Quantum Computing (CST Part II)

Lecture 16: Case Studies in Near-term Quantum Computation

In less than ten years quantum computers will begin to outperform

everyday computers, leading to breakthroughs in artificial intelligence,
the discovery of new pharmaceuticals and beyond.
Jeremy O’Brien (2016)

1 / 18
Timeline of quantum computing

Deutsch-Jozsa Quantum supremacy

Feynman Grover
full-scale era

’82 ’85 ’92 ’94 ’96 ’14 ’19 ??

NISQ era
Deutsch Shor VQE

Threshold

2 / 18
NISQs and full-scale fault-tolerant quantum computers

The successful quantum supremacy experiment, demonstrated by Google

in 2019, has heralded the start of the NISQ era. “NISQ” stands for noisy
intermediate-scale quantum (computer), a name coined by John Preskill.

The NISQ era is an exciting time, as experimentalists begin to use

small-scale quantum hardware to gain better understanding of, for
example, how to encode chemical properties into qubits. It is, however,
becoming increasingly clear that the variational algorithms available in
the NISQ era are unlikely to scale to give an actual quantum advantage.

The quantum algorithms we have studied in this course typically require

full-scale fault-tolerant quantum computers, and the time when such
technology exists I have termed the full-scale era in this lecture.

3 / 18
Quantum algorithms not covered in the course

Some of the most important quantum algorithms that we haven’t had

chance to study in the course are:

Quantum counting, quantum amplitude estimation (QAE) &

quantum Monte Carlo integration (QMCI);
HHL;
The quantum singular-value decomposition (QSVD).

4 / 18
Quantum counting, QAE & QMCI

Approximate quantum counting can be used to estimate the number of

marked elements, and hence the number of Grover iterates required in
unstructured search.
The same essential approach can be used to estimation the amplitude,
a = sin2 θ of any quantum state:

|ψ⟩ = cos θ |Φ0 ⟩ |0⟩ + sin θ |Φ1 ⟩ |1⟩

It turns out that this can be used as a sub-routine to achieve a quadratic

advantage in quantum Monte Carlo integration – which in turn speeds up
estimation and forecasting in a wide-variety of applications.

5 / 18
HHL

HHL is a quantum algorithm invented in 2008 for approximately solving

sparse systems of linear equations, known by the initials of its inventors,
Aram Harrow, Avinatan Hassidim and Seth Lloyd. From the HHL
Wikipedia article:
“Due to the prevalence of linear systems in virtually all areas of science
and engineering, the quantum algorithm for linear systems of equations
has the potential for widespread applicability.”
HHL has the interesting property that it combines many of the
fundamental quantum algorithms that we have studied:
Quantum phase estimation
Hamiltonian (quantum) simulation
Amplitude amplification (a generalisation of Grover’s algorithm)

6 / 18
The quantum singular-value transform

The quantum singular value transform is an extremely general framework,

dubbed the grand unifier of quantum algorithms as each of:
matrix inversion;
unstructured search;
factoring;
Hamiltonian simulation,
can be cast as instances thereof.

7 / 18
Overview of quantum algorithms

Quantum algorithm zoo lists over 60 quantum algorithms, some of the

main ones are:
Algorithm Function Speed-up Era
Shor factoring super-polynomial full-scale
Grover search polynomial full-scale
HHL linear algebra super-polynomial full-scale
QPE chemistry super-polynomial full-scale
QMCI estimation polynomial full-scale
VQE chemistry heuristic NISQ
Annealing optimisation heuristic NISQ
QAOA optimisation heuristic NISQ
Quantum machine learning various both?

8 / 18
Overview of quantum algorithms

Quantum algorithm zoo lists over 60 quantum algorithms, some of the

main ones are:
Algorithm Function Speed-up Era
Shor factoring super-polynomial full-scale
Grover search polynomial full-scale
HHL linear algebra super-polynomial full-scale
QPE chemistry super-polynomial full-scale
QMCI estimation polynomial full-scale
VQE chemistry heuristic NISQ
Annealing optimisation heuristic NISQ
QAOA optimisation heuristic NISQ
Quantum machine learning various both?

8 / 18
Quantum machine learning
“Quantum machine learning” is a buzz-word heavy slide title, but what
does it actually mean? Crudely, it can be divided into three categories:

Algorithm
quantum classical
classical quantum
Data

9 / 18
Quantum machine learning (cont.)
Quantum machine learning on classical data refers to the use of
quantum algorithms to enhance the performance of conventional
machine learning algorithms. For example, quantum optimisation
(annealing or QAOA), search (Grover) and / or linear system solving
(HHL) may be called as subroutines by some otherwise classical
machine learning algorithm.
Classical machine learning on quantum data refers to the use of
conventional, classical learning techniques to learn something about
some quantum data. Quantum state tomography is a basic example.
Quantum machine learning on quantum data is, it could be
argued, true quantum machine learning, in the sense that we want
to discern some information from a quantum data-set, which may
not be possible if that quantum data were simply measured and
classical learning applied.
In short, the second and third items differ because, in the former the
quantum state is measured and thus collapsed into classical data on
which classical machine learning is applied, whereas in the latter quantum
operations are applied to the quantum data.
10 / 18
Areas quantum computing is expected to impact

In chronological order (from most imminent to most distant), quantum

computing is expected to have a big impact on the following:

Chemistry: the expectation of exponential quantum advantage in

chemical simulations is already exciting people working in drug
discovery, oil and gas and many others, as well as myriad
applications in materials science.
Optimisation, estimation and QML: the hope of quadratic (and
potentially exponential) quantum advantage could potentially touch
virtually all areas of engineering and operations research.
Security: as has long been anticipated, when we have full-scale
quantum computers, Shor’s algorithm will be game-changer for
computer security. But so too, perhaps, will be QKD.

Note that this list is by no-means exhaustive, but is simply here to give a
flavour of an optimistic view of what things are to come.

11 / 18
How many qubits do we need?

Asking how many qubits we need to do something than cannot be

achieved classically is a bit like asking how long is a piece of string.
Nevertheless, to give a rough idea:
Quantum supremacy (a mathematically well-defined but useless
sampling problem): 53 qubits1 .
Quantum chemistry (simulation of Caffeine): 160 logical qubits2 .
Estimation (e.g. QMCI in finance): 100 – 1000 logical qubits3 .
Factoring: (breaking RSA-2048): 20M physical qubits4 .

1 https://www.nature.com/articles/s41586-019-1666-5
2 e.g. https://www.ft.com/content/154a1cf4-ad07-11e8-94bd-cba20d67390c
3 https://doi.org/10.1017/dce.2022.36
4 https://cacm.acm.org/news/237303-how-quantum-computer-could-break-2048-bit-rsa-encryption-in-8-hours/fulltext
12 / 18
From the NISQ era to the full-scale era

Fault-tolerance is the feature that distinguishes the full-scale era from the
NISQ era, and this will require an error correction overhead estimated to
be in the region 20–1000. That is, it will take 20–1000 physical qubits to
make each “clean” logical qubit.

Qubit fidelities and error correcting codes may well improve, bringing this
number down, but the fact remains that a serious scaling-up of the
number of qubits in a quantum computer is needed to build a
fault-tolerant quantum computer.

This in turn has led some in the quantum computing community to talk
about the need for a “quantum transistor” – a highly scalable physical
realisation for a qubit, that changes the game for quantum computing in
the same way that the transistor did for classical computing.

13 / 18
Types of qubit

There are various proposals for physically realising a qubit, of which the
most promising are superconducting qubits and trapped-ion qubits.

At present, superconducting quantum computers have the most

qubits, and superconducting qubits offer fast gate times.
On the other hand, trapped-ion qubits have the highest fidelity and
longest coherence times, and essentially have all-to-all connectivity,
unlike the restricted (planar) connectivity of most superconducting
quantum computers.
Other technologies include: silicon qubits; nitrogen-vacancy qubits;
and optical (photonic) qubits.

14 / 18
How good is a particular quantum computer?

The total number of qubits tends to grab the headlines, but how good a
particular quantum computer is actually depends on three factors:

1. The number of qubits.

2. The quality of those qubits (fidelity).
3. The connectivity (what overhead will be incurred to move the qubits
around such that they can interact).

Quantum volume is a measure that has been proposed to quantify how

good a given quantum computer is, incorporating these three factors.
The quantum volume of a quantum computer is given by:

Qv = 2(min(n,d))

where n is the number of qubits and d is the depth of a random circuit

that can be executed before an error is expected to occur.

15 / 18
Quantum volume
The depth term, d, is a function of both the fidelity and the depth
overhead incurred when executing a circuit consisting of random
2-qubit interactions.
Therefore the quantum volume is increased for quantum computers
with higher connectivity, as fewer SWAP gates will be needed to
rearrange interacting qubits to be local, and so the depth overhead
will be smaller.
The presence of the min term in the definition indicates whether the
performance of a given quantum computer is limited by a lack of
qubits or poor fidelity / connectivity of the qubits – i.e., a more
nuanced picture than simply quoting the number of qubits.
Quantum volume has been conceptualised so that it gives a
reasonable benchmark of the general performance of near-term
quantum computers.
However some researchers refute that random circuits are
appropriate for this, and instead assert that it part of the role of
quantum software design to execute algorithms in an efficient
manner, given the physical locality constraints of the hardware.
16 / 18
The state of the art in quantum hardware

Three players lead the way in quantum hardware:

Company Headline claim Qubits Quantum Volume

Google Quantum supremacy 72 –
IBM Most qubits 433 256 = 28
Quantinuum Highest quantum volume 20 32768 = 215

17 / 18
Quantum-inspired classical algorithms

One of the most important problems in data-science is the construction

of recommendation systems. Suppose that we have n products and a
purchase history of m users, from which we need to make product
recommendations:

Until 2016 only techniques which run in time O(poly(mn)) were

known.
In 2016 Iordanis Kerenidis and Anupam Prakash published a
quantum algorithm to achieve this task in time O(poly log(mn)).
That is, an exponential speed-up.
Then in 2018 Ewin Tang published a classical algorithm inspired by
Kerenidis and Prakash’s quantum algorithm that also achieves the
task in time O(poly log(mn)).

Therefore, even in a classical world, there is merit in thinking quantumly.

18 / 18
Quantum-inspired classical algorithms

One of the most important problems in data-science is the construction

of recommendation systems. Suppose that we have n products and a
purchase history of m users, from which we need to make product
recommendations:

Until 2016 only techniques which run in time O(poly(mn)) were

known.
In 2016 Iordanis Kerenidis and Anupam Prakash published a
quantum algorithm to achieve this task in time O(poly log(mn)).
That is, an exponential speed-up.
Then in 2018 Ewin Tang published a classical algorithm inspired by
Kerenidis and Prakash’s quantum algorithm that also achieves the
task in time O(poly log(mn)).

Therefore, even in a classical world, there is merit in thinking quantumly.

18 / 18