Algebraic Quantum Error Correction and Correctable Subalgebras

Exact algebraic quantum error correction asks which von Neumann subalgebra of logical observables can be restored by a normal channel. For a finite Kraus model on a code subspace, a logical algebra is correctable exactly when it commutes with every compressed error product $PE_i^*E_jP$. This replaces the scalar Knill–Laflamme equation and remains meaningful when only a subsystem algebra, rather than every operator on the code, is protected.

Required background. Normal von Neumann-algebra channels fix the Heisenberg orientation, and sufficiency and recovery distinguish an exact recovery theorem from an entropy heuristic. Helpful background. Haag–Kastler nets provide local observable algebras, while type-III entropy limits prevent density-matrix shortcuts. See error models and recovery conditions, operator-algebra QEC, erasure correctability, complementary recovery, locality and code distance, covariance constraints, subsystem codes, tensor-network encodings, and redundant reconstruction.

Correctable algebras and recovery

Let $P$ project onto a code Hilbert space $\mathcal H_C\subset\mathcal H$, and let a Schrödinger noise channel have Kraus operators $E_i$. Its Heisenberg action from output observables to code observables is

$$\mathcal N(A)=P\left(\sum_iE_i^*AE_i\right)P.$$

A von Neumann algebra $\mathfrak A\subset PB(\mathcal H)P$ is exactly correctable if there is a normal unital completely positive recovery map $\mathcal R$ such that

$$(\mathcal N\circ\mathcal R)(A)=A, \qquad A\in\mathfrak A.$$

For a finite or countable Kraus representation with the usual convergence, the criterion is

$$[PE_i^*E_jP,A]=0 \quad\text{for every }A\in\mathfrak A\text{ and all }i,j.$$

Equivalently, all environment observables generated by the error products lie in the commutant $\mathfrak A'$. Bény, Kempf, and Kribs prove the observable-algebra theorem and construct recovery from the commutant condition Bény, Kempf, and Kribs 2007, Theorems 2–3, article 042303.

The proof mechanism uses a Stinespring isometry $V\psi=\sum_iE_i\psi\otimes|i\rangle$. The compressed environment matrix elements are $V^*(1\otimes|i\rangle\langle j|)V=PE_i^*E_jP$. If these commute with $\mathfrak A$, the representation of $\mathfrak A$ can be transported through $V$ without conflict with the environment algebra; a commutant lifting supplies $\mathcal R$. Conversely, exact recovery means the environment cannot distinguish states solely through expectation values of $\mathfrak A$, forcing the commutators to vanish.

For $\mathfrak A=PB(\mathcal H)P$, the commutant consists only of scalars and the criterion reduces to $PE_i^*E_jP=\lambda_{ij}P$, the ordinary Knill–Laflamme relation. The algebraic criterion is weaker when a gauge or complementary subsystem may be damaged.

Knill and Laflamme originally derived the scalar condition by demanding preservation of inner products between all corrupted code words Knill and Laflamme 1997, §III, pp. 904–906. The algebraic generalization replaces scalar error products by elements of $\mathfrak A'$. If $\mathfrak A$ has a nontrivial center, its central projections represent classical superselection labels that may be protected even when off-diagonal coherences between sectors are not part of the logical algebra. Therefore the algebra itself, including its center, is indispensable input to the theorem.

Split-factor erasure model

Let a free net have a split inclusion, represented on

$$\mathcal H_C\simeq\mathcal H_1\otimes\mathcal H_2,$$

and embed a finite matrix algebra $M_k$ in the first type-I factor. Declare

$$\mathfrak A=M_k\otimes1$$

to be the logical observable algebra. Erase the complementary factor by the Schrödinger channel $\operatorname{Tr}_{2}$, restricting to a finite-dimensional auxiliary sector for this explicit calculation. A recovery appends a fixed normal state $\chi_2$:

$$\mathcal R_*(\rho_1)=\rho_1\otimes\chi_2.$$

In the Heisenberg picture, logical observables are embedded as $A\mapsto A\otimes1$, and erasure followed by this recovery fixes every $A\in\mathfrak A$. Choosing an orthonormal basis $|r\rangle$ of the erased factor, the error products are $1\otimes|r\rangle\langle s|$ on the code. They commute with $M_k\otimes1$, so the criterion holds term by term. This realizes the operator-algebra QEC construction inside a type-I factor supplied by a split inclusion; it does not assert that the sharp local field algebra itself is finite dimensional.

There is also an independent distinguishability check. If two code states have different restrictions to $M_k\otimes1$, tracing out factor 2 preserves those restrictions exactly. Therefore every logical measurement retains its outcome statistics and the appended state cannot alter them.

The converse check uses a reference system $R$. Entangle $R$ only with the first factor. Complementary erasure leaves the joint $R\mathcal H_1$ state unchanged, so all correlations with the protected algebra survive. If $R$ is instead entangled with the erased factor, those correlations disappear. This distinguishes exact algebra recovery from recovery of the whole encoded state.

Failure outside the commutant

Add a proposed logical generator $L=1\otimes Z$ on the erased factor. It fails to commute with $1\otimes|r\rangle\langle s|$ for suitable $r,s$. Choose two states with identical first-factor density but orthogonal second-factor eigenstates of $Z$. Erasure maps them to the same output, so no recovery channel can restore the expectation of $L$ for both: distinguishability has already been lost.

This adversarial pair also shows why correctability is algebra-dependent. Failure for $L$ does not damage $M_k\otimes1$, and success for that subalgebra does not recover correlations with the discarded factor. Approximate correctability similarly requires a norm, a state or energy class, and a quantitative information–disturbance theorem.

Exercises

1. Recover the scalar criterion. Show that if $\mathfrak A=PB(\mathcal H)P$, the commutant condition implies $PE_i^*E_jP=\lambda_{ij}P$.

Solution

The commutant of the full matrix algebra on the code is $\mathbb C P$. Each compressed error product lies in that commutant, so it is a scalar multiple of $P$.

2. Gauge subsystem. For $\mathcal H_C=\mathcal H_L\otimes\mathcal H_G$, suppose every error is $1_L\otimes F_i$. Identify a correctable algebra.

Solution

Take $\mathfrak A=B(\mathcal H_L)\otimes1_G$. Every error product $1_L\otimes F_i^*F_j$ commutes with it, so the theorem guarantees exact recovery of all logical observables even if the gauge factor changes irreversibly.

References

• Bény, Cédric, Achim Kempf, and David W. Kribs. “Generalization of Quantum Error Correction via the Heisenberg Picture.” Physical Review A 76 (2007): 042303. DOI.
• Knill, Emanuel, and Raymond Laflamme. “Theory of Quantum Error-Correcting Codes.” Physical Review A 55 (1997): 900–911. DOI.