Direct Ring Coupled-Cluster Doubles (drCCD)

• drCCD is a ring-truncated coupled-cluster doubles approach that retains only the direct, non-antisymmetrized particle-hole ring diagrams to yield the dRPA ground-state correlation energy.
• It bridges coupled-cluster, response theory, and Green’s-function formalism and is applied to electronic correlations in molecules, solids, and light-matter interactions.
• The method faces numerical challenges like solution multiplicity and finite-size errors, which are addressed by using preconditioners and staggered k-point meshes.

Searching arXiv for the cited papers on drCCD, RPA, and related developments. Direct ring coupled-cluster doubles (drCCD) is a ring-truncated coupled-cluster doubles approximation in which only the direct, non-antisymmetrized particle-hole ring terms are retained in the doubles amplitude equations. In electronic-structure practice, drCCD is the coupled-cluster realization of direct random phase approximation (dRPA): it sums direct ring diagrams to infinite order, provides the RPA ground-state correlation energy in a wavefunction parametrization, and serves as a bridge between coupled-cluster, response-theoretic, and Green’s-function formalisms (Shepherd et al., 2013, Berkelbach, 2018).

1. Definition and algebraic form

In the channel decomposition of the CCD doubles equations, the residual is partitioned into a driver, rings, crossed-rings, ladders, and mosaics. The schematic structure is

$$0=\textrm{driver}+\textrm{rings}+\textrm{crossed-rings}+\textrm{ladders}+\textrm{mosaics},$$

and the ring-only truncation retains only the driver and ring terms (Shepherd et al., 2013).

For the ring-only channel, the amplitude equation is written as

$\begin{split} 0 =\ & \bar{v}_{ij}^{ab} + (\epsilon_a+\epsilon_b-\epsilon_i-\epsilon_j)t_{ij}^{ab}\ &+ \bar{v}_{cj}^{kb}\, t_{ik}^{ac} + \bar{v}_{ci}^{ka}\, t_{jk}^{bc} + \bar{v}_{cd}^{kl}\, t_{lj}^{db} t_{ik}^{ac}, \end{split}$

with antisymmetrized two-electron integrals $\bar v$. In drCCD, the integrals are replaced by their direct, non-antisymmetrized parts (Shepherd et al., 2013). In matrix notation, the same structure appears as the Riccati equation

$R(T)=B^*+A^*T+TA+TBT=0,$

and the corresponding RPA correlation energy is

$$E_c^{\mathrm{RPA}}=E_c^{\mathrm{drCCD}}(T)=\frac{1}{2}\mathrm{Tr}(BT),$$

with $T=YX^{-1}$ when expressed through the RPA eigenvectors $X$ and $Y$ (Song et al., 14 Aug 2025).

This formulation preserves the exponential coupled-cluster ansatz while restricting the diagrammatic content to direct ring contractions. In closed-shell notation, the same direct-ring structure leads to the Riccati equation

$${}^1\mathbf{K}+{}^1\mathbf{L}\,{}^1\mathbf{T}+{}^1\mathbf{T}\,{}^1\mathbf{L}+{}^1\mathbf{T}\,{}^1\mathbf{K}\,{}^1\mathbf{T}=0$$

and to the dRPA energy expression

$$E_\mathrm{c}^\mathrm{dRPA}=\frac{1}{2}\mathrm{Tr}[{}^1\mathbf{K}\,{}^1\mathbf{T}],$$

which are standard ring-CCD forms for closed-shell systems (Toulouse et al., 2011).

2. Diagrammatic content and relation to RPA

The defining approximation in drCCD is the restriction to time-independent ring diagrams in the Goldstone sense. This is the subset of CCD terms that generates the direct particle-hole RPA series, and the equivalence between RPA and ring-CCD is exact for the ground-state correlation energy when only the direct terms are retained (Berkelbach, 2018).

Within the CCD channel language, drCCD excludes ladders, crossed-rings, and mosaics. Ladders represent particle-particle and hole-hole scattering, crossed-rings are exchange diagrams needed for proper antisymmetry, and mosaics generate effective one-body terms that renormalize single-particle energies (Shepherd et al., 2013). The direct approximation therefore omits exchange by construction. When antisymmetrization is retained in ring-only CCD, one obtains an rCCD variant rather than drCCD; the paper on the homogeneous electron gas explicitly distinguishes the direct and antisymmetrized cases and notes that only the direct, non-exchange ring summation is convergent in the thermodynamic limit (Shepherd et al., 2013).

The relation to standard RPA nomenclature is correspondingly precise. drCCD is equivalent to dRPA for the amplitudes, and if the energy expression is modified by retaining only the direct part with a factor of $\begin{split} 0 =\ & \bar{v}_{ij}^{ab} + (\epsilon_a+\epsilon_b-\epsilon_i-\epsilon_j)t_{ij}^{ab}\ &+ \bar{v}_{cj}^{kb}\, t_{ik}^{ac} + \bar{v}_{ci}^{ka}\, t_{jk}^{bc} + \bar{v}_{cd}^{kl}\, t_{lj}^{db} t_{ik}^{ac}, \end{split}$0, the result is the dRPA+SOSEX energy (Shepherd et al., 2013). In closed-shell range-separated formulations, several exchange-inclusive ring variants arise from different contractions of ring amplitudes with two-electron integrals, including RPAx-II, RPAx-SO1, and RPAx-SO2; these are ring-CCD-derived approximations rather than direct-ring ones (Toulouse et al., 2011).

Because the retained diagrams are particle-hole rings, drCCD emphasizes screening and collective particle-hole motion. The range-separated Brueckner CCD analysis states that the ring terms together with the driving terms yield the particle-hole RPA and “loosely, they treat electronic excitations as bosonic harmonic oscillators” (Shepherd et al., 2013).

3. Homogeneous electron gas and thermodynamic-limit behavior

The homogeneous electron gas provides a stringent benchmark for drCCD because metallic screening, infrared behavior, and the thermodynamic limit can all be probed explicitly. In finite 14- and 54-electron systems at the complete basis set limit over a wide density range, rings generally overcorrelate, ladders generally undercorrelate, and full CCD gives the closest agreement with Quantum Monte Carlo benchmarks (Shepherd et al., 2013).

The distinction between direct and antisymmetrized ring summations is especially important in the thermodynamic limit. The same analysis reports that MP2 diverges in the thermodynamic limit for 3D metals because of the unscreened long-range Coulomb interaction, whereas Gell-Mann and Brueckner showed that RPA removes this divergence if antisymmetrization is neglected. Numerical evidence in the CCD-channel study further indicates that rings-only CCD with antisymmetrized integrals diverges at a similar rate to MP2, but the divergence can be cured either by removing antisymmetrization on the four-index integrals to make direct RPA, or by inclusion of crossed-rings (Shepherd et al., 2013).

Mosaic terms change the situation qualitatively. The same work determines that mosaics, by forming the Brueckner Hamiltonian, open a gap in the effective one-particle eigenvalues at the Fermi energy. Methods based on this renormalization have convergent energies in the thermodynamic limit, including mosaic-only CCD, whereas single-channel ladder-only, ring-only, and crossed-ring-only approximations appear to yield divergent energies; incorporation of mosaic terms prevents this from happening (Shepherd et al., 2013).

The range-separated Brueckner CCD construction extends this thermodynamic-limit logic. It combines the short-range ladder channel with the long-range ring channel in the presence of a Brueckner-renormalized one-body interaction, using a Yukawa separation of the Coulomb interaction. In that framework, the long-range part is treated with ring diagrams, the short-range part with ladder diagrams, and the resulting ground-state energies are reported to have an accuracy of $\begin{split} 0 =\ & \bar{v}_{ij}^{ab} + (\epsilon_a+\epsilon_b-\epsilon_i-\epsilon_j)t_{ij}^{ab}\ &+ \bar{v}_{cj}^{kb}\, t_{ik}^{ac} + \bar{v}_{ci}^{ka}\, t_{jk}^{bc} + \bar{v}_{cd}^{kl}\, t_{lj}^{db} t_{ik}^{ac}, \end{split}$1 a.u./electron across a wide range of density regimes, while also curing the overcorrelation of approaches based on ring diagrams (Shepherd et al., 2013).

4. Periodic solids, $\begin{split} 0 =\ & \bar{v}_{ij}^{ab} + (\epsilon_a+\epsilon_b-\epsilon_i-\epsilon_j)t_{ij}^{ab}\ &+ \bar{v}_{cj}^{kb}\, t_{ik}^{ac} + \bar{v}_{ci}^{ka}\, t_{jk}^{bc} + \bar{v}_{cd}^{kl}\, t_{lj}^{db} t_{ik}^{ac}, \end{split}$2-point sampling, and SOSEX

For periodic insulating systems, drCCD is used as one of the two formalisms in which the RPA correlation energy can be evaluated, the other being the adiabatic-connection fluctuation-dissipation formalism. In the periodic drCCD formulation, the amplitudes satisfy a $\begin{split} 0 =\ & \bar{v}_{ij}^{ab} + (\epsilon_a+\epsilon_b-\epsilon_i-\epsilon_j)t_{ij}^{ab}\ &+ \bar{v}_{cj}^{kb}\, t_{ik}^{ac} + \bar{v}_{ci}^{ka}\, t_{jk}^{bc} + \bar{v}_{cd}^{kl}\, t_{lj}^{db} t_{ik}^{ac}, \end{split}$3-resolved ring equation, and the per-unit-cell energy is

$\begin{split} 0 =\ & \bar{v}_{ij}^{ab} + (\epsilon_a+\epsilon_b-\epsilon_i-\epsilon_j)t_{ij}^{ab}\ &+ \bar{v}_{cj}^{kb}\, t_{ik}^{ac} + \bar{v}_{ci}^{ka}\, t_{jk}^{bc} + \bar{v}_{cd}^{kl}\, t_{lj}^{db} t_{ik}^{ac}, \end{split}$4

while the SOSEX correction is

$\begin{split} 0 =\ & \bar{v}_{ij}^{ab} + (\epsilon_a+\epsilon_b-\epsilon_i-\epsilon_j)t_{ij}^{ab}\ &+ \bar{v}_{cj}^{kb}\, t_{ik}^{ac} + \bar{v}_{ci}^{ka}\, t_{jk}^{bc} + \bar{v}_{cd}^{kl}\, t_{lj}^{db} t_{ik}^{ac}, \end{split}$5

Once the drCCD amplitudes are solved, SOSEX is obtained with minimal additional cost (Xing et al., 2021).

The principal numerical issue in periodic RPA/drCCD calculations is finite-size error from Brillouin-zone discretization. The staggered mesh method addresses this by using two offset Monkhorst-Pack meshes, $\begin{split} 0 =\ & \bar{v}_{ij}^{ab} + (\epsilon_a+\epsilon_b-\epsilon_i-\epsilon_j)t_{ij}^{ab}\ &+ \bar{v}_{cj}^{kb}\, t_{ik}^{ac} + \bar{v}_{ci}^{ka}\, t_{jk}^{bc} + \bar{v}_{cd}^{kl}\, t_{lj}^{db} t_{ik}^{ac}, \end{split}$6 for occupied orbitals and a shifted $\begin{split} 0 =\ & \bar{v}_{ij}^{ab} + (\epsilon_a+\epsilon_b-\epsilon_i-\epsilon_j)t_{ij}^{ab}\ &+ \bar{v}_{cj}^{kb}\, t_{ik}^{ac} + \bar{v}_{ci}^{ka}\, t_{jk}^{bc} + \bar{v}_{cd}^{kl}\, t_{lj}^{db} t_{ik}^{ac}, \end{split}$7 for virtual orbitals, so that virtual and occupied $\begin{split} 0 =\ & \bar{v}_{ij}^{ab} + (\epsilon_a+\epsilon_b-\epsilon_i-\epsilon_j)t_{ij}^{ab}\ &+ \bar{v}_{cj}^{kb}\, t_{ik}^{ac} + \bar{v}_{ci}^{ka}\, t_{jk}^{bc} + \bar{v}_{cd}^{kl}\, t_{lj}^{db} t_{ik}^{ac}, \end{split}$8-points never coincide and zero-momentum-transfer pairs are avoided (Xing et al., 2021).

The theoretical result established in that work is that the finite-size error of each perturbative term in standard RPA or SOSEX using a uniform Monkhorst-Pack mesh scales as $\begin{split} 0 =\ & \bar{v}_{ij}^{ab} + (\epsilon_a+\epsilon_b-\epsilon_i-\epsilon_j)t_{ij}^{ab}\ &+ \bar{v}_{cj}^{kb}\, t_{ik}^{ac} + \bar{v}_{ci}^{ka}\, t_{jk}^{bc} + \bar{v}_{cd}^{kl}\, t_{lj}^{db} t_{ik}^{ac}, \end{split}$9. With the staggered mesh, the dominant contribution associated with non-smooth quadrature points is substantially reduced, and in quasi-1D systems the decay can be super-algebraic; for isotropic 2D and 3D systems with high symmetry, the error decays as $\bar v$0 or $\bar v$1 (Xing et al., 2021). In the adiabatic-connection formulation, the same staggered construction also avoids the need for head/wing corrections to the dielectric operator (Xing et al., 2021).

5. Excitations, equation-of-motion theory, and self-energy connections

The RPA/drCCD equivalence is not limited to the ground-state energy. Neutral excitation energies from RPA are identical to those from an approximation to EOM-CCD when three approximations are imposed: the ground-state doubles amplitudes are obtained from the ring-CCD equations, the EOM problem is truncated to the single-excitation subspace, and the similarity transformation of the Fock operator is neglected (Berkelbach, 2018). Under these conditions, the singles-projected EOM problem becomes

$\bar v$2

which is algebraically identical to a downfolded RPA eigenvalue problem (Berkelbach, 2018).

A later Green’s-function analysis pushes the correspondence further and states that the $\bar v$3 approximation is equivalent to an equation-of-motion treatment of drCCD. In that context, the extended coupled-cluster ansatz provides a unified framework connecting coupled-cluster and many-body perturbation theory, and is proposed as a route for incorporating vertex corrections while keeping a positive semi-definite self-energy (Tölle et al., 11 Feb 2026).

The self-energy perspective also yields a reverse connection, namely a derivation of the CCD amplitude equations from a particle-hole-time decoupled electronic self-energy. That analysis produces a ground-state correlation energy

$\bar v$4

exactly of CCD form, and identifies drCCD as a subset in which only ring diagrams are retained (Coveney, 25 May 2025). In the Hubbard dimer, the mapping is exact, and the Green’s-function viewpoint makes the interpretation of different Riccati roots explicit in terms of quasiparticle and satellite structure (Coveney, 25 May 2025).

6. Range separation, molecular applications, and extensions beyond electronic drCCD

Range separation was introduced early as a way to control the short-range deficiencies of ring approximations. In closed-shell ring-CCD with range-separated density-functional theory, the long-range random-phase approximation variants are combined with short-range density-functional approximations. On the S22 set of weakly interacting complexes, the two best variants were RPAx-SO1 and RPAx-SO2, with mean absolute errors on equilibrium interaction energies of about $\bar v$5 kcal/mol and mean absolute percentage errors of about $\bar v$6 with the aug-cc-pVDZ basis set (Toulouse et al., 2011). These are not pure drCCD methods, but they are derived from ring-CCD theory and illustrate the broader role of ring truncations in noncovalent interactions.

For metals, the direct-ring idea has also been extended upward in excitation rank. The ring-CCSDT method was introduced as an iterative triples approximation that resums the essential direct ring contributions in the $\bar v$7 sector. With density fitting or Cholesky decomposition, its iterative step can be reduced to $\bar v$8 scaling, matching CCSD(T), while avoiding the divergence that makes perturbative triples inapplicable to three-dimensional metals (Neufeld et al., 2023). In the uniform electron gas and solid lithium, connected triple excitations were reported to be essential for high accuracy, and ring-CCSDT achieved nearly the same accuracy as full CCSDT at lower cost (Neufeld et al., 2023).

The ring-CCD/RPA correspondence has also been generalized to cavity quantum electrodynamics. In the QED-ring-CCD model, the cluster operator contains double electron excitations, coupled single electron/single photon excitations, and double photon creation. The resulting QED-drCCD amplitudes satisfy a composite Riccati equation, and the correlation energy

$\bar v$9

is analytically equivalent to the particle-hole QED-RPA ground-state correlation energy (DePrince et al., 10 Feb 2026). This extends the standard electronic drCCD/RPA equivalence to explicit light-matter coupling.

7. Solution multiplicity and numerical stabilization

Although drCCD-based RPA is algebraically compact, its nonlinear Riccati structure can generate nonphysical solutions. A recent analysis states that the drCCD equation admits $R(T)=B^*+A^*T+TA+TBT=0,$0 distinct solutions, with only one corresponding to the physical RPA solution obtained by choosing the positive-eigenvalue RPA sector (Song et al., 14 Aug 2025). Unphysical solutions have energies lower than the true RPA energy and become especially problematic in small-gap systems, including stretched bonds, large conjugated systems, and metallic clusters (Song et al., 14 Aug 2025).

The same work gives a necessary and sufficient a posteriori physicality criterion: $R(T)=B^*+A^*T+TA+TBT=0,$1 A converged drCCD solution is physical if and only if the largest eigenvalue of $R(T)=B^*+A^*T+TA+TBT=0,$2 is less than unity (Song et al., 14 Aug 2025). This criterion is numerically practical because it avoids direct diagonalization of the full RPA matrix.

To stabilize the iterations, several preconditioners were proposed: a diagonal-Jacobian preconditioner, level shifting,

$R(T)=B^*+A^*T+TA+TBT=0,$3

and the regularized MP2 forms

$R(T)=B^*+A^*T+TA+TBT=0,$4

A two-stage algorithm uses one of these stabilizing preconditioners in the pre-convergence stage and then switches back to the standard MP2 preconditioner for faster final convergence (Song et al., 14 Aug 2025). The same strategy was extended to reduced-scaling, factorized drCCD equations, including quartic-scaling formulations, thereby addressing convergence failures in large-scale drCCD-based RPA calculations (Song et al., 14 Aug 2025).