing GW band gaps with experiment and identifying cases where GW alone is insufficient (requiring DMFT)

118.10.2 BSE Benchmarks#

  • 118.10.2.1 Implement BSE optical spectra for prototypical semiconductors: compute absorption spectra of Si, GaAs, and GaN, comparing peak positions and oscillator strengths against experiment — verifying E₁ and E₂ peak positions within 0.1 eV and correct line shapes
  • 118.10.2.2 Implement BSE exciton binding energy benchmarks: compute exciton binding energies for LiF (1.6 eV), MgO (0.6 eV), BN (0.7 eV), Si (15 meV), GaAs (4 meV), and monolayer MoS₂ (0.5 eV), comparing with experimental and published BSE values — testing both bulk 3D and 2D material cases
  • 118.10.2.3 Implement BSE convergence benchmark: systematic convergence study of BSE exciton energies w.r.t. k-point grid density, number of bands in the BSE kernel, and dielectric matrix size (G-vector cutoff), for Si and LiF as representative cases — documenting the convergence rates and establishing recommended parameters

118.10.3 DMFT Benchmarks#

  • 118.10.3.1 Implement Hubbard model Mott transition benchmark: compute the phase diagram of the single-band Hubbard model on the Bethe lattice (semi-circular DOS with half-bandwidth D) using DMFT+CT-HYB, reproducing the first-order Mott transition at U_c ≈ 1.5W, the coexistence region, and the critical endpoint temperature T_c ≈ 0.02W
  • 118.10.3.2 Implement SrVO₃ benchmark: compute the spectral function A(k,ω) for SrVO₃ (prototypical correlated metal) using DFT+DMFT with U ≈ 5 eV and J ≈ 0.7 eV, reproducing the lower and upper Hubbard bands at ~-1.5 eV and ~2.5 eV flanking the quasiparticle peak, comparing with ARPES data
  • 118.10.3.3 Implement NiO benchmark: compute the band gap of NiO (antiferromagnetic charge-transfer insulator, experimental gap ~4.3 eV) using DFT+DMFT, comparing with DFT (incorrect metallic), DFT+U (gap depends sensitively on U), and GW (underestimates gap) results
  • 118.10.3.4 Implement multi-orbital benchmark: compute the orbital-selective Mott transition in a two-orbital model with different bandwidths and Hund's coupling J, reproducing the phase diagram with metallic, orbital-selective Mott, and Mott insulating phases — validating the multi-orbital DMFT implementation

118.10.4 QMC Benchmarks#

  • 118.10.4.1 Implement homogeneous electron gas (HEG) benchmark: compute the total energy per electron of the 3D HEG at r_s = 1, 2, 3, 4, 5, 10, 20, 50, 100 using VMC and DMC, comparing with the Ceperley-Alder parametrization — the fundamental benchmark for any QMC code, targeting agreement within 1 mHa/electron
  • 118.10.4.2 Implement solid benchmark: compute the cohesive energy, lattice constant, and bulk modulus of Si, diamond, and NaCl using DMC, comparing with experiment — targeting lattice constant within 0.5%, cohesive energy within 0.05 eV/atom, and bulk modulus within 5%
  • 118.10.4.3 Implement finite-size correction benchmark: for Si at multiple supercell sizes (8, 16, 32, 64 atoms), compare raw DMC energies with TABC-corrected, KZK-corrected, and MPC-corrected values, verifying that corrections reduce the finite-size error from ~50 meV/atom to < 5 meV/atom

118.10.5 Coupled Cluster Benchmarks#

  • 118.10.5.1 Implement periodic MP2 benchmark: compute MP2 correlation energies for Si, diamond, and LiH in the thermodynamic limit (CBS extrapolated), comparing with published values from PySCF/CC4S and verifying O(N⁵) scaling behavior
  • 118.10.5.2 Implement periodic CCSD benchmark: compute CCSD correlation energies for diamond and BN, comparing with published values and verifying that CCSD corrects the MP2 overbinding, with convergence analysis w.r.t. basis set and k-point grid
  • 118.10.5.3 Implement CCSD(T) benchmark: compute the binding energy of diamond at the CCSD(T)/CBS level, comparing with the experimental cohesive energy of 7.37 eV/atom — the gold standard for weakly correlated solid-state systems

118.10.6 Cross-Method Validation#

  • 118.10.6.1 Implement DFT→G₀W₀→BSE pipeline validation: for Si, run the full pipeline from DFT ground state through G₀W₀ quasiparticle corrections to BSE optical spectrum, verifying that the indirect gap (1.17 eV), direct gap at Γ (3.4 eV), E₁ peak (3.4 eV), and E₂ peak (4.3 eV) are all reproduced within 0.1 eV of experiment
  • 118.10.6.2 Implement method comparison table: for a common set of 10 materials (Si, Ge, GaAs, ZnO, MgO, NiO, SrVO₃, LiF, diamond, MoS₂), compute band gaps from DFT-PBE, DFT+U, G₀W₀, scGW, DMFT (where applicable), and DMC (where feasible), presenting a comprehensive comparison table with experimental references and method-specific error statistics
  • 118.10.6.3 Implement automated regression test suite: for each method (GW, BSE, DMFT, QMC, CC), define a set of small benchmark calculations (< 5 min each) that run as part of the CI/CD pipeline, with reference values stored in the test fixtures and tolerance thresholds for pass/fail determination — catching regressions in numerical accuracy from code changes
  • 118.10.6.4 Implement performance benchmark suite: measure wall-clock time and peak memory for standard benchmark systems at increasing sizes (8, 16, 32, 64 atoms for QMC; 2×2×2, 4×4×4, 6×6×6 k-grids for GW/BSE), verifying the expected asymptotic scaling (O(N³) for GW space-time, O(N⁴) for canonical GW, O(N⁶) for CCSD, O(N⁷) for CCSD(T)) and identifying performance bottlenecks