13.07.2015 Views

PhD thesis in English

PhD thesis in English

PhD thesis in English

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.

2. Diagonalization of Transition Amplitudesstep ∆ are more complex, and follow from the fact that the k<strong>in</strong>etic energy operatorcannot be exactly represented on f<strong>in</strong>ite real-space grids. For example, a typicalnaive discretization of the k<strong>in</strong>etic energy operator gives <strong>in</strong> our notation the follow<strong>in</strong>gHamiltonian matrix elements [48]⎧⎪⎨ 1/∆ 2 + V (n∆) if n = mH nm = −1/(2∆⎪⎩2 ) if |n − m| = 10 otherwise.(2.4)Note that <strong>in</strong> the absence of a potential term V <strong>in</strong> the Hamiltonian, the above def<strong>in</strong>itioncorresponds to a tight-b<strong>in</strong>d<strong>in</strong>g model [48]. This prescription leads to numericalresults for eigenvalues which <strong>in</strong> the ∆ → 0 limit converge to the exact cont<strong>in</strong>uumvalues as ∆ 2 . The errors associated with this approach have non-variational behavior,i.e. the obta<strong>in</strong>ed results are not always upper bounds of the exact energy levels.Several papers discuss this issue and analyze the behavior of errors <strong>in</strong> the direct diagonalizationapproach (for more details, see Refs. [49, 50] and references there<strong>in</strong>).The state-of-the-art <strong>in</strong> this approach is a set of systematically improved prescriptionsfor discretization of the k<strong>in</strong>etic energy operator, which speeds up convergenceto the cont<strong>in</strong>uum limit to higher powers of ∆ 2 . However, with<strong>in</strong> this approach convergenceis always polynomial <strong>in</strong> ∆. Some recent results [51] also exist on extensionsof this approach that provide effective variational behavior of the discretized k<strong>in</strong>eticenergy operator.As outl<strong>in</strong>ed <strong>in</strong> the Introduction, we focus on an alternative approach, based onsolv<strong>in</strong>g the eigenproblem of the correspond<strong>in</strong>g transition amplitudes as proposed <strong>in</strong>[9]. The central equation isor <strong>in</strong> the discretized formN∑cut−1e −tĤ|ψ〉 = e −tE |ψ〉 , (2.5)m=−N cutA nm (t) 〈m∆|ψ〉 = e −t E(∆,L,t) 〈n∆|ψ〉 , (2.6)where A nm (t) = ∆·A(n∆, m∆; t) = ∆·〈n∆|e −tĤ|m∆〉. In this approach the time ofevolution t plays the role of an auxiliary parameter. This parameter is not relatedto the discretization, and <strong>in</strong> a cont<strong>in</strong>uous theory it does not affect the obta<strong>in</strong>edeigenvalues and eigenstates. However, <strong>in</strong> a discretized theory the numerically calcu-23

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

Saved successfully!

Ooh no, something went wrong!