Preconditioners in SciPy - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output of applying Jacobi preconditioner

Given the sparse matrix A and vector b, what is the output of applying the Jacobi preconditioner to b using scipy?

SciPy

import numpy as np
from scipy.sparse import diags
from scipy.sparse.linalg import LinearOperator

A = diags([4, -1, -1], [0, -1, 1], shape=(3,3))
b = np.array([1, 2, 3])

D_inv = diags(1 / A.diagonal())
M = LinearOperator(A.shape, matvec=lambda x: D_inv.dot(x))
result = M.matvec(b)
print(result)

A [0.25 0.5 0.75]

B [4 2 3]

C [0.5 1. 1.5]

D [1 2 3]

💡 Hint

Remember the Jacobi preconditioner uses the inverse of the diagonal elements of A.

The Jacobi preconditioner multiplies the vector by the inverse of the diagonal of A. Since A's diagonal is [4,4,4], the inverse diagonal is [0.25, 0.25, 0.25]. Multiplying by b = [1,2,3] gives [0.25, 0.5, 0.75].

🧠 Conceptual

intermediate

Purpose of Preconditioners in Iterative Solvers

What is the main purpose of using a preconditioner in iterative methods for solving linear systems?

A To convert the matrix into a diagonal matrix

B To increase the size of the matrix for better accuracy

C To change the solution vector to a random guess

D To reduce the condition number of the matrix and speed up convergence

💡 Hint

Think about how preconditioners affect the matrix properties.

Preconditioners improve the matrix's condition number, making iterative solvers converge faster by transforming the system into an easier one to solve.

🔧 Debug

advanced

Identify the error in incomplete LU preconditioner code

What error will this code raise when trying to create an incomplete LU preconditioner with scipy.sparse.linalg.spilu on a singular matrix?

SciPy

import numpy as np
from scipy.sparse import csr_matrix
from scipy.sparse.linalg import spilu

A = csr_matrix([[0, 1], [0, 0]])
ilu = spilu(A)

A ValueError: matrix is not square

B TypeError: unsupported operand type(s)

C RuntimeError: factor is exactly singular

D No error, returns a valid preconditioner

💡 Hint

Consider what happens if the matrix cannot be factorized.

The matrix is singular (zero row), so spilu cannot factorize it and raises a RuntimeError indicating the factor is exactly singular.

❓ data_output

advanced

Output shape of preconditioned vector

If you apply an incomplete LU preconditioner created by spilu on a vector of length 5, what will be the shape of the output vector?

SciPy

import numpy as np
from scipy.sparse import diags
from scipy.sparse.linalg import spilu

A = diags([1, 2, 3, 4, 5], 0)
ilu = spilu(A)
v = np.array([1, 2, 3, 4, 5])
result = ilu.solve(v)
print(result.shape)

A (5,)

B (1, 5)

C (5, 1)

D (25,)

💡 Hint

The preconditioner solves a system of the same size as the input vector.

The preconditioner returns a vector of the same shape as the input vector, which is (5,).

🚀 Application

expert

Choosing a preconditioner for a large sparse symmetric positive definite matrix

You have a large sparse symmetric positive definite matrix A. Which preconditioner is generally the best choice to speed up conjugate gradient solver convergence?

A Jacobi preconditioner

B Incomplete Cholesky factorization

C Incomplete LU factorization without symmetry

D No preconditioner

💡 Hint

Consider the matrix properties and the solver used.

For symmetric positive definite matrices, incomplete Cholesky is preferred as it preserves symmetry and positive definiteness, improving conjugate gradient convergence.