| dc.description.abstract |
In this dissertation, different parameter uniformly convergent numerical methods
are proposed to solve time-fractional singularly perturbed convection-diffusion
equations with/without delay in the time variable. The time-fractional derivative
is given in the sense of Caputo with order α ∈ (0, 1). These types of model
problems have wide applications in various branches of science and engineering,
which are characterized by a small perturbation parameter ϵ (0 < ϵ << 1) that
multiplies the diffusion term. The solution exhibits a regular boundary layer
near the right side of the spatial domain as the perturbation parameter ϵ → 0.
This means that the solution varies rapidly in the boundary region and behaves
smoothly away from the layer region. Due to this multiscale property of the solution,
standard numerical methods give oscillatory solutions on a uniform mesh,
which are not typical properties of the solution of the problems, unless the step
sizes must be smaller than the perturbation parameter, which is computationally
expensive. To overcome this difficulty, parameter uniformly convergent numerical
methods are needed. Hence, the aim of this dissertation is to develop simple, more
accurate, and parameter uniformly convergent numerical methods for the mathematical
problems considered. The L1, L2 − 1σ, and Crank-Nicolson schemes are
used to discretize the model problems in the temporal direction, and the fitted
operator, fitted mesh, and hybrid methods are used to discretize the model equations
in the spatial direction. The matrix inversion method is used to solve the
resulting linear system. The stability and convergence analysis of the developed
methods are the immediate consequence of the discrete maximum principle. The
parameter uniform convergence analysis of the developed methods is discussed in
the maximum norm. To validate the developed methods, different numerical examples
are considered. The results are presented with various graphs and tables.
The results support the theoretical findings. Moreover, comparisons are made
with some of the existing methods in the literature, and it is observed that the
proposed schemes in this dissertation outperform in terms of accuracy and order
of convergence. Finally, the suggested numerical methods are simple, accurate,
and parameter uniformly convergent. |
en_US |