See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/2198359

State tomography for two qubits using reduced densities

Article  in  Journal of Physics A General Physics · May 2006

DOI: 10.1088/0305-4470/39/34/020 · Source: arXiv

CITATIONS READS
11 16

4 authors, including:

Katalin M Hangos
Hungarian Academy of Sciences
335 PUBLICATIONS   2,105 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Katalin M Hangos on 26 April 2013.

The user has requested enhancement of the downloaded file.

 State tomography for two qubits
using reduced densities

D. Petz1,3 , K.M. Hangos2,3 , A. Szántó1,3 and F. Szöllősi1,3

arXiv:quant-ph/0605049v1 4 May 2006

1
Budapest University of Technology and Economics
2
Computer and Automation Research Institute

Abstract: The optimal state determination (or tomography) is studied for

a composite system of two qubits when measurements can be performed on
one of the qubits and interactions of the two qubits can be implemented.
The goal is to minimize the number of interactions to be implemented. The
algebraic method used in the paper leads to an extension of the concept of
mutually unbiased measurements.
PACS numbers: 03.67.-a, 03.65.Wj, 03.65.Fd
Key words: State determination, reduced density, unbiased measurement,
minimal realization, unbiased subalgebra, Pauli matrices.

1 Introduction
An n-level quantum system is described by an n-dimensional Hilbert space H, or equiva-
lently by the algebra Mn (C) of the n × n complex matrices. When an orthogonal basis of
H is chosen, operators acting on H correspond to n×n matrices. A positive operator ρ of
trace 1 is called state. If we choose and fix an orthonormal basis {e1 , e2 , . . . , en }, then a
state ρ is determined by the matrix elements ρij = hei |ρ|ej i. Determination of ρ involves
n2 − 1 real parameters, namely, ρii (1 ≤ i ≤ n − 1), Re ρij and Im ρij (1 ≤ i < j ≤ n).
A von Neumann measurement on the system P is a family M = {P1 , P2 , . . . , Pd } of
pairwise orthogonal projections such that i Pi = I. When the measurement M is
performed in the state ρ, the outcome 1 ≤ j ≤ d appear with probability pj = Tr ρPi for
each j [5, 6]. Independent measurements on several copies of our quantum system give
the relative frequencies fj for each outcome j and fj is an estimate of the probability pj .
3
Supported by the Hungarian Research Grants OTKA T042710, T063066 and T032662.

1
The repeated
P measurement provides d − 1 degree of freedom concerning the the density
ρ, since j pj = 1. The information we obtained is maximal if d = n which means that
all the projections Pj are of rank one. ρ is determined by n2 − 1 parameters, hence at
least n + 1 different measurements are to be performed to cover all degrees of freedom.
Of course, the n + 1 different measurements are sufficient in the case when they provide
“non-overlapping” information.
The literature of state tomography is very rich, there are several protocols, and the
efficiency of state reconstruction can be increased if the later measurements depend on
the outcomes of the former ones [1, 7, 9].
The composite system of two qubits is a 4-level quantum system which is described
on the space C4 ≡ C2 ⊗ C2 . A state is described by 15 real parameters. Therefore, at
least 5 kinds of elementary measurements should be made to determine the state of the
system.
Denote by A and B the two qubits. Then M4 (C) = B(HA ) ⊗ B(HB ), where B(HA )
and B(HB ) are isomorphic to M2 (C). Assume that we can perform measurements only on
the qubit A. If the total system has the statistical operator ρAB , then we can reconstruct
the reduced density
(1)
ρA := Tr B ρAB
after some measurements. In order to get more information, we switch on an interaction
between the two qubits. If H is the Hamiltonian, then the new state is
eiH ρAB e−iH = W1 ρAB W1∗ (1)
after the interaction. (For the sake of simplicity, the interaction is kept for a time unit.)
The new reduced density is
(2)
ρA := Tr B W1 ρAB W1∗ .
This procedure may be continued by using other interactions and ends with a sequence
(1) (2) (k)
of reduced states ρA , ρA , . . . , ρA . We want to determine the minimal k such that this
sequence of reduced densities determines ρAB . In other words, we want to minimize the
number of interactions between the two qubits. It turns out that the minimum number
is 5.
Minimal realizations play an important role in systems theory, too [3], because they
represent the state of the system with the minimum possible number of parameters.
Minimal realizations are known to be jointly controllable and observable for most of the
known system classes. The above problem of finding the minimum number of reduced
states can be regarded as a minimal representation problem for a system that consists
of a pair of coupled qubits.

2 Algebraic formulation
Instead of the transformation (1) of the density matrix ρAB , we can change the subalgebra
and we have an equivalent algebraic formulation. The total system is described by the

2
algebra M4 (C). We look for subalgebra A1 , A2 , . . . , Ak such that
1. Each Aj is algebraically isomorphic to M2 (C), 1 ≤ j ≤ k.
2. The linear span of the subspaces A1 , A2 , . . . , Ak is M4 (C).
Given a subalgebra Aj , there is a unitary Wj such that Wj∗ Aj Wj = B(HA ) ⊗ CIB . The
reduced density ρj ∈ Aj is the same as the reduction of Wj ρAB Wj∗ to the first spin.
Therefore, instead of the reduction of density after the interaction, we can work with
the reduced density of ρAB in Aj . The second condition makes sure that the reduced
densities in A1 , A2, . . . , Ak determine ρAB completely.
The traceless subspace of M4 (C) has dimension 15, while the traceless subspace of
Aj has dimension 3, therefore we need k ≥ 5 to fulfill the requirements. It will turn out
that k = 5 is possible.
The algebra M2 (C) is linearly spanned by the Pauli matrices:
       
1 0 0 1 0 −i 1 0
σ0 := , σ1 := , σ2 := , σ3 := .
0 1 1 0 i 0 0 −1
Recall that they satisfy the multiplication rules
3
X
σi σj = δij I + i ǫijk σk (1 ≤ i, j ≤ 3), (2)
k=1

where ǫijk is the Levi-Civita tensor:


 0 if ∃j, k such that ij = ik ,
ǫi1 i2 ···in = 1 if the permutation (i1 i2 · · · in ) is even,
−1 if the permutation (i1 i2 · · · in ) is odd.


The above rules can be essentially simplified by posing the following two requirements:
1. σj is a self-adjoint unitary (1 ≤ j ≤ k) and σ3 = −iσ1 σ2 .
2. σ1 σ2 + σ2 σ1 = 0.
When a triplet (S1 , S2 , S3 ) satisfies these condition, it will be called a Pauli triplet. For
such a triplet Tr Si = 0 and Tr Si Sj = 0 for i 6= j. The latter relation is interpreted
as the orthogonality of Si and Sj with respect to the Hilbert-Schmidt inner product
hA, Bi := Tr A∗ B. Furthermore, it can be seen that the two relations above imply (2).
Given a Pauli triplet (S1 , S2 , S3 ), the linear mapping defined as
σ0 7→ I, σ1 7→ S1 , σ2 7→ S2 , σ3 7→ −iS1 S2
is an algebraic isomorphism between M2 (C) and the linear span of the operators I, S1 , S2
and S3 .
In the algebra M4 (C), the elementary tensors σi ⊗ σj form an orthogonal basis (0 ≤
i, j ≤ 3). All these operators are self-adjoint unitaries and can be chosen to be Si ’s.
The next proposition is the main result of the paper.

3
Proposition 1 There are 5 sublalgebras of B(HA ) ⊗ B(HB ) such that each of them is
isomorphic to M2 (C) and the reduced states determine an arbitrary state ρAB of the two
qubits A and B.

Proof. First we take the following Pauli triplets consisting of elementary tensors:

{σ0 ⊗ σ1 , −σ1 ⊗ σ3 , σ1 ⊗ σ2 }
     

 0 1 0 0 0 0 −1 0 0 0 0 −i 

1 0 0 0   0 0 0 1   0 0 i 0 

=    ,   ,   ,

 0 0 0 1   −1 0 0 0   0 −i 0 0  
0 0 1 0 0 1 0 0 i 0 0 0
 

{σ3 ⊗ σ1 , σ1 ⊗ σ1 , σ2 ⊗ σ0 }
     

 0 1 0 0 0 0 0 1 0 0 −i 0 
1 0 0 0   0 0 1 0   0 0 0 −i 

=  , ,  ,


 0 0 0 −1   0 1 0 0   i 0 0 0 
0 0 −1 0 1 0 0 0 0 i 0 0
 

{σ1 ⊗ σ0 , σ2 ⊗ σ2 , σ3 ⊗ σ2 }
     

 0 0 1 0 0 0 0 −1 0 −i 0 0 
0 0 0 1   0 0 1 0   i 0 0 0 
  
=   1 , ,  ,

 0 0 0   0 1 0 0   0 0 0 i 
0 1 0 0 −1 0 0 0 0 0 −i 0
 

{σ0 ⊗ σ2 , σ2 ⊗ σ3 , σ2 ⊗ σ1 }
     

 0 −i 0 0 0 0 −i 0 0 0 0 −i 

i 0 0 0   0 0 0 i   0 0 −i 0 

=    i 0 0 0  ,  0 i 0 0  .
,   


 0 0 0 −i 
0 0 i 0 0 −i 0 0 i 0 0 0
 

Together with the identity, each triplet linearly spans a subalgebra Aj (1 ≤ j ≤ 4).
It is important to observe that all the matrices have vanishing diagonal, moreover the
matrices are pairwise orthogonal, therefore they are linearly independent.
If we find another Pauli triplet (S1 , S2 , S3 ) such that the diagonals are linearly inde-
pendent, then we have a fifth algebra A5 such that {Ak : 1 ≤ k ≤ 5} spans linearly
M4 (C). Indeed, if A is any matrix, then we can find T ∈ A5 such that A − T has 0
diagonal and this is in the linear hull of {Aj : 1 ≤ j ≤ 4}. It follows that the reduced
densities in {Aj : 1 ≤ j ≤ 5} determines ρAB uniquely.

4
 Here is an example of the above described triplet:
     
1 1 1 1 1 i i −1 −1 i 1 i
1 1
 1 −1 −1   1  −i −1 −1
 i  1 −i 1 i 1 
, , .
2 1 −1 −1
 1  2  −i −1 1 −i  2  1 −i 1 i 
1 −1 1 −1 −1 −i i −1 −i 1 −i −1

These matrices are not elementary tensors (but they are Hadamard matrices [11] up to
a constant multiple and were found by means of an exhaustive search algorithm on a
computer). 

Proposition 2 If all the matrices of the Pauli triplet generating the subalgebras Aj are
of the form ±σk ⊗ σl (0 ≤ k, l ≤ 3), then we need at least 6 triplets to span M4 (C).

Proof. Assume that a Pauli triplet (T1 , T2 , T3 ) in M4 (C) is such that every element is
of the form ±σi ⊗ σj (0 ≤ i, j ≤ 3).
If T1 = ±σi ⊗ σj and T2 = ±σk ⊗ σl , then
X
±i T3 = − (ǫikm ǫjln σm ⊗ σn )
m,n
 X X 
+i δik (ǫjln σ0 ⊗ σn ) + δjl (ǫikm σm ⊗ σ0 ) + δik δjl σ0 ⊗ σ0 .
n m

Since T3 is self-adjoint but iσi ⊗ σj is not, it follows that exactly one of the relations
i = k and j = l must hold. At least one of the operators Ti should be of the form σ0 ⊗ σj
or σj ⊗ σ0 .
We have three operators in the form σ0 ⊗σj and three in the form σj ⊗σ0 (1 ≤ j ≤ 3). If
we have 5 Pauli triplets, then at least one should contain two of the above tensor products
(up to a sign). One can see that σ0 ⊗ σj and σi ⊗ σ0 cannot be in a triplet, therefore
a triplet contains two operators in the form σ0 ⊗ σj or two operators like σj ⊗ σ0 . In
both cases, the third operator has similar form. Hence one of the operators σ0 ⊗ σj and
σj ⊗ σ0 appears in two triplets and in this case 5 triplet cannot span the whole space.
Six subalgebras described in the proposition can be given by the following Pauli
triplets:
{σ1 ⊗ σ1 , σ1 ⊗ σ2 , σ0 ⊗ σ3 },
{σ2 ⊗ σ2 , σ2 ⊗ σ3 , σ0 ⊗ σ1 },
{σ3 ⊗ σ3 , σ3 ⊗ σ1 , σ0 ⊗ σ2 },
(3)
{σ2 ⊗ σ2 , σ3 ⊗ σ2 , σ1 ⊗ σ0 },
{σ3 ⊗ σ3 , σ1 ⊗ σ3 , σ2 ⊗ σ0 },
{σ1 ⊗ σ1 , σ2 ⊗ σ1 , σ3 ⊗ σ0 }.
Together with I each triplet linearly spans a subalgebra Aj (1 ≤ j ≤ 6) and the 6
subalgebras linearly span the whole M2 (C) ⊗ M2 (C). 

5
3 Generalizations
Mutually unbiased bases (or measurements) are interesting from many point of view
[4, 2, 13] and the maximal number of such bases is not completely known [12]. The above
discussed setting of state determination is somewhat similar. In this setting we may
look for essentially orthogonal non-commutativ subalgebras while unbiased elementary
measurement are given essentially orthogonal maximal Abelian subalgebras, see Prop.
2.2 of [8]. The next statement is an analogue of Parthasarathy’s proposition.

Proposition 3 Let A1 and A2 be subalgebras of Mn (C) and assume that they are iso-
morphic to Mk (C). Then the following conditions are equivalent:

(i) If P ∈ A1 and Q ∈ A2 are minimal projections, then Tr P Q = n/k 2 .

(ii) The subspaces A1 ⊖ CI and A2 ⊖ CI are orthogonal in Mn (C)

Proof. It follows from the conditions that n = mk and Tr P = Tr Q = m for the minimal
projections. Therefore, condition (i) is equivalent to Tr (I − kP )(I − kQ) = 0 which
means that (I − kP ) ⊥ (I − kQ). Since the subspaces in (ii) are linearly spanned by
these operators, the proposition follows. 
Now we generalize Prop. 2 for n qubits.

Proposition 4 If all the matrices of the Pauli triplet generating the subalgebras Aj of
M2n (C) are of the form ±σk(1) ⊗ . . . ⊗ σk(n) (0 ≤ k(i) ≤ 3, 1 ≤ i ≤ n), then we need more
than (22n − 1)/3 triplets to span M2n (C).

Proof. First note that a matrix ±σk(1) ⊗ . . . ⊗ σk(n) has only real elements or only
imaginary elements. Among the three matrices of a Pauli triplet (T1 , T2 , T3 ), there is one
imaginary or there are three imaginary matrices. Let N be the number of triplets with 1
imaginary matrix and M be the number of triplets with 3 imaginary ones. If the N + M
triplets with identity linearly span the self-adjoint subspace, then 3(N + M) + 1 ≥ 22n .
Assume that
22n − 1
N +M = . (4)
3
Since the dimension of the subspace of self-adjoint matrices with imaginary elements is
(22n − 2n )/2, we must have
22n − 2n
N + 3M = . (5)
2
One can see that equations (4) and (5) do not have integer solution. 
We call a family M1 , M2 , . . . , Md of subalgebras strongly mutually unbiased if the
conditions in the proposition hold for any pair. The maximal number of strongly mu-
tually unbiased subalgebras is not know to us even in the simplest case when the large
algebra is M4 (C) and the subalgebras are isomorphic to M2 (C).

6
 Following [8], we may call a family M1 , M2, . . . , Md of subalgebras weakly mutually
unbiased if the subspaces M1 ⊖ CI, M2 ⊖ CI, . . . , Md ⊖ CI are linearly independent.
We showed that when the large algebra is M4 (C) and the subalgebras are isomorphic to
M2 (C), then the maximum number of weakly unbiased subalgebras is 5.

4 Discussion and conclusions

The optimal state tomography has been studied for a composite system of two qubits
when measurements can be performed on one of the qubits and interactions of the two
qubits can be implemented. Equivalently, we found physically realizable minimal set of
reduced densities. The transformation described by (1) is realized by a properly designed
measurement apparatus (see the experimental devices in [9]). Therefore, an additional,
but still unsolved problem is to find the unitary to each of the reduced densities that
transforms the state to be measured to another one belonging to the reduced density
in question. (Some preliminary results on the effect of the real 4-dimensional rotation
matrices on the reduced densities is found in [10].)
The construction of 5 Pauli triplets of 4×4 matrices from the tensor products of Pauli
matrices contains heuristic steps combined with an exhaustive search for the missing
final triplet. This makes it practically impossible to generalize the method for higher
dimensions. Even if the heuristic steps were removed the resulted algorithm would fall
to the NP-hard category because of the exhaustive search.
The questions we posed for two qubits can be asked about three (or more qubits). The
dimension of M8 (C) ⊖ CI is 63, so we need at least 21 Pauli triplets to span the whole
space. In the moment we can construct 22 spanning Pauli triplets by ad hoc method
(this will be discussed in another publication).
The problem investigated here motivates the definition of strongly and weakly unbi-
ased subalgebras. In relation with them, several questions can be raised.

References
[1] J.A. Bergou, V. Buzek, E. Feldman, U. Herzog and M. Hillery, Programmable
quantum state discriminators with simple programs, arXiv:quant-ph/0602164, 2006.

[2] P.O Boykin, M. Sitharam, P.H. Tiep and P. Wocjan, Mutually unbiased bases and
orthogonal decompositions of Lie algebras, arXiv:quant-ph/0506089, 2005.

[3] A. Isidoni, Nonlinear Control Systems, Springer-Verlag, Berlin, 1995.

[4] K. Kraus, Complementarity and uncertainty relations, Phys. Rev. D. 35, 3070-3075,
1987.

7
 [5] M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information,
Cambridge University Press, 2000.

[6] M. Ohya and D. Petz, Quantum Entropy and Its Use, Springer-Verlag, Berlin, 1993.

[7] M.G.A. Paris and J. Rehácek (eds), Quantum State Estimation, Lecture Notes in
Physics, Springer-Verlag, 2004.

[8] K.R. Parthasarathy, On estimating the state of a finite level quantum system, In-
finite Dimensional Analysis, Quantum Probability and Related Topics, 7, 607-617.
2004.

[9] J. Rehácek, B. Englert and D. Kaszlikowski, Minimal qubit tomography, Physical

Review A 70, 052321, 2004.

[10] A. Szántó, The reconstruction of the state of two qubits (in Hungarian), Student
Research Report, BUTE, 2005.

[11] W. Tadej and K. Zyczkowski, A concise guide to complex Hadamard matrices,

arXiv:quant-ph/0512154, 2005.

[12] R. Werner, http://www.imaph.tu-bs.de/qi/problems/13.html.

[13] W.K. Wooters and B.D. Fields, Optimal state determination by mutually unbiased
measurements, Annals of Physics, 191, 363–381, 1989.

View publication stats