Dicke State Initialization

• Dicke state initialization is the preparation of a many-body symmetric state with a fixed excitation number, defined by equal-weight superpositions in qubit systems.
• Techniques include projection-based methods, deterministic unitary circuit synthesis, and adiabatic or dissipative routes, each balancing resources like depth and ancilla requirements.
• Experimental implementations on platforms such as superconducting circuits, optical systems, and silicon-based spins demonstrate the practical impact of exploiting permutation symmetry.

Dicke state initialization is the preparation of a many-body state in a fixed-excitation, permutation-symmetric sector of Hilbert space. For qubits, the standard Dicke state of weight $k$ on $n$ sites is

$$\ket{D_{n,k}}=\binom{n}{k}^{-1/2}\sum_{\substack{x\in\{0,1\}^n\ \mathrm{HW}(x)=k}}\ket{x},$$

the equal-amplitude superposition of all computational-basis strings with Hamming weight $k$. In collective-spin notation it is the fully symmetric state $\ket{J=n/2,\;m=k-n/2}$. Initialization methods exploit this symmetry in distinct ways: some project onto Dicke eigenspaces by measurement, some synthesize exact unitaries, some use dissipative pumping into a steady state, and others follow adiabatic or resonance-selective trajectories through the Dicke ladder (Wang et al., 2021, Bärtschi et al., 2019).

1. State structure and symmetry classes

For spin-$\tfrac12$ systems, Dicke states are the orthonormal basis of the fully symmetric subspace, and the relevant label is the excitation number $k$, or equivalently the collective-spin magnetization $m=k-n/2$. In this representation, the symmetric sector of $N$ qubits has dimension $N+1$, which is the basis used explicitly in symmetry-reduced control constructions for Ising-coupled systems (Stojanovic et al., 2023).

The literature also distinguishes symmetric and asymmetric Dicke states. In the donor-spin silicon proposal, Dicke states are defined as common eigenstates of $n$0 and $n$1, with fully symmetric states corresponding to maximal total spin $n$2, while states with $n$3 are termed asymmetric Dicke states. That work further introduces a complete many-body basis $n$4 obtained by iterative partitioning of the nuclei into subsets, so Dicke-state initialization is not restricted to the fully symmetric manifold alone (Luo et al., 2011).

Higher-dimensional generalizations preserve the same organizing principle. Spin-$n$5 Dicke states are defined by repeated action of the total lowering operator,

$n$6

with $n$7, and reduce to ordinary qubit Dicke states for $n$8 (Nepomechie et al., 2024). A distinct generalization is the $n$9 qudit Dicke state $$\ket{D_{n,k}}=\binom{n}{k}^{-1/2}\sum_{\substack{x\in\{0,1\}^n\ \mathrm{HW}(x)=k}}\ket{x},$$0, the symmetric superposition of all permutations of a multiset with occupation vector $$\ket{D_{n,k}}=\binom{n}{k}^{-1/2}\sum_{\substack{x\in\{0,1\}^n\ \mathrm{HW}(x)=k}}\ket{x},$$1, $$\ket{D_{n,k}}=\binom{n}{k}^{-1/2}\sum_{\substack{x\in\{0,1\}^n\ \mathrm{HW}(x)=k}}\ket{x},$$2 (Nepomechie et al., 2023). These extensions matter because they recast initialization as preparation of symmetry-constrained states in either collective-spin or occupation-number language rather than as a qubit-specific task.

2. Projection-based and heralded initialization

A direct way to initialize a Dicke state is to measure the collective quantum number that labels the symmetric sector. In the phase-estimation protocol for a spin ensemble, the $$\ket{D_{n,k}}=\binom{n}{k}^{-1/2}\sum_{\substack{x\in\{0,1\}^n\ \mathrm{HW}(x)=k}}\ket{x},$$3 spin qubits are collectively coupled to one ancilla qubit through a $$\ket{D_{n,k}}=\binom{n}{k}^{-1/2}\sum_{\substack{x\in\{0,1\}^n\ \mathrm{HW}(x)=k}}\ket{x},$$4-type interaction,

$$\ket{D_{n,k}}=\binom{n}{k}^{-1/2}\sum_{\substack{x\in\{0,1\}^n\ \mathrm{HW}(x)=k}}\ket{x},$$5

or an equivalent collective Ising interaction. Because Dicke states have well-defined total $$\ket{D_{n,k}}=\binom{n}{k}^{-1/2}\sum_{\substack{x\in\{0,1\}^n\ \mathrm{HW}(x)=k}}\ket{x},$$6, they are eigenstates of the relevant unitary with eigenphase determined by the excitation number $$\ket{D_{n,k}}=\binom{n}{k}^{-1/2}\sum_{\substack{x\in\{0,1\}^n\ \mathrm{HW}(x)=k}}\ket{x},$$7. Iterative quantum phase estimation then reads out the binary digits of that phase, and the measurement outcome projects the spin ensemble onto the corresponding Dicke state. Since $$\ket{D_{n,k}}=\binom{n}{k}^{-1/2}\sum_{\substack{x\in\{0,1\}^n\ \mathrm{HW}(x)=k}}\ket{x},$$8 takes $$\ket{D_{n,k}}=\binom{n}{k}^{-1/2}\sum_{\substack{x\in\{0,1\}^n\ \mathrm{HW}(x)=k}}\ket{x},$$9 values, the scheme uses $k$0 ancilla measurements; the depth in ancilla rounds is therefore $k$1. The method is non-deterministic in the measurement-outcome sense, but conditional on the observed bit string the output state is the definite Dicke state $k$2 (Wang et al., 2021).

The same projection logic extends to qudits. For spin-$k$3 Dicke states, one may first prepare a product state $k$4 whose Dicke-basis amplitudes are biased toward the target $k$5, and then apply quantum phase estimation to the operator

$k$6

where $k$7 counts total digit sum. Measuring the QPE register projects onto $k$8. The paper gives a log-depth version with depth $k$9 and a constant-depth variant using higher-dimensional ancillas and fan-out, at the cost of probabilistic repetition (Kerzner et al., 17 Jul 2025).

Heralded linear-optical generation is another measurement-conditioned route, but it differs sharply from destructive postselection. In the LQG-based optical protocol, arbitrary $\ket{J=n/2,\;m=k-n/2}$0 states are generated with $\ket{J=n/2,\;m=k-n/2}$1 photons by encoding the target symmetry in a Dicke digraph $\ket{J=n/2,\;m=k-n/2}$2. The final state arises from the superposition of disjoint cycle covers of that digraph, and successful ancillary detections herald the run without destroying the output Dicke state. The protocol is explicitly designed to avoid the usual drawback of optical postselection, namely consumption of the resource state during verification (Kang et al., 24 Dec 2025).

3. Deterministic unitary circuit synthesis

Deterministic circuit constructions typically begin from an easily prepared unary seed state such as $\ket{J=n/2,\;m=k-n/2}$3 and implement a Dicke-state unitary $\ket{J=n/2,\;m=k-n/2}$4. A foundational ancilla-free method is the inductive “split and cyclic-shift” construction, based on the recurrence

$\ket{J=n/2,\;m=k-n/2}$5

Its circuit size is $\ket{J=n/2,\;m=k-n/2}$6, its depth is $\ket{J=n/2,\;m=k-n/2}$7, it requires no ancillas, and these asymptotics hold even on Linear Nearest Neighbor architectures (Bärtschi et al., 2019).

Subsequent work reorganized this recursion into divide-and-conquer form. One line of results introduces a weight-distribution block $\ket{J=n/2,\;m=k-n/2}$8 that coherently splits a unary-encoded Hamming weight between two subregisters with the correct binomial amplitudes, enabling recursive preparation on all-to-all and grid connectivities. This yields deterministic, ancilla-free circuits with total gate count $\ket{J=n/2,\;m=k-n/2}$9, depth $\tfrac12$0 for all-to-all connectivity, and depth $\tfrac12$1 for suitable grid layouts (Bärtschi et al., 2022).

A later refinement compresses the divide layers further by mixing unary, one-hot, and binary encodings inside the recursion. For all-to-all connectivity, the resulting depth is

$\tfrac12$2

improving the previous $\tfrac12$3 bound. On an $\tfrac12$4-grid, the same balanced recursion gives depth $\tfrac12$5 when $\tfrac12$6, and an optimal-depth $\tfrac12$7 construction when $\tfrac12$8. The paper also proves lower bounds of $\tfrac12$9 for all-to-all connectivity and $k$0 for grid connectivity, establishing near-optimality in the corresponding regimes (Yuan et al., 21 May 2025).

Hardware-oriented synthesis has focused on constant factors as much as asymptotics. On IBM heavy-hex devices, a divide-and-conquer implementation combined with improved Dicke-state unitaries was experimentally evaluated up to $k$1 qubits. For $k$2, the reported fidelity is about $k$3, compared with a previous best reported value of $k$4, and measurement error mitigation typically increases the measured quantum fidelity by about $k$5 absolute (Aktar et al., 2021). A related gate-level optimization study showed that exploiting partially defined unitaries can reduce the generic deterministic circuit to

$k$6

CNOT gates and

$k$7

single-qubit gates for $k$8, and that the improved $k$9 circuit performs better on ibmqx2 than the earlier realization (Mukherjee et al., 2020).

At the complexity-theoretic frontier, constant-depth exact preparation is known under stronger gate models. For $m=k-n/2$0, there is a clean depth-$m=k-n/2$1 $m=k-n/2$2 circuit of polynomial size that maps $m=k-n/2$3 to $m=k-n/2$4, and for $m=k-n/2$5 this becomes a plain $m=k-n/2$6 construction using only multi-qubit Toffoli gates and single-qubit unitaries (Gretta et al., 16 Apr 2026).

4. Driven-dissipative, adiabatic, and resonance-selective routes

Not all initialization schemes are gate-synthesis problems. In donor nuclear spins in silicon, cooperative pumping mediated by coupled donor electrons turns the electron system into a dissipative bridge that transfers population through a ladder of collective nuclear-spin states until the target Dicke state is the unique steady state. Only in situ Kane controls are required—electron initialization, ac A-gate control of hyperfine coupling, and J-gate control of exchange coupling—and the preparation is deterministic because the target is a steady state rather than the outcome of a finely timed coherent pulse. Numerical simulations with realistic parameters show that Dicke states of 10–20 qubits can be prepared with high fidelity; for 20 qubits, symmetric states are prepared sequentially with probability $m=k-n/2$7 in one control mode and $m=k-n/2$8 in an optimized faster mode, with preparation time $m=k-n/2$9 (Luo et al., 2011).

Adiabatic protocols instead reshape the Hamiltonian so that a simple initial ground state morphs into a Dicke-state ground state. In a collective atomic ensemble or two-mode Bose–Einstein condensate, one may start from the spin coherent ground state of $N$0 and interpolate toward

$N$1

whose ground state is $N$2. Counterdiabatic driving adds approximate compensating operators $N$3 to suppress diabatic leakage. For $N$4 and short evolution time $N$5, the fidelity rises from about $N$6 without compensation to about $N$7 with $N$8, $N$9 with $N+1$0 and $N+1$1, and essentially unit fidelity with three or four operators (Opatrný et al., 2015).

Rapid adiabatic passage in the Dicke basis uses a different spectral feature: the avoided-crossing ladder created by one-axis twisting. With

$N+1$2

a chirped longitudinal field $N+1$3 and transverse coupling $N+1$4 drive the system sequentially through neighboring Dicke states. Starting from the coherent spin state $N+1$5, the passage can be halted at the chosen avoided crossing to prepare $N+1$6 or interrupted earlier to generate a superposition relevant to extreme spin-squeezed metrological states. For small to moderate atom numbers, explicitly discussed up to roughly $N+1$7, the protocol is reported to be robust over a large parameter region, and the total time can be approximately independent of $N+1$8 if the chirp rate is scaled appropriately (Carrasco et al., 2023).

Selective-interaction schemes achieve initialization by isolating individual Dicke-ladder transitions. In the Dicke-Stark model, the Stark term makes the detunings depend on both photon number $N+1$9 and Dicke excitation number $n$00, so tuning $n$01 can render only one chosen transition resonant. For $n$02, successive selective TC and anti-TC resonances move the system from $n$03 to $n$04 in timed $n$05 pulses. The same machinery yields GHZ-state generation with reported final fidelity $n$06 for the chosen parameters (Mu et al., 2020).

A closely related deterministic idea uses uniform all-to-all Heisenberg exchange. Starting from a Dicke state $n$07 and appending one qubit, simple time evolution under the exchange Hamiltonian plus a local $n$08 correction deterministically expands either $n$09 or $n$10. This gives an $n$11-step preparation method for arbitrary Dicke states, and a generalized all-coupled Hamiltonian can reduce the depth further in favorable cases (Sharma et al., 2020).

5. Symmetry-restricted control and physical implementations

Initialization protocols often rely less on universal controllability than on controllability inside the symmetric manifold. For three qubits with all-to-all Ising coupling and global transverse controls,

$n$12

the relevant task lives in the four-dimensional symmetric sector spanned by the Dicke basis $n$13. Using the state-to-state controllability result attributed to Albertini and D’Alessandro for permutationally invariant initial and final states, a five-stage NMR-type sequence—three instantaneous global rotations and two finite Ising evolutions—prepares $n$14 from $n$15 with total duration

$n$16

and generalizes to $n$17 with $n$18 (Stojanovic et al., 2023).

The physical platforms proposed for Dicke-state initialization are correspondingly diverse. The phase-estimation projection scheme was analyzed for an ensemble of electronic spins at diamond NV centers collectively coupled to a superconducting flux qubit ancilla (Wang et al., 2021). Cooperative pumping was formulated in the Kane-style silicon donor architecture (Luo et al., 2011). Rapid adiabatic passage and counterdiabatic schemes target collective atomic systems and two-mode condensates (Opatrný et al., 2015, Carrasco et al., 2023). Exchange-based expansion was cast into a spintronic architecture of static spin qubits coupled indirectly through flying qubits emitted by a ferromagnetic reservoir (Sharma et al., 2020). Heralded linear-optical generation uses dual-rail qubits, asymmetric multiports, and $n$19-port Fourier multiports in an LQG translation of the Dicke digraph (Kang et al., 24 Dec 2025).

This variety reflects a common structural point: Dicke-state initialization is easiest when the hardware natively preserves permutation symmetry or provides a collective control primitive. A plausible implication is that the distinction between “algorithmic” and “physical” preparation is often secondary to whether the device exposes the right collective conserved quantities.

6. Robustness, limitations, and conceptual boundaries

The main error channels depend strongly on the preparation paradigm. In ancilla-based phase estimation, fidelity is reduced by phase-estimation precision limits, ancilla decoherence, imperfect global $n$20 coupling, measurement errors, and finite coherence times; because the protocol uses multiple sequential interrogations, errors accumulate, and feasibility requires interaction strength large enough that the total phase-estimation time remains short compared with decoherence times (Wang et al., 2021). In cooperative pumping, unwanted processes arise from off-resonant coupling to other electron excited states, leakage due to finite $n$21-dependent detuning, and nuclear dephasing, with leakage scaling roughly as $n$22 when $n$23 (Luo et al., 2011). In global-control pulse engineering, robustness analysis over the eight pulse parameters shows that the most sensitive parameter is the first rotation angle $n$24, while two-parameter error landscapes are approximately elliptical because the infidelity is quadratic to leading order near the optimum (Stojanovic et al., 2023).

A recurring boundary condition is preservation of the symmetric sector itself. In the driven Dicke model realized by transmons coupled to a 1D waveguide, the many-body reduction to collective operators $n$25 is valid when all transmons are identically driven, identically detuned, and placed at integer multiples of the wavelength. If the system is initialized in the symmetric ground state under those conditions, it remains in the $n$26 sector. However, the steady-state manifold is highly degenerate in the perfect model, and infinitesimal perturbations such as local dephasing, drive-phase gradients, spacing errors, or individual detunings can lift that degeneracy and qualitatively change the steady state, including selection of subradiant or dimerized attractors (Tong et al., 2024).

This should be distinguished from a different claim about superradiance. A non-Dicke initial state with nonzero dipole moment can still produce a superradiant burst; in that analysis, the decisive mechanism is decrease in the dispersion of the quantum phase, leading to synchronization of dipole-envelope phases, and the paper states explicitly that “the Dicke state is not necessary for SR” (Nefedkin et al., 2016). Taken together, these results suggest a precise conceptual boundary: Dicke-state initialization is not universally necessary for collective radiative enhancement, but it is necessary if the intended protocol depends on remaining inside an invariant permutation-symmetric Dicke manifold.

Higher-dimensional generalizations inherit the same trade-offs. Spin-$n$27 Dicke states admit a deterministic ancilla-free recursive circuit with total gate count $n$28 (Nepomechie et al., 2024); arbitrary qudit Dicke states admit deterministic recursive circuits with size/depth $n$29 in the general $n$30-level construction (Nepomechie et al., 2023); and exact canonical MPS representations provide deterministic sequential preparation, while QPE provides lower-depth probabilistic alternatives for both $n$31 spin-$n$32 and $n$33 Dicke states (Kerzner et al., 17 Jul 2025). The broad picture is therefore not a single optimal initializer, but a family of symmetry-exploiting constructions whose best choice depends on whether the limiting resource is depth, ancilla count, measurement overhead, dissipative control, or hardware-native collective interactions.