Upload
vandan
View
224
Download
0
Embed Size (px)
Citation preview
1
Introduction to Quantum ComputingLecture 2: Quantum Algorithms
Rod Van [email protected]
June 27-29WIDE University School of Internetwith numerous slides from E. Abe
Course Outline
● Lecture 1: Introduction● Lecture 2: Quantum Algorithms● Lecture 3: Devices and Technologies● Lecture 4: Quantum Computer Architecture● Lecture 5: Quantum Networking & Wrapup
2
Outline
● Introduction● Deutsch-Jozsa● Grover's search algorithm● Shor's factoring algorithm● Brief look at Quantum Error Correction (QEC)
How Do Quantum Algos Work?
● Runs are begun by creating a superposition of all possible input values.
● Executing a function gives a superposition of answers of all possible inputs! The hard part is extracting the answer we want.
● Every part of the superposition works independently on the algorithm.
● They all work by using interference. The phase of parts of the superposition are arranged to cancel out and leave only the interesting answer.
3
n⊗0
0 0 0 0 1
2
0 1
2
0 1
2
1
2 n 2 0 00 0 01 1 11
1
2 n 2 0 1 2 N 11
2 n 2x
N 1
x
H 1
2
1 11 1 2n N created superposition of all
possible inputs
Apply Hadamards to n-qubit register
NO! Final step is to measure the register, and the superposition collapses to ONE state! It's as if we did the computation exactly once.
Quantum Parallelism
12 n 2
For example, x=0,1,…,N-1 calculate f(x) for all values of x
Using the quantum superposition, we run the calculation all at once
So do we get all of the answers for f(x)?
H⊗ n
(|ƒ (0) > + ƒ (1) > + ƒ (2) > + - - - + | ƒ (N-1) > )
H⊗n
Entanglement and Superposition
● Two qubits can have dependent values
00 11
2
1
200 0 01 0 10 1
211
Measure either qubit. If youget 0, you know the otherqubit must be 0.Bit0 Bit1
0 0 50%0 1 0%1 0 0%1 1 50%
1
2
1001
4
Graphic Representation
Each bar is the amplitude of the wave function, that is, the square root of the probability, of finding the system in a particular state.
1
400 1
401 1
410 1
411
1
400 1
401 1
410 1
411
Example: Search
1
nN 2=
Start with all equalprobabilities:
Flip the phase ofthe answer:
Flip all states aboutthe mean (average):
The analog nature of phase figuresin strongly!
β
5
Quantum Algorithms● Deutsch-Jozsa(D-J)
– Proc. R. Soc. London A, 439, 553 (1992)● Grover's search algorithm
– Phys. Rev. Lett., 79, 325 (1997)● Shor's algorithm for factoring large numbers
– SIAM J. Comp., 26, 1484 (1997)
D. Deutsch R. Jozsa L. K. Grover P. W. Shor
Deutsch's ProblemDetermine whether a function is “constant” or “balanced”
quantumQuery once to becertain!
constant:f(x) is either 0 for all x, or 1 for all x
balanced: f(x) is 0 for half the values of x, 1 for half the values
e.g., n=2:
e.g., n=2:
classicalMust query 2n-1+1 times to be certain
f x
f(x) =(0,0,0,0) or f(x)=(1,1,1,1)
f(x) =(0,0,1,1)
f(0) =?
f(1) =?
f(2) =?
6
D-J Circuit
0
0 n
f fS
register
work bit
H x 1
2 z
x 1
2n 2z
∵ H 00 1
2H 1
0 1
2
Hadamard Gate
w h e r e , x x1
x2
xn
z z1
z2
zn
x z x 1 z 1 x 2 z 2 x n z n
transforms:
f Gate
S Gate
H⊗ n
H⊗ n⊗
(-1)xz|z>
(-1)xz|z>
ƒ|x>|y>|x>|y ƒ(x)>⊗
ƒ|x>|0> = |x>|0 ƒ(x)>=|x>|f(x)>⊗
ƒ|x>|f(x)>=|x>|f(x) f(x)>=|x>|0>⊗
S | x > | y > = (-1)y | x > | y >
What Happens in D-J
0 n0 H n f1
2n 2x
x 0 1
2 n 2x
x f x
7
D-J (More Efficient Version)
8
D-J and GroverDeutsch-Jozsa takes advantage of interference in the phase to cancel out unwanted terms in the superposition.
But, D-J uses only +1 and -1 in the phase and essentially calculates parity.
Grover's algorithm takes advantage of the full continuous nature of phase to create interference...
(Note: Remember, phase applies to the whole term in the superposition, not just a single qubit! Shift the phase on any qubitand you shift it on the whole term in the superposition.)
Grover's Search
nN 2= items
N 2 n unordered items, we're looking for item “β”
Classically, would have to check about N/2 items
To use Grover's algorithm, create superposition of all N inputs
N repetitions of the Grover iterator G will produce “β”
Hard task!!
β
9
Grover's Algorithm Circuit
oracle: sometimes called a “gate”, an unknown function that tells us if we have the answer
G Gate Analysis
O 2Ox x 2 x x x
H n H noracle 2 0 0 1
O 1 2
2 1k
k k2N k ,k' ,k''
k k' ' k' kk
k k
k2
kk ∵ α
k
αk
N
sign flip for the answer only!
“inversion about average”
0k 2
10
G Gate Analysis(2)
2 1 2cos 2sin 2
cos 2 sin 21 00 1
cos sinsin cos
O 1 2
1 00 1
2 01
0 1 1 00 1
G 2 1 Ocos sinsin cos
1
N 1x
x10
01
2
O
flipped with respect to
θ
flipped with respect to
rotate by angle θ
G
let alpha be the superposition of the N-1
possibilities that are wrong (e.g., not β)
N 1N
1
N
cos 2sin 2
∵θ 2arctan 1 N 1
2
2-D plane
2 30
Execution Example, N=4
1
4sin
2
changes in planeEvery value starts with equal amplitude (probability)
12
00 01 10 11
OO
12
00 01 10 11
2 11
2 1
60
1 execution of G gives 100% chance of correct answer (rare case)
2 k
214
12
00 214
12
01 214
12
10 214
12
11
10
11
Execution Example, N=8
1
8sin
2
G 5
281
28
Gθ41.4°
Every value starts with equal amplitude
2
G11
4 8 1
4 8
G
θ
Executing G twice takes us close to the answer; more takes us PAST the answer
changes in plane
Repeat CountHow many times do we have to repeat G to get the
answer?
12
Grover and Shor
Grover uses interference in phase to cancel unwanted terms.
Shor goes a step further, broadening the range of conditions in which useful interference occurs, by doing a Fourier transform...
Shor's Factoring Algorithm66554087 ? 6703 9929
An “efficient” classical algorithm for this problem is not known.
Using classical number field sieve, factoring anL-bit number is
Using quantum Fouriertransform (QFT), factoring an L-bit number is
Superpolynomial speedup! (Careful, technicallynot exponential speedup, and not yet proven thatno better classical algorithm exists.)
13
Quantum Fourier Transformj QFT
N
1
N k 0
N 1
exp 2 ijk N k
H R2
R3
H R2
HR
k1 00 exp 2 i 2k
Circuit for QFT8
QFT818
1 1 1 1 1 1 1 11 1 2 3 4 5 6 7
1 2 4 6 1 2 4 6
1 3 6 1 4 7 2 5
1 4 1 4 1 4 1 4
1 5 2 7 4 1 6 3
1 6 4 2 1 6 4 2
1 7 6 5 4 3 2 1
exp 2 i 8 iUnitary matrix for QFT8
i i 4 0
FFT in quantum form
14
Order-finding
15
Order-finding Process
0 43 H 4 14 x 0
15
x 3
Î 3
14 x 0
15
x x 3
14
0 4 8 12 3
14
1 5 9 13 1
14
2 6 10 14 7
14
3 7 11 15 0
measurementgives one of
0,4,8,12Run through classical computer to find r
14
0 4 8 12 3
14
0 i 4 8 i 12 1
14
0 4 8 12 7
14
0 i 4 8 i 12 0
QFT16
QFT16
1
16
1 1 1 11 1 2 15
1 2 4 14
1 15 14 1
exp 2 i 16i i 8 0
Shor's Algorithm Flow
ar 1 modL find smallest r
r e v e n
p gcd ar 2 1,Lq gcd ar 2 1,L
L pq
calculate
a 1 a LL is the number we wish to factor; pick a random
p , q L
1. even2. prime3. power of a prime
Algorithm fails if L is:
NO
NO
YES
YES
quantum order-findingcalculate
16
Factoring L=15
1 7 4 131 1 2 1 3 1
4 1ar 2 1 48 gcd 48,15 3ar 2 1 50 gcd 50,15 5
a 7
1 141 1
2 1
ar 2 1 13 gcd 13,15 1ar 2 1 15 gcd 15,15 15
a 14
r2 r7 r8 r13 4
Failure!!
1 111 1
2 1
ar 2 1 10 gcd 10,15 5ar 2 1 12 gcd 12,15 3
a 11
r4 r11 r14 2
a 2,4,7,8,11,13,14
17
Euclid's Method
494 133 3 95
133 95 1 38
95 38 2 19
38 19 2
95
133
3895
19
38
Algorithm for findinggreatest common
denominator
494
133
gcd(494,133)
Π Gate (Circuit)x 2n 1 x n 1 2 n 2 x n 2 2 x 1 x 0
a x mod L a2 n 1 x
n 1 a2n 2 x
n 2 a2 x
1 ax
0 mod L
a2 n 1 x
n 1 mod L a2 n 2 x
n 2 mod L ax
0 mod L
Î x y x axymodL
Î therefore,
U2jU20
U21U2n 1
x0
x1
xj
xn 1
U2jy a2j
y modL
can be constructedusing classical techniques
a2jmodL
Van Meter & ItohPRA 71, 052320 (2005)「xj is the “enable bit” for the
multiplication
18
Back to Factoring L=15
a 7Step.1 Pick random a
© Colin P. Williams (1997)
Step.2 Order-findingQFT example (N =28=256)
0 50 100 150 200
1
0.1
0.01
0.001
0.0001
0.00001
0 64 128 192
Step.3 Measure to recover r
r 4
Step.4p gcd( )ar 2 1,Lq gcd( )ar 2 1,L
calculate
ar 2 1 48 gcd(48,15)=3ar 2 1 50 gcd(50,15)=5
15 3 5Done!
Other Order-Finding Algos.
● Abelian subgroup, discrete logarithm● QFT based, but very different in classical portion
of algorithm● Hidden subgroup problems in general
19
Main Classes of Algorithms
● 1: Use the QFT to find periodicity● 2: Grover's algorithm and friends● 3: Simulating quantum physics● (D-J seems to fall outside these)
Architecture Affects Algorithm Efficiency
Long-distance gates
Neighbor onlyoperations requireswapping qubits
2-qubitgate
Single line layout, neighbor only
2-qubitgate
swapqubitvalues
20
Quantum Error Correction
● Discovered by Shor and Calderbank● Important work done by Steane● Based on known classical techniques, but with
important extensions● Must correct both bit errors and phase errors
3-bit Code EncodingKeep a logical qubit in triplicate:
0 1 enc odeL
000 111
This protects against single bit-flip errors,same as in classical coding.
1
0
0} L
Encoding circuit:
21
3-bit Code Correction
000 111 er r or 010 101
Bit flip error occurs:
How do you correct?Calculate parity pair-wise; results will tell you which bit is in error,then a simple NOT gate on that bit will correct.
1
2
0
Parity circuit:
This bit now holds parity of 1 & 2
Just measure that bottom bit and you get the parity of that pair;two parities can tell you which bit was in error.
Wait a Minute!You said measurement destroys the superposition, won'twe lose our important quantum state?
Not quite... you only lose as much superposition as youlearn about the state. This is called a “partial trace”. α| 000 > |0 + β | 111 > | 0 > parity
0 111 0 parity 010 parity 101 parity
Measuring this qubit...Tells us nothingabout these qubits.
But measuringthis one wouldtell us somethingand collapse the state!
22
What About Phase?
So far, that gets us bit flip protection. Now we need phase flip...
This is a basis change, and protects us against the phase flip error:
This code can be combined with the bit flip code to create the9-bit Shor code.
Wrap-Up on Algorithms
● “Quantum” algorithms actually have both quantum and classical parts
● Use of quantum interference based on complex,analog phase is critical
● Period-finding algorithms work well(exponential speedup over best known algos, but not yet proven better)