Composition of Linear Maps

In both the physical world and abstract mathematics, many complex processes are simply a sequence of individual actions. The concept of performing one action after another—like rotating an object, then scaling it—is formalized in linear algebra as the ​​composition of linear maps​​. This powerful idea addresses the challenge of analyzing long chains of transformations by providing a method to collapse them into a single, equivalent operation. This article explores the principle of composition, revealing it as a fundamental thread connecting diverse mathematical and scientific disciplines.

The following chapters will guide you through this essential concept. First, in "Principles and Mechanisms," we will delve into the core of composition, establishing how it is represented by matrix multiplication and exploring the critical consequences of this relationship, including the non-commutative nature of transformations and their effects on geometric properties like volume and dimensionality. Subsequently, in "Applications and Interdisciplinary Connections," we will witness how this principle extends far beyond simple geometry, serving as a foundational tool in fields from calculus and quantum mechanics to the abstract study of symmetry and category theory, demonstrating how the simple idea of "what comes next" unifies vast areas of modern science.

Imagine you are following a recipe. First, you chop the vegetables. Second, you place them in a hot pan. The final result depends crucially on this sequence. Chopping and then heating is quite different from heating and then attempting to chop! In mathematics, and indeed in the physical world, many processes can be thought of as a sequence of well-defined actions. When these actions are of a special, well-behaved kind called ​​linear maps​​ or ​​linear transformations​​, we can analyze their sequences with remarkable power and elegance. This process of performing one transformation after another is called ​​composition​​.

At its heart, composition is as simple as our cooking analogy. You start with an object (a vector), apply one transformation to it, and then apply a second transformation to the result. Let's call the first transformation $T$ and the second $S$. The composition, written as $S \circ T$, means "first do $T$, then do $S$."

Consider a vector in three-dimensional space, say $\vec{v} = (1, 2, 3)$. Let's subject it to two transformations. First, a transformation $T$ that swaps the second and third components and then negates the new second component. Applying $T$ to our vector gives:

Now, let's take this new vector and apply a second transformation, $S$, which adds the first component to the third.

So, the result of the composite action $(S \circ T)$ on the vector $(1, 2, 3)$ is the vector $(1, -3, 3)$. This step-by-step process is intuitive, but it can be cumbersome. What if we wanted to perform this sequence of actions on a million different vectors? Must we always perform two separate calculations? The beauty of linear algebra is that it allows us to package this entire sequence into a single, equivalent transformation.

Every linear transformation in a finite-dimensional space can be represented by a matrix. This matrix is not just a table of numbers; it is the DNA of the transformation. It encodes exactly how the transformation stretches, rotates, reflects, or projects space. The true magic appears when we consider composition.

The composition of two linear maps corresponds to the ​​multiplication of their matrices​​.

This is one of the most fundamental and useful ideas in linear algebra. If a map $T$ is represented by a matrix $M_T$ and a map $S$ is represented by a matrix $M_S$, then the composite map $S \circ T$ is represented by the product of the matrices, $M_{S \circ T} = M_S M_T$.

Notice the order. The transformation you apply first ($T$) corresponds to the matrix on the right ($M_T$). This might seem backward, but it makes perfect sense when you see it in action. A transformation acts on a vector $\vec{v}$ by matrix multiplication: $M_T \vec{v}$. If we then apply $S$ to this result, we get $M_S (M_T \vec{v})$. Because matrix multiplication is associative, this is the same as $(M_S M_T) \vec{v}$. The single matrix $M_S M_T$ now represents the entire two-step process. This powerful principle allows us to collapse a whole chain of transformations, no matter how long or complex, into a single matrix that tells the whole story.

Let's witness this principle create a beautiful geometric dance. Imagine we want to perform three distinct actions on every point in a 2D plane:

1. First, a counter-clockwise rotation $T_R$ by $90^\circ$ ($\frac{\pi}{2}$ radians).
2. Second, a non-uniform scaling $T_S$ that doubles the x-coordinate and triples the y-coordinate.
3. Third, a reflection $T_F$ across the y-axis.

Each of these is a linear transformation with a corresponding matrix:

• ​​Rotation ($M_R$):​​ $M_R = \begin{pmatrix} \cos(\frac{\pi}{2}) & -\sin(\frac{\pi}{2}) \\ \sin(\frac{\pi}{2}) & \cos(\frac{\pi}{2}) \end{pmatrix} = \begin{pmatrix} 0 & -1 \\ 1 & 0 \end{pmatrix}$
• ​​Scaling ($M_S$):​​ $M_S = \begin{pmatrix} 2 & 0 \\ 0 & 3 \end{pmatrix}$
• ​​Reflection ($M_F$):​​ $M_F = \begin{pmatrix} -1 & 0 \\ 0 & 1 \end{pmatrix}$

The composite transformation is $T = T_F \circ T_S \circ T_R$. Its matrix, $M$, is the product of the individual matrices in the reverse order of application:

Let's multiply them out. First, $M_S M_R$:

Now, we pre-multiply by $M_F$:

And there it is. The entire three-step dance—rotate, scale, reflect—is encapsulated in this one simple matrix. Any vector $\begin{pmatrix} x \\ y \end{pmatrix}$ subjected to this sequence will end up at $\begin{pmatrix} 0 & 2 \\ 3 & 0 \end{pmatrix} \begin{pmatrix} x \\ y \end{pmatrix} = \begin{pmatrix} 2y \\ 3x \end{pmatrix}$. What was a cumbersome procedure is now a single, elegant operation.

Our recipe analogy suggested that order matters, and matrix multiplication confirms it. In general, for two matrices $A$ and $B$, $AB \neq BA$. This property is called ​​non-commutativity​​. Far from being an inconvenience, it is a fundamental feature of the universe.

Let's explore this with a simple example. Consider a reflection $T_R$ across the line $y=x$ and an orthogonal projection $T_P$ onto the x-axis. Their matrices are:

• ​​Reflection ($[R]$):​​ $[R] = \begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}$
• ​​Projection ($[P]$):​​ $[P] = \begin{pmatrix} 1 & 0 \\ 0 & 0 \end{pmatrix}$

What happens if we reflect first, then project? This is $T_P \circ T_R$, with matrix $[P][R]$:

What if we project first, then reflect? This is $T_R \circ T_P$, with matrix $[R][P]$:

The results are clearly different. Performing the actions in a different order leads to a completely different outcome. This non-commutativity is not just a mathematical curiosity; it is at the heart of fields like quantum mechanics, where observing a particle's position and then its momentum gives a different result from observing its momentum and then its position. Sometimes, while two operations may not commute ($M_1 \neq M_2$), they are still related in a deeper, more symmetric way, such as being conjugates of each other ($M_1 = F M_2 F^{-1}$).

Beyond the mechanics of multiplication, composition reveals profound structural relationships and allows us to track fundamental properties, or "invariants," of geometric objects.

First, let's consider what happens when a composition results in nothing—that is, when it maps every vector to the zero vector. For a composition $S \circ T$ to be the zero map, there must be a special relationship between the two transformations. The set of all possible outputs of a map is called its ​​range​​, and the set of inputs it sends to zero is called its ​​kernel​​. The composition $S \circ T$ is the zero map if and only if the range of $T$ is contained entirely within the kernel of $S$.

This is a beautiful and intuitive condition. It means that everything the first transformation $T$ can possibly produce is precisely the kind of thing that the second transformation $S$ is designed to annihilate. It’s like a two-stage filter where the first stage isolates a specific substance, and the second stage neutralizes it completely.

Second, let's see how essential properties are affected by composition. The ​​determinant​​ of a matrix, for instance, tells us how the transformation scales volume and whether it flips the orientation of space. A key property is that the determinant of a product is the product of the determinants: $\det(AB) = \det(A)\det(B)$.

Consider a rotation $T$ in 3D space, followed by a reflection $R$ across a plane.

• A rotation preserves volume and orientation, so $\det(T) = +1$.
• A reflection preserves volume but flips orientation (like looking in a mirror), so $\det(R) = -1$.

The determinant of the composite map $T \circ R$ is simply $\det(T)\det(R) = (+1)(-1) = -1$. Without calculating a single entry of the final matrix, we know that the combined transformation acts like a reflection; it inverts the "handedness" of space.

Similarly, the ​​rank​​ of a transformation tells us the dimension of its output space—a line has rank 1, a plane has rank 2. The rank of a composite map can never be greater than the rank of any of its constituent maps. This makes sense: you cannot create dimensions out of thin air. If the first map squashes all of 3D space onto a plane (rank 2), the second map can't magically restore the third dimension. The composition principle gives us a precise rule for how dimensionality is lost: $\text{rank}(T \circ S) \le \min(\text{rank}(T), \text{rank}(S))$.

From simple sequences of actions to the deep structures of mathematics, the composition of linear maps is a unifying thread. It shows us how complex operations are built from simpler parts, how geometry is encoded in algebra, and how fundamental properties of space are preserved or altered through transformation.

What happens when you do one thing, and then another? You put on your left sock, then your left shoe. You rotate a photograph, then you enlarge it. This simple, everyday idea of sequential action is called composition. In the world of linear algebra, where transformations stretch, rotate, and reshape space, composition takes on a life of its own. As we saw in the previous chapter, the composition of two linear maps corresponds to the multiplication of their matrices. This might seem like a mere computational convenience, but it is so much more. It is a key that unlocks a breathtaking landscape of applications, revealing deep and often surprising connections between geometry, calculus, quantum physics, and the most abstract realms of modern mathematics. Join us on a journey to explore this landscape, guided by the simple question: "What happens next?"

Let's begin where our intuition is strongest: in the geometric world of shapes and movements. Imagine taking a deck of cards and pushing the top of the deck sideways, so the side is no longer a rectangle but a parallelogram. This is a "shear." A linear map can describe this action precisely. Now, what happens if you apply the same shear transformation a second time? Your intuition might suggest the shear effect simply doubles, and in this case, your intuition is spot on. If a single horizontal shear shifts a point $(x, y)$ to $(x+ky, y)$, applying it again results in a total shift to $(x+2ky, y)$. The composition of the shear with itself is a new shear, but with twice the strength. This simple example, $S_k \circ S_k$, shows how repeated actions, represented by matrix powers like $M^2$, can accumulate their effects.

Now, let's consider reflections. A single reflection, like looking in a mirror, flips the orientation of space. Any object with a distinct "handedness," like a glove, becomes its mirror image. This reversal of orientation is captured by the determinant of the transformation's matrix being $-1$. So, what happens if you compose a sequence of reflections? The determinant of a composition is the product of the individual determinants. If you compose two reflections, the total determinant is $(-1) \times (-1) = 1$. The orientation is restored! This is why standing between two parallel mirrors creates reflections of reflections that look just like you, not your mirror image. This simple principle is powerful enough to analyze the complex symmetries of elementary particles, where abstract "reflections" in higher-dimensional spaces are composed to form structures like Weyl groups.

This idea of multiplying scaling factors extends beautifully into multivariable calculus. When you perform a change of coordinates—say, from a familiar Cartesian grid $(x, y)$ to a polar grid $(r, \theta)$—a tiny rectangle in one system maps to a slightly distorted, slightly larger or smaller shape in the other. The factor by which the area scales is given by the Jacobian determinant of the transformation. If you then perform another transformation, it too will scale the area by its own Jacobian factor. The total scaling factor for the composite transformation is, just as we've come to expect, the product of the individual Jacobian determinants. Composition tells us how to chain together changes in space and keep track of how properties like area and volume evolve along the way.

The power of linear maps goes far beyond geometry. They can represent abstract operations. Consider the derivative from calculus. It takes a function and gives you a new function, its rate of change. You can easily verify that the derivative is a linear operator: the derivative of a sum is the sum of the derivatives. So, what is the second derivative, $\frac{d^2}{dx^2}$? It's simply the act of taking the derivative of the derivative! In our language, the second derivative operator, let's call it $D^2$, is the composition of the first derivative operator $D$ with itself: $D^2 = D \circ D$. This reframes a familiar tool from calculus as a repeated application of a single linear transformation. This perspective allows us to analyze differential equations using the tools of linear algebra, turning complex analytical problems into more manageable algebraic ones.

Once we realize we can compose transformations, a whole new world opens up. We can treat the transformations themselves as algebraic objects. We can "multiply" them (compose), add them, and scale them. This means we can form polynomials of transformations. For instance, a transformation $T$ might satisfy an equation like: $T^2 - 3T + 2I = \mathbf{0}$ where $T^2$ is $T \circ T$ and $I$ is the identity transformation. This isn't just a formal curiosity. We can manipulate this equation just like we would with numbers in high school algebra. By rearranging the terms to $2I = 3T - T^2$ and factoring out a $T$, we get $2I = (3I - T)T$. This equation gives us an explicit recipe for the inverse of $T$! It tells us that $T^{-1} = \frac{1}{2}(3I - T)$. We have found a way to undo a transformation using a combination of the transformation itself and the identity map. This is a glimpse of the deep connection between a transformation and its characteristic polynomial, a cornerstone of linear algebra known as the Cayley-Hamilton theorem.

This algebraic playground also allows us to prove profound properties about certain types of transformations. Consider a "projection," which acts like casting a shadow. If you project an object onto a plane, and then project that shadow onto the same plane, the shadow doesn't change. This defining property is captured by the equation $P^2 = P$. Now we can ask a question: can a projection be invertible? Can you reliably reconstruct a 3D object from its 2D shadow? Intuitively, the answer is no; information is lost. Our algebra proves it. If $P$ were invertible, we could take the equation $P^2 = P$ and multiply on the left by $P^{-1}$. This gives $(P^{-1}P)P = P^{-1}P$, which simplifies to $IP = I$, and finally $P = I$. This means the only invertible projection is the identity transformation—the one that doesn't do anything in the first place! The algebraic structure born from composition gives us a powerful and elegant way to reason about the nature of transformations.

The concept of composition is a golden thread that weaves through countless scientific disciplines, binding them together.

• ​​Quantum Information Science:​​ In a quantum computer, the state of two separate quantum bits (qubits) is described in a combined space known as the tensor product. If you apply an operation $T$ to the first qubit and an operation $S$ to the second, the composite system evolves according to the map $T \otimes S$. A crucial question is whether this evolution is reversible; can you always recover the initial quantum state? The theory of composition gives a beautifully clear answer: the combined operation $T \otimes S$ is injective (and thus reversible) if and only if both of the individual operations, $T$ and $S$, are injective. The properties of the whole are determined directly by the properties of its parts, a principle that is fundamental to building reliable quantum computers.

• ​​Symmetry and Representation Theory:​​ Composition is the very heart of group theory, the mathematical language of symmetry. If you perform one symmetry operation on an object (say, a rotation) and then another, the result is yet another symmetry operation. The set of all symmetries is "closed" under composition. In representation theory, we study symmetries by representing them as linear maps. For this representation to be faithful, the algebraic structure must be preserved. A composition of two symmetry operations must correspond to the composition of their representative matrices. This is the essence of a G-homomorphism, a map that respects the group's structure. And, as you might guess, the composition of two such structure-preserving maps is itself a structure-preserving map, ensuring the integrity of the representation.

• ​​Information Flow and Bottlenecks:​​ Imagine a chain of transformations, $L$, then $M$, then $N$, as an assembly line. $L$ takes raw materials from a space $U$ and produces a set of components in a space $V$. $M$ assembles these into sub-assemblies in a space $W$. Finally, $N$ performs the final packaging into space $X$. The rank of a linear map can be thought of as a measure of the diversity or "information" it can produce. If one of the intermediate stages, say map $M$, is a significant bottleneck—meaning it has a very low rank—it doesn't matter how sophisticated $L$ and $N$ are. The variety of the final product will be severely limited. The rank of the total composition $N \circ M \circ L$ is constrained by the ranks of the individual maps and the dimensions of the intermediate vector spaces. This is the intuition behind deep results like Sylvester's rank inequality, which provides precise bounds on how much information can pass through a chain of linear processes.

We have seen composition appear in geometry, calculus, and algebra, in the quantum world and the study of symmetry. This concept is so universal, so utterly fundamental, that it forms the bedrock of one of the most powerful and abstract frameworks in modern mathematics: category theory.

In its grandest vision, category theory describes a universe of mathematical objects (like vector spaces) and the "arrows" between them (linear maps). The single most important rule in this universe is that if you have an arrow from object A to B and another arrow from B to C, you can compose them to get a direct arrow from A to C. But it doesn't stop there. One can define functors, which are maps between entire categories, and then natural transformations, which are maps between functors. And yes, you can compose these as well! This "vertical composition" of natural transformations is like stacking one structured mapping on top of another. The idea of "one thing after another" is a fractal pattern, reappearing at ever-higher levels of abstraction.

From the simple act of shearing a deck of cards to the deepest structures of mathematics, the composition of linear maps is a unifying principle. It is far more than matrix multiplication; it is a fundamental pattern of thought that allows us to build complex operations from simple ones, to analyze chains of events, and to find the elegant, underlying unity in a world of endless transformation.