Matrices and Systems of Linear Equations

Introduction and Context

Business analytics frequently represents many related quantities at once: products and resources, assets and returns, customers and characteristics, or sectors and supplier relationships. Matrices provide a compact structure for storing and transforming these quantities. Systems of linear equations then allow analysts to determine unknown values subject to several simultaneous relationships.

For example, a manager may need to determine how many units of two products to produce when each uses different amounts of labor and materials. A financial analyst may represent portfolio weights and expected returns as vectors and matrices. An economist may estimate the production required throughout an economy to satisfy final demand. These applications share the same mathematical foundation.

Learning Objectives

By the end of this chapter, you should be able to:

• identify matrix dimensions, elements, rows, columns, and special matrices;
• perform matrix addition, scalar multiplication, multiplication, and transposition;
• formulate business problems as systems of linear equations;
• solve systems using Gaussian elimination, Gauss–Jordan elimination, and matrix inverses;
• interpret uniqueness, inconsistency, and infinitely many solutions;
• use computational tools responsibly to verify and analyze solutions;
• apply matrices to resource allocation, portfolio representation, transition models, and Leontief input–output analysis.

1. Matrix Fundamentals

A matrix is a rectangular array of numbers. A matrix with m rows and n columns has order m × n and is written A = [aᵢⱼ], where aᵢⱼ is the element in row i and column j.

For example,

has 2 rows and 3 columns, so its order is 2 × 3. The element a₂₃ is 4.

A row matrix has one row; a column matrix has one column. A square matrix has the same number of rows and columns. A zero matrix contains only zeros. The identity matrix Iₙ is an n × n square matrix with ones on the main diagonal and zeros elsewhere:

The identity matrix behaves like the number 1 under multiplication, provided dimensions are compatible.

A diagonal matrix has nonzero entries only on its main diagonal. A triangular matrix has all entries above or below the main diagonal equal to zero. A symmetric matrix satisfies Aᵀ = A, where Aᵀ denotes the transpose of A.

Equality and dimensions

Two matrices are equal only when they have the same dimensions and corresponding elements are equal. Dimensions are essential because matrix operations are defined by shape as well as by numerical values.

2. Matrix Operations

Addition and subtraction

Matrices of the same order can be added or subtracted element by element:

The matrices must have identical dimensions. If A represents monthly sales for one product group and B represents monthly sales for another group using the same row and column definitions, A + B may represent combined sales.

Scalar multiplication

For a scalar c and matrix A,

Every element is multiplied by c. A scalar may represent a price change, percentage adjustment, or scaling factor.

Matrix multiplication

If A is m × n and B is n × p, then AB is defined and has order m × p. Its elements are calculated by multiplying corresponding entries from a row of A and a column of B and summing:

The inner dimensions must match: the number of columns in A must equal the number of rows in B. Matrix multiplication is generally not commutative; AB may exist while BA does not, or both may exist but have different values.

Worked example: multiplication

Let

Then

In contrast,

Thus AB ≠ BA.

Transpose

The transpose changes rows into columns:

If A is m × n, Aᵀ is n × m. Transposition is useful when converting a row of observations into a column vector or expressing quadratic forms such as portfolio variance.

Important rules include (AB)ᵀ = BᵀAᵀ and (A + B)ᵀ = Aᵀ + Bᵀ.

3. Determinants and Inverses

The determinant is defined for square matrices and provides information about scaling, singularity, and solvability. For a 2 × 2 matrix,

A square matrix is singular when det(A) = 0. It is nonsingular or invertible when det(A) ≠ 0.

For

its inverse, when ad − bc ≠ 0, is

The inverse satisfies AA⁻¹ = A⁻¹A = I. An inverse exists only for a square, nonsingular matrix. In practical numerical work, a very small determinant or a poorly conditioned matrix may make direct inversion unstable even if the determinant is not exactly zero.

4. Systems of Linear Equations

A system of m linear equations in n unknowns can be written as

where A is the m × n coefficient matrix, x is the n × 1 vector of unknowns, and b is the m × 1 vector of constants or requirements. A linear equation contains variables only to the first power; products such as xy and terms such as x² are not linear.

For example,

has coefficient matrix, variable vector, and constant vector

The augmented matrix combines A and b:

Types of solutions

A system may have:

1. one unique solution;
2. no solution, when equations are inconsistent;
3. infinitely many solutions, when at least one variable is free.

For a square system, det(A) ≠ 0 guarantees a unique solution. If det(A) = 0, the system may have no solution or infinitely many solutions; the determinant alone does not distinguish these cases.

5. Gaussian and Gauss–Jordan Elimination

Elementary row operations preserve the solution set of a system:

• interchange two rows;
• multiply a row by a nonzero constant;
• add a multiple of one row to another row.

Gaussian elimination transforms the augmented matrix into row-echelon form and then uses back-substitution. Gauss–Jordan elimination continues until reduced row-echelon form is obtained, making the solution directly visible.

Worked example: elimination

Solve

The augmented matrix is

Interchange the rows and eliminate x from the second row:

The second row gives −5y = −15, so y = 3. Substitution into x + 3y = 13 gives x = 4. Therefore,

Verification: 2(4) + 3 = 11 and 4 + 3(3) = 13.

Detecting special cases

A row of the form

represents 0 = c and therefore indicates no solution. A completely zero row does not create a contradiction. If one or more variables do not have pivot positions and no contradiction occurs, those variables are free and infinitely many solutions result.

6. Solving by the Inverse Matrix

For Ax = b, if A is square and invertible, premultiplication by A⁻¹ gives

This method is conceptually important and useful when the same coefficient matrix is used with many different right-hand sides. However, computational software usually solves Ax = b directly rather than explicitly calculating A⁻¹, because factorization methods are generally faster and more numerically stable.

Worked example: inverse method

For

we have det(A) = 2(3) − 1(1) = 5, so

Therefore,

The result agrees with elimination.

7. Rank, Consistency, and Model Interpretation

The rank of a matrix is the number of linearly independent rows or columns. In a system Ax = b, consistency can be described using ranks:

• a solution exists if rank(A) = rank([A | b]);
• a unique solution exists if this common rank equals the number of unknowns;
• infinitely many solutions exist if the common rank is less than the number of unknowns.

This criterion is valuable in analytics because it separates insufficient information from contradictory information. For instance, duplicate business measures may add no independent information, while conflicting reported constraints may make a planning model infeasible.

8. Business Applications

8.1 Resource allocation

Suppose a firm produces products X and Y. Product X requires 2 labor hours and 1 machine hour; product Y requires 1 labor hour and 3 machine hours. If available resources are 100 labor hours and 90 machine hours, the exact-use model is

where x and y are production quantities. Solving gives y = 16 and x = 42. The solution uses all resources. In a realistic planning problem, constraints are usually inequalities, such as 2x + y ≤ 100, with x ≥ 0 and y ≥ 0. Matrix methods describe the constraints, while linear programming is required when an objective such as profit maximization is included.

8.2 Portfolio representation

Let w be a vector of portfolio weights, r the vector of expected asset returns, and 1 a vector of ones. The full-investment condition is

The expected portfolio return is

If Σ is the covariance matrix of asset returns, portfolio variance is

The assumptions are that the weights are defined consistently, returns have been estimated on the same time basis, and Σ is compatible with those units. Short selling, transaction costs, and constraints such as wᵢ ≥ 0 must be modeled explicitly rather than assumed away.

8.3 Transition models

A transition matrix describes movement among states over time. If pₜ is a column vector of state proportions and P is a column-stochastic transition matrix, then

Each column of P must sum to 1 under this convention. If a row-vector convention is used instead, the update is pₜ₊₁ = pₜP and rows must sum to 1. Mixing conventions is a common source of error.

For example, if 80% of customers who are currently active remain active next period and 20% become inactive, while 30% of inactive customers return, then with states active and inactive and a column-vector convention,

The matrix can forecast customer-state distributions. A long-run distribution p satisfies Pp = p together with 1ᵀp = 1. Such a distribution exists under standard conditions, but convergence depends on the transition structure, including issues such as periodicity and reducibility.

8.4 Leontief input–output analysis

In an input–output model, aᵢⱼ is the amount of input from sector i required to produce one unit of output in sector j. The matrix A = [aᵢⱼ] is the technical-coefficient matrix. If x is the total output vector and d is final demand, then intermediate demand is Ax and the balance equation is

Rearranging gives

If I − A is invertible, the solution is

The matrix (I − A)⁻¹ is the Leontief inverse. It captures direct and indirect production requirements.

The model assumes fixed input proportions, a linear production technology, sufficient capacity, and a defined period and price basis. It does not automatically account for substitution, price changes, technological innovation, imports, or capacity constraints.

9. Detailed Applied Case: Regional Production Planning

The following is an illustrative case.

Background

A regional food processor produces meal kits and snack boxes. Each meal kit requires 3 kilograms of ingredients and 2 labor hours. Each snack box requires 1 kilogram of ingredients and 1 labor hour. Weekly availability is 1,200 kilograms of ingredients and 700 labor hours. Management wants a production plan that uses all available resources exactly.

Problem formulation

Let x be meal kits and y be snack boxes. The resource equations are

In matrix form,

Analysis and solution

Subtracting the second equation from the first gives x = 500. Substitution into 2x + y = 700 gives 1,000 + y = 700, so y = −300. A negative production quantity is not feasible.

Interpretation

The algebraic system has a unique solution, but it is not operationally feasible because y cannot be negative. This distinction is important: mathematical solvability does not guarantee managerial feasibility. The exact-use requirement is inappropriate for the stated resource levels. A feasible model should use inequalities:

Management must then specify an objective, such as maximizing profit. If meal-kit profit is p₁ and snack-box profit is p₂, the objective is

subject to the resource constraints. This becomes a linear programming problem.

Outcome and learning points

The case shows that matrices efficiently encode resource relationships, elimination reveals the implied quantities, and nonnegativity and inequality constraints are essential for operational interpretation. Analysts should always verify both the equations and the practical domain of the answer.

10. Computational Methods and Best Practice

Spreadsheet software, statistical languages, and numerical libraries can perform matrix operations and solve systems. A typical workflow is:

1. define rows, columns, units, and ordering clearly;
2. construct A and b from documented data;
3. check dimensions and missing values;
4. solve Ax = b using a reliable solver;
5. calculate the residual r = b − Ax;
6. inspect whether the residual is acceptably close to zero;
7. test sensitivity to changes in inputs;
8. document assumptions, precision, and versioned data.

A small residual does not prove that the model is appropriate; it only indicates that the computed solution satisfies the supplied equations approximately. Analysts should also examine conditioning. An ill-conditioned system may produce large changes in x from small changes in A or b. In such cases, pivoting, QR factorization, singular value decomposition, regularization, or reformulation may be appropriate.

For large sparse systems, storing only nonzero entries can reduce memory and computation. In production analytics, reproducible code, data validation, access controls, and independent checks are preferable to an undocumented manual calculation.

11. Limitations, Governance, and Ethics

Matrix models can create an impression of precision while embedding uncertain or biased assumptions. Important risks include:

• measurement error in coefficients and demands;
• aggregation that hides differences among customers or regions;
• historical bias in transition probabilities or allocation data;
• inappropriate use of forecasts as causal explanations;
• privacy risks when matrices contain individual-level records;
• unjustified decisions based solely on an automated output.

Governance practices should include data lineage, clear ownership, documented assumptions, validation against observed outcomes, access controls, and human review for high-impact decisions. When a matrix model affects employment, credit, pricing, or public resources, stakeholders should be able to understand the relevant inputs, constraints, and limitations.

12. Contemporary Developments

Modern analytics extends classical matrix methods through high-dimensional optimization, recommendation systems, matrix factorization, graph-based models, and machine learning. Singular value decomposition can reduce a large data matrix to lower-dimensional patterns. However, dimensionality reduction may remove information relevant to fairness or rare events. AI systems also rely heavily on matrix operations, but computational scale does not eliminate the need for valid data, interpretable assumptions, and monitoring for drift.

Chapter Summary

Matrices organize multidimensional business information and support addition, scalar multiplication, multiplication, transposition, and representation of linear relationships. Matrix multiplication requires compatible dimensions and is generally noncommutative. Determinants identify whether a square matrix is invertible, while the inverse method solves Ax = b when A is square and nonsingular. Gaussian elimination is a general and practical solution method and reveals whether a system has one, none, or infinitely many solutions.

Matrix models support resource allocation, portfolio calculations, customer transition forecasting, and Leontief input–output analysis. A mathematically valid solution must still be tested for feasibility, unit consistency, sensitivity, and ethical appropriateness. Computational solvers are valuable, but analysts should verify dimensions, residuals, assumptions, and model limitations.

Key Terms

Matrix: Rectangular array of numbers arranged in rows and columns.

Order: Number of rows and columns of a matrix.

Element: Individual entry of a matrix.

Scalar: Single number used to multiply a matrix.

Identity matrix: Square matrix that acts as 1 under matrix multiplication.

Transpose: Matrix formed by exchanging rows and columns.

Determinant: Scalar associated with a square matrix that indicates, among other properties, invertibility.

Singular matrix: Square matrix with determinant zero.

Inverse matrix: Matrix A⁻¹ satisfying AA⁻¹ = I.

Coefficient matrix: Matrix containing coefficients of the variables in a linear system.

Augmented matrix: Matrix formed by appending the constants column to the coefficient matrix.

Gaussian elimination: Row-operation method for solving linear systems.

Pivot: Leading nonzero entry used to eliminate other entries.

Rank: Number of linearly independent rows or columns.

Residual: Difference b − Ax after a computed solution is substituted into a model.

Transition matrix: Matrix describing movement among states.

Portfolio variance: Variability measure represented by wᵀΣw.

Technical-coefficient matrix: Input–output matrix describing inputs required per unit of sectoral output.

Leontief inverse: Matrix (I − A)⁻¹ capturing direct and indirect production requirements.

Ill-conditioning: Numerical sensitivity in which small input changes can cause large output changes.

Review Questions

5 Short Answer Questions

1. What dimensions must two matrices have for their sum to be defined?
2. State the condition under which a square matrix has an inverse.
3. What does a contradictory row in an augmented matrix indicate?
4. Distinguish a coefficient matrix from an augmented matrix.
5. Why is matrix multiplication generally not commutative?

5 Long Answer Questions

1. Explain Gaussian elimination and illustrate the three elementary row operations.
2. Compare solving Ax = b by elimination with solving it using A⁻¹b.
3. Explain how portfolio weights, expected returns, and covariance matrices are represented using vectors and matrices.
4. Derive the Leontief equation x = (I − A)⁻¹d and discuss its assumptions.
5. Discuss why a mathematically unique solution may still be unsuitable for a business decision.

3 Analytical / Application Questions

1. A firm uses 4 labor hours and 2 kilograms of material per unit of Product A, and 3 labor hours and 5 kilograms per unit of Product B. With 180 labor hours and 150 kilograms of material, formulate the matrix system for exact resource use and determine whether the resulting quantities are feasible.
2. A two-state customer transition matrix is P = [[0.75, 0.20], [0.25, 0.80]] under a column-vector convention. Explain how to forecast the next-period distribution and identify the conditions needed for a long-run distribution.
3. An input–output model has technical matrix A and final-demand vector d. Explain how a manager would use x = (I − A)⁻¹d, and identify at least three reasons why the forecast might differ from actual production.

Multiple Choice Questions with Answers

Question 1

If A is 3 × 2 and B is 2 × 4, what is the order of AB?

A. 3 × 4

B. 2 × 2

C. 4 × 3

D. 3 × 2

Correct Answer: A. 3 × 4

Question 2

For A = [[2, 1], [1, 3]], what is det(A)?

A. 4

B. 5

C. 6

D. 7

Correct Answer: B. 5

Question 3

Which condition guarantees that a square matrix A is invertible?

A. det(A) = 0

B. A has more rows than columns

C. det(A) ≠ 0

D. All entries of A are positive

Correct Answer: C. det(A) ≠ 0

Question 4

What does a row [0 0 | 5] in an augmented matrix represent?

A. A unique solution

B. A free variable

C. An identity equation

D. An inconsistency

Correct Answer: D. An inconsistency

Question 5

For the system 2x + y = 11 and x + 3y = 13, what is the solution?

A. x = 3, y = 4

B. x = 4, y = 3

C. x = 5, y = 1

D. x = 2, y = 7

Correct Answer: B. x = 4, y = 3

Question 6

Which expression gives expected portfolio return when w is the weight vector and r is the return vector?

A. w + r

B. wᵀr

C. wrᵀ only when w is square

D. w⁻¹r

Correct Answer: B. wᵀr

Question 7

Under a column-vector transition convention, what must each column of a transition matrix sum to?

A. 0

B. −1

C. 1

D. The number of states

Correct Answer: C. 1

Question 8

In the Leontief model, what does d represent?

A. The total technical-coefficient matrix

B. Final demand

C. The identity matrix

D. The determinant

Correct Answer: B. Final demand

Question 9

What is the standard residual for a computed solution x to Ax = b?

A. Ax + b

B. A + xb

C. b − Ax

D. A⁻¹ + b

Correct Answer: C. b − Ax

Question 10

Why might a numerical solver be preferred to explicitly computing A⁻¹?

A. An inverse exists for every matrix

B. Solvers are often more stable and efficient

C. Solvers remove all modeling assumptions

D. The inverse method cannot solve square systems

Correct Answer: B. Solvers are often more stable and efficient