Local Unitary Jastrow Ansatz

• The local unitary Jastrow ansatz is a variational many-electron wave function that dresses a Hartree–Fock reference with unitary Jastrow correlators for efficient correlation recovery.
• Its structure employs local number–number operators and orbital rotations, enabling shallow circuits with O(N²) parameter scaling and reduced gate complexity.
• Practical implementations rely on exact circuit compilation and optimized initialization strategies to balance expressivity, computational efficiency, and hardware constraints.

Searching arXiv for recent and foundational papers on local/unitary cluster Jastrow ansätze and related Jastrow formulations. The local unitary Jastrow ansatz is a family of variational many-electron wave functions in which a reference state, typically a Hartree–Fock Slater determinant, is dressed by a product of unitary Jastrow correlators generated by number–number operators in an orbital basis and conjugated by orbital rotations. In its most common electronic-structure form, the ansatz is written as

$$\ket{\Psi}=\prod_{\mu=1}^{L} e^{\hat{K}_\mu} e^{i\hat{J}_\mu} e^{-\hat{K}_\mu}\ket{\Psi_{\mathrm{HF}}},$$

with

$$\hat{K}_\mu=\sum_{pq,\sigma} K_{pq}^\mu\,\hat{a}^\dagger_{p\sigma}\hat{a}_{q\sigma},\qquad \hat{J}_\mu=\sum_{pq,\sigma\tau} J_{pq,\sigma\tau}^\mu\,\hat{n}_{p\sigma}\hat{n}_{q\tau}.$$

The matrices $K^\mu$ are anti-Hermitian one-body generators, while $J^\mu$ are purely imaginary symmetric matrices, so that the exponentials are unitary. The qualifier “local” refers to restricting the number–number couplings in $\hat{J}_\mu$ to hardware- or graph-local pairs, yielding shallow circuits with $O(N^2)$ parameter and gate scaling rather than the $O(N^4)$ structure characteristic of generic UCC doubles. This places local unitary Jastrow constructions at the intersection of Hilbert-space Jastrow theory, low-rank coupled-cluster factorization, and hardware-efficient variational quantum circuits (Matsuzawa et al., 2019).

1. Origins in Jastrow and geminal wave functions

The broader Jastrow tradition in quantum chemistry and many-body physics predates unitary circuit formulations. In variational Monte Carlo, a standard correlated wave function has the form

$$\Psi(\mathbf{X})=J(\mathbf{X})\,\Psi_{\det}(\mathbf{X}),$$

with $J(\mathbf{X})=\exp(U)$ a positive multiplicative correlator and $\Psi_{\det}$ a Slater determinant or an antisymmetrized geminal power (AGP). In the Jastrow correlated AGP used for the $$\hat{K}_\mu=\sum_{pq,\sigma} K_{pq}^\mu\,\hat{a}^\dagger_{p\sigma}\hat{a}_{q\sigma},\qquad \hat{J}_\mu=\sum_{pq,\sigma\tau} J_{pq,\sigma\tau}^\mu\,\hat{n}_{p\sigma}\hat{n}_{q\tau}.$$0 model, the determinant part is an AGP written as the determinant of a pairing matrix, and the Jastrow factor contains electron–ion and electron–electron terms, including a localized geminal-like two-body expansion (1905.06971). In that formulation,

$$\hat{K}_\mu=\sum_{pq,\sigma} K_{pq}^\mu\,\hat{a}^\dagger_{p\sigma}\hat{a}_{q\sigma},\qquad \hat{J}_\mu=\sum_{pq,\sigma\tau} J_{pq,\sigma\tau}^\mu\,\hat{n}_{p\sigma}\hat{n}_{q\tau}.$$1

with $$\hat{K}_\mu=\sum_{pq,\sigma} K_{pq}^\mu\,\hat{a}^\dagger_{p\sigma}\hat{a}_{q\sigma},\qquad \hat{J}_\mu=\sum_{pq,\sigma\tau} J_{pq,\sigma\tau}^\mu\,\hat{n}_{p\sigma}\hat{n}_{q\tau}.$$2 diagonal in coordinate representation and real. This is explicitly non-unitary: $$\hat{K}_\mu=\sum_{pq,\sigma} K_{pq}^\mu\,\hat{a}^\dagger_{p\sigma}\hat{a}_{q\sigma},\qquad \hat{J}_\mu=\sum_{pq,\sigma\tau} J_{pq,\sigma\tau}^\mu\,\hat{n}_{p\sigma}\hat{n}_{q\tau}.$$3 rescales amplitudes, does not preserve norm, and does not directly change the sign of the wave function.

A related line of work in local orbital bases showed that a network of location-specific Jastrow factors can restore size consistency to AGP by acting as partial particle-number projectors. The Jastrow network is written as

$$\hat{K}_\mu=\sum_{pq,\sigma} K_{pq}^\mu\,\hat{a}^\dagger_{p\sigma}\hat{a}_{q\sigma},\qquad \hat{J}_\mu=\sum_{pq,\sigma\tau} J_{pq,\sigma\tau}^\mu\,\hat{n}_{p\sigma}\hat{n}_{q\tau}.$$4

with each $$\hat{K}_\mu=\sum_{pq,\sigma} K_{pq}^\mu\,\hat{a}^\dagger_{p\sigma}\hat{a}_{q\sigma},\qquad \hat{J}_\mu=\sum_{pq,\sigma\tau} J_{pq,\sigma\tau}^\mu\,\hat{n}_{p\sigma}\hat{n}_{q\tau}.$$5 built from occupation projectors on local spin-orbitals. In that setting the Jastrow is again non-unitary, but its locality structure already anticipates later unitary orbital-space correlators (Neuscamman, 2012).

These non-unitary constructions established two principles that remain central in local unitary Jastrow ansätze. First, strong correlation can be compactly represented by combining a structured reference, such as AGP, with local pair correlators. Second, the correlation operator and the reference must be optimized or chosen together: in the $$\hat{K}_\mu=\sum_{pq,\sigma} K_{pq}^\mu\,\hat{a}^\dagger_{p\sigma}\hat{a}_{q\sigma},\qquad \hat{J}_\mu=\sum_{pq,\sigma\tau} J_{pq,\sigma\tau}^\mu\,\hat{n}_{p\sigma}\hat{n}_{q\tau}.$$6 study, simultaneous optimization of Jastrow and AGP was reported as essential for recovering an almost perfect nodal surface, while pure AGP could exhibit qualitatively wrong behavior near the square geometry (1905.06971).

2. Unitarization in orbital space

The explicitly unitary version emerged from Hilbert-space cluster Jastrow decompositions designed for low-depth quantum circuits. A central formulation is the unitary cluster Jastrow, or uCJ, ansatz,

$$\hat{K}_\mu=\sum_{pq,\sigma} K_{pq}^\mu\,\hat{a}^\dagger_{p\sigma}\hat{a}_{q\sigma},\qquad \hat{J}_\mu=\sum_{pq,\sigma\tau} J_{pq,\sigma\tau}^\mu\,\hat{n}_{p\sigma}\hat{n}_{q\tau}.$$7

where each factor is a unitary variant of Neuscamman’s cluster Jastrow operator. In its product form,

$$\hat{K}_\mu=\sum_{pq,\sigma} K_{pq}^\mu\,\hat{a}^\dagger_{p\sigma}\hat{a}_{q\sigma},\qquad \hat{J}_\mu=\sum_{pq,\sigma\tau} J_{pq,\sigma\tau}^\mu\,\hat{n}_{p\sigma}\hat{n}_{q\tau}.$$8

with

$$\hat{K}_\mu=\sum_{pq,\sigma} K_{pq}^\mu\,\hat{a}^\dagger_{p\sigma}\hat{a}_{q\sigma},\qquad \hat{J}_\mu=\sum_{pq,\sigma\tau} J_{pq,\sigma\tau}^\mu\,\hat{n}_{p\sigma}\hat{n}_{q\tau}.$$9

Here $K^\mu$0 is complex anti-Hermitian and $K^\mu$1 is purely imaginary symmetric, so both $K^\mu$2 and $K^\mu$3 are unitary (Matsuzawa et al., 2019).

This construction can be interpreted as a low-rank decomposition of generalized UCC doubles. The $K^\mu$4-uCJ amplitudes are written as

$K^\mu$5

so increasing $K^\mu$6 systematically enlarges the reachable doubles manifold. In the near-term quantum-computing context, the defining advantage is that the Jastrow generators are built entirely from number operators. Under Jordan–Wigner,

$K^\mu$7

so $K^\mu$8 becomes a sum of commuting $K^\mu$9-type strings, eliminating Trotter error for the Jastrow block (Matsuzawa et al., 2019).

A broader generalization introduced alternative one-body generators beyond the earlier real-rotation form. For a reference $J^\mu$0, the $J^\mu$1-fold uCJ ansatz is

$J^\mu$2

with three variants distinguished by the structure of $J^\mu$3: Re-uCJ, Im-uCJ, and g-uCJ. Re-uCJ restricts orbital rotations to real anti-Hermitian generators, Im-uCJ uses imaginary orbital rotations, and g-uCJ permits fully complex anti-Hermitian $J^\mu$4, thereby enlarging expressivity while preserving exact, Trotter-free implementation (Tkachenko et al., 16 May 2025).

A different but conceptually related continuous-variable perspective appears in the time-dependent Jastrow ansatz for exact many-body dynamics. There, a complex Jastrow state is written as a real Jastrow amplitude multiplied by a unitary phase factor

$J^\mu$5

with $J^\mu$6 a sum of one- and two-body coordinate functions. In that framework, the phase piece is literally a local unitary Jastrow correlator, and it provides a bridge between exact dynamics and shortcuts to adiabaticity (Yang et al., 2022). This suggests that “local unitary Jastrow” is not restricted to qubit chemistry circuits but also has a coordinate-space, continuum many-body realization.

3. Locality, circuit structure, and exact compilation

The local unitary Jastrow ansatz derives much of its practical significance from exact circuit compilation with quadratic gate scaling. The Jastrow block is diagonal in the occupation basis: $J^\mu$7 where $J^\mu$8 and $J^\mu$9 are locality-restricted orbital and pair sets. The on-site opposite-spin term penalizes or favors double occupancy, while the inter-orbital same-spin terms encode short-range density correlations. Because all number operators commute, the full Jastrow exponential factorizes into commuting two-qubit controlled-phase gates after fermion-to-qubit mapping (Motta et al., 2024).

The orbital-rotation blocks are one-body fermionic Gaussian unitaries. In the quantum-circuit literature they are compiled through Givens rotations acting on orbital pairs. For complex orbital rotations, generalized fermionic Givens rotations extend the exact decomposition to Im-uCJ and g-uCJ. The practical consequence is that a single uCJ layer can be implemented with $\hat{J}_\mu$0 two-qubit gates and linear depth, and exact counts reported for the generalized construction give a total CNOT cost

$\hat{J}_\mu$1

for a single layer in the worst-case estimate (Tkachenko et al., 16 May 2025).

Locality in the stricter “LUCJ” sense is obtained by pruning $\hat{J}_\mu$2 to hardware-neighbor pairs. In the LUCJ formulation optimized with stochastic reconfiguration and the linear method, the full ansatz is

$\hat{J}_\mu$3

with

$\hat{J}_\mu$4

so for fixed locality pattern the parameter count scales as $\hat{J}_\mu$5 (Motta et al., 2024). This locality is operational rather than physical in the real-space sense: the retained Jastrow couplings are chosen to match qubit-connectivity graphs such as square or heavy-hex layouts, thereby removing SWAP overhead and making the ansatz hardware efficient.

The same emphasis on local two-body gates appears in impurity-style applications. In the local unitary cluster Jastrow ansatz used for ghost Gutzwiller embedding, each layer is

$\hat{J}_\mu$6

with nearest-neighbor hopping-like $\hat{J}_\mu$7 and density–density $\hat{J}_\mu$8 that include an impurity on-site $\hat{J}_\mu$9 term and nearest-neighbor same-spin terms. Its Jordan–Wigner form consists of $O(N^2)$0-type orbital rotations and $O(N^2)$1-type Jastrow gates, and the Jastrow block can be implemented in constant depth 2 when acting on disjoint pairs (Sriluckshmy et al., 26 Jun 2025).

4. Expressivity, optimization, and initialization

The appeal of local unitary Jastrow ansätze lies in achieving high correlation recovery with a shallow, structured circuit. For $O(N^2)$2, generalized uCJ models already capture substantial correlation. Numerical benchmarks showed that g-uCJ frequently recovers all or nearly all correlation energy in small active spaces, while Im-uCJ is systematically more accurate than the earlier Re-uCJ and often remains within chemical accuracy. On $O(N^2)$3 in STO-3G, g-uCJ reproduced FCI exactly for all bond distances, while Im-uCJ closely tracked FCI and Re-uCJ showed noticeably larger errors (Tkachenko et al., 16 May 2025).

For more challenging systems, optimization becomes decisive. Quantum algorithms based on stochastic reconfiguration and the linear method were developed specifically for LUCJ-type wave functions. In that setting, the overlap matrix, gradient vector, and tangent-space Hamiltonian matrix are estimated with polynomial circuit and shot complexity, even though classical evaluation for such orbital-space Jastrow-cluster states would be exponentially costly. Classical simulations showed that linear-method optimization consistently found lower-energy LUCJ solutions than L-BFGS-B across the dissociation curves of $O(N^2)$4 and $O(N^2)$5, and LUCJ predictions deviated from exact diagonalization by 1 kcal/mol or less at all points on the potential energy curve (Motta et al., 2024). A plausible implication is that the expressivity of the ansatz and the quality of the optimizer cannot be separated: shallow unitary Jastrow forms may underperform primarily because of optimization traps rather than intrinsic limitations.

Initialization has likewise become a distinct topic. CCSD-derived parameters are a natural starting point because UCJ/LUCJ can be viewed as a double-factorized representation of UCC doubles. A later study proposed compressed double factorization to improve CCSD-based initialization under layer truncation and locality constraints, minimizing a least-squares mismatch to target CCSD doubles while respecting the LUCJ architecture. A second method used approximate tensor-network simulation to optimize parameters for sample-based algorithms such as QSCI and SQD. Both methods improved sample quality and energy accuracy on systems up to 52 qubits in state-vector simulation and up to 65 qubits on superconducting processors (Lin et al., 27 Nov 2025).

Yet a complementary study within the quantum-centric supercomputing framework found that SQD energies are largely insensitive to how closely the initial $O(N^2)$6 amplitudes resemble CCSD. Across twelve molecular systems and three basis sets, random initialization produced performance competitive with CCSD, while zero initialization gave the worst energy agreement despite smaller deviations from CCSD amplitudes. The reported conclusion was that configuration recovery within SQD, rather than circuit initialization, is the dominant factor governing energy accuracy (Jones et al., 12 Jun 2026). This suggests that initialization matters differently depending on the downstream algorithm: expectation-value VQE benefits more directly from amplitude-faithful initializations, whereas sample-based subspace reconstruction can tolerate highly imperfect starting points.

5. Applications and variants

A number of applications have clarified which physical regimes local unitary Jastrow ansätze handle well. In low-depth VQE, the original $O(N^2)$7-uCJ ansatz was benchmarked on $O(N^2)$8 dissociation in STO-6G. Using exact matrix exponentiation, $O(N^2)$9 achieved a maximum error of about $O(N^4)$0 along the potential curve, effectively attaining chemical accuracy, while $O(N^4)$1 was insufficient in the multireference region (Matsuzawa et al., 2019). This remains one of the clearest demonstrations that a sequence of number–number Jastrow layers and orbital rotations can emulate a large portion of generalized doubles physics at $O(N^4)$2 parameter cost.

The local unitary cluster Jastrow ansatz has also been used as a trial state for sample-based quantum-selected configuration interaction in ghost Gutzwiller embedding. There the aim is not direct variational energy minimization but generation of CI configurations for an embedding Hamiltonian. In that role, shallow LUCJ circuits combined with circuit cutting and QSCI were sufficient to recover converged ghost Gutzwiller calculations and to capture the metal-to-insulator transition in the Bethe-lattice Fermi–Hubbard model while sampling as little as $O(N^4)$3 of the total CI basis states (Sriluckshmy et al., 26 Jun 2025). This use case highlights a different strength of local unitary Jastrow states: they need not be the final wave function if they sample the important determinant support.

Non-unitary analogues remain relevant in NISQ settings. A non-unitary VQE with Jastrow-factor-inspired qubit operators combined with localized active-space references was reported to recover interfragment correlation and reach chemical accuracy with a small number of gates in both $O(N^4)$4 and square cyclobutadiene. In that approach the non-unitary $O(N^4)$5 does not increase circuit depth but instead shifts cost into measuring $O(N^4)$6 and metric operators. This is not a unitary ansatz, but it provides a contrasting design trade-off: circuit-depth savings at the expense of measurement complexity (Wang et al., 23 Jan 2025).

Another related route uses cascaded VQE to incorporate a non-unitary Jastrow–Gutzwiller operator classically while implementing only a unitary Thouless circuit on hardware. For the Hubbard model, the Jastrow–Gutzwiller operator

$O(N^4)$7

was handled through measurement reweighting rather than direct unitary implementation, but the work explicitly noted that a unitary local Jastrow analogue would be

$O(N^4)$8

namely a product of local $O(N^4)$9-type phase gates (Stenger et al., 2023). This makes clear that the local unitary Jastrow ansatz is part of a broader landscape that includes non-unitary projective correlators, unitary correlator-product circuits, and hybrid quantum–classical reformulations.

6. Conceptual significance and open directions

Several themes recur across the literature. The first is the separation of static and dynamic correlation. In non-unitary JAGP, AGP handles static correlation while the Jastrow recovers short-range dynamical correlation (1905.06971). In local unitary cluster Jastrow circuits, the orbital-rotation-dressed density correlator plays an analogous role: it provides low-rank, pairwise entanglement that is far more expressive than a bare Slater determinant yet far cheaper than generic UCC doubles (Matsuzawa et al., 2019). A plausible implication is that local unitary Jastrow ansätze are best understood not as simplified coupled-cluster truncations, but as structured correlator-product states in an optimized orbital frame.

The second theme is exact implementability. Unlike many UCC-like circuits, the Jastrow blocks commute and the Gaussian one-body blocks admit exact fermionic compilation. This is why Re-uCJ, Im-uCJ, and g-uCJ can all be implemented without Trotter decomposition and still retain $$\Psi(\mathbf{X})=J(\mathbf{X})\,\Psi_{\det}(\mathbf{X}),$$0 gate scaling (Tkachenko et al., 16 May 2025). That property is central for NISQ relevance.

The third theme is locality as an architectural prior. In LUCJ, locality means restricting density couplings to a graph compatible with device connectivity (Motta et al., 2024). In impurity embedding, it means nearest-neighbor cluster couplings plus a distinguished impurity interaction (Sriluckshmy et al., 26 Jun 2025). In continuous-variable time-dependent Jastrow theory, locality refers to one- and two-body functions of coordinates (Yang et al., 2022). These are distinct notions, but they share the idea that the correlator should be assembled from few-body local generators whose exponentials remain tractable.

There are also persistent limitations. A single $$\Psi(\mathbf{X})=J(\mathbf{X})\,\Psi_{\det}(\mathbf{X}),$$1 layer can fail when higher-rank excitations are essential, as in stretched multi-bond dissociation, motivating larger $$\Psi(\mathbf{X})=J(\mathbf{X})\,\Psi_{\det}(\mathbf{X}),$$2 (Tkachenko et al., 16 May 2025). LUCJ circuits can produce highly accurate energies but may yield nonsmooth potential curves unless additional symmetry constraints or projection methods are imposed (Motta et al., 2024). CCSD-based initialization is not always a reliable predictor of downstream sample-based performance (Jones et al., 12 Jun 2026). And although locality improves hardware efficiency, it necessarily discards some long-range orbital couplings; later work on improved initialization was motivated precisely by the degradation caused by truncating layers and discarding nonlocal interactions (Lin et al., 27 Nov 2025).

Taken together, the literature presents the local unitary Jastrow ansatz as a distinct variational paradigm: unitary, number-preserving, pair-correlator based, exactly compilable, and systematically improvable by increasing the number of layers or relaxing locality restrictions. It inherits conceptual foundations from Jastrow–Slater and JAGP theory, but reinterprets them in a circuit language suited to quantum hardware. Within that framework, Re-uCJ, Im-uCJ, g-uCJ, and LUCJ are best viewed not as isolated models but as members of a single family of low-rank correlator ansätze whose practical value depends jointly on locality design, compilation strategy, initialization, and optimizer quality (Tkachenko et al., 16 May 2025).