4 0 0 2 n a J 1 2 BERRY PHASE WITH ENVIRONMENT: ] CLASSICAL VERSUS QUANTUM l l a h - s e m Robert S. Whitney . t D´epartement de Physique Th´eorique, Universit´e de Gen`eve, a m CH-1211 Gen`eve 4, Switzerland - d n Yuriy Makhlin o Institut fu¨r Theoretische Festko¨rperphysik, Universita¨t Karlsruhe, 76128 Karlsruhe, Germany c [ Landau Institute for Theoretical Physics, Kosygin st. 2, 117940 Moscow, Russia 1 v 6 Alexander Shnirman 7 Institut fu¨r Theoretische Festko¨rperphysik 3 Universita¨t Karlsruhe, 76128 Karlsruhe, Germany 1 0 4 0 Yuval Gefen / t Department of Condensed Matter Physics a m The Weizmann Institute of Science, Rehovot 76100, Israel - d n Abstract We discuss the concept of the Berry phase in a dissipative system. We o showthatonecanidentifyaBerryphaseinaweakly-dissipativesystem c andfindtherespectivecorrection tothisquantity,inducedbytheenvi- : v ronment. Thiscorrectionisexpressedintermsofthesymmetrizednoise i powerandisthereforeinsensitivetothenatureofthenoiserepresenting X the environment, namely whether it is classical or quantum mechani- r cal. It is only the spectrum of the noise which counts. We analyze a a model of a spin-half (qubit) anisotropically coupled to its environment and explicitly show the coincidence between the effect of a quantum environment and aclassical one. Keywords: adiabaticity, Berry phase, dissipative dynamics, Lamb shift 1 2 Introduction Three papers published independently in 1932 by Zener, Landau and Stueckelberg (Landau, 1932; Zener, 1932; Stueckelberg, 1932) have in- troducedthephenomenonknowntodayasLandau-Zenertunneling. The idea is to consider a 2-level system, where the energy of each level varies linearly with a classical variable (which, in turn, is varied linearly in time). As function of time, t, the energy levels should intersect but for theinter-levelcoupling∆whichgivesrisetoan“avoided crossing”inthe spectrum, cf. Fig.1. Using the spin notation, one can write the Hamil- tonian as Hˆ = αtS +∆S . HereS =σ/2, andσ ,σ arePauli spin-1/2 z x z x operators; α is the rate of change of the energy of the pseudo-spin at asymptotic times. The avoided crossing gap is ∆. The probability of transition from, say, the lower level at time −∞ , to the upper level at time +∞ is given by P = exp[−(π/2)∆2/α]. LZ Besides being ubiquitous in physics and chemistry, the Landau-Zener framework appears to suggest a natural definition for the notion of adiabaticity. The adiabatic limit is approached when P << 1, i.e., LZ α << ∆2. The latter inequality involves a comparison of the rate of change (of the time dependent term in the Hamiltonian) with the gap in the spectrum, ∆. This notion of the adiabatic limit has become widespread. Acloser look suggests that, ingeneral, adiabaticity cannot be associated with comparing the rate of change to the gap. Indeed, on one hand any finite, discrete-spectrum system is coupled, however weakly, to the rest of the universe. Hence the emerging spectrum is, at least in principle, always continuous and gapless. The naive view would then imply that the adiabatic limit cannot be approached. This, on the other hand cannot be correct: if we consider a finite system with a dis- crete spectrum, for which adiabaticity is well defined, it is inconceivable that an infinitesimal coupling to the continuum (rendering the overall spectrum continuous) will change its physics in a dramatic way. The resolution of this problem is provided by the observation that the cri- terion for adiabaticity involves not only spectral properties but also the matrix elements of the system-environment coupling. To gain some insight into this problem we focus here on the analysis of the Berry phase (Berry, 1984) in a weakly dissipative system. It is particularly timely to address this issue now given the recent experi- mental activities in realization of controlled quantum two-level systems (qubits), and in particular, the interest in observing a Berry phase (BP) (see, e.g., (Falci et al., 2000)). For instance, the superconducting qubits have a coupling to their environment, which is weak but not negligi- ble (Nakamura et al., 1999; Vion et al., 2002; Chiorescu et al., 2002), Berry Phase with Environment:Classical versus Quantum 3 and thus it is important to find both the conditions under which the Berry phase can be observed and the nature of that Berry phase. In this paper we appeal to a simple analysis of the problem. We first, in Section 2, consider a quantum-mechanical framework, where a perturbative approach is taken. When the environment is replaced by a single oscillator, a second-order perturbation analysis is straightforward and produces a result which allows for a simple interpretation. We then generalizethecalculationforahostofenvironmentalmodes. InSection3 weconsideratoymodelwheretheenvironmentisreplacedbyaclassical stochasticforce. Thequantities ofinterest, theLambshiftandtheBerry phase, are then calculated, and simple heuristic arguments are given to interpret the results. To complete the analogy with the analysis of the previous section, here the “single-oscillator environment” is replaced by a simple periodic classical force (of random amplitude). In Section 4 we summarize the relation between the quantum mechanical approach and the classical model in more general terms. 1. The system: spin + environment We begin in the conventional way by writing the Hamiltonian for the “universe” (system + environment) as Hˆ = Hˆ +Hˆ +Vˆ (1) syst env coupling The system is defined as the set of those quantum degrees of freedom thatoneisinterestedtocontrolandmeasure;theenvironmentconsistsof all the rest, namely those degrees-of-freedom we can neither control nor measure. Thecouplingbetween thesystem andenvironmentis V . coupling Thepropertiesof theenvironmentarecontrolled bymacroscopic param- eters, such as temperature. Our treatment below applies to a reservoir at either zero or a finite temperature. For our purposesit is sufficient to represent the environment by a sin- gle operator X which couples to a spin. The Hamiltonian then becomes Hˆ = −1µgB·σˆ − 1Xσ +Hˆ . (2) 2 2 z env Hereafter we put µg = 1. Below we express our results in terms of the statistical properties (correlators) of the environment’s noise, X(t). De- pending on the physical situation at hand, one can choose to model the environment via a bath of harmonic oscillators (Feynman and Vernon, 1963; Caldeira and Leggett, 1983). In this case the generalized coordi- nate of the reservoir is defined as X = λ x , where {x } are the coor- i i i dinate operators of the oscillators and {Pλi} are the respective couplings. Eq. 2 is then referred to as the spin-boson Hamiltonian (Leggett et al., 4 1987). Another example of a reservoir could be a spin bath (Prokof’ev and Stamp, 2000) 1. However, in our analysis below we do not specify the type of the environment. We will only assume that the reservoir gives rise to markovian evolution on the time scales of interest. More specifically, the evolution is markovian at time scales longer than a cer- taincharacteristictimeτ ,determinedbytheenvironment2. Weassume c thatτ is shorterthan thedissipativetime scales introducedbytheenvi- c ronment,suchasthedephasingorrelaxationtimesandtheinverseLamb shift (the scale of the shortest of which we denote as T , τ ≪ T ). diss c diss We further assume that τ ≪ t , the characteristic variation time of c P the field B(t). Moreover, under these conditions we may consider only lowest-order (in the system-environment coupling) contributions to the quantities of interest: energy shifts, BP and relaxation rates. Indeed, if one divides the evolution time interval into short domains (≪ t ), P longer than τ but shorter than T , fluctuations at different domains c diss areuncorrelatedandtheireffectcanbeanalyzedseparately. Atthesame time, for each domain (≪ T ) the effect of the noise is weak. Thus, diss to the leading order corrections to the dynamics may be described as corrections to the rates (energies) of the spin dynamics, which may be estimated perturbatively. We also consider an underdamped spin, with the dissipative times longer than the period of the coherent dynamics, T ≫ 1/B. This implies that the time windows alluded to above diss consist of numerous oscillations, in other words they are ≫ 1/B. Wehavechosen ananisotropic spin-environmentcoupling, ∝ σ . This z is a realistic model, e.g., for many designs of solid-state qubits, where thedifferentcomponentsofthe“spin”areinfluencedbyentirelydifferent environmental degrees of freedom (Nakamura et al., 1999; Vion et al., 2002; Chiorescu et al., 2002). While our analysis can be generalized to account for multiple-directional fluctuatingfields (Whitney et al., 2004), here we focus on unidirectional fluctuations (along the z axis). 1For any reservoir in equilibrium the fluctuation-dissipation theorem provides the rela- tion between the symmetrized and antisymmetrized correlators of the noise: SX(ω) = AX(ω)coth(ω/2T). Yet, the temperature dependence of SX and AX may vary depending on the type of the environment. For an oscillator bath, AX (also called the spectral den- sityJX(ω)) istemperature-independent, sothat SX(ω)=JX(ω)coth(ω/2T). Ontheother hand, foraspinbath SX istemperature-independent andisrelated tothe spins’densityof states,whileAX(ω)=SX(ω)tanh(ω/2T). 2Thistimemaybegivenbythecorrelationtimeofthefluctuations,butingeneralisamore subtlecharacteristic ofthe spectrum related toits roughness near qubit’s frequencies. Note furtherthatforsingularspectraτc maybeilldefinedandtheperturbativeanalysismayfail. See, e.g., (Bloch, 1957; Redfield, 1957; Slichter, 1978; Makhlin et al., 2003; Wilhelm et al., 2004;WhitneyandGefen,2004). Berry Phase with Environment:Classical versus Quantum 5 Another remark to be made concerns the possibility to observe a (weak) dissipative correction to Berry phase in spite of the dephasing and relaxation phenomena. While the respective time scales (T , T and 1 2 the inverse of the correction to the Berry phase) scale similarly with the strength of fluctuations (inversely proportionally to the noise power), they are dominated by different frequency domains. Indeed, the dephas- ing and relaxation are known to be dominated by resonant fluctuations with frequencies close to B (for the relaxation and the corresponding contribution to dephasing) and 0 (for the pure dephasing), cf. Eq. (15) below. In contrast, as we shall see below, the Lamb shiftand the correc- tion to the Berry phase accumulate contribution from the entire range of frequencies. Thus, one may think of (engineering) a system with an environment whose fluctuations at ν ∼ B and ν ∼ 0 are suppressed. In this case, one can easily observe an observable correction to the Berry phase at times when the dephasing and relaxation are still negligible. 2. Quantum-mechanical analysis In this section we consider a two-level system coupled to an environ- ment which we treat as a quantum-mechanical system. We begin with a discussion of the Lamb shift and then show, in Subsection 2.3, how the results for the Lamb shift may be used to find the environment-induced correction to the Berry phase and the relaxation times. 2.1 Lamb shift as level repulsion Consider first, for illustration, a simple system of the spin coupled to a single oscillator, with the Hamiltonian H = −1Bσ − 1cσ (a†+a)+ω a†a , (3) 2 z 2 x 0 where c is the coupling constant. Let |ni denote the n-th level of the oscillator; the second-order corrections to the energies of thestates |↑,0i and |↓,0i are |h↑,0|V|↓,1i|2 1 c2 (2) E = − = − , (4) ↑ ω +B 4 ω +B 0 0 and |h↓,0|V|↑,1i|2 1 c2 (2) E = − = − , (5) ↓ ω −B 4 ω −B 0 0 where V ≡ (c/2)σ (a† + a) is the perturbation. This results in the x following correction to the level spacing E −E : ↓ ↑ c2 B (2) (2) E −E = . (6) ↓ ↑ 2 B2−ω2 0 6 This correction (the Lamb shift) has different signs for fast (ω > B) 0 and slow (ω < B) oscillators. As one can see from Eqs. (4), (5), this 0 result can be understood in terms of the level repulsion (Wilhelm et al., 2004): theperturbationcouplesthelevel|↑,0ito|↓,1iand|↓,0ito|↑,1i. The levels of the latter pair are closer, and the coupling has a stronger effect on their energies. They repel each other due to the coupling, thus reducingthe distance between |↑,0i and |↓,0i for ω > B and increasing 0 it for ω < B. 0 2.2 Second-order perturbative analysis In this section we find the Lamb shift using the lowest-, second-order perturbativeanalysis. IntheHamiltonian (2)wetreatthecouplingterm V = −1Xσ as a perturbation: H = H + V. The eigenstates of H 2 z 0 0 are |α,ii, where α =↑ /↓ denotes the eigenstates of the spin without B B dissipation, withthespindirection parallel or antiparallel to thefiledB, and i denotes eigenstates of the environment. The perturbation theory gives for the corrections to their eigenenergies: |hα,i|V|β,ji|2 (2) E = − . (7) α,i (0) (0) (0) (0) Xβ,j Eβ +Ej −Eα −Ei −i0 For V = −1Xσ we notice that h↑ |σ |↑ i2 = h↓ |σ |↓ i2 = cos2θ 2 z B z B B z B and h↑ |σ |↓ i2 = h↓ |σ |↑ i2 = sin2θ, and find for the environment- B z B B z B (2) (2) averaged quantities E ≡ ρ E (see the discussion of these quan- α i i α,i tities at the end of this subsPection): cos2θ ρ |hi|X|ji|2 sin2θ ρ |hi|X|ji|2 (2) i i E = − − . ↑ 4 Xi,j Ej(0) −Ei(0) −i0 4 Xi,j B+Ej(0) −Ei(0) −i0 (8) Thecorrection toE isobtainedbysubstitutingB → −B intotheabove ↓ equation. Now using the identity 1 ∞ = i dte−i(E−i0)t, (9) E −i0 Z 0 we rewrite Eq. (8) as i ∞ E(2) = − dthX(t)X(0)i cos2θ+sin2θe−iBt e−0t, (10) ↑ 4Z0 (cid:16) (cid:17) where we have used the relation hX(t)X(0)i = ρ hi|X|jihj|X|iie−i(Ej−Ei)t. (11) i Xi,j Berry Phase with Environment:Classical versus Quantum 7 In terms of the the Fourier transform hX2i ≡ dthX(t)X(0)ieiνt we ν obtain R 1 dν hX2i 1 dν hX2i E(2) = − cos2θ ν − sin2θ ν . (12) ↑ 4 Z 2π ν −i0 4 Z 2π ν +B −i0 (2) (2) (2) For the Lamb shift E ≡ ℜe(E −E ) this gives a principal value Lamb ↓ ↑ integral dν S (ν) ∞ dν S (ν) E(2) = 1 sin2θ P X = Bsin2θ P X , (13) Lamb 2 Z 2π B −ν Z 2π B2−ν2 0 where S (ν)≡ 1(hX2i+hX2 i) = 1 dth[X(t),X(0)] ieiνt. (14) X 2 ν −ν 2Z + Thus the Lamb shift is expressed in terms of the symmetrized correlator S and is insensitive to the antisymmetric part of the noise spectrum. X As one can see from Eq. (13), in agreement with the discussion in the previous section, the high-frequency noise (ν > B) reduces the en- ergy gap between the spin states (Leggett et al., 1987), while the low frequency modes (ν < B) increase the energy gap. Similarly, from Eq. (12) one can evaluate the dephasing time: 1 cos2θ sin2θ (2) (2) = −ℑm(E +E ) = S (ν = 0)+ S (ν = B). (15) T ↑ ↓ 4 X 4 X 2 This expression correctly reproduces the contribution of the transverse fluctuations (∝ sin2θ) to the dephasing rate, but underestimates the longitudinal contribution (∝ cos2θ) by a factor of two (cf. Ref. (Bloch, 1957; Redfield, 1957; Weiss, 1999)). One can show that an accurate evaluation of this contribution, as well as the analysis of the relaxation, requires taking into account corrections to the eigenstates, and not only to the eigenenergies (7). More precisely, our calculation of the correc- tions to the eigenenergies in this subsection corresponds to evaluation only of the four left diagrams in Fig. 7 of Ref. (Makhlin et al., 2003); the term i0 in the denominators allows one to find also the outgoing transition rates from the eigenstates (and the respective contribution, ∝ sin2θ, to dephasing) but only the part of the ‘pure-dephasing’ rate, ∝ 1 cos2θ. Analysis of the two remaining diagrams in Fig. 7 and those 4 in Fig. 6 allows one to find also the pure dephasing rate (as well as the incoming transition rates, the latter though do not require an extra evaluation due to probability conservation). 8 2.3 From Lamb shift to Berry phase So far we have analyzed the environment-induced correction to the level splitting (the Lamb shift). Using the results above one can eval- uate also the environment-induced correction to the Berry phase for a slow cyclic variation of the magnetic field B (Whitney and Gefen, 2001; Whitney and Gefen, 2003; Whitney et al., 2004; Whitney and Gefen, 2004). Indeed, consider the simplest case of conic variations of the field around the z-axis (to which the environment is coupled), as shown in Fig. 1: the field varies at a constant rate, with the low angular velocity ω , and traverses the circle after the period t ≡ 2π/ω . The analysis B P B of the spin dynamics is considerably simplified by going to the frame, rotating with the angular velocity ω zˆ, where zˆis the unit vector along B the z-axis. In this frame the spin is subject to the fluctuating field Xzˆ and the field B+ω zˆ, which is stationary. Thus, in this frame one can B usetheresults of theanalysis above to obtain theLamb shift,if onesub- stitutes B by B+ω zˆ. In other words, the correction to the Lamb shift B associatedwiththevariationofthefieldB intime,isgivenbytakingthe derivative ω ∂ oftheLambshift(13)andmultiplyingbytheperiodof B Bz variation, t . After a full period the basis of the rotating frame makes a P complete circle and returns to its initial position, i.e. coincides with the laboratory frame’s basis. Hence the phases accumulated in the rotating and laboratory frames coincide, and it is sufficient to evaluate it in the rotating frame. Thus, one finds the environment-induced correction to the Berry phase to be ∂E (B) Lamb δΦ = 2π . (16) BP ∂B z Taking the derivative of Eq. (13), we find: S (ν)(2ν −3B) δ(2)ΦBP = cosθsin2θ PZ dν X2B(B −ν)2 . (17) (Notice the convention: this expression gives the correction to the rela- tive Berry phase between the spin-up and spin-down states, rather than to the phases of each of these states.) As for the Lamb shift, the contri- butionsofthehigh-andlow-frequency fluctuationsareof oppositesigns. For the Berry phase the contribution changes sign at ν = 3B/2. Inpassingwenotethatthisanalysiscanbegeneralizedtoanarbitrary (but adiabatic) path B(t), this enables one to see that the correction to the Berry phase is geometric, but that its geometric nature is very Berry Phase with Environment:Classical versus Quantum 9 z ω B B θ X(t) Figure 1. different from the Berry phase of an isolated spin-half (Whitney et al., 2004). In Section 3 we shall find exactly the same expression for the Lamb shift and therefore for the Berry phase in the case of classical environ- ment. 2.4 High-frequency noise: renormalization of the transverse B-field Consider now the influence of the high-frequency fluctuations in the environment only (ν ≫ B). Since the frequencies of the fluctuations are much higher than the typical spin-dynamics frequencies, one may eliminate these high-frequency fluctuations using the adiabatic (Born- Oppenheimer)approximation,asdescribed,e.g.,byLeggettetal. (Leggett et al., 1987). Indeed, consider the spin-boson model, with the Hamiltonian H = −1(B +Xzˆ)σ+H , (18) 2 env † † where X = c (a + a ) and H = ω a a . Let us ignore the i i i i env i i i i low-frequencyPoscillators and focus on thoPse at high frequencies ν ≫ B. Thesefastoscillatorsadjustalmostinstantaneouslytotheslowlyvarying spin state. For the last two terms of the Hamiltonian (18) two lowest- energy states are ˜↑ = |↑i g↑ and ˜↓ = |↓i g↓ . Here g↑ i i i i i (cid:12) E (cid:12) E (cid:12) E (cid:12) E (cid:12) E denotes the ground(cid:12) state of tQhe(cid:12)ith oscilla(cid:12)tor correspQon(cid:12)ding to the (cid:12)spin (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) † † ↓ state |↑i, i.e. thegroundstate of ω a a +c (a +a ), and g is defined i i i i i i i (cid:12) E similarly; further eigenstates of the last two terms are se(cid:12)parated by a (cid:12) gap ∼ ν. 10 Consider now the matrix elements of the first term −1Bσ in this 2 two-state low-energy subspace; one finds that its transverse component is suppressed by the factor ∞ dν J (ν) g↑ g↓ = exp(−c2/2ω2) = exp − X , (19) Yi D i(cid:12)(cid:12) iE Yi i i (cid:18) Z0 2π ν2 (cid:19) (cid:12) where J (ν) ≡ π c2δ(ν −ω ) is the spectral density of the oscillator X i i i bath. At a finite tPemperature T each high-frequency oscillator remains in its thermal equilibrium state (subject to the spin state), rather than the ground state, and on the rhs of Eq. (19) the spectral density J (ν) X is replaced by the thermal noise power S (ν) = J (ν)coth(ν/2k T). X X B Thustheroleofthehigh-frequencyoscillators istosuppressthetrans- versefield component(inother words, thetransverseg-factor). If weare interested only in the contribution to the level spacing (the Lamb shift), one should consider only the longitudinal (k B) part of the renormaliza- tion, i.e. multiply the result by sinθ, to obtain Eq. (13). 2.5 Effective-action analysis Onecanstudythespindynamicsintegrating outtheenvironmentand using the effective action for the spin. We derive the effective action us- ing the Feynman-Vernon-Keldysh technique. For the interaction −Xs z with the z-component of the spin, the effective action (the influence functional) reads 1 iΦ = − dt dt′s (t)·s (t′)[iG (t,t′)], (20) infl z z X 2Z Z CK CK where we assumed the Gaussian statistics of X, and defined the Green function G as iG (t,t′) = hT X(t)X(t′)i. The time ordering here X X CK refers to the Keldysh time contour C , and in Eq. (20) we integrate K over C ; accordingly each of the time dependent variables assumes a K ‘Keldysh index’ u,d indicating the upper/lower branch of this contour. After the Keldysh rotation one obtains the influence functional in terms of the classical and quantum components, sc ≡ (su +sd)/2 and z z z sq ≡ su−sd: z z z 1 Φ = − dtdt′ sq(t)GR(t−t′)sc(t′)+ sq(t)GK(t−t′)sq(t′) , infl Z (cid:20) z X z 4 z X z (cid:21) (21) in terms of the retarded and Keldysh Green functions, GR ≡ −iθ(t− X t′)h[X(t),X(t′)] i and GK ≡−ih[X(t),X(t′)] i = −2iS (t−t′). − X + X