Entanglement-Defined Controller

• Entanglement-defined controller is a family of quantum control constructions where entanglement properties, rather than fixed gate sequences, specify the control law.
• It encompasses methodologies from gate-level optimization using perfect-entangler functionals to state-targeted protocols and resource-based network configurations.
• Key insights include invariant reduction and target-set enlargement, enabling robust, adaptable control even under noise and system imperfections.

Across several strands of quantum-control and quantum-network research, an entanglement-defined controller can be understood as a control paradigm in which entanglement, rather than a fixed basis-dependent trajectory or a single canonical gate, specifies what is being controlled. In this sense, the controller may be defined by the full set of two-qubit perfect entanglers, by a target multipartite state such as $|D_n^{(m)}\rangle$, by the Schmidt coefficients of a bipartite state under fast local-unitary reduction, or by a shared entangled resource whose LOCC reconfiguration selects a network connectivity graph. A related line of work treats prior entanglement with a controller as the decisive resource enabling communication or coordination tasks that are otherwise impossible (Watts et al., 2014, Malvetti, 2024, Mazza et al., 14 May 2026, Guha et al., 2022).

1. Conceptual scope and principal variants

A recurring pattern across the literature is that the word controller does not always denote a classical supervisory device. In some works, the controller is an optimization objective over entangling gates; in others, it is a reduced dynamical system on entanglement invariants; in still others, it is a multipartite resource state or a third-party system entangled with the data being processed. This suggests that “entanglement-defined controller” is best treated as a family of constructions rather than a single formalism.

Locus of control	Defining object	Representative sources
Two-qubit gate synthesis	Perfect-entangler set; entangling capacity	(Watts et al., 2014, Patra, 2023)
State-engineering protocol	Target Dicke state; $A,B,C$ entanglement vectors; Schmidt coefficients	(Vu et al., 9 Feb 2026, Uskov et al., 2023, Malvetti, 2024)
Quantum-network configuration	Shared graph-state resource; number of disjoint entangled paths; remote control register	(Mazza et al., 14 May 2026, Gyongyosi et al., 2019, Riera-Sàbat et al., 2022)
Coordination resource	Sender-controller entanglement; controller-controller entanglement	(Guha et al., 2022, Truong et al., 2013)
Dynamical phase control	Measurement-plus-feedback transition between entangled and disentangled phases	(Iadecola et al., 2022)

Two distinctions recur. First, some schemes define the controller through a target entanglement property: the optimizer seeks any perfect entangler, any state in a Dicke manifold, or any point on a Schmidt sphere. Second, other schemes define the controller through an entanglement resource: a shared graph state, a maximally entangled state between sender and controller, or an auxiliary control system whose internal state determines which interactions are effectively active. In both cases, entanglement is not merely an output diagnostic; it specifies the control law, the admissible configuration space, or the resource budget.

A common misconception is that such controllers must always be measurement-based or feedback-based. The record is mixed. Some constructions are explicitly closed-loop or measurement-conditioned, but others are deterministic open-loop syntheses based on invariant functionals, collapse-and-revival physics, or fast local-unitary reduction. The unifying feature is not feedback per se, but the fact that entanglement geometry, entanglement reachability, or entanglement accessibility is the operative variable.

2. Gate-level and operator-level formulations

At the two-qubit gate level, the most explicit entanglement-defined objective is the perfect-entangler functional developed for optimal control in $SU(4)$. Any gate $U\in SU(4)$ is written in Cartan form as $U=k_1Ak_2$, with local operations $k_1,k_2\in SU(2)\otimes SU(2)$ and nonlocal content

$$A=\exp\!\left(-\frac{i}{2}\sum_{j=1}^3 c_j\,\sigma_j\otimes\sigma_j\right).$$

The local equivalence class is encoded by the Makhlin invariants

$$g_1=\frac{1}{16}\Re\{(\mathrm{tr}\,m)^2\},\qquad g_2=\frac{1}{16}\Im\{(\mathrm{tr}\,m)^2\},\qquad g_3=\frac{1}{4}\left[(\mathrm{tr}\,m)^2-\mathrm{tr}(m^2)\right],$$

with $m=U_B^TU_B$ in the Bell basis. Perfect entanglers occupy a polyhedral region in the Weyl chamber bounded by

$$c_1+c_2=\frac{\pi}{2},\qquad c_1-c_2=\frac{\pi}{2},\qquad c_2+c_3=\frac{\pi}{2}.$$

The resulting functional

$A,B,C$0

vanishes if and only if $A,B,C$1 is locally equivalent to a perfect entangler, and is otherwise positive. Here $A,B,C$2 is supplemented by a sector identifier $A,B,C$3 obtained from the roots of a cubic determined by the same invariants. The significance of this construction is that the controller is defined by the entangling capability of the reachable gate, not by synthesis of one specified representative such as CNOT (Watts et al., 2014).

This gate-level perspective has a geometric counterpart in the study of entangling capacity. There the entangling capacity of a unitary $A,B,C$4 is its distance to the set of local/product unitaries $A,B,C$5, evaluated only on separable inputs: $A,B,C$6 With the trace norm, this becomes

$A,B,C$7

The dual quantity

$A,B,C$8

is a maximin problem and satisfies $A,B,C$9. Within this framework, generalized control operators are defined by

$SU(4)$0

so the first subsystem selects which $SU(4)$1 acts on the second. Their decisive characterization is that maximal dual entangling capacity,

$SU(4)$2

occurs if and only if there exists $SU(4)$3 such that the vectors $SU(4)$4 are mutually orthogonal. In this line of work, the controller is defined by how far the operator sits from the local-unitary manifold and by how strongly its controlled branches separate product inputs in Hilbert space (Patra, 2023).

Taken together, these two approaches establish two distinct but compatible meanings of entanglement-defined gate control: one invariant-theoretic, based on the perfect-entangler region in Weyl space, and one metric-geometric, based on distance from product unitaries and distance from the separable-state manifold.

3. State-targeted and reduced-coordinate control

A second major variant defines the controller directly through a target entangled state. In repeated-interaction architectures for symmetric Dicke-state preparation, the target $SU(4)$5 determines the controller prescription. The architecture uses two disjoint qubit registers and $SU(4)$6 ancillary shuttle qubits, with total qubit count $SU(4)$7, initial state

$SU(4)$8

and excitation-preserving partial-SWAP collisions

$SU(4)$9

The design variables are $U\in SU(4)$0 and $U\in SU(4)$1, the shuttle–register and intra-register collision angles. One round is

$U\in SU(4)$2

and after $U\in SU(4)$3 rounds the fidelity to the Dicke target is

$U\in SU(4)$4

The objective is

$U\in SU(4)$5

optimized on $U\in SU(4)$6 by multi-start L-BFGS-B with grid spacing $U\in SU(4)$7 rad, $U\in SU(4)$8, maxiter = 100, and ftol = 10^{-6}. A secondary optimization over local $U\in SU(4)$9 rotations aligns phases. The important point is that the controller is not a fixed gate sequence: the desired Dicke state itself defines the policy, and robustness under missing collisions, depolarization, dephasing, and amplitude damping is expressed mainly through an increased number of required rounds rather than catastrophic loss of fidelity (Vu et al., 9 Feb 2026).

A different reduction appears in fast-local-unitary control of bipartite pure states. For a finite-dimensional system on $U=k_1Ak_2$0 with bilinear Schrödinger dynamics,

$U=k_1Ak_2$1

fast local control allows all local-unitary degrees of freedom to be factored out. The reduced control variables are precisely the Schmidt coefficients, i.e. the singular values of the state matrix. The reduced dynamics is

$U=k_1Ak_2$2

with analogous bosonic and fermionic forms based on Autonne–Takagi and Hua factorization. The equivalence theorem states that if $U=k_1Ak_2$3 solves the full system then the ordered singular-value vector solves the reduced system, and conversely any reduced solution lifts to an approximate solution of the full system up to arbitrarily small error and local-unitary gauge. Controllability, stabilizability, reachable-set characterization, and quantum speed limits can therefore be formulated on the Schmidt sphere. In this setting the controller is entanglement-defined because the state space of the reduced system is exactly the space of entanglement invariants for pure bipartite states (Malvetti, 2024).

An analytically different but conceptually related reduction is the three-vector description of pure three-qubit entanglement. A state

$U=k_1Ak_2$4

is mapped to three complex vectors $U=k_1Ak_2$5, built from Plücker coordinates associated with the bipartitions $U=k_1Ak_2$6, $U=k_1Ak_2$7, and $U=k_1Ak_2$8. Local $U=k_1Ak_2$9 operations induce $k_1,k_2\in SU(2)\otimes SU(2)$0 rotations on the corresponding vector, while two-qubit couplings induce $k_1,k_2\in SU(2)\otimes SU(2)$1 rotations on appropriate $k_1,k_2\in SU(2)\otimes SU(2)$2-component assemblies such as $k_1,k_2\in SU(2)\otimes SU(2)$3. The three-tangle and two-tangles become vector invariants,

$k_1,k_2\in SU(2)\otimes SU(2)$4

$k_1,k_2\in SU(2)\otimes SU(2)$5

This representation supports explicit analytic control tasks, including $k_1,k_2\in SU(2)\otimes SU(2)$6-to-GHZ conversion and maximization of $k_1,k_2\in SU(2)\otimes SU(2)$7 under access only to qubits $k_1,k_2\in SU(2)\otimes SU(2)$8 and $k_1,k_2\in SU(2)\otimes SU(2)$9. Here again, the controller is defined by entanglement coordinates rather than by the full wavefunction amplitudes (Uskov et al., 2023).

4. Resource-defined controllers in quantum networks and distributed architectures

In quantum-network settings, the controller may be the entanglement resource itself. A resource-driven framework for configurable entanglement treats a shared multipartite graph-state resource as a latent “whatever channel” that “does not determine a priori which end-to-end entangled links are activated,” but can be configured through LOCC into different entanglement-connectivity graphs. The framework restricts to Generalized Tree-like two-colorable graph states with partition

$$A=\exp\!\left(-\frac{i}{2}\sum_{j=1}^3 c_j\,\sigma_j\otimes\sigma_j\right).$$0

where $$A=\exp\!\left(-\frac{i}{2}\sum_{j=1}^3 c_j\,\sigma_j\otimes\sigma_j\right).$$1 are orchestration qubits and $$A=\exp\!\left(-\frac{i}{2}\sum_{j=1}^3 c_j\,\sigma_j\otimes\sigma_j\right).$$2 are peer qubits. Its structural design parameters are the peer degree

$$A=\exp\!\left(-\frac{i}{2}\sum_{j=1}^3 c_j\,\sigma_j\otimes\sigma_j\right).$$3

the $$A=\exp\!\left(-\frac{i}{2}\sum_{j=1}^3 c_j\,\sigma_j\otimes\sigma_j\right).$$4-rank bridges

$$A=\exp\!\left(-\frac{i}{2}\sum_{j=1}^3 c_j\,\sigma_j\otimes\sigma_j\right).$$5

and the bridge degree

$$A=\exp\!\left(-\frac{i}{2}\sum_{j=1}^3 c_j\,\sigma_j\otimes\sigma_j\right).$$6

Entanglement Rolling, implemented by Pauli-$$A=\exp\!\left(-\frac{i}{2}\sum_{j=1}^3 c_j\,\sigma_j\otimes\sigma_j\right).$$7 measurements on orchestration qubits, supplies the local transition rule. The single-step effect theorem states that a support bridge becomes the center of a star connecting all neighbors of the measured orchestration qubit, while non-bridge neighbors are transferred toward the next orchestration qubit. A corollary states that peer qubits can be made adjacent using a number of $$A=\exp\!\left(-\frac{i}{2}\sum_{j=1}^3 c_j\,\sigma_j\otimes\sigma_j\right).$$8-measurements equal to their peer proximity

$$A=\exp\!\left(-\frac{i}{2}\sum_{j=1}^3 c_j\,\sigma_j\otimes\sigma_j\right).$$9

For $$g_1=\frac{1}{16}\Re\{(\mathrm{tr}\,m)^2\},\qquad g_2=\frac{1}{16}\Im\{(\mathrm{tr}\,m)^2\},\qquad g_3=\frac{1}{4}\left[(\mathrm{tr}\,m)^2-\mathrm{tr}(m^2)\right],$$0, the protocol instantiates the maximum number $$g_1=\frac{1}{16}\Re\{(\mathrm{tr}\,m)^2\},\qquad g_2=\frac{1}{16}\Im\{(\mathrm{tr}\,m)^2\},\qquad g_3=\frac{1}{4}\left[(\mathrm{tr}\,m)^2-\mathrm{tr}(m^2)\right],$$1 of disjoint Bell pairs concurrently. Noise propagation is analyzed with the Noisy Stabilizer Formalism, yielding closed-form updated noise maps and supporting a benchmark in which extracted Bell-pair and GHZ-state fidelities remain above $$g_1=\frac{1}{16}\Re\{(\mathrm{tr}\,m)^2\},\qquad g_2=\frac{1}{16}\Im\{(\mathrm{tr}\,m)^2\},\qquad g_3=\frac{1}{4}\left[(\mathrm{tr}\,m)^2-\mathrm{tr}(m^2)\right],$$2 over a broad noise range (Mazza et al., 14 May 2026).

A more classical network-control variant appears in entanglement access control for the quantum Internet. There the controller regulates access by choosing the number $$g_1=\frac{1}{16}\Re\{(\mathrm{tr}\,m)^2\},\qquad g_2=\frac{1}{16}\Im\{(\mathrm{tr}\,m)^2\},\qquad g_3=\frac{1}{4}\left[(\mathrm{tr}\,m)^2-\mathrm{tr}(m^2)\right],$$3 of end-to-end connection-disjoint entangled paths made available to a user. For a single path $$g_1=\frac{1}{16}\Re\{(\mathrm{tr}\,m)^2\},\qquad g_2=\frac{1}{16}\Im\{(\mathrm{tr}\,m)^2\},\qquad g_3=\frac{1}{4}\left[(\mathrm{tr}\,m)^2-\mathrm{tr}(m^2)\right],$$4 with $$g_1=\frac{1}{16}\Re\{(\mathrm{tr}\,m)^2\},\qquad g_2=\frac{1}{16}\Im\{(\mathrm{tr}\,m)^2\},\qquad g_3=\frac{1}{4}\left[(\mathrm{tr}\,m)^2-\mathrm{tr}(m^2)\right],$$5 links,

$$g_1=\frac{1}{16}\Re\{(\mathrm{tr}\,m)^2\},\qquad g_2=\frac{1}{16}\Im\{(\mathrm{tr}\,m)^2\},\qquad g_3=\frac{1}{4}\left[(\mathrm{tr}\,m)^2-\mathrm{tr}(m^2)\right],$$6

while for $$g_1=\frac{1}{16}\Re\{(\mathrm{tr}\,m)^2\},\qquad g_2=\frac{1}{16}\Im\{(\mathrm{tr}\,m)^2\},\qquad g_3=\frac{1}{4}\left[(\mathrm{tr}\,m)^2-\mathrm{tr}(m^2)\right],$$7 disjoint paths

$$g_1=\frac{1}{16}\Re\{(\mathrm{tr}\,m)^2\},\qquad g_2=\frac{1}{16}\Im\{(\mathrm{tr}\,m)^2\},\qquad g_3=\frac{1}{4}\left[(\mathrm{tr}\,m)^2-\mathrm{tr}(m^2)\right],$$8

The integrated criterion combines connection probability and node fidelity through

$$g_1=\frac{1}{16}\Re\{(\mathrm{tr}\,m)^2\},\qquad g_2=\frac{1}{16}\Im\{(\mathrm{tr}\,m)^2\},\qquad g_3=\frac{1}{4}\left[(\mathrm{tr}\,m)^2-\mathrm{tr}(m^2)\right],$$9

and the local mismatch

$m=U_B^TU_B$0

Access control is implemented by adapting $m=U_B^TU_B$1 until user constraints are met. With $m=U_B^TU_B$2, $m=U_B^TU_B$3, and $m=U_B^TU_B$4, the reported success probabilities are approximately $m=U_B^TU_B$5 for a single path, $m=U_B^TU_B$6 for $m=U_B^TU_B$7, and $m=U_B^TU_B$8 for $m=U_B^TU_B$9. Here the controller is not a gate-level entangler but an allocation mechanism that directly shapes usable end-to-end entanglement, reliability, and fidelity by controlling path multiplicity (Gyongyosi et al., 2019).

A third networked construction uses a remotely controllable auxiliary system $$c_1+c_2=\frac{\pi}{2},\qquad c_1-c_2=\frac{\pi}{2},\qquad c_2+c_3=\frac{\pi}{2}.$$0 coupled by always-on distance-dependent Ising interactions to a target register $$c_1+c_2=\frac{\pi}{2},\qquad c_1-c_2=\frac{\pi}{2},\qquad c_2+c_3=\frac{\pi}{2}.$$1. The control system is encoded into the logical subspace

$$c_1+c_2=\frac{\pi}{2},\qquad c_1-c_2=\frac{\pi}{2},\qquad c_2+c_3=\frac{\pi}{2}.$$2

and the effective coupling to target qubit $$c_1+c_2=\frac{\pi}{2},\qquad c_1-c_2=\frac{\pi}{2},\qquad c_2+c_3=\frac{\pi}{2}.$$3 is

$$c_1+c_2=\frac{\pi}{2},\qquad c_1-c_2=\frac{\pi}{2},\qquad c_2+c_3=\frac{\pi}{2}.$$4

Thus the internal state $$c_1+c_2=\frac{\pi}{2},\qquad c_1-c_2=\frac{\pi}{2},\qquad c_2+c_3=\frac{\pi}{2}.$$5 defines the coupling pattern $$c_1+c_2=\frac{\pi}{2},\qquad c_1-c_2=\frac{\pi}{2},\qquad c_2+c_3=\frac{\pi}{2}.$$6, equivalently $$c_1+c_2=\frac{\pi}{2},\qquad c_1-c_2=\frac{\pi}{2},\qquad c_2+c_3=\frac{\pi}{2}.$$7. If $$c_1+c_2=\frac{\pi}{2},\qquad c_1-c_2=\frac{\pi}{2},\qquad c_2+c_3=\frac{\pi}{2}.$$8 is large enough and $$c_1+c_2=\frac{\pi}{2},\qquad c_1-c_2=\frac{\pi}{2},\qquad c_2+c_3=\frac{\pi}{2}.$$9 is full rank, any desired coupling pattern can be engineered. The target register, initially in $A,B,C$00, can then be driven to states generated by diagonal unitaries, including graph states, hypergraph states, locally maximally entangled states, and GHZ states. The paper explicitly presents gate-sequence, measurement-based, and control-rotation constructions, and it emphasizes increased error tolerance with respect to position fluctuations (Riera-Sàbat et al., 2022).

5. Entanglement as coordination resource, measurement resource, and dynamical order parameter

Another cluster of results defines the controller through prior entanglement with a distinct control system. In coherently controlled quantum networks, Alice’s data can be initially entangled with Charlie’s control system. For orthogonal information-erasing channels, coherent control of channel order or channel choice yields a network $A,B,C$01 with a decoherence-free subspace $A,B,C$02. This enables two tasks that are impossible without the appropriate sender-controller entanglement. Theorem 1 states that a classical $A,B,C$03-it can be communicated with no leakage to the controller through $A,B,C$04 orthogonal information-erasing channels in $A,B,C$05 coherently controlled configurations iff control and target are initially in a $A,B,C$06-dimensional maximally entangled state. Theorem 3 gives the analogous iff statement for establishment of a maximally entangled two-qudit state, with multipartite GHZ distribution extending the same principle to multiple receivers. In this formulation, the controller is neither passive side information nor a classical switch; it is part of the entanglement resource structure that makes the task feasible (Guha et al., 2022).

Measurement-based feedback can also define a controller through its effect on entanglement phases. In probabilistic control of chaos, the classical Bernoulli map is promoted to a many-qubit monitored circuit with chaotic evolution $A,B,C$07 and control operation

$A,B,C$08

where $A,B,C$09 is a local measurement-and-feedback step on qubit $A,B,C$10. The corresponding Kraus operators are

$A,B,C$11

As the control rate $A,B,C$12 varies, the system undergoes a diffusive transition between a chaotic volume-law entangled phase and a disentangled controlled phase. The local order parameter

$A,B,C$13

detects the same transition. The conceptual novelty is that the control transition and the entanglement transition coincide, so the controller is defined by the entanglement structure of the dynamical phase rather than by trajectory stabilization alone (Iadecola et al., 2022).

Entanglement can also improve control even when the plant and all I/O are classical. In a discretized Witsenhausen problem, two controllers $A,B,C$14 and $A,B,C$15 act on a signal line with cost

$A,B,C$16

For strategy classes $A,B,C$17, $A,B,C$18, and $A,B,C$19, the paper constructs a family $A,B,C$20 for which

$A,B,C$21

while

$A,B,C$22

The entangled controllers still read and write only classical signals, but they share a bipartite quantum state and use local measurements indexed by their classical inputs. This shows that an entanglement-defined controller need not control a quantum system at all; it may instead be a classically interfaced controller whose internal coordination resource is entanglement (Truong et al., 2013).

A deterministic, non-feedback version is the coherent control of multipartite entanglement in remote Jaynes–Cummings subsystems. $A,B,C$23 remote qubits begin in a GHZ state, each interacts with its own local coherent field $A,B,C$24, and the coherent-state amplitude fixes the collapse-and-revival timescale. The central distinction is between all-party entanglement and weak inseparability. For $A,B,C$25, “the death of all-party entanglement is permanent after an initial collapse,” whereas weak inseparability can be “deterministically managed for an arbitrarily large number of qubits almost indefinitely.” The entanglement trajectory is interpreted geometrically as motion relative to the convex sets of biseparable and fully separable states, so the controller is defined by how it steers the system across separability boundaries (Rafsanjani et al., 2014).

6. Recurring principles, limitations, and points of interpretation

Several technical principles recur across these otherwise diverse constructions. One is invariant reduction. Makhlin invariants remove basis dependence in two-qubit gate control; Schmidt coefficients quotient out fast local unitaries in bipartite control; the $A,B,C$26 vectors package three-qubit entanglement into rotational coordinates; and network resource frameworks characterize admissible reconfigurations independently of a specific implementation mechanism. This suggests that entanglement-defined control tends to appear when local gauge degrees of freedom are stripped away and the remaining variables are entanglement invariants or entanglement accessibility variables (Watts et al., 2014, Malvetti, 2024, Uskov et al., 2023, Mazza et al., 14 May 2026).

A second principle is target-set enlargement. Optimizing over the full perfect-entangler region rather than a single gate enlarges the accessible target space; optimizing over Dicke-state fidelity over repeated collisions chooses whichever collision strengths and minimal round count best realize the target under constraints; resource-driven network control resolves a latent shared state into any admissible connectivity graph rather than a single hardwired link pattern. A plausible implication is that entanglement-defined controllers are especially useful when the reachable set contains many locally inequivalent or topologically distinct entangled realizations (Watts et al., 2014, Vu et al., 9 Feb 2026, Mazza et al., 14 May 2026).

The limitations are equally consistent. Some constructions are only nontrivial when the reachable set has sufficient dimensionality: if the two-qubit reachable set collapses essentially to a one-dimensional line, optimizing over all perfect entanglers offers no real advantage over direct gate synthesis. Some tasks require maximal prior entanglement as a necessity, not merely as a convenience: the communication and entanglement-distribution results for coherently controlled erasing channels are iff theorems. Some controllers preserve weaker entanglement notions but not stronger ones: in the Jaynes–Cummings collapse-and-revival model, weak inseparability can revive for large $A,B,C$27, while genuine multipartite entanglement cannot for $A,B,C$28. And in classical control enhanced by entanglement, the separation is established for a specially constructed family rather than for all control circuits (Watts et al., 2014, Guha et al., 2022, Rafsanjani et al., 2014, Truong et al., 2013).

A final source of confusion concerns robustness. In several papers robustness does not mean invariance of the optimal protocol under noise; rather, it means that the controller can trade other resources for performance. In the Dicke-state collision model, imperfections “manifest primarily as a modest increase in the required number of collision rounds.” In configurable network entanglement, the Noisy Stabilizer Formalism yields closed-form effective noise maps rather than a claim of noise immunity. In access-controlled quantum networks, reliability improves by granting additional disjoint paths. Robustness therefore often appears as reallocation in entanglement space, time, or path multiplicity rather than as elimination of noise sensitivity (Vu et al., 9 Feb 2026, Mazza et al., 14 May 2026, Gyongyosi et al., 2019).

Under this synthesis, the entanglement-defined controller is not a single architecture but a recurrent research motif: control is formulated in the language of entanglement regions, entanglement spectra, entanglement vectors, entanglement resources, or entanglement-assisted coordination. What unifies these formulations is that entanglement is promoted from a performance metric to the defining variable of the control problem itself.