Abstract:In classic statistical physics, an isolated system corresponds to a constant energy shell in the phase space, which can be described by the microcanonical ensemble. While, for an isolated quantum system, the conventional treatment is to subject the system to a narrow energy window in the Hilbert space instead of the energy shell in classical phase space, and then confine the participating eigen states of system wave function in the narrow window, so that the microcanonical ensemble can be recovered in the framework of quantum mechanics. Apart from the traditional theory, there is a more self-consistent description for the isolated quantum system, that is, the quantum microcanonical (QMC) ensemble. The QMC ensemble abandons the narrow energy window assumption, and allows all the eigen states to contribute to the system wave function on condition that the system average energy is fixed at a given value. At the same time, the total occupation probability of these eigen states is conserved to unity. The most probable probability distribution obtained in the Hilbert space for an isolated quantum system according to the constraints specified above is called the QMC statistics. There is a clear difference between the QMC distribution and the traditional Gibbs distribution having an exponential form. Through the external periodic drives, an isolated quantum system may produce the QMC distribution, which is a consequence of the interplay between internal origins and external drives. In this paper, we investigate the conditions for the formation and suppression of QMC distribution by using the exact diagonalization method based on the one-dimensional Ising model. We start with the one-dimensional Ising model and focus on three different cases of periodic drives: systems under vertical (along the z axis), horizontal (along the x axis), horizontal magnetic field together with random internal (along the y axis) magnetic field. For all these three cases, the external magnetic fields are set to be ordinary rectangular pulses and the Gibbs distributions are taken as the initial states. We then study the evolutions and their asymptotic tendencies to the QMC distributions of the eigen state occupation probability under the effect of external periodic magnetic field. The results show that under the vertical magnetic field, the eigen state occupation probability does not change, and the system cannot produce the QMC distribution; under the horizontal magnetic field, the system tends to display a QMC distribution, but only partly; under horizontal and random internal magnetic fields at the same time, the transition to QMC distribution can be fully realized, and finally the system is almost completely thermalized. In order to clarify the different behaviors of the Ising model in the three cases, we also calculate the information entropy of the eigen state of Floquet operator in the eigen representation of the unperturbed Hamiltonian. We find that as the information entropy of the Floquet eigen state increases, the convergence to the QMC distribution in the Hilbert space is improved. We also notice that the mechanism for the emergence of QMC distribution is closely related to the thermalization effect of the isolated quantum system. Our analyses show that when the magnetic field is vertical, it cannot trigger the thermalization of the system. When the magnetic field is horizontal, the system becomes partly, but not completely, thermalized. When we add a horizontal periodic magnetic field and a random internal magnetic field at the same time, the system can be completely thermalized to infinite temperature. Thus, the asymptotic behavior towards the QMC statistics is a reflection of the fact that the isolated quantum system is thermalizable under periodic drives. Keywords:micro-canonical ensemble/ Ising model/ Floquet representation/ thermalization
3.计算结果这里将对前面所述的三种情况进行数值模拟. 首先考察伊辛模型加上纵向(沿$z$轴)磁脉冲的情况, 结果如图2所示, 此时占据概率p不随脉冲数$n$变化. 在图中, 绿线和红线是在给定平均能量, 即在$E_{{\rm{av}}}^{(n)} =\displaystyle \sum\nolimits_\alpha {E_\alpha ^{(n)}p_\alpha ^{(n)}}$约束条件下分别做e指数和QMC拟合的结果, 显然占据概率p始终保持初始时刻的吉布斯(e指数)分布没有变化. 图 2 一维伊辛模型在纵向(沿$z$轴)周期磁脉冲作用下本征态占据概率不发生任何变化, 其中$n$是磁脉冲作用的次数, 图上每一数据点代表相邻32个态的平均值, 绿线和红线分别是e指数和QMC拟合的结果 Figure2. The eigenstate occupation numbers of one-dimensional Ising model keep invariant under longitudinal periodic magnetic pulses (along the $z$-axis). Here $n$ is the number of magnetic pulses, each point in the graph represents the mean of 32 neighboring states, the green and red curves are the exponential and QMC fitting results, respectively.
图2所示结果来自分段平均, 其中每个点代表32个相邻能态数据的平均值. 图3计算了三种伊辛模型的能态密度(density of states, DOS), 其中粗红线对应${H_1}$, 蓝线对应${H_2}$, 细绿线对应${H_3}$. 可见${H_1}$和${H_2}$具有分立的能谱, 而且各能态有不同的简并度, 这种分立谱将导致对能态做分段平均时产生较大误差. 这个问题在图2中并不明显, 数据点与拟合曲线符合很好, 这是因为伊辛模型在纵向脉冲作用下占据概率实际上并没有发生演化. 如果将纵向脉冲改为横向脉冲, 则占据概率将发生明显变化, 这时能级不均匀性导致的误差将会显现. 为了减小这一影响, 本文在模拟横向脉冲作用的伊辛模型时, 取${H_3}$作为无扰动哈密顿量, 把磁场的短暂关闭作为一个小的扰动(如图1), 而${H_1}$则作为扰动后的哈密顿量. 由图3可见, ${H_3}$能态分布的均匀性和连续性都优于${H_1}$, 因而可较好地抑制误差. 同时${H_3}$比${H_1}$更加远离可积模型, 因而长脉冲具有更强的热化作用. 图 3 一维伊辛模型的态密度, 其中粗红线: 无外场伊辛模型; 蓝线: 纵场伊辛模型; 细绿线: 横场伊辛模型 Figure3. DOS of one-dimensional Ising model. Thick red curve: Ising model without external field; blue curve: Ising model with a longitudinal field; thin green curve: Ising model with a transverse field.
图4描绘了横向周期磁场作用下伊辛模型的情况, 这里占据概率分布随时间发生一些改变后就不再继续演化, $n$ 从2到超过100都基本停留在从吉布斯向QMC分布过渡的中间状态. 图 4 一维伊辛模型在横向(沿$x$轴)周期磁脉冲作用$n$次后本征态占据概率的分布 (a) n = 0; (b) n = 2; (c) n = 16; (d) n = 103. 图中每一点代表相邻32个态的平均值, 绿线和红线分别是e指数和QMC拟合的结果 Figure4. Distribution of eigenstate occupation numbers of one-dimensional Ising model after $n$ periodic transverse (along x-axis) magnetic pulses: (a) n = 0; (b) n = 2; (c) n = 16; (d) n = 103. Each point in the graph represents the mean of 32 states, the green and red curves are the exponential and QMC fitting results, respectively.
对于第三种情况, 横向磁场和局域随机磁场同时作用, 为便于和前面的情况比较, 外加磁场依然采用图1所示的长脉冲, 随机内磁场则不随时间变化. 模拟结果见图5, 在4个脉冲后占据概率已到达中间状态, 在14个脉冲后就表现出明显的QMC分布的特征, 即QMC统计比e指数统计更好地描述了占据概率分布. 这种状态可以持续一段较长时间, $n = 14$是其中的一个代表. 如果继续施加更多脉冲, 比如100个脉冲之后, 系统逐渐完全热化, 所有能态几乎被等几率地占据, 这时吉布斯分布和QMC分布变得难以区分. 总体上看, 当增加局域随机磁场后, 系统产生QMC分布的速度明显加快, 与吉布斯分布的差异也得到强化. 图 5 本征态占据概率分布 (a) n = 0; (b) n = 4; (c) n = 14; (d) n = 100. 与图4类似的情况, 区别是在模型中加入了微弱的随机局域磁场 Figure5. Distribution of eigenstate occupation numbers: (a) n = 0; (b) n = 4; (c) n = 14; (d) n = 100. The situation is similar to that of Fig. 4 except for the introduction of weak random local magnetic fields.
这里${C_{\mu k}}$为展开系数, ${S_\mu }$度量了${\psi _\mu }$与${\varphi _k}$的交叠程度, 即同${\psi _\mu }$有关联的${\varphi _k}$的多寡. 图6(a)和图6(b)分别计算了前面第2和第3种情况下系统所有的弗洛凯本征态的信息熵, 这里$\ln \left( {0.48 N} \right)$是高斯正交系综信息熵的最大值[22], $N$代表希尔伯特空间的维数, ${\varepsilon _\mu }$是弗洛凯算符的本征值, ${\varepsilon _\mu }\left( {{t_{{\rm{off}}}} + {t_{{\rm{on}}}}} \right)$表示弗洛凯相位, 取值在$ - {\text{π}}$到${\text{π}}$之间. 图中的红、绿、蓝点分别对应${t_{{\rm{off}}}} = $1, 2, 5时的信息熵. 作为参照, 图中的水平虚线标出了第1种情况下信息熵的大小, 显然这时${S_\mu }$处处为零, 表明${\psi _\mu }$与${\varphi _k}$正交, 因而将弗洛凯算符作用于系统时不引起任何变化. 对于第2种情况, 图6(a)显示${S_\mu } > 0$但数值不是很大, 表明${\psi _\mu }$与${\varphi _k}$有微弱的交叠, 这导致时间演化过程中不同${\varphi _k}$之间可能通过F的作用实现跃迁, 从而改变占据概率的分布. 但此时希尔伯特空间存在对应不同波矢的独立子空间, 占据概率的转移只发生在各个子空间内, 不能遍及整个希尔伯特空间. 如果引入局域随机场, 这些子空间将能实现联通, 占据概率传播的范围也随之扩大. 所以对于第3种情况, ${S_\mu }$显示${\psi _\mu }$与${\varphi _k}$交叠明显并且随着${t_{{\rm{off}}}}$的增大逐渐趋向深度混合, 表明在时间演化算符作用下, 一个本征态可以同其他多个态发生强烈的耦合, 使得占据概率的转移遍及整个希尔伯特空间, 进而能够表现出最可几分布, 即QMC统计规律. 图 6 一维伊辛模型在横向(沿$x$轴)周期磁脉冲作用下弗洛凯本征态在哈密顿量${H_{{\rm{on}}}}$的本征态表象下的信息熵 (a) 系统不含随机局域磁场; (b) 系统包含随机局域磁场. 其中红、绿、蓝点分别对应磁脉冲的时间间隔为${t_{{\rm{off}}}} = 1, \;2, \;5$, 黑色虚线是一维伊辛模型在纵向(沿$z$轴)周期磁脉冲作用下的信息熵分布 Figure6. Information entropy of the Floquet eigenstates in the eigenstate representation of ${H_{{\rm{on}}}}$ for the one-dimensional Ising model under transverse (along x-axis) periodic magnetic pulses: (a) Systems without random local magnetic fields; (b) systems with random local magnetic fields, where the red, green, and blue points correspond to magnetic pulse interval ${t_{{\rm{off}}}} = 1, \;2, \;5$, respectively, and the black dashed lines are the distribution of information entropy for the one-dimensional Ising model under longitudinal (along z-axis) periodic magnetic pulses.
上面的分析表明, QMC分布显现的机理与量子系统热化机制[23]有着密切关联. 图7计算了单自旋平均能量${E_{{\rm{av}}}}/{N_{\rm{S}}}$随施加脉冲数$n$的变化, 图7(a)和图7(b)分别对应第2和第3种情况. 其中图7(a)的平均能量明显只在某个非零值附近涨落而不会趋向于零, 即使改变${t_{{\rm{off}}}}$也没有明显变化, 表明系统发生了部分热化但并不彻底; 而图7(b)的演化则逐渐趋向于零点, 特别是${t_{{\rm{off}}}}$增大后更加明显, 说明此时系统可完全热化至无限温度状态. 在图7中, 黑色虚线给出了第1种情况下能量平均值的演化过程, 它呈现水平状且没有任何涨落, 表明此时系统完全没有被热化. 这些结果与三种情况下占据概率趋向QMC统计的演化特点, 以及信息熵揭示的希尔伯特空间的结构特征是一致的. 信息熵的计算无需对量子系统做长时间的动力学演化, 从它可以比较方便地判断系统能否热化以及热化发生的快慢. 由于QMC统计的显现不仅依赖希尔伯特空间的结构, 还依赖演化初态的选取等因素, 因此可否热化是QMC统计形成的一个必要但不充分条件. 图 7 一维伊辛模型在横场(沿x轴)周期磁脉冲作用下单个自旋平均能量随脉冲数n的变化 (a) 系统不含随机局域磁场; (b) 系统含随机局域磁场. 这里红方块、绿圆圈、蓝三角分别对应磁脉冲的时间间隔为${t_{{\rm{off}}}} = 1, \;2, \;5$, 黑色虚线是一维伊辛模型在纵向(沿z轴)周期磁脉冲作用下的单自旋平均能量 Figure7. Average energy per spin versus pulse number n for the one-dimensional Ising model under transverse (along x-axis) periodic magnetic pulses: (a) Systems without random local magnetic fields; (b) systems with random local magnetic fields, where the red squares, green circles, and blue triangles correspond to magnetic pulse interval ${t_{{\rm{off}}}} = 1, \;2, \;5$, respectively, and the black dashed lines are the average energy per spin for the one-dimensional Ising model under longitudinal (along z-axis) periodic magnetic pulses.