• <tr id="yyy80"></tr>
  • <sup id="yyy80"></sup>
  • <tfoot id="yyy80"><noscript id="yyy80"></noscript></tfoot>
  • 99热精品在线国产_美女午夜性视频免费_国产精品国产高清国产av_av欧美777_自拍偷自拍亚洲精品老妇_亚洲熟女精品中文字幕_www日本黄色视频网_国产精品野战在线观看 ?

    Variational quantum eigensolvers by variance minimization

    2022-12-28 09:52:26DanBoZhang張旦波BinLinChen陳彬琳ZhanHaoYuan原展豪andTaoYin殷濤
    Chinese Physics B 2022年12期
    關(guān)鍵詞:陳彬

    Dan-Bo Zhang(張旦波) Bin-Lin Chen(陳彬琳) Zhan-Hao Yuan(原展豪) and Tao Yin(殷濤)

    1Guangdong–Hong Kong Joint Laboratory of Quantum Matter,Frontier Research Institute for Physics,

    South China Normal University,Guangzhou 510006,China

    2Guangdong Provincial Key Laboratory of Quantum Engineering and Quantum Materials,School of Physics and Telecommunication Engineering,South China Normal University,Guangzhou 510006,China

    3Guangzhou Educational Infrastructure and Equipment Center,Guangzhou 510006,China

    4Yuntao Quantum Technologies,Shenzhen 518000,China

    Keywords: quantum computing,quantum algorithm,quantum chemistry

    1. Introduction

    The variational principle is instrumental for understanding physical theories and also becomes a powerful tool for solving computational physics problems. In recent years,the power of quantum computing of near-term noisy quantum processors is expected to be exploited with variational methods,[1–24]which refers to hybrid quantum-classical algorithms for optimization. A representative example is variational quantum eigensolver (VQE),[1,3,7,15,19,22]which receives special interests due to its fundamental roles in quantum chemistry,[1,3,7,22,25]quantum many-body systems,[13,19]and many other applications.[9,26,27]

    Interestingly,though called eigensolver,the original VQE is designed to solve the ground state for a Hamiltonian,since it uses the energy as a cost function to minimize,which promises to give an upper bound for the ground state energy. For being an eigensolver, many variants of VQE have been developed to solve excited states,which can be understood as minimizing energy in a subspace, constrained by symmetries,[28]enforced by orthogonality,[29]obtained by eliminating the space of lower energy states,[16]or using quantum subspace expansion.[5]As excited states are responsible for physical processes such as chemical reactions, advances in VQE for calculating excited states are arguably important.

    While those variants of VQE can be successful to some extent, an alternative approach is to directly access whether a state is an eigenstate and use this criterion as a cost function to optimize. A simple answer for this is to use zero energy variance to judge an eigenstate, since an eigenstate should have zero energy variance. Variational methods based on the zero-energy variance principle that run on classical computers,[30–36]in fact,can be originated in the early age of quantum mechanics for solving the Helium atom.[30]Remarkably,zero variance puts a very strong constraint on the wavefunction structure, and it can be powerful for solving quantum many-body problems when a well-approximated wavefunction is required.[35]The zero-variance principle has been applied experimentally to self verify if a quantum simulator can prepare a ground state for a Hamiltonian.[13]It is, however,awaiting for exploiting the power of quantum computing to develop a variational quantum eigensolver that is based on minimizing energy variance,which can solve ground state and excited states on the same footing.

    In this paper, we develop a variational quantum eigensolver based on minimizing energy variance,and we call it the variance-VQE. For solving excited states of a Hamiltonian,we give two approaches, which represent eigenstates in different ways. One uses a single wavefunction ansatz, where different parameters correspond to different eigenstates. The other uses a set of orthogonal wavefunction ansatzs,and incorporates energy variances into one cost function for optimization. We numerically solve the energy potential surface for molecules, which is a fundamental quantum chemistry problem, and demonstrate the properties of variance VQE as well as explore its potential advantages. We also show that optimizing a combination of energy and variance may be more efficient in finding a set of eigenstates with lowest energies,compared with one that optimizes energy or variance alone.Moreover,we investigate stochastic gradient descent for optimizing the variance-VQE with Hamiltonian sampling, which can be useful for optimization with fewer quantum resources.Our work demonstrates that VQE by minimizing energy variance can be useful for calculating excited states and also for working out an avenue to reach efficient optimization.

    The paper is organized as follows. We first propose variance VQE in Section 2,and then apply it to solve excited states for quantum chemistry problems in Section 3.In Section 4,we investigate the stochastic gradient for optimizing the variance-VQE.Finally,we give a summary in Section 5.

    2. Variance VQE

    In this section, we first review the traditional approach of variational quantum eigensolver that minimizes the energy.Then, we formulate a new type of VQE that minimizes the energy variance and then discusses its optimization.

    2.1. VQE by energy minimization

    We consider a HamiltonianHas a summation of local terms,H= ∑Ni=1ciLi, where a local HamiltonianLican be written as a tensor product of a few Pauli matrices. We denotec=(c1,c2,...,cN)TandL=(L1,L2,...,LN)T. Thus we can writeH=cTL.

    The variational quantum eigensolver works as follows.One uses an ansatz|ψ(β)〉=U(β)|R〉to represent a candidate ground state.Here|R〉usually is taken as an approximation for the ground state ofH,e.g.,Hartree–Fock state for the quantum chemistry problem.U(β)is a unitary operator parameterized withβ. The task then is to optimizeβfor some cost function.The traditional way is to minimize the energy defined as

    Here, we have denotedψ(β)=|ψ(β)〉〈ψ(β)|. By designing, VQE based on minimizing energy is suitable for finding ground state of a Hamiltonian.

    The optimization is completed with a hybrid quantumclassical algorithm. The quantum processor preparesψ(β)and performs measurements to evaluate?(β), which can be reduced into?(β) =cT?(β), where?(β) = Tr(ψ(β)L).Here, quantum average of each component ofLcorresponds to a joint measurement on multiple qubits. The classical computer updates parametersβaccording to received data from the quantum processor, which forms a hybrid quantumclassical optimization loop.

    2.2. VQE by variance minimization

    As the name indicates, variational quantum eigensolver should aim at solving eigenstates. However, VQE, based on minimizing energy,is prone to find only an eigenstate with the lowest energy, named as the ground state. Other VQEs have been developed for solving excited states. Essentially, those VQEs are realized by minimizing energy in a subspace,which is enforced with symmetry, or by eliminating space of lower energy states with an orthogonality condition.

    Can variational quantum eigensolver directly solve excited states? An answer for this is to design a cost function that can assign the same cost to all eigenstates and higher cost for other states. A natural choice is energy variance, which is zero only for eigenstates and positive for others. By minimizing energy variance, one can then find the eigenstates of a Hamiltonian. We call these two methods the minimized energy variance-VQE and energy-VQE,respectively.

    Let us formulate variance-VQE. The energy variance of HamiltonianHwith a wavefunctionψ(β)can be written as

    The non-negative energy variance consists ofG(β),which is a semi-positive matrix. Note that each element ofG(β)can be obtained on a quantum computer. Then, the energy variance can be evaluated based on the quantum covariance matrix.

    As a variational method,one key ingredient for variance-VQE is whether the wavefunciton ansatz has enough expressive power to parameterize the eigenstates. Unlike energy-VQE which looks for a ground state,the chance that an ansatz can parameterize some eigenstates can be greater as there are 2Meigenstates for a quantum system ofMqubits, although it is in general hard for an ansatz to parameterize all eigenstates. Notably, with an intermediate-size variational quantum circuit,it is expected that lower-lying eigenstates may be parameterized, as those states typically will be characterized with smaller entanglement entropy.[40]This can be useful as most applications are interested in lower-lying eigenstates. In addition,the variance-VQE has the self-verifying property that it can verify whether it indeed obtains an eigenstate by checking if the energy variance is zero.

    Another important aspect is optimization, namely minimizing the energy variance, which can be carried with gradient free or gradient descents. The gradient descent method requires to calculate the gradient of energy variance,which is

    SinceG(β) should be evaluated withO(N2) elements,calculating energy variance costsO(N2),which is larger than a cost ofO(N) for calculating energy (see Appendix for details as well as a demonstrative comparison using a 2-qubit Hamiltonian).As a result,direct gradient descent for variance-VQE can be more resource-costing than energy-VQE. Nevertheless, we can use Hamiltonian sampling to reduce this cost,[42]which estimates the energy variance and its gradient by sampling only a portion of the quantum covariance matrix. In Section 4, we will give a stochastic gradient descent algorithm for optimizing variance-VQE by Hamiltonian sampling,[42]which can be efficient under a minimal sampling rate and thus can reduce the overload significantly. Moreover,one may reduce the overload of evaluating the energy variance by using some smarter strategies that can simultaneously access a set of quantum covariance matrix elements,[43,44]or use the quantum power method introduced in Ref.[45].

    3. The variance-VQE for excited states

    As all eigenstates always have zero variance, optimizing the variance-VQE by minimizing energy variance may lead to any eigenstate. Thus, the variance-VQE can be used for obtaining excited states. In this section,we propose different approaches for solving excited states of molecule Hamiltonians with the variance VQE. We first show that the cost function variance-VQE can have many global minimums of zero variance,corresponding to different excited states. However,this approach can be inefficient if a specified eigenstate is required.To overcome this,we propose the variance-VQE that solves a set of excited states jointly, using a set of orthogonal wavefunction ansatzs. We further demonstrate that optimization of a linear combination of energy and variance can be more efficient than optimizing energy or variance alone.

    For demonstration, we use the hydrogen moleculeH2as an example. We would consider a more complicated system,H4, in Section 4. All Hamiltonians are calculated withOpenFermion,[46]and the numeral simulation for the variance VQE is conducted with the HiQ simulator framework.[47]Under STO-3G basis, we use four qubits to describe the Hamiltonian ofH2. A UCC ansatz is taken with trotter stepk=1(see Appendix B),and there are 9 parameters(6 for single excitations and 3 for double excitations). The reference state is chosen as|0011〉,or others in the subspace of two electrons.

    Though demonstrated with UCC ansatz for quantum chemistry problems, it should be stressed that the framework of variance-VQE can be applied for solving general quantum systems with proper chosen ansatz. The choosing of ansatz is often domain specific and can be important for efficiently applying VQE.For instance, to solve quantum lattice models one may adopt the Hamiltonian variational ansatz.[48]

    3.1. Single ansatz with multiple global minimums

    For the variance VQE,the criterion for being eigenstates is that energy variance?(β)=0. If a wavefunction ansatz can represent all eigenstates,then we can expect that there are many solutions ofβcorresponding to?(β)=0. In practice,as high energy eigenstates usually have higher complexity,an ansatz may only capture some eigenstates.In the optimization,one can select those minimums that are close to zero. This is indeed one advantage of the variance-VQE: one can verify if an eigenstate is obtained by checking whether the variance is zero.[13]A direct way to calculate excited states with the variance VQE is as follows. First, chose an ansatz|ψ(β)〉that can express considered excited states. For quantum chemistry,we can use some modified UCC ansatzs with enough singleparticle and double-particle excitations (see Appendix A).Second, minimizing?(β) with random choice of initial parametersβ0∈[0,2π]K. Third,select minimums of?(β)that are close to zero. Lastly, with optimized parameters, we can calculate their corresponding energies.

    Fig. 1. Distribution of global minimums in the parameter space β,where the high dimension β is visualized on a plane by dimensional reduction,using multi-dimensional scaling method. E0,E1,E2,and E3 correspond to energy levels in ascending order.

    As for demonstration, we calculate the spectrum ofH2.With uniformly sampled initialβ,we optimize the energy variance with default optimizer inScipy. All obtained minimums are almost zero (less than 10?8), and their corresponding energies are eigen-energies successfully. However, we find the chance to be a given eigenstate varies,e.g.,solutions to ground states are far less than excited states. To illustrate this, we visualize the distribution of solutions in the parameter space,as seen in Fig.1, by projecting the 9-dimensional parameters onto a plane,using a multidimensional scaling method that can preserve well information of distance. Note that we use a distance metricd(β,β′)=∑icos(θi ?θ′i) asθis an angle such thatθ1=θ2mod 2π. This indicates that the ground state is relatively hard to find in this method. It is also observed that multiple points of parameters may correspond to excited states with the same energy. For non-degenerate eigenstates, corresponding toE0,E2,E3respectively, it can be explained that differentU(β)can transform|R〉to the same target state. As excited states forE1are three-fold degenerate(will be revealed later),there are many solutions ofU(β)that transforms|R〉to this subspace. We remark that the above phenomena hold for other reference states,e.g.,|0101〉and|0110〉.

    3.2. A set of orthogonal ansatzs

    While variance-VQE can be optimized to obtain all eigenstates for a single ansatz with different optimized parameters,it may be hard to get the ground state or other specified states as it requires choosing a proper initial parameter. Here we develop another method,using a set of orthogonal ansatz wavefunctions.

    The orthogonality of ansatzs for different eigenstates can be easily enforced by using|ψn(β)〉=U(β)|Rn〉, where{|Rn〉}are orthogonal to each other. Note that all ansatz uses the sameU(β)and the orthogonality of{|ψn(β)〉}only holds under the sameβ. To optimizeβ,we use a cost function that is equal-weighting summation of all energy variances, which can be written as

    where?n(β)is the energy variance ofHfor the state|ψn(β)〉,andwn= 1/k. The formula is similar to that of subspacesearch VQE, which requires a specified weighting{wn}that enforces an ordering of eigenstates. We note that Eq. (7) has been used in Ref.[35]to calculate all eigenstates of quantum many-body systems with unitary matrix product state ansatz.

    While other methods usually calculate excited states one by one, this method incorporates all into a single cost function to optimize. In spirit, it can be viewed as a multi-task learning.[49]As those tasks are closely related, solving them in a package may benefit each other.

    Solving excited states then can be carried out by optimizingCvar(β)with gradient descent,

    Other gradient-free methods are also applicable. We present the numeral results forH2. Reference states are computational basis on a subspace of two occupied electrons, which has 6 states. As the UCC ansatz is particle-conserving, the wavefunction ansatz is in the subspace of 2 particles. The energy potential curve fits perfectly to that by exact diagonalization,as seen in Fig. 2. It is interesting to note that there are only four energy levels,while the subspace is 6-dimensional. This can be explained that the first excited state is three-fold degenerate.

    Fig. 2. The spectrum of H2 with different bond lengths. Blue dash lines are results from exact diagonalization in the whole Hilbert space.Red markers are obtained from the variance-VQE in a subspace of two electrons that H2 is electronic neutral.

    Moreover, evaluation of the cost function Eq. (7) can be more efficient with ancillary qubits, which is given in Appendix C. This is useful when lots of eigenstates are needed,where a direct summation can be impractical. For instance,there is an exponentially large number of terms when all eigenstates are needed.

    It is interesting to compare the result of the variance-VQE with orthogonal ansatz to that of subspace-search VQE.[29]The subspace-search VQE focuses on a subspace of eigenstates function and optimizes a number ofklow-energy eigenstates simultaneously. The cost function is a weighted combination of all energies,

    where|ψn(β)〉=U(β)|n〉. The weightings satisfyw0>w1>w2>···>wk?1, such that energies at optimizedβ?will satisfy?0(β?)≤?1(β?)≤···≤?k?1(β?) (see the proof in Ref.[29]). It is noted that the weightings can be learned,as in Ref. [20], which self adjusts the weightings when variational preparing Gibbs state for a quantum system.

    One requirement for the subspace-search VQE is that the order of eigenstates needs to be given. It is possible that the optimization may go a long way to make all eigenstates rest in a given order, as revealed in Fig. 3. This can increase the complexity of the optimization. To illuminate this, we compare the evolution of energies in the optimization process for both subspace-search VQE and variance-VQE. The ordering of eigenstates in the subspace-search VQE is fixed by reference states|0011〉,|0101〉,|1001〉,|0110〉,|1010〉,|1100〉,which give to ascending order of energies after optimization.As seen in Fig.3,initialized with the same parameters and thus the same energies, evolutions of energies for subspace-VQE will have complicated trajectories toward convergence.In contrast, all orthogonal ansatz wavefunctions flow directly to the nearest eigenstates in the variance-VQE.We also check other ordering of eigenstates for the subspace-search VQE,and find the same result.

    Fig.3. Evolution of energy levels in the process of optimization. Two methods are compared: the subspace-search VQE and the variance-VQE.The bond length for H2 is λ =1.

    3.3. Minimizing a combination of energy and variance

    For the orthogonal ansatz, if we only choose a subset of the Hilbert space or subspace with a fixed number, then the variance-VQE can not promise that the solutions are lowest energy states. Just as the previous approach uses one ansatz,the final result depends on the initial parameter and the optimization process. On the other hand,for a cost function with an equal-weighting energies for orthogonal ansatz, it can get the lowest total energy, but can not promise each optimized state to be an eigenstate. While subspace-search VQE can solve this problem, the optimization can not be efficient by assigning an order for ansatzs as eigenstates.Here,we demonstrate that a simple combination of energies and energy variances can make the best of both to find low energy excited states.[33]The cost function as a combination of energies and energy variances can be written as

    In this regard, the optimization includes gradient information of both energy and energy variance,andηvplays a role in adjusting the step size between them. In Fig. 4, we show the result of mixed VQE with differentηv,and it can be seen that nonzeroηvis critical to get the right eigenstates,and increasingηvcan raise the efficiency for optimization. Moreover, if only energy variance is used in the cost function,the solution may not be the lowest energy eigenstates. Thus, the result demonstrates that a combination of energy and energy variance can be useful and efficient for VQE to solve low-energy excited states.

    Fig. 4. Optimization processes for mixed cost function with different mixing factor ηv of energy variances. Three orthogonal ansatzs are used. The bond length for H2 is λ =0.8.

    4. Hamiltonian sampling

    So far,we have demonstrated that the variance-VQE can efficiently solve excited states for a Hamiltonian. However,the overload of calculating variance and its gradient descent is massive since it scales with the number of Hamiltonian terms asO(N2). In this section, we propose stochastic gradient descent for the variance-VQE by Hamiltonian sampling, which can significantly reduce the overload.

    The Hamiltonian sampling randomly choose some components ofc(setci=0 ifciis not chosen),which is denoted as ?c. Then,the estimated energy variance is A prefactor|c|2/?c2is added to account for the fact that coefficients for different terms of a Hamiltonian vary largely, so there should be a term for normalization. Whenever ?ci=0,elementsG(β)i jdo not need to be evaluated. For a sampling rates(defined as a ratio thatciis sampled),the number of elements inG(β)ijrequired to be calculated can be reduced to a factors2.

    With a Hamiltonian sampling, the gradient can be estimated as???(β)/?β, which shall have a distribution due to sampling. With this gradient, a stochastic gradient descent algorithm can be applied for optimizing the variance-VQE,which updates parameters as

    We apply the stochastic gradient descent for solving eigenstates ofH2andH4with the variance-VQE. The moleculeH4is investigated in a trapezoidal structure. Under sto-6g basis,we use a Hamiltonian of 6 qubits to describeH4.The UCC ansatz is chosen ask=1 with 21 parameters,including 15 for single-particle excitations and 6 for double-particle excitations(we only consider double-particle excitations from occupied orbitals to empty orbitals). Note that there are more than one hundred terms for the Hamiltonian,which means that Hamiltonian sampling is badly demanded to evaluate the variance.

    Numeral simulation results are displayed in Fig. 5 for solving two eigenstates with orthogonal ansatzs. ForH2, it shows that Hamiltonian sampling with a small sampling rates=0.1,0.2 can have comparable convergent behavior with the case without sampling. However, the variance ceases to converge at small values, due to the fluctuation of energy for a subsystem even when the whole system has zero energy variance. To solve this issue, we can turn to no sampling (sampling rates=1) in the later stage of optimization. ForH4,similar phenomena can be observed. It is noted that only one ansatz reaches the true eigenstate, while the other fail (a local minimum or the ansatz lacks the capacity to express this eigenstate), which can be seen from the nonzero energy variance.This again verifies that the variance-VQE can self-verify whether an eigenstate is obtained.

    Fig.5. Optimization of the variance-VQE for H2 (upper row)and H4 (down row)with different Hamiltonian sampling rates s. Sampling rates can be adjustable in an optimization process,for instance,s=0.1 and 1 means that s changes from 0.1 to 1 in the later stage of optimization.The learning rate is 0.05 for all cases.

    We can explain why Hamiltonian sampling fails to work around the optimized point of parameters,where gradients by Hamiltonian sampling can give meaningless direction for optimization. This is because energy variance is not additive: ifH=HA+HBand the wavefunction is|ψ〉, then the variance ofHis not a summation of the variances ofHAandHB. For example,if|ψ〉is an eigenstate ofH,|ψ〉will not be an eigenstate ofHAorHB,if A and B are entangled.As a result,even at the optimized parameter of zero variance,gradients by Hamiltonian sampling can be finite. Such a property is in sharp contrast to the case of machine learning.Machine learning usually uses a set of independent and identically distributed samples,and the cost function is a summation of independent contributions from all samples. Thus,gradients can be estimated with a batch of samples, giving stochastic gradient descents with well convergence behavior.[50]

    From the above discussion, a practical strategy is to use Hamiltonian sampling with a small sampling rate to optimize the variance to a small value, then turn to a large sampling rate or no sampling. Also,one may then use energy-VQE.We note that stochastic gradient descent with few shots of measurements can further reduce the overload,[42,51]which leaves for further investigation.

    5. Summary

    In summary,we have proposed the variance-VQE that optimizes the energy variance to obtain eigenstates for a Hamiltonian. Compared with the VQE that minimizes the energy which is prone to find ground state, the variance-VQE naturally is suitable for solving and self-verifying arbitrary eigenstates. We have adopted different strategies to solve excited states with the variance-VQE, using moleculesH2andH4as examples. Remarkably, it has been shown that optimizing a set of orthogonal ansatzs with the same parameterized circuit can be very efficient for calculating a set of eigenstates. Also,we have demonstrated that minimizing a linear combination of energy and variance can outperform the case of optimizing either energy or variance alone. Moreover, we have proposed stochastic gradient descent for minimizing the variance by Hamiltonian sampling, using only a few terms of the Hamiltonian to estimate the variance and its gradient. It has been shown that the variance can decrease quickly under a small sampling rate. While numeral simulations suggested that the variance fails to converge, Hamiltonian sampling can be useful in the early stage to locate the zone of optimized parameters. Our work has demonstrated that the variance-VQE can be useful and practical for variational solving excited states in a quantum system on a quantum computer.

    Appendix A: Comparison between measurements of energy and variance

    For a HamiltonianH= ∑Ni=1ciLi, whereLiis a Pauli string, it is known that measurement of the energy with regard to a wavefunction|ψ〉can be carried out by measuring everyLiseparately and then making a linear combination of results ofNPauli measurements,namely,The capacity of decomposing the energy measurement to simple Pauli measurements is key for implementing variational quantum eigensolver.

    The situation is similar for measuring the energy variance on the quantum computer. One can make a decomposition for the first term in Eq.(2),

    Appendix B:UCC ansatz

    The UCC ansatz is widely used in the field of quantum chemistry,as it can represent a parameterized wavefunction of electronic structure of a molecule efficiently. The parameterized wavefunction can be written as Here,T=T1+T2usually includes single-particle excitationsT1and double-particle excitationsT2. We adopt a notation that does not distinguish occupied and empty states in the reference state|R〉(which typically is a Hartree–Fock state). ThenT1andT2can be expressed as

    Appendix C:Efficient evaluation of cost function

    While the goal is to solve a number ofkexcited states,evaluation of the cost function (the total energy variance) is time-consuming for largek, since it should calculate energy variance one by one. For instance, for a system ofMqubits,the total number of eigenstates is 2M. Here, we discuss how to efficiently evaluateCvar(β),similar to Ref.[35]. Let us first considerk=2M. Using then it can be verified that

    To evaluate the second part of Eq. (7), the summation in Eq. (C4) should be replaced byk ?1. The total number of qubits isM+M+W=2M+W.

    Acknowledgements

    This work was supported by the National Natural Science Foundation of China (Grant No. 12005065) and the Guangdong Basic and Applied Basic Research Fund (Grant No.2021A1515010317).

    猜你喜歡
    陳彬
    Microwave preionization and electron cyclotron resonance plasma current startup in the EXL-50 spherical tokamak
    努力做一件事的樂趣
    努力做一件事的樂趣
    Controllable four-wave mixing response in a dual-cavity hybrid optomechanical system*
    陳彬:以孝賢之心,傾情守護(hù)別樣“家人”
    杭州(2020年23期)2021-01-11 01:04:36
    十幾年如一日守護(hù)兩位“家人”
    學(xué)生小高組特等獎作品
    中國篆刻(2019年10期)2019-10-28 07:06:54
    回報(bào)
    愛到最后
    愛到最后
    短篇小說(2016年8期)2016-09-20 01:20:17
    99久久人妻综合| 精品第一国产精品| 如日韩欧美国产精品一区二区三区| 久久女婷五月综合色啪小说| 亚洲国产精品一区二区三区在线| 日本色播在线视频| 精品国产乱码久久久久久小说| 亚洲精品日韩在线中文字幕| 香蕉国产在线看| 一本色道久久久久久精品综合| 精品国产一区二区久久| 亚洲国产av新网站| 天美传媒精品一区二区| 午夜日韩欧美国产| 国产精品99久久99久久久不卡 | 精品国产超薄肉色丝袜足j| 999久久久国产精品视频| 丝袜喷水一区| 18禁国产床啪视频网站| 大话2 男鬼变身卡| 中文字幕人妻熟女乱码| 国产一区亚洲一区在线观看| 国产99久久九九免费精品| 人妻人人澡人人爽人人| 日本爱情动作片www.在线观看| 在线亚洲精品国产二区图片欧美| 国产精品.久久久| 水蜜桃什么品种好| 自拍欧美九色日韩亚洲蝌蚪91| 久久久欧美国产精品| xxxhd国产人妻xxx| 精品国产一区二区久久| 欧美最新免费一区二区三区| 久久鲁丝午夜福利片| av视频免费观看在线观看| 国产亚洲一区二区精品| 久久久久人妻精品一区果冻| av又黄又爽大尺度在线免费看| 视频在线观看一区二区三区| 亚洲一码二码三码区别大吗| 美女高潮到喷水免费观看| 国产一卡二卡三卡精品 | 这个男人来自地球电影免费观看 | 国产熟女欧美一区二区| 18在线观看网站| 国产精品秋霞免费鲁丝片| 国产成人啪精品午夜网站| 日日摸夜夜添夜夜爱| 亚洲欧美中文字幕日韩二区| 亚洲图色成人| 精品少妇黑人巨大在线播放| 十八禁人妻一区二区| 日本一区二区免费在线视频| 亚洲国产精品一区三区| 曰老女人黄片| 亚洲欧美色中文字幕在线| 亚洲男人天堂网一区| 国产欧美亚洲国产| 色精品久久人妻99蜜桃| 成年美女黄网站色视频大全免费| 国产麻豆69| 日韩中文字幕欧美一区二区 | 久久精品国产综合久久久| 国产高清不卡午夜福利| 欧美成人午夜精品| 嫩草影院入口| 男男h啪啪无遮挡| 免费观看av网站的网址| a级毛片在线看网站| 日韩精品有码人妻一区| 欧美日韩亚洲国产一区二区在线观看 | 99香蕉大伊视频| av一本久久久久| 我要看黄色一级片免费的| 午夜福利视频在线观看免费| 成人亚洲欧美一区二区av| 只有这里有精品99| 国产精品人妻久久久影院| 999久久久国产精品视频| 久久国产亚洲av麻豆专区| 王馨瑶露胸无遮挡在线观看| 亚洲欧美中文字幕日韩二区| av线在线观看网站| 欧美变态另类bdsm刘玥| 十八禁人妻一区二区| 亚洲美女黄色视频免费看| 亚洲成国产人片在线观看| 成人影院久久| 人妻人人澡人人爽人人| 久久久久久免费高清国产稀缺| 超碰97精品在线观看| 精品亚洲成国产av| 久久精品国产亚洲av涩爱| a级毛片在线看网站| 操美女的视频在线观看| 亚洲欧美中文字幕日韩二区| 99国产精品免费福利视频| 亚洲天堂av无毛| 看免费av毛片| 男女午夜视频在线观看| 国产女主播在线喷水免费视频网站| 色吧在线观看| 国产精品99久久99久久久不卡 | bbb黄色大片| 嫩草影院入口| 久久婷婷青草| 欧美日韩成人在线一区二区| 午夜福利视频精品| 亚洲欧美色中文字幕在线| 久久99一区二区三区| 欧美中文综合在线视频| 日韩中文字幕欧美一区二区 | 久久久亚洲精品成人影院| 国产精品三级大全| 欧美日韩亚洲高清精品| 亚洲欧美一区二区三区黑人| 亚洲欧洲精品一区二区精品久久久 | 亚洲精品第二区| 国产爽快片一区二区三区| 午夜福利免费观看在线| 91aial.com中文字幕在线观看| 精品国产一区二区久久| 97人妻天天添夜夜摸| 另类亚洲欧美激情| bbb黄色大片| netflix在线观看网站| 日韩制服丝袜自拍偷拍| 国产女主播在线喷水免费视频网站| 精品久久蜜臀av无| 一二三四在线观看免费中文在| 日韩欧美一区视频在线观看| 一区二区三区精品91| 精品少妇久久久久久888优播| 免费看av在线观看网站| 啦啦啦 在线观看视频| 少妇人妻久久综合中文| 97精品久久久久久久久久精品| 国产午夜精品一二区理论片| 国产国语露脸激情在线看| 天天躁日日躁夜夜躁夜夜| 巨乳人妻的诱惑在线观看| 久久久亚洲精品成人影院| 欧美日韩视频高清一区二区三区二| 免费日韩欧美在线观看| 丝袜脚勾引网站| 国产不卡av网站在线观看| 欧美日韩视频精品一区| 高清欧美精品videossex| 一区福利在线观看| 精品国产乱码久久久久久小说| 毛片一级片免费看久久久久| 秋霞伦理黄片| bbb黄色大片| 亚洲精品国产av成人精品| 天天影视国产精品| 观看av在线不卡| 国产在视频线精品| 18禁国产床啪视频网站| 日韩大片免费观看网站| 777久久人妻少妇嫩草av网站| 国产一区有黄有色的免费视频| 啦啦啦 在线观看视频| 电影成人av| 国产精品熟女久久久久浪| 啦啦啦中文免费视频观看日本| 91精品伊人久久大香线蕉| 韩国精品一区二区三区| 午夜福利网站1000一区二区三区| 伦理电影免费视频| 99久久人妻综合| 高清黄色对白视频在线免费看| 美女扒开内裤让男人捅视频| 亚洲精品国产av蜜桃| 欧美日韩一区二区视频在线观看视频在线| xxx大片免费视频| av国产久精品久网站免费入址| 精品免费久久久久久久清纯 | 精品免费久久久久久久清纯 | 狠狠婷婷综合久久久久久88av| 桃花免费在线播放| 日韩一区二区三区影片| 欧美激情高清一区二区三区 | 中文字幕人妻丝袜一区二区 | 亚洲精品日韩在线中文字幕| 久久综合国产亚洲精品| 精品亚洲成国产av| 一区在线观看完整版| 中文字幕av电影在线播放| 男女午夜视频在线观看| 久久性视频一级片| 五月天丁香电影| 中文字幕人妻丝袜一区二区 | 18禁观看日本| 久久久久久久久久久久大奶| 日韩欧美精品免费久久| 蜜桃在线观看..| 你懂的网址亚洲精品在线观看| 国产又爽黄色视频| 一级片'在线观看视频| 成人漫画全彩无遮挡| 中文乱码字字幕精品一区二区三区| 国产亚洲av片在线观看秒播厂| 人成视频在线观看免费观看| 高清不卡的av网站| 另类亚洲欧美激情| 久久久久国产一级毛片高清牌| 久久久久国产精品人妻一区二区| 亚洲国产毛片av蜜桃av| 国产欧美日韩综合在线一区二区| 亚洲四区av| 热99久久久久精品小说推荐| 日本wwww免费看| 曰老女人黄片| 99re6热这里在线精品视频| 狂野欧美激情性xxxx| 一边摸一边做爽爽视频免费| 男女免费视频国产| 亚洲精品久久午夜乱码| 另类亚洲欧美激情| 中国三级夫妇交换| 久久久久久免费高清国产稀缺| 成人免费观看视频高清| 91老司机精品| 久久久精品免费免费高清| 少妇人妻 视频| 少妇 在线观看| 久久国产精品大桥未久av| 亚洲欧洲精品一区二区精品久久久 | 国产爽快片一区二区三区| av网站免费在线观看视频| 啦啦啦在线免费观看视频4| 国产色婷婷99| 精品久久久久久电影网| 黄网站色视频无遮挡免费观看| 90打野战视频偷拍视频| 亚洲情色 制服丝袜| 午夜影院在线不卡| 精品一区二区免费观看| 国产成人欧美| 午夜日本视频在线| 欧美 亚洲 国产 日韩一| 久久av网站| 99热网站在线观看| 丝袜在线中文字幕| 一级片免费观看大全| 精品卡一卡二卡四卡免费| 亚洲精华国产精华液的使用体验| 亚洲一级一片aⅴ在线观看| 亚洲熟女毛片儿| 国产伦人伦偷精品视频| 毛片一级片免费看久久久久| 精品少妇黑人巨大在线播放| 中国国产av一级| 国产精品久久久av美女十八| 午夜av观看不卡| 成年人午夜在线观看视频| 国产在线视频一区二区| 免费久久久久久久精品成人欧美视频| 亚洲成人一二三区av| 叶爱在线成人免费视频播放| 啦啦啦中文免费视频观看日本| 乱人伦中国视频| 日韩av不卡免费在线播放| 丝袜美足系列| 岛国毛片在线播放| 91老司机精品| 久久毛片免费看一区二区三区| 亚洲天堂av无毛| 777久久人妻少妇嫩草av网站| 国产成人精品福利久久| 青青草视频在线视频观看| 天天躁狠狠躁夜夜躁狠狠躁| 精品一区二区三区四区五区乱码 | 亚洲国产精品成人久久小说| 亚洲男人天堂网一区| 高清av免费在线| 精品亚洲乱码少妇综合久久| 91国产中文字幕| 美女国产高潮福利片在线看| 肉色欧美久久久久久久蜜桃| 大片免费播放器 马上看| 久久亚洲国产成人精品v| 可以免费在线观看a视频的电影网站 | 亚洲欧美一区二区三区黑人| 欧美日韩国产mv在线观看视频| 久久毛片免费看一区二区三区| 极品人妻少妇av视频| 亚洲精品中文字幕在线视频| 久久精品亚洲av国产电影网| 亚洲精品第二区| 制服丝袜香蕉在线| 我要看黄色一级片免费的| 女人久久www免费人成看片| 国产精品女同一区二区软件| 午夜免费男女啪啪视频观看| 免费久久久久久久精品成人欧美视频| 午夜福利视频在线观看免费| 国产成人91sexporn| 另类精品久久| 十八禁人妻一区二区| 成年人午夜在线观看视频| 国产成人精品福利久久| 成年女人毛片免费观看观看9 | 亚洲天堂av无毛| 色婷婷av一区二区三区视频| 老司机靠b影院| 欧美在线一区亚洲| 一边摸一边抽搐一进一出视频| 国产欧美亚洲国产| av在线播放精品| 国产午夜精品一二区理论片| 黑人猛操日本美女一级片| 日本欧美视频一区| 国产成人免费观看mmmm| 日日啪夜夜爽| 一区在线观看完整版| 亚洲三区欧美一区| 免费高清在线观看视频在线观看| 王馨瑶露胸无遮挡在线观看| 看免费成人av毛片| 成人漫画全彩无遮挡| 国产又爽黄色视频| 狂野欧美激情性xxxx| 精品人妻一区二区三区麻豆| 色婷婷av一区二区三区视频| 女人高潮潮喷娇喘18禁视频| 极品人妻少妇av视频| 午夜精品国产一区二区电影| 欧美人与性动交α欧美精品济南到| 日日撸夜夜添| 国产精品久久久久久人妻精品电影 | 9热在线视频观看99| 午夜免费观看性视频| 麻豆av在线久日| 亚洲色图综合在线观看| 欧美日韩av久久| 精品国产一区二区三区四区第35| 成年人免费黄色播放视频| 操美女的视频在线观看| 一边摸一边抽搐一进一出视频| 亚洲国产看品久久| 国产精品一区二区在线观看99| 亚洲国产av影院在线观看| 成年人午夜在线观看视频| 日本黄色日本黄色录像| 18禁裸乳无遮挡动漫免费视频| 中文乱码字字幕精品一区二区三区| 蜜桃在线观看..| 国产一区二区在线观看av| 老司机在亚洲福利影院| 看免费成人av毛片| 色综合欧美亚洲国产小说| 热re99久久精品国产66热6| 91精品三级在线观看| 色吧在线观看| 亚洲五月色婷婷综合| 尾随美女入室| 777米奇影视久久| 蜜桃国产av成人99| av有码第一页| 国产亚洲午夜精品一区二区久久| 麻豆乱淫一区二区| 国产av一区二区精品久久| 视频区图区小说| 亚洲欧美一区二区三区国产| 不卡视频在线观看欧美| 日本猛色少妇xxxxx猛交久久| 一本大道久久a久久精品| 99国产精品免费福利视频| 成年女人毛片免费观看观看9 | 亚洲精品日韩在线中文字幕| 黄片播放在线免费| 亚洲av在线观看美女高潮| 久久精品国产亚洲av高清一级| 一区二区av电影网| av女优亚洲男人天堂| 五月开心婷婷网| 国产亚洲午夜精品一区二区久久| 亚洲精品国产区一区二| 最近最新中文字幕大全免费视频 | 少妇人妻久久综合中文| 黄网站色视频无遮挡免费观看| 看十八女毛片水多多多| www.熟女人妻精品国产| 中文字幕亚洲精品专区| 男的添女的下面高潮视频| 国产成人免费无遮挡视频| 午夜影院在线不卡| 久久99精品国语久久久| 欧美成人午夜精品| 老熟女久久久| 亚洲欧洲精品一区二区精品久久久 | 丝袜人妻中文字幕| 丝袜美足系列| 热re99久久国产66热| 国精品久久久久久国模美| 亚洲精品国产区一区二| 欧美激情高清一区二区三区 | 男人操女人黄网站| 国产亚洲最大av| 中文字幕制服av| 香蕉丝袜av| 亚洲色图综合在线观看| 亚洲,欧美,日韩| 久久99热这里只频精品6学生| 欧美日韩亚洲高清精品| 国产免费福利视频在线观看| 亚洲,一卡二卡三卡| 在线精品无人区一区二区三| 久久午夜综合久久蜜桃| av视频免费观看在线观看| 又粗又硬又长又爽又黄的视频| 国产一级毛片在线| 岛国毛片在线播放| 亚洲情色 制服丝袜| 久久精品aⅴ一区二区三区四区| 一区二区三区精品91| 黑人猛操日本美女一级片| 中文字幕亚洲精品专区| 亚洲av中文av极速乱| 在线 av 中文字幕| 午夜福利在线免费观看网站| 午夜免费鲁丝| 日韩一区二区视频免费看| 性少妇av在线| 久久人人97超碰香蕉20202| 亚洲一区二区三区欧美精品| 婷婷色综合www| 丝袜美腿诱惑在线| 亚洲,一卡二卡三卡| 欧美老熟妇乱子伦牲交| 国产精品三级大全| 伊人久久国产一区二区| 99热国产这里只有精品6| 亚洲三区欧美一区| 久久人妻熟女aⅴ| 亚洲成人手机| 久久女婷五月综合色啪小说| 国产精品二区激情视频| 亚洲激情五月婷婷啪啪| 国产精品嫩草影院av在线观看| 在线天堂中文资源库| 777米奇影视久久| 亚洲伊人色综图| 别揉我奶头~嗯~啊~动态视频 | 黑丝袜美女国产一区| 青春草亚洲视频在线观看| tube8黄色片| 婷婷色综合大香蕉| 成年人午夜在线观看视频| 最近中文字幕2019免费版| 久久精品久久精品一区二区三区| 欧美日韩亚洲高清精品| 女人高潮潮喷娇喘18禁视频| 观看av在线不卡| 建设人人有责人人尽责人人享有的| 大陆偷拍与自拍| 日本爱情动作片www.在线观看| 男人添女人高潮全过程视频| 99热网站在线观看| 中文字幕制服av| 制服诱惑二区| 亚洲精品成人av观看孕妇| 男女免费视频国产| 欧美日韩国产mv在线观看视频| 另类亚洲欧美激情| 一边亲一边摸免费视频| 日本91视频免费播放| 色综合欧美亚洲国产小说| 免费看不卡的av| 国产av国产精品国产| 欧美日韩亚洲综合一区二区三区_| 天天躁日日躁夜夜躁夜夜| 欧美在线一区亚洲| 亚洲成人一二三区av| 日本wwww免费看| 国产在线一区二区三区精| 男人操女人黄网站| 国产日韩欧美视频二区| 亚洲四区av| 国产在线视频一区二区| 欧美国产精品一级二级三级| 亚洲少妇的诱惑av| 亚洲一区二区三区欧美精品| 精品少妇一区二区三区视频日本电影 | 精品少妇久久久久久888优播| av在线app专区| 国产国语露脸激情在线看| 激情视频va一区二区三区| 美女大奶头黄色视频| 老司机影院成人| 国产成人精品福利久久| 黄片播放在线免费| 亚洲,欧美精品.| 青春草亚洲视频在线观看| 亚洲视频免费观看视频| svipshipincom国产片| 欧美在线黄色| 免费久久久久久久精品成人欧美视频| 97人妻天天添夜夜摸| 久久天躁狠狠躁夜夜2o2o | 色婷婷久久久亚洲欧美| 成年美女黄网站色视频大全免费| 日韩欧美精品免费久久| 国产在线免费精品| 午夜福利视频精品| 老司机影院毛片| 卡戴珊不雅视频在线播放| av有码第一页| 美女福利国产在线| 亚洲精品美女久久av网站| 久久久久精品国产欧美久久久 | 成人18禁高潮啪啪吃奶动态图| 亚洲精品美女久久久久99蜜臀 | 亚洲av综合色区一区| 别揉我奶头~嗯~啊~动态视频 | a级毛片黄视频| 视频在线观看一区二区三区| 纯流量卡能插随身wifi吗| 可以免费在线观看a视频的电影网站 | 大香蕉久久成人网| 亚洲少妇的诱惑av| 亚洲欧洲精品一区二区精品久久久 | 一级片免费观看大全| 国产精品久久久久久精品古装| 别揉我奶头~嗯~啊~动态视频 | 国产亚洲av高清不卡| 久久精品久久久久久噜噜老黄| 亚洲av在线观看美女高潮| 免费av中文字幕在线| 伦理电影大哥的女人| 久久久久网色| av免费观看日本| 午夜影院在线不卡| 日本爱情动作片www.在线观看| 亚洲成人av在线免费| 九色亚洲精品在线播放| 免费在线观看黄色视频的| 国产又爽黄色视频| 看免费av毛片| 国产女主播在线喷水免费视频网站| 亚洲五月色婷婷综合| 国产极品粉嫩免费观看在线| 日韩欧美一区视频在线观看| 少妇猛男粗大的猛烈进出视频| 成人亚洲精品一区在线观看| a 毛片基地| 午夜av观看不卡| 亚洲少妇的诱惑av| 久久午夜综合久久蜜桃| 国产视频首页在线观看| 久久国产精品大桥未久av| 九色亚洲精品在线播放| 精品一区二区三区四区五区乱码 | 十八禁高潮呻吟视频| 免费不卡黄色视频| 日韩电影二区| 久久青草综合色| 99国产综合亚洲精品| 亚洲欧美日韩另类电影网站| 亚洲美女视频黄频| 国产精品久久久人人做人人爽| 激情五月婷婷亚洲| 美女高潮到喷水免费观看| 欧美精品亚洲一区二区| 日韩伦理黄色片| 日韩电影二区| 热re99久久国产66热| 午夜日本视频在线| 久久久久久久国产电影| 亚洲综合色网址| 久久亚洲国产成人精品v| 天天躁日日躁夜夜躁夜夜| 国产精品成人在线| av在线老鸭窝| 成人影院久久| 亚洲中文av在线| 亚洲欧美精品综合一区二区三区| 免费人妻精品一区二区三区视频| 高清在线视频一区二区三区| 国产精品欧美亚洲77777| 夫妻性生交免费视频一级片| 久久人人爽人人片av| 久久久久精品久久久久真实原创| 免费黄频网站在线观看国产| 久久久久久久久免费视频了| av.在线天堂| 国产精品久久久久久精品古装| 亚洲精品乱久久久久久| 不卡视频在线观看欧美| 国产 精品1| 在线观看免费午夜福利视频| 日韩一区二区视频免费看| 99九九在线精品视频| 青草久久国产| 美女中出高潮动态图| 久久 成人 亚洲| 成年美女黄网站色视频大全免费| 日韩 亚洲 欧美在线| 菩萨蛮人人尽说江南好唐韦庄| 国产色婷婷99| 久久99热这里只频精品6学生| 国产免费福利视频在线观看| 国产精品久久久久久精品古装| 国产精品人妻久久久影院| 在线天堂中文资源库| 在线免费观看不下载黄p国产| 欧美日韩一级在线毛片| 日韩欧美一区视频在线观看| 国语对白做爰xxxⅹ性视频网站| 久久久久久久久久久久大奶| av线在线观看网站| 久久热在线av| 久久99一区二区三区| 乱人伦中国视频|