在Python中求解齐次线性方程组的非平凡解

Hey there, I’ve run into this exact issue before—most standard linear algebra solvers in Python will spit out the trivial all-zero solution for homogeneous systems by default, but finding non-zero solutions just means we need to look at the null space of your coefficient matrix. Let’s break this down with concrete steps and code examples.

First, Verify Your System Actually Has Non-Zero Solutions #

Before diving into code, double-check that your coefficient matrix doesn’t have full rank. A homogeneous system ( A\mathbf{x} = \mathbf{0} ) has non-zero solutions if and only if ( \text{rank}(A) < n ), where ( n ) is the number of unknowns (3 in your case).

Use NumPy to calculate the rank:

import numpy as np

# Replace this with your actual coefficient matrix values
A = np.array([
    [a, b, c],
    [d, e, f],
    [g, h, i]
])

rank = np.linalg.matrix_rank(A)
print(f"Rank of matrix A: {rank}")

if rank == 3:
    print("Oops, your matrix is full rank—only the trivial (all-zero) solution exists.")
else:
    print("Great, non-zero solutions exist! Let's find them.")

Use NumPy’s null_space to Find Basis Vectors for Non-Zero Solutions #

NumPy has a built-in function np.linalg.null_space() that returns the basis vectors of the null space of ( A ). Any non-zero linear combination of these vectors is a valid non-zero solution to your system.

Example code:

# Continuing with matrix A from above
null_basis = np.linalg.null_space(A)

print("Basis vectors for the null space (each column is a basis vector):")
print(null_basis)

# Generate a non-zero solution by scaling a basis vector (use any non-zero scalar)
non_zero_solution = 5 * null_basis[:, 0]  # Multiply first basis vector by 5
print("\nA non-zero solution:")
print(f"x1 = {non_zero_solution[0]}, x2 = {non_zero_solution[1]}, x3 = {non_zero_solution[2]}")

# Or combine multiple basis vectors if the null space has dimension >1
if null_basis.shape[1] > 1:
    another_solution = 2 * null_basis[:,0] + 3 * null_basis[:,1]
    print("\nAnother non-zero solution:")
    print(another_solution)

Why Your Previous Methods Only Returned the Trivial Solution #

Most general-purpose solvers like np.linalg.solve() require the coefficient matrix to be invertible (full rank) to return a unique solution. For homogeneous systems with full rank, that unique solution is the all-zero vector. But when the matrix is rank-deficient, these solvers will throw an error or default to the trivial solution—hence why you weren’t seeing non-zero solutions.

Alternative: Use SciPy’s null_space #

If you’re working with SciPy, it has a similar function scipy.linalg.null_space() that behaves almost identically to NumPy’s version. Some users prefer it for edge cases with poorly conditioned matrices:

from scipy.linalg import null_space

scipy_null_basis = null_space(A)
print("SciPy null space basis:")
print(scipy_null_basis)

Quick Note on Complex Solutions #

If your matrix has complex eigenvalues, the null space basis might include complex vectors. If you know your solution should be real, you can take the real part of the basis vectors (as long as the matrix itself is real):

real_null_basis = np.real(null_basis)
real_solution = 3 * real_null_basis[:,0]

内容的提问来源于stack exchange，提问作者hamza