PreprintPDF Available

# Quantum annealing for solving a nurse-physician scheduling problem in Covid-19 clinics

Authors:
Preprints and early-stage research may not have been peer reviewed yet.

## Abstract and Figures

Quantum annealing (QA) is a metaheuristic methodology that has recently been used to analyze the performance of first generation adiabatic quantum processors, like the one available from D-wave Systems, in solving combinatorial optimization problems. In this paper, we formulate QA representation of problems essential to mitigating the effects of the COVID-19 pandemic, namely, the Nurse Scheduling Problem (NSP), the Physician Scheduling Problem (PSP), and the Nurse-Physician Scheduling Problem (NPSP). The proposed objective functions for each problem are scripted to the Ising model and then transformed into the quadratic unconstrained binary optimization (QUBO) format. An optimal solution is sought for a set of constraints, including the number of nurses or physicians per shift and the maximum number of shifts for each nurse or physician. After reducing the proposed NSP, PSP, and NPSP QUBOs to a novel Ising type Hamiltonian, we obtain solutions from a classical (simulated) annealer, and then from the D-wave 2000Q forward and reverse annealer. Results from the reverse quantum annealer show a dramatic improvement in solution quality as compared to those from the classical annealer. We observe that the reverse annealing method gives us the most satisfactory output from among three heuristic processes used, for each scheduling problem.
Content may be subject to copyright.
Quantum annealing for solving a nurse-physician scheduling
problem in Covid-19 clinics
1*Kunal Das, 1@Sahil Zaman, 2#Arindam Sadhu, 2$Asmita Banerjee, 3&Faisal Shah Khan 1Department of Computer Science, Acharya Prafulla Chandra College (Govt. Sponsored), New Barrackpur, Kolkata 700131, India;2Narula Institute of technology, 81, Nilganj Road, Agarpara, Kolkata-700109, India;3Center on Cyber Physical Systems, Khalifa University, Abu Dhabi, UAE *kunal@apccollege.ac.in, @sahilzaman30@gmail.com, #arindam.hit11@gmail.com,$asmibanerjee42@gmail.com,
&quantumsheikh@gmail.com
Abstract
Quantum annealing (QA) is a metaheuristic methodology that has recently been used to analyze the performance of
first generation adiabatic quantum processors, like the one available from D-wave Systems, in solving combinatorial
optimization problems. In this paper, we formulate QA representation of problems essential to mitigating the effects
of the COVID-19 pandemic, namely, the Nurse Scheduling Problem (NSP), the Physician Scheduling Problem (PSP),
and the Nurse-Physician Scheduling Problem (NPSP). The proposed objective functions for each problem are scripted
to the Ising model and then transformed into the quadratic unconstrained binary optimization (QUBO) format. An
optimal solution is sought for a set of constraints, including the number of nurses or physicians per shift and the
maximum number of shifts for each nurse or physician. After reducing the proposed NSP, PSP, and NPSP QUBOs to
a novel Ising type Hamiltonian, we obtain solutions from a classical (simulated) annealer, and then from the D-wave
2000Q forward and reverse annealer. Results from the reverse quantum annealer show a dramatic improvement in
solution quality as compared to those from the classical annealer. We observe that the reverse annealing method gives
us the most satisfactory output from among three heuristic processes used, for each scheduling problem.
Keywords: simulated annealing; forward annealing; reverse annealing; ground state Hamiltonian; QUBO; Ising
model
1. Introduction:-
The origins of quantum computation (QC) lie in the ideas of Benioff relating to quantum mechanical versions of the
Turing machine [1], and of Feynman pertaining the need of a simple quantum computational system that can
efficiently simulate more complicated ones [2]. In the years that have ensued since these ideas were put forward, the
notion of a quantum computer processing quantum information has been scientifically well established, with several
equivalent models of quantum computing are available. Two most prominent ones are the quantum circuit model [3],
which is popular due to its similarity to the electrical circuit model that prevails in engineering disciplines, and the
adiabatic quantum computation (AQC) model [4], which is based on the adiabatic theorem of quantum mechanics in
which a quantum system’s lowest energy state stays invariant under a slow” enough evolution of the system’s
Hamiltonian. Either model has advantages depending on the nature of the analysis one is interested in performing.
The two models can be transformed into each other in polynomial time.
Simulated annealing was introduced by S. Kirkpatrick et. al. to solve real-world optimization problems [5].They used
thermal fluctuation property to find a global minimum value by escaping local minimums that a system can get stuck
in.Geman and Geman proved that if thermal fluctuation is decreased at the rate of T=c/ln t or slower, then it will reach
the global minimum value (where the constant c depends on the systems size) [6]. Otherwise, it will stay stuck at
some local minimum value. T sallis et. al. proposed another probabilistic method on the basis of conventional
Boltzmann-type probability distributions to find out better convergence to solutions than the thermal fluctuation
methodology [7].
In 1998, Kadowki et al. proposed quantum annealing [8] as a way to solve combinatorial optimization problems.
Compared to the thermal fluctuation methodology, QA is a more efficient metaheuristic for detecting global minimum.
However, there is no guarantee of getting global minimum from adiabatic process [3-4, 11-14].Many combinatorial
optimization problems are reported as an application of QA [15-22], while similar efforts of our proposed work have
been made regarding nursing scheduling problem [23].
D-wave Systems is the first commercial vendor of a quantum computer based on quantum annealing to solve
combinatorial optimization problems. In this paper, we use this quantum computer to find an optimal solution of nurse
and physician scheduling problem in the context of the COVID-19 pandemic. We first construct binary variable model
or QUBO model and then convert it into the equivalent Ising or bipolar model. Basic QUBO formulation is expressed
as
QUBO: Y= xTQx or 
where x is vector of binary variables and Q is N×N upper bound triangular matrix. The entries Qi,i of Q are the linear
diagonal coefficient and off diagonal non zero coefficients are represented by Qi,j. In this manuscript, upper bound
triangular structure of Q matrix is used to define our constraints in QUBO model.
This paper is presented as follows. Section 2 consists of discussion regarding Hamiltonian function formation for
nurse and physician. Hamiltonian function regarding nursing scheduling problem (NSP) is demonstrated in sub section
2.1. Physician scheduling problem (PSP) is illustrated in subsection 2.2. The subsection 2.3 is concentrated on
Hamiltonian function of Nursing Physician Scheduling Problem (NPSP). The outcomes from D-Wave 2000Q
processor for above mentioned scheduling problems are summarized in section 3. Again in this section, the importance
of reverse annealing to find optimal results is also discussed and compared. Finally, we concluded in section 4.
2. Scheduling Problem for Covid-19 Hospital
The NSP, PSP, and NPSP deal with creating a rotating roster for both the nurse and the physician according to some
given constraints for a clinic providing medical care to patients of COVID-19.
2.1 Nurse Scheduling Problem (NSP) for Covid-19 Hospital
We set a four-shift system per day: morning time (MT), day time (DT), early night (EN) and late night (LN). Each
shift consists of 6 hours, for some specified N number of days. Due to overload of working shifts, we set as our first
hard constraint the maximum number of shifts that each nurse can work per day, so as to ensure that the nurse can
take sufficient amount of rest in-between shifts. Due to the high pressure working environment in the ward, we set the
second constraint, a soft constraint, to be that each nurse/physician can do at most two consecutive shifts. According
to WHO guideline, maintaining distance between individuals is extremely important to mitigate the spread of the virus
that causes COVID-19, even for the front-line medical workers; hence, we set our third constraint, another hard
constraint, to be the maximum number of nurses/physicians that can be present during each shift. Our constraints then
are:
1. Hard Constraint 1: The maximum number of shifts, each nurse and physician can do per day/week.
2. Hard Constraint 2: The maximum number of nurse and physician, each shift can have.
3. Soft Constraint: The number of consecutive shifts each nurse and physician can do.
(2.1.a) Hamiltonian formulation of NSP:
For combinatorial optimization problem to be solved on a quantum annealer, it is a best practice to formulate it as a
quadratic unconstrained binary optimization problem (QUBO). The QUBO can then be easily transformed into an
Ising formulation which a quantum annealer like the one developed from D-Wave can accept as input.
In QUBO format, a number of nurses n (1, 2, 3… N), are assigned to complete the set of working shift s, where s
(1,2,3,……S). In our QUBO formulation, each day is divided into four shift. Hence duration of each shift is 6 hours.
Now, binary vector variable of our proposed QUBO function qt,s is belongs to (0,1). If qt,s =1, then nurses is assigned
for shift s. Otherwise scheduling process will not work. Now, QUBO formulation with hard constraint 1 (First
constraint) is presented in equation 1. In this case, we assigned that each nurse can do maximum two shift in a day as
a hard constraint 1. Means maximum duty hours of a nurse is 12 hours. The QUBO function with hard constraint 1 is
depicted in equation 1.



= 
   

 




 



The above constraints are satisfied when the cost function is minimum, which happens only when the left-hand side
and the right-hand side terms of Eq. (1) are equal to zero for some unknown binary variables . The variable is
the positive Lagrange constant, set by the programmer, to tune the function. The expressions
andare the diagonal and off-diagonal terms of our Q matrix, respectively.
For our second hard constraint, we set a total number of hours that each shift can have, combining all the nurses
working hours that are placed in that shift to be a = h * m, where m is the maximum number of members that can be
present in each shift. Hence the QUBO form of our constraint is:




 








 




 



 
The above constraint is satisfied (i.e. the cost function is minimum) only when the left-hand side and the right-hand
side terms of Eq. (2) equal zero, for some unknown binary variables . The variable is the positive Lagrange
constant, set by the programmer, to tune the function. Again, and are the diagonal and off-diagonal
terms of our Q matrix, respectively.
For our third constraint (soft constraint), we take two composite indices, i(n,s) and j(n, s), as functions of the number
of nurses and shifts. Hence the QUBO objective function’s form for our constraint, that is, at most two consecutive
shifts, is set for each team member and takes on the form


 (3)
To minimize the cost function, Equation (3) must be zero, otherwise the constant ‘c’ is set by the programmer to tune
the objective functions for most optimal result.
The complete Hamiltonian of our problem is


 

 

 
We want to find the value of  to minimize our cost function Eq. (4), therefore, we transform our QUBO form to
the equivalent Ising model, where the binary variable  equivalent to the variables {-1, +1},
representing bipolar spin variable  . During the conversion of the QUBO form to Ising form for the D-
Wave quantum annealer, the linear terms and quadratic terms are separated in the form
  



where, are linear terms,  are quadratic terms of our matrix, and corresponds to a physical Ising spin from set
{-1, +1}.
2.2 Hamiltonian formulation of PSP: For our physician scheduling problem, we use the same hard and soft
constraint and the objective function that we formulated for our NSP. For the case of PSP, the upper limits and the
number of shift per day may vary from the NSP. We take the number of physician to be p {1, 2, …, P} and working
shift to be s {1,2,… , S}. From the Hamiltonian (Eq. 4) we follow our QUBO formalism for our PSP as expressed
as equation 5.


 

 

 
From Eq. (5), we state that the constant b represents the total shift hours each physician can do per day, a = h * m,
where h is each shift hour, m is the maximum number of physician allowed per shift.  is binary variable from set
{0,1} where  is physician ‘p’ is placed at shift ‘s’. The Lagrange constants  , are set by the programmer,
to tune the Hamiltonian accordingly.
2.3 Hamiltonian formulation of NPSP: We assume a team of nurses and physicians, working for ‘S’ shift. The team
consist of set of nurse and physician , t {1,2,3,... , T} , we assume the last two members of the set {1,2,3,…, T} ;
T-1 and T be the physician and rest nurse. Let the matrix, for our QUBO form be Q. Let the number of working shift
be s {1, 2,… ,S}, for our case we have four shift, each consist of h = 6 hour, to cover a day of 24 hour. Let q t,s {0
, 1} be a binary variable, where if qt,s = 1, then member ‘t’ is placed on the shift s, otherwise zero. For our first
constraint, we set the maximum number of shift, each nurse and physician can do per day, by total hour as b = 12;
hence each nurse and physician can do at most 2 shift each day. Hence our QUBO form follows:



= 
   

 




 


 
The above constraint is satisfied, i.e. the cost function is minimum, only when the left-hand side and the right-hand
side terms of Eq. (6) become zero, for some unknown binary variables . The variable is the positive Lagrange
constant, set by the programmer, to tune the function.  and are the diagonal and off-diagonal terms
of our Q matrix, respectively.
For our second constraint, we set a total duration of hour that each shift can have, combining all the team members’
working hour, that are placed in that shift be a = h * m, where m, is the maximum number of member that can be
present in each shift. Hence the QUBO form of our constraint is:




 








 




 



 
The above constraint is satisfied, i.e. the cost function is minimum, only when the left-hand side and the right-hand
side terms of Eq. (7), become zero, for some unknown  binary variables. The variable is the positive Lagrange
constant, set by the programmer, to tune the function. and are the diagonal and off-diagonal terms
of our Q matrix, repsectively.
For our third constraint, we take two composite indices as,i(t,s) and j(t, s) as the function of team member and shift.
Hence the QUBO Objective function form for our constraint, that is at most two consecutive shift is set for each team
member is:
  

 (8)
To satisfy our objective function i.e. to minimize the cost function; Eq (7) must be zero otherwise ‘c’ which is a
constant, set by the programmer, to tune the objective functions for most optimal result. Hence our complete
Hamiltonian of our problem is:


   

 

 (9)
We want to find the value of  to minimize our cost function Eq. (9), hence we transform our QUBO form to
equivalent Ising model, where the binary variable  equivalent to  {-1, +1}, which are bipolar
spin variable  . During the conversion of QUBO form to Ising form for D-Wave Quantum Computer, the
linear terms and quadratic terms are separated in the form:
  



where are linear terms,  are quadratic terms of our matrix, and corresponds to a physical Ising spin from set
{-1, +1}.
3. Result and Discussion
We use D-wave’s quantum annealer to solve the NSP, PSP and NPSP. D-wave recently launched Leap 2, a classical-
quantum hybrid sampler that supports up to 10,000 variables fully connected to the quantum processor. First we shall
explore the result obtained from simulated annealing (SA), then we shall discuss the results from quantum forward
annealing (FA) and quantum reverse annealing (RA) in the following subsections respectively. During our
experiments, we have fixed the annealing time to be 200µs and total number of samples used as 1,000. Throughout
our experiments we have fixed the positive Lagrange constants .The standard schedule provided by D-
wave is used for all forward annealing schedules. On the other hand, scheduled annealing process is illustrated for
all the experiments as per Fig 1. The reverse scheduling process is started from time t= 0  with annealing parameter
value 1. The process is continued till 2.75  and ended at the annealing parameter value of 0.45. Then annealing
parameter value s is fixed at 0.45 till 83.3. Finally forward annealing process is started at time 83.3 and completed
the process at time 83.46 with annealing parameter value s=1.
We present the results for NSP problem considering N=4, 5 and 6 nurses and number of shift per day being 4. The
total number shifts for each case are S=4 to 20. The two hard constraints, namely the maximum number of nurses per
shift are fixed at 2 and the maximum number of shifts for each nurse are fixed at S/2 for S=4 to 20.
The result of simulated annealing shows that the ground state Hamiltonian was found to be satisfying all constraints
for the values of , described in table 1 and fig 2. For demonstration, we consider a case of N=4 and
S=12. The graphical representation of optimal ground state solution from simulated annealing for N=4, S=12 is shown
in fig 3. It can be noted that the result of simulated annealing does not satisfy all the hard constraints, like in fig 3 (a)
where the maximum number of shifts per nurse does not meet.
Now, let us discuss about FA and RA for the NPS with above mentioned values of N and S. The results of each type
is shown in table 1 for the value of . The graphical representation of optimal ground state solution from
FA and RA, for N=4, S=12, is shown in fig 3(b) and (c).The schedules are satisfying all hard constraints.
Fig 1:Annealing schedule. Reverse annealing schedule ( t=0 to 2.75  and s= 1 to 0.45) , Pause state (t=2.75 to
83.3  and s= 0.45), Forward annealing process(t=83.3 to 83.46  and s= 0.45 to 1)
Table 1 : Generated ground states from different annealing process for different constraint.
Variables
Variables
Constraint 1
Constraint 2
Simulated
Annealing
Reverse
Annealing
Nurse
Shift
Max Nurse
per shift
Max Shift for
each Nurse
Ground State
H
Ground State H
4
4
2
2
-5333.76
-5644.8
4
8
2
4
-32738.4
-33638.4
4
12
2
6
-97653.6
-100569.6
4
16
2
8
-219859.2
-223027.2
4
20
2
10
-414504
-417600
5
4
2
2
-5950.08
-6019.2
5
8
2
4
553.04
-35798.4
5
12
2
6
-103341.6
-105609.6
5
16
2
8
-228391.2
-232027.2
5
20
2
10
-427464
-432000
6
4
2
2
-6391.44
-6393.6
6
8
2
4
-36719.28
-37238.4
6
12
2
6
-106545.6
-109209.6
6
16
2
8
-36359.28
-37238.4
6
20
2
10
-435744
-441360
(a)
(b)
(c)
Fig 2: Ground state energy with respect to number of shift is achieved by simulated annealing, forward annealing and
reverse annealing for (a) N=4, (b) N=5 and (c) N=6
Next, we consider NPSP scheduling for COVID’19 hospitals. The NPSP problem considers two teams of Nurse-
Physician working for 15 days each, after which time an entire team will be quarantined for the following 15 days.
One Nurse-Physician team will remain as a buffer, in case any member of a given team tests positive for COVID’19.
In such a situation, the entire effected team will be replaced by the buffer team. Let us next consider aNurse-Physician
team consisting of 6 nurses and 2 physicians, on aschedule consisting of 5 days(or 20 shifts). The hard constraints for
nurses are that two nurses must be present during a shift, and that each nurse can work for a maximum of 2shifts per
day. On the other hand, a physician can work every alternative shift, and total two shifts per day. Since the team is
working in a COVID’19 ward, team members are expected to maintain distancing measures, but not take a day off, as
soft constraint, and every fifteen days the entire team will be quarantined or on leave.
We fix the positive Lagrange constants and the SA, FA, and RA scheduling setups remain the same as
those described for NSP. The schedule for optimal ground state Hamiltonian from the three annealing regimens are
described in fig 4 (a), (b) and (c). The graphical representation shows that in case of SA results, all the hard constraints
are not satisfied. On the other hand, FA and RA results show that all of the hard constraints are satisfied. Fig 4(a),(b)
and (c) demonstrate that the physician’s duty is assigned every alternative shift, while a nurse can do two consecutive
shifts. At the same time, for each shift, a maximum of two nurses and one physician are available.
(a)
(b)
(c)
Fig 3: NSP scheduling four nurses for 12 shifts (a) simulated annealing result (b) forward annealing result (c)
reverse annealing result.
(a)
(b)
(c)
Fig 4: NPSP scheduling six nurses and two physician for 20 shifts (a) simulated annealing result (b) forward
annealing result (c) reverse annealing result.
3.1 Comparative Study
Our examples of nurses and shifts per day comprised of 4 Nurse and 4 shifts per day for COVID’19 Hospitals with
different hard and soft constraints as described in section 2.1. In this section, we compare our results with the previous
report by Kazuki Ikeda et al in [23]. The comparison serves to demonstrate the advantage of using Leap2 D-wave
hybrid solver with respect to just the D-wave solver. In this process we have considered the same example used in
[23], that is, 3 nurses and 3 shifts per day, and a total of 6 shifts considered. The shift of a day labeled as day time
(DT), early night (EN) and night time (NT), although both the three shift and four shift systems are commonly utilized
in COVID’19 clinics. However, for the three shift system, same constraints are considered as stated earlier in section
2.1. Maximum number of nurses per shift is 2 and the maximum number of duty is 2 for given roster as a case study
reported in [23], both are considered as hard constraints. The soft constraint is a day off i.e. three consecutive shifts
off. A soft constraint is defined as follows.
Let  be the priority of day off i.e. three consecutive shifts off and defined as



The soft constraint term  the extra external Hamiltonian magnetic field is added with eq. 4 to allow
the soft constraint as reported in [23]. The final Hamiltonian of the stated system becomes


 

 

 

 
By heuristic tuning of positive Lagrange parameter , the day off (or the three consecutive shifts off) requests are
considered for the NSP. This constraint is treated as soft constraint, however, while higher priority is imposed, it is
suppose to meet the day off criteria. On the other hand, hard constraints like the maximum number of nurses per shift,
or the maximum number of shifts per nurse, are satisfied in the nurse-shift schedule. For the above experiment on D-
wave Q2000TM, the hybrid sampler is used. As a result, only one best optimal lowest ground state energy sample with
probability 1 is found. For this experiment, which means that the middle priority of day off is imposed.
To demonstrate the higher priority day off for nurses, the experiment is also carried out for  for specific
nurse and from specific shift, the day off requested. In figure 5, we demonstrate the schedule of three nurses for six
shifts using (a) SA simulator (b) QA and (c) RA with , results shows that
none of the schedule are satisfying 100% day off criteria as day off is soft constraints but all hard constraints are
satisfied. On the other hand, when we impose  for specific nurse and from specific shift for day off, the
results shows in (d) SA simulator (e) FA and (f) RA and the entire schedule are stratifying day off criteria. Fig 6 is
described the best optimal solution of above mentioned problem with minimum ground state energy -3194,-2304 and
-3196 for SA, FA and RA, respectively.
(a)
(d)
(b)
(e)
(c)
(f)
Fig. 5: Optimal scheduling solution for comparative study with kazukilkeda et al. [23], for three shift NSP with
hard and soft constraints. For  (a) SA result (b) FA result using D-wave hybrid solver (c) RA result
using D-wave hybrid solver. For , (d) SA result (e) FA result using D-wave hybrid solver (f) RA result
using D-wave hybrid solver.
Figure 5 shows the RA using D-wave hybrid solver to find lowest ground state energy for the optimal solution sample.
The results show that ground state energy are GSA= -3194, GFA=-2304 and GRA=-3196 respectively while the soft
constraint is of middle priority as depicted in fig 5 (a), (b) and (c). Similarly, ground state energy of figures 5 (d)-(f)
are GSA= -4387, GFA=-3600 and GRA=-4752 respectively, while the soft constraint is of high priority.The ground state
Hamiltonian for SA, FA and RA during middle priority and high priority soft constraint evaluation are shown in fig
6. During comparison with earlier reported NSP with D-wave [23], it can be observed that the day-off constraint is
partially satisfied. It is also to be noted that the above mention report [23] used only the quantum sampler whereas we
have used Leap2 hybrid sampler for these experiments.
Fig. 6: Ground State energy for SA,FA and RA for the middle priority of day-off soft constraint and high priority
day off soft constraint
4. Conclusion
We have presented nurse and physician scheduling problems specifically for the COVID-19 pandemic situation. The
proposed Hamiltonian functions of NSP, PSP, and NPSP are reducing in the QUBO model first and then transform to
the Ising Hamiltonian model. Hence Ising Hamiltonian functions easily solved by commercially available quantum
annealer D-Wave 2000Q. In comparison with earlier reported NSP with D-wave 2000Q [23], it can be observed that
the day off constraint is partially satisfied. It also to be noted that the above mentioned scientific report [23] used
quantum sampler called D-wave sampler to get probabilistic optimized ground state. On the other hand, we have used
Leap2 hybrid sampler for these experiments to get accurate and optimized global minima. In this paper, we fixed all
constraints (Soft and hard both) following the WHO (World Health Organization) specified guidelines. The outcome
of variety of quantum annealing processes proves that all assigned constraints satisfied successfully. Our result also
demonstrates that the proposed Hamiltonian Ising function will provide an accurate, optimized, and uniform solution
with increasing roaster and nurse or physician numbers in fixed annealing duration. It is also noted that among variety
quantum annealing process, the reverse annealing process provides us the most optimized global minimum for all the
above-mentioned scheduling problems. Adding extra soft or hard constraints, our proposed Hamiltonian Ising function
can be tuned more as the quantum annealing process is in its early phase. Yet, our presented work will create a good
impact on society in this pandemic situation.
Acknowledgement
This work has been supported by D-Wave that allowed us to use the commercial D-Wave Quantum Annealer 2000QTM
related to COVID-19 specific scheduling problem.
References
1. Benioff, P. Models of quantum Turing machines. Fortschritte der Physik: Progress of Physics, 46(45),
pp.423-441 (1998).
2. Feynman, R., RP,“Simulating physics with computers,”. International Journal of Theoretical Physics, 21,
pp.6-7 (1982).
3. Humble, T.S., McCaskey, A.J., Bennink, R.S., Billings, J.J., DʼAzevedo, E.F., Sullivan, B.D., Klymko, C.F.
and Seddiqi, H. An integrated programming and development environment for adiabatic quantum
optimization. Computational Science & Discovery, 7(1), p.015006 (2014).
4. Klymko, C., Sullivan, B.D. and Humble, T.S. Adiabatic quantum programming: minor embedding with hard
faults. Quantum information processing, 13(3), pp.709-729 (2014).
5. Kirkpatrick, C.D. Gelatt, and MP Vecchi. Optimization by simulated annealing. Science, 220(4598), pp.671-
680 (1983).
6. Geman, S. and Geman, D. Stochastic relaxation, Gibbs distributions, and the Bayesian restoration of
images. IEEE Transactions on pattern analysis and machine intelligence, (6), pp.721-741(1984).
7. Tsallis, C. and Stariolo, D.A. Generalized simulated annealing. Physica A: Statistical Mechanics and its
Applications, 233(1-2), pp.395-406 (1996).
8. Kadowaki, T. and Nishimori, H. Quantum annealing in the transverse Ising model. Physical Review E, 58(5),
p.5355 (1998).
9. Nishimori, H. and Nonomura, Y. Quantum effects in neural networks. Journal of the Physical Society of
Japan, 65(12), pp.3780-3796 (1996).
10. Chakrabarti, B.K., Dutta, A. and Sen, P. Quantum Ising phases and transitions in transverse Ising
models (Vol. 41). Springer Science & Business Media (2008).
11. Farhi, E., Goldstone, J., Gutmann, S., Lapan, J., Lundgren, A. and Preda, D. A quantum adiabatic evolution
algorithm applied to random instances of an NP-complete problem. Science, 292(5516), pp.472-475 (2001).
12. Aharonov, D., Van Dam, W., Kempe, J., Landau, Z., Lloyd, S. and Regev, O. Adiabatic quantum
computation is equivalent to standard quantum computation. SIAM review, 50(4), pp.755-787 (2008).
13. Albash, T. and Lidar, D.A. Adiabatic quantum computation. Reviews of Modern Physics, 90(1), p.015002
(2018).
14. Fujii, K. Quantum speedup in stoquastic adiabatic quantum computation. arXiv preprint arXiv:1803.09954
(2018).
15. Johnson, M.W., Amin, M.H., Gildert, S., Lanting, T., Hamze, F., Dickson, N., Harris, R., Berkley, A.J.,
Johansson, J., Bunyk, P. and Chapple, E.M. Quantum annealing with manufactured spins. Nature, 473(7346),
pp.194-198 (2011).
16. Venturelli, D., Mandra, S., Knysh, S., O’Gorman, B., Biswas, R. and Smelyanskiy, V. Quantum optimization
of fully connected spin glasses. Physical Review X, 5(3), p.031040 (2015).
17. Venturelli, D., Marchand, D.J. and Rojo, G. Quantum annealing implementation of job-shop
scheduling. arXiv preprint arXiv:1506.08479 (2015).
18. Rosenberg, G., Haghnegahdar, P., Goddard, P., Carr, P., Wu, K. and De Prado, M.L. Solving the optimal
trading trajectory problem using a quantum annealer. IEEE Journal of Selected Topics in Signal
Processing, 10(6), pp.1053-1060 (2016).
19. Neukart, F., Compostella, G., Seidel, C., Von Dollen, D., Yarkoni, S. and Parney, B. Traffic flow
optimization using a quantum annealer. Frontiers in ICT, 4, p.29 (2017).
20. Stollenwerk, T., O’Gorman, B., Venturelli, D., Mandrà, S., Rodionova, O., Ng, H., Sridhar, B., Rieffel, E.G.
and Biswas, R. Quantum annealing applied to de-conflicting optimal trajectories for air traffic
management. IEEE transactions on intelligent transportation systems, 21(1), pp.285-297 (2019).
21. Neukart, F., Compostella, G., Seidel, C., Von Dollen, D., Yarkoni, S. and Parney, B. Traffic flow
optimization using a quantum annealer. Frontiers in ICT, 4, p.29 (2017).
22. Bian, Z., Chudak, F., Macready, W., Roy, A.,Sebastiani, R. and Varotti, S. Solving sat and maxsat with a
quantum annealer: Foundations, encodings, and preliminary results. arXiv preprint arXiv:1811.02524
(2018).
23. Ikeda, K., Nakamura, Y. and Humble, T.S. Application of Quantum Annealing to nurse Scheduling
problem. Scientific reports, 9(1), pp.1-10 (2019).
24. Choi, V. Minor-embedding in adiabatic quantum computation: I. The parameter setting problem. Quantum
Information Processing, 7(5), pp.193-209 (2008).
25. D-Wave System. Getting Started with the D-Wave System; Technical Report for D-Wave Systems. 2019.
26. Boothby, K., Bunyk, P., Raymond, J. and Roy, A. Next-generation topology of d-wave quantum
processors. arXiv preprint arXiv:2003.00133 (2020).
27. D-Wave Launches Leap 2, Opening Door to In-Production Quantum Applications
Available online:https://www.dwavesys.com/press-releases/d-wave-launches-leap-2-opening-door-
production-quantum-applications
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Quantum annealing is a promising heuristic method to solve combinatorial optimization problems, and efforts to quantify performance on real-world problems provide insights into how this approach may be best used in practice. We investigate the empirical performance of quantum annealing to solve the Nurse Scheduling Problem (NSP) with hard constraints using the D-Wave 2000Q quantum annealing device. NSP seeks the optimal assignment for a set of nurses to shifts under an accompanying set of constraints on schedule and personnel. After reducing NSP to a novel Ising-type Hamiltonian, we evaluate the solution quality obtained from the D-Wave 2000Q against the constraint requirements as well as the diversity of solutions. For the test problems explored here, our results indicate that quantum annealing recovers satisfying solutions for NSP and suggests the heuristic method is potentially achievable for practical use. Moreover, we observe that solution quality can be greatly improved through the use of reverse annealing, in which it is possible to refine returned results by using the annealing process a second time. We compare the performance of NSP using both forward and reverse annealing methods and describe how this approach might be used in practice.
Article
Full-text available
The Sherrington-Kirkpatrick model with random $\pm1$ couplings is programmed on the D-Wave Two annealer featuring 509 qubits interacting on a Chimera-type graph. The performance of the optimizer compares and correlates to simulated annealing. When considering the effect of the static noise, which degrades the performance of the annealer, one can estimate an improvement on the comparative scaling of the two methods in favor of the D-Wave machine. The optimal choice of parameters of the embedding on the Chimera graph is shown to be associated to the emergence of the spin-glass critical temperature of the embedded problem.
Article
Full-text available
Adiabatic quantum programming defines the time-dependent mapping of a quantum algorithm into an underlying hardware or logical fabric. An essential step is embedding problem-specific information into the quantum logical fabric. We present algorithms for embedding arbitrary instances of the adiabatic quantum optimization algorithm into a square lattice of specialized unit cells. These methods extend with fabric growth while scaling linearly in time and quadratically in footprint. We also provide methods for handling hard faults in the logical fabric without invoking approximations to the original problem, and illustrate their versatility through numerical studies of embeddabilty versus fault rates in square lattices of complete bipartite unit cells. The studies show these algorithms are more resilient to faulty fabrics than naive embedding approaches, a feature which should prove useful in benchmarking the adiabatic quantum optimization algorithm on existing faulty hardware.
Article
Adiabatic quantum computing (AQC) started as an approach to solving optimization problems and has evolved into an important universal alternative to the standard circuit model of quantum computing, with deep connections to both classical and quantum complexity theory and condensed matter physics. This review gives an account of the major theoretical developments in the field, while focusing on the closed-system setting. The review is organized around a series of topics that are essential to an understanding of the underlying principles of AQC, its algorithmic accomplishments and limitations, and its scope in the more general setting of computational complexity theory. Several variants are presented of the adiabatic theorem, the cornerstone of AQC, and examples are given of explicit AQC algorithms that exhibit a quantum speedup. An overview of several proofs of the universality of AQC and related Hamiltonian quantum complexity theory is given. Considerable space is devoted to stoquastic AQC, the setting of most AQC work to date, where obstructions to success and their possible resolutions are discussed.
Article
We present the mapping of a class of simplified air traffic management (ATM) problems (strategic conflict resolution) to quadratic unconstrained boolean optimization (QUBO) problems. The mapping is performed through an original representation of the conflict-resolution problem in terms of a conflict graph, where nodes of the graph represent flights and edges represent a potential conflict between flights. The representation allows a natural decomposition of a real world instance related to wind- optimal trajectories over the Atlantic ocean into smaller subproblems, that can be discretized and are amenable to be programmed in quantum annealers. In the study, we tested the new programming techniques and we benchmark the hardness of the instances using both classical solvers and the D-Wave 2X and D-Wave 2000Q quantum chip. The preliminary results show that for reasonable modeling choices the most challenging subproblems which are programmable in the current devices are solved to optimality with 99% of probability within a second of annealing time.
Conference Paper
Quantum annealers (QA) are specialized quantum computers that minimize objective functions over discrete variables by physically exploiting quantum effects. Current QA platforms allow for the optimization of quadratic objectives defined over binary variables, that is, they solve quadratic unconstrained binary optimization (QUBO) problems. In the last decade, QA systems as implemented by D-Wave have scaled with Moore-like growth. Current architectures provide 2048 sparsely-connected qubits, and continued exponential growth is anticipated.
Article
We make an analogy between images and statistical mechanics systems. Pixel gray levels and the presence and orientation of edges are viewed as states of atoms or molecules in a lattice-like physical system. The assignment of an energy function in the physical system determines its Gibbs distribution. Because of the Gibbs distribution, Markov random field (MRF) equivalence, this assignment also determines an MRF image model. The energy function is a more convenient and natural mechanism for embodying picture attributes than are the local characteristics of the MRF. For a range of degradation mechanisms, including blurring, nonlinear deformations, and multiplicative or additive noise, the posterior distribution is an MRF with a structure akin to the image model. By the analogy, the posterior distribution defines another (imaginary) physical system. Gradual temperature reduction in the physical system isolates low energy states (annealing''), or what is the same thing, the most probable states under the Gibbs distribution. The analogous operation under the posterior distribution yields the maximum a posteriori (MAP) estimate of the image given the degraded observations. The result is a highly parallel relaxation'' algorithm for MAP estimation. We establish convergence properties of the algorithm and we experiment with some simple pictures, for which good restorations are obtained at low signal-to-noise ratios.