Coordinate systems and change of basis
The same vector looks different in different bases. Change-of-basis matrices translate between coordinate systems, simplifying problems by choosing the right point of view.
Coordinates relative to a basis
If \(B = \{\mathbf{b}_1, \dots, \mathbf{b}_n\}\) is a basis of \(\mathbb{R}^n\), every \(\mathbf{x}\) has unique coordinates \([\mathbf{x}]_B = (c_1, \dots, c_n)\) with \(\mathbf{x} = c_1\mathbf{b}_1 + \dots + c_n\mathbf{b}_n\). Standard coordinates are coordinates relative to the standard basis, which is why they look like the vector's entries.
The change-of-coordinates matrix
Let \(P_B\) be the matrix whose columns are the basis vectors. Then \(\mathbf{x} = P_B[\mathbf{x}]_B\), converting B-coordinates to standard coordinates. Because the columns form a basis, \(P_B\) is invertible, and \([\mathbf{x}]_B = P_B^{-1}\mathbf{x}\) converts the other way, which in practice means solving \(P_B\mathbf{c} = \mathbf{x}\).
Changing between two bases
For two bases B and C of the same space, the change-of-coordinates matrix \(P_{C \leftarrow B}\) has columns \([\mathbf{b}_1]_C, \dots, [\mathbf{b}_n]_C\), and \([\mathbf{x}]_C = P_{C \leftarrow B}[\mathbf{x}]_B\). It is invertible, with \(P_{B \leftarrow C} = (P_{C \leftarrow B})^{-1}\). In \(\mathbb{R}^n\), it can be computed by row reducing \([C \mid B]\) to \([I \mid P_{C \leftarrow B}]\).
Coordinates in abstract spaces
Coordinates turn any n-dimensional vector space into R^n. Choosing the basis 1, t, t^2 for quadratic polynomials turns questions about polynomials, such as whether three polynomials are independent, into questions about vectors in R^3, solved with matrices and row reduction.
Why change basis
A well-chosen basis can make a problem simple. Rotating axes to align with an ellipse removes the cross term from its equation; choosing a basis of eigenvectors turns a complicated matrix into a diagonal one; choosing an orthonormal basis makes coordinates simple dot products. Much of linear algebra is about finding the best basis for a given problem.
Similarity
If a linear transformation has matrix A in standard coordinates, its matrix in basis B is \(P^{-1}AP\), where P has the basis vectors as columns. Matrices related this way are similar; they represent the same transformation in different coordinates and share its determinant, trace, rank, and eigenvalues.
Applications
Computer graphics converts between object coordinates, world coordinates, and camera coordinates many times per frame. Robotics expresses positions in the frame of each joint. Signal processing changes from a basis of time samples to a basis of sinusoids (the Fourier basis), where filtering becomes simple, and image compression uses cosine and wavelet bases.
Worked example
Let \(B = \{(1, 1), (1, -1)\}\). For \(\mathbf{x} = (5, 1)\), solve \(c_1(1, 1) + c_2(1, -1) = (5, 1)\): \(c_1 + c_2 = 5\) and \(c_1 - c_2 = 1\), so \(c_1 = 3\), \(c_2 = 2\), and \([\mathbf{x}]_B = (3, 2)\). Converting back: \(P_B[\mathbf{x}]_B = 3(1, 1) + 2(1, -1) = (5, 1)\), as expected.
Key takeaways
Coordinates relative to a basis B are the unique weights expressing x in terms of B's vectors. P_B, with basis vectors as columns, converts B-coordinates to standard ones; its inverse converts back. P_{C <- B} has columns [b_i]_C and is found by row reducing [C | B]. Coordinates turn any n-dimensional space into R^n. Choosing good bases, aligned, eigenvector, orthonormal, or Fourier, simplifies problems. In basis B a transformation's matrix is P^{-1} A P; similar matrices share determinant, trace, rank, and eigenvalues. Graphics, robotics, and signal processing change bases constantly.
Flashcard
[x]_B, the weights relative to B
Check your understanding
In the basis B = {(2, 0), (0, 3)}, what are the coordinates of x = (4, 9)?
