text stringlengths 0 8.13M |
|---|
it was shown that almost any n-bit gate with n 2 is universal. The question |
≥ |
as to which gates are not universal was clarified in [51], a result which we will |
return to below. |
Single qubit and CNOT universality |
Relying on the fact that the Deutsch gate Λ (U) is universal, and the gate |
2 |
constructions of section6.2.10, where it was shownthat any Λ (U) gate can be |
2 |
decomposedintermsofsinglequbitandCNOTgates,itfollowsthatsinglequbit |
and CNOT gates are universal for quantum computation. This is approximate |
universality, since the Deutsch gate is approximately universal. |
Howeverit is possible to show thatsingle qubit andCNOT gates arein fact |
exactly universal. This stronger results follows from the exact decomposition |
of a general N-dimensional unitary matrix in terms of 2-dimensional unitaries |
acting on two-dimensional subspaces. Relying on the construction in section |
6.2.11, this can be simulated with single-qubit gates and CNOT-gates. |
6.2.16 Discrete sets of gates |
Finally we arrive at a discussion on finite sets of practical one qubit gates to- |
gether with CNOT gates. Relying on the previous discussion, it now suffices to |
effectively approximate general one-qubit gates with a discrete set of one-qubit |
gates. |
The standard set of gates consists of the Hadamard gate H, the π/8 gate T |
and the phase gate S [53]. The phase gate is not really needed for universality, |
but it is needed in order to perform the approximations fault tolerantly, an |
important topic that will be briefly mentioned below. |
The proof that the gates H, π/8 and T are sufficient to approximate any |
single qubit gate, is quite complicated. It will not be repeated here, the reader |
is instead referred to [39]. |
147 |
Efficiency of the gate constructions |
In order to estimate the efficiency of this construction, consider an n-qubit |
circuit approximating a U(2n) operation. Suppose the circuit requires f(n) |
gates,mostofwhichneednotbeintheuniversalsetofgatessothatmostofthem |
must be approximated. Suppose furthermore that the total tolerance to errors |
is required to be ǫ. Then each gate must be approximated to within an error |
δ =ǫ/f(n) since the errors adds at most linearly. The overall efficiency is thus |
determinedbytheefficiencywithwhichasinglequbitgatecanbeapproximated. |
Naively,onewouldsuspectthatroughlyΘ(1/δ)gatesfromthediscretesetwould |
be needed, resulting in an overall efficiency of Ω(f(n)/δ)=Ω(f(n)/(ǫ/f(n))= |
Ω(f(n)2/ǫ). |
However, according to the Solovay-Kitaev theorem (which unfortunately is |
too complicatedto provehere)a generalsingle qubit gatecanbe approximated |
within an error δ using (logc(1/δ)) gates from the discrete set. The constant |
O |
c is number approximatelyequalto 2. Therefore,the overallefficiency becomes |
(f(n)logc(f(n)/δ)),whichisapolylogarithmicincreaseinthenumberofgates |
O |
as compared to original circuit. |
What about the function f(n)? An upper bound for this function can be |
calculated based on the gate constructions in sections 6.2.11 and 6.2.14. From |
the decomposition of an arbitrary U(2n) matrix in terms of two-level unitaries |
(section 6.2.14), we know that it takes 2n(2n 1)/2 gates, i.e. (4n). Then, |
− O |
accordingtosection6.2.11,eachsuchtwo-levelunitarygate(actingonthestate |
space of n qubits) can be implemented using (n2) single qubit and CNOT |
O |
gates. Therefore, the function f(n) is in (n24n). |
O |
Putting everything together we see that the overall complexity is |
(n24nlogc(n24n/δ)), which is exponential in n. But the exponential behav- |
O |
ior arises not from the approximation of single qubit gates by discrete single |
qubit gates, but from the complexity of breaking down general U(2n) matrices |
in terms of two-level matrices. Thus, fast quantum algorithms cannot rely on |
naive universality constructions. |
6.2.17 General results |
The particular universality results reviewed to above are beautifully subsumed |
under a general theorem [51]. In order to state the theorem, the notion of an |
imprimitive gate must be defined. |
A two-qubit gate V is said to be primitive if it maps decomposable states |
intodecomposablestates,i.e. if x and y arequbits, thenthere arequbits u |
| i | i | i |
and v such that V x y = u v . V is imprimitive if it is not primitive. |
| i | i| i | i| i |
There is a simple condition to determine whether a gate is primitive or |
not. Let P be an operator that swaps the two qubits in a product state, i.e. |
P x y = y x . Then it can be shownthat V is primitive if and only if V can |
| i| i | i| i |
be expressed as S T or as (S T)P for some single qubit gates S and T, so |
⊗ ⊗ |
that V acts as V x y =S x T y or as V x y =S y T x . |
| i| i | i⊗ | i | i| i | i⊗ | i |
148 |
Theorem |
Given a two-qubit gate V, the following conditions are equivalent |
the collection of all single qubit gates together with V is approximately |
• |
universal |
the collectionofallsingle qubitgatestogetherwith V is exactlyuniversal |
• |
V is imprimitive |
• |
Of course, a discrete set of single qubit gates can never yield exact univer- |
sality. |
149 |
Bibliography |
[1] J. Gruska, Quantum Computing, McGraw-Hill, London 1999. |
[2] A. Steane, Quantum Computing, Rep. Prog. Phys. 61 (1998) 117-173. |
[3] A.GalindoandM.A.Mart´in-Delgado,Information andcomputation: Clas- |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.