11.03.2014 Views

Itinerant Spin Dynamics in Structures of ... - Jacobs University

Itinerant Spin Dynamics in Structures of ... - Jacobs University

Itinerant Spin Dynamics in Structures of ... - Jacobs University

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

92 Chapter 5: <strong>Sp<strong>in</strong></strong> Hall Effect<br />

result is presented <strong>in</strong> Fig.5.3. For small V one would expect the Van Hove s<strong>in</strong>gularity at<br />

half fill<strong>in</strong>g. Due to f<strong>in</strong>ite Rashba SOC it is split to f<strong>in</strong>ite energies E = ±(2t− √ 4+α 2 2 t),<br />

<strong>in</strong>dicated <strong>in</strong> Fig.5.1 for the energy below half fill<strong>in</strong>g. If the impurity strength is <strong>in</strong>creased,<br />

a preformed impurity band is created. Us<strong>in</strong>g exact diagonalization 1 and apply<strong>in</strong>g the Kubo<br />

formalism, Eq.5.22, we calculate the SHC. Exemplarily we show the result for σ SH (E) at<br />

V = −2.8t <strong>in</strong> Fig.5.4. The SHC is strongly reduced but shows an additional maximum at<br />

energy where the preformed impurity band is located, as can be seen <strong>in</strong> comparison with<br />

the DOS, Fig.5.4(b).<br />

Toanalyze thereduction <strong>of</strong>SHCfor agiven fill<strong>in</strong>gn, wevary theimpuritystrength<br />

up to V = −5t and keep the concentration constant at 10%. Similar to the results <strong>in</strong> the<br />

case <strong>of</strong> block distribution <strong>of</strong> impurity strength, we see a monotone suppression at all fill<strong>in</strong>gs,<br />

even <strong>in</strong> the preformed impurity band, Fig.5.5.<br />

5.3.2 Kernel Polynomial Method<br />

The numerical calculations presented <strong>in</strong> the previous section, which were based<br />

on exact diagonalization us<strong>in</strong>g LAPACK[LAP] rout<strong>in</strong>es, are limited to small system sizes.<br />

This leads to f<strong>in</strong>ite size effects like oscillations <strong>in</strong> the SHC, e.g. Fig.5.4(b) above half fill<strong>in</strong>g.<br />

For further calculations concern<strong>in</strong>g the role <strong>of</strong> the impurity band it is necessary to do a<br />

f<strong>in</strong>ite size scal<strong>in</strong>g analysis and consider system-sizes beyond L = 64 which makes an exact<br />

treatment on current hardware impossible: for a D-dimensional matrix such a calculation<br />

requires memory <strong>of</strong> the order <strong>of</strong> D 2 , and the LAPACK rout<strong>in</strong>e scales as D 3 .<br />

Another problem is the adjustment <strong>of</strong> the cut<strong>of</strong>f η, see Eq.(5.22), which has to be taken<br />

with care as analyzed e.g. by Nomura et al., Ref.[NSSM05].<br />

Toovercome thelimitation onsmall systems, therearedifferent numerical order-Dmethods.<br />

Oneprocedureisthetimeevolution projection methoddevelopedbyTanakaandItoh[TI98],<br />

which was already used to calculate SHC[MMF08, MM07]. However, the algorithm requires<br />

both the choice <strong>of</strong> a sufficient number <strong>of</strong> time steps and an adjustment <strong>of</strong> cut<strong>of</strong>f η. A more<br />

effective method, which uses Chebyshev expansion based on Kernel Polynomial Method,<br />

will be presented <strong>in</strong> the follow<strong>in</strong>g.<br />

1 us<strong>in</strong>g LAPACK[LAP] and OpenMP[OMP] <strong>in</strong> C++

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

Saved successfully!

Ooh no, something went wrong!