and extending this definition to all vectors in H AB by the complex bilinearity of the inner product.

Transcription

1 Multiple systems, the tensor-product space, and the partial trace Carlton M Caves 2005 September 15 Consider two quantum systems, A and B System A is described by a d A -dimensional Hilbert space H A, and system B is described by a d B -dimensional Hilbert space H B What we want to do is to construct the Hilbert space that describes the composite system consisting of both A and B The composite space certainly must have a way of describing the situation where A has the pure state ψ and B has the pure state φ We denote the corresponding composite state by ψ φ, and we call such a state a product state For the present, the product symbol, read as o-times, is simply a way of separating the state of A from the state of B, with the state of A on the left and the state of B on the right More generally, of course, we can apply this product to unnormalized vectors from A and B, and we do so freely in what follows without paying much attention to whether we are taling about normalized or unnormalized vectors The set of all product states is not a vector space; it is the Cartesian product of H A and H B, ie, the set of all ordered pairs consisting of a vector from H A and a vector from H B Now suppose we write ψ as a superposition of two other states of A, ie, ψ = a χ + b ξ It is certainly reasonable to write ψ φ = a χ + b ξ φ = a χ φ + b ξ φ, 1 since all this is saying is that superposing two states of A and then saying B has state φ is the same as saying B has state φ and then superposing the corresponding two composite states This innocuous assumption, along with the same considerations for B, already says, however, that the o-times product is bilinear in both its inputs We can see that the Cartesian product is not a vector space by noting that a linear combination of two product vectors, a ψ φ + b χ ξ, 2 is not generally a product vector It being a general principle of quantum mechanics that systems are described by complex vector spaces, we now assume, in accordance with this principle, that the appropriate state space for the composite system is not just the Cartesian product, but rather the entire vector space spanned by the product states This is a fateful assumption, because it leads to the phenomenon of quantum entanglement, entangled pure states being precisely the composite states that are not product states The vector space spanned by the product states, denoted by H A H B = H AB, is called the tensor product of H A and H B The o-times symbol is now read as tensor product The inner product on H AB is defined by defining the inner product of two product vectors as ψ φ χ ξ = ψ χ φ ξ 3 and extending this definition to all vectors in H AB by the complex bilinearity of the inner product 1

2 We often use capital letters to denote vectors in the tensor-product space, as in Ψ, and when confusion threatens, we use subscripts or superscripts to indicate to which system a vector belongs, as in ψ A for a vector in H A or Ψ AB for a vector in H AB Any vector Ψ in H AB can be written as a linear combination of product vectors, all the vectors in the products can be expanded in orthonormal bases, e j, j = 1,, d A, for A and f, = 1,, d B, for B, and the bilinearity of the tensor product can then be used to write Ψ as a linear combination of the orthonormal product vectors e j f, showing that these product vectors are a basis for H AB The number of these vectors is d A d B, which means that the dimension of the tensor-product space is d A d B We often leave out the inner-product symbol in these basis vectors, writing, in increasing order of omitting redundancies, such things as e j f = e j f = e j, f = j, = j 4 The expansion of an arbitrary vector in H AB loos lie Ψ = j, e j, f e j, f Ψ = j, c j e j, f = j, c j e j f 5 The expansion coefficients c j = e j, f Ψ can be written as a matrix, or in accord with our usual notation for representations of vectors, they can be written as a column vector: Ψ c 11 c 1dB c 21 c 2dB c da 1 c da d B = c 1 c da 6 In the second form, we introduce d B -dimensional column vectors c j1 c j = c jdb, 7 thus showing how we can thin of a vector in the tensor-product space as a d A -dimensional vector whose components are themselves d B -dimensional vectors One should recognize that Eqs 6 and 7 are simply the column-vector version of the following way of writing Ψ : Ψ = e j c j f 8 j 2

3 The expansion coefficients of a product vector ψ φ are the outer product of the expansion coefficients for ψ = j a e and φ = b f, ie, c j = e j, f ψ φ = e j ψ f φ = a j b, 9 which means that the column vectors in Eq 7 are A-dependent multiples of the column vector for φ : b 1 c j = a j 10 Once we have the tensor product under our belt, we realize that we can mae sense of what might be called the partial inner product, φ B Ψ AB This is defined to be the et in H A whose inner product with any vector ψ A in H A is the same as the complete inner product of ψ A φ B with Ψ AB, ie, b db ψ A φ B Ψ AB = ψ A φ B Ψ AB 11 Woring this out explicitly in the product basis e j, f, we get φ B Ψ AB = c j e j φ B f = e j b c j 12 j, j The final form on the right maes clear that the partial inner product is the inner product of b with the second system B index of c j, with the first system A index left over to form a vector for system A The partial inner product with a product state is obviously ξ B ψ A φ B = ψ A ξ B φ B We can, of course, define in the same way a partial inner product ψ A Ψ AB It is worth noting that the partial inner product e j Ψ AB = c j f is the vector in H B that is represented by the column vector c j Thus another way of thining of the partial inner product is that it is a way of generating the column vectors c j We re now ready to go on to operators acting on the tensor-product space The basic operators are outer products ψ φ χ ξ = ψ χ φ ξ 13 The first form is in standard outer-product notation in the tensor-product space; we now how to handle this form because we have defined the inner product on the tensor-product space The second form re-arranges the outer product as a tensor product of outer-product operators for A and B This innocuous re-arrangement defines the tensor product of operators Equation 13 defines the definition of a tensor product of outer products; the definition is extended to the outer product of any two operators by assuming that the operator tensor product is bilinear, ie, A B = A jl e j e l B m f f m j,l = j,,l,m = j,,l,m,m A jl B m e j e l f f m A jl B m e j, f e l, f m 14 3

4 Once we re woring in the tensor-product space, we really should write an operator A acting on H A as A I B, ie, A I B = j,l, A jl e j e l f f = j,l, A jl e j, f e l, f 15 Liewise, an operator B acting on H B should be written as I A B We usually ignore this nicety unless doing so causes confusion It is easy to verify from Eq 14 that A 1 B 1 A 2 B 2 = A 1 A 2 B 1 B 2 An arbitrary operator O acting on the tensor-product space can be expanded as O = O j,lm e j, f e l, f m = O j,lm e j e l f f m 16 j,l,,m j,,l,m Using our usual notation for representing an operator as a matrix, we write O 11,11 O 11,1dB O 11,dA 1 O 11,dA d B O 1dB,11 O 1dB,1d B O 1dB,d A 1 O 1dB,d A d B O O da 1,11 O da 1,1d B O da 1,d A 1 O da 1,d A d B O da d B,11 O da d B,1d B O da d B,d A 1 O da d B,d A d B 17 = O 11 O 1dA O da 1 O da d A In the second form, we introduce d B d B -dimensional matrices O j1,l1 O j1,ldb O jl =, 18 O jdb,l1 O jdb,ld B thus showing how we can thin of a matrix in the tensor-product space as a d A d A - dimensional matrix whose components are themselves d B d B -dimensional matrices One should recognize that Eqs 17 and 18 are simply the matrix version of the following way of writing O: O = e j e l O j,lm f f m 19 j,l,m This realization should cure one of wanting to write the explicit matrix forms ever again, but it is useful to able to loo at Eq 19 and to be able to reconstruct the matrices 17 and 18 in your head 4

5 Using the partial inner product, we can go on to define a partial matrix element of a composite operator O, φ B O ξ B = e j e l O j,lm φ B f f m ξ B, 20 j,l,m which is an operator acting on system A It should be obvious that the partial matrix element of a tensor product is φ B A B ξ B = A φ B B ξ B Nearly as obvious is that φ B A I B O ξ B = A φ B O ξ B and φ B OA I B ξ B = φ B O ξ B A In the same way, we can, of course, define a partial matrix element ψ A O χ A It is worth noting that e j O e l =,m O j,lm f f m 21 is the operator that has the matrix representation 18 We have all the ingredients now to define the partial trace of a composite operator O The partial trace of O with respect to system B is an operator on system A: tr B O = f O f = j,l O j,l e j e l 22 The partial trace on B is a linear map from joint operators to operators on A The partial trace of O with respect to A is similarly defined as tr A O = j e j O e j =,m O j,jm f f m 23 Notice that the complete trace can be obtained by doing two partial traces: tro = j, O j,j = tr A trb O = tr B tra O 24 In the same way as the complete trace, the partial trace is linear and independent of the orthonormal basis used to calculate it It should be obvious that tr B A B = AtrB Nearly as obvious is that tr B A IB O = Atr B O and tr B OA IB = tr B OA It is not generally true that tr B NO = tr B ON, as one discovers by trying to duplicate the proof for the complete trace: tr B NO = f NO f =,m f N f m f m O f 25 When dealing with the complete trace, one can switch the order of the matrix elements, since they are complex numbers, thus changing the order of the product In contrast, the partial matrix elements in Eq 25 are operators on system A, and these operators generally don t commute It is true, however, that tr B IA BO = tr B OIA B, 26 5

6 because in this case the partial matrix element f I A B f m = I A f B f m is a multiple of the unit operator, so it does commute with f m O f In the same way, we can also show that tr B OIA φ B ξ B =,m f O f m f m φ B ξ B f =,m ξ B f f O f m f m φ B 27 = ξ B O φ B, so that the partial trace with respect to B does turn an outer product on B into a partial matrix element with respect to B The main thing to be alert to in performing these manipulations is which objects are operators and which are complex numbers in Eq 27, f O f m is an operator on system A, whereas f m φ B and ξ B f are complex numbers since we use the same notation that, in the case of a single system, made everything a complex number We use the partial trace for a lot of things we really couldn t do without it but it receives its main justification from the notion of a marginal density operator for a subsystem of a composite system If the composite system has the density operator ρ AB, a measurement in the basis e j on system A yields result j with probability p j = p j = e j, f ρ AB e j, f, 28 In writing this expression, we imagine that in addition to the measurement in the basis e j on system A, a measurement in an arbitrary basis f is made on system B To find the probability for j, we sum the joint probabilities over the results of the measurement on B, in accordance with the rules for classical probabilities The probability for result j can be put in the form p j = e j f ρ AB f }{{} = tr B ρ AB = ρ A e j = e j ρ A e j 29 where ρ A = tr B ρ AB, obtained by taing the partial trace of ρ AB with respect to B, is called the marginal density operator of system A We use this terminology because as far as measurements on A are concerned, all probabilities are calculated as though ρ A were the density operator of system A 6

7 We can put p j in other useful forms through the following manipulations: p j = tr ρ AB e j, f e j, f = tr ρ AB e j e j f f = tr ρ AB e j e j f f = tr ρ AB P j I B = tr A tr B ρab P j I B = tr A trb ρ AB P j = tr A ρ A P j In the last form, we get bac to Eq 29 In the middle, we find that when we are woring with the composite state ρ AB, the projection operator that goes with a particular result on A is P j I B = e j e j I B, which is a multi-dimensional projector with ran d B, corresponding to the fact that result j is degenerate, with d B different possibilities for system B Let s put this machinery into action in a particular case Suppose we have two qubits in the pure state Ψ AB = cos θ 00 + sin θ The corresponding composite density operator is ρ AB = Ψ AB Ψ AB = cos 2 θ sin 2 θ cos θ sin θ A measurement of Z A = on A yields +1 with probability p +1 = tr ρ AB P ez I B = Ψ AB P ez I B Ψ AB = Ψ AB Ψ AB = Ψ AB Ψ AB = cos 2 θ We can get the same result by calculating the marginal density operator ρ A = tr B ρ AB = cos 2 θ tr B sin 2 θ tr B cos θ sin θ tr B trb = cos 2 θ sin 2 θ cos θ sin θ = cos 2 θ sin 2 θ 1 1, from which we calculate p +1 = tr A ρa P ez = tra ρa 0 0 = 0 cos 2 θ sin 2 θ = cos 2 θ