Материал: искусственный интеллект

The possibility to expand one basis with respect to another basis induces connection between the probabilities P(a = α) and P(b = β). Let us expand the vectors eαa with respect to the basis eb

where uαβ = eαa , eβb . Thus d1 = c1u11 + c2u21, d2 = c1u12 + c1u22. We obtain the quantum rule for transformation of probabilities:

On the other hand, by the deÞnition of quantum conditional probability, see (28), we obtain

By combining (33), (34) and (37), (38) we obtain the quantum formula of total probabilityÑthe formula of the interference of probabilities:

α

+2 cos θ P(a = α1)P(b = β|a = α1)P(a = α2)P(b = β|a = α2).

In general cos θ = 0. Thus the quantum FTP does not coincide with the classical FTP (10) which is based on the BayesÕ formula.

We presented the derivation of the quantum FTP only for observables given by Hermitian operators acting in the two dimensional Hilbert space and for pure states. In Sect. 9.1, we give (without proving) the formula for spaces of an arbitrary dimension and states represented by density operators (see [42] for quantum FTP for observables represented by POVMs).

9.1Växjö (Realist Ensemble Contextual) Interpretation of Quantum Mechanics

The VŠxjš interpretation [33] is the realist ensemble contextual interpretation of QM. Thus, in contrast to Copenhagenists or QBists, by the VŠxjš interpretation QM is not complete and it can be emergent from a subquantum model. This interpretation is the ensemble interpretation This interpretation is contextual, i.e., experimental contexts have to be taken into account really seriously.

By the VŠxjš interpretation the probabilistic part of QM is a special mathematical formalism to work with contextual probabilities for families of contexts, which are, in general, incompatible. Of course, the quantum probabilistic formalism is not the only possible formalism to operate with contextual probabilities.

The main distinguishing feature of the formalism of quantum probability is its complex Hilbert space representation and the BornÕs rule. All quantum contexts can be unified with the aid of a quantum state ψ (wave function, complex probability amplitude). A state represents only a part of context, the second part is given by an observable a. Thus the quantum probability model is not just a collection of Kolmogorov probability spaces. These spaces are coupled by quantum states.

Each theory of probability has two main purposes: descriptive and predictive. In classical probability theory its predictive machinery is based on Bayesian inference and, in particular, FTP (Sect. 2, formula (10)).

Can the probabilistic formalism of QM be treated as a generalization of Bayesian inference?

My viewpoint is that the quantum FTP with the interference term (Sect. 9, formula (39)) is, in fact, a modiÞed rule for the probability update. QM provides the following inference machinery. There are given a mixed state represented by density operator ρ and two quantum observables a and b represented mathematically by Hermitian operators a and b with purely discrete spectra. The Þrst measurement of a can be treated as collection of information about the state ρ. The result a = αi appears with the probability

This is generalization of the BornÕs rule to mixed states.

Postulate 6L (the projection postulate in the LŸdersÕ form) can be extended to mixed states. Initial state ρ is transferred to the state

Then, for each state ρai , we perform measurement of b and obtain probabilities

These are quantum conditional (transition) probabilities for the initial state given by a density operator (generalization of the formalism of Sect. 7).

We now recall the general form of the quantum FTP [42]:

k

+2 cos φj ;k,m p(b = β|a = αk )p(a = αk )p(b = β|αm)p(a = αm).

k<m

Thus we can predict the probability of the result βj for the b-observable on the basis of the probabilities for the results αi for the a-observable and conditional probabilities. Of course, the main nonclassical feature of this probability update rule is the presence of phase angles. In the case of dichotomous observables of the von NeumannÐLŸders type the phase angles φj can be expressed in terms of probabilities.

10 Quantum Logic

von Neumann and Birkhoff [11, 61] suggested to represent events (propositions) by orthogonal projectors in complex Hilbert space H.

For an orthogonal projector P , we set HP = P (H ), its image, and vice versa, for subspace L of H, the corresponding orthogonal projector is denoted by the symbol

PL.

The set of orthogonal projectors is a lattice with the order structure: P ≤ Q iff HP HQ or equivalently, for any ψ H, ψ|P ψ ≤ ψ|Qψ .

We recall that the lattice of projectors is endowed with operations ÒandÓ ( ) and ÒorÓ ( ). For two projectors P1, P2, the projector R = P1 P2 is deÞned as the projector onto the subspace HR and the projector S = P1 P2 is deÞned as the projector onto the subspace HR deÞned as the minimal linear subspace containing the set-theoretic union HP1 HP2 of subspaces HP1 , HP2 : this is the space of all linear combinations of vectors belonging to these subspaces. The operation of negation is deÞned as the orthogonal complement: P = {y H :y|x = 0 for all x HP }.

In the language of subspaces the operation ÒandÓ coincides with the usual settheoretic intersection, but the operations ÒorÓ and ÒnotÓ are nontrivial deformations of the corresponding set-theoretic operations. It is natural to expect that such deformations can induce deviations from classical Boolean logic.

and Pv Pe2 = 0, but Pv (Pe1 Pe2 ) = Pv . Hence, for quantum events, in general the distributivity law is violated:

The lattice of orthogonal projectors is called quantum logic. It is considered as a (very special) generalization of classical Boolean logic. Any sub-lattice consisting of commuting projectors can be treated as classical Boolean logic.

At the Þrst sight the representation of events by projectors/linear subspaces might look exotic. However, this is simply a prejudice which springs from too common usage of the set-theoretic representation of events (Boolean logic) in the modern classical probability theory. The tradition to represent events by subsets was Þrmly established by A. N. Kolmogorov in 1933. We remark that before him the basic

classical probabilistic models were not of the set-theoretic nature. For example, the main competitor of the Kolmogorov model, the von Mises frequency model, was based on the notion of a collective.

As we have seen, quantum logic relaxes some constraints posed on the operations of classical Boolean logic, in particular, the distributivity constraint. This provides novel possibilities for logically consistent reasoning.

Since human decision makers violate FTP [32, 38]Ñthe basic law of classical probability, it seems that they process information by using nonclassical logic. Quantum logic is one of the possible candidates for logic of human reasoning. However, one has to remember that in principle other types of nonclassical logic may be useful for mathematical modeling of human decision making.

11 Space of Square Integrable Functions as a State Space

Although we generally proceed with Þnite-dimensional Hilbert spaces, it is useful to mention the most important example of inÞnite-dimensional Hilbert space used in QM. Consider the space of complex valued functions, ψ : Rm → C, which are square integrable with respect to the Lebesgue measure on Rm :

Rm

It is denoted by the symbol L2(Rm). Here the scalar product is given by

A delicate point is that, for some measurable functions, ψ : Rm → C, which are not identically zero, the integral

Rm

We remark that the latter equality implies that ψ(x) = 0 a.e. (almost everywhere). Thus the quantity deÞned by (45) is, in fact, not norm: ψ = 0 does not imply that ψ = 0. To deÞne a proper Hilbert space, one has to consider as its elements not simply functions, but classes of equivalent functions, where the equivalence relation is deÞned as ψ φ if and only if ψ(x) = φ(x) a.e. In particular, all functions satisfying (46) are equivalent to the zero-function.

72 A. Khrennikov

12 Operation of Tensor Product

Let both state spaces be L2-spaces, the spaces of complex valued square integrable functions: H1 = L2(Rk ) and L2(Rm).

Take two functions: ψ ≡ ψ(x) belongs to H1 and φ ≡ φ(y) belongs to H2. By multiplying these functions we obtain the function of two variables(x, y) = ψ(x) × φ(y), where × denotes the usual point wise product.5 It is easy to check that this function belongs to the space H = L2(Rk+m). Take now n functions, ψ1(x), . . . , ψn(x), from H1 and n functions, φ1(y), . . . φn(y), from H2 and consider the sum of their pairwise products:

This function also belongs to H.

It is possible to show that any function belonging to H can be represented as (47), where the sum is in general inÞnite. Multiplication of functions is the basic example of the operation of the tensor product. The latter is denoted by the symbol . Thus in the example under consideration ψ φ(x, y) = ψ(x) × φ(y). The tensor product structure on H = L2(Rk+m) is symbolically denoted as H = H1 H2.

Consider now two arbitrary orthonormal bases in spaces Hk , (ej(k)), k = 1, 2.

Then functions (eij = ei(1) ej(2)) form an orthonormal basis in H : any H can be represented as

Consider now two arbitrary Þnite-dimensional Hilbert spaces, H1, H2. For each pair of vectors ψ H1, φ H2, we form a new formal entity denoted by ψ φ. Then we consider the sums = i ψi φi . On the set of such formal sums we can introduce the linear space structure. (To be mathematically rigorous, we have to constraint this set by some algebraic relations to make the operations of addition and multiplication by complex numbers well deÞned.) This construction gives us

the tensor product H = H1 H2. In particular, if we take orthonormal bases in Hk , (ej(k)), k = 1, 2, then (eij = ei(1) ej(2)) form an orthonormal basis in H, anyH can be represented as (48) with (49).

5Here it is convenient to use this symbol, not just write as (x, y) = ψ(x)φ(y).