The same vector gets a different list of numbers depending on which basis you measure it against. Picture the same point marked on two overlapping grids drawn at different angles: the point on the page never moves, but its row and column count changes with the grid. Change of basis is the arithmetic that converts the numbers from one grid to the numbers from another, without ever moving the point. This page states the coordinate map, builds the change of basis matrix, checks why that matrix always has an inverse, and works two examples by exact arithmetic, following the treatment in Margalit and Rabinoff's Interactive Linear Algebra, Section 2.7 and Section 5.3 on similar matrices.
Coordinates relative to a basis
Fix a basis $B = \{\mathbf{b}_1, \mathbf{b}_2, \ldots, \mathbf{b}_n\}$ of a vector space. Because a basis spans the space and is linearly independent, as the basis and dimension explainer shows, every vector $\mathbf{v}$ can be written in exactly one way as $\mathbf{v} = c_1\mathbf{b}_1 + c_2\mathbf{b}_2 + \cdots + c_n\mathbf{b}_n$. The column of weights $[\mathbf{v}]_B = (c_1, c_2, \ldots, c_n)$ is the coordinate vector of $\mathbf{v}$ relative to $B$. David Austin's Understanding Linear Algebra, Section 3.2 opens with the same picture used above: a city grid of streets gives directions such as four blocks one way and three blocks another, a coordinate system that is more natural than the usual east and north grid for that city, even though both describe the same walk.
When $B$ is the standard basis, $[\mathbf{v}]_B$ is just $\mathbf{v}$ written the usual way. Change the basis and the list of numbers changes, even though the arrow itself stays put. This is the entire content of the page: track that one fact carefully enough to compute with it.
The change of basis matrix
Collect the basis vectors of $B$, written in standard coordinates, as the columns of a matrix $P = \begin{pmatrix}\mathbf{b}_1 & \mathbf{b}_2 & \cdots & \mathbf{b}_n\end{pmatrix}$. Multiplying $P$ by the coordinate vector rebuilds $\mathbf{v}$ exactly, by the definition of matrix times vector as a combination of columns:
$$\mathbf{v} = P\,[\mathbf{v}]_B \qquad\text{and equivalently}\qquad [\mathbf{v}]_B = P^{-1}\mathbf{v}.$$
$P$ turns $B$ coordinates into standard coordinates. $P^{-1}$ runs the conversion the other way. Margalit and Rabinoff state the same pair of equations in Section 2.8 of Interactive Linear Algebra and call $P$ the matrix whose columns are a basis, used exactly this way.
Why P always has an inverse
The columns of $P$ are the vectors of a basis, so they are linearly independent, as the linear independence explainer defines the term. A square matrix with independent columns has a nonzero determinant, because dependent columns are exactly what flattens the unit square or cube to zero volume. A nonzero determinant is exactly the condition for an inverse to exist. So $P$ is invertible for the same reason the basis is a basis in the first place: nothing in it is redundant.
Worked example 1: coordinates in a tilted basis
Let $B = \left\{\begin{pmatrix}1\\1\end{pmatrix}, \begin{pmatrix}1\\-1\end{pmatrix}\right\}$ and $\mathbf{v} = \begin{pmatrix}3\\1\end{pmatrix}$. Solve $c_1\begin{pmatrix}1\\1\end{pmatrix} + c_2\begin{pmatrix}1\\-1\end{pmatrix} = \begin{pmatrix}3\\1\end{pmatrix}$, which is $c_1 + c_2 = 3$ and $c_1 - c_2 = 1$. Adding the two equations gives $2c_1 = 4$, so $c_1 = 2$, and then $c_2 = 1$. So $[\mathbf{v}]_B = \begin{pmatrix}2\\1\end{pmatrix}$.
Now build $P$ from the same basis vectors, as columns: $P = \begin{pmatrix}1 & 1\\1 & -1\end{pmatrix}$. Its determinant is $1(-1) - 1(1) = -2$, nonzero as expected. Using the formula for a 2 by 2 inverse, $P^{-1} = \dfrac{1}{-2}\begin{pmatrix}-1 & -1\\-1 & 1\end{pmatrix} = \begin{pmatrix}\tfrac12 & \tfrac12\\\tfrac12 & -\tfrac12\end{pmatrix}$. The 2 by 2 matrix calculator will match this inverse once you type in $P$.
Changing between two nonstandard bases
Suppose $C$ is a second basis with its own change of basis matrix $Q$, built from $C$'s vectors the same way $P$ was built from $B$'s. Since $\mathbf{v} = P\,[\mathbf{v}]_B$ and also $\mathbf{v} = Q\,[\mathbf{v}]_C$, the two right hand sides equal the same $\mathbf{v}$, so:
$$[\mathbf{v}]_C = Q^{-1}P\,[\mathbf{v}]_B.$$
Standard coordinates are the bridge that makes this work, even though the formula never writes them down. Take $C = \left\{\begin{pmatrix}2\\0\end{pmatrix}, \begin{pmatrix}0\\1\end{pmatrix}\right\}$, so $Q = \begin{pmatrix}2 & 0\\0 & 1\end{pmatrix}$ and $Q^{-1} = \begin{pmatrix}\tfrac12 & 0\\0 & 1\end{pmatrix}$. For the same $\mathbf{v} = (3,1)$ from worked example 1, solving $2c_1 = 3$ and $c_2 = 1$ directly gives $[\mathbf{v}]_C = \left(\tfrac32, 1\right)$.
A linear map in a new basis: similar matrices
If $A$ is the matrix of a linear map $T$ in standard coordinates, the matrix of the same map written in $B$ coordinates is $[T]_B = P^{-1}AP$. Margalit and Rabinoff's Interactive Linear Algebra calls $A$ and $P^{-1}AP$ similar matrices and shows that similar matrices describe the same map, only in different coordinate systems, and that they always share the same characteristic polynomial, hence the same eigenvalues. Choosing $B$ to be a basis of eigenvectors is exactly what makes $P^{-1}AP$ diagonal, the payoff behind diagonalization: a diagonal matrix just scales each basis direction, which is as simple as a matrix can act. That page states when $A = PDP^{-1}$ exists and checks a 2 by 2 and a 3 by 3 factorization by exact arithmetic.
Worked example 2: a map that becomes diagonal
Let $A = \begin{pmatrix}2 & 1\\1 & 2\end{pmatrix}$ and reuse $B = \left\{\begin{pmatrix}1\\1\end{pmatrix}, \begin{pmatrix}1\\-1\end{pmatrix}\right\}$ and $P = \begin{pmatrix}1 & 1\\1 & -1\end{pmatrix}$ from worked example 1. Multiply $A$ by each basis vector directly: $A\begin{pmatrix}1\\1\end{pmatrix} = \begin{pmatrix}2+1\\1+2\end{pmatrix} = \begin{pmatrix}3\\3\end{pmatrix} = 3\begin{pmatrix}1\\1\end{pmatrix}$, and $A\begin{pmatrix}1\\-1\end{pmatrix} = \begin{pmatrix}2-1\\1-2\end{pmatrix} = \begin{pmatrix}1\\-1\end{pmatrix} = 1\begin{pmatrix}1\\-1\end{pmatrix}$. Each basis vector of $B$ is an eigenvector of $A$, with eigenvalues $3$ and $1$. The 2 by 2 eigenvalue calculator reports those same two roots from the trace $4$ and determinant $3$ of $A$.
Orthonormal bases: P inverse for free
When the columns of $P$ are unit vectors that are pairwise orthogonal, the Gram Schmidt process explainer shows that $P^{-1} = P^{\mathsf T}$, so the inverse is just the transpose, with no row reduction at all. Orthogonal in that sentence means the ordinary dot product. A different inner product moves which pairs count as perpendicular. See inner product spaces. Take $\mathbf{b}_1 = \begin{pmatrix}3/5\\4/5\end{pmatrix}$ and $\mathbf{b}_2 = \begin{pmatrix}-4/5\\3/5\end{pmatrix}$, both unit vectors with dot product $\tfrac{3}{5}\!\cdot\!\left(-\tfrac{4}{5}\right) + \tfrac{4}{5}\!\cdot\!\tfrac{3}{5} = 0$, so they are orthonormal. With $P = \begin{pmatrix}3/5 & -4/5\\4/5 & 3/5\end{pmatrix}$ and $P^{\mathsf T} = \begin{pmatrix}3/5 & 4/5\\-4/5 & 3/5\end{pmatrix}$:
Checkpoints
- For $B = \{(1,0), (1,1)\}$ and $\mathbf{v} = (4,1)$, solve for $[\mathbf{v}]_B$ directly, then build $P$ and check with $P^{-1}\mathbf{v}$.
- Explain in one sentence why a set of vectors that fails to span the space could never be used as the columns of a change of basis matrix.
- For the $A$ and $B$ in worked example 2, compute $A\begin{pmatrix}2\\0\end{pmatrix}$ directly, then check it against $3\begin{pmatrix}1\\1\end{pmatrix} + \begin{pmatrix}1\\-1\end{pmatrix}$ using linearity, since $(2,0) = (1,1)+(1,-1)$.
Common mistakes
- Writing the basis vectors as rows instead of columns. $P$ is built with each basis vector as a column. Rows give the transpose of $P$, which is only the correct inverse when the basis happens to be orthonormal.
- Mixing up $P$ and $P^{-1}$. $P$ turns $B$ coordinates into standard coordinates. $P^{-1}$ runs the other way. Multiplying a standard coordinate vector by $P$ again, instead of by $P^{-1}$, gives a vector with no clear meaning.
- Forgetting the order in $P^{-1}AP$. Matrix multiplication does not commute, so $AP^{-1}P$ or $PAP^{-1}$ is not the same computation and usually gives a different, meaningless matrix. The $P^{-1}$ belongs on the left and the $P$ on the right.
- Assuming every basis gives a diagonal $P^{-1}AP$. Worked example 2 used a basis made of eigenvectors of $A$ on purpose. A basis chosen for any other reason usually leaves $P^{-1}AP$ full of nonzero entries, still equal in trace and determinant to $A$, but not diagonal.
Related reading
- Basis and dimension of a vector space explained and what is a vector space?
- Linear independence explained and what a determinant means geometrically
- Eigenvalues in plain language, diagonalization explained, and the Gram Schmidt process explained
- Orthogonal projection and least squares explained and the rank nullity theorem explained
- Linear algebra topic map
- Glossary: basis, matrix, linear independence, all explainers and the sources page
Related tools
The 2 by 2 matrix calculator finds the determinant and inverse used to build $P^{-1}$ in worked example 1. The 2 by 2 eigenvalue calculator checks the eigenvalues used in worked example 2. The RREF calculator is a general way to solve the coordinate equations for a basis in more than two dimensions. The Gram Schmidt calculator builds an orthonormal basis, so that the orthonormal shortcut in this page applies.
Sources and further reading
- Dan Margalit and Joseph Rabinoff, Interactive Linear Algebra, Section 2.7 Basis and Dimension (the basis theorem and B coordinates). textbooks.math.gatech.edu (accessed 18 Nov 2026)
- Dan Margalit and Joseph Rabinoff, Interactive Linear Algebra, Section 5.3 Similarity (the change of basis matrix, $v = P[v]_B$, and $A = PDP^{-1}$ style similarity). textbooks.math.gatech.edu (accessed 18 Nov 2026)
- David Austin, Understanding Linear Algebra, Section 3.2 Bases and coordinate systems. understandinglinearalgebra.org (accessed 18 Nov 2026)
Last reviewed: 20 November 2026.
Common questions
What does the coordinate vector [v]_B actually mean?
It is the list of weights that rebuild v from the basis B in order. If B has basis vectors b1 through bn and v equals c1 times b1 plus c2 times b2 and so on, then the coordinate vector is the column (c1, c2, up to cn). It depends on which basis you chose and on the order you listed its vectors in. The vector v itself does not change. Only its address inside a particular basis changes.
Why is the change of basis matrix always invertible?
Its columns are the vectors of a basis, written in standard coordinates, and a basis is linearly independent by definition. A square matrix with linearly independent columns has a nonzero determinant, so it has an inverse. If any column were a repeat of the others, the determinant would be zero, but that would mean the original set was not a basis in the first place.
How do I convert coordinates between two bases that are both nonstandard?
Build the change of basis matrix for each one, P for basis B and Q for basis C, using standard coordinates as the bridge. Since v equals P times the B coordinates and also equals Q times the C coordinates, the C coordinates equal Q inverse times P times the B coordinates. You never need the standard coordinates written out separately. The two matrices carry you there and back.
What does an orthonormal basis do to the change of basis matrix?
When the columns of P are unit vectors that are pairwise at a right angle, P transpose equals P inverse, so finding the inverse is just copying rows into columns with no arithmetic at all. The Gram Schmidt process is the usual way to turn an ordinary basis into an orthonormal one before using this shortcut.