Solutions to Seminar 2
- a) false,
Ax = b --> x = A^-1 * b
b) true, col(s) = rows(S^-1)
c) true, because: eigenvalue is zero, singular || by gauss one col can be zeroed out
d) true, because of how matrix multiplication works
e) true, rows indep => A singular => col indep
f) false, PA permutes rows, AP permutes cols
g) true, ||x||_1 is sum, ||x||_inf
is only the largest
h) false, matrix = {{0,0},{1,0}} has determinant zero, but norm 1
i) false, S=S^T, V=V^T -> (SV)^T = V^T * S^T = V*S
j) true, ||A|| * ||A^-1||
k) false, one solution, infinite or none are the only possibilities
l) false, but true if you allow their row permutations
m) PA=LU
L and U are both non invertible or both invertivle
- a) nothing, ordering of the rows is the ordering of the equations, if
permuted the order of solutions changes as well
b) not a big deal, if the columns change the order of the solution changes
c)
Ax = b => MAx = Mb
, x remains the same. If multiplied by a singular
the amount of solutions is infinite, but one of them is the original
- a) 4
b) 6 (largest abs of elem in first col)
c) if scaled partial 5 (largest for its row)
-
A^-1 * B = X => B = AX
- b, the first one is multiplying xx^T == matrix then multiplying matrices.
The second one is multiplying a vector by a vector.
- Its transpose
-
PU=LU => A^T * P^T = U^T * L^T => A^T * P^T * P = U^T * L^T * P => A^T*x = U^T*L^T*Px = b
Meaning, if you have the LU for A, you can use it for its transpose
- a)
k(D) = max|lambda_i|/min|lambda_i| = 10^10 / 10^-10 = 10^20
, Ill cond
b) k(D)=1
the best!
c) k(D)=1
the best!
d) k(A) = ||A|| * ||A^-1|| = 6*1.5*10^7 ~= 10^8
, because inf norm:
6/4*10^7 = 1.5*10^7
, might loose 8 digits of accuracy
- no,
res = A*x_hat - b
if the matrix is well conditioned and the residual
is small, the solution is accurate
- for several runs, b
- first and second are solvable, but last one can not be made strictly
diagonally dominant. This means that the last one is possibly not
solvable with GS
- Solve with
r_k
... don't get