Numerical-analysis-functions

A collection of functions that solve some known problems in Matlab.
Problems and algorithms for solving them are shown below.

Linear systems

Problem description: solving a linear system Ax = b using the most suitable type of algorithm given the properties of the coefficients' matrix A.

Function	Description	Computational cost
my_diag.m	Function that solves the linear system Dx = b, where D is a diagonal matrix.	$O(n)$
my_tril.m	Function that solves the linear system Lx = b, where L is a lower triangular matrix.	$O(\frac{n^2}{2})$
my_triu.m	Function that solves the linear system Ux = b, where U is an upper triangular matrix.	$O(\frac{n^2}{2})$
my_lu.m	Function that calculates the LU factorization of a matrix A. It uses Gauss factorization method.	$O(\frac{n^3}{3})$
my_lu_piv.m	Function that calculates the LU factorization of a matrix A. It uses Gauss factorization method with partial pivoting.	$O(\frac{n^3}{3})$
my_ldl.m	Function that calculates the LDL factorization of a matrix A. It uses the Gauss factorization method for symmetric matrices.	$O(\frac{n^3}{6})$
my_jacobi.m	Function that calculates the solution of a linear system Ax=b using Jacobi method. It also returns the number of iterations.	$O(n^2)$ per iteration
my_GSeidel.m	Function that calculates the solution of a linear system Ax=b using Gauss-Seidel method. It also returns the number of iterations.	$O(n^2)$ per iteration

In order to solve complex problems that requires factorization, the solution is given by decomposing the system into several triangular or diagonal subsystems using the matrices obtained from the factorization.
Whereas Jacobi and Gauss-Seidel algorithms are iterative methods, the others are direct methods.

Non-linear equations

Problem description: finding a zero of a given function.

Function	Description	Computational cost
my_bisection.m	Function that approximates a zero of a function with one real variable.	One function evaluation per iteration.

The functions whose root is to be found can be stored in my_function.m, fruthermore using the script plot_my_function_zero.m the calculated root can be graphically displayed.

Interpolation

Problem description: given n fixed points in the cartesian plane, finding a function $g:[a, b] &#8594 R$ that passes through all the points.

Function	Description	Computational cost
my_vandermonde.m	Takes as input a vetor x of length n that represents the x-coordinates of interpolation points (nodes) and produces the Vandermonde matrix.	$O(n^3)$
my_lagrange_interpol.m	Function which, given n+1 point, evaluates the interpolating polynomial of degree n. It is calculated using the Lagrange interpolation theorem.	$O(n^2)$

The calculation of vandermonde matrix is part of the so called method of undeterminate coefficients: it is about solving the linear system $V \alpha = y$, the solution are the coefficients of the interpolation polynomial $p_n(x)$ in canonical form.
In order to avoid Runge phenomenon, Chebychev nodes can be generated using my_ceb_nodes.m for a better nodes partitioning method.

Least squares

Problem description: finding the function that aproximates better the given points, knowing the mathematical model of the function.

Function	Description	Computational cost
my_least_squares.m	Function that solves the least squares problem (general linear case). Valid for the non-degenerate case, i.e. when the rank of the matrix associated with the model is equal to the number of parameters.	$O(\frac{2n^3}{3})$
my_least_squares_poly.m	Function that solves the least squares problem (polynomial case). Valid for the non-degenerate case, i.e. when the rank of the matrix associated with the model is equal to the number of parameters.	$O(\frac{2n^3}{3})$

Credits

Numerical analysis lectures of the computer science course at UniMoRe.