View in A-Z Index
Number of views:
33080

Let D y(x) = f(x) be a differential equation defined by a given differential operator D. For simplicity, let us assume that the equation is given by

(1)

with the initial condition y(0) = 1, for x in the interval [0, 1]. The solution of this equation is y(x) = e–x, which has the infinite series expansion

(2)

Let us assume that we use only the first three terms of it and let y2(x) stand for this polynomial, of degree 2. If we take for y(x) the approximate expression y2(x) and replace it into the differential equation (1) we obtain one with a nonzero right hand side:

(3)

The higher the order of approximation n, the smaller the right hand side shall be (in our example it will be xn/n! for x in the interval 0,1). It is well known that for any given n the monomials Kxn with a nonzero coefficient K, are not the closest (or best uniform) approximation of zero (the original right hand side of Equation (1) in the interval 0,1. In fact for every value of n, the Chebyshev Polynomials Tn(x) have that singular property.

Therefore, it seems appropriate to reconsider the problem in a completely new way: instead of finding the error term (xn/n! in our concrete example) after fixing the approximation Yn(x), we should fix the error term to be minimal (i.e., a Chebyshev polynomial) and then determine the approximation Yn(x) which satisfies it. This original idea was proposed by the eminent mathematician Cornelius Lanczos in his formulation of the Tau Method, first conceived while working under Albert Einstein in problems of relativity theory [see Lanczos (1956) for an account of the original formulation of this method].

Assume that for a given differential operator D we can determine a sequence of polynomials Qn(x), n, = 0, 1, 2,..., called canonical polynomials, associated with D, and such that for any value of n,

(4)

(i.e., they generate a basis in the image space). The Chebyshev polynomials Tn(x) have the form

(5)

where the coefficients ck are known for all k and n. To find a Tau approximation Yn(x) we can start using τTn(x) as the minimal right hand side. (This term is usually referred to as the Tau Method perturbation term.) The free parameter τ is a multiplier introduced for us to be able to satisfy the initial condition (y(0) = 1 in our example, more generally, initial, boundary or multipoint boundary conditions). Then, if we set

(6)

the Tau approximate solution Yn(x) is immediately given in our concrete example by

(7)

where the parameter τ is such that

(8)

The main problem then is to find the sequence Qn(x) associated with D for all n. It has been proved that a sequence of canonical polynomials is uniquely associated to every differential operator D and that it can be generated recursively through a simple expression (iteratively for nonlinear operators), given exclusively in terms of the coefficients of the differential equation [see Oritz (1968) for further details]. For example, for Equation (1) the sequence of canonical polynomials is recursively generated by

(9)

that is:

(10)

Suitable generalizations of the Tau Method have been applied successfully to the numerical approximation of complex systems of nonlinear partial differential equations involving large gradients in the function and or derivatives, such as in the case of soliton interactions and also in problems of fracture mechanics, fluid mechanics, mathematical physics and molecular biology. It has also been used in the approximation of differential eigenvalue problems where the eigenvalue parameter appears nonlinearly; such questions arise in difficult fluid dynamics problems see Oritz, (1994), for an overview of recent research into the Tau Method.

One advantage of the Tau Method, which accounts for the high precision of its approximations is that it does not require a discretization of the given differential operator, a process which often alters the behavior of solutions.

Several numerical approximation techniques can be interpreted as special cases of the Tau Method, corresponding to different choices of perturbation terms, and can then be treated in a more unified and systematic way. Among these are the Spectral Method, Collocation and several types of projection and finite difference and finite element methods. (See for example Spectral Analysis and Numerical Methods.)

REFERENCES

Lanczos, C. (1956) Applied Analysis, Prentice-Hall, New Jersey.

Ortiz, E. L. (1968) The Tau Method, SIAM J. Numer. Anal., 6, 480–492.

Ortiz, E. L. (1994) The Tau method and the numerical solution of differential equations: past and recent research, Proc. Cornelius Lanczos International Centenary Conference, 77–81, J. David Brown, Ed., SIAM, Philadelphia.

References

  1. Lanczos, C. (1956) Applied Analysis, Prentice-Hall, New Jersey.
  2. Ortiz, E. L. (1968) The Tau Method, SIAM J. Numer. Anal., 6, 480–492.
  3. Ortiz, E. L. (1994) The Tau method and the numerical solution of differential equations: past and recent research, Proc. Cornelius Lanczos International Centenary Conference, 77–81, J. David Brown, Ed., SIAM, Philadelphia.
Back to top © Copyright 2008-2024