Linear and integrable nonlinear evolution of the qutrit

The nonlinear generalization of the von Neumann equation preserving convexity of the state space is studied in the nontrivial case of the qutrit. This equation can be cast into the integrable classical Riccati system of nonlinear ordinary differential equations. The solutions of such system are investigated in both the linear case corresponding to the standard von Neumann equation and the nonlinear one referring to the generalization of this equation. The analyzed dynamics of the qutrit is rich and includes quasiperiodic motion, multiple equilibria and limit cycles.


Introduction
In recent paper [1] the evolution of the density matrix was studied of the form where {·, ·} designates the anticommutator. We recall [1] that the following nonlinear Schrödinger equation implied by (1.2) when ρ(t) = |φ(t) φ(t)| is the pure state was originally introduced by Gisin [2] as a nonlinear candidate for description of quantum evolution of dissipative systems. It is worthwhile to note that this equation is recognized as "the only sensible candidate for a dissipative Schrödinger equation" [3]. Another application of (1.3) written in the form i d dt |ψ(t) = H|ψ(t) + (1 − P ψ(t) )U |ψ(t) , (1.4) where P ψ(t) = |ψ(t) ψ(t)| is the projection operator and U is an arbitrary linear or nonlinear operator, introduced by Grigorenko [4] is the modelling the collapse of the wave function when a measurement is made on a quantum system. An important property of the complete positive map specified by (1.1) such that , (1.5) where A(t) = e t(G−iH) , is that it preserves the convex structure of the state space. Namely, we have [1,5,6] where ρ 1 and ρ 2 are density matrices, and satisfies λ ′ ∈ [0, 1]. Another relevant feature of the dynamics described by Eq.
(1.1) is that it maps pure states into pure ones. We remark that non-unitary evolution in general takes pure states to mixed states. Finally, a desired property of the investigated nonlinear generalization of the von Neumann equation is that it does not allow superluminal messages [7]. We recall that the superluminal signalling is one of the crucial arguments against nonlinear generalizations of quantum mechanics such as for example the Schrödinger-Newton equation [8].
The quantum dynamics described by the nonlinear equation (1.2) was illustrated in [1] by the example of the qubit. To be more specific, we set

9)
Linear and integrable nonlinear evolution of the qutrit 3 where σ i , i = 1, 2, 3 are the Pauli matrices, the dot designates the scalar product, the Bloch vector ξ 0 ∈ R 3 satisfies |ξ 0 | 1, and H = a · σ, G = b·σ. (1.10) On substituting (1.9) and (1.10) into (1.1) we arrive at the following form of the density matrix ρ(t) ρ(t) = 1 2 (1 + ξ(t) · σ), (1.11) where ξ(t) is an explicit function of t and ξ 0 (see [1]). On the other hand, inserting (1.11) into the nonlinear von Neumann equation (1.2) we find that ξ(t) is the solution of the following nonlinear system of ordinary differential equationsξ = 2b + 2a × ξ − 2(b · ξ)ξ, ξ(0) = ξ 0 . (1.12) Thus, it turns out that the nonlinear quantum evolution equation (1.2) can be reduced to the nonlinear classical system (1.12). (1.14) The solutions to (1.12) were analyzed in a great detail in [1]. In particular an interesting property of the nonlinear dynamics of the system (1.12) was found -the global asymptotic stability of stationary (equilibrium) solutions corresponding to evolution of the qubit from mixed states to pure ones. An example of a physical application of the discussed approach is the relativistic quantum spin one-half particle in electromagnetic field analyzed in Ref. 1. More precisely, the evolution is introduced therein such that the ρ(t) and A(t) in (1.5) are connected with the Bargmann-Michel-Telegdi equations [10] describing relativistic motion of a particle with a magnetic moment in the external electromagnetic field. The vectors a and b from (1.10) are identified with the external magnetic and electric field, respectively. Another interesting physical example is the quantum dynamics with the su(1, 1) Hamiltonian given by (1.10) with time-dependent a and b introduced in Ref. 11, regarded as a Rabi problem with a complex transverse magnetic field. Finally, utilizing a nonlinear generalization of the Gorini-Kossakowski-Sudarshan-Lindblad equation in the case of the qubit, based on the nonlinear von Neumann equation (1.2) as a point of departure, the extension of the celebrated Jaynes-Cummings model was introduced in Ref. 12 describing the interaction of a two-level atom with a single mode of the electromagnetic field.
In this work following the approach taken up in the case of the qubit we study the much more complicated case of the qutrit. The paper is organized as follows. In Sect. 2 we introduce the nonlinear system that is the counterpart of (1.12) in the case with the qutrit. Section 3 is devoted to the linear case referring to the standard von Neumann equation. Section 4 deals with the nonlinear system. All the necessary identities corresponding to the su(3) algebra are collected in Appendix.

(2.6)
We remark that there exist in the literature the alternative parametrizations of ∂Ω p (see for example [14]).
Linear and integrable nonlinear evolution of the qutrit 5 Now proceeding analogously as with the qubit states we set and On substituting (2.7) and (2.8) into (1.2) and using the identities (A.30), (A.32) and (A.33) we arrive at the following Riccati system of nonlinear ordinary differential equationṡ where a ∧b is the antisymmetric product of vectors a and b (see Appendix) . It thus appears that the nonlinear quantum evolution equation (1.2) with ρ(t), H and G given by (2.7) and (2.8), respectively, reduces to the nonlinear classical system (2.9).

Linear evolution of a qutrit: periodic and quasiperiodic solutions
We now restrict to the case b = 0. The system (2.9) reduces then to the linear one corresponding to the linear von Neumann equation. Namely, we havė In opposition to the evolution of the qubit discussed in [1] even in such a linear case the problem of finding the solution to (3.1) by means of the transformation (1.1) is in general complex and leads to cumbersome formulas. To be more specific consider first the problem of calculating the exponential e a·λ . Referring to (1.1) and (2.1) we point out that in view of (A. 19) and (A.24) an arbitrary power of a · λ has the expansion of the form (a · λ) n = c n + d n a · λ + e n (a * a) · λ, n = 0, 1, 2, . . . , (3.2) where c n , d n and e n are the scalar coefficients. On using (A. 19), (A.24) and (A.30) we arrive at the following system of recurrence equation 3) subject to the initial data c 0 = 1, The characteristic equation corresponding to (3.4) is of the form The discriminant of the cubic equation (3.5) is where |a| designates the norm of the vector a. The discriminant (3.6) fulfills Q 0. The simplest case Q = 0 refers to the condition a * a = ±|a|a and will be discussed later. For Q < 0 we have the trigonometric solution [15] to (3.5) such that where cos α = a · (a * a) |a| 3 .
Linear and integrable nonlinear evolution of the qutrit 7 An immediate consequence of (3.9), (3.10) and (3.11) is the following formula for the exponential e τ a·λ e τ a·λ = 1 12 As a matter of fact, setting τ = −it in (3.12) and making use of (1.1), (2.7) and (2.8) one can obtain the solution to (3.1). Nevertheless, as mentioned earlier the formulas are cumbersome and we decided not to present them herein. The numerical integration of (3.1) shows that the typical trajectory is the quasiperiodic one such as that depicted in Fig. 1. We remark that quasiperiodic trajectories were absent in the case with the evolution of the qubit [1]. The remaining cases include periodic (see Fig. 2) and stationary solutions. In the next sections we introduce explicit solutions to (3.1) in the special cases with a * a = ±|a|a, a · (a * a) = 0, and the diagonal generator a · λ.
Consider now the stationary solutions to (3.1). By virtue of (A.20) the most general form of these solutions is where µ and ν are constants. From (2.2) it follows that the statesξ correspond to the following region in the (µ, ν) plane The pure states (2.4) lying in the boundary ∂Ω p are given by (3.13), where µ and ν are the solution of the system where the second equation of (3.15) is a consequence of the relationξ ·(ξ * ξ) = 1. Finally, the mixed states (2.5) on the boundary ∂Ω m are specified by The concrete examples of the stationary solutions to the system (3.1) are discussed in the following sections.
Linear and integrable nonlinear evolution of the qutrit 9 3.1 The case of a * a = |a|a We first recall that the condition a * a = |a|a refers to Q = 0 in (3.6). Taking into account (3.2) we find that in this case Hence, proceeding as with (3.2) and using (A.31) we obtain the system of recurrence equations satisfied by c n and d n . Solving the second order recurrence satisfied by d n we get On putting τ = −it and making use of (1.1), (2.7) and (2.8) we arrive after some calculation to the following formula for the solution to (3.1) Using (3.13) and (3.15) we find that the system (3.1) has in the discussed case the stationary solution representing the pure state of the form Evidently, the stationary solution (3.20) is not asymptotically stable (this is not any limit of t going to infinity of (3.19)). Furthermore, the stationary solution representing the mixed state on the boundary of the state space is given byξ .
Finally, we have a family of stationary solutions corresponding to mixed states in the interior of the space of states such that where σ > 2. Of course,ξ ′′ approaches 0 referring to maximally mixed state in the limit σ → ∞.
The projections of the periodic trajectory (3.19) onto the planes (ξ i , ξ j ), i = j, i, j = 1, 2, . . . , 8 are circles, ellipses, segments and points so it is a plausible counterpart of the periodic evolution of the qubit.

The case with a * a = −|a|a
It can be easily checked that the solution to (3.1) in the case of a * a = −|a|a can be obtained from the solution (3.19) to (3.1) with a * a = |a|a by the formal replacement a → −a and t → −t. Hence we get for a * a = −|a|a Analogously, applying the transformation a → −a to (3.20), (3.21) and (3.22) we arrive at the stationary solutions of the form (mixed state on the boundary of the state space), , σ > 2 (mixed state in the interior of the space of states).

(3.26)
It is clear that qualitative behavior of solutions to (3.1) given by (3.23) is the same as in the case with a * a = |a|a.

3.3
The condition a · (a * a) = 0 Using the system of recurrence equations (3.3) for a · (a * a) = 0, we easily find (3.27) On setting τ = −it and using (1.1), (2.7) and (2.8) we obtain after some calculation the solution to (3.1) such that Further, setting ν = 0 and a 2 µ 2 + |a| 4 ν 2 = 1 3 in (3.16) with a · (a * a) = 0, we obtain the following stationary solutions corresponding to the mixed states on the boundary of the space of states and Finally, assuming that a 2 µ 2 + |a| 4 ν 2 = 1 3 we find the stationary solutions (3.13) referring to mixed states in the interior of the space of states specified by (3.32)
We point out that since a * a = (0, 0, 2a 3 a 8 , 0, 0, 0, 0, a 2 3 − a 2 8 ), the vector a satisfies the condition a * a = |a|a for a 3 =0 and a 8 0, and for a 3 = ± √ 3a 8 and a 8 > 0, so we then have the case discussed in Sect. 3.1. Analogously, for a 3 = 0 and a 8 0 as well as for a 3 = ± √ 3a 8 and a 8 < 0, we get the condition a * a = −|a|a and we then deal wit the case investigated in Sect. 3.2. Furthermore, for a 8 = 0 and the condition a 8 = 0 and a 3 = ± 1 √ 3 a 8 , we have a · (a * a) = 0, so we then have the case analyzed in Sect. 3.3.
Now it can be easily checked that .

(3.36)
On setting τ = −it and maing use of (1.1), (2.7) and (2.8) we arrive after some calculation at the following solution of (3.1) (3.37) The solution (3.37) is in general three-frequency quasi-periodic solution. This observation is consisitent with the general formula (3.12) for the exponential of a · λ. Examples of the trajectories given by (3.37) that are the Lissajous curves are illustrated in Fig. 3. We now discuss the stationary solutions. From (3.13) and (A.10) it follows that the stationary solutions in the discussed case are of the formξ = (0, 0,ξ 3 , 0, 0, 0, 0,ξ 8 ). Hence we find that the stationary solutions referring to the pure states areξ 1 = (0, 0, 0, 0, 0, 0, 0, −1), andξ 2,3 = (0, 0, ± √ 3 2 , 0, 0, 0, 0, 1 2 ). The stationary solutions corresponding to the mixed states on the boundary of the state space can be easily obtained from (2.5). Namely, we havē where κ is the parameter defined byξ 2 3 +ξ 2 8 = κ, so in the general case of the mixed states on the boundary it satisfies κ ∈ (0, 1). In the particular case of κ = 1 4 the relations (3.38) reduce toξ 8 = 1 2 andξ 3 = 0. Further, the trigonometric solution to the cubic equation satisfied byξ 8 results in state space are specified by the system of the inequalities such that (see (2.2)) ξ2 3 +ξ 2 8 < 1, 2ξ 3 8 − 6ξ 2 3ξ8 + 3(ξ 2 3 +ξ 2 8 ) < 1. (3.40) The region in the plane defined by (3.40) is depicted in Fig. 4. The vertices of the triangle correspond to the pure states and its edges refer to the mixed states on the boundary of the space of states. The parametrization of edges is given by (3.38) and (3.39). We finally point out that the case of the diagonal Gell-Mann matrices was analyzed in the context of the properties of the states of qutrits in [16,17].

Nonlinear evolution of a qutrit
We first confine ourselves to the case of a = 0, i.e. H = 0. The system (2.9) takes then the forṁ Proceeding analogously as with (3.1) we first consider the case with b * b = |b|b. Taking into account (3.18) with τ = t, (1.1), (2.7) and (2.8), we get the solution to (4.1) of the form where Taking into account (4.1) one can easily check that we have in the discussed case the following stationary solution It follows from numerical calculations that the stationary solution (4.5) is asymptotically stable. An easy example of solution to (4.1) going toξ as t → ∞ can be obtained from (4.4) by setting ξ 0 = 0. The problem of finding the stationary solutions to (4.1) that are not asymptotically stable seems to be a difficult task. Such solution proportional to b is of the form .
The system (4.1) has in the discussed case the stationary solution such that . (4.9) The numerical simulations indicate that the stationary solution (4.9) is asymptotically stable. The solution (4.9) represents the mixed state on the boundary of the space of states. We point out that there are no asymptotically stable stationary solutions representing mixed states in the case of the qubit [1]. The asymptotically stable stationary solution (4.9) is not globally stable. Consider for example the case of b = (0, 0, The solution (4.10) represents the pure state. It is clear that it is not the unique one.

The condition
We now discuss the case that is the nonlinear counterpart of the condition investigated in Sect. 3.3. Applying the algorithm used for calculation of (3.28) we arrive at the following solution to (4.1) given by (4.2) and the relations (4.12) On making the ansatz of the form (3.13) with a replaced by b one can easily obtain the following stationary solutions to (4.1) All the stationary solutions represent the pure states. The numerical calculations show that the solutionξ is asymptotically stable and the solutionsξ ′ andξ ′′ are unstable.

Rational solution
We finally discuss the case when the series representing the exponential from (1.1) such that e −it(H+iG) = e −it(a+ib)·λ truncates and becomes a polynomial, so the solution to (2.9) is rational. This is the only simple example involving nonvanishing both a and b. We recall that in the case of the qubit the rational solution was intermediate between the periodic and hyperbolic motion. Taking into account (A.31) we find that whenever a 2 = b 2 , a · b = 0, and a * a = b * b, a * b = 0, then e −it(a+ib)·λ = 1 − it(a + ib) · λ.
(4.21) We remark that the conditions satisfied by a and b in the case of the rational solution are the most natural generalizations of those taking place for the qubit such that a 2 = b 2 and a · b = 0. It should also be noted that relations satisfied by a and b imply a · (a * a) = 0 and b · (b * b) = 0 following directly from (A.27). Now, making use of (4.21) we get after some calculation the following solution to (2.9) expressed by (4.2) with On putting ξ 0 = 0 and taking the limit lim t→∞ ϕ(t) , we arrive at the asymptotically stable stationary solution of the form This stationary solution satisfies ξ 2 = 1 2 and 3ξ 2 − 2ξ · (ξ * ξ) = 1, so it represents a mixed state on the boundary of the space of states. An interesting nontrivial property of the nonlinear system (2.9) absent in the case of the qubit, is the existence of the limit cycles. More precisely, it follows from the numerical calculations that whenever a 2 > b 2 , a · b = 0, a · (a * a) = 0, a · (b * b) = 0, and b · (a * a) > 0, then the trajectories go to the periodic solution represented in the projections onto the planes (ξ i , ξ j ) or (ξ i , ξ j , ξ k ), i, j, k = 1, 2, . . . , 8 by ellipses. We point out that the first two conditions a 2 > b 2 and a · b = 0 refer to periodic solutions in the case of the nonlinear evolution of the qubit [1]. The limit cycle of the system (2.9) is illustrated in the Fig. 5. It should be noted that the parameters of the limit cycles depend on initial data. As with the nonlinear evolution of the qubit there exist for a · b = 0 the spiral trajectories going to equilibrium points (stationary solutions) corresponding to the pure states that are combinations of periodic and hyperbolic motion. An example is of the form a = (1, 1, 0, 2, −2, 1, 0, 1), b = (0, 0,   (4.11). In opposition to the case of the qubit we have also the spiral solutions corresponding to a·b = 0. The conditions for parameters can be then obtained from that holding in the case of the limit cycle by omitting some of requirements. An example is a = (0, 1, 0, 2, −2, 1, 0, 0) and b = (0, 0,  Fig. 6. We finally point out that the Gell-Mann matrices λ 1 , λ 2 and λ 3 are generators of the SU (2) group, so we can recover all solutions obtained in the case of the qubit [1] by putting a = (a 1 , a 2 , a 3 , 0, 0, 0, 0, 0), b = (b 1 , b 2 , b 3 , 0, 0, 0, 0, 0) and ξ 0 = (ξ 01 , ξ 02 , ξ 03 , 0, 0, 0, 0, 0).

Dynamics of entropy for qutrit states
Our purpose now is to study the temporal development of entropy under the evolution of the qutrit state given by (1.1), (2.7) and (2.8). Solving the characteristic equation det[ρ(t) − νI] = 0, where ρ(t) is given by (2.7), that is cubic in ν, we get the following trigonometric solution Hence, we obtain the following expression for the entropy of the qutrit states Notice that the maximum value 1 of the entropy S(t) corresponds to |ξ(t)| = 0 i.e. maximally mixed state, and its minimum value 0 is reached at |ξ(t)| = 1, ξ(t) * ξ(t) = ξ(t) referring to the pure state (see (2.4)). The entropy is periodic function of time for periodic solutions to (2.9) corresponding to mixed states and as it follows from numerical calculations, decreasing function of time involving possibility of dumped oscillations, for solutions going to equilibrium points on the boundary of the space of states except of those with the initial data corresponding to pure states. Further, the entropy is constant for stationary solutions and trajectories on the boundary ∂Ω p . The novel behavior of the entropy that is absent in the evolution of qubit states includes the case of the auto-oscillations connected with the existence of the limit cycles illustrated in Fig. 7. Another example are the horizontal asymptotes corresponding to the asymptotically stable stationary solutions connected with the mixed states on the boundary of the state space.

Conclusions
In this work we study the nonlinear generalization of the von Neumann equation. An advantage of such nonlinear generalization in comparison with alternative approaches is that it preserves the convexity of the space of quantum states. This means that the introduced nonlinearity does not violate the probabilistic structure of quantum mechanics. Another asset of the approach taken   Fig. 4, entropy is then decreasing function of time. Clearly such situation cannot occur in the case of an isolated system. As a concrete realization of the nonlinear von Neumann equation we have chosen the nontrivial case of the qutrit. As one would expect, this three-level system exhibits richer dynamics than does the qubit analyzed in the recent work [1]. In particular, the classical Riccati system (2.9) corresponding to the qutrit has quasiperiodic solutions and limit cycles that are absent in the case of the qubit. Interestingly, the quasiperiodic trajectories in the investigated 24 Krzysztof Kowalski three-level quantum system such as that illustrated in Fig. 3 have counterpart in classical orbital mechanics called Lissajous orbits that an object can follow around the Lagrangian point of a three-body system. The existence of the limit cycles of the system (2.9) and consequently self-oscillations of the qutrit is remarkable. The author does not know any other example of the limit cycle in a simple quantum system like that described in this work. For instance, one can find in the literature limit cycles in renormalization group behaviour of quantum Hamiltonians [24]. Self-oscillations as dynamical peculiarity of open systems are well-known in classical physics. For example we can read in Ref. 25 that: "Self-oscillation system is an open system because emergence and maintenance energy is given by an external source." or [26]: "Self-oscillatory systems have some properties that differ from those of harmonic oscillators. First, they exhibit undamped oscillations by taking and dissipating energy from various sources; thus, the systems have typical characteristics of nonequilibrium open systems." Referring to the harmonic oscillator mentioned in the second quotation we point out that the closed, periodic trajectory depicted in Fig. 5 is reached only asymptotically. An example of connection between self-oscillations of energy and entropy of a mechanical system was provided in Ref.
27. An important example of self-oscillations in open systems far from equilibrium is the temporal celebrated dissipative structures in the form of sustained oscillations, illustrated by biological rhythms [28] playing the fundamental role in the biological self-organization. Bearing in mind the existence of the limit cycle in the case of the qutrit and its lack for the qubit's time evolution, it is worthwhile to note that occurrence of dissipative structure in the case of chemical reactions takes place for three-particle reactions [29] that are less probable than two-particle ones. Referring back to the ringing structure in the entropy in Fig. 7 we point out an intriguing interpretation of such entropy oscillations [30] as a process of self-organization and process of disorganization of an open system arising after reaching some critical level of organization. It is further suggested that these oscillations can model action of mankind such as constructive one like: "building houses and factories, partition off rivers by dams and etc." and destructive: such as "disintegration of ecosystems, destructive fluctuations of the climate." and so on. As an advantage of this approach the author of Ref. 30 indicates that "The understanding of the reasons of the given tendency not only enables to foresee future hardships, but also prompts a way to avoid them." The fundamental character of the notion of the qutrit in quantum information theory suggests that the observations of this work concerning the nonlinear evolution of the qutrit would be of importance in quantum information processing. We also remark that some of the identities presented in Appendix concerning star and wedge products related to the structure constants of the su(3) algebra are most probably new. It seems that they would be a useful tool in the study of the qutrit and other problems connected with the SU (3) symmetry.