Deterministic Entanglement Compilation

Updated 10 July 2026

Deterministic entanglement compilation is the process of translating abstract entangled resources into predetermined physical operations, eliminating reliance on probabilistic post-selection.
Techniques include sequential photon emission, native two-qudit synthesis via QR decomposition, and graph-state partitioning with local complementation that reduce gate counts and circuit duration.
Protocols leverage measurement feedback, holonomic transport, and network recursion to achieve reliable state generation while managing trade-offs between theoretical exactness and practical fidelity.

Deterministic entanglement compilation denotes the translation of a target entanglement resource into a concrete sequence of physical operations whose success is not delegated to probabilistic post-selection or ad hoc circuit search. In the cited literature, the target may be a multipartite photonic state, an arbitrary two-qudit unitary, an amplitude-encoded register, a graph state on emitter-photonic hardware, a distributed entangled link in a quantum network, or a Bell state created by measurement and feedback (Besse et al., 2020, Mato et al., 2023, Gundlapalli et al., 2021, Ren et al., 20 Mar 2025, Meng et al., 2021, Ristè et al., 2013). The common theme is not a single hardware platform or a single mathematical formalism, but a design objective: to compile entanglement structure into deterministic dynamics, deterministic measurements with feedback, or deterministic local operations and classical communication (LOCC), while optimizing a platform-dependent cost such as entangling-gate count, circuit duration, photon-loss exposure, concurrence, or fidelity.

1. Operational scope and meanings of determinism

The cited works use “deterministic” in several operational senses. In sequential emitter-photonic generation, it means that the emitted photonic state is left in a pure, deterministic entangled state after a final disentangling operation on the auxiliary system (Besse et al., 2020). In qudit compilation, it means a fully-automatic workflow that produces correct gate sequences for any two-qudit unitary over an arbitrary native entangling gate set, once the standardized mediating gates have been precompiled (Mato et al., 2023). In measurement-based schemes, it may mean that a random measurement branch is converted into a fixed target entangled state by conditional feedback, or that all measurement outcomes are equiprobable and locally interconvertible, so no branch is a failure branch (Ristè et al., 2013, J et al., 2024).

Modality	Deterministic object	Representative result
Sequential emission	Cluster, GHZ, W photonic states	Full tomography up to $N=4$ (Besse et al., 2020)
Native-gate compilation	Any two-qudit unitary	Fully-automatic workflow (Mato et al., 2023)
Graph-state compilation	Emitter-photonic graph states	CNOT and duration reductions up to 52% and 56% (Ren et al., 20 Mar 2025)
Network transmission	Distributed bipartite entanglement	DET improves on CEP (Meng et al., 2021)
Parity measurement with feedback	Bell-state preparation	66% fidelity on demand (Ristè et al., 2013)

A recurrent misconception is that determinism implies exact, noiseless realization. The cited experiments and compiler studies do not support that interpretation. Deterministic protocols may still exhibit finite fidelity, decoherence-limited performance, or approximation error. Another recurrent ambiguity concerns heralding. In one strand of the literature, “deterministic and heralded” refers to remote-state-generation protocols in which measurement outcomes certify which entangled state was created, without requiring entangled-state measurements or multiple pre-shared entangled states (Liu et al., 5 Mar 2025). In another, “effectively deterministic” refers to concentration protocols in which every branch yields the same entanglement up to local unitaries (J et al., 2024). By contrast, “near-deterministic” is used for large many-body generation when the final-state distribution is strongly concentrated but not strictly single-shot deterministic (Luo et al., 2017).

2. Sequential emitters, photonic state synthesis, and graph-state compilation

A central formulation of deterministic entanglement compilation is the sequential emission protocol for multipartite-entangled photonic qubits (Besse et al., 2020). The hardware consists of an auxiliary transmon qutrit $(|g\rangle, |e\rangle, |f\rangle)$ and an emitter transmon qubit, tunably coupled via a flux-tunable interferometric coupler, with strong decay into a transmission line. The auxiliary is initialized by microwave pulses, after which CNOT-like operations, SWAP gates, or programmable $\text{SWAP}(\theta)$ gates are interleaved with photon emission into well-defined time-bin modes $P_i$ . Local auxiliary control $U$ between emissions enables state compilation, and a final SWAP disentangles the auxiliary so that the photonic output alone carries the target entanglement.

The implemented circuits cover cluster states, GHZ states, and W states. Cluster states are created with repeated CNOT gates and interleaved Hadamard gates on the auxiliary; GHZ states are constructed entirely with CNOT gates; W states require tunable partial exchanges with

$\theta_i = 2\arcsin\left[(N-i+1)^{-0.5}\right].$

This architecture deterministically generates a wide class of matrix-product states with bond dimension $d=2$ , making the compilation target more general than a fixed family of stabilizer states. Full quantum state tomography of the multi-mode photonic output is performed for up to $N=4$ time-bin modes using heterodyne detection and the $R\rho R$ iterative maximum-likelihood algorithm. Reported fidelities are $F=0.77{-}0.82$ for cluster states, $(|g\rangle, |e\rangle, |f\rangle)$ 0 for GHZ states, and $(|g\rangle, |e\rangle, |f\rangle)$ 1 for W states for $(|g\rangle, |e\rangle, |f\rangle)$ 2. For larger $(|g\rangle, |e\rangle, |f\rangle)$ 3, the emission cycle is characterized by a stationary process map

$(|g\rangle, |e\rangle, |f\rangle)$ 4

which is iterated in a matrix product density operator description. Using negativity,

$(|g\rangle, |e\rangle, |f\rangle)$ 5

A later compiler framework generalizes the deterministic emitter-photonic perspective from specific sequential circuits to large graph states (Ren et al., 20 Mar 2025). The target graph is partitioned into subgraphs, optionally transformed by local complementation, then compiled and scheduled so as to maximize emitter resource utilization and suppress photon loss. The optimization problem explicitly models cut-edge minimization after depth-limited local complementations, while subgraph synthesis is guided by a time-reversed reduction sequence and a photon-loss metric,

$(|g\rangle, |e\rangle, |f\rangle)$ 9

The scheduling stage uses an as-late-as-possible policy and a priority coefficient

$\text{SWAP}(\theta)$ 0

Benchmarking on lattice, tree, and random graphs reports average emitter-emitter CNOT reduction of 25–37% and up to 52%, average circuit-duration reduction of 32–43% and up to 56%, and photon-loss suppression improvements of $\text{SWAP}(\theta)$ 1 to $\text{SWAP}(\theta)$ 2. The significance is architectural: deterministic state generation becomes a compilation stack that spans graph rewriting, hardware-aware decomposition, and resource-constrained scheduling, rather than a single pulse-level protocol.

3. Gate- and state-level compilation algorithms

Deterministic entanglement compilation also appears in algorithmic form, particularly for native-gate synthesis and amplitude encoding. For two-qudit systems, a complete workflow compiles any two-qudit unitary into an arbitrary, user-chosen native entangling gate set (Mato et al., 2023). The target unitary $\text{SWAP}(\theta)$ 3, of dimension $\text{SWAP}(\theta)$ 4, is reshaped as a virtual single-qudit unitary in dimension $\text{SWAP}(\theta)$ 5 and decomposed by QR factorization as

$\text{SWAP}(\theta)$ 6

After mapping back by $\text{SWAP}(\theta)$ 7, only two standardized two-level entangling gates remain: controlled rotations $\text{SWAP}(\theta)$ 8 and partial swaps $\text{SWAP}(\theta)$ 9. Arbitrary instances are reduced by permutations to canonical forms such as $P_i$ 0 and $P_i$ 1, then decomposed into local operations and a controlled-exchange gate $P_i$ 2. The final step uses parametrized quantum circuits interleaving local unitaries with the native entangling gate,

$P_i$ 3

optimized offline to the desired fidelity. Case studies for the qutrit and ququart CSUM gates report, for $P_i$ 4, 44 controlled rotations, 24 partial swaps, and 184 $P_i$ 5 gates, with layered compilation achieving target infidelity $P_i$ 6 in both $P_i$ 7 and $P_i$ 8 gate sets; for $P_i$ 9, 92 controlled rotations and 36 partial swaps are required.

A different deterministic compiler adapts state-preparation cost to the entanglement of the target state through a matrix product state representation (Gundlapalli et al., 2021). Any $U$ 0-qubit state is decomposed into MPS cores, optionally truncated by singular-value decomposition, and synthesized by $U$ 1 sequential variable-size contiguous blocks. The block size is

$U$ 2

so product structure yields single-qubit gates, while larger bond dimensions trigger multi-qubit gates only where required. This compiler is deterministic and non-variational, but it explicitly allows approximate entanglement reduction to improve actual device performance. For the normal distribution, the reported resource trade-off is 5 CNOTs and depth 44 for the approximate MPS circuit, versus 77 CNOTs and depth 123 for exact isometry, with experimental fidelity 0.980 versus 0.864 at theoretical fidelities 0.990 and 1.0, respectively. The significance is methodological: compilation is driven by entanglement profile rather than by a worst-case isometric template.

Deterministic gate synthesis in photonics has also been formulated at the interferometer level (Scala et al., 2023). In dual-rail encoding, a universal set of single- and two-qubit gates is synthesized by concatenating hopping regions and nonlinear free-propagation segments under the Hamiltonian

$U$ 3

The total unitary is a product of block unitaries, and parameters are optimized by a bound-constrained L-BFGS-B routine. For CNOT and Mølmer–Sørensen gates, fidelities above 99.9% are reported at $U$ 4 with 10–20 blocks. Here the deterministic feature is that the logical subspace is recovered at the output without ancilla modes or probabilistic measurements, even though intermediate amplitudes may transiently occupy non-logical states.

4. Measurement, holonomy, and indefinite causal order

Deterministic entanglement need not arise from direct unitary compilation alone. One major route is measurement plus feedback. In superconducting circuits, a time-resolved continuous parity measurement of two transmon qubits in 3D cQED is engineered so that the measured observable is

$U$ 5

with $U$ 6 in the ideal parity-measurement regime (Ristè et al., 2013). Postselection on the odd-parity outcome yields concurrence $U$ 7 at moderate thresholding, rising to $U$ 8 under more aggressive thresholding. Real-time FPGA feedback then maps the even-parity branch to the target odd-parity Bell state by a conditional $U$ 9-pulse, producing fully deterministic entanglement with 66% fidelity to the target Bell state and deterministic concurrence $\theta_i = 2\arcsin\left[(N-i+1)^{-0.5}\right].$ 0. The conceptual importance is that the stochasticity of measurement is absorbed into a compiled control loop.

An earlier interferometric scheme shows the same logic in abstract form (Lorenzo, 2015). An ancillary particle propagates through a Mach–Zehnder interferometer and interacts with subsystem 1 or subsystem 2 depending on the arm. The interaction is parameterized by

$\theta_i = 2\arcsin\left[(N-i+1)^{-0.5}\right].$ 1

After the second beamsplitter, post-selection on detector outcomes $\theta_i = 2\arcsin\left[(N-i+1)^{-0.5}\right].$ 2 or $\theta_i = 2\arcsin\left[(N-i+1)^{-0.5}\right].$ 3 yields different two-system states; with strong, equal coupling $\theta_i = 2\arcsin\left[(N-i+1)^{-0.5}\right].$ 4, both branches are maximally entangled and a local phase flip on one subsystem converts the $\theta_i = 2\arcsin\left[(N-i+1)^{-0.5}\right].$ 5 branch to the $\theta_i = 2\arcsin\left[(N-i+1)^{-0.5}\right].$ 6 branch. Determinism therefore arises from post-selection followed by conditional feedback, whereas for weak interaction only the success probability is reduced.

A distinct route uses adiabatic non-Abelian holonomy in an integrated photonic chip (Bhattacharya et al., 2024). For total photon number $\theta_i = 2\arcsin\left[(N-i+1)^{-0.5}\right].$ 7, the system has an $\theta_i = 2\arcsin\left[(N-i+1)^{-0.5}\right].$ 8-dimensional dark-state subspace, and adiabatic transport along a closed loop $\theta_i = 2\arcsin\left[(N-i+1)^{-0.5}\right].$ 9 in coupling space implements

$d=2$ 0

Its matrix representation in the dark-state basis is equivalent to the $d=2$ 1-dimensional irreducible representation of the rotation operator with $d=2$ 2. For $d=2$ 3, an input $d=2$ 4 state is transformed into a maximally entangled bipartite qutrit at $d=2$ 5, with entanglement $d=2$ 6 ebits. No measurement or nonlinear photon-photon interaction is required, so determinism is purely geometric.

Indefinite causal order furnishes another remote-generation mechanism (Liu et al., 5 Mar 2025). Multiple parties pre-share a maximally entangled GHZ control state, and each party applies two single-qubit gates to a local target qubit in an order coherently controlled by the corresponding control qubit. Measurement of the control qubits in the $d=2$ 7 basis heralds the generated target-state branch. The protocol generates deterministic and heralded $d=2$ 8-qubit entangled states among distant parties using only pre-shared maximally entangled states, single-qubit operations, and local measurements, while avoiding complex entangled-state measurements and multiple pre-shared entangled states required by conventional entanglement swapping.

5. Quantum-network transmission and deterministic concentration

In quantum networks, deterministic entanglement compilation has been formulated as a recursive transmission theory on series-parallel topologies (Meng et al., 2021). Concurrence percolation theory motivates a deterministic entanglement transmission scheme fully analogous to resistor network analysis. For links with Schmidt vectors $d=2$ 9 and $N=4$ 0, deterministic swapping over a series connection is

$N=4$ 1

while deterministic concentration over parallel links maps the tensor-product Schmidt vector to a majorization-optimal $N=4$ 2-vector $N=4$ 3. The relevant optimality metrics are generalized $N=4$ 4-concurrences,

$N=4$ 5

with topology-dependent statements: simple series optimizes average $N=4$ 6-concurrence, simple parallel optimizes average $N=4$ 7, parallel-then-series optimizes average $N=4$ 8-concurrence, and qubit series-parallel networks optimize worst-case $N=4$ 9-concurrence. The DET scheme is stated to be optimal over the nested repeater protocol for distilling entanglement from partially entangled qubits and to yield higher success probabilities of obtaining a maximally entangled state than classical entanglement percolation. Experimental implementation on IBM’s platform reached about 92.4% fidelity for the series rule and about 78.2% for the parallel rule.

Deterministic concentration in qubit-qudit systems is based on a different principle: Hilbert-space truncation by generalized XZ-basis measurements (J et al., 2024). Starting from $R\rho R$ 0 qubit-qudit states

$R\rho R$ 1

a joint multi-qubit measurement on the qubit subsystems projects $R\rho R$ 2 onto a single logical qubit space $R\rho R$ 3. The protocol is effectively deterministic because all outcomes occur with equal probability $R\rho R$ 4 and all post-measurement states are interconvertible by local unitaries on the qudit side. The multi-qubit measurement is equivalent to a repetitive two-qubit protocol, and for two inputs with concurrences $R\rho R$ 5 and $R\rho R$ 6 the concentrated entanglement is

$R\rho R$ 7

For identical inputs this generalizes to

$R\rho R$ 8

and for unequal inputs the iterative rule

$R\rho R$ 9

implies the bounds

$F=0.77{-}0.82$ 0

The same formalism is used to create generalized GHZ states and, by local $F=0.77{-}0.82$ 1 measurements on ancillary qubits, maximally entangled pairs between arbitrary nodes.

6. Performance metrics, trade-offs, and broader realizations

The metrics used to assess deterministic entanglement compilation are markedly heterogeneous. Photonic sequential emission emphasizes full-state fidelity for small $F=0.77{-}0.82$ 2 and inferred entanglement length for larger $F=0.77{-}0.82$ 3 (Besse et al., 2020). Graph-state compilers optimize emitter-emitter CNOT count, circuit duration, and photon-loss exposure (Ren et al., 20 Mar 2025). Two-qudit native-gate synthesis reports gate counts and target infidelity (Mato et al., 2023). Network protocols optimize generalized concurrence monotones and success probability relative to percolation baselines (Meng et al., 2021). Measurement-based superconducting schemes emphasize concurrence, Bell-state fidelity, and generation rate (Ristè et al., 2013). This suggests that “compilation” in this area is best understood as target-aware optimization under hardware-specific constraints, rather than as a uniform notion of optimal circuit depth.

A second trade-off concerns exactness versus practical fidelity. The amplitude-encoding MPS compiler makes this explicit: reducing theoretical fidelity by SVD truncation can increase actual fidelity on NISQ hardware by lowering entangling-gate count and circuit depth (Gundlapalli et al., 2021). A related trade-off appears in weakly nonlinear photonic interferometers, where moderate rather than extreme nonlinearity is the optimal regime for deterministic entangling gates (Scala et al., 2023). In measurement-based schemes, determinism may reduce to compiled correction of stochastic branches, as in feedback-stabilized parity measurement or strong-interaction ancilla interferometry (Ristè et al., 2013, Lorenzo, 2015).

Adjacent many-body and matter-wave realizations broaden the scope of deterministic entanglement engineering beyond circuit compilation in the narrow sense. Driving a $F=0.77{-}0.82$ 4Rb Bose–Einstein condensate through two consecutive quantum phase transitions yields a near-deterministic twin-Fock state of about $F=0.77{-}0.82$ 5 atoms, with number squeezing of $F=0.77{-}0.82$ 6 dB, normalized collective spin length $F=0.77{-}0.82$ 7, phase sensitivity of about 6 dB beyond the standard quantum limit, entanglement breadth of about 910 atoms, and conversion efficiency $F=0.77{-}0.82$ 8 (Luo et al., 2017). A different proposal uses controlled dissociation of diatomic Feshbach molecules to generate deterministic entangled atom pairs in spin, position-momentum, and path degrees of freedom, with scalable integration into matter-wave waveguide arrays and atomtronic networks (Li et al., 26 Sep 2025). These works are not compilers in the software sense, but they reinforce a broader research trajectory: deterministic entanglement generation is increasingly treated as a systems-engineering problem in which control sequence, resource layout, and target correlation structure are co-designed.

The field therefore has no single canonical architecture. Its unifying principle is the replacement of probabilistic entanglement generation by compiled deterministic workflows: sequential emission with auxiliary disentangling, graph partitioning with local complementation and scheduling, QR- and MPS-based circuit synthesis, feedback-enabled measurement, holonomic transport in degenerate subspaces, concurrence-preserving network recursion, and effectively deterministic concentration by generalized-basis measurements (Besse et al., 2020, Ren et al., 20 Mar 2025, Mato et al., 2023, Gundlapalli et al., 2021, Ristè et al., 2013, Bhattacharya et al., 2024, Meng et al., 2021, J et al., 2024). The technical differences are substantial, but the research direction is coherent: entanglement is increasingly being treated as a resource that can be compiled deterministically across gates, states, networks, and hardware stacks.