25.07.2013 Views

Ph.D. Thesis - Physics

Ph.D. Thesis - Physics

Ph.D. Thesis - Physics

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Digital<br />

Analog<br />

Classical Quantum<br />

Space: 2nO(log 1/ǫ)<br />

Time: T22n Space: 2n Time: T<br />

Precision: O(log 1/ǫfix)<br />

Space: n<br />

Time: Tn 2 O(1/ǫ r )<br />

Space: n<br />

Time: T O(1/ǫ)<br />

Table 3.2: A comparison of the resources required for digital and analog simulation of<br />

quantum systems, using both classical and quantum systems, and taking into account the<br />

precision obtained. We consider a system of n qubits simulated for a total time T. The error<br />

due to projection noise is denoted ǫ, while the fixed error of a classical analog computer is<br />

denoted ǫfix. Here r ≥ 2 when error correction is required.<br />

the quantum simulation is still more efficient than a classical algorithm, provided no<br />

efficient classical approximation exists.<br />

2. In general, use of the Trotter formula or some related technique will be necessary<br />

in digital quantum simulation, as most non-trivial Hamiltonians have mutually non-<br />

commuting terms. We have shown that, when fault tolerance is required, this process<br />

increases the number of required gates by a broadly exponential factor. The conclusion<br />

here is roughly the same as in item 1: this class of simulations is inefficient with respect<br />

to the precision, but moreso when error correction is needed. This result is included<br />

in Table 3.2.<br />

3. Control errors are a problem that afflicts any quantum simulator. The type that<br />

appear in NMR arise from the “always-on” scalar coupling interaction. Although<br />

for small systems it is relatively easy to compensate (to first order) for these errors,<br />

and thereby bring the solution to within the theoretical bound, it will not be so<br />

straightforward in larger systems. Fortunately, in scalable systems such as ion traps,<br />

the two-body interactions are controlled more directly by the experimenter. There<br />

will be control errors in that system as well, but of a different sort.<br />

4. All quantum simulations are limited by natural decoherence times. The bound on<br />

the number of qubits that may be used to obtain an answer to within a given error<br />

depends on the number of gates that may be performed within the coherence time,<br />

unless error correction is used. Error correction becomes necessary for larger systems,<br />

but then the overall scaling properties become less favorable, as explained in items 1<br />

and 2.<br />

We would like to point out that the above results apply to digital quantum simulation<br />

generally, regardless of whether the classical discrete Fourier transform (DFT) or the quan-<br />

tum Fourier transform (QFT) is used. Our results apply to both for the reason that they<br />

deal with the implementation of the simulated Hamiltonian; whether the DFT or QFT is<br />

79

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!