Lesson 7 of 16 · Matrices
Identity and Inverse
Understand the identity matrix, the inverse as an undo button, the 2 x 2 inverse formula, singular matrices that cannot be undone, and why np.linalg.solve beats inverting in practice.
- 9 min read
- Intermediate
- Reviewed
- What the identity matrix is and why it is the 'do nothing' matrix
- What the inverse of a matrix is
- How to find the inverse of a 2 × 2 matrix
- What a singular matrix is and why it has no inverse
- Why programmers rarely compute inverses directly
The matrix that does nothing
With ordinary numbers, multiplying by 1 changes nothing: 7 × 1 = 7. Matrices have their own "1".
[ 1 0 0 ]
I₃ = [ 0 1 0 ]
[ 0 0 1 ]import numpy as np I = np.eye(3) # "eye" sounds like I print(I) x = np.array([4, 7, 2]) print(I @ x) # unchanged
The inverse: an undo button
With numbers, 5 × (1/5) = 1. The number 1/5 "undoes" multiplying by 5. Some matrices have an undo button too.
import numpy as np
A = np.array([[2.0, 1.0],
[1.0, 1.0]])
A_inv = np.linalg.inv(A)
print(A_inv)
print(np.round(A @ A_inv, 10) + 0.0) # the identity (rounding hides tiny float errors)
x = np.array([3.0, 5.0])
y = A @ x # apply A
print("y =", y)
print("back to x:", A_inv @ y) # undo itThe 2 × 2 inverse formula
For a 2 × 2 matrix there is a neat formula:
A = [ a b ] A⁻¹ = 1 [ d -b ]
[ c d ] ------- × [ -c a ]
ad - bcIn words: swap a and d, flip the signs of b and c, then divide everything by ad - bc.
When there is no inverse
The number ad - bc is so important it has a name, the determinant (its own lesson comes later). If it is 0, the formula would divide by zero.
import numpy as np
S = np.array([[1.0, 2.0],
[2.0, 4.0]]) # the second row is just double the first
det = np.linalg.det(S)
print("determinant:", round(det, 6) + 0.0)
# Always check before inverting
if abs(det) < 1e-10:
print("No inverse: this matrix is singular")
else:
print(np.linalg.inv(S))Why can it not be undone? S turns many different vectors into the same result, so you cannot tell which one you started with:
import numpy as np
S = np.array([[1, 2],
[2, 4]])
print(S @ np.array([2, 0]))
print(S @ np.array([0, 1])) # a different input, the same outputLike a blender: once two different fruits come out as the same smoothie, you cannot un-blend them.
Solving equations: use solve, not inv
Inverses are the key to solving equations of the form A x = b ("which x does A turn into b?"): mathematically, x = A⁻¹b. In code, there is a better way:
import numpy as np
A = np.array([[2.0, 1.0],
[1.0, 3.0]])
b = np.array([5.0, 10.0])
x = np.linalg.solve(A, b) # faster and more accurate than inv(A) @ b
print(x)
print(A @ x) # check: gives back bCommon mistakes
Write your own 2 × 2 inverse, and check it the safe way.
import numpy as np
A = np.array([[4.0, 7.0],
[2.0, 6.0]])
# 1. Write inverse_2x2(M) using the formula: swap a and d, flip the signs of b and c,
# divide by ad - bc. Return a NumPy array.
# 2. Print the inverse of A, rounded to 2 decimal places.
# 3. Check that A @ inverse is the identity with np.allclose, and print the result.
# Output must be exactly:
# [[ 0.6 -0.7]
# [-0.2 0.4]]
# Is identity: True
Which matrix has no inverse?
Summary
- The identity matrix I has 1s on the main diagonal and leaves vectors unchanged.
- The inverse A⁻¹ undoes A:
A⁻¹ @ A = I. - For 2 × 2: swap a and d, flip the signs of b and c, divide by ad - bc.
- A singular matrix has determinant 0, squashes information and has no inverse.
- To solve
A x = bin code, usenp.linalg.solve, notinv.
Next: the transpose, which flips a matrix over its diagonal.
Finished reading? Tick it off to track your progress.