The Integration Algorithm

A quantum computer could integrate a function in less

computational time then a classical computer...
1 1 1
I = ∫ ∫ ...∫ f ( x1 , x2 ,...xn )dx1dx2 ...dxn
0 0 0

y = f(x)

y The integral of a one dimensional

function, f(x), is the area between the
f(x) and the x-axis.

x
 Integration via Summation
y=f(x) y=f(x)

y y

x x

The integral, I, can be approximated by a sum, S. Taking

more equally spaced points in the summation, leads to a
better the approximation of the integral.
 Summation
M
 x 
We first evaluate the sum ∑1
f where M is the number of
M 
points used in the approximation. This sums the height of all
the boxes. Multiplying this by the width of each box gives
the area under the boxes.
 1 M  x 
y=f(x)
S =  ∑ f  
M  1 M 
 x 
y Defining f (a ) = f  M  , we see
 
that S is equal to the average value
of f(a).

x
 Quantum Averaging
The average of a function can be found on a quantum
computer in the following way...

Initial state of quantum computer

0 00...0

1 work qubit log2(M) function qubits - these

qubits store the number for which
we will evaluate the function, f(a).
 The Hadamard Transform
The Hadamard transform, H, takes a qubit from a ‘classical’ 0
or 1 state, to a superposition of 0 and 1.

H0 =
1
(0 +1) H1 =
1
(0 −1)
2 2
Hence, Hadamards on all function qubits in the initial state of
our quantum computer will give an equal superposition of all
possible states, a, allowing us to evaluate f(a) for all input
states.
M −1
0 ( 00...0 + 00...1 + ... + 11...1 ) =
1 1
M M
∑0
a=0
a
 Quantum Averaging
We now conditionally rotate the work qubit by an amount f(a)
depending on the state of the function qubits. This puts our
quantum computer into the state...
M −1
1
M
∑
a =0
1 − f (a) 0 a + f (a) 1 a

If we now perform another set of Hadamards on the

function qubits the state 1 00...0 will have an amplitude
M −1
1
of
M
∑ f (a) from which we can get S.
a =0
 Quantum Averaging via NMR
Measurement of a quantum system in a superposition state is
probabilistic. Therefore, we can only extract the amplitude of a
particular state by repeated experiments and measurements of
the system. The more experiments the closer we can estimate
the amplitude.

An NMR quantum information processor allows us to read out

the entire state of our system exactly - allowing us to bypass
methods necessary to amplify the amplitude.
 Integration Gate Sequence
bit

0
work

Extract
0 H H amplitude
0 evaluate of
H H
function bits

f(a) 1 00...0
0 H H state

0 H H

Sequence of conditional rotations - rotate work bit

by some angle if the function bit is 1.
 Integrating Sinusoidal Functions
To integrate a sinusoidal function between 0 and 1 would require each
state, a, to conditionally rotate the work bit by aφ , where φ = freq( f ( x)) M − 1
bit

0 φ 2n− 2 φ 2 n −1φ 2n φ
work

0 H H Extract
amplitude
0 H H
function bits

of
0 H H 1 00...0
state

0 H H

a is stored as a binary number a = an an −1...al ...a0 . Thus the

sequence to evaluate f(a) is a series of conditional gates that
rotate the work bit by an amount 2 φ .
l
 Integration of f ( x) = sin (πx )
1
Actual integration
1
yields:
∫ sin (πx ) dx = .637
0

0 1

The integration algorithm taking the four data points

shown above yields:
1 3  πx 
∑ sin   = .433
4 x =0  3 
 Integrating sin ( πx )
1

work bit 0 1
π 2π
0 3 3
Extract
amplitude
function bits

of
0 H H 1 00
0 H H state

conditional rotations
 Integration Algorithm for sin (πx )
Amplitude of 010
state = .433

Pseudo pure state

Hadamard on function bits

Bits 1 and 3 are

function bits.
Hadamard on
function bits
Conditional rotation
from most significant
function bit
Conditional rotation
from least significant
function bit
  3π 
Integration of f ( x) = sin  2
x
1
 2 
Actual integration
1 yields:
 3π 
∫0 sin  2 x dx = .5
2


0 1
The integration algorithm taking the four data points
shown above yields:
1 3 2π 
∑
4 x =0
sin 
2
x  = .5

  3π 
Integrating 2
sin  x
 2 
1

work bit 0 1
0 Extract
amplitude
function bits

of
0 H H 1 00
0 H H state

0 ( 00 + 10 ) + 1 ( 01 + 11 )
Controlled-NOT gate
 Integration Algorithm Using
CNOT

Initial 000 state CNOT31

Amplitude of 100
state = .5

Hadamard on Hadamard on
function bits function bits
 Quantum Information Processing
using NMR
Spectrometer Nuclear Spins as qubits

ADC for data acquisition

RF synthesizer and amplifier  0 1B
Gradient control

wave guides sample

test tube
I JIS S
9.6 T

RF Wave

RF wave
High field magnet 2-3 Dibromothiophene
 Internal Hamiltonian

• The evolution of a spin system is generated

by Hamiltonians
– Internal Hamiltonian:

Hint =ω IIz+ω SSz+2π JIS Iz I JIS S

9.6 T
Sinteraction
z with B field

spin-spin coupling

2-3 Dibromothiophene
 External Hamiltonian
– Experimentally Controlled Hamiltonian:
Hext (t) =ω RFx (t)·(Ix+Sx)+ω RFy (t)·(Iy+Sy)
spins couple to RF field
I JIS S
– Total Hamiltonian:
9.6 T
Htotal (t) = Hint + Hext (t)
Htotal (t)
controlled via RF wave
Hext (t) 2-3 Dibromothiophene
 The Alanine Spin System
ω 2 = −2286.5Hz
J12 = 54.1 J23 = 35.0

ω1 = 7167.8 Hz C1 C2 C3

J13 = -1.3
ω 3 = −4881.4 Hz

n n
H int = ∑ωk I zk + ∑∑ 2πJ kl I zk I zl
k =1 k =1 l <k
 Radio Frequency Pulses
RF pulses are designed to implement a single unitary operator on
any number of spins. A computer program designed for the specific
spin system is used to search for such a pulse based on the
parameters: duration of pulse, power, phase, and frequency offset.

This pulse
rate (radians)

implements
RF nutation

a Hadamard
gate on the
second and
third spins.

time
 Quantum Error Correction
Start with an initial state and some extra spins
α β Encode
α β
Single bit errors become correlated errors
Decode

No Error α β Measure the extra

Flip Bit 1 α β bits to collapse to
one error and learn
Flip Bit 2 α β what error occurred.
Flip Bit 3 α β Then correct it.
Never need to know the original state!
 Decoherence Free Subspace
Encode Decode
α β Engineered α β
Noise

Encoded
0.8

0.6 Un-Encoded
noi t a mr of nI

0.4
30 60 90
Noise strength (Hz)
 Noiseless Subsystem Experiment
Weak Noise Strong Noise Limit
1
Info
No Encoding, Y Noise
Information

0.8 Un-Encoded
Z-X Noise 0.24
NS-Encoded
0.6 No Noise 0.70
Encoded, Y, Z Noise
Z-X Noise 0.70
0.4 Z-Y Noise 0.70
d
0 10 20 30
Noise Strength (Hz)
ve
e

N
r

e
ne
C
c
e

o
s
et
E
i
g
n
o

Encode Decode
i
l
i
l

† †
U
3 U
1

U
1U
3
†
U
2 α
β
α
β U
2 ε
1ε
0
 Tomography
Not all elements of the density matrix are observable on an
NMR spectra.

σ 2
x σ σ2
x
3
z
To observe the other elements of the density matrix
requires repeating the experiment 7 times with
readout pulses appended to the pulse program.
This is done without changing any other parameters
of the pulse program.
 Creation of a Pseudo-Pure State

thermal state 72o spin 2 rotation and Control2 90o y on

gradient 1&3

Add some
identity

gradient Fake ‘swap’ 1 &2

Pseudo-pure state
 NMR sin ( πx ) Simulation

Hadamard on function bits

Pseudo-pure
state

Hadamard on
function bits
Conditional rotation
from most significant
function bit
Conditional rotation
from least significant
function bit Simulator correlation -.92
 NMR CNOT Simulation

Pseudo-pure CNOT31
state

Hadamard on
function bits

Hadamard on
function bits Simulator correlation -.99
 NMR Experiment
Pseudo-pure state CNOT31

projection = .98 correlation = .97

Hadamard on function Hadamard on function bits

bits

correlation = .92 correlation = .91

 Integration Results
The 100 element gives the result of the integration.

100 element

Amplitude = .497
 Conclusions
•Concrete mapping between integration algorithm and NMR
QIP implementation.
•Sufficient control with current NMR quantum information
processors to execute integration in small Hilbert spaces.
•NMR QIP version of algorithm does not require amplitude
amplification.
•General approach for integrating sinusoidal functions.