8 Applications of Linear Algebra

Modeling with Vectors and Matrices

Linear algebra provides a common language for situations involving many interacting quantities. A vector can represent a state, a point, a set of unknowns, or quantities such as product amounts and regional populations. A matrix organizes relationships among those quantities or describes how inputs are converted into outputs.

For example, if a factory makes tables and chairs, the production vector is

where $t$ and $c$ are the numbers of tables and chairs. The resource-use matrix is

so

The matrix captures the structure of the production process, while the vector describes the particular plan being evaluated.

Takeaway: Use vectors for quantities or states and matrices for relationships, coefficients, or transformations.

Linear Systems and Gaussian Elimination

Many applied problems ask which unknown quantities produce specified totals. These problems are represented by

where $A$ is the coefficient matrix, $\mathbf{x}$ contains the unknowns, and $\mathbf{b}$ contains observed totals or required outputs.

Gaussian elimination solves the system by forming the augmented matrix $[A\mid\mathbf{b}]$, applying elementary row operations, and interpreting the resulting pivots and free variables. Row operations preserve the solution set because they replace equations with equivalent equations or combinations of existing equations.

Consider a mixture requiring $10$ liters at an acid concentration of $30\%$, using a $20\%$ solution and a $50\%$ solution. If $x$ and $y$ are the amounts used, then

Equivalently,

Solving gives

so approximately $6.67$ liters of the $20\%$ solution and $3.33$ liters of the $50\%$ solution are needed.

A consistent system can have exactly one solution or infinitely many solutions. An inconsistent system has no solution. Free variables often represent design choices, while inconsistency may signal conflicting constraints or measurements.

Takeaway: Translate each condition into an equation, solve systematically, and interpret the number of solutions in context.

Determinants and Invertibility

For a square matrix, the determinant helps assess invertibility and geometric scaling. The key test is

For a $2\times2$ matrix,

When the determinant is nonzero, the inverse satisfies

The inverse represents the operation that reverses the original matrix model. In practice, elimination is usually preferred to explicitly calculating an inverse for a large system.

The determinant also has a geometric meaning. In two dimensions, $\lvert\det(A)\rvert$ is the area-scaling factor. A negative determinant reverses orientation. If $\det(A)=0$, the transformation collapses the space into a lower-dimensional set, so independent directions become dependent.

Takeaway: A nonzero determinant supports unique recovery and invertibility; a zero determinant signals collapse, dependence, or possible nonuniqueness.

Transformations, Kernels, and Images

A linear transformation preserves addition and scalar multiplication:

After bases are chosen, it can be represented as

The columns of $A$ are the images of the basis vectors. If two transformations are represented by $A_1$ and $A_2$, their composition is represented by

The order matters: generally $A_2A_1\neq A_1A_2$.

A shear provides a useful example:

Horizontal lines remain horizontal, but points shift in proportion to their height. Since $\det(S)=1$, the shear preserves area even though it changes shape.

The kernel is the set of inputs mapped to the zero vector, while the image is the set of possible outputs. A nontrivial kernel indicates that some input information is lost.

Takeaway: Matrices describe transformations, composition corresponds to multiplication, and the multiplication order reflects the order in which transformations occur.

Vector Spaces and Coordinate Choices

A vector space is a collection in which vectors can be added and multiplied by scalars while satisfying the vector-space rules. Examples include coordinate vectors, polynomials, matrices, functions, and solution sets of homogeneous systems.

A set is linearly independent when

has only the trivial solution

A basis is a linearly independent spanning set. It provides a coordinate system for the vector space, and the number of basis vectors is the dimension.

Changing the basis changes the matrix used to describe a transformation, but not the underlying transformation itself. If $P$ converts coordinates from a new basis to the standard basis, then

This similarity relationship explains why a difficult matrix may become simpler in a well-chosen coordinate system.

Takeaway: Independence prevents redundancy, spanning ensures coverage of the space, and a basis combines both properties into a useful coordinate system.

Eigenvalues and Long-Term Behavior

An eigenvector is a nonzero vector whose direction is preserved by a matrix. Its associated eigenvalue gives the scaling factor:

Eigenvalues are found from the characteristic equation

and the corresponding eigenvectors are found by solving

Repeated application is simple along an eigenvector:

If a matrix has a basis of eigenvectors, diagonalization gives

This is useful for transition models. If a state vector evolves according to

then a stable distribution, when it exists, satisfies

Thus the stable distribution is an eigenvector with eigenvalue $1$. Other eigenvalues indicate how quickly deviations decay, persist, or grow: magnitudes below $1$ describe decay, while magnitudes above $1$ indicate growth in the corresponding modes.

Takeaway: Eigenvectors reveal invariant directions, eigenvalues describe scaling or stability, and diagonalization simplifies repeated updates.

A Strategy for Applied Problems

A reliable modeling workflow connects the algebra to the application:

1. Define the quantities. State what every variable or vector component represents and include units.

2. Choose the structure. Use vectors for states or collections of values and matrices for coefficients, transitions, or transformations.

3. Write the governing equation. Common forms are

   $$A\mathbf{x}=\mathbf{b}, \qquad \mathbf{x}_{k+1}=A\mathbf{x}_k, \qquad T(\mathbf{x})=A\mathbf{x}.$$

4. Check dimensions. Matrix multiplication is valid only when the inner dimensions agree.

5. Select a method. Use elimination for systems, determinants for invertibility tests, and eigenvalue methods for repeated applications or stable modes.

6. Interpret the result. Check signs, units, feasibility, and whether the result makes sense in context.

7. Assess sensitivity. Systems close to singularity can respond strongly to small data changes, and eigenvalues near or above magnitude $1$ can make repeated updates especially sensitive.

For a network with unknown flows $x_1$, $x_2$, and $x_3$, conservation conditions might produce

The matrix records how flows enter or leave locations, while $\mathbf{b}$ records external supplies or demands. If the flows change over time, a separate update model such as

can describe the dynamics. The determinant of the current-state matrix addresses uniqueness, while the eigenvalues of the update matrix reveal whether disturbances decay, persist, or grow.

Takeaway: Good applied linear algebra requires accurate modeling, dimension checks, an appropriate method, and interpretation of the result.