Menu

Numerics.TDSE.Chebychev

Burkhard Schmidt Ulf Lorenz
Attachments
bessel_def.gif (4162 bytes)
bessel_def2.gif (4350 bytes)
chebychev_def.gif (2362 bytes)
coefficient.gif (1762 bytes)
expansion.gif (3196 bytes)
result.gif (5234 bytes)
result2.gif (2562 bytes)

General scheme

The basic idea of the Chebychev method is an expansion into Chebychev polynomials. To recall, the idea of a (truncated) expansion in an orthonormal basis of functions is to write a given function f(x) as

f(x) = \sum_{n=0}^N f_n \varphi_n(x)

For a given metric induced by a scalar product, the expansion is optimal if the coefficients are chosen as

f_n = \langle \varphi_n | f \rangle

Chebychev expansion

For the particular problem described in the section on propagators in quantum dynamics, we choose an expansion of the form

e^{-i\alpha x} = \sum_{n=0}^N c_n(\alpha) T_n(x)

Here, x is an arbitrary eigenvalue of the normalized Hamiltonian in the range from -1 to +1, and we choose to consider α as a scaling factor that determines the coefficients, while Tn are Chebychev polynomials of the first kind. These are orthogonal, but not normalized with respect to the scalar product

\langle T_n | T_m \rangle = \int_{-1}^1 T_n(x) T_m(x) \frac{dx}{\sqrt{1-x^2}} = \frac{\pi}{K_n} \delta_{nm}

Here, K=1 for n=0 and K=2 otherwise. The coefficients are then calculated according to

c_n(\alpha) = \frac{K_n}{\pi} \langle T_n | e^{-i\alpha x} \rangle = \frac{K_n}{\pi} \int_{-1}^1 \frac{e^{-i\alpha x} T_n(x)}{\sqrt{1-x^2}} dx

(the factors come from the missing normalization). Question: How do we evaluate the integral? For this, we use an integral representation for the Bessel functions of first kind (see Abramowitz&Stegun alias DLMF, eq. 10.9.2),

J_n(z) = \frac{i^{-n}}{\pi} \int_0^\pi e^{iz\cos\theta} \cos(n\theta) d\theta

We can transform the integral using two steps:

  • We use the property of Chebychev polynomials T_n(\cos\theta) = \cos(n\theta) (see Abramowitz&Stegun alias DLMF, eq. 18.5.1),
  • We substitute the integration variable to x=cosθ

and obtain the result

T_n(\cos\theta) = \cos(n\theta)

which looks almost like what we want for z=-α. Note that from the definition and the real-valuedness of the Bessel function, it is easy to show that Bessel functions are even (/odd) for even (/odd) n. We end up with an expression for the coefficients that is

c_n(\alpha) = i^n (-1)^n K_n J_n(\alpha)

Putting all this together, and replacing x by the normalized Hamiltonian, we end up with the final form of the Chebychev propagator

e^{-i \alpha \hat H_\text{norm}} = \sum_{n=0}^N i^n (-1)^n K_n J_n(\alpha) \ T_n(\hat H_\text{norm})

If you look up the original references by Tal-Ezer and Ronny Kosloff, you will find that their expansion looks considerably different. There, some of the pre-factors are included in the polynomials, thus defining new polynomials as

\phi_n(x) = i^n (-1)^n T_n(x)

If you plug in these definitions into the recursive definition of the Chebychev polynomials, you get the recursion relations for the new polynomials φn that are shown on the page for solution methods for the time-dependent Schrödinger equation.


Related

Wiki: Numerics.TDSE

Want the latest updates on software, tech news, and AI?
Get latest updates about software, tech news, and AI from SourceForge directly in your inbox once a month.