14.09.2014 Views

pdf, 9 MiB - Infoscience - EPFL

pdf, 9 MiB - Infoscience - EPFL

pdf, 9 MiB - Infoscience - EPFL

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.

48 CHAPTER 2. NUMERICAL METHODS<br />

with φ † m = ∑ j<br />

c † jm and Φ′ = e U Φ, where U is a square matrix whose elements<br />

are given by U ij and B = e U is therefore an N × N square matrix as well. In<br />

conclusion, the operation of ˆB on |φ〉 simply involves multiplying an N × N<br />

matrix by an N × M matrix. This allows to apply the exponential of the kinetic<br />

term e λK on the Salter determinant |ψ MF 〉. Furthermore, for two states |L〉 and<br />

|R〉 represented by Slater determinants L and R respectively, the single-particle<br />

right and left Green functions are written as :<br />

〈<br />

〉<br />

∣<br />

L ∣c †<br />

G L ij ≡ i c j∣ R [<br />

= R ( L T R ) ]<br />

−1<br />

L<br />

T<br />

(2.85)<br />

〈L |R〉<br />

ij<br />

〈<br />

〉<br />

∣<br />

L ∣c i c † G R ij ≡ j∣ R [<br />

= δ ij − R ( L T R ) ]<br />

−1<br />

L<br />

T<br />

(2.86)<br />

〈L |R〉<br />

ij<br />

In order to evaluate all the expectation values of the observables, we generate a<br />

Monte Carlo sample by using the importance sampling with the following weight<br />

function :<br />

ω = det ( )<br />

ψMF T eV (u1) e λ1K e V (u2) e λ2K ...e λ2K e V (u2m−1) e λ1K e V (u2m) ψ MF (2.87)<br />

where {u i } are the Ising spins of the different species i =1, .., 2m lying on the<br />

lattice. The Monte Carlo algorithm consists to update the Ising variables, from<br />

the old s i to the new one s ′ i , the and the ratio r = ω′ /ω is calculated to determine<br />

whether to accept or reject the new configuration. In the process updating the<br />

spin, the Green function can be used to calculate very fast the ratio of the the<br />

weights r:<br />

r = det (1 + G L ∆) (2.88)<br />

where ∆ = ( e V ′ −V − 1 ) and the two states L and R (in the notations of equation<br />

(2.85)) that define G L are :<br />

L = ψMF T eV (u1) e λ1K e V (u2) e λ2K ...e V (u i)<br />

(2.89)<br />

R = e λiK ...e V (u2m−1) e λ1K e V (u2m) ψ MF (2.90)<br />

u i is the spin species that is changed during the move. When a change of the<br />

spin of the corresponding species is accepted, the Green function is updated like<br />

:<br />

G ′ L = G L − G L∆(1 − G L )<br />

(2.91)<br />

1+(1 − G L )∆<br />

Moreover, Steve White and collaborators have proposed a very efficient way to<br />

carry on the simulations [62] by proposing moves successively for each of the<br />

species of spin, that allows for a fast sweep over all the site of the lattice and over<br />

all the species of spin. Finally, since the AFQMC is a non-interacting theory,

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

Saved successfully!

Ooh no, something went wrong!