76 4. TOP
C D
0:04 0:95i 0:01 0:30i
0:07 C 0:29i 0:25 0:92i
; and D D
0:87 0:43i 0:15 C 0:20i
0:08 0:24i 0:68 0:68i
and
A D
0:67 0:72i 0:19 0:03i
0:16 C 0:10i 0:30 0:93i
; B D
0:50 0:52i 0:50 C 0:47i
0:19 C 0:66i 0:70 C 0:20i
;
C D
0:04 C 0:95i 0:07 0:29i
0:01 C 0:30i 0:25 C0:92i
; and D D
0:87 C 0:43i 0:15 0:20i
0:08 C 0:24i 0:68 C 0:68i
:
In Chapter 2, we investigated the synthesis of all 24 classical reversible two-bit circuits as
well as all 40,320 classical reversible three-bit circuits. We investigated the statistics of the gate
cost of the resulting gate cascades. Here, we cannot synthesize all 2-qubit or 3-qubit circuits,
for the simple reason there are an infinite number of them. e best we can do is generate,
for example, 1,000 random U(4) or U(8) matrices [56, 65] and synthesize the corresponding
quantum circuits. However, we can predict that all these circuits will have a same gate cost.
Indeed, as explained in Section 4.2, the resulting schematic consists of
5
12
4
w
2
3
(un)controlled
U(2) gates.
Because there exist only two different P(2) blocks, there is one chance out of two such
a block equals the IDENTITY block. In contrast, because there exist 1
4
different U(2) blocks,
the probability it equals the
IDENTITY
block is negligible. Hence, none of the 1,000 circuits
will contain a controlled IDENTITY gate that would lead to a cost reduction. Hence, all 2-qubit
circuits will have quantum-gate cost equal to 6, and all 3-qubit circuits will have quantum-gate
cost equal to 26. Decomposition of each U(2) block into two XU(2) and three ZU(2) blocks
(see Section 3.3) does not alter the conclusion: the odds that a NEGATOR gate or a PHASOR gate
equals the IDENTITY gate are 1 to 1
1
.
In contrast to the numerical approach in the above first example, we will now perform an
analytic decomposition of a second example:
U D
0
B
B
@
1
cos.t/ sin.t/
sin.t/ cos.t /
1
1
C
C
A
;
i.e., a typical evolution matrix for spin-spin interaction, often discussed in physics. We have the
following four matrix blocks and their polar decompositions
1
:
1
In fact, the presented polar decompositions are only valid if 0 t =2 (i.e., if both c 0 and s 0). However, the
reader can easily treat the three other cases.