Quantum Purity Amplification

• Quantum Purity Amplification is a process that transforms several noisy copies of a mixed state into one or more high-fidelity pure states using coherent quantum operations.
• It leverages representation-theoretic methods like Schur–Weyl duality and optimal covariant channels to target principal eigenstates with finite-sample fidelity guarantees.
• Operational implementations span entanglement distillation for QKD, SWAP test-based qubit purification, and continuous-variable schemes employing Gaussian and non-Gaussian strategies.

Quantum purity amplification denotes a family of procedures that transform noisy quantum resources into outputs closer to pure states. In the modern many-copy formulation, the task is to process $n$ copies of a mixed state coherently and output one or more high-fidelity copies of a chosen eigenstate. In entanglement-based quantum communication, the same phrase historically appears under the name quantum privacy amplification, meaning local quantum processing and post-selection that convert noisier shared entanglement into fewer, higher-quality pairs. The literature therefore uses “purity” in two related but nonidentical senses: the strict density-operator quantity $\mathrm{Tr}(\rho^2)$, and operational surrogates such as Bell-state fidelity, quantum bit error rate, secure key rate, or reduced-state purity (Li et al., 20 May 2026, Sohr et al., 2024, Man'ko et al., 2010).

1. Problem formulations and terminological scope

In the general many-copy setting, the input state is written as

$\rho=\sum_{i=1}^d p_i\,\psi_i,\qquad \psi_i=\ketbra{\psi_i}{\psi_i},$

with sorted spectrum $\boldsymbol p=(p_1,\dots,p_d)$, and the task is to amplify the $k$-th eigenstate $\psi_k$. The target maps are

$${}^k\Gamma(\rho)\coloneqq \psi_k^{\otimes m},\qquad {}^k\gamma(\rho)\coloneqq \psi_k,$$

and a QPA protocol is a channel $$\mathcal T:\mathbb H^{\otimes n}\to \mathbb H^{\otimes m}$$ that sends $\rho^{\otimes n}$ to a state close to $\psi_k^{\otimes m}$. The theory distinguishes an all-site criterion, where the joint output should approximate $\mathrm{Tr}(\rho^2)$0, from a one-site criterion, where only a single marginal needs to approximate $\mathrm{Tr}(\rho^2)$1. The default loss is infidelity, though the same framework also treats trace distance, purified distance, Bures-type losses, and cross entropy through monotone relations or pure-target simplifications (Li et al., 20 May 2026).

A complementary principal-eigenstate convention orders the spectrum as

$\mathrm{Tr}(\rho^2)$2

so that the target is the unique top eigenvector $\mathrm{Tr}(\rho^2)$3. In that notation, the success criterion for a channel $\mathrm{Tr}(\rho^2)$4 is

$\mathrm{Tr}(\rho^2)$5

This formulation emphasizes finite-sample fidelity to the principal eigenstate and makes the role of the spectral gap $\mathrm{Tr}(\rho^2)$6 explicit (Scharnhorst et al., 26 May 2026).

The communication-theoretic usage is older and operationally different. In entanglement-based QKD, “entanglement distillation is known as Quantum Privacy Amplification (QPA),” and the protocol acts on a noisy shared entangled resource before classical error correction and classical privacy amplification. In that setting, the paper emphasizes entanglement quality, Bell-state fidelity, QBER, secure key rate, yield, and gain, rather than $\mathrm{Tr}(\rho^2)$7 itself (Sohr et al., 2024).

2. Optimal coherent protocols and representation-theoretic structure

The modern solution of QPA is representation-theoretic. By Schur–Weyl duality,

$\mathrm{Tr}(\rho^2)$8

so $\mathrm{Tr}(\rho^2)$9 decomposes into symmetry sectors labeled by Young diagrams $\rho=\sum_{i=1}^d p_i\,\psi_i,\qquad \psi_i=\ketbra{\psi_i}{\psi_i},$0. Sectorwise optimization is then possible because the global utility is an average over the Schur–Weyl distribution of these sectors. The optimal covariant channel is described by the overhang removal rule: in each sector, one removes $\rho=\sum_{i=1}^d p_i\,\psi_i,\qquad \psi_i=\ketbra{\psi_i}{\psi_i},$1 boxes beginning at row $\rho=\sum_{i=1}^d p_i\,\psi_i,\qquad \psi_i=\ketbra{\psi_i}{\psi_i},$2, exhausting successive overhangs until the required output size is reached. Equivalently, the optimal protocol has a universal three-step form: Schur sampling, application of an intertwining isometry $\rho=\sum_{i=1}^d p_i\,\psi_i,\qquad \psi_i=\ketbra{\psi_i}{\psi_i},$3, and tracing out the environment irrep $\rho=\sum_{i=1}^d p_i\,\psi_i,\qquad \psi_i=\ketbra{\psi_i}{\psi_i},$4 (Li et al., 20 May 2026).

For global depolarizing noise, the optimization can be solved exactly as a semidefinite program over the Choi matrix of the channel. The noisy family is

$\rho=\sum_{i=1}^d p_i\,\psi_i,\qquad \psi_i=\ketbra{\psi_i}{\psi_i},$5

with the ideal input $\rho=\sum_{i=1}^d p_i\,\psi_i,\qquad \psi_i=\ketbra{\psi_i}{\psi_i},$6 Haar-random over pure states in $\rho=\sum_{i=1}^d p_i\,\psi_i,\qquad \psi_i=\ketbra{\psi_i}{\psi_i},$7. The exact optimizer projects, in each Schur sector, onto the mixed-symmetry irrep obtained by removing the bottom-right box, equivalently the smallest feasible row index. This resolves what the paper describes as a two-decade open problem for general finite-dimensional systems under global depolarizing noise. The resulting asymptotic optimal fidelity is

$\rho=\sum_{i=1}^d p_i\,\psi_i,\qquad \psi_i=\ketbra{\psi_i}{\psi_i},$8

and the operational form is again a three-step procedure: Schur-sample the input, apply a unitary correction to a standard tableau, and output the last register (Li et al., 2024).

The same paper also provides concrete implementations. One route is based on generalized quantum phase estimation and Schur-type transforms. A second route is the sparse and shallow three-input circuit SWAPNET12-13, which exactly implements the optimal $\rho=\sum_{i=1}^d p_i\,\psi_i,\qquad \psi_i=\ketbra{\psi_i}{\psi_i},$9 protocol without compiling a full Schur transform. SWAPNET alternates SWAP tests between copies $\boldsymbol p=(p_1,\dots,p_d)$0 and $\boldsymbol p=(p_1,\dots,p_d)$1, and between copies $\boldsymbol p=(p_1,\dots,p_d)$2 and $\boldsymbol p=(p_1,\dots,p_d)$3, and in simulation it reaches near-optimal performance with only a small number of repetitions (Li et al., 2024).

For qubits, a separate simplification is available: repeated random two-qubit SWAP tests can emulate Schur sampling itself. The protocol detects and removes singlet pairs, and once all singlets are removed, the surviving qubits lie in the totally symmetric subspace carrying the optimal purified information. This gives the same purification fidelity as the Schur transform while using only elementary SWAP tests (Brahmachari et al., 7 Aug 2025).

3. Performance laws, finite-sample guarantees, and coherent advantage

In the intensive-output regime with fixed $\boldsymbol p=(p_1,\dots,p_d)$4, the asymptotic all-site loss of optimal QPA obeys

$\boldsymbol p=(p_1,\dots,p_d)$5

In the extensive regime $\boldsymbol p=(p_1,\dots,p_d)$6, the performance develops explicit phase-like regimes: the relevant formula changes when the output rate $\boldsymbol p=(p_1,\dots,p_d)$7 crosses spectral-gap thresholds $\boldsymbol p=(p_1,\dots,p_d)$8. By contrast, the one-site loss still scales as $\boldsymbol p=(p_1,\dots,p_d)$9, so all-site and one-site criteria diverge sharply once $k$0 (Li et al., 20 May 2026).

The nonasymptotic theory makes these statements usable at finite $k$1. For a mixed state with ordered eigenvalues $k$2 and unique top eigenvalue $k$3, there exists an absolute constant $k$4 such that

$k$5

copies suffice to produce a state with fidelity at least $k$6 with $k$7. The proof gives the explicit sufficient condition

$k$8

so the result is dimension-independent, valid for arbitrary spectra satisfying $k$9, and naturally extended to multiple outputs (Scharnhorst et al., 26 May 2026).

A companion line of work places QPA inside coherent quantum inference and compares it with any measurement-mediated alternative. For fixed spectral gap $\psi_k$0, coherent QPA achieves

$\psi_k$1

and for $\psi_k$2 this becomes dimension-independent: $\psi_k$3 By contrast, any incoherent protocol, formalized through the entanglement-breaking limit, requires

$\psi_k$4

For principal eigenstate targets, the paper summarizes the separation as coherent $\psi_k$5 versus incoherent $\psi_k$6. Since $\psi_k$7 for $\psi_k$8 qubits, this is an exponential advantage in system size (Li et al., 20 May 2026).

4. Experimental and algorithmic realizations

A direct experimental realization of QPA in quantum communication uses hyperentangled photon pairs. The input state is entangled in polarization and energy-time, local polarizing beam splitters implement deterministic single-photon two-qubit CNOTs, the target degree of freedom is measured, and post-selection keeps only events with matching target outcomes. The retained resource is the polarization entanglement, now with lower QBER and higher secure key rate. In that proof-of-principle experiment, the authors report a positive-gain region extending down to about $\psi_k$9 QBER in polarization and $${}^k\Gamma(\rho)\coloneqq \psi_k^{\otimes m},\qquad {}^k\gamma(\rho)\coloneqq \psi_k,$$0 in energy-time, extending up to $${}^k\Gamma(\rho)\coloneqq \psi_k^{\otimes m},\qquad {}^k\gamma(\rho)\coloneqq \psi_k,$$1 QBER in either degree of freedom, and a maximum secure-key-rate gain

$${}^k\Gamma(\rho)\coloneqq \psi_k^{\otimes m},\qquad {}^k\gamma(\rho)\coloneqq \psi_k,$$2

They also show regimes in which key generation becomes possible although it was impossible before QPA. The demonstration is explicitly for noise that factorizes across the two degree-of-freedom subspaces, and the implementation carries a fixed $${}^k\Gamma(\rho)\coloneqq \psi_k^{\otimes m},\qquad {}^k\gamma(\rho)\coloneqq \psi_k,$$3 loss from the Franson interferometer (Sohr et al., 2024).

For qubit purification, a simple random-SWAP protocol achieves the same fidelity as Schur-transform purification. Starting from

$${}^k\Gamma(\rho)\coloneqq \psi_k^{\otimes m},\qquad {}^k\gamma(\rho)\coloneqq \psi_k,$$4

the protocol repeatedly performs SWAP tests on random active pairs, removes detected singlets, and returns one surviving qubit. After approximately

$${}^k\Gamma(\rho)\coloneqq \psi_k^{\otimes m},\qquad {}^k\gamma(\rho)\coloneqq \psi_k,$$5

random SWAP tests, the probability of detecting any new singlet drops exponentially with $${}^k\Gamma(\rho)\coloneqq \psi_k^{\otimes m},\qquad {}^k\gamma(\rho)\coloneqq \psi_k,$$6, and after

$${}^k\Gamma(\rho)\coloneqq \psi_k^{\otimes m},\qquad {}^k\gamma(\rho)\coloneqq \psi_k,$$7

the protocol implements weak and unitary Schur sampling with error $${}^k\Gamma(\rho)\coloneqq \psi_k^{\otimes m},\qquad {}^k\gamma(\rho)\coloneqq \psi_k,$$8. The purified single-qubit fidelity converges to the Schur-optimal value with additive error bounded by $${}^k\Gamma(\rho)\coloneqq \psi_k^{\otimes m},\qquad {}^k\gamma(\rho)\coloneqq \psi_k,$$9 (Brahmachari et al., 7 Aug 2025).

A thermodynamic realization concentrates purity into one target qubit at the expense of two auxiliaries. For three nearly identical qubits, the target ground-state population transforms as

$$\mathcal T:\mathbb H^{\otimes n}\to \mathbb H^{\otimes m}$$0

so $$\mathcal T:\mathbb H^{\otimes n}\to \mathbb H^{\otimes m}$$1 whenever $$\mathcal T:\mathbb H^{\otimes n}\to \mathbb H^{\otimes m}$$2. This is a circuit-based entropy-compression protocol rather than heat-bath algorithmic cooling: the gates redistribute entropy across the small register while the control pulses supply work. On IBM superconducting processors, the experiment reports purification capability down to about $$\mathcal T:\mathbb H^{\otimes n}\to \mathbb H^{\otimes m}$$3 mK for the realistic three-qubit purification task, and about $$\mathcal T:\mathbb H^{\otimes n}\to \mathbb H^{\otimes m}$$4 mK in the best low-depth idealized refrigeration implementation. The dominant obstacle is gate noise (Solfanelli et al., 2022).

5. Continuous-variable variants, purity diagnostics, and mitigation-related uses

In continuous-variable optics, the task can interpolate between amplification and purification. One starts from $$\mathcal T:\mathbb H^{\otimes n}\to \mathbb H^{\otimes m}$$5 copies of a noisy coherent state,

$$\mathcal T:\mathbb H^{\otimes n}\to \mathbb H^{\otimes m}$$6

and aims to produce $$\mathcal T:\mathbb H^{\otimes n}\to \mathbb H^{\otimes m}$$7 approximate copies of $$\mathcal T:\mathbb H^{\otimes n}\to \mathbb H^{\otimes m}$$8. A beamsplitter network reduces the problem exactly to a single effective mode with

$$\mathcal T:\mathbb H^{\otimes n}\to \mathbb H^{\otimes m}$$9

When $\rho^{\otimes n}$0, the optimal purification is deterministic, Gaussian, and implemented by a beamsplitter attenuator; deterministic and probabilistic fidelities coincide. When $\rho^{\otimes n}$1, the optimal deterministic process is still Gaussian, realized by a quantum-limited phase-insensitive amplifier, while non-Gaussian heralded operations of the Ralph–Lund type can provide a probabilistic advantage in a finite parameter regime. The paper is explicit that exact purification to a pure coherent state is impossible for genuinely noisy inputs, so the task is one of optimal approximation quantified by global fidelity (Zhao et al., 2017).

Because many protocols need an independent purity benchmark, continuous-variable homodyne tomography supplies a complementary tool. In that setting, purity is defined explicitly as

$\rho^{\otimes n}$2

and for single-mode Gaussian states it can be obtained from the covariance matrix through

$\rho^{\otimes n}$3

The paper also derives an approximate reconstruction-free estimator from the tomographic uncertainty function $\rho^{\otimes n}$4,

$\rho^{\otimes n}$5

and demonstrates experimentally that a homodyne detector can serve as an online “purity meter” for thermal states, with a maximum relative error below about $\rho^{\otimes n}$6 for the $\rho^{\otimes n}$7-based estimator (Man'ko et al., 2010).

Purity can also enter as an auxiliary variable in error mitigation rather than as a direct purification target. Purity-assisted zero-noise extrapolation uses

$\rho^{\otimes n}$8

to infer an effective noise scale under a Pauli-diagonal model, improving or stabilizing extrapolation relative to routine ZNE. The method does not physically amplify the purity of the hardware state; instead, it uses purity loss to estimate cleaner expectation values. The reported effect is modest bias improvement but a wider effective error-rate threshold than routine ZNE, together with lower standard deviation in the examined regimes (Jin et al., 2023).

6. Conceptual distinctions, adjacent notions, and open issues

A persistent source of confusion is that several neighboring ideas are not QPA in the strict state-purification sense. Heralded optical amplification with a deterministic phase-insensitive amplifier followed by photon subtraction improves effective amplitude gain, fidelity to a target coherent state, and phase variance, but the paper does not define purity as $\rho^{\otimes n}$9 and does not prove a purity increase (Jeffers, 2011). “Purification of noisy quantum measurements” improves the effective POVM or detector statistics, not the state itself (Dall'Arno et al., 2010). A one-dimensional photonic-crystal “entanglement degrees amplifier” rebalances amplitudes of specific pure polarization-entangled states and increases an entanglement-entropy measure after renormalization; it is better understood as probabilistic entanglement concentration than mixed-state purity amplification (Wu et al., 2015). In quantum programming languages, “purity” can even mean absence of entanglement with unowned qubits, as in Twist’s type-theoretic notion of a pure expression, which is a semantic separability condition rather than a physical purification process (Yuan et al., 2022).

Even within the direct QPA literature, several limitations remain. The main asymptotic theory assumes a nondegenerate target eigenvalue, and one-site optimality can fail if nonsymmetric output sectors are allowed for $\psi_k^{\otimes m}$0; degenerate spectra and tighter finite-$\psi_k^{\otimes m}$1 analyses remain open directions (Li et al., 20 May 2026). The exact optimal depolarizing protocol is information-theoretically complete, but generic scalable implementations still depend on Schur- and Clebsch–Gordan-type transforms, while explicit sparse constructions such as SWAPNET are currently available only for three inputs (Li et al., 2024). The hyperentangled QKD experiment demonstrates advantage only for noise that factorizes across degree-of-freedom subspaces and suffers the Franson interferometer’s fixed loss (Sohr et al., 2024). Purity-assisted mitigation can require costly purity estimation, and thermodynamic circuit-based purification remains gate-noise limited on present hardware (Jin et al., 2023, Solfanelli et al., 2022).

Taken together, these results support a precise but plural understanding of quantum purity amplification. In the narrow modern sense, it is a coherent many-copy eigenstate-extraction problem with exact symmetry-based optimal channels, dimension-uniform sample complexity, and provable coherent advantages. In operational settings, the same idea appears as entanglement purification for QKD, qubit purification by SWAP tests or thermodynamic entropy concentration, and continuous-variable amplification–purification tradeoffs. What unifies these threads is the conversion of a larger noisy resource into a smaller, cleaner one; what distinguishes them is the target object—eigenstate, entangled pair, coherent state, or certified subsystem purity—and the metric used to quantify “cleaner.”