Table of Contents
Fetching ...

General-order degenerate coupled-cluster theory

So Hirata

TL;DR

This work introduces ΔCC, a general-order, degenerate extension of coupled-cluster theory that remains size-extensive and black-box while handling both degenerate and nondegenerate references. By formulating an internal M-by-M reference space with a C-condition, it unifies the treatment of excited, ionized, and electron-attached states under a single framework, recovering full FCI in the limit. ΔCC$(S)$ and ΔCC$(SD)$ provide highly accurate energies that, in many benchmark cases, surpass EOM-CC in accuracy and reliability, especially for multi-electron excitations and challenging degenerate references. The authors develop determinant-based and algebraic implementations and demonstrate that ΔCC offers a viable, complementary alternative to MBGF and traditional EOM-CC methods with promising future extensions to derivatives, higher-order corrections, and periodic or finite-temperature regimes.

Abstract

A size-extensive, converging, black-box coupled-cluster ($Δ$CC) ansatz is introduced that computes the energies and wave functions of stationary states from any degenerate or nondegenerate Slater-determinant references with any numbers of $α$- and $β$-spin electrons, any patterns of orbital occupancy, any spin multiplicities, and any spatial symmetries. For a nondegenerate reference, it is identical to the single-reference, projection coupled-cluster ansatz. For degenerate references, it is a natural coupled-cluster extension of degenerate RS perturbation ($Δ$MP) theory, and is closely related to, but distinct from Li and Paldus's state-universal multireference coupled-cluster (SUMRCC) theory. For ionized and electron-attached references, it can be viewed as a coupled-cluster Green's function, although the present theory is convergent toward the full-configuration-interaction (FCI) limits, while many-body Green's function (MBGF) theory generally is not. Its single-excitation instance is a projection (nonvariational) Hartree-Fock theory for a degenerate or nondegenerate reference as per the Thouless theorem, whose practical utility seems rather limited except for core ionizations, high-spin states, and possibly electron affinities. A determinant-based, general-order algorithm is implemented, generating $Δ$CC energies through connected octuple excitations, which are compared with the results from CI, equation-of-motion coupled-cluster (EOM-CC), and SUMRCC theories up to the FCI limits as well as from $Δ$MP and MBGF theories up to the 19th order. An algebraic, optimal-scaling, order-by-order algorithm is also computer-synthesized at the levels of single excitations only and of single and double excitations. The order of performance is: $Δ$CC $>$ EOM-CC $>$ CI at the same order or $Δ$CC $>$ $Δ$MP $>$ MBGF at the same cost scaling.

General-order degenerate coupled-cluster theory

TL;DR

This work introduces ΔCC, a general-order, degenerate extension of coupled-cluster theory that remains size-extensive and black-box while handling both degenerate and nondegenerate references. By formulating an internal M-by-M reference space with a C-condition, it unifies the treatment of excited, ionized, and electron-attached states under a single framework, recovering full FCI in the limit. ΔCC and ΔCC provide highly accurate energies that, in many benchmark cases, surpass EOM-CC in accuracy and reliability, especially for multi-electron excitations and challenging degenerate references. The authors develop determinant-based and algebraic implementations and demonstrate that ΔCC offers a viable, complementary alternative to MBGF and traditional EOM-CC methods with promising future extensions to derivatives, higher-order corrections, and periodic or finite-temperature regimes.

Abstract

A size-extensive, converging, black-box coupled-cluster (CC) ansatz is introduced that computes the energies and wave functions of stationary states from any degenerate or nondegenerate Slater-determinant references with any numbers of - and -spin electrons, any patterns of orbital occupancy, any spin multiplicities, and any spatial symmetries. For a nondegenerate reference, it is identical to the single-reference, projection coupled-cluster ansatz. For degenerate references, it is a natural coupled-cluster extension of degenerate RS perturbation (MP) theory, and is closely related to, but distinct from Li and Paldus's state-universal multireference coupled-cluster (SUMRCC) theory. For ionized and electron-attached references, it can be viewed as a coupled-cluster Green's function, although the present theory is convergent toward the full-configuration-interaction (FCI) limits, while many-body Green's function (MBGF) theory generally is not. Its single-excitation instance is a projection (nonvariational) Hartree-Fock theory for a degenerate or nondegenerate reference as per the Thouless theorem, whose practical utility seems rather limited except for core ionizations, high-spin states, and possibly electron affinities. A determinant-based, general-order algorithm is implemented, generating CC energies through connected octuple excitations, which are compared with the results from CI, equation-of-motion coupled-cluster (EOM-CC), and SUMRCC theories up to the FCI limits as well as from MP and MBGF theories up to the 19th order. An algebraic, optimal-scaling, order-by-order algorithm is also computer-synthesized at the levels of single excitations only and of single and double excitations. The order of performance is: CC EOM-CC CI at the same order or CC MP MBGF at the same cost scaling.
Paper Structure (29 sections, 68 equations, 39 figures)

This paper contains 29 sections, 68 equations, 39 figures.

Figures (39)

  • Figure 1: The diagonal energy, $H_{II}$, of $\Delta$CCSD [Eq. (\ref{['DCCSD3']})]. The dashed line with circled $I$ denotes $(f_I)^{\text{out}}_{\text{in}}$, while the dashed line translates to $(v)^{\text{left out, right out}}_{\text{left in, right in}}$. The solid line with $I$ designates $(t_I)^{\text{out}}_{\text{in}}$ or $(t_I)^{\text{left out, right out}}_{\text{left in, right in}}$.
  • Figure 2: The left-hand side, $\langle I^{a}_{i} |\hat{H}_Ie^{\hat{T}_I}|I\rangle$, of the $t$-amplitude equation of $\Delta$CCSD [Eq. (\ref{['DCCSD2']})] in the external single-excitation space ($\epsilon_a -\epsilon_i \neq 0$). The double line with $I$ ( 2a) designates the CC effective Hamiltonian $(\hat{H}_I e^{\hat{T}_I})^\text{out}_\text{in}$, which is unlinked through diagram 2h.
  • Figure 3: The left-hand side, $\langle I^{a}_{i} |\hat{H}_Ie^{\hat{T}_I}|I\rangle$, of the $t$-amplitude equation of $\Delta$CCSD [Eq. (\ref{['DCCSD2']})] in the internal single-excitation space, i.e., the off-diagonal energy, $H_{JI}$, with $|J\rangle = |I^a_i\rangle$. The red oblong denotes the fictitious resolvent line that demands $\epsilon_a -\epsilon_i = 0$. The double line with $I$ ( 3a) now designates the connected CC effective Hamiltonian $(\hat{H}_I e^{\hat{T}_I})^\text{out}_\text{in}$. See Fig. 10.2 of Shavitt and Bartlettshavitt for a complete list of diagrams.
  • Figure 4: The C-condition, $S_{JI}=0$, of $\Delta$CCSD [Eq. (\ref{['DCCSD5']})], where $|J\rangle = |I^a_i\rangle$ with $\epsilon_a -\epsilon_i = 0$.
  • Figure 5: The left-hand side, $\langle I^{ab}_{ij} |\hat{H}_Ie^{\hat{T}_I}|I\rangle$, of the $t$-amplitude equation of $\Delta$CCSD [Eq. (\ref{['DCCSD1']})] in the external double-excitation space ($\epsilon_a +\epsilon_b -\epsilon_i -\epsilon_j \neq 0$). The double line with $I$ ( 5a) designates the CC effective Hamiltonian $(\hat{H}_I e^{\hat{T}_I})^\text{left out, right out}_\text{left in, right in}$, which is unlinked through diagrams 5i and 5j.
  • ...and 34 more figures