Linear Algebra: Linear Transformations

Linear transformations are the fundamental language of change in vector spaces. While you may be comfortable with matrices as arrays of numbers, understanding them as functions that move and reshape space is what unlocks their true power in engineering, from computer graphics and robotics to data science and control systems. This concept bridges abstract algebra and concrete application, providing a precise toolkit for modeling any operation that preserves the structure of linearity.

1. What is a Linear Transformation?

A linear transformation is a special type of function, $T$, that maps vectors from one vector space (the domain) to another (the codomain), while preserving the operations of vector addition and scalar multiplication. Formally, for any vectors $\mathbf{u}$ and $\mathbf{v}$ in the domain and any scalar $c$, a transformation $T$ is linear if it satisfies two properties:

1. Additivity: $T(\mathbf{u}+\mathbf{v})=T(\mathbf{u})+T(\mathbf{v})$
2. Homogeneity (Scalar Multiplication): $T(c\mathbf{u})=cT(\mathbf{u})$

These two rules ensure the transformation is "well-behaved" and doesn't warp the grid lines of the vector space in a curved or discontinuous way. To verify a given function is a linear transformation, you must test these two properties. For example, consider $T:{\mathbb{R}}^2\to {\mathbb{R}}^2$ defined by $T(x,y)=(2x-y,x+3y)$. For vectors $\mathbf{u}=(u_1,u_2)$ and $\mathbf{v}=(v_1,v_2)$:

• Check Additivity: $T(\mathbf{u}+\mathbf{v})=T(u_1+v_1,u_2+v_2)=(2(u_1+v_1)-(u_2+v_2),(u_1+v_1)+3(u_2+v_2))$. This equals $(2u_1-u_2,u_1+3u_2)+(2v_1-v_2,v_1+3v_2)=T(\mathbf{u})+T(\mathbf{v})$.
• Check Homogeneity: $T(c\mathbf{u})=T(c{u}_1,c{u}_2)=(2c{u}_1-c{u}_2,c{u}_1+3c{u}_2)=c(2u_1-u_2,u_1+3u_2)=cT(\mathbf{u})$.

Since both hold, $T$ is linear. Functions that break these rules, like $T(x,y)=(x^2,y)$ or $T(x,y)=(x+1,y)$, are not linear transformations.

2. The Kernel and Image: Fundamental Subspaces

Every linear transformation defines two crucial subspaces that reveal its behavior. The kernel (or null space) of $T$, denoted $\ker (T)$, is the set of all vectors in the domain that map to the zero vector in the codomain: $\ker (T)=\{\mathbf{v}|T(\mathbf{v})=0\}$. It measures what the transformation "squashes" into nothingness. If the kernel contains only the zero vector, the transformation is injective (one-to-one); no two distinct input vectors produce the same output.

The image (or range) of $T$, denoted $\mathrm{im}(T)$, is the set of all possible outputs: $\mathrm{im}(T)=\{T(\mathbf{v})|\mathbf{v}\text{ is in the domain}\}$. It is a subspace of the codomain and represents everything the transformation can "reach." The dimensions of these spaces are linked by the Rank-Nullity Theorem: $\dim (\text{domain})=\dim (\ker (T))+\dim (\mathrm{im}(T))$. In engineering, the kernel might represent unobservable states in a sensor system, while the image represents all possible sensor readings.

3. The Standard Matrix and Composition

The power of linear algebra is that every linear transformation $T:{\mathbb{R}}^n\to {\mathbb{R}}^m$ can be represented by matrix multiplication. You can find the standard matrix $A$ of $T$ by applying $T$ to the standard basis vectors ${\mathbf{e}}_1,{\mathbf{e}}_2,...,{\mathbf{e}}_n$ of ${\mathbb{R}}^n$. The outputs become the columns of $A$: $A=[T({\mathbf{e}}_1)|T({\mathbf{e}}_2)|...|T({\mathbf{e}}_n)]$. Then, for any vector $\mathbf{x}$, $T(\mathbf{x})=A\mathbf{x}$.

This leads to a profound connection: composition of transformations corresponds to matrix multiplication. If $T:{\mathbb{R}}^n\to {\mathbb{R}}^m$ has matrix $A$ and $S:{\mathbb{R}}^m\to {\mathbb{R}}^p$ has matrix $B$, then the composite transformation $(S\circ T)(\mathbf{x})=S(T(\mathbf{x}))$ is represented by the product $BA$. The order matters profoundly. In engineering, this is how you chain operations: rotating an object in 3D space (transformation $T$), then projecting it onto a 2D screen (transformation $S$) is achieved by multiplying the projection matrix by the rotation matrix.

4. Key Geometric Transformations in the Plane

Many fundamental operations in engineering graphics and physics are linear transformations with intuitive geometric interpretations. Their standard matrices are worth memorizing.

• Rotation by an angle $\theta$ counterclockwise:

$$R_{\theta }=\left[\begin{array}{cc}\cos \theta & -\sin \theta \\ \sin \theta & \cos \theta \end{array}\right]$$

• Reflection across a line through the origin with unit direction vector $\mathbf{u}=(\cos \alpha ,\sin \alpha )$:

The matrix is derived from the projection formula and is $2\mathbf{u}{\mathbf{u}}^T-I$.

• Orthogonal Projection onto a line through the origin with unit direction vector $\mathbf{u}$:

The matrix is $\mathbf{u}{\mathbf{u}}^T$.

These transformations are the building blocks for more complex manipulations. For instance, in computer-aided design (CAD), scaling and shearing transformations (also linear) are combined with these to model parts.

5. Examples in Engineering Contexts

Linear transformations are not just abstract math; they are the operational engines in countless engineering domains.

• Robotics & Kinematics: The relationship between joint angles (in configuration space) and the position/orientation of a robot's end-effector in 3D space (task space) is modeled by a linear transformation when considering small motions or velocities. The Jacobian matrix of this transformation is central to motion planning and control.
• Signal Processing: The Discrete Fourier Transform (DFT) is a linear transformation that maps a time-domain signal vector to its frequency-domain representation. The transformation matrix is composed of complex exponentials, and its properties (like the Fast Fourier Transform algorithm) rely entirely on linearity.
• Structural Analysis: In finite element analysis, the stiffness matrix $K$ acts as a linear transformation. It maps the vector of nodal displacements $\mathbf{u}$ to the vector of nodal forces $\mathbf{f}$ via the equation $\mathbf{f}=K\mathbf{u}$. Solving this system for displacements under load is the core of the analysis.
• Control Systems: The state-space representation $\dot{\mathbf{x}}=A\mathbf{x}+B\mathbf{u}$ describes how the system state $\mathbf{x}$ evolves. The matrix $A$ is a linear transformation describing the internal dynamics of the system—how the current state linearly determines the rate of change of the state.

Common Pitfalls

1. Assuming All Functions are Linear: A common error is to see a function with $x$ and $y$ and assume it's linear. Always check the two defining properties. The function $T(x,y)=(xy,0)$ fails additivity: $T((1,1)+(1,1))=T(2,2)=(4,0)$, but $T(1,1)+T(1,1)=(1,0)+(1,0)=(2,0)$. They are not equal.
2. Confusing the Order of Composition: When performing sequential transformations $T$ then $S$, the matrix for the overall operation is $[S][T]$, not $[T][S]$. Remember, the matrix closest to the input vector acts first. Think of it as applying operators from right to left: $S(T(\mathbf{v}))$ means do $T$, then do $S$.
3. Misidentifying the Kernel: The kernel is a set of input vectors. Don't confuse it with the set of output vectors that are zero (that's just $\{0\}$ in the codomain). To find the kernel, solve $A\mathbf{x}=0$.
4. Overlooking the Domain and Codomain: The transformation $T(x)=(x,x,x)$ maps ${\mathbb{R}}^1$ to ${\mathbb{R}}^3$. Its image is a line in 3D space (all vectors of the form $(a,a,a)$), and its kernel is just $\{0\}$ in ${\mathbb{R}}^1$. Clearly stating spaces prevents dimension-related errors.

Summary

• A linear transformation is a function between vector spaces that preserves vector addition and scalar multiplication, defined by $T(\mathbf{u}+\mathbf{v})=T(\mathbf{u})+T(\mathbf{v})$ and $T(c\mathbf{u})=cT(\mathbf{u})$.
• The kernel consists of all inputs that map to zero, revealing what is "lost" in the transformation. The image is the set of all possible outputs, showing what can be "reached."
• Every linear transformation between ${\mathbb{R}}^n$ and ${\mathbb{R}}^m$ can be represented by a standard matrix $A$, where $T(\mathbf{x})=A\mathbf{x}$, found by applying $T$ to the standard basis vectors.
• Performing one transformation after another corresponds to matrix multiplication, with the order of operations critical: $(S\circ T)(\mathbf{x})$ is represented by $[S][T]\mathbf{x}$.
• Fundamental geometric transformations like rotation, reflection, and projection have standard matrix representations and serve as core building blocks for modeling physical and digital systems in engineering.