Equivalent-System Purification in Quantum Systems

• Equivalent-System Purification is an operational framework that represents a mixed state as the marginal of a pure state on an extended system, preserving the reduced observables of interest.
• It employs techniques such as density-matrix purification, virtual subsystems, and localized virtual purification to simulate open-system dynamics and correct measurement noise.
• The approach enables controlled error protection, effective quantum coding, and dynamic analysis of both Markovian and non-Markovian evolutions in quantum settings.

Equivalent-System Purification denotes a class of operational constructions in which a target mixed state, open-system evolution, local observable, or noisy measurement is replaced by a purified or effectively idealized system that reproduces the relevant reduced description. In the literature, the equivalence may be exact, asymptotic, or local: a mixed state on $S$ may be represented as a pure state on $S\otimes E$; a global $k$-copy purification may be replaced by a local purification on a region $R$; many noisy detectors may become equivalent to a single ideal detector after preprocessing; or conditional monitoring may purify the reduced density matrix and dynamically project evolution into a protected code space [(Inoue et al., 2018); (Schlimgen et al., 2022); (Hakoshima et al., 2023); (Dall'Arno et al., 2010); (Gullans et al., 2019); (Chiribella et al., 2009)]. This suggests a unifying criterion: the substitute system must preserve the operational content of interest while relocating entropy, noise, or irreversibility into auxiliary degrees of freedom, measurement records, or discarded subsystems.

1. Conceptual structure and formal definitions

At its most elementary level, purification means representing a mixed state as the marginal of a pure bipartite state. For a spectral decomposition $\rho_S=\sum_i p_i |i\rangle\langle i|$, a standard construction is

$$|\Psi\rangle=\sum_i \sqrt{p_i}\,|i\rangle_S\otimes|i\rangle_E,$$

with $$\rho_S=\operatorname{Tr}_E(|\Psi\rangle\langle\Psi|)$$. In the open-system setting, one may instead demand dynamical equivalence: a pure initial state on $S+B$ must reproduce the same reduced trajectory as a mixed product initial state under the physical joint unitary $\hat V(t)$. In the localized-virtual-purification setting, the relevant object is not the full state but the purified expectation of a local observable, estimated from a reduced state $\rho_R$ by

$S\otimes E$0

In monitored dynamics, purification is dynamical entropy reduction along a quantum trajectory, quantified by

$S\otimes E$1

For noisy measurements, purification means that suitable preprocessing and postprocessing make many noisy detectors equivalent, in their outcome statistics, to a single ideal detector [(Inoue et al., 2018); (Hakoshima et al., 2023); (Gullans et al., 2019); (Dall'Arno et al., 2010)].

Regime	Equivalent object	Canonical expression
State purification	Pure state on $S\otimes E$2	$S\otimes E$3
Local virtual purification	Reduced $S\otimes E$4-copy purification on $S\otimes E$5	$S\otimes E$6
Monitored dynamics	Conditional reduced state $S\otimes E$7	$S\otimes E$8
Noisy measurement purification	Effective ideal POVM statistics	$S\otimes E$9 before purification

Several papers make clear that the equivalence is operational rather than merely representational. In generalized probabilistic theories, purification is unique up to reversible channels on the purifying system, and purifying environments that differ can still be related by reversible dynamics and a pure ancilla. In detector purification, equivalence is defined by equality or convergence of the effective POVM and thus of all outcome probabilities. In localized virtual purification, equivalence means agreement of local purified expectation values up to a controllable error, often exponentially small in the buffer size. This suggests that Equivalent-System Purification is best understood as a family resemblance among constructions that preserve a specified reduced description, rather than as a single universally fixed formalism [(Chiribella et al., 2009); (Dall'Arno et al., 2010); (Hakoshima et al., 2023)].

2. Ancilla, environment, and virtual-subsystem realizations

A central realization of Equivalent-System Purification is the embedding of non-unitary reduced dynamics into unitary dynamics on a larger Hilbert space. In “Quantum Simulation of Open Quantum Systems Using Density-Matrix Purification” and “Unitary Dynamics for Open Quantum Systems with Density-Matrix Purification,” any $k$0 system density matrix is recast as a pure state on $k$1 with bath dimension $k$2, so that

$k$3

The purified state is written as

$k$4

and the total Hamiltonian is

$k$5

These works emphasize the “equivalent-system” case $k$6, contrasting it with conventional Stinespring dilations whose ancilla dimension is bounded above by $k$7. They also state that the purification-based extension map is non-linear in $k$8, allowing the framework to model both Markovian and non-Markovian dynamics and to extend beyond complete positivity while maintaining positivity and normalization at the joint $k$9 level (Schlimgen et al., 2022, Delgado-Granados et al., 2024).

A different environment-based formulation appears in “Typical Purification Reproducing the Time Evolution of an Open Quantum System.” There the problem is state-specific rather than channel-universal: given the actual product initial state $R$0 and the physical joint unitary $R$1, one asks whether there exists a pure initial state on $R$2 whose reduced dynamics reproduces the actual open-system evolution. For a macrosystem bath, the paper constructs

$R$3

with typical bath vectors in the relevant ensemble, proves

$R$4

and provides concentration bounds showing that the purified dynamics reproduces the reference trajectory with overwhelming probability when $R$5, or more generally when

$R$6

This establishes an equivalent pure-state surrogate for mixed-bath evolution under the actual Hamiltonian, not merely under a constructed dilation (Inoue et al., 2018).

The virtual-subsystem formulation sharpens the resource question. “Quantum resources for purification and cooling: fundamental limits and opportunities” assumes initial factorization $R$7 and arbitrary joint unitary controllability. The main theorem states that exact purification is possible if and only if the environment contains a virtual subsystem $R$8 isomorphic to $R$9 and initialized in the desired pure state, within a decomposition

$\rho_S=\sum_i p_i |i\rangle\langle i|$0

Approximate purification is controlled by the environment spectrum through the quantities

$\rho_S=\sum_i p_i |i\rangle\langle i|$1

with exact cooling ruled out unless such a pure virtual subsystem exists. Operationally, purification is implemented by a generalized swap between $\rho_S=\sum_i p_i |i\rangle\langle i|$2 and the virtual copy $\rho_S=\sum_i p_i |i\rangle\langle i|$3 (Ticozzi et al., 2014).

At the broadest structural level, “Probabilistic theories with purification” elevates the idea from Hilbert-space quantum mechanics to causal generalized probabilistic theories. There purification of every mixed state, unique up to reversible channels on the purifying system, is equivalent to the existence of a reversible realization of every physical process. The same framework yields a transformation–state isomorphism with the structural properties of the Choi–Jamiołkowski isomorphism and underlies statements such as no information without disturbance, no cloning, teleportation, no programming, no bit commitment, complementarity between correctable and deletion channels, and characterization of entanglement-breaking channels as measure-and-prepare channels (Chiribella et al., 2009).

3. Dynamical purification under monitoring and repeated measurements

In monitored many-body dynamics, Equivalent-System Purification acquires a distinctly non-equilibrium form. “Dynamical purification phase transitions induced by quantum measurements” studies a system $\rho_S=\sum_i p_i |i\rangle\langle i|$4 interacting with an environment whose outputs are continuously monitored and used for conditioning but not for feedback. For mixed initial states, purification means a decrease of $\rho_S=\sum_i p_i |i\rangle\langle i|$5 along the conditioned trajectory, with instantaneous rate $\rho_S=\sum_i p_i |i\rangle\langle i|$6. In a $\rho_S=\sum_i p_i |i\rangle\langle i|$7D random Clifford circuit with two-site gates and single-site projective $\rho_S=\sum_i p_i |i\rangle\langle i|$8 measurements at rate $\rho_S=\sum_i p_i |i\rangle\langle i|$9, the paper identifies two phases: a locally purifying phase for $$|\Psi\rangle=\sum_i \sqrt{p_i}\,|i\rangle_S\otimes|i\rangle_E,$$0, where entropy decays at a system-size-independent rate and purification time is $$|\Psi\rangle=\sum_i \sqrt{p_i}\,|i\rangle_S\otimes|i\rangle_E,$$1-independent up to logarithmic finite-size effects, and a mixed phase for $$|\Psi\rangle=\sum_i \sqrt{p_i}\,|i\rangle_S\otimes|i\rangle_E,$$2, where the late-time entropy density remains nonzero and the purification time scales as $$|\Psi\rangle=\sum_i \sqrt{p_i}\,|i\rangle_S\otimes|i\rangle_E,$$3. The reported critical point and exponents are

$$|\Psi\rangle=\sum_i \sqrt{p_i}\,|i\rangle_S\otimes|i\rangle_E,$$4

with finite-size scaling

$$|\Psi\rangle=\sum_i \sqrt{p_i}\,|i\rangle_S\otimes|i\rangle_E,$$5

and $$|\Psi\rangle=\sum_i \sqrt{p_i}\,|i\rangle_S\otimes|i\rangle_E,$$6 at intermediate times (Gullans et al., 2019).

The same work ties purification directly to channel capacity and emergent quantum error correction. A nonzero residual entropy density in the mixed phase implies an extensive channel-capacity density and the formation of an error-protected code space. The late-time mixed density matrix defines a code that saturates the single-use channel-capacity bound for the future evolution, is often highly degenerate, and, near $$|\Psi\rangle=\sum_i \sqrt{p_i}\,|i\rangle_S\otimes|i\rangle_E,$$7 in $$|\Psi\rangle=\sum_i \sqrt{p_i}\,|i\rangle_S\otimes|i\rangle_E,$$8D, has contiguous code length scaling $$|\Psi\rangle=\sum_i \sqrt{p_i}\,|i\rangle_S\otimes|i\rangle_E,$$9. The mutual information of an initially completely mixed state grows sublinearly in time in these $$\rho_S=\operatorname{Tr}_E(|\Psi\rangle\langle\Psi|)$$0D random Clifford circuits, a behavior attributed to the dynamical formation of the error-protected subspace. In spatially local $$\rho_S=\operatorname{Tr}_E(|\Psi\rangle\langle\Psi|)$$1D models, the purification transition for mixed states occurs concurrently with the entanglement transition for pure states at the same $$\rho_S=\operatorname{Tr}_E(|\Psi\rangle\langle\Psi|)$$2 and $$\rho_S=\operatorname{Tr}_E(|\Psi\rangle\langle\Psi|)$$3 (Gullans et al., 2019).

“Asymptotic Purification of Quantum Trajectories under Random Generalized Measurements” generalizes the trajectory viewpoint to repeated random measurements in a stationary random environment. The trajectory $$\rho_S=\operatorname{Tr}_E(|\Psi\rangle\langle\Psi|)$$4 is a time-inhomogeneous Markov chain in a random environment with updates

$$\rho_S=\operatorname{Tr}_E(|\Psi\rangle\langle\Psi|)$$5

The paper defines asymptotic purification as almost-sure convergence to pure states, equivalently $$\rho_S=\operatorname{Tr}_E(|\Psi\rangle\langle\Psi|)$$6, and proves an exact criterion: purification occurs if and only if the measurable correspondence of random dark subspaces is empty almost surely. Dark subspaces are those on which the measurement effects are scalar and the Kraus maps act as partial isometries, so the outcomes reveal no internal information and mixedness can persist (Ekblad et al., 2024).

“Purification in Rapid Repeated Interaction Systems” studies a different discrete-to-continuum limit. For a repeated interaction map $$\rho_S=\operatorname{Tr}_E(|\Psi\rangle\langle\Psi|)$$7, purification is possible if and only if the map is non-unital, equivalently $$\rho_S=\operatorname{Tr}_E(|\Psi\rangle\langle\Psi|)$$8, or at the generator level $$\rho_S=\operatorname{Tr}_E(|\Psi\rangle\langle\Psi|)$$9. In ancillary bombardment, the rapid-repetition interpolation is $S+B$0. The work shows that efficient purification requires first-order non-unitality, because decoherence also appears at first order. Pure product couplings $S+B$1 satisfy $S+B$2 and $S+B$3, so purification appears only at third order; by contrast, interactions of the form $S+B$4 purify efficiently when

$S+B$5

Many common light–matter interactions, including single electric multipoles and their linear combinations, are reported not to purify efficiently in this rapid-repetition regime (Grimmer et al., 2016).

4. Localized purification and purification of noisy measurements

Equivalent-System Purification also appears as a localization principle. “Localized Virtual Purification” addresses the measurement overhead of global $S+B$6-copy virtual purification by restricting entangled measurements to the vicinity of the observable of interest. For a local observable $S+B$7 supported on $S+B$8, one chooses a buffer $S+B$9, defines $\hat V(t)$0, and estimates

$\hat V(t)$1

Equivalently,

$\hat V(t)$2

The deviation from the global $\hat V(t)$3-copy estimator is written as an integral of a generalized canonical correlation function, and under exponential clustering, Lieb–Robinson locality, and localization of $\hat V(t)$4 near the boundary, the bias is bounded by terms that decay exponentially in $\hat V(t)$5. For unique gapped ground states in $\hat V(t)$6D, the error is $\hat V(t)$7. Numerically, the method is verified on the $\hat V(t)$8D transverse-field Ising model, on noisy ground states, and on free fermions in $\hat V(t)$9D and $\rho_R$0D; away from the stated conditions it can still converge, but critical systems and $\rho_R$1 free fermions exhibit power-law rather than exponential decay. The resource advantage is explicit: for Gibbs states, the variance bound depends on $\rho_R$2 rather than $\rho_R$3, and the measurement cost is exponentially reduced by the discarded region size (Hakoshima et al., 2023).

A more classical operational equivalence is developed in “Purification of noisy quantum measurements.” There purification means using preprocessing and classical postprocessing so that many noisy detectors become equivalent to a single ideal detector. For qubit $\rho_R$4 measurements, noisy POVM elements are modeled as

$\rho_R$5

and the orthogonal cloning channel

$\rho_R$6

copies the classical $\rho_R$7 information into $\rho_R$8 registers. Maximum-likelihood estimation on the counts yields a variance that decreases with $\rho_R$9 and approaches the ideal lower bound $S\otimes E$00, while the added variance term scales as $S\otimes E$01. For a continuous ensemble of pure qubit states, the mutual information approaches the ideal benchmark

$S\otimes E$02

almost exponentially fast in $S\otimes E$03. For observables with unbounded spectrum, purification is achieved asymptotically by preamplification: photon counting by gain $S\otimes E$04, homodyne detection by squeezing $S\otimes E$05, and heterodyne detection by phase-insensitive gain $S\otimes E$06, with inefficiency-induced noise driven to zero as $S\otimes E$07, $S\otimes E$08, or $S\otimes E$09 (Dall'Arno et al., 2010).

The conceptual commonality between localized virtual purification and measurement purification is the replacement of a costly or noisy global object by an operationally equivalent local or amplified surrogate. In one case the surrogate is a reduced purified subsystem; in the other it is an effective ideal POVM produced by preprocessing. The equivalence criterion remains the same: preservation of the relevant expectation values or outcome statistics [(Hakoshima et al., 2023); (Dall'Arno et al., 2010)].

5. Information-theoretic content, coding, and reversibility

Several strands of the literature connect Equivalent-System Purification to channel capacity, coding, and reversible simulation. In monitored many-body systems, the mixed phase below the purification threshold is not merely slow to purify; it contains a nonzero residual entropy density that implies an extensive quantum channel capacity. The late-time density matrix defines an error-protected code space, often highly degenerate, with encoded information density identified with a channel-capacity density $S\otimes E$10, and the code can be decoded from the measurement record and gate history. In this sense, purification failure at the trajectory level and robust information storage are two sides of the same monitored channel (Gullans et al., 2019).

The generalized probabilistic formulation shows that such links are structural rather than model-specific. If every mixed state has a purification, unique up to reversible dynamics on the purifying system, then every physical process admits a reversible dilation, and conversely. The same framework produces a transformation–state isomorphism, supports teleportation, implies no information without disturbance, excludes universal cloning, and characterizes entanglement-breaking channels as measure-and-prepare channels. Equivalent-system purification is therefore not only a recipe for state preparation; it is a statement about when irreversibility can be represented as reversible dynamics plus discarding (Chiribella et al., 2009).

The virtual-subsystem theorem gives the same idea a resource-theoretic form. Purity cannot be created unitarily; it can only be redistributed. Exact purification or exact ground-state cooling of $S\otimes E$11 is possible if and only if the environment already contains a pure virtual subsystem of the same dimension. Approximate purification is governed entirely by how much spectral weight of $S\otimes E$12 can be concentrated into a $S\otimes E$13-dimensional subspace, and the generalized swap makes this resource conversion explicit (Ticozzi et al., 2014).

The density-matrix-purification approach to open-system simulation extends this reversible logic to non-CP settings. Because the joint $S\otimes E$14 state remains pure and evolves unitarily, the reduced state remains physical even when no single global linear CPTP map on $S\otimes E$15 is assumed. This does not reproduce the universal Stinespring theorem for all inputs; rather, it yields a state-dependent equivalent-system realization whose resource count can be smaller when $S\otimes E$16 suffices (Delgado-Granados et al., 2024, Schlimgen et al., 2022).

6. Validity regimes, limitations, and implementations

The effectiveness of Equivalent-System Purification depends sharply on dynamical, geometric, and control assumptions. Localized virtual purification is most effective for short-range interactions, finite Lieb–Robinson velocity, exponential clustering, and local observables with small supports; it becomes less effective for long-range interactions, criticality with algebraic decay of correlations, strong nonlocal noise, or nonlocal observables. In monitored circuits, the purification perspective remains meaningful even in long-range or all-to-all models where conventional area-versus-volume entanglement language becomes ambiguous, and the all-to-all “bag-of-bits” model exhibits a purification critical point satisfying

$S\otimes E$17

By contrast, the random-dark-subspace theory is formulated for finite-dimensional systems with stationary random environments and does not directly cover infinite-dimensional Hilbert spaces, continuous-time diffusive measurements, or non-stationary noise. In rapid repeated interaction systems, non-unitality remains necessary but, in infinite dimensions, is not sufficient. These works collectively indicate that purification-based equivalences are robust but not universal: their formal strength depends on locality, finite dimensionality, and the existence of suitable environment or measurement structure (Hakoshima et al., 2023, Gullans et al., 2019, Ekblad et al., 2024, Grimmer et al., 2016).

Resource scaling is likewise model-dependent. Global virtual purification suffers from an exponentially large measurement overhead because $S\otimes E$18 decays exponentially with system size; LVP reduces this to dependence on the local region $S\otimes E$19. Density-matrix purification of an $S\otimes E$20-qubit open system uses $S\otimes E$21 bath qubits rather than a worst-case $S\otimes E$22-qubit bath for Stinespring dilation, so the total register size is $S\otimes E$23 rather than up to $S\otimes E$24. The cost is shifted into classical optimization: in the 2022 formulation, a $S\otimes E$25 Hermitian generator with $S\otimes E$26 carries a worst-case parameter count of $S\otimes E$27; in the 2024 formulation, time-dependent $S\otimes E$28 is fitted by optimization, including L-BFGS-B procedures and state-dependent constructions such as

$S\otimes E$29

This suggests a recurrent trade-off between ancilla size and control synthesis complexity (Schlimgen et al., 2022, Delgado-Granados et al., 2024).

Experimental and computational implementations already span several architectures. For monitored purification transitions, proposed observables include purity $S\otimes E$30, local entropy or Rényi-2 entropy, single-reference-qubit entropy $S\otimes E$31, and mutual information $S\otimes E$32; platforms listed include superconducting circuits, trapped ions, and Rydberg arrays, with random brickwork entangling gates interleaved with tunable mid-circuit measurements (Gullans et al., 2019). Localized virtual purification is implemented by local cyclic-shift or controlled-SWAP networks acting only on $S\otimes E$33, with ancilla-based SWAP-test-like circuits and adaptive buffer enlargement when needed (Hakoshima et al., 2023). Density-matrix purification has been executed on quantum-simulator backends and hardware: the 2022 work reports average Frobenius-norm errors $S\otimes E$34 in classical simulations, and on IBMQ Lagos an average density-matrix error $S\otimes E$35 with $S\otimes E$36 shots; the 2024 work describes a $S\otimes E$37-qubit realization on Qiskit FakeBogotaV2, using $S\otimes E$38 unitaries, $S\otimes E$39 CNOT gates and $S\otimes E$40 single-qubit rotations per time step (Schlimgen et al., 2022, Delgado-Granados et al., 2024).

A common misconception is that purification must always mean a static ancilla representation of a mixed state. The surveyed works show a broader landscape: purification can be dynamical and measurement-induced, local rather than global, asymptotic rather than exact, or defined through effective POVM statistics instead of reduced density matrices. Another misconception is that purification is automatically synonymous with complete positivity. The open-system density-matrix-purification framework explicitly distinguishes its non-linear embedding from the linear product-state assignment of standard CP theory, while typical purification and monitored-circuit purification are state-specific or trajectory-specific rather than universal channel constructions (Delgado-Granados et al., 2024, Inoue et al., 2018, Gullans et al., 2019).

Equivalent-System Purification therefore functions as an organizing principle across several branches of quantum information and open-system physics. It unifies ancilla embeddings, virtual subsystems, monitored trajectory purification, local virtual distillation, detector purification, and reversible dilation principles under a single operational theme: the relevant subsystem or process is replaced by an equivalent purified realization that preserves the reduced quantities of interest while exposing the resources, thresholds, and obstructions that govern entropy removal, error protection, and controllable irreversibility.