Matt Purkeypile's Doctoral Dissertation Defense Slides

Cove: A Practical Quantum
Computer Programming
Framework

Matt Purkeypile
Doctorate of Computer Science
Dissertation Defense
June 26, 2009

Outline
• This presentation will cover the following:
– A brief introduction to quantum computing.
– Walking through a simple factoring example.
– Programming quantum computers.
– Cove: A new solution for programming
quantum computers.
– Questions

Quantum Computing
• Existing computers (classical) operate on bits,
which can hold the value of 0 or 1.
• Quantum computers operate on qubits, which
can hold the value of 0, 1, or a combination of
the two.
– Utilizes probability amplitudes, which means they can
reinforce or cancel out.
• What known problems can quantum computers
do better?
– Factor numbers, which means RSA can be cracked.
• A simple example will be shown.
– Simulate quantum systems.
– Unsorted searches.

Classical and quantum comparison

• The bit is just the poles of a qubit.
• The probabilistic bit is just a line through the
poles of a qubit.

Mathematically
• General state of an arbitrary qubit:
α 0 
ψ = α 0 0 + α1 1 =  
α 1 
2 2
α 0 + α1 = 1
• α1 and α2 are complex numbers and represent
probability amplitudes.
– Hence the total of 1.
– cos(θ ) 0 + eiϕ sin(θ ) 1 in polar form, not commonly used.
2 2
• n qubits are described by 2n complex numbers.
• Operations on n qubits are described by a 2n x
2n matrix of complex numbers.

Limitations of quantum computers
• There are several limitations of quantum
computers.
– Although qubits can hold many possible
values, only one classical result can be
obtained from every run.
• Hence the output is probabilistic.
• Repeated runs may be necessary to obtain the
desired result.
– The computation must be reversible.
– It is impossible to copy qubits (no-cloning
theorem)

Practical Example: Factoring
• Shor’s algorithm for factoring (1994) is perhaps
the most famous practical quantum computing
example.
– It is exponentially faster than the classical solution.
– A quantum computer is utilized for only part of the
algorithm.
• This means you still have to do classical computation.
• Factoring means you can break codes such as
RSA.
– RSA is frequently utilized.
– If N=pq, it is easy to calculate N when given p and q,
but very hard to determine p and q when only given
N.
• Also known as a one-way function.

High Level View of Factoring

• Except for step 2, the algorithm is carried out classically.
• A probabilistic algorithm: may have to repeat runs until
the answer is achieved.

Trivial Example
• Goal: Factor 15.
●
Result is 3 and 5.
●
This has been done on quantum computers in the lab.
●
Can be worked out by hand.
• Step 1, let:
●
N = 15 (the number we are factoring)
●
n = number of (qu)bits needed to express N, in this
case 4.
●
m = 8 (a randomly selected number between 1 and N)

Step 2
• Calculate: f ( x) = m x mod N
●
Need to calculate with enough x’s to find the period.
● x = 0,1, 2,...
●
In general, go to at least N2 values.
●
It seems like guessing would be faster, but isn’t.
●
For this example we’ll just do 0 – 15.
• Given this we can find the period (P).
●
Essentially where f ( x) repeats.
●
In other words f ( x + P ) = f ( x) for every x.
• Performing all these calculations where we need
only one answer (P) is how we can exploit a
quantum computer.

Result
f (0) = 80 mod15 = 1
f (1) = 81 mod15 = 8
f (2) = 82 mod15 = 4
f (3) = 83 mod15 = 2
f (4) = 84 mod15 = 1
f (5) = 85 mod15 = 8
f (6) = 86 mod15 = 4
f (7) = 87 mod15 = 2
f (8) = 88 mod15 = 1
f (9) = 89 mod15 = 8
f (10) = 810 mod15 = 4
f (11) = 811 mod15 = 2
f (12) = 812 mod15 = 1
f (13) = 813 mod 15 = 8
f (14) = 814 mod15 = 4
f (15) = 815 mod15 = 2

Can easily see the period
graphically

Using the period (P)
• The period is 4
●
It repeats 1, 8, 4, 2,…
●
This concludes step 2
• Step 3: is P even?
●
If not we start over using a different randomly
selected m, however in this case it is even.
• Step 4: Utilize P:
8 + 1 = 8 + 1 = 65
4/2 2

8 − 1 = 8 − 1 = 63
4/2 2

Check the result
• gcd(65, 15) = 5 and gcd(63, 15) = 3
●
Can be done efficiently on classical computers [1].
• Step 5: we have found the factors 5 and 3.
●
May only obtain one of the factors.
●
Simple to obtain the second factor if not found.
●
Basic algebra: pq=N, we know N and either p or q.
●
Start over with a different m if the gcd of the results
are 1.

[1] M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information, 1
ed. Cambridge, UK: Cambridge University Press, 2000.

How does a quantum computer
help?
• A quantum computer speeds things up by doing step 2
(finding the period) efficiently.
● Qubits are put in a superposition to represent all possible x’s at
once (in the first register).
● In the case of factoring 15 we need 12 qubits (2(4) + 4, as we
need two registers) [2]
• Next f ( x) is performed on the qubits in superposition.
● One calculation on a quantum computer, many more classically.
● The result is put in the second register.
• Measure Register 2- collapses the superpositions.
• The period is then obtained via the Quantum Fourier
Transform (QFT) followed by a measurement.
• The rest of the algorithm is done classically.
[2] N. S. Yanofsky and M. A. Mannucci, Quantum Computing for Computer
Scientists, 1 ed. New York, NY: Cambridge University Press, 2008.

What is really happening after first
measurement?

f(x) for m=8, N=15 after measurement of 4

5
4
3
f(x)

2
1

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
x

How about QFT and the second
measurement?

Scaling
• 15 is a trivial example, how about a 128 bit number?
• We need at least 384 qubits (128 * 3) to do the quantum
part of the algorithm. (scratch qubits not accounted for)
● The quantum operations that are performed are done once, just
on more qubits.
● Similar to adding two integers: same technique, more bits.
• If we do it classically we have to calculate f(x) many
times.
● It isn’t how easy it is to calculate f(x), it is how many times.
● Need to go from 0 to N2 , this is a huge number of calculations
for a 128 bit number! This could be 2(2*128) or ~1.16 x 1077
● The results have to be stored somewhere (taking up memory)
and then we still have find the period!
● Or we can just use 384 qubits and run through a set of quantum
operations once per attempt, so the quantum computer scales
quite well.
• Likewise, Quantum Fourier Transform also finds the
period in one operation.

What do you need to program
quantum computer?
• Fundamentally, there are only three things
needed to perform quantum computation:
– Initialization of a register (collection of multiple qubits)
to a classical value.
– Manipulation of the register via (reversible)
operations.
– Measurement, which “collapses” the system to a
classical result.
• Hence input and outputs are classical values.
• Like programming classical computers, this is
harder than it sounds.

Programming Quantum
Computers?
• Quantum computers hold immense power, but how do
you program them?
– The operate fundamentally different from classical computers, so
classical techniques don’t work.
• With the exception of one technique [3], all existing
proposals are new languages.
– New languages may be able to perform quantum computation,
but lack power for classical computation.
– Quantum computing is typically only part of the solution, as in
factoring.
– Often geared more towards mathematicians and physicists more
than programmers.

[3] S. Bettelli, "Towards an architecture for quantum programming," in Mathematics. vol.
Ph.D. Trento, Italy: University of Trento, 2002, p. 115.

Grover’s algorithm in Bettelli’s:

Deutsch’s algorithm in Tafliovich’s:

A new solution: Cove
• Cove is a framework for programming quantum
computers.
– This means classical computation is handled by the
language it is built on (C#)
– It designed to be extended by users.
– Key concept: programming against interfaces, not
implementations.
• The current work includes a simulated quantum
computer to execute code.
– All simulations of quantum computers experience an
exponential slow down.

Why is Cove a new contribution?
• Provides extensibility not present in Bettelli’s
solution.
– Like Bettelli, classical computation is handled by the
existing language.
• Provides an object oriented approach for
quantum computing.
• Documentation is as important as the
framework.
– Available online, within code, intellisense, and a help
file.
• Attempts to avoid numerous usability flaws that
are present in all existing proposals to various
degrees.

Example: Entanglement
• Measurement of one qubit impacts the state of
another.
– This doesn’t happen in a classical computer, bits are
manipulated independently- no impact on other bits.

Example: Implementation of Sum

(documentation of method excluded)

Reflections
• Unit testing led to a much more solid design and
implementation.
– Forced code to be written that utilized Cove.
– Takes hours to run tests with just a handful of qubits.
• Implementation of the local simulation was much harder
than anticipated.
– Many problems with implementation aren’t documented well:
• Reordering operations.
• Expanding operations to match register size.
– Memory and time constraints limit what can be done.
• Ran into memory constraints early on.
• Applying an operation to a 20 qubit register requires 220 + (220)2
=1,099,512,676,352 complex numbers!
• Makes debugging difficult.

Areas for future work
• Make the prototype implementation more robust
and complete.
– Utilize remote resources?
• Investigation into the expanded QRAM model.
– Essentially how classical and quantum computers
interact.
• Provide solutions for other algorithms such as
Grover’s (unsorted search).
• The number of quantum algorithms is small, so
that is an area for work as well.

Conclusion
• Quantum computers can carry out tasks
that can never be done on classical
computers, no matter how fast or powerful
they become.
• Existing quantum programming
techniques suffer from numerous flaws.
• Cove is a new method of programming
quantum computers that tries to avoid
flaws of existing techniques.

Questions?

https://cove.purkeypile.com/

(Source code, documentation, dissertation,
presentations and more)

Matt Purkeypile
mpurkeypile@acm.org

Matt Purkeypile's Doctoral Dissertation Defense Slides

More Related Content

What's hot (20)

Similar to Matt Purkeypile's Doctoral Dissertation Defense Slides (20)

Recently uploaded (20)

Matt Purkeypile's Doctoral Dissertation Defense Slides