The emergence of quantum technologies: challenges and opportunities

Abolfazl Bayat
白安之

University of Electronic Science and Technology of China, Chengdu

电子科技大学，成都

1
 Industrial revolutions
First industrial revolution (~𝟏𝟕𝟔𝟎): steam engine (James Watt)

Second industrial revolution (~𝟏𝟗𝟎𝟎): electricity (Thomas Edison)

Third industrial revolution (~𝟏𝟗𝟒𝟎): information technology and computers (Steve Jobs & Bill Gates)

Fourth industrial revolution (~𝟏𝟗𝟖𝟎): digital revolution, Internet and fusion of different technologies
(Google, Facebook, Uber, Airbnb, Amazon)

2
Investment in quantum technologies
Governments:
➢ The US quantum initiative program ($1b)
➢ The European quantum flagship program (€1b)
➢ Quantum hubs in the UK (£560m)
➢ The National Quantum Strategy plan in Australia ($1b)

Companies:
➢ Giant Companies: Google, IBM, Microsoft, Amazon, Huawei, Tencent, Alibaba
➢ Startups: IonQ, PsiQ, Xanada, Zapata, …

3
Moor’s law: Processing power

50 years

1970 2020

The CPU of iphone is ~3000 times faster

than Apollo spacecraft’s

4
Moor’s law: Memory

20 years

Launched 1999
Launched 2020
Weight~ 150 g
Weight~ 160 g
Memory= 16 MB
Memory= 256 GB

16000 times more memory

with the same weight !!

5
Moor’s law

The number of transistors were doubled every two years for a period of ~50 years

Moor’s law is now violated as transistors have reached the atomic scale
6
Quantum Physics

7
Quantum Superposition
 0 + 1 ȁ0ۧ = 1 , ȁ1ۧ =
0
0 or 1 0 1
0&1

𝑎0 ȁ0ۧ + 𝑎1 ȁ1ۧ 2 numbers are encoded by 1 particle

𝑎00 ȁ00ۧ + 𝑎01 ȁ01ۧ + 𝑎10 ȁ10ۧ + 𝑎11 ȁ11ۧ 22 numbers are encoded by 2 particle

𝑎000 ȁ000ۧ + 𝑎001 ȁ001ۧ + 𝑎010 ȁ010ۧ + 𝑎011 ȁ011ۧ + 23 numbers are encoded by 3 particle
𝑎100 ȁ100ۧ + 𝑎101 ȁ101ۧ + 𝑎110 ȁ110ۧ + 𝑎111 ȁ111ۧ

2300 ~ is equal to the number of atoms

By 𝑁 particles one can encode 2𝑁 numbers
In the whole universe
Can quantum mechanics help for big data analysis? 8
 Entanglement
The first consequence of super position principle is quantum entanglement

Can any bipartite system AB be described as: ȁ Ψ𝐴𝐵 ۧ ȁ𝜙𝐴 ۧ ȁ𝜙𝐵 ۧ

ȁ Ψ𝐴𝐵 ۧ = ȁ00ۧ + ȁ11ۧ / 2

ȁ𝜙𝐴 ۧ = 𝑎0 ȁ0ۧ + 𝑎1 ȁ1ۧ
ȁ𝜙𝐴 ۧ ȁ𝜙𝐵 ۧ= 𝑎0 𝑏0 ȁ00ۧ + 𝑎0 𝑏1 ȁ01ۧ + 𝑎1 𝑏0 ȁ10ۧ + 𝑎1 𝑏1 ห11ۧ = ȁ Ψ𝐴𝐵 ۧ
ȁ𝜙𝐵 ۧ = 𝑏0 ȁ0ۧ + 𝑏1 ȁ1ۧ
1
𝑎0 𝑏0 =
2
𝑎0 𝑏1 = 0 It is impossible to be satisfied
𝑎1 𝑏0 = 0
𝑎1 𝑏1 = 1/ 2

While we can precisely describe the whole system, we cannot describe the subsystems by a single quantum state

9
 Quantum evolution (closed systems)

Every quantum state can evolve to another state through unitary operation

ȁΨۧ = 𝑈 ȁ𝜙ۧ 𝑈𝑈 † = 𝑈 † 𝑈=I

For instance (mathematical description):

0 1
ȁΨ(𝜃)ۧ = 𝑒 −𝑖𝜃𝜎𝑥 ȁ0ۧ = cos 𝜃 ȁ0ۧ − 𝑖 sin(𝜃)ȁ1ۧ 𝜎𝑥 =
1 0

In the lab, this unitary rotation is implemented by a magnetic field along the 𝑥 direction

10
 Quantum measurement
Unlike classical physics, quantum measurement changes the state of the system

𝑝0 = 𝑎0 2 : ȁ0ۧ
ȁψۧ = 𝑎0 ȁ0ۧ + 𝑎1 ȁ1ۧ 𝜎𝑧 measurement
𝑝1 = 𝑎1 2 : ȁ1ۧ

Basis of the measurement matter:

0 + ȁ1ۧ + + ȁ−ۧ
+ = 0 =
2 2 𝑎0 + 𝑎1 𝑎0 − 𝑎1
ȁψۧ = 𝑎0 ȁ0ۧ + 𝑎1 ȁ1ۧ = + + −
0 − ȁ1ۧ + − ȁ−ۧ 2 2
− = 1 =
2 2
ȁ𝑎0 + 𝑎1 ȁ2
𝑝+ = : ȁ +ۧ
2
𝜎𝑥 measurement
ȁ𝑎0 − 𝑎1 ȁ2
𝑝− = : ȁ−ۧ 11
2
Quantum technologies timeline

12
Interdisciplinary subject

Physics
& Engineering
Mathematic
Quantum
Tech.

Computer Science

13
What is quantum technology?
Quantum communications:
❑ Quantum key distributions
❑ Quantum Internet

Quantum computation/simulations:
❑ Quantum computers/simulators
❑ Quantum algorithms
❑ Quantum machine learning

Quantum sensing:
❑ Quantum probes
❑ Quantum enhanced sensitivity

14
Application 1: quantum communication

15
Communication

Ideal channel

Message: 001101101001

16
 Secure communication (random shared key)

Message: 0 0 1 1 0 1 Random key: 1 0 1 0 0 1

Random key: 1 0 1 0 0 1 ⊕

Message to send: 1 0 0 1 0 0 Message to send: 1 0 0 1 0 0 ⊕

Faulty channel

Message: 0 0 1 1 0 1
This is random too

Secure communication needs a random shared key

Classically there is no way to provide absolute security for key distribution

17
 Quantum key distribution

Alice Message Bob Basis

basis basis matching
X 0 Z X
Measures in either
𝜎𝑥 : 0: + , 1: ȁ−ۧ X 0 Z X 𝜎𝑥 or 𝜎𝑧
Z 1 X X
𝜎𝑧 : 0: 0 , 1: ȁ1ۧ
X 1 X √ Alice sends − and Bob measures it right
Z 0 X X
Z 1 Z √ Alice sends 1 and Bob measures it right
X 0 X √ Alice sends + and Bob measures it right

18
Quantum key distribution in Space

S.-K. Liao, et. al., Nature 549, 43 (2017)

19
Application 2: quantum computation/simulation

20
Quantum computer
What is a quantum computer: 𝑈
➢ Programmable machine
ȁ𝜙ۧ
➢ Implements any unitary operator ȁΨۧ
➢ Can convert any quantum state into another

Every unitary operator can be decomposed into:

➢ Arbitrary single qubit rotations (SU(2) rotation)

𝑈 𝛼, 𝛽, 𝛾 = 𝑒 −𝑖(𝛼𝜎𝑥 +𝛽𝜎𝑦+𝛾𝜎𝑧 )
0 1 0 −𝑖 1 0
𝜎𝑥 = , 𝜎𝑦 = , 𝜎𝑧 =
1 0 𝑖 0 0 −1

➢ One two-qubit entangling gate

1 0 0 0 ȁ00ۧ → ȁ00ۧ
ȁ01ۧ → ȁ01ۧ
𝑈𝐶𝑁𝑂𝑇 = 0 1 0 0 𝑈𝐶𝑁𝑂𝑇 :
ȁ10ۧ → ȁ11ۧ
0 0 0 1
0 0 1 0 ȁ11ۧ → ȁ10ۧ 21
NISQ era

Noisy Intermediate Scale Quantum (NISQ) devices are available.

➢ Noisy ~ Limited coherence time (~ 300 − 400 CNOT gates)
➢ No error correction
➢ Intermediate Scale (devices with ~200 − 300 qubits are available)
➢ Limited qubit connectivity
22
Quantum simulation
𝑗 𝑖
Hubbard model (High-T superconductors, … )

𝐻 = −𝐽 ෍ 𝑐Ƹ † 𝑖,𝜎 𝑐𝑗,𝜎
Ƹ + 𝑈 ෍ 𝑛ො 𝑖,↑ 𝑛ො 𝑖,↓
<𝑖,𝑗>,𝜎 𝑖

Classical computers cannot handle exponential problems

2^300 > Number of protons in the universe!!
More than 180 cold atom laboratories worldwide !
Complex
quantum
system

Quantum
simulator Controllable
quantum
Salfi, et al, Nat. Commun. 7
system
11342 (2016) 23
Why quantum simulators?
➢ Classical computers are not capable of solving quantum problems due to the exponential growth
of the Hilbert space (~ 2N)

➢ Better controllability (e.g. high Tc superconductivity)

➢ Some theoretical models do not exist in nature (e.g. Kitaev toric model)

➢ Classical big-data problems might be solved faster on quantum

computers/simulators

24
Quantum simulators
Cold atoms in optical lattices
Ion traps Rydberg atoms

Superconducting
Quantum dot arrays quantum simulators

25
NISQ era
Noisy Intermediate Scale Quantum (NISQ) devices are available.
➢ Noisy ~ Limited coherence time (~ 300 − 400 CNOT gates)
➢ No error correction
➢ Intermediate Scale (devices with ~200 − 300 qubits are available)
➢ Limited qubit connectivity

26
Variational Quantum Algorithms
Measuring a
cost function

ȁ0ۧ
𝐶 𝜃Ԧ = 𝜓(𝜃)
Ԧ 𝐻 𝜓(𝜃)
Ԧ
ȁ0ۧ
ȁ0ۧ Circuit Ԧ ඀
ቚ𝜓(𝜃)
ȁ0ۧ 𝜃𝑘

ȁ0ۧ
Updating the
ȁ0ۧ Circuit parameters 𝜽

➢ Complexity is divided between a quantum simulator (i.e. a shallow circuit) and a classical optimizer
➢ Only problems which can be written variationally can be solved
➢ By choosing the observable to be the Hamiltonian then the final output becomes the ground state27
Resources

VQE is a hybrid algorithms using both quantum circuit and classical optimizations

1. Quantum circuit:
➢ Number of layers or number of CNOTs

2. Classical minimization:
➢ Number of iterations in gradient decent algorithm (convergence speed)
➢ The number of parameters to optimize

Classical resources: CR=Number of parameters × Iterations

28
Comparison between adiabatic and VQE

VQE can be implemented on a shallow circuit

29
Implementing symmetries in VQE
𝐻, 𝑆 = 0 𝐻 and 𝑆 have common eigenvectors

2
1- Penalizing the cost function: 𝐶 𝜃 = 𝐻 + 𝑆 − 𝑆𝑡𝑎𝑟

2- Implementing the symmetries in the quantum circuit so that ȁ𝜓(𝜃)ۧ naturally

conserves the symmetry

ȁ0ۧ
ȁ0ۧ
Symmetry
ȁ0ۧ preserving The circuit output naturally conserves
ȁ𝜓(𝜃)ۧ the symmetry: 𝜓(𝜃) 𝑆 𝜓(𝜃) = 𝑆𝑡𝑎𝑟
ȁ0ۧ Circuit
𝜃𝑘
ȁ0ۧ Implementation of symmetries in the hardware is more
ȁ0ۧ resource (both quantum and classical) efficient
30
Quantum machine learning

31
 Machine learning
1- Supervised learning: To predict the label of an unknown input data,
e.g. classification.

2- Unsupervised learning: To group the inputs according to their similarities,

e.g. clustering.

3- Reinforcement learning: There is no data to train. The system learns by

rewarding (or punishing) the desired (undesired) outputs.

32
 Classification Problems
Thanks to the Internet, we have loads of labeled data

Classification is one of the main machine learning algorithms

Cats vs Dogs (2 classes: binary) Handwriting digits (10 classes)

33
Quantum classifiers

➢ Can quantum computers solve classification problems?

➢ If so, can they outperform classical classifiers?

➢ Datasets can be either classical (handwriting) or quantum (ground state of a Hamiltonian).

➢ For quantum datasets it is likely that quantum computers are useful.

➢ For classical datasets it is still an open problem whether quantum computers can provide
any advantage.

34
 Encoding classical datasets
Amplitude encoding: It provides an exponential advantage. Number of qubit=log(𝑁)
ȁ0ۧ
ȁ0ۧ
Input data: 𝒙𝒊 = (𝑥𝑖1 , 𝑥𝑖1 , … , 𝑥𝑖𝑁 ) Encoder
⋮ 𝑥𝑖 ⋮ 𝒙𝒊
ȁ0ۧ
ȁ0ۧ

Rotation encoding: It is easy for experiments but without advantage in scaling. Number of qubit=𝑁

ȁ0ۧ 𝑅𝑦 (𝑥𝑖1 )

ȁ0ۧ 𝑅𝑦 (𝑥𝑖2 ) ȁΨ(𝒙𝒊 )ۧ

⋮
ȁ0ۧ 𝑅𝑦 (𝑥𝑖𝑁 )
35
Variational classifiers
𝒙𝒊 Measurement Class Probability
outcome
ȁ0ۧ
ȁ0ۧ 00 0 𝑃00
Parameterized
Encoder
⋮ 𝑥𝑖 ⋮ circuit 01 1 𝑃01
ȁ0ۧ 𝜃Ԧ 10 2 𝑃10
ȁ0ۧ 11 3 𝑃11

➢ The largest probability determines the class

➢ For probability vector 𝑃𝑖 = (𝑃𝑖1 , 𝑃𝑖2 , … ) The loss function can be defined as: 𝑌𝑖 = (0,0,1,0)
ℒ 𝜃Ԧ = − ෍ 𝑌𝑖𝑇 log(𝑃𝑖 )
𝑖

36
MNIST dataset (odd digits)

➢ Each image is a 8 × 8 pixels (64 features)

➢ Four different classes (odd numbers)

37
Quantum circuit

We use amplitude encoding but the protocol works for rotation encoding too

Error rate: Where 𝕀 True =1 and 𝕀 False =0

38
The effect of layers

➢ In a noise-free quantum computer the accuracy increases by increasing the layers

➢ Training and test errors remain close to each other showing the absence of over-fitting

39
Application 3: quantum sensing

40
Sensing procedure

Unknown parameter
Probe Measurement Estimator
is estimated

The probe, measurement basis and estimators should be optimal

41
 Classical Fisher information

One can measure a random variable 𝑋 for measuring an unknown parameter 𝜃

𝑋 can be the height of the thermometer’s liquid

With the probability of 𝑃𝑥 (𝜃) the height is
𝜃 can be temperature measured to be 𝑥

1
Cramer-Rao inequality: 𝛿𝜃 2 = 𝜃𝑒𝑠𝑡 − 𝜃𝑟𝑒𝑎𝑙 2
≥ M is the number of repetition
𝑀𝐹(𝜃)
2 2
𝑑 log(𝑃𝑥 (𝜃)) 1 𝑑 𝑃𝑥 (𝜃)
Fisher information: 𝐹 𝜃 = ෍ 𝑃𝑥 (𝜃) =෍
𝑑𝜃 𝑃𝑥 (𝜃) 𝑑𝜃
𝑥 𝑥

Larger Fisher information implies better sensitivity 42

Quantum Fisher information
𝜌(𝜃) Estimate 𝜃

1. One can choose both measurement basis and the estimator

2. For any choice of measurement we have projectors: {Mk}

3. The probability of each outcome is: 𝑝𝑘 𝜃 = 𝑡𝑟[𝜌(𝜃)𝑀𝑘 ]

4. Then quantum Fisher information: Fq(θ) = Max{Mk}[Fc(θ)] 𝐹𝑞 ≥ 𝐹𝑐

For a classical probe of size N, at best one can achieve: 𝐹𝐶 = 𝑁

43
 Quantum Fisher Information
𝜕𝜌 𝜃 𝐿 𝜃 𝜌 𝜃 + 𝜌 𝜃 𝐿(𝜃)
𝜌 𝜃 = ෍ 𝑞𝑘 (𝜃)ȁ𝜓𝑘 (𝜃)ۧ‫)𝜃( 𝑘𝜓ۦ‬ȁ SLD =
𝜕𝜃 2
𝑘

𝐹𝑞 𝜃 = 𝑇𝑟 𝜌 𝜃 𝐿2 (𝜃)

2
For pure states: 𝜌 𝜃 = ȁΨ(𝜃)ۧ‫ۦ‬Ψ(𝜃)ȁ 𝐹𝑞 𝜃 = 4 ർ𝜕𝜃 Ψ 𝜃 ȁ𝜕𝜃 Ψ 𝜃 ۧ − ൻΨ 𝜃 ȁ𝜕𝜃 Ψ 𝜃 ۧ

➢ Optimal measurement basis will be the eigenstates of 𝐿 𝜃

➢ The optimal basis might depend on the unknown parameter 𝜃

➢ Super linear scaling (i.e. quantum enhanced sensitivity) might be possible

44
 Quantum sensing
ȁ0,0, … , 0ۧ +ȁ1,1, … , 1ۧ −𝑖𝐻𝑡 ȁ
ȁ0,0, … , 0ۧ +𝑒 −𝑖𝑁𝐵𝑡 ȁ1,1, … , 1ۧ
𝐻 = 𝐵 ෍ 𝜎𝑘𝑧 ȁΨ0 ۧ = ȁΨ𝑡 ۧ = 𝑒 Ψ0 ۧ =
2 2
𝑘

𝑃𝑥+ = 𝑐𝑜𝑠 2 𝑁𝐵𝑡

Proper measurement
𝑃𝑥− = 𝑠𝑖𝑛2 (𝑁𝐵𝑡)

1
𝛿𝐵2 ≥ Heisenberg limit
Fisher information: 𝐹 𝐵 = 4𝑁 2 𝑡 2 Precision 4𝑁 2 𝑡 2

V. Giovannetti, S. Lloyd, and L. Maccone, Science 306, 1330 (2004).

V. Giovannetti, S. Lloyd, and L. Maccone, Phys. Rev. Lett. 96, 010401 (2006).
J. P. Dowling, Contemp. Phys. 49, 125 (2008).
F. Frowis and W. Dur, Phys. Rev. Lett. 106, 110402 (2011).
D. Dobrzanski, J. Ko lodynski, and M. Guta, Nat. Commun.3, 1063 (2012).
H. Kwon, K. C. Tan, T. Volkoff, and H. Jeong, Phys. Rev. Lett. 122, 040503 (2019).
J. Joo, W. J. Munro, and T. P. Spiller, Phys. Rev. Lett. 107, 083601 (2011).
S. Slussarenko, et al., Nature Photonics 11, 700 (2017). 45
 Criticality enhanced sensing

𝐻 = 𝜃𝐻1 + 𝐻2 ȁ𝐺𝑆(𝜃)ۧ Can one estimate 𝜃 directly from the ground state?

Near criticality (𝜃 ≈ 𝜃𝑐 ): 𝐹𝑞 𝜃 ~𝑁 2/𝑑𝑣

𝐹𝑞 𝜃 ~𝑁 𝜃 − 𝜃𝑐 𝑑𝜐−2
Far from criticality:

For Ising in transverse field (𝑑 = 1, 𝜐 = 1) around criticality: 𝐹𝑞 𝜃 ~𝑁 2 Heisenberg limit

P. Zanardi and N. Paunkovic, Phys. Rev. E 74, 031123 (2006).

P. Zanardi, H. T. Quan, X. Wang, and C. P. Sun, Phys. Rev. A 75, 032109 (2007).
P. Zanardi, M. G. A. Paris, and L. Campos Venuti, Phys. Rev. A 78, 042105 (2008).
C. Invernizzi, M. Korbman, L. C. Venuti, and M. G. A. Paris, Phys. Rev. A 78, 042106 (2008).
M. Skotiniotis, P. Sekatski, and W. Dur, New J. Phys. 17, 073032 (2015).
S.-J. Gu, Int. J. Mod. Phys. B 24, 4371 (2010).
S. Gammelmark and K. Mølmer, New J. Phys 13, 053035 (2011). 46
Quantum enhanced sensitivity
What is responsible for quantum enhanced sensitivity?

❑ Scale invariance
❑ Symmetry-breaking
❑ Long-range entanglement/correlations
❑ Gap closing

47
Symmetry protected topological Systems
Topological properties don’t change under continuous deformations.

Phase transitions in such systems are not captured by Landau’s theory:

➢ They are captured by of a global order parameter

➢ There is no symmetry breaking

➢ There is no long-range entanglement

➢ The gap closing still remains valid

➢ Zero energy edge states emerge in the system

Around the critical point the edge states show Heisenberg precision

This shows that gap closing is the key feature for quantum enhanced sensitivity
48
Summary
Quantum technologies may change our lives in a fundamental way in coming decades

Quantum features (such superposition and measurement) can be exploited for surpassing the
performance of classical devices.

The most important aspects of quantum technologies are:

➢ Quantum communications
➢ Quantum simulation
➢ Quantum sensing

49