Numerical methods. Note 14
Home ] Feynman's path integral method in quantum mechanics. Numerical Methods. Note  « 14 »

Feynmans path integral.' Consider a one-dimentional motion of a particle with mass m, coordinate x and Lagrangian L(x,x´ )=m/22-V(x). Suppose |xa⟩ and |xb⟩ are the eigenfunctions of the coordinate operator with the eigenvalues xa and xb. Suppose U(tb,ta)=e-iH(tb-ta) is the time-evolution operator: Ψ(x,tb)=U(tb,ta)Ψ(x,ta). Then the propagator Kxbxa(tb,ta) ≡ ⟨xb|U(tb,ta)|xa⟩ is intuitively defined defined as the Feynman's integral over all classical paths x(t) connecting (xa,ta) and (xb,tb),

Kxbxa(tb,ta) =  x(t) D[x(t)] ei/h S[x(t)]
where D[x(t)] is certain functional meausure, and S[p]=  dt L(x,x´ ) is the action calculated along the path x(t).

Connection to classical mechanics. The stationary phase approach says that when h→0 only the stationary path, that is where δS/δx=0, gives nonvanishing contribution to the integral. This is apparently the classical variational principle which defines the classical path.

Lattice discretization (and regularization). Let us discretize time in N equidistant slices t0=ta,t1,...,tN=tb, with Δt=(tb-ta)/N

Kxbxa(tb,ta)  ≈  (m/2π i h Δt)N/2   dx(t1)...dx(tN-1) exp(i/hn Sn)
where the short time action is defined using the midpoint rule,
Sn = Δt L[1/2(xn+xn-1),1/Δt(xn-xn-1)]
The (broken-line) path is now defined by the set of values {x(t1),x(t2),...,x(tN-1)}

Monte-Carlo evaluation of the path integral: generate random paths {x(t1),x(t2),...,x(tN-1)} using some algorithm and add up their contributions to the path integral.

Space discretization With the space discretized also, x={x0,x1,...,xD}, the short-time propagator becomes a complex matrix

Kij(Δt)=(m/2π i h Δt)1/2 exp(iΔt/h L[1/2(xj+xi),1/Δt(xj-xi)])
and consequently the finite time propagator Kij(t)=⟨xi|U(t)|xj⟩ becomes a product of short-time propagators, K(t)=ΔxN-1K(Δt)N.

Imaginary time propagation. Partition function e-βH.

Real time propagation. The energy levels can be extracted by a Fourier transform of the trace of the propagator:

Tr U(t) = ∑n ⟨ n | exp(-i/hHt) | n ⟩ = ∑n exp(-i/hEn t)

Problems

  1. Consider (the lowest states of) a one-dimentional oscillator with time and space discretized path integral method with real- and complex-time propagation.

"Copyleft" © 2004 D.V.Fedorov (fedorov at phys dot au dot dk)