On quantum circuits employing roots of the Pauli matrices
Mathias Soeken,1 D. Michael Miller,2 and Rolf Drechsler1 1Institute of Computer Science, University of Bremen, Germany 2Department of Computer Science, University of Victoria, BC, Canada The Pauli matrices are a set of three 2×2 complex Hermitian, unitary matrices. In this article, we investigate the relationships between certain roots of the Pauli matrices and how gates implementing those roots are used in quantum circuits. Techniques for simplifying such circuits are given. In particular, we show how those techniques can be used to find a circuit of Clifford+T gates starting from a circuit composed of gates from the well studied NCV library.
I. INTRODUCTION The NOT operation is described by the Pauli X ma- trix [10]. Note that the controlled NOT could be drawn The well-studied NCV quantum gate library [1] con- as a controlled X but we use the normal convention of a tains the gates: NOT (X), controlled NOT (CNOT), and ⊕ as shown in the figures above. The matrices for the V † both the single controlled square root of NOT as well as and V operations as used in (1) are square roots of the its adjoint, denoted V and V †, respectively. In their sem- Pauli X matrix. Similarly, the T gate operation is given inal paper [2], Barenco et al. presented a general result by a matrix (2) that is the fourth root of the Pauli Z that as one instance shows how a classical reversible Tof- matrix. foli gate can be realized by five NCV gates as follows: The use of gates associated with different Pauli matri- ces and their roots within the same circuit, as illustrated in (3), motivated us to explore the relations between the = (1) Pauli matrices and their roots, and to investigate how the V V † V associated gates can be used in constructing quantum circuits. This article presents our findings and demon- This result has been used in several works on synthesis strates their applicability in deriving the optimal circuits and optimization of quantum circuits [3–5]. from [8] from known NCV circuits rather than using ex- Another quantum gate library called Clifford+T [1], haustive search techniques. which consists of the controlled NOT, the phase gate S and the Hadamard gate H, plus the T gate II. PRELIMINARIES 1 0 T = iπ , (2) 0 e 4 The three Pauli matrices [10] are given by and the adjoint gates S† and T † has also received consid- erable attention. The Clifford+T library has the advan- 0 1 0 −i 1 0 σ1 = , σ2 = , σ3 = (4) tage over the NCV library with respect to fault-tolerant 1 0 i 0 0 −1 computing [6]. Recently different synthesis results based on the The alternate naming X = σ1, Y = σ2, and Z = σ3 is Clifford+T gate library have been presented [7–9]. One often used and we use it whenever we refer to a specific common aim in these works is to reduce the so-called Pauli matrix. T -depth, i.e. the number of T -stages where each stage Matrices describing rotations around the three axes of consists of one or more T or T † gates that can operate the Bloch sphere are given by simultaneously on separate qubits. In [8] the authors de-
arXiv:1308.2493v1 [quant-ph] 12 Aug 2013 θ θ scribe a search-based algorithm that finds optimal circuit Ra(θ) = cos 2 I − i sin 2 σa (5) realizations with respect to their T -depth. One of their circuits that realizes a Toffoli gate is [8, Fig. 13]: where a ∈ {1, 2, 3} with θ being the rotation angle and I the identity matrix [1]. Each Pauli matrix specifies a T T † half turn (180◦) rotation around a particular axis up to = T T † T † (3) a global phase, i.e.
H T T H iπ σa = e 2 Ra(π). (6) This circuit has a T -depth of 3 and a total depth of 10. (Note that the gates surrounded by the dashed rectangle The conjugate transpose of Ra(θ) is found by negating together have a depth of 1 and are drawn in sequence only the angle θ or by multiplying it by another Pauli ma- for clarity.) The approach in [8] produces optimal circuits trix σb from both sides, i.e. but since the technique’s complexity is exponential, it is † only applicable to small circuits. Ra(θ) = Ra(−θ) = σbRa(θ)σb (7) 2 where a 6= b. Note that it does not matter which of the Given a unitary 2n × 2n matrix U, called the target two possible σb is used. Since σb is Hermitian, we also operation, we define four controlled operations † have Ra(θ) = σbRa(θ)σb. U U Rotation matrices are additive with respect to their , , , and (18) angle, i.e. R (θ )R (θ ) = R (θ + θ ), so one can de- a 1 a 2 a 1 2 U U rive the kth root of the Pauli matrices as well as their conjugate transpose from (6) as which are described by
√ iπ π √ † − iπ † π k 2k k 2k C (U) = |0ih0| ⊗ I n + |1ih1| ⊗ U, σa = e Ra k and σa = e Ra k . (8) 1 2 C2(U) = I2n ⊗ |0ih0| + U ⊗ |1ih1| , For brevity, we term these matrices the Pauli roots. − (19) Using (8) the rotation matrices can also be expressed C1 (U) = |0ih0| ⊗ U + |1ih1| ⊗ I2n , and − in terms of the roots of the Pauli matrices, i.e. C2 (U) = U ⊗ |0ih0| + I2n ⊗ |1ih1| ,