A Probabilistic Imaginary Time Evolution Algorithm Based on Non-unitary Quantum Circuit Hao-Nan Xie1Shi-Jie Wei2yFan Yang1 Zheng-An Wang2 Chi-Tong Chen34 Heng Fan32 and Gui-Lu Long12z

2025-04-27 0 0 3.91MB 13 页 10玖币
侵权投诉
A Probabilistic Imaginary Time Evolution Algorithm Based on Non-unitary Quantum
Circuit
Hao-Nan Xie1,Shi-Jie Wei2,Fan Yang1, Zheng-An Wang2, Chi-Tong Chen3,4, Heng Fan3,2, and Gui-Lu Long1,2
1Department of Physics, Tsinghua University, Beijing 100084, China
2Beijing Academy of Quantum Information Sciences, Beijing 100193, China
3Institute of Physics, Chinese Academy of Sciences, Beijing 100190, China and
4School of Physical Sciences, University of Chinese Academy of Sciences, Beijing 100190, China
(Dated: October 12, 2022)
Imaginary time evolution is a powerful tool applied in quantum physics, while existing classical
algorithms for simulating imaginary time evolution suffer high computational complexity as the
quantum systems become larger and more complex. In this work, we propose a probabilistic al-
gorithm for implementing imaginary time evolution based on non-unitary quantum circuit. We
demonstrate the feasibility of this method by solving the ground state energy of several quantum
many-body systems, including H2, LiH molecules and the quantum Ising chain. Moreover, we
perform experiments on superconducting and trapped ion cloud platforms respectively to find the
ground state energy of H2and its most stable molecular structure. We also analyze the successful
probability of the algorithm, which is a polynomial of the output error and introduce an approach
to increase the success probability by rearranging the terms of Hamiltonian.
I. INTRODUCTION
Imaginary time evolution (ITE), as a mathematical
tool, has impacted many problems in quantum physics,
such as solving the ground state of a Hamiltonian [1–
3], studying finite temperature properties [4–6] and the
quantum simulation of non-Hermitian systems [7, 8].
The concept of ITE can be understood by defining the
imaginary time β=it and substituting it into the
Schr¨odinger’s equation, i∂t|Φi=H|Φi, where His a
Hermitian Hamiltonian, which gives us the imaginary-
time Schr¨odinger’s equation:
β|Φβi=H|Φβi.(1)
Given the initial state |Φ0i, the solution of Eq. 1 is
|Φβi=AeβH|Φ0i, where the corresponding evolution
operator eβHis non-unitary, and Ais the normaliza-
tion constant. In classical simulations, one can directly
calculate eβHand apply it to the initial state vector
|Φ0i, or employ some classical techniques such as quan-
tum Monte Carlo [9] and tensor networks [10]. However,
the dimension of the Hilbert space grows exponentially
with the size of the quantum system, making the tasks
intractable for classical computers [11].
Quantum computer is one of the promising tools for
efficiently simulating quantum systems [11–19]. For the
real time simulation, the evolution operator eitHcan
be realized directly or be simply decomposed into a se-
quence of unitary quantum gates; but it is not the case
for imaginary time simulation, where the evolution oper-
ator eβHis non-unitary. Therefore alternative methods
are required. Recently, some hybrid quantum-classical
xiehn19@mails.tsinghua.edu.cn
weisj@baqis.ac.cn
gllong@mail.tsinghua.edu.cn
algorithms for simulating ITE have been proposed. For
example, variational quantum simulation methods [20–
22] utilize variational ansatz and simulate the evolution
of quantum states with classical optimization of parame-
ters; quantum imaginary time evolution (QITE) [23, 24]
finds a unitary operator to approach the ideal ITE in each
evolution step. The main drawbacks of these methods
include systematic error due to fixed parametrization,
complexity of classical optimization [25, 26], limitation
of correlation length [3, 24], etc.
In 2004, Terashima and Ueda [27] proposed a method
to implement non-unitary quantum circuit by quantum
measurement. By applying unitary operations in the ex-
tended Hilbert space, one can obtain the desired final
state in a certain subspace of the auxiliary qubits. Non-
unitary quantum circuit has been widely used in sim-
ulating non-Hermitian dynamics [8, 28], linear combina-
tion of unitary operators (LCU) [29], full quantum eigen-
solver (FQE) [30] and other algorithms. Based on non-
unitary circuit, Ref. [31] proposes a form of non-unitary
gate which applies to two-qubit ITE process; Refs. [3, 8]
show the form of the unitary operator acting on the total
Hilbert space to implement ITE, and give the examples
of circuits for two-qubit cases.
In this work, we propose a probabilistic imaginary time
evolution (PITE) algorithm which utilizes non-unitary
quantum circuit with one auxiliary qubit. In contrast to
previous works, we explicitly illustrates the construction
of the required quantum circuits using single- and double-
qubit gates, which applies to any number of qubits and
more generic Hamiltonians. We numerically apply the
PITE algorithm to calculate the ground-state energy of
several physical systems, and perform experiments on su-
perconducting and trapped ion cloud platforms. We also
give a detailed analysis about the computational com-
plexity of this method.
This paper is organized as follows. In section II we
give a description of the PITE method. Section III shows
arXiv:2210.05293v1 [quant-ph] 11 Oct 2022
2
some experimental and numerical simulation results, and
the analysis about the error and the successful probabil-
ity. In Section IV, we discuss the generalization of the
PITE algorithm to the cases where Hamiltonian is not
composed of Pauli terms, and introduce a method to in-
crease the success probability.
II. METHOD
An n-qubit Hamiltonian, H=Pm
k=1 ckhk, is composed
of mPauli product terms, in which ckis a real coefficient
and hk=n
j=1σj
αj, where σj
αjis a Pauli matrix or the
identity acting on the j-th qubit, with αj∈ {0, x, y, z}
(here we use the notation σ0=I). We assume that
the Hamiltonian does not contain the identity term In
because it merely shift the spectrum of Hamiltonian.
Our goal is to implement the non-unitary operator
eβHin quantum circuits. We first apply the Trotter
decomposition [32, 33]
eβH=ec1h1t. . . ecmhmtβ/t+O(∆t).(2)
For a single Trotter step, we wish to obtain |Φ0i=
eckhkt|Φi. Define e
Tk=eckhkt. Due to the fact
that ckhkonly has two different eigenvalues ±|ck|, each
with the degeneracy of 2n1, there exists a unitary Uk
satisfying
UkckhkU
k=−|ck|σlk
z, lk∈ {1, . . . , n}(3)
which is a single qubit operator. Thus we have
e
Tk=U
kexp −|ck|σlk
ztUk.(4)
In Fig. 1, we show how to implement e
Tkin a quantum
circuit. The construction of Ukrequires O(n) CNOT
gates and single qubit gates at most (see Appendix B for
details). After the action of Uk, we write the work-qubit
state as Uk|Φi=a0|ψ0i+a1|ψ1i, where |ψ0iand |ψ1iare
the projection of Uk|Φion the subspace where the lk-th
qubit is |0iand |1i, respectively. In Appendix B we will
show that a0, a1are also the amplitude of the projection
of |Φion the ground-state subspace and excited-state
subspace of ckhk, respectively.
To realize e−|ck|σzton the lk-th work qubit, we add an
ancillary qubit |0i, and apply the controlled-Ryoperation
|0ih0| I+|1ih1| Ry(θk) on the lk-th work qubit and
the ancilla qubit, where
Ry(θ) = eiθσy/2=cos θ/2sin θ/2
sin θ/2 cos θ/2,
θk= 2 cos1(e2|ck|t),
(5)
which gives the state
a0|ψ0i|0ianc +a1e2|ck|t|ψ1i|0ianc
+a1p1e4|ck|t|ψ1i|1ianc .(6)
Then we measure the ancilla qubit, and if the result is 0,
we obtain
s1
|a0|2+|a1|2e4|ck|ta0|ψ0i+a1e2|ck|t|ψ1i(7)
which is equivalent to the result of e−|ck|σztacting on
the lk-th work qubit up to normalization.
The probability of obtaining 0 in the measurement of
ancilla is |a0|2+|a1|2e4|ck|t. If we success, then the
last step is to apply U
kon the work qubits. The output
state will be exactly e
Tk|Φiup to normalization.
In summary, the non-unitary operator e
Tkacting on a
quantum state |Φican be written as
e
Tk|Φi=U
k[h0|anc Rk·(Uk|Φi⊗|0ianc)] (8)
up to a constant coefficient, where Uktransforms hkinto
a Pauli matrix acting on the lk-th work qubit, and Rk
represents the controlled-Rygate acting on the lk-th work
qubit and the ancilla qubit, with the rotation angle θk=
2 cos1e2|ck|t.
n work
qubits
|Φ
1-st
lk-th
n-th
auxiliary qubit
exp(-ckσz
lkΔt)
UkUk
Ry(θk)
FIG. 1. Quantum circuit for implementing
e
Tk.
III. RESULTS
A. Calculation of H2, LiH and quantum Ising
model
To illustrate the performance of the PITE algorithm,
we apply the algorithm in the calculation of the ground-
state energy of three physical systems: H2molecules, LiH
molecules, and a quantum Ising spin chain with both
transverse and longitudinal fields. The calculations of
H2are carried out on the Quafu’s 10-qubit superconduct-
ing quantum processor and IonQ’s 10-qubit trapped ion
QPU, and the calculations of LiH and the Ising model
are carried out on a numerical simulator to study the
influence of noises and the success probability of the al-
gorithm.
3
To calculate the ground state of H2and LiH on a
quantum computer, we first need to encode the molec-
ular Hamiltonian onto qubits. Here we choose the STO-
3G basis [34] and use the Jordan-Wigner transforma-
tion (JWT) (see details in Appendix C). We eventu-
ally obtain Hamiltonians composed of Pauli matrices
H(R) = Pkck(R)σ1
α1. . . σn
αn, which can be acted on n
qubits, and the coefficients ckvary with the interatomic
distance R. In this way the Hamiltonian for H2and LiH
can be encoded onto 4 and 6 qubits, respectively. Fur-
ther mapping is applied on H2to compactly encode the
H2Hamiltonian onto 2 qubits (see details in Ref. [35]),
which gives the H2Hamiltonian in the form of
HH2=c0(R) + c1(R)σ1
z+c1(R)σ2
z
+c2(R)σ1
zσ2
z+c3(R)σ1
xσ2
x,(9)
where the coefficients at different Rare available in Ap-
pendix C. The Hamiltonian for LiH at its lowest-energy
interatomic distance (bond distance) is given explicitly
also in Appendix C.
In our experiments, we use 2 qubits as the work qubits
to represent the H2molecule, and 1 qubit as the ancilla
qubit. The Hatree-Fock state of H2is |ΦHFi=|00iin
the qubit representation, which is chosen as the initial
state of the work qubits. Following the PITE method,
we first do the calculation at a fixed interatomic dis-
tance R= 0.75˚
A, and the experiments are carried out on
Quafu’s superconducting QPU P-10 and IonQ’s trapped
ion QPU (see information about Quafu in Appendix A).
After each Trotter step, the quantum state of work qubits
is tomographed, with 2000 shots on Quafu and 1000 shots
on IonQ, and then we use the state to calculate the en-
ergy value, and set it as the initial state for the next step.
The quantum circuits and related details are shown in
Appendix D. The results of the energy expectation value
as a function of the imaginary time βare shown in Fig.
2(a), compared with the theoretical PITE results. As β
increases, the energy rapidly converges to the exact solu-
tion in 5 evolution steps, within an error of 104a.u,
which is in the chemical precision.
To obtain the most stable molecular structure, we vary
the interatomic distance and plot the potential-energy
surface for H2molecule, as shown in Fig. 2(b). The
series of experiments is only carried out on Quafu’s P-
10, and the results (β= 0.5 and β= 1) are compared
with the Hatree-Fock state energies (β= 0) and the exact
ground-state energies obtained by diagonalization. The
lowest energy in the potential-energy surface corresponds
to the bound distance of H2molecule, which is around
0.75˚
A.
In our numerical simulations, we calculate the ground-
state energy of LiH and quantum Ising chain to study the
success probability and the influence of noises. For LiH,
we use 6 work qubits and 1 ancilla qubit. The Hatree-
Fock state is |ΦHFi=|110000iin the qubit represen-
tation. Here |ΦHFiis very close to the exact ground
state, so it takes few steps for the state to converge.
Therefore, to show more about the convergence process,
FIG. 2. Experimental results of PITE on Quafu and IonQ
cloud platform. (a) H2energy expectation value as a func-
tion of β, at a fixed interatomic distance R= 0.75˚
A. The
identity term in the Hamiltonian is considered when calculat-
ing the energy value, but not considered when executing the
algorithm. (b) H2energy as a function of βand the inter-
atomic distance R. The black line is the exact ground state
energy obtained by diagonalization.
we use a superposition of |ΦHFiand an excited state,
|Φ0i=0.99 |ΦHFi+ 0.1|000011i, as the initial state.
Fig. 3(a) and Fig. 3 (b) show the convergence of the
energy E(β) and the fidelity Fas a function of β, re-
spectively, where Fis the fidelity between |Φβiand the
exact ground state |ψGi. The influence of quantum noises
are studied by applying quantum channels on all qubits
before each measurement of the ancilla qubit. The noise
is described by
E(ρ) =
3
X
ν=1
ˆ
Eνρˆ
E
ν(10)
with Kraus operators
ˆ
E1=1 0
01rd,
ˆ
E2=0d
0 0 ,ˆ
E3=0 0
0r(11)
where rand dare relaxation parameter and dephasing
parameter, respectively (see details in Appendix G). The
simulation results are also shown in Fig. 3(a) and Fig.
摘要:

AProbabilisticImaginaryTimeEvolutionAlgorithmBasedonNon-unitaryQuantumCircuitHao-NanXie1,Shi-JieWei2,yFanYang1,Zheng-AnWang2,Chi-TongChen3;4,HengFan3;2,andGui-LuLong1;2z1DepartmentofPhysics,TsinghuaUniversity,Beijing100084,China2BeijingAcademyofQuantumInformationSciences,Beijing100193,China3Institu...

展开>> 收起<<
A Probabilistic Imaginary Time Evolution Algorithm Based on Non-unitary Quantum Circuit Hao-Nan Xie1Shi-Jie Wei2yFan Yang1 Zheng-An Wang2 Chi-Tong Chen34 Heng Fan32 and Gui-Lu Long12z.pdf

共13页,预览3页

还剩页未读, 继续阅读

声明:本站为文档C2C交易模式,即用户上传的文档直接被用户下载,本站只是中间服务平台,本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私,请立即通知玖贝云文库,我们立即给予删除!
分类:图书资源 价格:10玖币 属性:13 页 大小:3.91MB 格式:PDF 时间:2025-04-27

开通VIP享超值会员特权

  • 多端同步记录
  • 高速下载文档
  • 免费文档工具
  • 分享文档赚钱
  • 每日登录抽奖
  • 优质衍生服务
/ 13
客服
关注