Scribd
Upload
140 views

LIN1 Solved

This document discusses solving systems of linear equations. It provides an example of using a system of linear equations to analyze forces on a truss structure. The document then covers determining if a unique solution exists, and different methods in MATLAB for solving systems of linear equations, including LU decomposition, QR factorization, and Cholesky decomposition.

Uploaded by

Sheeraz Ahmed
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
140 views

LIN1 Solved

This document discusses solving systems of linear equations. It provides an example of using a system of linear equations to analyze forces on a truss structure. The document then covers determining if a unique solution exists, and different methods in MATLAB for solving systems of linear equations, including LU decomposition, QR factorization, and Cholesky decomposition.

Uploaded by

Sheeraz Ahmed
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

System of Linear Equations

Systems of linear equations are very important in applied and numerical mathematics. Since engineers are
dealing mostly with linear models, this requires solving systems of linear equations in engineering problems.

Consider a problem in statics. Let us analyze the forces on a truss with equilateral triangles as shown on the
figure below.

Compute member forces by the method of joints. Make the following approximations: and

( and ).

Equations of equilibrium of rectangular components of forces and moments are used to determine reactions
at supports ( , , ). The following system of linear equations can be formulated by considering forces at
nodes, where represent tension in members:

The above linear system is equivalent to the matrix equation .

In general form:

1
Existence and uniqueness of solution
The system of equations can be inhomogeneous (b ≠ 0) or homogeneous (b = 0). When the system is
homogeneous, we have a nontrivial solution (x ≠ 0) only if the determinant of A is zero: det(A)=0.

Discussion of solutions when the system is inhomogeneous:

Solution exists and is unique


Solution exists, if rank of matrix A (number of its linearly independent columns, r(A)) is the same as the rank of
the augmented matrix A|b, i.e. r(A)=r(A|b). The augmented matrix is obtained by appending column vector b to
matrix A.

2
The solution is unique, if r(A)=r(A|b)=n or det(A) ≠ 0, that is rank of matrix A is equal to the rank of the
augmented matrix and this rank is full (equals to the number of columns).

Load A and b from file truss.txt.

Ab=[0.500000000000000,1,0,0,0,0,0,1000;...
0.866000000000000,0,0,0,0,0,0,-2067;...
-0.500000000000000,0,0.500000000000000,1,0,0,0,-1000;...
0.866000000000000,0,0.866000000000000,0,0,0,0,0;...
0,-1,-0.500000000000000,0,0.500000000000000,1,0,0;...
0,0,0.866000000000000,0,0.866000000000000,0,0,5000;...
0,0,0,-1,-0.500000000000000,0,0.500000000000000,0];
% Ab = load('truss.txt')
A = Ab(:,1:end-1)
b = Ab(:,end)

Check whether we have a unique solution

rank(A), rank([A b]) % 7=7, solution exists and unique: n=7

Check whether the determinant is non-zero:

det(A) % =-0.3247 – we have a unique solution

Solution methods in matlab if there is a unique solution


Both direct and iterative methods may be used for solving systems of linear equations. Let us consider now
direct methods. Consider four possibilities:

x1 = inv(A)*b; % computation of inverse (time consuming)


x2 = A\b; % nxn: Cholesky or LU factorization; mxn: QR factorization
x2 = mldivide(A,b); % the same as above
x3 = pinv(A)*b; % pseudoinverse by SVD factorization
x4 = linsolve(A,b); % LU or QR factorization can be specified
[x1 x2 x3 x4] % results are shown side by side

Check our solutions by computing the length (norm - function norm) of the residual vector if it is zero or is
within the required tolerance.

norm(A*x1-b) % 1.0904e-12
norm(A*x2-b) % 5.5695e-13
norm(A*x3-b) % 5.1348e-12
norm(A*x4-b) % 6.0157e-13

The most precise method is x2 = A\b.

Matrix decompositions

Solution of linear equations by LU decomposition


LU decomposition for A consists of three matrices (L, U, P– permutation matrix) such that

3
.

Matlab will produce an LU decomposition with pivoting for a matrix Aa with the following command:

[L U P] = lu(A)

To solve we first multiply both sides with the permutation (pivot) matrix:

Substituting for we get

Then we define , which is unknown since x is unknown. Using forward substitution, we can (easily)
solve

• (for y, where L is lower triangular and )

and then using back substitution we can (easily) solve

• (for x, where U is upper triangular)

Solve the equation system of the truss problem by LU decomposition

[L U P] = lu(A)
d = P*b;
opt1.LT=true
y = linsolve(L,d,opt1);
opt2.UT=true
x = linsolve(U,y,opt2)
residual = norm(A*x - b) % check

Cholesky decomposition
Cholesky decomposition is very similar to LU decomposition: .

Steps of the solution using :

• (where is lower triangular)


• (where L is upper triangular)

Conditions:

• Matrix A is symmetric, i.e. and


• positive definite, i.e. it has got only positive eigenvalues

Matrix A of the truss problem is symmetric and positive definite:

norm(A-A') % 1.5946

4
real(eig(A)) % [0.5000; -0.7136; -0.7136; -0.7136; 1.2136; 1.2136; 1.2136]
%chol(A) % issues an error message: Matrix must be positive definite

Instead consider the following symmetric and positive definite matrix:

A = pascal(4) % Pascal matrix composed of binomial coefficients


norm(A-A') % symmetric because A = A’
[V D] = eig(A) % compute eigenvectors and eigenvalues
min(diag(D)) % positive definite since even the smallest eigenvalue >0
L = chol(A) % Cholesky decomposition is allowed
norm(A-L'*L) % check
b = rand(4,1) % arbitrary vector b

Solution of a linear system is similar to LU decomposition (there is no permutation, hence instead of d we put
vector b, put L' in place of L and put L in place of U):

opt1.LT=true
y = linsolve(L',b,opt1);
opt2.UT=true
x = linsolve(L,y,opt2)
residual = norm(A*x - b) % check

You might also like