All-Particle FSRCC: Relativistic Multireference Method

• The method is a valence-universal multireference framework that correlates both core and valence electrons using a four-component Dirac–Coulomb (or Breit) Hamiltonian.
• It employs a Fock-space sector approach to sequentially access ionized, electron-attached, and excited states, providing high accuracy for heavy-element systems.
• The technique overcomes intruder state challenges through careful model-space design and intermediate Hamiltonian constructions, ensuring reliable property predictions.

All-particle multireference Fock-space relativistic coupled-cluster, usually denoted FSRCC or simply FSCC, is a valence-universal coupled-cluster framework for relativistic many-electron structure in which a closed-shell reference is correlated first and families of open-shell, ionized, electron-attached, or excited states are then reached by moving through Fock-space sectors labeled by valence holes and particles. In contemporary heavy-atom and heavy-molecule work it is typically formulated with a four-component no-pair Dirac–Coulomb or Dirac–Coulomb–Breit Hamiltonian, finite nuclear size, and explicit correlation of core and valence electrons, while multireference character enters through a model space of near-degenerate determinants and diagonalization of an effective Hamiltonian in that space (Pašteka et al., 18 Aug 2025, Oleynichenko et al., 2020).

1. Terminology and conceptual scope

In the recent literature, FSRCC and FSCC are effectively synonymous for the four-component relativistic implementation. The qualifier “all-particle” is used in more than one closely related sense. In most spectroscopic applications it denotes an all-electron treatment in which core and valence electrons are correlated explicitly and Fock-space sectors are indexed by holes and particles relative to a closed-shell vacuum. In some two-valence atomic developments it also designates an attachment-only construction that proceeds through the $(0,0)$, $(0,1)$, and $(0,2)$ sectors. The 2025 review distinguishes these electronic uses from the genuinely all-particle QED extension, Double Fock-Space Coupled Cluster, in which photons are included alongside electrons in Fock space (Pašteka et al., 18 Aug 2025, Mani et al., 2010, Oleynichenko et al., 2020).

The method is valence-universal rather than state-specific. A single wave operator is constructed for a chosen model space and then used to recover several target states simultaneously. This contrasts with single-reference CCSD(T), which is most accurate when one determinant dominates, and with EOM-CC, which accesses attached, detached, or excited states by diagonalizing a similarity-transformed Hamiltonian in a linear excitation manifold. The BaF hyperfine benchmark makes this distinction explicit: the excited $A\,{}^2\Pi$ states show large single-reference $T_1$ values of about $0.16$, supporting the use of FSCC rather than a single-reference treatment in those manifolds (Denis et al., 2022).

Because the model space is built from selected valence spinors or configurations, FSRCC is multireference in a precise operational sense rather than by formal analogy. Near-degenerate determinants are treated on equal footing in the effective Hamiltonian, while dynamic correlation is summed to all orders within the chosen cluster truncation. This feature is central in heavy atoms, superheavy atoms, and spin-orbit-coupled molecules, where strong configuration mixing and relativistic splittings coexist (Pašteka et al., 18 Aug 2025).

2. Relativistic Hamiltonian, wave operator, and sector structure

The standard starting point is the no-pair projected relativistic Hamiltonian. In compact form,

$H_{\mathrm{NP}}=\Lambda^{(+)}H\Lambda^{(+)},$

where $\Lambda^{(+)}$ projects onto the positive-energy electronic subspace. In second quantization, with respect to a closed-shell Fermi vacuum, the Hamiltonian is written as

$$H=\sum_{pq} h_{pq}\{a_p^\dagger a_q\}+\frac14\sum_{pqrs}V_{pqrs}\{a_p^\dagger a_q^\dagger a_s a_r\},$$

with $h_{pq}$ derived from the four-component Dirac operator and $(0,1)$0 from Coulomb or Coulomb–Breit interactions. Finite nuclear size is standard, and Breit, vacuum-polarization, and self-energy corrections are included when the required accuracy demands them (Oleynichenko et al., 2020, Pašteka et al., 18 Aug 2025).

Several equivalent parameterizations of the wave operator are used. Common notations are

$(0,1)$1

and

$(0,1)$2

The closed-shell cluster operator $(0,1)$3 correlates the $(0,1)$4 vacuum, while $(0,1)$5 or $(0,1)$6 carries the sector-specific valence amplitudes. The model space $(0,1)$7 is spanned by selected determinants or configuration state functions, with $(0,1)$8. The fundamental Bloch relations are

$(0,1)$9

or, after similarity transformation,

$(0,2)$0

Diagonalization of $(0,2)$1 in $(0,2)$2 yields the correlated roots of interest (Oleynichenko et al., 2020, Denis et al., 2022).

Sector labels $(0,2)$3 count valence holes and particles relative to the closed-shell reference. Standard electronic sectors include $(0,2)$4 for ionization, $(0,2)$5 for electron attachment, $(0,2)$6 for excitations, and higher sectors such as $(0,2)$7, $(0,2)$8, and $(0,2)$9 for increasingly open-shell manifolds. The sectors are solved sequentially, with the subsystem embedding condition ensuring that amplitudes in a given sector depend only on lower or equal sectors. In FS-RCCSD the truncation is

$A\,{}^2\Pi$0

while selected sectors now also admit FS-CCSDT-$A\,{}^2\Pi$1 and full FS-CCSDT treatments (Oleynichenko et al., 2020, Pašteka et al., 18 Aug 2025).

Because $A\,{}^2\Pi$2 is non-Hermitian, modern implementations retain biorthonormal left and right eigenvectors,

$A\,{}^2\Pi$3

with $A\,{}^2\Pi$4. This biorthogonal structure is not a technical afterthought; it enters directly into property evaluation, transition moments, and density matrices (Oleynichenko et al., 2020).

3. Model spaces, intruder states, and numerical stabilization

The practical success of FSRCC depends strongly on model-space design. A balanced $A\,{}^2\Pi$5 space must contain the valence spinors or configurations needed for the target manifold while avoiding excessive expansion that invites intruders. In heavy-atom applications this usually means including the dominant valence configurations together with enough auxiliary configurations to stabilize the effective Hamiltonian and preserve quasi-degeneracy structure (Pašteka et al., 18 Aug 2025).

Intruder states arise when denominators in the amplitude equations become small or positive because an external $A\,{}^2\Pi$6-space configuration approaches the energy region of the model space. Several remedies are established. Intermediate-Hamiltonian constructions, associated in the literature with Evangelisti–Malrieu and Kaldor’s mixed-sector variants, partition the model space into main and auxiliary parts and suppress problematic couplings. EXP-T also implements denominator shifts of the form

$A\,{}^2\Pi$7

for ill-defined denominators, followed by Padé extrapolation to $A\,{}^2\Pi$8, which damps spurious high-lying roots while preserving low-lying spectra (Oleynichenko et al., 2020).

An alternative strategy is model-space engineering. In the two-valence Sr, Ba, and Yb study, a quasi-complete incomplete model space built from $A\,{}^2\Pi$9, $T_1$0, and $T_1$1 configurations was used precisely to avoid intruders while retaining the correct subduction to lower sectors. In the neutral Bi $T_1$2 calculation, the minimal active space of $T_1$3 bispinors did not generate intruders, so no intermediate-Hamiltonian machinery was required. In molecular work, balanced inclusion of physically relevant valence spinors is equally important; in BaF, the $T_1$4, $T_1$5, and $T_1$6 spinors of the neutral manifold were used in the $T_1$7 sector to access the $T_1$8 and $T_1$9 states within a single calculation (Mani et al., 2010, Skripnikov et al., 2021, Denis et al., 2022).

The consequence is methodological rather than merely numerical. FSRCC does not separate formal theory from active-space choice: the multireference content of the method is realized through the model space, and the quality of $0.16$0 depends directly on how that space resolves the physically relevant near-degeneracies (Pašteka et al., 18 Aug 2025).

4. Property theory, finite fields, and reduced density matrices

Property evaluation in FSRCC proceeds through similarity-transformed operators and biorthogonal model-space eigenvectors. In the common formulation,

$0.16$1

and transition moments are written as $0.16$2. Ground-state properties can be formulated through a CC Lagrangian, whereas open-shell sector properties usually rely on effective-operator matrix elements or finite-field differentiation (Oleynichenko et al., 2020, Pašteka et al., 18 Aug 2025).

Finite-field methods are especially important because fully analytic derivatives for intermediate-Hamiltonian or extrapolated-IH variants are not yet generally available. In BaF, magnetic hyperfine constants were obtained from numerical energy derivatives with respect to a perturbation $0.16$3, using $0.16$4 for $0.16$5 and $0.16$6 for $0.16$7, both verified to lie in the strictly linear regime. In Bi, electric field gradients for extraction of $0.16$8 were also computed by finite field (Denis et al., 2022, Skripnikov et al., 2021).

Perturbed-cluster extensions retain the same Fock-space logic while introducing tensorial perturbation operators. In one-valence systems, FS-PRCC has been used to compute static polarizabilities and nuclear-spin-dependent parity-nonconserving amplitudes, with perturbed rank-1 cluster operators that subsume sum-over-states expressions. For Cs, this yielded recommended $0.16$9 NSD-PNC amplitudes with an estimated upper-bound theoretical uncertainty of about $H_{\mathrm{NP}}=\Lambda^{(+)}H\Lambda^{(+)},$0, and identified a largest cumulative Breit+QED correction of about $H_{\mathrm{NP}}=\Lambda^{(+)}H\Lambda^{(+)},$1 (Pandey et al., 2024).

A complementary development is the finite-order effective-operator approach to density matrices. In this scheme, pure-state and transition reduced density matrices are obtained by substituting the Fock-space CC wave operator into effective-operator expressions and truncating at quadratic order in the cluster amplitudes. For hole-only and particle-only sectors under intermediate normalization, the resulting pure-state reduced density matrices are connected; transition analyses can be stabilized by working with products such as $H_{\mathrm{NP}}=\Lambda^{(+)}H\Lambda^{(+)},$2, which also underpin relativistic natural transition spinors (Oleynichenko et al., 2024).

5. Representative applications in molecules, atoms, and clocks

Molecular hyperfine structure has been one of the clearest demonstrations of four-component FSRCC. In BaF, a combined CCSD(T)/FSCC finite-field strategy gave $H_{\mathrm{NP}}=\Lambda^{(+)}H\Lambda^{(+)},$3 hyperfine constants $H_{\mathrm{NP}}=\Lambda^{(+)}H\Lambda^{(+)},$4 MHz and $H_{\mathrm{NP}}=\Lambda^{(+)}H\Lambda^{(+)},$5 MHz for the $H_{\mathrm{NP}}=\Lambda^{(+)}H\Lambda^{(+)},$6 ground state, and $H_{\mathrm{NP}}=\Lambda^{(+)}H\Lambda^{(+)},$7 MHz, $H_{\mathrm{NP}}=\Lambda^{(+)}H\Lambda^{(+)},$8 MHz for $H_{\mathrm{NP}}=\Lambda^{(+)}H\Lambda^{(+)},$9, together with $\Lambda^{(+)}$0 MHz and $\Lambda^{(+)}$1 for $\Lambda^{(+)}$2. The same study found uncertainties of about $\Lambda^{(+)}$3 for the ground state but $\Lambda^{(+)}$4 to $\Lambda^{(+)}$5 for excited-state constants because the diffuse $\Lambda^{(+)}$6 manifolds were much more basis-sensitive (Denis et al., 2022). In KCs, four-component FS-RCCSD was used to compute diagonal and off-diagonal magnetic hyperfine matrix elements as functions of internuclear separation; the couplings were nearly constant beyond $\Lambda^{(+)}$7 Å, while deviations from atomic values reached about $\Lambda^{(+)}$8 near the equilibrium region, and core-core correlation was found to be negligible compared with core-valence correlation and core relaxation (Oleynichenko et al., 2020).

Atomic and ionic applications span one-, two-, and three-particle sectors. For two-valence Sr, Ba, and Yb, an all-particle Fock-space treatment built on $\Lambda^{(+)}$9, $$H=\sum_{pq} h_{pq}\{a_p^\dagger a_q\}+\frac14\sum_{pqrs}V_{pqrs}\{a_p^\dagger a_q^\dagger a_s a_r\},$$0, and $$H=\sum_{pq} h_{pq}\{a_p^\dagger a_q\}+\frac14\sum_{pqrs}V_{pqrs}\{a_p^\dagger a_q^\dagger a_s a_r\},$$1 model spaces yielded excitation energies, magnetic hyperfine constants, electric quadrupole hyperfine constants, and E1 matrix elements, with the one-valence sector simultaneously supplying properties of Sr$$H=\sum_{pq} h_{pq}\{a_p^\dagger a_q\}+\frac14\sum_{pqrs}V_{pqrs}\{a_p^\dagger a_q^\dagger a_s a_r\},$$2, Ba$$H=\sum_{pq} h_{pq}\{a_p^\dagger a_q\}+\frac14\sum_{pqrs}V_{pqrs}\{a_p^\dagger a_q^\dagger a_s a_r\},$$3, and Yb$$H=\sum_{pq} h_{pq}\{a_p^\dagger a_q\}+\frac14\sum_{pqrs}V_{pqrs}\{a_p^\dagger a_q^\dagger a_s a_r\},$$4 (Mani et al., 2010). For neutral Bi, a $$H=\sum_{pq} h_{pq}\{a_p^\dagger a_q\}+\frac14\sum_{pqrs}V_{pqrs}\{a_p^\dagger a_q^\dagger a_s a_r\},$$5 calculation with full iterative triples in the three-particle sector produced electric field gradients from which the recommended quadrupole moment $$H=\sum_{pq} h_{pq}\{a_p^\dagger a_q\}+\frac14\sum_{pqrs}V_{pqrs}\{a_p^\dagger a_q^\dagger a_s a_r\},$$6 mb was extracted; the triples corrections changed the EFGs by $$H=\sum_{pq} h_{pq}\{a_p^\dagger a_q\}+\frac14\sum_{pqrs}V_{pqrs}\{a_p^\dagger a_q^\dagger a_s a_r\},$$7 for the ground state and $$H=\sum_{pq} h_{pq}\{a_p^\dagger a_q\}+\frac14\sum_{pqrs}V_{pqrs}\{a_p^\dagger a_q^\dagger a_s a_r\},$$8 for the excited state, demonstrating that sector-specific triples are indispensable in $$H=\sum_{pq} h_{pq}\{a_p^\dagger a_q\}+\frac14\sum_{pqrs}V_{pqrs}\{a_p^\dagger a_q^\dagger a_s a_r\},$$9 systems (Skripnikov et al., 2021).

Clock and precision-spectroscopy studies have made the same point in different observables. The hyperfine-induced $h_{pq}$0 transition in Al$h_{pq}$1 was described with a two-valence all-particle FSRCC method, giving a metastable lifetime of $h_{pq}$2 s in excellent agreement with the experimental $h_{pq}$3 s, and showing that triples, Breit, and QED effects are all essential at the targeted level (Kumar et al., 2020). In Pb$h_{pq}$4, an all-particle multireference FSRCC treatment of the $h_{pq}$5 clock transition yielded a recommended clock-state lifetime of $h_{pq}$6 s and a ground-state polarizability of $h_{pq}$7 a.u.; the analysis identified valence-valence correlation from higher configurations, perturbative triples, and Breit/QED corrections as essential (Gakkhar et al., 2024). For fermionic and bosonic Sr, the same framework produced clock-state lifetimes of $h_{pq}$8 s for $h_{pq}$9Sr and $(0,1)$00 s for $(0,1)$01Sr, together with isotope-shift parameters and a PRCC ground-state polarizability of $(0,1)$02 a.u. (Gakkhar et al., 17 Oct 2025).

Superheavy-element studies extend the range of observables. In nobelium, all-particle multireference FSRCC plus PRCC gave an ionization potential of $(0,1)$03 cm$(0,1)$04, a ground-state polarizability of $(0,1)$05 a.u., and nuclear moments $(0,1)$06 and $(0,1)$07 eb extracted from hyperfine data. The same work found a perturbative-triples contribution of about $(0,1)$08 to the $(0,1)$09 transition rate and a largest cumulative Breit+QED contribution of about $(0,1)$10 to the magnetic-dipole hyperfine constant of $(0,1)$11 (Kumar et al., 28 Jul 2025).

6. Accuracy, implementation, limitations, and current directions

The principal high-performance implementation described in the supplied literature is the EXP-T package. It reduces tensor contractions to block supermatrix multiplications through a TTGT strategy, partitions tensors with direct-product decomposition, exploits double-group and Kramers symmetry, and uses OpenMP on CPUs and CUDA on GPUs together with MKL and cuBLAS. At the FS-CCSD level the leading cost is $(0,1)$12 per iteration and memory is $(0,1)$13, while the dominant bottleneck is often I/O associated with the $(0,1)$14 integral blocks stored on disk (Oleynichenko et al., 2020).

Accuracy is controlled through basis choice, model-space design, correlation level, relativistic corrections, and uncertainty analysis. The review recommends four-component finite-nucleus calculations with Dyall or ANO-RCC basis families, tight functions for nuclear-region properties, diffuse augmentation for anions and response properties, and large positive-energy virtual cutoffs when deep-core correlation matters. For hyperfine constants and field gradients in heavy systems, correlating all electrons and including Breit and leading QED effects can be necessary; for many energy predictions, valence plus outer-core correlation may suffice (Pašteka et al., 18 Aug 2025). The BaF benchmark provides a concrete uncertainty protocol in which basis incompleteness, missing correlation, virtual-space cutoff, and omitted relativistic effects are varied independently and combined in quadrature; there the dominant error for excited-state hyperfine constants was basis-set incompleteness rather than correlation truncation (Denis et al., 2022).

The principal limitations are also well documented. Intruder states remain the classic difficulty, though intermediate-Hamiltonian variants, dynamic shifts plus Padé extrapolation, and careful model-space balancing have greatly improved robustness. Model-space dependence is intrinsic to the method; poor active-space choices can destabilize $(0,1)$15 or misrepresent quasidegeneracies. Full triples remain expensive outside selected sectors, and a general open-shell FS-RCCSD(T) correction is explicitly described as nontrivial and not broadly available. On the property side, analytic density matrices and effective property operators for IH/XIH formulations are still under development, and the finite-order density-matrix scheme guarantees connectivity only in particle-only and hole-only sectors, not in mixed sectors such as $(0,1)$16 without additional machinery (Oleynichenko et al., 2020, Oleynichenko et al., 2024, Pašteka et al., 18 Aug 2025).

Current directions therefore follow naturally from the existing architecture. The 2025 review identifies broader use of FS-CCSDT in higher sectors, improved Breit and QED treatments, wider deployment of XIH and Padé-IH, analytical property formalisms, tensor-train and GPU acceleration, and the development of Dyall $(0,1)$17 basis sets. It also points to DFSCC as the true all-particle QED generalization beyond the electronic no-pair framework (Pašteka et al., 18 Aug 2025). Taken together, these developments place all-particle multireference FSRCC at the intersection of relativistic many-body theory, precision spectroscopy, and uncertainty-quantified prediction for heavy and superheavy systems.