Quantum adiabatic algorithms using unitary interpolation
Zhang Shuo1, Duan Qian-Heng1, Li Tan1, Fu Xiang-Qun1, Huang He-Liang1, Wang Xiang1, Bao Wan-Su1, 2, †
Henan Key Laboratory of Quantum Information and Cryptography, SSE IEU, Zhengzhou 450001, China
Synergetic Innovation Center of Quantum Information and Quantum Physics, University of Science and Technology of China, Hefei 230026, China

 

† Corresponding author. E-mail: bws@qiclab.cn

Project supported by the National Natural Science Foundation of China (Grant Nos. 11504430, 11805279, 61501514, and 61502526).

Abstract

We present two efficient quantum adiabatic algorithms for Bernstein–Vazirani problem and Simon’s problem. We show that the time complexities of the algorithms for Bernstein–Vazirani problem and Simon’s problem are O(1) and O(n), respectively, which are the same complexities as the corresponding algorithms in quantum circuit model. In these two algorithms, the adiabatic Hamiltonians are realized by unitary interpolation instead of standard linear interpolation. Comparing with the adiabatic algorithms using linear interpolation, the energy gaps of our algorithms keep constant. Therefore, the complexities are much easier to analyze using this method.

1. Introduction

In 2001, Farhi et al. proposed a novel quantum computing model, the adiabatic quantum computation (AQC).[1] The adiabatic quantum computation has proved to be computational equivalent to the quantum circuit model[2,3] and robustness against thermal fluctuations.[46] Therefore, it has received extensive attention, and several quantum adiabatic algorithms (QAAs) have been proposed and demonstrated, e.g., adiabatic algorithms for unstructured search,[7] Deutsch–Jozsa problem,[812] Bernstein–Vazirani problem,[13] Simon’s problem,[1315] and integer number factorization problem.[1618]

A general QAA is achieved as follows. One first designs the target Hamiltonian Hp, whose ground state encodes the solution of the problem, and then prepare the system in the ground state of the initial Hamiltonian H0. After adiabatically evolving from H0 to Hp, which is achieved by means of a linearly interpolating Hamiltonian H (s(t)) = (1 − s)H0 + sHp, with s(0) = 0 and s(T ) = 1, the system will finally evolve to the ground state of Hp, and the solution is therefore obtained.

The quantum adiabatic condition[19] max0tT|E1(t)|tH(t)|E0(t)||Δ(t)|2ε1

guarantees the adiabaticity during the evolution, where |E0 (t)〉 and |E1 (t)〉 are the instantaneous ground state and first excited state of H (t), Δ(t) denotes the energy gap between these two energy levels. The computational complexity of a QAA, regarding to the run time T, can be estimated using the adiabatic condition (1) as well. However, the estimation of Δ(t) in expression (1) is a hard problem. Hence the complexity of some QAAs might be difficult to analyze.

In 2005, Siu introduced the unitary interpolation method[20] to achieve the time-dependent Hamiltonian H (t), in which the time dependent Hamiltonian has the form H(t)=U(t)H0U(t),

where U (0) = I, and U (T )H0U (T ) = Hp. Then Sarandy and Lidar proposed an adiabatic implementation of the Deutsch– Jozsa algorithm using unitary interpolation.[10] It can be proved that the energy gap keep invariant in such method. Therefore, it is much easier to obtain the time complexity using unitary interpolation.

In this paper, we present two alternative QAAs for Bernstein–Vazirani problem and Simon’s problem, where the time-dependent Hamiltonians H (t) are achieved by means of unitary interpolation. We show that the Bernstein–Vazirani problem and Simon’s problem can be solved by using quantum adiabatic algorithms in O(1) and O(n) runtime, respectively, which is the same as the quantum circuit model.

2. The QAA for Bernstein–Vazirani problem

In Bernsterin–Vazirani problem, we are given an n-bit function fa(x)=ax:=a0x0a1x1an1xn1.

The task is to determine the n-bit a. Classically, the problem can be solved within n queries of f (·), while the Bernsterin–Vazirani algorithm[21] in quantum circuit model only requires a single query of f (·) to achieve the same task.

The initial Hamiltonian for solving the Bernstein– Vazirani problem is H0=I|++|,

where we have normalized the energy scale to unity.

The ground state of H0 is |+〉, which is an equal superposition over all n-bit string |+=12nx{0,1}n|x.

The target Hamiltonian is Hp=UH0U=IU|++|U,

with the unitary operator U=x{0,1}neiπ(ax)|xx|,

and the ground state of Hp is U|+=12nx{0,1}n(1)(ax)|x.

The QAA for Bernsterin–Vazirani problem is achieved as follows.

Step 1 Prepare |+〉, which is the ground state of H0, as the initial state of the quantum system |ψ(t=0)=|+.

Step 2 Adiabatically evolve the system from H0 to the target Hamiltonian HP, under the following time-dependent Hamiltonian H(t)=U(t)H0U(t),

where the unitary operator U(t)=x{0,1}neiπ(ax)tT|xx|.

The unitary operator can be realized by U (t) = exp(iHat) with Hamiltonian Ha=πTx{0,1}n(ax)|xx|.

Under the adiabatic approximation, the system finally evolves to the ground state of Hp, which is |ψ(t=T)U|+=12nx{0,1}neiπ(ax)|x.

The final state can be rewritten in x-basis. Define and , then |0=12(|0+|1),|1=12(|0|1).

The above equations can be summarized in more concise form as |x=12y=0,1(1)xy|y.

For the n-bit case, the state |x〉 is |x=12ny{0,1}n(1)xy|y.

Then the final state can be rewritten as |ψ(t=T)U|+=12nx{0,1}ny{0,1}n(1)xy+xa|y.

Notice that 12nx{0,1}ny{0,1}n(1)xy+xa=δay,

then (t = T)〉 ≈ U |+〉 = |a′〉.

Step 3 At the end of the evolution, measure each qubit of the system in the {|0′〉,|1′〉} basis, and a can be obtained.

The run time of the algorithm is determined by the adiabatic condition max0tT|E1(t)|tH(t)|E0(t)||Δ(t)|2ε1.

Obviously, the energy gap is invariant during the adiabatic evolution, i.e., Δ(t) = Δ(0) = 1. On the other hand, |E1(t)|tH(t)|E0(t)|=|E1(t)|(tU(t))H0U(t)|E0(t)+E1(t)|U(t)H0(tU(t))|E0(t)|.

As , and Ha is commutative with U (t) and U (t), therefore |E1(t)|tH(t)|E0(t)|=|E1|[H0,Ha]|E0|=Δ(0)|E1|Ha|E0|.

Note that any eigenvalue λ of Ha satisfies |λ| ≤ π/T, therefore |E1|Ha |E0〉| ≤ π/T . Thus max0tT|E1(t)|tH(t)|E0(t)||Δ(t)|2πTε,

and the total evolution time Tπε,

that means the algorithm can be achieved in O(1) run time, which is the same order as the algorithm using quantum circuit mode.

3. The QAA for Simon’s problem

In Simon’s problem, a function from n-bit strings to n-bit strings is given by f:{0,1}n{0,1}n,xf(x),

where we are given the promise that there is one unknown string a = a1a2 ⋯an so that f (x) = f (y) (xy) if and only if y = xa.

In order to find a, namely the period of the function, classical algorithm requires exponentially queries of f (·), while Simon’s algorithm in quantum circuit model requires only O(n) queries.[22]

Here, we propose a QAA for Simon’s problem. We choose the initial Hamiltonian H0=I2n|++||00|.

The algorithm requires two n-bit registers with the first one encodes the inputs and the second one records the output of f (·).

Meanwhile, the target Hamiltonian is designed as Hp=UH0U=I2nU|++||00|U,

The unitary operator is U=x{0,1}n|xx|e[i(π/2)f1(x)σx,1]e[i(π/2)f2(x)σx,2]e[i(π/2)fn(x)σx,n],

where σx,i is the σx operator acting on the i-th bit of the second register, and fi (x) denotes the i-th bit of f(x).

The ground state of Hp is U|+|0=12nUx{0,1}n|x|0=12nx{0,1}n|x|f1(x)|f2(x)|fn(x)=12nx{0,1}n|x|f(x).

The quantum adiabatic algorithm for Simon’s problem proceeds as the following steps.

Step 1 Prepare the ground state of H0, as the initial state of the quantum system. |ψ(t=0)=|+|0.

Step 2 Adiabatically evolve the system from H0 to the target Hamiltonian Hp, by means of the time-dependent Hamiltonian H(t)=U(t)H0U(t),

where the unitary operator U(t)=x{0,1}n|xx|exp[iπ2tTi=1nfi(x)σx,i].

The unitary operator can be realized by U (t) = exp(iHat) with Hamiltonian Ha=π2Tx{0,1}n(|xx|i=1nfi(x)σx,i).

At the end of the evolution, we approximately obtain the ground state of Hp |ψ(t=T)U|+=12nx{0,1}n|x|f(x).

Step 3 Measure the second register, and obtain some values |f (y)〉, then the first register will be collapsed to the state , which can be rewritten in the {|0′〉,|1′〉} basis as 12(|y+|ya)=12n(z{0,1}n(1)yz|z+z{0,1}n(1)(ya)z|z)=12n1a:z=0(1)yz|z.

Step 4 Measure the first register in the {|0′〉,|1′〉} basis, some values z which satisfy a · z = 0 can be obtained randomly.

Repeating from Step 1 to Step 4 m times, it yields the following linear equation {az1=0,az2=0,azm=0.

It can be proved that for m = n + O(1), a can be uniquely determined from Eq. (24) with probability at least 2/3,[23], so the QAA for Simon problem requires O(n) evaluations of f .

Same as in Section 2, energy gap Δ(t) = Δ(0). From the adiabatic condition, max0tT|E1(t)|tH(t)|E0(t)||Δ(t)|2=max0tT|E1|Ha|E0||Δ(t)|π2Tε,

the run time of each cycle is Tπ/2ɛ. Therefore, the complexity of the algorithm is O(n).

4. Discussion and conclusion

We first compare our scheme with Ref. [13]. In Ref. [13], I. Hen has proposed the QAAs for the same two problems as well, but his algorithms are significantly different from ours. On the one hand, in our work, the algorithms are devised using unitary interpolation, while in Ref. [13] the algorithms are achieved in linear interpolation. On the other hand, in Ref. [13], the ground states of the Hamiltonian is highly degenerate, and higher levels of coherence are needed,[13] in addition, the adiabatic evolution is achieved in each invariant subspaces, which is different from the traditional AQC model. While, in our work the ground state is unique, which indicates that the two alternative algorithms we proposed can be more robust against noises.

We now discuss the physical realization. The initial and target Hamiltonians in our schemes are one-dimensional projectors, i.e., H = I − |α〉〈α|. It is n-local Hamiltonian, which cannot be physically realized. In the QAA for Bernstein–Vazirani problem, we can take alternative 1-local initial Hamiltonian H0=σx,1σx,1σx,n

with energy gap Δ(0) = 1, and the time-dependent Hamiltonian is H(t)=U(t)H0U(t)=k[σx,kcos(πTakt)+σy,ksin(πTakt)].

which keeps 1-local during the evolution. From the adiabatic condition (1), the algorithm can also be achieved in O(1) running time.

However, in the QAA for Simon’s problem, the two registers are correlated after adiabatic evolution, and might be entangled. It means that the target Hamiltonian might be n-local and not physically realizable, even the initial Hamiltonian is chosen to be 1-local. Nevertheless, like other oracle-based adiabatic algorithms with n-local Hamiltonians, our algorithm provides a meaningful example to investigate some aspects of QAC, such as the role of coherence,[24] and performances in open system.[10,25]

In summary, we have proposed two quantum adiabatic algorithms for Bernstein–Vazirani problem and Simon’s problem. The algorithm for Bernstein–Vazirani problem has an O(1) complexity, and the complexity of the algorithm for Simon’s problem is O(n). Both of the two algorithms have the same complexities as the algorithms in quantum circuit model. In further studies, we would like to find more realizable algorithms using local Hamiltonian, and extend the unitary interpolation method to devise other quantum adiabatic algorithms, such as unstructured search problem and prime factorization problem.

Reference
[1] Farhi E Goldstone J Gutmann S Lapan J Lundgren A Preda D 2001 Science 292 472
[2] Mizel A Lidar D A Mitchell M 2007 Phys. Rev. Lett. 99 070502
[3] Yu H Huang Y Wu B 2018 Chin. Phys. Lett. 35 110303
[4] Aharonov D van Dam W Kempe J Landau Z Lloyd S Regev O 2007 SIAM J. Comput. 37 166
[5] Childs A M Farhi E 2001 Phys. Rev. 65 012322
[6] Roland J Nicolas J C 2005 Phys. Rev. 71 032330
[7] Roland J Cerf N J 2002 Phys. Rev. 65 042308
[8] Das S Kobes R Kunstatter G 2002 Phys. Rev. 65 062310
[9] Wei Z Ying M 2006 Phys. Lett. 354 271
[10] Sarandy M S Lidar D A 2005 Phys. Rev. Lett. 95 250503
[11] Sun J Lu S Liu F Gao C 2014 Quantum Inf. Process. 13 731
[12] Sun J Lu S Liu F Gao C Zhou Q Zhang Z 2014 Chin. Phys. Lett. 31 070304
[13] Hen I 2014 Europhys. Lett. 105 50005
[14] Rao M V P 2003 Phys. Rev. 67 052306
[15] Long Y Feng G Tang Y Qin W Long G 2013 Phys. Rev. 88 012306
[16] Du J Xu N Peng X Wang P Wu S Lu D 2010 Phys. Rev. Lett. 104 030502
[17] Li Z Dattani N S Chen X Liu X Wang H Tanburn R Chen H Peng X Du J 2017 arXiv: 1706.08061
[18] Kieu T D 2018 arXiv: 1808.02781
[19] Messiah A 1962 Quantum Mechanics, Vol. II Amsterdam North Holland Publishing Company
[20] Siu M S 2005 Phys. Rev. 71 062314
[21] Bernsterin E Vazirani U 1997 SIAM J. Comput. 26 1411
[22] Simon D R 1997 SIAM J. Comput. 26 1474
[23] Kaye P Laflamme R Mosca M 2007 An Introduction to Quantum Computing New York Oxford University Press Inc.
[24] Li F Bao W Zhang S Huang H Li T Wang X Fu X 2018 Phys. Lett. 382 2709
[25] Wild D S Gopalakrishnan S Knap M Yao N Y Lukin M D 2016 Phys. Rev. Lett. 117 150501