Generalizations of Pauli channels

Denes Petz, Hiromichi Ohno

Introduction

As a generalization of the Pauli channel on a qubit, we define a linear mapping α:Mn→Mn\alpha:M_{n}\to M_{n} such that

Let σ0=I\sigma_{0}=I and σ1,σ2,σ3\sigma_{1},\sigma_{2},\sigma_{3} be Pauli matrices, i.e.,

and let E:M2→M2{\cal E}:M_{2}\to M_{2} be defined as

Density matrices are sent to density matrices if and only if

It is not difficult to compute the representing block matrix X:=∑i,jE(Eij)⊗EijX:=\sum_{i,j}{\cal E}(E_{ij})\otimes E_{ij}, we have

According to Choi’s theorem the positivity of this matrix is equivalent to the complete positivity of E{\cal E}. XX is unitarily equivalent to the matrix

This matrix is obviously positive if and only if

This is necessary and sufficient condition of complete positivity. □\square

It is not obvious that condition (3) is symmetric in the three variables λ1,λ2,λ3\lambda_{1},\lambda_{2},\lambda_{3}. Condition (3) actually determines the tetrahedron which is the convex hull of the points (1,1,1)(1,1,1), (1,−1,−1)(1,-1,-1), (−1,1,−1)(-1,1,-1) and (−1,−1,1)(-1,-1,1).

Now we show the idea leading to the generalization. The mapping E{\cal E} in (2) has the form

From the expansion of E(σj){\cal E}(\sigma_{j}) we can get equations and the solution is the following:

If μi≥0\mu_{i}\geq 0 for every ii, then E{\cal E} is a completely positive mapping. Therefore,

or together this is (3). (Actually, this argument gives that (3) is a sufficient condition for the complete positivity.)

Pauli channels form an important and popular subject in quantum information theory . The mappings (1) were studied in the paper in the case when the subalgebras are maximal Abelian and pairwise complementary. Our method is different and we allow non-commutative subalgebras as well.

The mapping (1) restricted to Ai{\cal A}_{i} has the form

on density matrices DD. If 0≤λi≤10\leq\lambda_{i}\leq 1, then we can say that DD does not change with probability λi\lambda_{i} and with probability 1−λi1-\lambda_{i} it is sent to the tracial state. Such mappings are usually called as depolarizing channels .

A simple example including non-commutative subalgebras is the following.

Consider M4=M2⊗M2M_{4}=M_{2}\otimes M_{2} and the complementary F-subalgebras A1,…,A4{\cal A}_{1},\dots,{\cal A}_{4} generated by the following triplets of unitaries:

We take also the M-subalgebra A5{\cal A}_{5} generated by σ1⊗σ1,σ2⊗σ2,σ3⊗σ3\sigma_{1}\otimes\sigma_{1},\sigma_{2}\otimes\sigma_{2},\sigma_{3}\otimes\sigma_{3}. The conditional expectations Ej:M4→AjE_{j}:M_{4}\to{\cal A}_{j} are convex combinations of automorphisms

where Uj1=IU_{j1}=I and UjiU_{ji}’s are orthogonal unitaries from Aj′{\cal A}_{j}^{\prime}. Since A5{\cal A}_{5} is an M-subalgebra, A5′=A5{\cal A}_{5}^{\prime}={\cal A}_{5}. The subalgebras A1′,…,A4′{\cal A}_{1}^{\prime},\dots,{\cal A}_{4}^{\prime} are F-subalgebras generated by the following unitaries:

(The above triplets generating Aj{\cal A}_{j} and Aj′{\cal A}_{j}^{\prime} (1≤j≤41\leq j\leq 4) are Pauli triplets, see for details.) Moreover,

The linear mapping (1) has the concrete form

where the conditional expectations EjE_{j} is expressed by the commutant, see (4). (The condition for complete positivity of α\alpha is in Theorem 4.) □\square

Our main result is the necessary and sufficient condition for the complete positivity of mappings like (1) which can be called generalized Pauli channel.

Generalized Pauli channels

Let A{\cal A} be a (unital *-) subalgebra of MnM_{n}. Our aim is to describe the conditional expectation onto A{\cal A} by means of an orthogonal system in the commutant.

Up to unitary equivalence, a subalgebra A{\cal A} of MnM_{n} can be written as

The commutant A′{\cal A}^{\prime} in MnM_{n} is

Let N=∑i=1kni2N=\sum_{i=1}^{k}n_{i}^{2} and let PiP_{i} be a minimal central projection of A{\cal A}, that is, Pi=Ini⊗ImiP_{i}=I_{n_{i}}\otimes I_{m_{i}}.

Let {Ui}i=1N\{U_{i}\}_{i=1}^{N} be an orthonormal basis of A{\cal A}. Then the completely positive map FF from MnM_{n} onto A′{\cal A}^{\prime} given by

In particular, if all ni/min_{i}/m_{i} are equal, then ndimAF\frac{n}{{\rm dim}{{\cal A}}}F is the trace-preserving conditional expectation from MnM_{n} onto A′{\cal A}^{\prime}.

Proof: If all ni/min_{i}/m_{i} are equal, then their ratio is equal to dimAn{{\rm dim}{\cal A}\over{n}}. Therefore it is sufficient to prove the first assertion.

Let {eij(l)}i,j=1nl\{e_{ij}^{(l)}\}_{i,j=1}^{n_{l}} and {fij(l)}i,j=1ml\{f_{ij}^{(l)}\}_{i,j=1}^{m_{l}} be matrix units of MnlM_{n_{l}} and MmlM_{m_{l}}, respectively. Then UiU_{i} is written by

for 1≤i≤N1\leq i\leq N, 1≤l≤k1\leq l\leq k and 1≤s,t≤nl1\leq s,t\leq n_{l} is unitary. Indeed, WW can be considered as the matrix which takes the orthonormal basis {1mlest(l)}\{{1\over\sqrt{m_{l}}}e_{st}^{(l)}\} of A{\cal A} into the orthonormal basis {Ui}\{U_{i}\}. Hence we have

Let TT be a partial isometry with T∗T=es1s1(l1)⊗ft1t1(l1)T^{*}T=e_{s_{1}s_{1}}^{(l_{1})}\otimes f_{t_{1}t_{1}}^{(l_{1})} and TT∗=es2s2(l2)⊗ft2t2(l2)TT^{*}=e_{s_{2}s_{2}}^{(l_{2})}\otimes f_{t_{2}t_{2}}^{(l_{2})}. Then we obtain

by (5) so that FF maps the off-diagonal part to , that is, if l1≠l2l_{1}\neq l_{2} then F(T)=0F(T)=0.

Now let T=es2s1(l)⊗ft2t1(l)T=e_{s_{2}s_{1}}^{(l)}\otimes f_{t_{2}t_{1}}^{(l)}. Then we obtain

which shows the first assertion. □\square

The commutant of M-subalgebras and F-subalgebras are again M-subalgebras and F-subalgebras, and in both types it is possible to choose an orthogonal basis consisting of unitaries, only. Thus by an application of the previous proposition, for such a subalgebra A{\cal A}, the trace-preserving conditional expectation is the convex combination of automorphisms:

where {Ui′}\{U_{i}^{\prime}\} is an orthogonal basis of A′{\cal A}^{\prime} consisting of unitaries. Bases consisting of unitaries are important also in quantum state teleportation .

Let {Ui:1≤i≤m}\{U_{i}:1\leq i\leq m\} be an orthonormal system in MnM_{n}. Then the linear mapping

is completely positive if and only if μi≥0\mu_{i}\geq 0 for every 1≤i≤m1\leq i\leq m.

Proof: If μi≥0\mu_{i}\geq 0 for every 1≤i≤m1\leq i\leq m, it is clear that α\alpha is completely positive. To prove the converse, we first show that

is a projection for every 1≤k≤m1\leq k\leq m. To show that they are pairwise orthogonal, we compute the trace of PkPlP_{k}P_{l}:

is positive, therefore μk≥0\mu_{k}\geq 0. □\square

Proof: Since both sides are bilinear in the variables XX and YY, it is enough to check the case X=EabX=E_{ab} and Y=EcdY=E_{cd}. Simple computation gives that left-hand-side is δabδcd\delta_{ab}\delta_{cd}. A physicist might make a different proof of the lemma:

We also need the next lemma; the proof can be found in .

Let V1,V2,…,Vn2V_{1},V_{2},\dots,V_{n^{2}} be matrices in MnM_{n}. Then the following conditions are equivalent:

The next result includes important particular cases which are formulated afterwards.

Let A1,…,Ar{\cal A}_{1},\dots,{\cal A}_{r} be pairwise complementary subalgebras of MnM_{n} such that their commutants A1′,…,Ar′{\cal A}_{1}^{\prime},\dots,{\cal A}_{r}^{\prime} are pairwise complementary as well. Then the trace-preserving conditional expectations Ej:Mn→AjE_{j}:M_{n}\to{\cal A}_{j} can be expressed by the orthonormal bases Uj1,Uj2,…,Ujn(j)∈Aj′U_{j1},U_{j2},\dots,U_{jn(j)}\in{\cal A}_{j}^{\prime}, where Uj1=1nIU_{j1}={1\over\sqrt{n}}I, via the formula

and the generalized Pauli channel (1) is completely positive if and only if

To prove this theorem we prepare the following proposition.

Let A1{\cal A}_{1} and A2{\cal A}_{2} be complementary subalgebras of MnM_{n}. Then A1′{\cal A}_{1}^{\prime} and A2′A_{2}^{\prime} are complementary if and only if A1A2{\cal A}_{1}{\cal A}_{2} linearly spans MnM_{n}. Moreover, in this case the trace-preserving conditional expectation E1:Mn→A1′E_{1}:M_{n}\to{\cal A}_{1}^{\prime} can be expressed as

where {Ui}\{U_{i}\} is an orthonormal basis of A1{\cal A}_{1}.

Proof: Assume A1′{\cal A}_{1}^{\prime} and A2′{\cal A}_{2}^{\prime} are complementary. Let {Ui′}\{U_{i}^{\prime}\} and {Vj′}\{V_{j}^{\prime}\} be orthonormal bases of A1′{\cal A}_{1}^{\prime} and A2′{\cal A}_{2}^{\prime}, respectively, which consist of scalar multiple of their matrix units. Then the trace-preserving conditional expectations onto A1{\cal A}_{1} and A2{\cal A}_{2} are given by the linear combinations of Ui′∗( ⋅ )Ui′U_{i}^{\prime*}(\,\cdot\,)U_{i}^{\prime} and Vj′∗( ⋅ )Vj′V_{j}^{\prime*}(\,\cdot\,)V_{j}^{\prime}, respectively, thanks to Proposition 1.

Conversely assume A1A2{\cal A}_{1}{\cal A}_{2} linearly spans the whole space MnM_{n}. Since A1{\cal A}_{1} is a subalgebra of MnM_{n}, A1{\cal A}_{1} can be written as

Let QQ be a minimal central projection in A2{\cal A}_{2} and let {Ui(s)}\{U_{i}^{(s)}\} and {Vj}\{V_{j}\} are orthonormal bases of A1{\cal A}_{1} and A2{\cal A}_{2}, respectively, with the assumption Ui(s)∈Mns⊗ImsU_{i}^{(s)}\in M_{n_{s}}\otimes I_{m_{s}}. Since A1{\cal A}_{1} and A2{\cal A}_{2} are complementary and span{A1A2}=Mn{\rm span}\{{\cal A}_{1}{\cal A}_{2}\}=M_{n}, {nUi(s)Vj}\{\sqrt{n}U_{i}^{(s)}V_{j}\} is an orthonormal basis of MnM_{n}. Therefore by Lemma 2 and Proposition 1, we have

for some c>0c>0. These equations imply, for 1≤s≤k1\leq s\leq k,

where PsP_{s} is a central projection Ins⊗ImsI_{n_{s}}\otimes I_{m_{s}}. Now we take the trace to the above equation. Then we have

Hence ns/ms{n_{s}/m_{s}} is equal to 1/c=dimA1n{1/c}={{\rm dim}{\cal A}_{1}\over n} for all 1≤s≤k1\leq s\leq k and so

is the trace-preserving conditional expectation onto A1′{\cal A}_{1}^{\prime} by Proposition 1. Similarly,

is the trace-preserving conditional expectation onto A2′{\cal A}_{2}^{\prime}. Since

Proof of Theorem 2. The first assertion is already proven in the above proposition. Due to the Lemma 2, we have

and the coefficient of An=(1nI)A(1nI){A\over n}=\left({1\over\sqrt{n}}I\right)A\left({1\over\sqrt{n}}I\right) is

Theorem 1 tells us that completely positivity holds if and only if both are positive. □\square

Assume that MnM_{n} contains pairwise complementary M-subalgebras A1,…,Ar{\cal A}_{1},\dots,{\cal A}_{r}. Then the generalized Pauli channel is completely positive if and only if

Bipartite channels

In this section we consider subalgebras of Mn⊗MnM_{n}\otimes M_{n}. A subalgebra isomorphic to MnM_{n} will be called F-subalgebra. An M-subalgebra is a maximal Abelian subalgebra. Both kinds of subalgebras are subspaces of dimension n2n^{2}.

Assume that A1{\cal A}_{1} and A2{\cal A}_{2} are F- or M-subalgebras of Mn⊗MnM_{n}\otimes M_{n}. If they are complementary, then the commutants A1′{\cal A}_{1}^{\prime} and A2′{\cal A}_{2}^{\prime} are complementary as well.

Proof: Since both kinds of subalgebras are subspaces of dimension n2n^{2}, the dimension of A1A2{\cal A}_{1}{\cal A}_{2} is n4n^{4} so that A1A2=Mn⊗Mn{\cal A}_{1}{\cal A}_{2}=M_{n}\otimes M_{n}. Therefore the commutants A1′{\cal A}_{1}^{\prime} and A2′{\cal A}_{2}^{\prime} are complementary by Proposition 2. □\square

Assume that Mn⊗MnM_{n}\otimes M_{n} is decomposed to pairwise complementary F- and M-subalgebras Ai{\cal A}_{i} (1≤i≤n2+11\leq i\leq n^{2}+1). The trace-preserving conditional expectation of Mn⊗MnM_{n}\otimes M_{n} onto Ai{\cal A}_{i} is denoted by EiE_{i}. The linear trace-preserving mapping acting as

Proof: Theorem 3 allows to use Theorem 2 and the result follows. □\square

The theorem can be applied in Example 2. Note that decompositions of M2⊗M2M_{2}\otimes M_{2} into F- and M-subalgebras are discussed in , while decomposition of Mn⊗MnM_{n}\otimes M_{n} into F-subalgebras is constructed in if n=pkn=p^{k} with a prime number p>2p>2.

Acknowledgement. The authors thank to Professor Tsuyoshi Ando for communication and to the project of JSPS and Hungarian Academy of Sciences for support.

References