Chapter 3 Key Takeaways — Systems of Linear Equations

The big ideas

  • A system of linear equations is a list of constraints the unknowns must satisfy all at once. Each linear equation allows many possibilities; the system asks for the overlap. Solving it is finding that overlap — and the "system of equations meaning" is exactly this simultaneous satisfaction.

  • There are only three possible solution counts: none, exactly one, or infinitely many. Never two, never seventeen. This trichotomy is a privilege of linearity — geometrically, the solution set is always a flat object (a point, line, plane, …), a direct consequence of the superposition principle from Chapter 1.

  • Every system has two equally important geometric readings. The row picture: each equation is a line (2D) or plane (3D), and the solution is where the surfaces intersect. The column picture: writing the system as $A\mathbf{x} = \mathbf{b}$, the solution is the recipe of weights that builds $\mathbf{b}$ as a linear combination of the columns of $A$. Same equations, same answer, two complementary stories.

  • The column picture is the doorway to the rest of the book. "Does $A\mathbf{x} = \mathbf{b}$ have a solution?" becomes "is $\mathbf{b}$ a combination of the columns?" — which becomes "is $\mathbf{b}$ in the column space $C(A)$?" of Chapter 13. The matrix–vector product $A\mathbf{x}$ from Chapter 1 is that combination.

  • Inconsistent vs. dependent have precise geometric and algebraic meanings. Inconsistent (no solution): surfaces share no common point; $\mathbf{b}$ lies outside the span of the columns; honest manipulation yields a contradiction like $0 = 3$; and $\operatorname{rank}([A\mid\mathbf{b}]) > \operatorname{rank}(A)$. Dependent (infinitely many): equations are redundant; the solution set carries free variables; $\operatorname{rank}([A\mid\mathbf{b}]) = \operatorname{rank}(A) < n$.

  • Rank decides everything (the Rouché–Capelli criterion). Consistent $\iff \operatorname{rank}(A) = \operatorname{rank}([A\mid\mathbf{b}])$. If consistent: unique when $\operatorname{rank}(A) = n$ (number of unknowns), infinitely many when $\operatorname{rank}(A) < n$, with exactly $n - \operatorname{rank}(A)$ free variables. The number of free variables equals the dimension of the solution set.

  • For a square system, it's a clean dichotomy. $A$ invertible ($\det A \neq 0$, full rank) → exactly one solution for every $\mathbf{b}$. $A$ singular ($\det A = 0$) → no solution or infinitely many, never exactly one.

  • Solution sets have the structure "particular + homogeneous." Every solution of a consistent $A\mathbf{x} = \mathbf{b}$ is $\mathbf{x}_0 + \mathbf{h}$, where $\mathbf{x}_0$ is one particular solution and $\mathbf{h}$ ranges over the homogeneous solutions ($A\mathbf{h} = \mathbf{0}$, the null space). The solution set is a subspace translated to pass through $\mathbf{x}_0$.

Skills you gained

  • Recognize and write a linear equation; distinguish coefficients, unknowns, and the constant.
  • Read any 2×2 or 3×3 system in both the row and column pictures, and sketch the corresponding lines/planes or column arrows.
  • Classify a system as none / one / infinitely many without solving it, using geometry and the rank test.
  • Predict what np.linalg.solve returns: the unique vector, or LinAlgError: Singular matrix; and use np.linalg.matrix_rank and np.linalg.lstsq to diagnose the singular case.
  • Form the augmented matrix $[A\mid\mathbf{b}]$ and apply the consistency / uniqueness criterion.
  • Translate a real problem (economic input–output, circuit currents, network flow, web ranking) into a system $A\mathbf{x} = \mathbf{b}$ and interpret its solution set.

Terms to know

linear equation, system of linear equations, coefficient, unknown, constant / right-hand side, solution set, consistent, inconsistent, row picture, column picture, linear combination, coefficient matrix, augmented matrix $[A\mid\mathbf{b}]$, homogeneous system, trivial solution, free variable, rank, singular, dependent.

How this connects (recurring themes & the road ahead)

  • Geometry and algebra are two views of one object (Theme 2): the row picture (intersection) and column picture (combination) are the same question asked two ways — the habit of holding both is the working style this book trains.
  • The four fundamental subspaces (Theme 5): the column picture seeds the column space $C(A)$ (Chapter 13); homogeneous systems seed the null space $N(A)$ (Chapter 13); the free-variable count matures into rank–nullity (Chapter 14).
  • Linear algebra is the most applied branch of pure math (Theme 4): the very same $A\mathbf{x} = \mathbf{b}$ balances an economy, finds circuit currents, routes network flow, and — scaled to billions of unknowns — ranks the entire web.

Forward references. How to solve systematically is Chapter 4 (Gaussian elimination and row reduction), where you implement gaussian_elimination in toolkit/linear_systems.py. The invertibility/determinant conditions are Chapters 9 and 11. The column space and null space are Chapters 13–14. And the anchor we seeded — ranking the web as a giant linear system whose solution is an eigenvector — is solved in Chapter 29 (PageRank), building on eigenvectors from Chapter 23. You now know what a system means; next you learn how to solve it.