4.3. Change of Basis#

If we work with the basis \(\mathcal B\) the transformation takes on a very simple form. Any vector \(\vect\) can be written as

In this section we will explain how such a change of basis works.

4.3.2. Coordinates with Respect to a Basis#

We start with a proposition.

If \(\mc = \_1, \vect_2,\ldots,\vect_m\>\) is a basis for a subspace \(S\) in \(\R^n\) , then any vector \(\vect\) in \(S\) can written as a linear combination of \(\vect_1, \vect_2,\ldots,\vect_m\) in a unique way, i.e.

for unique constants \(c_1,c_2,\ldots,c_m\) in \(\R\) .

Proof. From the definition of a basis it follows that

for some constants \(c_1,c_2,\ldots,c_m\) .

To prove that the linear combination is unique, suppose that also

for constants \(d_1,d_2,\ldots,d_m\) . Then we have to show that in fact

This can be done as follows: subtracting the two expressions for \(\vect\) gives

So, we have a linear combination of the vectors \(\vect_i\) equal to the zero vector \(\vect\) .

From the linear independence of the vectors \(\vect_1, \vect_2,\ldots,\vect_m\) it then follows that all the coefficients \((c_i-d_i)\) must be 0. Thus we find

from which we can conclude that

and we are done.

In this subsection we will be interested mainly in the case where the subspace \(S\) is the whole of \(\R^n\) .

Proposition 4.3.1 shows that the following definition makes sense.

Let \(\mc = \_1, \vect_2,\ldots,\vect_m\>\) be a basis for a subspace \(S\) in \(\R^n\) , and \(\vect\) a vector in \(S\) . Then the coefficients \(c_1,c_2, \ldots, c_m\) for which

are called the coordinates of the vector \(\vect\) with respect to the basis \(\mc\) ,

[\vect]_ <\mc B>= \begin c_1 \\ c_2 \\ \vdots \\ c_m\end \end\]

is called the coordinate vector of \(\vect\) with respect to the basis \(\mc\) .

We find the coordinates of the vectors \(\vect= \begin 6 \\ -2\end\) and \(\vect=\begin -2 \\ 4\end\) with respect to the basis \(\mc = \_1, \vect_2 \> = \left\ < \begin1 \\ 3\end, \begin 2 \\ 1\end \right\>\) of \(\R^2\) .

For \(\vect\) we have to solve the vector equation

\[\begin c_1\begin 1 \\ 3\end + c_2\begin 2 \\ 1\end\,\,=\,\,\begin6 \\ -2\end. \end\]

The solution \(c_1 = -2,\,c_2 = 4\) gives the answer:

\[\begin [\vect]_ <\mc B>= \begin -2 \\ 4\end. \end\]

Likewise the vector \(\vect\) has the coordinate vector \([\vect]_ <\mc B>= \begin 2 \\ -2\end\) .

Figure 4.3.2 may help to see what is going on geometrically.

A note of warning. Strictly speaking a basis is an ordered set of vectors. By this we mean that the set \(\mathcal = \_1,\vect_2\>\) and the set \(\mathcal' = \_2,\vect_1\>\) are the same. However, as bases we should consider them as different. This becomes important when we work with coordinates.
\([\vect]_<\mathcal> = \beginc_1\\c_2 \end\) means that \(\vect = c_1\vect_1+c_2\vect_2\) ,
whereas \([\vect]_<\mathcal'> = \beginc_1\\c_2 \end\) should be interpreted \(\vect = c_1\vect_2+c_2\vect_1\) .
To avoid ambiguities like this the way out is to use the notation

\[ \mathcal = \big(\vect_1,\vect_2, \ldots, \vect_n\big), \]

when we talk about a basis. However, we won’t, as is quite customary. In case of a basis we will always tacitly assume that the vectors are ordered as they are written down.

Recall that the standard basis of \(\R^n\) is given by

With respect to the standard basis it is very easy to find coordinates.

If \(\vect = \begin a_1 \\ a_2 \\a_3\end\) is an arbitrary vector in \(\R^3\) , then

\[\begin \vect = a_1 \begin 1 \\ 0 \\ 0\end + a_2 \begin 0 \\ 1 \\ 0\end + a_3 \begin 0 \\ 0 \\ 1\end = a_1 \vect_1 + a_2 \vect_2 + a_3 \vect_3. \end\]

This means that

\[\begin [\vect]_<\mc> = \begin a_1 \\ a_2 \\a_3\end = \vect. \end\]

So with respect to the standard basis any vector in \(\R^3\) is at the same time its coordinate vector.

In fact, this last observation holds more general. That is the content of the following proposition.

For each vector \(\vect\) in \(\R^n\)

\[ [\vect]_<\mc> = \vect, \]

where \(\mc\) is the standard basis of \(\R^n\) .

In Example 4.3.1 we have seen that finding the coordinates of a vector with respect to the basis \(\mc = \_1, \vect_2 \> = \left\ < \begin1 \\ 3\end, \begin 2 \\ 1\end \right\>\) amounts to solving a vector equation. The converse, finding \([\vect]_<\mc>\) when \([\vect]_<\mc>\) is given, requires less work. The following example illustrates this.

Suppose that with respect to the basis \(\mc = \left\ < \begin1 \\ 3\end, \begin 2 \\ 1\end \right\>\) it is given that

\[\begin [\vect]_<\mc> = \begin f_1 \\ f_2\end. \end\]

Then it follows that

The rule to go from basis \(\mc\) to the standard basis can thus be written as

(4.3.1)# \[[\vect]_ <\mc E>= P[\vect]_<\mc>, \quad \text\quad P = \left[\,\vect_1 \,\,\vect_2 \,\right].\]

The procedure of Example 4.3.3 can be generalized to bases in \(\R^n\) . Before we state how, we first define another useful concept.

Let \( <\mathcal B>= \_1, \ldots, \vect_n \>\) be a basis of \(\R^n\) . The change-of-coordinates matrix from \(<\mathcal B>\) to \(<\mathcal E>\) is the matrix

\[ P_<\mc> = \left[\,\vect_1\,\,\vect_1\,\, \ldots\,\, \vect_n\, \right]. \] \[ \vect = P_<\mc> [\vect]_<\mc B>. \] \[\begin [\vect]_ <\mc B>= \begin p_1 \\ p_2 \\ \vdots \\ p_n \end. \end\]

This is shorthand for

\[ \vect = p_1\vect_1+ p_2\vect_2+\ldots + p_n\vect_n, \]

and this linear combination of \(n\) vectors in \(\R^n\) can be written as a matrix-vector product:

\[\begin \vect = \left[\begin \vect_1 \,\,\vect_2 \,\,\ldots\,\, \vect_n \end\right]\begin p_1 \\ p_2 \\ \vdots \\ p_n \end = P_<\mc> [\vect]_<\mc B>. \end\]

Show that every change-of-coordinates matrix \(P_<\mc>\) is invertible.

Solution to Exercise 4.3.1 (click to show) \[ P_<\mc> = [\,\vect_1\,\,\vect_2\,\,\ldots\,\,\vect_n\,]. \]

This is an \(n \times n\) matrix with \(n\) linearly independent columns, so it is an invertible matrix.

The following proposition is an immediate consequence of Proposition 4.3.3 and Exercise 4.3.1 .

Suppose \(\mc\) is a basis for \(\R^n\) , and \(P_<\mc>\) is the change-of-coordinates matrix. Then the coordinates of any vector \(\vect\) with respect to basis \(\mc\) are given by

\[ [\vect]_ <\mc B>= P^_<\mc> \vect. \]

Have another look at Example 4.3.1 where we found the coordinates of the vector \( \vect= \begin 6 \\ -2\end\) with respect to the basis \(\mc = \_1, \vect_2 \> = \left\ < \begin1 \\ 3\end, \begin 2 \\ 1\end \right\>\) .

The change-of-coordinates matrix from basis \(\mc\) to the standard basis is given by

\[\begin P_<\mc> = \begin \vect_1 & \vect_2 \end = \begin 1& 2 \\ 3&1\end. \end\]

The inverse matrix then becomes

\[\begin P^_<\mc> = \begin 1& 2 \\ 3&1\end^ = -\frac15 \begin 1& -2 \\ -3&1\end. \end\]

So the coordinates of the vector \(\vect\) can now also be found as follows:

We conclude this subsection with an example from physics.

Suppose \((x,y,z,t)\) and \((x',y',z',t')\) are the coordinates of an event in two frames with the origins coinciding at \(t = t'= 0\) where the primed frame is seen from the unprimed frame as moving with speed \(v\) along the \(x\) -axis.

The connection between two systems can be rewritten by a four-dimensional ‘space-time’ coordinate transformation

\[\begin \begin x'= \gamma (x - vt) \\ y'= y \\ z'= z \\ t'= \gamma \left( t - \dfrac x \right) \end \end\]

Here \(c\) is the speed of light, and \(\gamma = \left(\sqrt>\right)^\) the so-called Lorentz factor.

Systems that move with a constant velocity with respect to each other are called inertial systems. In each reference frame, an observer can use a local coordinate system (usually Cartesian coordinates in this context) to measure lengths, and a clock to measure time intervals. An event is something that happens at a point in space at an instant of time, or more formally a point in space-time. The transformation connects the space and time coordinates of an event as measured by an observers in the different frames.

4.3.3. The Matrix of a Linear Transformation with Respect to a Basis#

In Section 3.1 we have seen that every linear transformation \(T:\R^n \to \R^m\) can be represented by its standard matrix

(4.3.2)# \[A = \begin T(\vect_1) & T(\vect_2) & \ldots & T(\vect_n) \end.\]

This means that for each vector \(\vect\) in \(\R^n\) it holds that

\[ T(\vect) = A\,\vect. \]

In this way we have expressed everything into coordinates with respect to the standard bases \(\mc\) and \(\mc'\) for \(\R^n\) and \(\R^m\) respectively.

We will see that formula (4.3.2) can be generalized almost automatically to arbitrary bases for \(\R^n\) and \(\R^m\) .

Suppose that \(T:\R^n \to \R^m\) is a linear transformation, that \(\mc =\_1, \ldots, \vect_n\>\) is a basis for \(\R^n\) and \(\mc\) a basis for \(\R^m\) . Then for each vector \(\vect\) in \(\R^n\) :

Proof. We use the definition of the coordinate vector and the linearity of the transformation. In fact we can play copycat with the proof of Theorem 3.1.1 in the section of linear transformations.

Indeed, suppose that \(\vect\) is a vector in \(\R^n\) , and that

\[\begin [\vect]_<\mc> = \begin f_1 \\ \vdots \\ f_n \end, \quad \text \quad \vect = f_1\vect_1 + f_2\vect_2 + \ldots + f_n\vect_n. \end\]

Because of the linearity of \(T\) we then have

(4.3.3)# \[T(\vect) = f_1T(\vect_1) + f_2T(\vect_2) + \ldots + f_nT(\vect_n).\]

Now we take the coordinate vectors with respect to basis \(\mc\) :

(4.3.4)# \[[T(\vect)]_<\mc> = f_1[T(\vect_1)]_<\mc> + f_2[T(\vect_2)]_<\mc> + \ldots + f_n[T(\vect_n)]_<\mc>. \]

Noting that the term on the right side of Equation (4.3.4) is a linear combination of \(m\) -vectors, we can write the identity as

Prove the identity

\[ [c_1\vect_1 +c_2\vect_2+ \ldots + c_n\vect_n ]_<\mc> = c_1[\vect_1 ]_<\mc> + c_2[\vect_2 ]_<\mc> + \ldots + c_n[\vect_n ]_<\mc> \]

that is used to go from Equation (4.3.3) to Equation (4.3.4) in the proof of Proposition 4.3.5.

Hint: first show the correctness of the identities

Proposition 4.3.5 is fairly general. We will mostly use it in the situation where the linear transformation goes from \(\R^n\) to itself, and where we use the same basis \(\mc\) in the domain and the codomain. Before we do so, we introduce a new notation for the general case.

Let \(T:\R^n \to \R^m\) be a linear transformation, and let \(\mc = \_1,\ldots,\vect_n\>\) be a basis for \(\R^n\) , \(\mc\) a basis for \(\R^m\) . Then the matrix of \(T\) with respect to bases \(\mc\) and \(\mc\) is defined by

In the case where \(n = m\) and \(\mc = \mc\) we will use the slightly simpler notation

and call this simply the matrix of \(T\) with respect to basis \(\mathcal\) .

If we let \(\mc\) be the standard basis \(\mc\) of \(\R^n\) and \(\mc\) is the standard basis \(\mc\) of \(\R^m\) we get the following:

This is the “good old” standard matrix as in Section 3.1.4 .

Let us have a second look at the transformation mentioned in the introduction, i.e., the reflection in the plane along the line \(3x-2y=0\) . We found that the reflection behaves in a very simple way with respect to the basis

\[\begin \mc = \_1, \vect_2 \> = \left\ 2\\3 \end, \,\begin 3\\-2 \end \right\>. \end\]

Recall that this basis was chosen so that \(\vect_1\) lies on \(\mathcal\) and \( \vect_2\) is perpendicular to \(\mathcal\) , from which we deduced that

\[ T(\vect_1) = \vect_1 \quad \text \quad T(\vect_2) = -\vect_2. \]

From this we can immediately write down the matrix of \(T\) with respect to \(\mc\) :

\[\begin [T]_<\mathcal> = \begin 1 & 0 \\ 0 & -1 \end. \end\]

We will find the matrix of the linear transformation \(\) sending \(\vect_1 = \begin 1\\3\end\) to \(\begin 5 \\ 5\end\) and \(\vect_2 = \begin 2\\1\end\) to \(\begin 1 \\ -2\end\) with respect to the basis \(\mathcal =\< \vect_1, \vect_2\>\) .

The only thing that remains to be done before we can write down

\[ [T]_<\mc> = \big[\, [T(\vect_1 )]_<\mc>\,\,\,[T(\vect_2 )]_<\mc>\,\big] \]

is to find the coordinates of the given images with respect to the basis \(\mathcal\) . This can be done as in Example 4.3.1 or by using the change-of-coordinates matrix as in Example 4.3.4:

Grasple Exercise 4.3.1

Finding \([T]_<\mathcal>\) for a ‘nice’ linear transformation.

Show/Hide Content

In the next example we will find the matrix of the rotation \(R\) about the origin through an angle \(\frac13\pi\) . In Section 3.3 we have already seen the standard matrix of this linear transformation, involving sines and cosines. As the example shows a change of basis may lead to an interesting alternative.

Let \(\mc\) be the basis

\[\begin \left\_1, \vect_2 \right\> = \left\ < \begin2 \\ 0 \end, \begin -1 \\ \sqrt \end \right\>. \end\]

Note that \(\vect_1\) and \( \vect_2\) are two vectors of length 2 subtending an angle \(\frac23\pi\) . See Figure 4.3.3 . The vector \(\vect_1+\vect_2 = \begin 1 \\ \sqrt \end\) , is also indicated in the figure.

Now let us find the matrix of the rotation \(R\) with respect to basis \(\mc\) . From the figure we read off

\[ R(\vect_1) = \vect_1+\vect_2 \quad\text \quad R(\vect_2) = -\vect_1. \]

From this we can immediately write down the matrix of \(R\) with respect to the \(\mc\) -basis:

In Example 4.3.6 and Example 4.3.8, the reflection and the rotation, we came up first with bases that fitted the geometric context. And then we found the (very simple) matrix with respect to these bases. However, usually we prefer to have the standard matrix. In the last section we will see how a change of basis affects the matrix of a linear transformation.

4.3.4. The relation between \([T]_<\mc>\) and \([T]_<\mc>\) #

We have seen how to convert vectors from one coordinate system (i.e., basis) to another, and also how to construct the matrix of a linear transformation with respect to an arbitrary basis. In this section we will present a ready-made formula that connects the matrices with respect to two different bases. In this subsection we will restrict ourselves to the common situation of a linear transformation from \(\R^n\) to itself, where one of the bases is the standard basis.

We start with an example that illustrates the underlying ideas.

Consider the linear transformation \(T:\R^2\to\R^2\) with the standard matrix \(A = \begin 1 & 4 \\ 2 & 1\end\) , i.e.

\[\begin T(\vect) = A\vect = \begin 1 & 4 \\ 2 & 1\end\vect. \end\]

We want to find the matrix of \(T\) with respect to the basis

\[\begin \mc = \left\_1,\vect_2\right\> = \left\ 2\\ -1 \end,\, \begin 1\\ 1 \end\right\>. \end\]

Let \(P_<\mc>\) denote the change-of-coordinates matrix.

So we have, for each vector \(\vect\) :

\[\begin \vect = [\vect]_<\mc> = P [\vect]_<\mc>,\quad\text\, P = \PB = \begin 2 & 1\\ -1 & 1\end. \end\]

To obtain the matrix \([T]_<\mc>\) we need the coordinate vectors

\[ [T(\vect_1)]_<\mc> \quad \text \quad [T(\vect_2)]_<\mc>. \]

Let us start with the vector \(\vect_1\) :

For the second basis vector \(\vect_2\) we could proceed likewise.

Alternatively, we can start from the coordinates of an arbitrary vector with respect to the basis \(\mc\) , and then step by step work our way to the matrix \([T]_<\mc>\) such that:

\[ \left[T(\vect) \right]_<\mc> = [T]_<\mc>\,\left[\vect \right]_<\mc>. \]

1. First we find the coordinates of the vector \(\vect\) with respect to the standard basis \(\mc\) :

\[ \vect = \left[\vect\right]_<\mc> = P\left[\vect\right]_<\mc>. \] \[ T(\vect) = A \vect = A\,P\left[\vect\right]_<\mc>. \] \[ \left[T(\vect)\right]_<\mc> = P^ T(\vect)= P^AP\left[\vect\right]_<\mc>. \]

We see that the matrix \(P^AP\) does the job:

(4.3.5)# \[[T]_<\mc> \,=\, P^AP.\]

For the given transformation and basis we find

Verify that the first column indeed corresponds to the image of \(\vect_1\) as we found earlier.

Note that the product \(PAP^\) in Equation (4.3.5) cannot be simplified to \(PP^A\) , which would be equal to \(A\) itself!

Example 4.3.9 contains an informal proof, or rather, a “proof by example”, of the following proposition.

Suppose \(\mc\) is a basis of \(\R^n\) , and \(P = P_<\mc>\) is the change-of-coordinates matrix. If \(T:\R^n \to \R^n\) is a linear transformation with standard matrix \(A\) , then the matrix of \(T\) with respect to the basis \(\mc\) is given by

(4.3.6)# \[[T]_<\mc> = P^AP.\]

The following diagram illustrates what is going on:

Note that Equation (4.3.6) is equivalent to

\[ A = P [T]_<\mc>P^, \quad \text\, [T]_<\mc> = P_<\mc>[T]_<\mc>\left(P_<\mc>\right)^, \]

and as such gives a quick way to go from an alternative basis to the standard basis.

Suppose \(P\) is an invertible matrix.

Show that for matrices \(A\) and \(B\)

\[ B = P^<-1>A\,P \,\, \iff \,\,A = P\,B\,P^<-1>. \] Solution to Exercise 4.3.3 (click to show)

Starting from \(B = P^A\,P\) multiply both sides from the left by \(P\) and from the right by \(P^\) , and use that \(P^P = PP^ = I\) :

\[ B = P^<-1>AP \iff PBP^ <-1>= P(P^<-1>AP)P^ <-1>= PP^<-1>APP^ <-1>= A. \]

We want to find the matrix of the orthogonal projection \(T\) in \(\R^3\) onto the plane

\[ V: x - y - 2z = 0. \]

A suitable basis here is a basis \(\mc = \_1, \vect_2,\vect_3\>\) where the first two vectors lie in the plane, and the third vector is perpendicular to the plane. See Figure 4.3.4 .

For instance, we can take

The orthogonal projection maps \(\vect_3\) to the origin and leaves the other two basis vectors invariant:

\[ T(\vect_1) = \vect_1, \quad T(\vect_2) = \vect_2, \quad T(\vect_3) = \vect. \] \[\begin [T]_<\mc> = \begin1 & 0 & 0\\0& 1& 0\\0&0&0 \end, \end\]

and we can write down the standard matrix in one stroke:

This can be evaluated to yield

(4.3.8)# \[\begin [T]_<\mc> = \frac16 \begin5 & 1 & 2\\1& 5& -2\\2&-2&2 \end.\end\]

As was shown in Proposition 3.3.4 of Section 3.3.1 the matrix \(A\) of a projection has to satisfy the identity \(A^2=A\) . Show that the matrix \([T]_<\mc>\) of Example 4.3.10 indeed satisfies this identity.

(To verify this you don’t have to compute the matrix as given by Equation (4.3.8).)

Solution to Exercise 4.3.4 (click to show)

Obviously the matrix \(B = [T]_<\mc> = \begin1 & 0 & 0\\0& 1& 0\\0&0&0 \end\) has the property \(B^2 = B\) .

It follows that the matrix \(A = P_<\mathcal> B P_<\mathcal>^ = PBP^\) has it too:

\[ A^2 = PBP^<-1>PBP^ <-1>= PB^2 P^ <-1>= PBP^ <-1>= A. \]

In the last subsection we will generalize the formula to the more general situation of a linear transformation from \(\R^n\) to \(\R^m\) and arbitrary bases \(\mc,\mc'\) for \(\R^n\) , and bases \(\mc,\mc'\) for \(\R^m\) .

4.3.5. General Transformation Formula for Matrices of Linear Transformation#

In the previous subsections we studied the relations of the coordinates with respect to the standard basis and one other alternative basis \(\mc\) . In this section we go one step further and omit the restriction that one of the bases is the standard basis. We will see that the generalizations are pretty straightforward.

First of all we will extend the notion of a change-of-coordinates matrix.

Let \( <\mathcal B>= \_1, \ldots, \vect_n \>\) and \( <\mathcal C>= \_1, \ldots, \vect_n \>\) be two bases of \(\R^n\) . The change-of-coordinates matrix from \(<\mathcal B>\) to \(<\mathcal C>\) is the matrix

Note that in the situation where \(\mc\) is the standard basis, we have that

\[ P_ <\mc\leftarrow \mc> = P_ <\mc\leftarrow \mc> = P_<\mc>. \]

Again, as the name suggests, the change-of-coordinates matrix can be used to switch from coordinates with respect to one basis to coordinates with respect to another basis.

If \( <\mathcal B>= \_1, \ldots, \vect_n \>\) and \( <\mathcal C>= \_1, \ldots, \vect_n \>\) are two bases of \(\R^n\) , then for any vector \(\vect\) in \(\R^n\) it holds that

\[ [\vect]_<\mc> = \CoBmatrix [\vect]_<\mc>. \]

Proof. Again we will make use of the identity

\[ [p_1\vect_1+\ldots + p_k\vect]_ <\mc B>= p_1[\vect_1]_<\mc B>+ \ldots +p_k[\vect_k]_<\mc B>. \]

Thus, suppose that

\[\begin [\vect]_ <\mc B>= \begin p_1 \\ \vdots \\ p_n \end \quad \text \quad \vect = p_1\vect_1+ \ldots + p_n\vect_n. \end\]

\[ [\vect]_ <\mc C>= [p_1\vect_1+ \ldots + p_n\vect_n]_ <\mc C>= p_1[\vect_1]_<\mc C>+ \ldots + p_n[\vect_n]_<\mc C>. \]

The last expression is a linear combination of \(n\) vectors in \(\R^n\) and consequently can be written as a matrix-vector product:

as this is exactly how \(\CoBmatrix[\vect]\) was defined in Definition 4.3.2.

Suppose \(\mc\) and \(\mc\) are two bases of \(\R^n\) . Show that the change-of-coordinates matrix \(P_<\mc\leftarrow\mc>\) is invertible, and that

\[ P_<\mc\leftarrow\mc> = \left(P_<\mc\leftarrow\mc>\right)^. \]

Suppose \(\mc, \mc\) and \(\mc\) are three bases of \(\R^n\) . What is the relation between the change-of-coordinates matrices \( P_<\mc\leftarrow\mc>, \quad P_<\mc\leftarrow\mc> \quad \text \quad P_<\mc\leftarrow\mc>\) ?

Solution to Exercise 4.3.6 (click to show)

The defining relation of a change-of-coordinates matrix like \(P_<\mc\leftarrow\mc>\) is that for every vector \(\vect\) in \(\R^n\) we have

\[ P_<\mc\leftarrow\mc>[\vect]_ <\mc D>= [\vect]_<\mc B>. \]

As such, we see that for every vector \(\vect\) it also holds that

From these identities it follows that

\[ P_<\mc\leftarrow\mc> = P_<\mc\leftarrow\mc> P_<\mc\leftarrow\mc>. \]

Note how nicely the notation for change-of-coordinates matrices comes out here.

With the chosen notation for the change-of-coordinates matrix, the transformation formula for the matrices of a linear transformation with respect to different bases becomes very natural as well. As is stated in the next theorem.

Suppose \(\mc\) and \(\mc'\) are two bases of \(\R^n\) , and \(\mc\) and \(\mc'\) are two bases of \(\R^m\) . If \(T:\R^n \to \R^m\) is a linear transformation, then the relation between the two matrices \(A = [T]_<\mc\leftarrow \mc>\) and \(A' = [T]_<\mc'\leftarrow \mc'>\) is as follows

The following diagram, strikingly similar to the diagram in Proposition 4.3.6, illustrates what is going on.

Proof. To find \([T(\vect)]_<\mc'>\) when \( [\vect]_<\mc'>\) is given one can either

• Multiply by \(A'= [T]_<\mc'\leftarrow \mc'> \)

Find the coordinates of \(\vect\) with respect to basis \(\mc\) :

\[ [\vect]_<\mc> = P_<\mc\leftarrow\mc> [\vect]_<\mc'> \] \[ [T(\vect)]_<\mc> = A[\vect]_<\mc> = AP_<\mc\leftarrow\mc> [\vect]_<\mc'> \]

Since we have that

for each coordinate vector \([\vect]_<\mc'> \) , that is, for each vector in \(\R^n\) , it follows that the two matrices

\[ [T]_<\mc'\leftarrow \mc'> \quad \text \quad \left( P_<\mc\leftarrow\mc'>\right)^AP_<\mc\leftarrow\mc> \]

4.3.6. Grasple Exercises#

Grasple Exercise 4.3.2

To find \([\vect]_<\mathcal>\) for some vector \(\vect\) and basis \(\mathcal\) in \(\R^2\) .

Show/Hide Content

Grasple Exercise 4.3.3

One more like the previous.

Show/Hide Content

Grasple Exercise 4.3.4

To find \([\vect_i]_<\mathcal>\) for several vectors \(\vect_i\) in \(\R^2\) .

Show/Hide Content

Grasple Exercise 4.3.5

Finding \([\vect]_<\mathcal>\) using a figure.

Show/Hide Content

Grasple Exercise 4.3.6

One more like the previous.

Show/Hide Content

Grasple Exercise 4.3.7

Given \(\vect \) and some basis \(\mathcal\) for a subspace in \(\R^3\) , to find \([\vect]_<\mathcal>\) .

Show/Hide Content

Grasple Exercise 4.3.8

Expressing a vector \(\vect \in \R^3\) in a basis \(\mathcal = \_1,\vect_2,\vect_3\>\) (and vice versa).

Show/Hide Content

Grasple Exercise 4.3.9

Expressing a vector \(\vect \in \R^3\) in a basis \(\mathcal = \_1,\vect_2,\vect_3\>\) .

Show/Hide Content

Grasple Exercise 4.3.10

To explain the relation between \(\vect\) and \([\vect]_<\mathcal>\) .

Show/Hide Content

Grasple Exercise 4.3.11

To find out how coordinates change when a basis is reordered.

Show/Hide Content

Grasple Exercise 4.3.12

Show/Hide Content

The remaining exercises are about matrix representations of linear transformations.

Grasple Exercise 4.3.13

Show/Hide Content

Grasple Exercise 4.3.14

To transform from \([T]_<\mathcal>\) to \([T]_<\mathcal_2\leftarrow\mathcal_1>\) for \(T\) from \(\R^2\) to \(\R^3\) .

Show/Hide Content

Grasple Exercise 4.3.15

To compute \([T]_<\mathcal\leftarrow\mathcal>\) for \(T\) from \(\R^2\) to \(\R^2\) with respect to several bases.

Show/Hide Content

Grasple Exercise 4.3.16

To compute \([T]_<\mathcal\leftarrow\mathcal>\) , for \(T\) from \(\R^2\) to \(\R^2\) . with respect to several bases.

Show/Hide Content

Grasple Exercise 4.3.17

Like the previous, for a linear transformation \(T\) from \(\R^3\) to \(\R^2\) .

Show/Hide Content

4.2. Basis and Dimension

By a team of lecturers and developers from the Delft Institute of Applied Mathematics from the TU Delft University of Technology.

Have you spotted a mistake or an error on this page? Click here to tell us!

This work is licensed under a Creative Commons Attribution 4.0 International License.