Synthesis of Logical Clifford Operators via Symplectic Geometry

Narayanan Rengaswamy , Robert Calderbank , Swanand Kadhe , and Henry D. Pfister

† † ∗ †

†
Information Initiative at Duke (iiD), Department of Electrical and Computer Engineering, Duke University, Durham, NC, USA
∗
Department of Electrical Engineering and Computer Science, University of California, Berkeley, CA, USA

Motivation and Contribution Stabilizer Codes and Logical Pauli Operators

• k-dimensional Stabilizer: commutative subgroup S ⊂ HWN generated by linearly independent Hermitian operators
• Fault-tolerance: Given a quantum error-correcting code (QECC), if a quantum operation T
ab
E(aj , bj ) , ι D(aj , bj ), j = 1, . . . , k.
is performed on an encoded block of qubits, and a single component of the circuit fails, then the
number of errors in the output state should be within the error-correcting capacity of the code. • [ m, m − k, d]] Stabilizer Code: The 2m−k dimensional
n subspace V (S) jointly fixed
o by all elements of the stabilizer S,
i.e., V (S) , |ψi ∈ CN : g |ψi = |ψi ∀ g ∈ S .
• Part of the goal: For a chosen code, determine the circuits that realize non-trivial operations on
the logical qubits. These physical circuits are called the logical operators for the code. • The [ 6, 4, 2]] CSS Code: S , hg X , X ⊗6 = E(111111, 000000), g Z , Z ⊗6 = E(000000, 111111)i.

• Many works have concentrated on constructing codes with good properties and also on optimizing • CSS Construction: Let C be the [6, 5, 2] single-parity check code (m = 6). The dual C ⊥ ⊂ C is the [6, 1, 6] repetition
a given circuit for complexity or fault-tolerance, with respect to a chosen gate set. code with generator GC ⊥ = HC = [1 1 1 1 1 1]. Two possible generator matrices for the coset space C/C ⊥ are:
       0
1 1 0 0 0 0 h1 0 1 0 0 0 1 h1
• We provide a systematic and efficient algorithm for synthesizing logical Clifford operators on
1 0 1 0 0 0 h2 0 0 1 0 0 1 h0 
stabilizer codes. We also reveal the exact degeneracy in realizing these encoded operations. Our GC/C ⊥ = 
X  =:   or G ⊥ = 
1 0 0 1 0 0 h3
Z
C/C
 =:  20  .
0 0 0 1 0 1 h  (5)
enumeration of all valid circuits can be useful in a compiler choosing codes even dynamically. 3
1 0 0 0 1 0 h4 0 0 0 0 1 1 h04
unreliable circuit • So if we have a 4-qubit logical state |xiL then the CSS code will encode this into the physical state
 +
 E X  E X   X4
 1  1
initial arbitrary logical operation g L desired |ψxi ≡ v + C ⊥ , q c + x · GX ⊥ = √ c +
 xj hj . (6)
|xiL |x̃iL C/C 2
final |C ⊥| c∈C ⊥ c∈C ⊥

logical j=1
state state • For the [ 6, 4, 2]] CSS code the logical Pauli operators are: X̄j = D(hj , 0) = X1Xj+1, Z̄j = D(0, h0j ) = Zj+1Z6.
Need to
translate
QECC for given QECC Synthesis of Logical Clifford Operators for Stabilizer Codes
encode QECC decode
• Conditions on ḡ: ḡ X̄j ḡ † = h̄ if g LXjL(g L)† = hL ∈ HW2m−k and ḡ Z̄j ḡ † = h̄0 if g LZjL(g L)† = (h0)L ∈ HW2m−k .

relevant physical operation ḡ • Synthesizing g L = CZL 12 for the [ 6, 4, 2]] CSS code: Find physical operator ḡ = CZ12 that normalizes S and satisfies
|ψx i |ψx̃ i 

X̄1Z̄2 if j = 1,
† †
CZ12X̄j CZ12 , Z̄1X̄2 if j = 2, , CZ12Z̄j CZ12 , Z̄j ∀ j = 1, 2, 3, 4. (7)
We do this for logical Clifford operations on stabilizer QECCs 

X̄j 6 1, 2
if j =
Our algorithms, along with more utilities, are available open-source at: • Using the symplectic representation translate these into constraints on the desired symplectic matrix for CZ12:
https://github.com/nrenga/symplectic-arxiv18a † CZ12 γ,φ
CZ12X̄1CZ12 = X̄1Z̄2 ⇒ X̄1 = X1X2 7−→ X1X2Z3Z6 ⇐⇒ [110000, 000000]FCZ = [110000, 001001]
12
† CZ12 γ,φ
CZ12X̄2CZ12 = Z̄1X̄2 ⇒ X̄2 = X1X3 7−→ X1X3Z2Z6 ⇐⇒ [101000, 000000]FCZ = [101000, 010001]
12
.. .. .. .. ..
Logical Clifford Our All physical circuits ḡ
Operator g L Algorithm that realize g L & fix S X † X ⊗6 CZ12 ⊗6 γ,φ
CZ12 g CZ12 = g ⇒ X 7−→ X = X1X2 · · · X6 ⇐⇒ [111111, 000000]FCZ = [111111, 000000]
12
Z † Z ⊗6 CZ12 ⊗6 γ,φ
CZ12 g CZ12 = g ⇒ Z 7−→ Z = Z1Z2 · · · Z6 ⇐⇒ [000000, 111111]FCZ = [000000, 111111].
12
 
Stabilizer S Logical Paulis X̄i , Z̄i One possible solution 0 0 0 0 0 0
(defines the code) (Got97, Wil09) 0 0 1 0 0 1
    2
I6 B 0 1 0 0 0 1
Figure 1: (top) Problem of Encoded Computation. (bottom) An abstract representation of our contribution. ⇒ FCZ12 = ,B = 
0
 ←→ 3
0 I6  0 0 0 0 0
0 0 0 0 0 0 6 Z
⇒

Heisenberg-Weyl Group and Symplectic Vector Spaces   0 1 1 0 0 0

vBv T
CZ12 = diag ι Z6
• The single qubit Pauli or Heisenberg-Weyl operators are given by Not captured in FCZ12 – added to fix signs
        = CZ36 CZ26 CZ23 Z6
1 0 0 1 1 0 0 −ı √
I2 , , X, , Z, , Y , ı · XZ = ; ı , −1. (1)
0 1 1 0 0 −1 ı 0 • We solve such symplectic systems of linear equations using binary symplectic transvections.
• Bit-flip (X |vi = |v ⊕ 1i) and phase-flip (Z |vi = (−1)v |vi) anti-commute: XZ = −ZX. • Definition: Given a row vector h ∈ F2m , the corresponding symplectic transvection Z : F2m → F2m is defined as
2 h 2 2

m-qubit Pauli (or) Heisenberg-Weyl Group HWN (N = 2m): Operators ικD(a, b), where Zh(x) , x + hx, hish ⇔ Fh , I2m + ΩhT h ∈ Sp(2m, F2). (8)
D(a, b) , X a1 Z b1 ⊗ X a2 Z b2 ⊗ · · · ⊗ X am Z bm ∈ U2m , (2)
Our Generic Algorithm
2 , κ ∈ {0, 1, 2, 3} and UN is the unitary group.
a = (a1, . . . , am), b = (b1, . . . , bm) ∈ Fm
1 Determine the target ḡ by specifying its action on X̄i, Z̄i: ḡ X̄iḡ † = X̄i0, ḡ Z̄iḡ † = Z̄i0. Add conditions to normalize or centralize S.
• Example: D(a, b) |vi = (−1) vbT
|v + ai ⇒ D(11010, 10110) |10101i = |01111i. 2 Using the maps γ, φ, transform these relations into linear equations on Fḡ ∈ Sp(2m, F2), i.e., γ(X̄i)F = γ(X̄i0), γ(Z̄i)F = γ(Z̄i0). Add
(XZ ⊗ X ⊗ Z ⊗ XZ ⊗ I2) |10101i = XZ |1i ⊗ X |0i ⊗ Z |1i ⊗ XZ |0i ⊗ I2 |1i = |01111i. the conditions for normalizing the stabilizer S, i.e., γ(S)F = γ(S 0).
• Symplectic Inner Product: For row vectors [a, b], [a0, b0] ∈ F2m Find the feasible symplectic solution set Fḡ using symplectic transvections and “nullspace-like” properties of symplectic matrices.
2 , define
3
  4 Factor each F ∈ F into a product of elementary symplectic transformations, possibly using the algorithm given in [Can17], and
0 Im compute the physical Clifford operator ḡ.
h[a, b], [a0, b0]is , a0bT + b0aT = [a, b] Ω [a0, b0]T (mod 2), where Ω = . (3)
Im 0 5 Check for conjugation of ḡ with S, X̄i, Z̄i. If some signs are incorrect, post-multiply by an element from HWN as necessary to satisfy
0 0 h[a,b],[a 0,b0]i
s D(a0, b0)D(a, b) ⇒ commute iff h[a, b], [a0, b0]i = 0. these conditions (apply [NC10, Prop. 10.4] to S ⊥ = hS, X̄i, Z̄ii). Note that every Pauli operator in HWN induces the symplectic
• D(a, b)D(a , b ) = (−1) s transformation I2m, since HWN is the kernel of the map φ, so post-multiplication does not change the target symplectic matrix F .
6 Express ḡ as a sequence of Clifford gates, obtained from the factorization in step 4, which yields the desired physical circuit.
2 defined as γ(D(a, b)) , [a, b].
Isomorphism γ : HWN /hικIN i → F2m

Clifford Group and Symplectic Matrices 1 H H X 1 H H

2 H 2 H H
CliffN , NUN (HWN ): all g ∈ UN s.t. gHWN g † = HWN (normalizer of HWN in UN ). 6 Z 6 H
Gate Unitary Matrix Action on Paulis Figure 2: Logical Hadamard operator H̄1, synthesized by Chao and Reichardt [CR17] (left), and using our generic algorithm (right). This
illustrates that, while our algorithm yields all symplectic solutions for the desired logical operator ḡ, the decomposition we use from [Can17]
  may not yield lowest circuit complexity or fault-tolerance. Hence, our circuits can potentially be further optimized for such purposes.
1 1 1 HXH † = Z
Hadamard H, √
2 1 −1 HZH † = X Summary of Our Technical Results
 
1 0 P XP † = Y • For an [ m, m − k]] stabilizer code, the number of symplectic solutions for each logical Clifford operator is 2k(k+1)/2. Our generic
Phase P , algorithm above details the steps to determine all solutions and their circuits, using a particular decomposition of symplectic matrices.
0 ı P ZP † = Z
• For an [ m, m − k]] stabilizer code with stabilizer S, each physical realization of a given logical Clifford operator that normalizes S can
  be converted into a circuit that centralizes S, i.e., commutes with every element of S, while realizing the same logical operation.
I2 0 †
Controlled-NOT CNOT1→2 , CNOT1→2(X ⊗ I2)CNOT1→2 = X ⊗ X = X1X2 • Given a sequence of binary vectors xi, yi, i = 1, . . . , t ≤ 2m s.t. hxi, xj is = hyi, yj is, there exists a symplectic matrix F , expressible as
0 X
a product of at most 2t transvections, s.t. xiF = yi. We also given an explicit algorithm to compute such a matrix.
 
I2 0 † • Let {(ua, va), a ∈ {1, . . . , m}} be a collection of pairs of binary vectors that form a symplectic basis for F2m
2 , where ua , va ∈ F2 .
2m
Controlled-Z CZ12 , CZ12(X ⊗ I2)CZ12 = X ⊗ Z = X1Z2 Consider a system of linear equations uiF = u0i, vj F = vj0 , where i ∈ I ⊆ {1, . . . , m}, j ∈ J ⊆ {1, . . . , m} and F ∈ Sp(2m, F2). Let
0 Z
α , |Ī| + |J¯ |. Then there are 2α(α+1)/2 solutions F to the system. We also give an algorithm to efficiently enumerate them.
T
Symplectic Representation: Define E(a, b) , ι D(a, b). If g ∈ CliffN then
ab
  References
† Ag Bg

gE(a, b)g = ±E [a, b]Fg , where Fg = is symplectic, (4)
Cg Dg
1 N. Rengaswamy, R. Calderbank, S. Kadhe, and H. D. Pfister, “Synthesis of Logical Clifford Operators via Symplectic Geometry,” to appear in
i.e., Fg ΩFgT = Ω, and hence preserves inner products: h[a, b], [a0, b0]is = h[a, b]Fg , [a0, b0]Fg is. Proc. 2018 IEEE Int. Symp. Inform. Theory, arXiv preprint arXiv:1803.06987, 2018, [Online]. Available: http://arxiv.org/abs/1803.06987.
  2 D. Gottesman, “A Theory of Fault-Tolerant Quantum Computation,” arXiv preprint arXiv:quant-ph/9702029, 1997, [Online]. Available:
1 0 0 1
0 1 1 0
http://arxiv.org/pdf/quant-ph/9702029.pdf.

E.g., g = CZ12, Fg =   † †
: g(X ⊗ I2)g = gE(10, 00)g = E [10, 00]Fg = E(10, 01) = X1Z2. M. M. Wilde, “Logical operators of quantum codes,” Phys. Rev. A, vol. 79, no. 6, p. 062 322, 2009. doi: 10.1103/PhysRevA.79.062322.
1 0 3

0 1
Homomorphism φ : CliffN → Sp(2m, F2) defined as φ(g) , Fg , where Sp(2m, F2) is the binary
symplectic group. Note that for g ∈ HWN we have Fg = I2m, i.e., HWN is the kernel of the map φ.
4 T. Brun, I. Devetak, and M.-H. Hsieh, “Correcting quantum errors with entanglement,” Science, vol. 314, no. 5798, pp. 436–439, 2006. doi:
10.1126/science.1131563. [Online]. Available: http://science.sciencemag.org/content/314/5798/436.
5 T. Can, “An algorithm to generate a unitary transformation from logarithmically many random bits,” Research Independent Study, Preprint, 2017.
6 R. Chao and B. W. Reichardt, “Fault-tolerant quantum computation with few qubits,” arXiv preprint arXiv:1705.05365, 2017, [Online]. Available:
http://arxiv.org/pdf/1705.05365.pdf.