equivalence of 2d color codes (without translational ...abhagoji/files/isit_presentation_v2.pdf ·...
TRANSCRIPT
Equivalence of 2D color codes (without translationalsymmetry) to surface codes
Arjun Bhagoji1 Pradeep Sarvepalli1
1Department of Electrical EngineeringIndian Institute of Technology, Madras
IEEE International Symposium on Information Theory 2015, HongKong
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 1 / 20
Motivation
Surface codes, a class of topological codes, have many advantages(local stabilizer generators, low complexity decoders etc.)
Limited in terms of transversal gates that can be implemented
Color codes, another class of topological codes, can implement entireClifford group transversally
So, they are inequivalent?No!
Bombin, Duclos-Cianci and Poulin (2012) showed that translationallyinvariant 2D color codes can be mapped to a finite number of copies of
Kitaev’s toric code.Question: What about translation variant codes?
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 2 / 20
Surface codes
Topological codes are a class of quantum codes where information isstored in topological degrees of freedomSurface codes are topological codes with qubits attached to the edgesof a lattice embedded in a given closed surfaceEach face (plaquette) and vertex of the lattice defines a stabilizeracting on the surrounding qubits
XX
XX
Z
Z
Z Z
E1
E2
Figure: Left: Stabilizers on a copy of Kitaev’s toric code. Right: E1 is a homologicallytrivial error, can be detected and corrected. E2 is homologically non-trivial, cannot bedetected, interacts with the encoded qubits.
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 3 / 20
Color codes
Different class of topological codes, embedded on 2-colexes (trivalent,3-face-colorable complexes)Qubits attached to every vertexEach face of the lattice defines a stabilizer acting on the qubits in itsboundary
X
X
X
X
X
X
Z
ZZ
Z
Z
Z
Figure: Hexagonal color code
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 4 / 20
Charges and syndromes
Think of each violated check (stabilizer anti-commuting with error) asa charge
4 types of charges on a surface code: i) electric (ε) on vertices, ii)magnetic (µ) on the plaquettes, iii) composite electric and magneticcharge (εµ) on both plaquettes and vertices, and iv) the vacuum (ι)
Only two of these are independent, which we choose to be ε (due toZ -type) and µ (due to X -type)
Color code charges also have color associated with them.
16 types of charges on a color code (Bombin et.al., 2012) with 2independent pairs
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 5 / 20
Hopping operators
Any element of the Pauli group that moves chargesStabilizers can be viewed as a combination of hopping operatorsmoving charges back to their initial position
Figure: Hopping operators on the surface code (left) and color code (right)
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 6 / 20
Mapping between color codes and surface codes
Observation:1 4 types of charges on a surface code and 16 types of charges on a color
code.2 2 pairs of independent charges on the color code
⇒ Decompose the color code into 2 surface codes, mapping eachcharge pair from the color code to one of the surface codes
Electric charges on the surface codes live on the vertices while themagnetic charges live on the plaquettes.
Consider the charges on the color code Γ: all live on plaquettes
How do we relate the two?
Notation: For what follows, we denote the faces, edges and vertices of the embedding of a graph Γ by F(Γ), E(Γ) and V(Γ).Bσ
f =∏
v∈fσv denotes the stabilizers on the color code, where σ ∈ X , Z .
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 7 / 20
Desirable constraints on map
The map π between the color code and the surface codes should have thefollowing desirable properties:
1 Linear2 Invertible (i.e. map should be bijective)3 Local4 Efficiently computable5 Preserve the commutation relations between the Pauli error operators
on V(Γ), i.e. PV(Γ)6 Consistent in the description of the movement of charges on the color
code and surface codes.
Focus will be on the consistency of charge movement and the preservationof commutation relations. Time permitting, other constraints will bediscussed.
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 8 / 20
Surface codes obtained from color codes
Contract all the c-colored plaquettes in the embedding of Γ. This will giverise to a new graph τc(Γ).
Figure: Illustrating the contraction of a color code via τc and the resultant surface code.Only portions of the codes are shown.
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 9 / 20
Charge Mapping
Each c-colored face in Γ can host εc and µc .Both these charges live on the vertices of a contracted latticeIdentify εc ≡ ε1 and µc ≡ ε2.
Now choose the magnetic charges on Γ1 and Γ2 from εc′ , εc′′ and µc′ , µc′′ .How?
Lemma (Charge mapping)
Let c, c ′, c ′′ be three distinct colors. Then, {εc , µc′} and {εc′ , µc} arepermissible pairings of the charges so that the color code on Γ can bemapped to a pair of surface codes on Γi = τc(Γ). In other words, ε1 ≡ εc ,µ1 ≡ µc′ , ε2 ≡ µc and µ2 ≡ εc′ , where εi and µi are the electric andmagnetic charges of the surface code on Γi .
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 10 / 20
Elementary hopping operators (electric)
Figure: Mapping the electric charge hopping operators from the color code to thesurface codes
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 11 / 20
Mapping elementary hopping operators
Lemma (Elementary hopping operators)
Let f , f ′ ∈ Fc(Γ) where the edge (u, v) is incident on f and f ′. Then, thefollowing choices reflect the charge movement on Γ onto the surface codeson Γi .
π(Hεcu,v ) =
[Zτ(u)
]1
=[Zτ(v)
]1
(1)
π(Hµcu,v ) =
[Zτ(u)
]2
=[Zτ(v)
]2, (2)
where [T ]i indicates the instance of the surface code on which T acts.Now if f , f ′ ∈ Fc′(Γ) and (u, v) ∈ Ec′(Γ) such that u ∈ f and v ∈ f ′ andHεc′
u,v and Hµc′u,v are chosen to be independent hopping operators of f , then
π(Hεc′u,v ) =
[Xτ(u)Xτ(v)
]2
; π(Hµc′u,v ) =
[Xτ(u)Xτ(v)
]1. (3)
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 12 / 20
Dependent hopping operators
Not all the hopping operators on a particular face are independent.
Lemma (Dependent hopping operators)Let f ∈ Fc′′(Γ) and 1, . . . , 2`f be the vertices in its boundary so that(2i − 1, 2i) ∈ Ec(Γ), (2i , 2i + 1) ∈ Ec′(Γ) for 1 ≤ i ≤ `f and 2`f + 1 ≡ 1.If π is invertible, then π(Bσ
f ) 6= I and there are 4`f − 2 independentelementary hopping operators along the edges of f .
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 13 / 20
Mapping the single qubit errorsLet the operators Hεc′
1,2`f and Hµc′2m,2m+1 be the dependent ones.
Z1Z2Z3Z4
...Z2`−1Z2`
X1X2X3X4
...X2`−1X2`
π7→
ZZ
. . .Z
ZZ
. . .Z
1∪2
(4)
Z2Z3Z4Z5
...Z2`−2Z2`−1
π7→
X X
X X. . .
X X
2
(5)
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 14 / 20
Single qubit error splitting
X2X3...
X2m−2X2m−1X2m+2X2m+3
...X2`X1
π7→
X X. . .X X
X X. . .
X X
1
(6)
Lemma (Splitting)
The following choices lead to an invertible π while respecting thecommutation relations with the independent hopping operators.
π(gX1) = [Xτ(1)]1 where g ∈ {I,BXf ,BY
f ,BZf } (7)
π(gZ2m) = [Xτ(2m)]2 where g ∈ {I,BXf } (8)
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 15 / 20
Projection theorem
Theorem
Any 2D color code (on a 2-colex Γ without parallel edges) is equivalent toa pair of surface codes τ(Γ) under the map π.
Mapping can be done locally for each c ′′-colored face
Stabilizers on the color code are mapped to stabilizers on the surfacecode, so code capabilities are preserved
Decoding of color codes is a possible application
Mapped errors can be decoded on the surface code, and uniquelylifted back to the color code, as the map is bijective
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 16 / 20
Map for X -type errors on the hexagonal color code
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 17 / 20
Map for X -type errors on the hexagonal color code
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 18 / 20
Comparison with previous work
Authors Translationinvariancereqd.
No. ofcopies
Ancillaqubitsreqd.
Injective Conseq. fordecoding
Bombinet. al.
Yes Finite,could bemany
Sometimes Yes High com-plexity
Delfosseet. al.
No 3 Not sure Yes No pre-image forsome errors
Presentwork
No 2 No Yes Non-CSSnature, upto 2 de-coders persurface code
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 19 / 20
That’s all folks!
Bhagoji, Sarvepalli Equivalence of 2D color codes (without translational symmetry) to surface codes ISIT 2015 20 / 20