Quantum algorithms

Lecture 4: Quantum Error Correction

Benoı̂t Vermersch
October 12, 2023
LPMMC Grenoble

1
Outline

What is an error in quantum computing?

The bit flip code

The Steane code

The surface code and fault-tolerance

2
Outline

What is an error in quantum computing?

The bit flip code

The Steane code

The surface code and fault-tolerance

3
An error in a quantum computer?

• Example: Spontaneous emission with an atomic qubit |ψi = |1i

p √
|1i → 1 − p |1i |0iphoton + p |0i |1iphoton (1)

• Spontaneous emission process corresponds to a ‘bitflip error’ |ψi → X |ψi

|ψi → |ψi |E iI + X |ψi |E iX (2)

4
An error in a quantum computer?

• For a general qubit state |ψi = (α |0i + β |1i), a decoherence process can always
be interpretated as a sum of ‘Pauli Errors’:

|ψi → |ψi |E iI + X |ψi |E iX + Y |ψi |E iY + Z |ψi |E iZ (3)

• Quantum error correction: How to detect an error without destroying the

quantum superposition?

5
Outline

What is an error in quantum computing?

The bit flip code

The Steane code

The surface code and fault-tolerance

6
The bit flip code

• Our first code: The bit flip code

|ψi = α |0iL + β |1iL (4)

with a logical qubit that is made of three physical qubits

|0iL = |000i |1iL = |111i (5)

• The code aims at tracking and correcting X errors occurring on one of the three
physical qubits
X
|ψi → |ψi |E iI + Xi |ψi |E iXi →QEC |ψi (6)
i=1,2,3

7
The bit flip code

• There are two mesurements to be made hZ1 Z2 i, hZ2 Z3 i, giving rise to unique
error syndromes, independently of the qubit superposition state.
Error State hZ1 Z2 i, hZ2 Z3 i
none α |000i + β |111i 1,1
X1 α |100i + β |011i -1,1
X2 α |010i + β |101i -1,-1
X3 α |001i + β |110i 1,-1
• Code distance: Number of errors that map one logical state to the other. Here
it’s d = 3. For a general d, we can correct t errors if d ≥ 2t + 1.
• How to measure and correct errors?

8
The bit flip code: Collective measurements

• We require a collective measurement of hZ1 Z2 i with two measurement outcomes

(eigenvalues)  = ±1:

Z1 Z2 = |00i h00| + |11i h11| − (|01i h01| + |10i h10|)

| {z } | {z }
P1 P−1

• A measurement on |ψ 0 i gives a mesurement outcome  and a projection

|ψ 0 i → P |ψ 0 i with probability hψ|P |ψi

• If |ψ 0 i is proportional to |ψi , X1 |ψi , X2 |ψi, we obtain a deterministic

measurement  = 1, or  = −1, and the state is unchanged.
• For a quantum superposition of errors, the outcome is probabilitic, but the
post-measured state is compatible with such outcome.
9
The bit flip code: Collective measurements

• Example with a bitflip process on qubit 3, |ψ 0 i = X3 |ψi |E iX

• Measurement of Z1 Z2 : Projection onto the same state with |ψ 0 i with outcome 1.
• Measurement of Z2 Z3 : I measure outcome −1. I can perform error recovery by
applying X3 and obtain X32 |ψi = |ψi
• Note: With a ‘naive’ non-collective measurement sequence of Z1 , Z2 , Z3 , I would
always project the state |ψ 0 i in a classical state, such as |000i and destroy
quantum superposition α |000i + β |111i.

10
The bit flip code: Implementation aspects

• Step 1: Encoding from a physical qubit state |ψ1 i = α |0i + β |1i:

|ψ1 i

|0i |ψi = α |000i + β |111i

|0i

• Side remark: This is very different from quantum cloning |ψi → |ψi⊗3 , which can
be proven to be strictly impossible.

11
The bit flip code: Implementation aspects

• Step 2: Error syndromes and recoveries: One requires ancilla qubits (see also
Exercices 4)

|ψ1 i

|0i Gates with errors

|0i

|0i → Z1 Z2

|0i → Z2 Z3

• The logical gates XL = |0iL h1| + h.c = X1 X2 X3 , ZL = |0iL h0| − |1iL h1| = Z1 , 12
The bit flip code: Limitations

• The bit flip code fails for two and three qubit bit flip errors with probability

pL = 3p 2 (1 − p) + p 3 (7)

with p the single qubit error

• Notion of threshold: Quantum error correction is only useful when the logical
qubit lifetime is larger than the physical qubit lifetime, i.e when pL ≤ p, this
means when p ≤ 1/2.
• What about combined presence of X , Y , Z errors?

13
Outline

What is an error in quantum computing?

The bit flip code

The Steane code

The surface code and fault-tolerance

14
Steane code

• One logical qubit made of seven physical qubits.

• The error syndromes are defined as the set
S = {Z4 Z5 Z6 Z7 , Z2 Z3 Z6 Z7 , Z1 Z3 Z5 Z7 , X4 X5 X6 X7 , X2 X3 X6 X7 , X1 X3 X5 X7 }.
• These operators commute, i.e errors can be measured successively
• The ‘code world’ (distance d = 3)
√
|0iL = 1/ 8 (|0000000i + |1010101i + |0110011i + |1100110i
+ |0001111i + |1011010i + |0111100i + |1101001i)
|1iL = X1 X2 X3 |0iL (8)
• The code is ‘stabilized’ by S: For any |ψi = α |0iL + β |1iL , for any g ∈ S,
g |ψi = |ψi.
• The logical gates are XL = i Xi , ZL = i Zi
Q Q

15
Steane code

• The Steane code is an example of stabilizer codes, whose error syndromes are
elements of a commuting Pauli subgroup.
• For the purpose of this lecture, we will simply check that the syndromes do the
job.
• General rules:
• If Zi is present in an error syndrome g , it will detect Xi errors (because
Xi Zi Xi = −Zi , and operators acting on different sites i, j commute.)
• Similarly, Zi errors are detected by Xi operators .
• Y = iXZ , therefore a Y error is a Z error followed by an X error.

16
Steane code

Error Z4 Z5 Z6 Z7 Z2 Z3 Z6 Z7 Z1 Z3 Z5 Z7 X4 X5 X6 X7 X2 X3 X6 X7 X1 X3 X5 X7
none 1 1 1 1 1 1
X1 1 1 -1 1 1 1
X2 1 -1 1 1 1 1
X3 1 -1 -1 1 1 1
X4 -1 1 -1 1 1 1
X5 -1 1 -1 1 1 1
X6 -1 -1 1 1 1 1
X7 -1 -1 -1 1 1 1
Z1 1 1 1 1 1 -1
..
.
Y1 1 1 -1 1 1 -1
..
. 17
Steane code: Conclusion

• The Steane corrects any single qubit errors.

• As the bitflip code, it does not corrected double errors (ex: X1 X2 ).
• A first option to achieve Fault tolerance (reaching arbitrary precision in presence
of a finite error probability): Concatenated Steane Codes.
• Another approach: Surface codes.

18
Outline

What is an error in quantum computing?

The bit flip code

The Steane code

The surface code and fault-tolerance

19
Surface code

• Kitaev, Bravyi (1997), following works

on ‘Toric codes’.
Z Z Z Z Z

X X X X
• The physical qubits sit on a 2D lattice.
Z Z Z Z Z • The stabilizer operators, i.e the
X X X X measurements to be made for error
Z Z Z Z Z
detection, are
X X X X

Z Z Z Z Z Zi1 Zi2 Zi3 Zi4 on plaquettes

X X X X
Xj1 Xj2 Xj3 Xj4 on vertices
Z Z Z Z Z

• Code world is ‘stabilized’ by all such

operators g |ψi = |ψi

20
Surface code: Error detection

• A green qubit indicates an X error.

Z Z Z Z Z
• A pink qubit indicates an Z error.
X X X X

Z
−1
Z Z Z Z
• This is detected by the neighboring
X X X X
plaquettes.
−1
Z Z Z Z Z • Any error can be detected provided
X X X X the lattice is sufficiently large.
Z Z Z Z Z
−1 −1 • One can then apply recovery
X X X X
operations (or adapt in the software
Z Z Z Z Z
the definition of the code with
g → −g )

21
Surface code: Error detection

Z Z Z Z Z

X X X X
−1 • Example inspired from Google’s
Z Z Z Z Z

X X X X
demonstration of the toric code
Z Z
−1
Z Z Z
[Science 374, 1237-1241 (2021)]
X X X X • I observe stabilizer expectations
Z Z Z Z Z g = −1 as indicated here. What is the
X X X X
error recovery operation?
Z Z Z Z Z

22
Surface code: Error detection

• Apparent ambiguity: I have two

‘explanations’ (X1 X2 or X3 X4 )
Z Z Z Z Z • However these are fixed by the same
X X X X recovery operation
−1
Z Z 3 Z Z Z • Assume eg the error that happened is
X 1 X 4
−1
X X
|ψ 0 i = X3 X4 |ψi, but you apply the
Z Z 2 Z Z Z
‘other’ recovery operation X1 X2 , you
X X X X
still obtain the right state
Z Z Z Z Z

X X X X
X1 X2 X3 X4 |ψi = |ψi (9)
Z Z Z Z Z

• All these properties can be checked

using the stabilizer formalism (eg
Preskill notes). 23
Surface code: Single qubit initialization and quantum logic

• Probabilistic initialization: Take a

random state. Measure stabilizers.
Z Z Z Z Z Perform error recovery. I obtain a
X X X X random state in the code world |ψi.
• Let ZL = j Zi=L/2,j . We can prepare
Z Z Z Z Z
Q
X X X X
our logical |0L i by measuring ZL , such
Z Z Z Z Z XL that ZL |0iL = |0iL (Note that ZL
X X X X
commutes with all stabilizer
Z Z Z Z Z
operators).
X X X X

• Let XL = i Xi,j=L/2 , and we define

Q
Z Z Z Z Z

ZL our logical |1iL = XL |0iL . Exercice:

prove that |1iL also belongs to the
code world and that h0L |1L i = 0. 24
Notion of fault tolerance

• ‘Macroscopic errors’ cannot be detected with a surface code (eg an error of the
type XL , see also Exercices 4)
• Possible fix: increase the code size. But, if I add more and more noisy qubits, I
also increase the probabilities of individual errors . . .
• Quantum threshold theorem [Knill,Laflamme,Zurek, Aharonov,Ben-or,Kitaev], we
can achieve arbitrary precision small  on arbitrary quantum circuits with quantum
error correction, provided the physical qubit error probability is below a threshold
p < pth .
• If p < pth , adding more qubits help in achievieng quantum error correction.

25
Fault tolerance in the surface code

• Exercises 4: We will roughly estimate

the logical error probability pL versus
p, for different system sizes d.
• Optimistic scenario for practical
applications: More than 1000 physical
required for implemented a single
logical qubit. We are not there yet. . .
• Recent highlights
• Google’s toric code
• Fault tolerant two qubit gates
(Postler et al, Nature 2022)

26