quantum error correction - lu
TRANSCRIPT
Quantum Error Correction
• In principle: whole chapter 10
• What is error correction? (classically)• Introduction to quantum errors• Some formalism to help us• What are the boundries for correctability
• Fault-tolerant threshold• Example: Surface codes with superconducting qubits
Today’s lecture:
In the book:
• TL;DR version: sec. 10.1-10.2, 10.6
Classical Error Correction
Everyday example: Noisy phoneline
B D VRAVO
ELTA
ICTOR
Classical Error Correction
Example:
Error:With probability pany bit is flipped
000
Repetition coding:00001111
31 pProbability: 213 pp pp 13 2 3p
001
100010
110
011101000 111
Majority voting Scheme fails
?13 32 pppp 021
232 pp
Repetition better if
21 p
When do we gain by error correction?
Parity check:
Classical Error CorrectionRepetition code has heavy cost, +200 % of original message length
byte parity bit = 1 if odd number of 1’s
Error: byte and parity bit missmatch = resend
Works well, if p is very lowExample: computers, where ε < 10-17
1010 001 1
1010 101 1Cost is only +~15%
Quantum Error Correction
Differences from classical:
• No cloning: Cannot use repetition coding directly, becausewe cannot duplicate arbitrary states
• Measurement collapse: Everytime we try to detect whatstate we have, it collapses to the basis
• Continuous errors: Infinite ways that errors can occur,think rotations in the Bloch sphere
Despite all this, quantum error correction still works!
Quantum Error Correction
00000 L
11111 L
0110
XQuantum bit flip = Pauli X operator,
Define logical qubit:
11100010
No cloning… But what does this circuit do?
First task: correct for bit flip
Cannot copy… but can still add redundancy
One quantum bit flipError-detection circuit for original state 111000 orig
Applycorrection:bit flip
I X1 X2 X3
3210 eeeeeeorigtot cccc
1111110000000 P
0110111001001 P
1011010100102 P
1101100010013 P
Projective measurement set:
orig , α and β are untouched!
Quantum Error CorrectionHow much improvement from the error correction?
Measured by fidelity: ,origF
Without error correction:
XpXp )1(
XXpppppp 3223 )1(3)1(3)1(
F XXpp )1( p 1
With error correction:
...)1(3)1( 23 pppF 32 231 pp
21 p
Quantum Error Correction
Quantum phase flip:
10 orig 10
Change basis:
2
10
2
10
L0
L1
1001
Z
Z,
Same procedure as before! )( HZHX
10, X
Quantum Error Correction
Both phase and bit flip at the same time?
First, encode01
, then, encode each of these according to the bit flip:
22
11100011100011100000
L
22
11100011100011100011
L
The Shor code:9 qubit code!
XZeZeXeIeE 3210
Continuous errors?
Saved by the projective measurement!
Ex: How to measure the error syndrome
111000 orig
tot P0 P1 P2 P3Z1Z2 Z2Z3
32121 IZZZZ 11001001
,
Z
2121 11001100ZZ
1010010111110000
+1-1
+1-1
1111110000000 P
0110111001001 P
1011010100102 P
1101100010013 P
Projective measurement set:
Eigenvalues +1 and -1
Error Correction Formalism
• How can we find better codes?• Can we know if we have found the best code?• How can we build real circuits from the theory?• How big error is allowed for a full scale fault-tolerant
quantum computer?
Why do we need to develop formalism?
Error Correction Formalism
Generators (classical)
Definition: a linear code C, encoding k bits of information into an n bitcode space, is specified by an n by k generator matrix G
Example: 3-bit repetion code
111
G code Gxy , where x is the k bit message
0x Ly 0000
1x Ly 1
111
[n,k] code = [3,1] here
Error Correction Formalism
Parity check matrix, H (classical)
0Hy
Example: 3-bit repetion code
, where H is an n – k by k matrix
0HGx 0 HG , so the rows of H must be orthogonalvectors to the columns of G(modulo 2)
111
G
110
,011
11 vv
110011
HHy is only zero forthe code words(0,0,0) and (1,1,1)
Error Correction Formalism
Error detection with parity matrix
Gxy eyy HeHeHyyH
Special case: (classical) Hamming code, a [7,4] code
101010111001101111000
HIf ej is an erroron the j’th bit
Hej is the binaryrepresentation of j
Error Correction Formalism
Generators (quantum)Stabilizers
},,,,,,,{1 iZZiYYiXXiIIG Pauli group:
Suppose S is a subgroup of Gn and define VS to be the set of n qubit states which are fixed by every element of S.
S is then said to be the stabilizer of the space VS.
Definition:
21100
EPR
EPR
XX 21
EPR
EPRZZ 21
EPR ?
?
Stabilizers: example
Error Correction Formalism
},,,{3
313221 ZZZZZZISn
21ZZ
111
110
001
000
32ZZ
111
011
100
000
111,000Common base (Vs):
221ZZI
322131 ZZZZZZ
3221 , ZZZZS
Z1Z2 Z2Z3tot
3-qubit flip code!
Generator (quantum):
The stabilizers that generate our logical qubits tellus how to measure the error syndrome!
Realization:
(not unique)
Check matrix (quantum parity)
Error Correction Formalism
101010111001101111000
H
101010111001101111000000000000000000000000
000000000000000000000101010111001101111000
H
Name Operator1g IIIXXXX2g IXXIIXX3g XIXIXIX4g IIIZZZZ5g IZZIIZZ
6g ZIZIZIZ
1101001011110010110100001111
1100110011001110101010000000810
L
0010110100001101001011110000
0011001100110001010101111111811
L
7 qubit Steane code:
Quantum case is bigger because errors are more complex, not only bit flips but also phase errors
Can be used to findthe generators:
Error Correction Measurement
Measure arbitrary operator M, as a controlled-M
Specific case: X as controlled-X
:temporary ancilla qubit
𝑍
𝑍
Z1Z2 Z2Z3
tot 3-qubit flip code!
Error Correction Measurement7 qubit Steane code (standard form):
Name Operator
𝑔 𝑍𝑍𝑍𝐼𝐼𝑍𝐼
𝑔 𝑋𝐼𝐼𝐼𝑋𝑋𝑋𝑔 𝐼𝑋𝐼𝑋𝐼𝑋𝑋
𝑔 𝐼𝑍𝑍𝐼𝑍𝐼𝑍𝑔 𝑍𝐼𝑍𝑍𝐼𝐼𝑍𝑔 𝐼𝐼𝑋𝑋𝑋𝑋𝐼
Error Correction Bounds
What is the smallest possible code that protects against any errors?
Quantum Singelton Bound (ch12):
tkn 4k is the original number of qubitsn is the encoded number of qubitst is the max number of qubit errors
141 nExample: 5 n
5-qubit code:
Error Correction BoundsFault-tolerant quantum computing:
Single error: p After block:Fails with cp2
• c depends on number of components• In how many ways can two components fail? c ~10 ways• Improvement if 𝑐𝑝 𝑝
Fault-tolerant threshold, important for experiments
→ 𝑝 ~10
Example: Surfaces codesSuperconducting qubits are designed two-dimensionally:
Gambetta et al, npj Quantum Information 3, 2 (2017)
Intel’s chip
• Fault tolerant error threshold: 1% but…
• Trade-off between error threshold and number of qubits
• Only nearest neighbor interactions
Example: Surfaces codes
Detailed walkthrough from Fowler et al, PRA 86, 032324 (2012)
How many qubits are required for a technologically relevant implementation of Shor’s algorithm?
Classical record for integer factorization:
• 768-bit RSA took 2 years running of several hundred machines• 1024-bit RSA would take a factor of 1000 longer
Quantum test:Find prime factors for a 2000-bit RSA number in ~1 day
Trade-off between space and time
Shor’s:
Example: Surfaces codes
Number of logical qubits: 2N = 4000
Total number of logical gates: 40N3
Ancilla qubits per (Toffoli) gate: 7
Total number of ancilla qubits: 7 ⋅ 40𝑁 280 ⋅ 4000 10
Final error per operation must be1
280𝑁 10
Estimating the total number of gates:
Fowler et al, PRA 86, 032324 (2012)
N = 2000
(Logical)
(Code is limited by error correction)
Example: Surfaces codes
Realistically:
For perfect gates: 13 physical qubits for 1 logical qubit
0.1% errors
2) 3600 physical qubits for 1 logical qubit
Operation times:Readout time: ~100 nsGate time: ~0.1 – 10 µs (Limited partly by classical post
processing)
Starting from 0.1% = 10-3 , two stages of error correction are needed:
Final error per logical operation must be1
280𝑁 10
Scheme can work with 1% physical qubit error
How much error correction is needed to get there?
1) 900 physical qubits for 1 logical qubit
Total footprint of 1 logical ancilla qubit: 200 000 physical qubits
(low error limit)
Example: Surfaces codesSumming up
→ ~1000 logical ancilla qubits needed in parallel:
1013 logical ancilla qubits are needed
Fast gate time ~100 ns → ~1010 logical ancilla qubits within a day
Total qubits needed:
For the error correction…qubits involved in calculation = 20 million
1000 logical ⋅ 200 000 physical/logical
= 200 million physical qubits
Fowler et al, PRA 86, 032324 (2012)
From before:
Quantum Error CorrectionSummary
• Difference with quantum error correction• No cloning• Continuous errors• Measurement collapse
• Simple case: bit flip and phase flips• Systematic treatment using stabilizers
• 7-qubit Steane code• Fault-tolerant bound: 𝑒~10• Technologically relevant algorithms require a
lot (200 million) qubits!
(Steane code)