Overview of quantum error correction

[Link] image, @quantshah

 GOAL OF ERROR CORRECTION
 To preserve messages sent through a noisy transmission channel by encoding the messages
in an error-correcting code
• More precisely: to make sure the rate of corruption of encoded (i.e., logical) information
is lower than that of the same information sent without the extra encoding step.
message
𝜌 → ℰ ncoder 𝒩 oise 𝒟ecoder → 𝒟 𝒩 ℰ ( 𝜌 ) ≈ 𝒟 ℰ ( 𝜌 )
space- or time-like
 Fault tolerance: to successfully apply
(reversible) operations on the message
within a noisy environment:
~
𝒟 𝒢 𝒩 ℰ ( 𝜌 ) =𝒟 𝒩 𝒢 ℰ (𝜌)
𝒟𝒢𝒩ℰ ( 𝜌 ) ≈ 𝒟 𝒢 ℰ(
~ −1
( 𝒩 =𝒢 𝒩 𝒢 )
 QEC: ORIGINS & GROWTH
Google Scholar search: "quantum error correction"
3000

2500
Peter Shor:
2000
 Quantum error-correcting codes (1995)
 Fault-tolerant syndrome measurement (1996) 1500
 Fault-tolerant universal quantum gates (1996)
 Using QEC to prove security of QKD (2000) 1000

500

0
005 006 007 008 009 010 011 012 013 014 015 016 017 018 019 020 021 022
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2

Alexei Yu. Kitaev: Other pioneers:

 Topological quantum codes (1996-2003)  Stabilizer codes (Gottesman + Calderbank, Rains, Shor, Sloane)
 Physically protected quantum computing (1997)  FT error correction (Shor, Steane, Knill)
 Computing with nonabelian anyons (1997)  QEC conditions (Knill, Laflamme)
 CSS-to-homology dictionary (1998)  Concatenated threshold theorem (Aharonov, Ben-Or)
 Magic state distillation (1999-2004)
 Majorana modes in quantum wires (2000) Some material from: J. Preskill, QEC 2017 talk
 INTRO TO QEC
(questions encouraged)
 BASICS OF ERROR CORRECTION: REPETITION
CODE
• Code: subspace of a quantum system that allows protection against some noise channel.
• Consider the repetition code, protecting against bit-flip noise ( acting on one qubit).

Codewords:
• Quantum information is encoded into the codespace via coefficients . A logical state is:

• Bit flips map code space to an error space. E.g., flipping the 2nd qubit via :

error words
• Since the two spaces can be distinguished without collapsing the logical information within
them, bit flips are detectable. This is true for any error taking information out of codespace.

Q.E.D.
 ONE ROUND OF QEC Measure the

𝑈
syndrome, not the
data!

¿𝜓 ⟩

Lookup table

for
|+1 ⟩
for

for
|+1 ⟩
for
• To resolve code and error spaces (error diagnosis), measure eigenvalues of commuting set of
observables (check operators; here, and with ) and apply recovery conditional on parity-
check eigenvalue (error syndrome). This is one round of correction:
1. Diagnose: measure error syndromes using ancillary qubits.
2. Decode: given a syndrome, determine which recovery to apply.
• Correction rounds generalize straightforwardly to other types of errors and other codes.
 MY FIRST QUANTUM CODE
• Repetition code fails vs. phase errors:
• Four-qubit code detects both bit and phase errors:

• Single bit flips are detectable due to spacing between computational basis states participating
in superpositions of each codeword:
1st qubit bit flip 2nd qubit bit flip

 Each error takes one out of codespace  detectable.

? Both errors map to same error space (degeneracy). Are they both correctable?
 Both errors are not correctable because they induce a different operation on the
codewords.
 Flipping qubit one is equivalent to first undergoing a logical bit-flip and then flipping
qubit two: .
 MY FIRST QUANTUM CODE
• Four-qubit code detects both bit and phase errors:

• Single phase flips are detectable due to each codeword itself being in superposition:
1st or 2nd qubit phase flip 3rd or 4th qubit bit flip

 All four phase flips take one out of codespace  detectable.

 1st & 2nd qubit phase flips map to same error space, induce different effect than 3rd & 4th
qubit phase flips  not correctable.

• Similar behavior happens to bit-phase flips .

All single-qubit Pauli errors are detectable, but not correctable.
 ERROR SPACE STRUCTURE Noise is continuous,
but measured errors
are discrete!

1. Check operator measurement collapses system

onto codespace or an error space. |0 𝐿 ⟩ ,|1𝐿 ⟩
2. Paulis are a basis for single-qubit operators
 General errors are detectable!
|0 𝑋 ⟩ ,|1 𝑋 ⟩

Example: Z-axis rotation:

|0 𝑍 ⟩ ,|1 𝑍 ⟩

The result superposition of error space collapses to one |0𝑌 ⟩ ,|1𝑌 ⟩

error space upon a round of EC:
SIMPLE CODES  IMPORTANT CODES
 FOUR-QUBIT CODE  STABILIZER CODE
 Recall four-qubit code:

 Recall observation of operator for which the codespace is a +1-eigenvalue eigen-subspace:

 Operators and satisfy this as well. The three mutually commuting operators generate the
code’s stabilizer group :

 Example: Four-qubit code is an code. [ [ 𝑛 , 𝑘, 𝑑 ] ]

Distance
 Advantages of stabilizer codes (over other codes): # logical qubits
# physical qubits
 Efficient presentation in terms of stabilizer generators.
 Syndromes obtained for free: group generators are check operators.
 General idea: works for bosons, fermions, modular qudits, Galios qudits, molecules.
[Link]
OUR-QUBIT CODE  SURFACE CODE
 Arrange the four qubits ( )
into a square and observe
geometrical pattern formed
by stabilizer generators:

𝑍𝑍𝐼𝐼 𝑋𝑋𝑋𝑋 𝐼𝐼𝑍𝑍

Geometries with holes Hyperbolic geometries Fractal geometries
arXiv:1111.4022 arXiv:1506.04029 arXiv:2201.03568
 Springboard to other geometries;
connection to topological phases.

Nontrivial boundaries
Exotic manifolds >=4D
 toric code 3D version
arXiv:math/0002124,
quant-ph/9707021 arXiv:1406.4227
Rotated surface codes: arXiv:1404.3747 arXiv:1310.5555
 BREAK EVEN & THRESHOLD THEOREM
 The primary goal of error correction is to yield a
probability of corruption of the encoded information
[The threshold is]
() that is smaller than a reference probability of everything and nothing.
unencoded information corruption ().

 A more ambitious goal in the quantum setting is made possible by the notion of an error-
correcting threshold (), below which it should be possible to construct error-correcting
protocols with arbitrarily large gain by scaling up the physical degrees of freedom

 Existing threshold theorems require infinite families of tensor-product codes (e.g., for ) as well
as the thermodynamic limit. Few-qubit/mode codes won’t cut it.
RECENT (AND NOT-SO-RECENT) HIGHLIGHTS
 HARDNESS OF DECODING
 We picked most likely error (maximum-
likelihood or ML decoding), based on the
assumption that low-weight Paulis are more
likely than high-weight Paulis.

 General stabilizer recovery consists of three parts: 𝑈=𝑇 ⋅ 𝐿⋅ 𝑆

 ML decoding: Which do we pick? Stabilizer group equivalence
Residual logical operations
Map back to codespace

 ML decoding generally NP-complete. arXiv:1009.1319; Kuo, Lu IEEE 2012 Surface: arXiv:1405.4883

 Can be efficient: e.g., surface code or some bosonic codes. Bosonic: arXiv:2209.04573

 Degenerate ML decoding solves for most likely error class  #P-complete. arXiv:1310.3235

 stat-mech mapping relates

maximization to partition
function of classical model.
arXiv:quant-ph/0110143
 STABILIZER EXPERIMENTS
Parameters Name Platforms
NMR (Waterloo), SC circuits (Google, IBM), silicon (RIKEN, Delft), NV centers
[[n,1]] Repetition (Wratchup, Kosaka, Hanson groups), ions (Blatt group)
Photonic (Rarity group), ions (IonQ), SC circuits (IBM, Google, Delft, Wallraff,
[[4,1,2]] variants Four-qubit Monz groups)
NMR (Waterloo), SC circuits (Pan group), ions (Quantinuum), NV centers
[[5,1,3]] Five-qubit perfect (Delft)
[[7,1,3]] Steane Ions (Blatt, Monz groups, Quantinuum), Rydberg arrays (Lukin group)
[[9,1,3]] Shor Ions (Linke group, IonQ), photonics (Pan group)
[[9,1,3]] Bacon-Shor subsystem Ions (IonQ)
[[m^2,m,3]], m=2,3 Heavy-Hexagon subsystem SC circuits (IBM)
[[m^2,m,3]], m<=7 Quantum Parity / Shor Ions (Linke group)
[[9,1,3]] Surface-17 SC circuits (Wallraff, Pan groups)
n=19 planar, 24 toric Kitaev surface Rydberg arrays (Lukin group)
n=9,25, d=3,5 planar XZZX surface SC circuits (Google)

Is our system good enough that

adding these extra qubits actually
improves logical performance?
For references, see:
[Link]
BOSONIC EXPERIMENTSCOMMERCIAL PROPOSAL
modes Name Platforms Evaluate ideas
Niset-Andersen- quickly and don’t
4 Cerf Photonics (Anderson group) get attached.
Lloyd-Slotine

ve
wa
9 analog Photonics (Furusawa group)

cro
Microwave cavities (Devoret,

mi
1 Binomial Sun groups) arXiv:2012.04108, 2103.06994

ic
[Link]

on
Microwave cavities (Devoret,

on
Leghtas, Wang groups, Alice & Cat/GKP + rep-n/surface

ph
arXiv:1907.11729

ve
1 Two-legged cat Bob)

wa
arXiv:2204.09128
Microwave cavities (Schoelkopf
Cat + repetition

cro
1 Four-legged cat group; break-even QEC)

mi
arXiv:2010.02905
Trapped ions (Home group), [Link]

al
Microwave cavities (Devoret

tic
GKP + cluster states

op
1 GKP group)

FBQC arXiv:2101.09310,
Interleaving arXiv:2103.08612,
[Link]

al
For references, see: Dual-rail + fusion-based QC
tic
op
[Link]
 PAST QEC CONFERENCES
QEC 2017 QEC 2019
Yoder, Jones, Litinski, Browne, Terhal, Litinski, Takita, Chubb, Tuckett, B. Brown,
Topological codes 9 Kubica, Beverland, B. Brown, Krishna 6 Kubica
General fault Flammia, Gottesman, Bravyi, Haah,
tolerance 8 Fowler, Campbell, Poulin, Chao 3 Beverland, N. Brown, Babbush
Taminlau (NV centers), Martinis (SC
circuits), Paik (SC circuits), Schoelkopf
Experiments & (SC circuits), Linke (ions), Marinelli (ions),
platforms 8 Gutierrez (ions), Flühmann (ions) 3 Home (ions), Cai (silicon), Boter (silicon)
Panteleev, Fawzi, Krishna, Breuckmann,
QLDPC 0 9 Vuillot, Debroy, Crosson, Pryadko, Groues
Bosonic/CV coding 1 Niu 5 Puri, Albert, Heinze, Noh, Pantaleoni
Holography & QEC 4 Harlow, Yoshida, Pastawski, Swingle 1 Cubbitt
General coding 2 Kastoryano, Nezami 1 Nezami
Circuits/algorithms 0 3 Koenig, Kim, Benjamin
QEC for sensing 1 Cappellaro 1 Jiang
AQC 1 Lidar 1 Marvian
Quantum control 0 1 Biercuk
Overview 1 Preskill 0

Some material from: J. Preskill, QEC 2017 talk

 PAST QEC CONFERENCES
QEC 2017 QEC 2019
Yoder, Jones, Litinski, Browne, Terhal, Litinski, Takita, Chubb, Tuckett, B. Brown,
Topological codes 9 Kubica, Beverland, B. Brown, Krishna 6 Kubica
General fault Flammia, Gottesman, Bravyi, Haah,
tolerance 8 Fowler, Campbell, Poulin, Chao 3 Beverland, N. Brown, Babbush
Taminlau (NV centers), Martinis (SC
circuits), Paik (SC circuits), Schoelkopf
Experiments & (SC circuits), Linke (ions), Marinelli (ions),
platforms 8 Gutierrez (ions), Flühmann (ions) 3 Home (ions), Cai (silicon), Boter (silicon)
Panteleev, Fawzi, Krishna, Breuckmann,
QLDPC 0 9 Vuillot, Debroy, Crosson, Pryadko, Groues
Bosonic/CV coding 1 Niu 5 Puri, Albert, Heinze, Noh, Pantaleoni
Holography & QEC 4 Harlow, Yoshida, Pastawski, Swingle 1 Cubbitt
General coding 2 Kastoryano, Nezami 1 Nezami
Circuits/algorithms 0 3 Koenig, Kim, Benjamin
QEC for sensing 1 Cappellaro 1 Jiang
AQC 1 Lidar 1 Marvian
Quantum control 0 1 Biercuk
Overview 1 Preskill 0

Some material from: J. Preskill, QEC 2017 talk

SURFACE CODE:
 Kitaev 1998-2003: definitions
NEW
Key
FRIENDS
papers:  Bravyi, Kitaev 1998: boundaries
 Dennis et al 2002: decoding, topological
threshold, stat-mech mapping
 Bombin 2010: twist-defect encoding
 Fowler et al 2012: defect encoding, resource est.
 Horsman et al 2012: lattice surgery

XY, XZZX, etc

For references, see:

[Link]
 OTHER NOTABLE CODES
 Holographic codes: encoding circuit is a tensor network in a hyperbolic geometry.
[Link]

 QEC for sensing: separating noise from signal of interest within a codespace.
[Link]
[Link]

 Subsystem codes: use only some of the logical qubits for storage, set the rest to be malleable
gauge qubits. [Link]

 Floquet codes: logical qubits are generated through a particular sequence of check-operator
measurements. [Link]

 Generalized homological product codes: chain-complex math taken to the extreme in the
race to good QLDPC. [Link]

 Magic-state distillation codes: particular properties useful for magic-state factories.

[Link]
 OBSERVATIONS & OPPORTUNITIES
Errors don’t all go to
 Beginning to scale up and see favorable scaling hell [as we scale up].
 exponential decrease with repetition codes [Fancy papers] are nice, but what is
 nonzero decrease with quantum codes going to keep everyone employed
 Noise remains local is making the qubits better.
 Daunting systems engineering / control issues require industrial efforts.
 Companies not upfront about much money/people-power has been invested.
 Once a roadblock is overcome, others can easily replicate (I am told…).
 Graduate career may no longer be enough to single-handedly make progress.

 Characterizing physical noise in materials.

‼ Low-lying fruit can only last us so long before we hit operational ceiling.
 Threshold and associated scaling from experimentally observed data.
 For quantum memory or, more ambitiously, logical gates.
 Interpolating between QLDPC and geometrically local codes.
 Tailoring QEC and fault-tolerant protocols to experiments.