One vector, two addresses
Definition 3.8.1
(Coordinates in an ordered basis)
A basis is an ordered list, so coordinates are addresses relative to that list. If is a basis of , put the basis vectors into the columns of . If , then . The vector has not changed: is its -coordinate column, while is its standard-coordinate column. Because the columns form a basis, is invertible and .
Theorem 3.8.2
(Conversion between coordinate columns)
Let and be ordered bases of a finite-dimensional real vector space, and let be the matrix whose -th column is . Then for every vector , the matrix is invertible, and is the matrix whose -th column is . When is the standard basis of and , then is the basis matrix.
Proof
(Expanding in the other basis)
For two ordered bases and of a finite-dimensional real space, a conversion matrix can be constructed without choosing standard coordinates: its th column is . Expanding in the basis gives
The reverse conversion exists because each also has unique coordinates. Composing the two conversions returns every coordinate column unchanged, so the reverse matrix is . This argument also applies to a plane in , where a basis matrix is rectangular and must not be inverted as though it were square.
A linear map in two coordinate systems
Definition 3.8.3
(Matrix of a linear map in a basis)
Let be linear and let be an ordered basis of . The matrix of in is the matrix with for every . Its -th column is .
Theorem 3.8.4
(Change of basis for a linear map)
Let be linear, let and be ordered bases of , and let be the conversion matrix with . Then . In particular and are similar.
Proof
(Following one input around the diagram)
Let be linear. Its matrix in ordered basis is defined by ; the th column is . If converts coordinates into coordinates, follow an input around the diagram:
Consequently . These products are in this order because the rightmost operation occurs first. It is the same map and the same input and output vectors; only their coordinate columns have changed. For this similarity formula, each description uses the same basis for input and output; an endomorphism may also be represented using different input and output bases.
Proof X-ray
Why the inverse sits on the right in the conjugation formula
The derivation follows one input through three conversions and reads off the matrix that acts on . Two points carry the argument. First, conversion matrices act on coordinate columns, not on vectors, so the product is written in the order of the operations, rightmost first: undo to get back to coordinates, apply , then convert the output with . Second, and appear as a pair only because input and output use the same basis; with different bases the sandwich would be with , and the result is in general not similar to . The argument closes because was arbitrary, and a matrix is determined by its action on all coordinate columns. A quick column check settles the orientation: the -th column of must be .
Theorem 3.8.5
(Matrix of a map in a new basis of )
Let be the standard matrix of a linear map and let be a basis with basis matrix . Then the matrix of in is , whose -th column is . If uses a basis with matrix of for the input and a basis with matrix of for the output, the matrix is .
Proof
(Reading the formula from right to left)
A common special case starts with the standard matrix on and a new basis with basis matrix . Here converts coordinates to standard coordinates, so converts standard coordinates to coordinates. Hence
Read this from right to left: turn the new input coordinates into a standard vector, apply , then express the output in coordinates. The formula is not an instruction to transform the basis vectors twice. Similarity does not say that and have the same entries; it says that they describe the same endomorphism in different ordered bases.
The orientation can be checked column by column. The th column of is , because . This is an effective check when a formula is easy to reverse. If a linear map has different domain and codomain, the two coordinate conversions generally differ: for standard matrix , domain basis matrix and codomain basis matrix , the new matrix is , provided these are bases of their respective full coordinate spaces. Similarity is the special case for an endomorphism.
Worked examples
Worked example 3.8.6
(A basis that diagonalizes)
Let have standard matrix
Find and use it on .
Solution. With the given order,
The determinant of is , so is a basis. The two columns of are and , hence
Equivalently and , so the columns in coordinates are and . For the given , . Applying the new matrix gives , which represents . Direct multiplication gives , confirming both conversion directions.
Worked example 3.8.7
(The theorem also works for polynomials)
Let be the real polynomials of degree at most two, with bases and . Define . Find the matrix of in and check it on .
Solution. is linear and maps into itself. The vectors form a basis: their -coordinate columns give the triangular matrix
whose determinant is . This converts coordinates into coordinates; the polynomials are vectors, not ordinary numerical columns. In , , and , so . By the conversion theorem, .
We can find its columns without computing the inverse. Put . Then , , and . Therefore
Check the matrix identity directly: , so . Now , so . Directly and , so the coefficient column agrees.
Worked example 3.8.8
(The identity map in one basis and in two)
Let on and , so . With the same basis on both sides the matrix is . If instead the input uses and the output uses the standard basis, the matrix is , whose columns are the vectors themselves. For we have , and applying returns . The same map therefore has the matrix or according to whether the two bases agree, which is why the similarity formula needs one basis on both sides.
Common mistakes
Common mistake
(Reversing the conjugation)
For and , the product is a valid product of three matrices but not the matrix of in . Its second column would say , whereas direct multiplication gives . The correct order is , read from right to left.
Common mistake
(Feeding coordinates in the wrong basis)
In the polynomial example, applying to the -coordinate column gives , which is not . Computing the matrix correctly is not enough: the input column must be in the basis that the matrix expects.
Exercises
Quick check
(For find when .)
Solution · Answer
Solve : and give , . So . Check: .
Quick check
(Let on and . Find .)
Solution · Guided solution
Use the columns . Here and , so . Check with and : and .
Quick check
(If and , what is ?)
Solution · Answer
It is : , because . Similar matrices have the same determinant.