πŸ”’ Private Site

This site is password-protected.

Engineering Mathematics-II β€” Unit I Notes

Course: Engineering Mathematics-II (Linear Algebra and Differential Equations)
Institute: Indian Institute of Information Technology Pune β€” F.Y. B.Tech Semester II
Syllabus Coverage: Systems of linear equations & Gauss-elimination; Vector space, subspace, spanning sets; Linear independence & dependence; Basis & dimension; Rank of matrix; Inner product; Gram-Schmidt process (08 Hrs)
Text Books: Serge Lang β€” Introduction to Linear Algebra; Erwin Kreyszig β€” Advanced Engineering Mathematics


Table of Contents

1 β€” Systems of Linear Equations & Gauss Elimination

2 β€” Vector Spaces

3 β€” Subspaces & Spanning Sets

4 β€” Linear Independence & Dependence

5 β€” Basis & Dimension

6 β€” Rank of a Matrix

7 β€” Inner Product & Gram-Schmidt Process

8 β€” Coordinates of a Vector (with Respect to a Basis)

9 β€” Linear Transformations (Linear Mappings)

Comprehensive Unit I Practice Set (with Answers)


Glossary β€” Key Terms at a Glance

Term Meaning
System of Linear Equations A collection of linear equations involving the same set of unknowns.
Augmented Matrix The matrix $[A \mid b]$ combining coefficient matrix $A$ and constant vector $b$.
Elementary Row Operations Row addition, row multiplication (by non-zero scalar), row interchange β€” operations that don’t change the solution.
Pivot The first non-zero entry in a row of a row echelon form matrix.
Row Echelon Form (REF) A matrix where: all-zero rows are at bottom, each pivot is to the right of the one above, and all entries below each pivot are zero.
Gauss Elimination A method to solve linear systems by reducing the augmented matrix to REF, then back-substituting.
Vector Space A set $V$ with addition and scalar multiplication operations satisfying 10 axioms (closure, commutativity, associativity, identities, inverses, distributivity).
Subspace A non-empty subset $W \subseteq V$ that is itself a vector space under the same operations.
Span The set of all possible linear combinations of a collection of vectors.
Linear Independence A set of vectors where the only way to get $\vec{0}$ as a linear combination is with all-zero coefficients.
Linear Dependence A set of vectors where at least one vector is a linear combination of the others.
Basis A linearly independent spanning set of a vector space.
Dimension The number of vectors in any basis of a vector space.
Rank The number of linearly independent rows (or columns) of a matrix; equivalently, the number of pivots in REF.
Nullity The dimension of the null space (solution space of $Ax = 0$).
Inner Product A generalization of the dot product β€” a function $\langle u, v \rangle$ satisfying symmetry, linearity, and positive-definiteness.
Orthogonal Two vectors $u, v$ are orthogonal if $\langle u, v \rangle = 0$.
Orthonormal A set is orthonormal if every pair is orthogonal and every vector has unit norm.
Gram-Schmidt Process An algorithm that converts any basis into an orthonormal basis.
Coordinate Vector The tuple of scalars $[\vec{v}]_B = (c_1, \ldots, c_n)^T$ representing $\vec{v}$ as $c_1\vec{v}_1 + \cdots + c_n\vec{v}_n$ in a given basis $B$.
Transition Matrix The matrix $P_{B \to B’}$ that converts coordinate vectors from basis $B$ to basis $B’$: $[\vec{v}]{B’} = P{B \to B’}[\vec{v}]_B$.
Linear Transformation A function $T: V \to W$ between vector spaces that preserves addition and scalar multiplication.
Kernel (Null Space) The set of all vectors mapped to $\vec{0}$ by a linear transformation: $\ker(T) = \lbrace\vec{v} : T(\vec{v}) = \vec{0}\rbrace$.
Image (Range) The set of all output vectors of a linear transformation: $\text{Im}(T) = \lbrace T(\vec{v}) : \vec{v} \in V\rbrace$.
Isomorphism A bijective (one-to-one and onto) linear transformation between vector spaces.

1 β€” Systems of Linear Equations & Gauss Elimination


1.1 What is a System of Linear Equations?

Definition: A system of $m$ linear equations in $n$ unknowns has the form:

\[a_{11}x_1 + a_{12}x_2 + \cdots + a_{1n}x_n = b_1 \\ a_{21}x_1 + a_{22}x_2 + \cdots + a_{2n}x_n = b_2 \\ \vdots \\ a_{m1}x_1 + a_{m2}x_2 + \cdots + a_{mn}x_n = b_m\]

where $a_{ij}$ are coefficients, $x_j$ are unknowns, and $b_i$ are constants.

Why do we care?

Almost every engineering problem β€” circuit analysis, force equilibrium, traffic flow, chemical balancing, economics β€” reduces to solving a system of linear equations. Gauss elimination gives us a systematic, computer-friendly algorithm to solve any such system.

Classification of Systems

Type Meaning Example
Consistent Has at least one solution $x+y=3,\; x-y=1$ β†’ $(2,1)$
Inconsistent Has no solution $x+y=3,\; x+y=5$ (parallel lines)
Unique Solution Exactly one solution Lines intersecting at one point
Infinitely Many Entire family of solutions Dependent equations (same line/plane)

1.2 Matrix Representation & Augmented Matrix

Any system $Ax = b$ can be written compactly:

\[A = \begin{bmatrix} a_{11} & a_{12} & \cdots & a_{1n} \\ a_{21} & a_{22} & \cdots & a_{2n} \\ \vdots & \vdots & \ddots & \vdots \\ a_{m1} & a_{m2} & \cdots & a_{mn} \end{bmatrix}, \quad x = \begin{bmatrix} x_1 \\ x_2 \\ \vdots \\ x_n \end{bmatrix}, \quad b = \begin{bmatrix} b_1 \\ b_2 \\ \vdots \\ b_m \end{bmatrix}\]

The Augmented Matrix is formed by appending $b$ to $A$:

\[[A \mid b] = \begin{bmatrix} a_{11} & a_{12} & \cdots & a_{1n} & b_1 \\ a_{21} & a_{22} & \cdots & a_{2n} & b_2 \\ \vdots & \vdots & \ddots & \vdots & \vdots \\ a_{m1} & a_{m2} & \cdots & a_{mn} & b_m \end{bmatrix}\]

This is the starting point for Gauss elimination.

Quick Example: The system

\[\begin{cases} x + 2y = 8 \\ 3x + 4y = 18 \end{cases}\]

has augmented matrix:

\[[A \mid b] = \begin{bmatrix} 1 & 2 & 8 \\ 3 & 4 & 18 \end{bmatrix}\]

1.3 Elementary Row Operations

These are the only operations we perform during Gauss elimination. They do NOT change the solution of the system.

Operation Notation Description
Row Addition $R_i \leftarrow R_i + k \cdot R_j$ Add $k$ times row $j$ to row $i$
Row Multiplication $R_i \leftarrow k \cdot R_i$ Multiply row $i$ by non-zero scalar $k$
Row Interchange $R_i \leftrightarrow R_j$ Swap two rows

Why are these safe? Each row operation corresponds to multiplying both sides of an equation by a constant, adding two equations together, or reordering equations β€” none of which changes the solution set.


1.4 Gauss Elimination β€” Forward Elimination

Gauss Elimination is a two-phase algorithm:

Phase 1 β€” Forward Elimination: Transform $[A \mid b]$ β†’ upper triangular form (all zeros below the diagonal).

Phase 2 β€” Back Substitution: Solve for unknowns starting from the last row upward.

Forward Elimination Step-by-Step

For each column $k = 1, 2, \ldots, n-1$:

  1. Identify the pivot: Element $a_{kk}$ (diagonal entry in row $k$, column $k$).
  2. If pivot is zero: Swap row $k$ with a row below that has a non-zero entry in column $k$.
  3. Eliminate below: For each row $i > k$, compute the multiplier:
\[m_{ik} = \frac{a_{ik}}{a_{kk}}\]

Then perform: $R_i \leftarrow R_i - m_{ik} \cdot R_k$

This creates a zero in position $(i, k)$.

After processing all columns, the matrix is in upper triangular form.


1.5 Back Substitution

Starting from the last row (which has only one unknown) and working upward:

\[x_n = \frac{b'_n}{a'_{nn}}\]

For rows $i = n-1, n-2, \ldots, 1$:

\[x_i = \frac{b'_i - \sum_{j=i+1}^{n} a'_{ij} x_j}{a'_{ii}}\]

Think of it as: Solve the simplest equation first (bottom row), then plug those values into the equation above, and so on. Like β€œunwinding” a chain of equations.


1.6 Row Echelon Form (REF)

A matrix is in Row Echelon Form if:

  1. All rows consisting entirely of zeros are at the bottom.
  2. The first non-zero entry (pivot) of each non-zero row is to the right of the pivot of the row above.
  3. All entries below each pivot are zero.

REF example:

\[\begin{bmatrix} \boxed{1} & 2 & 3 & 4 \\ 0 & \boxed{5} & 6 & 7 \\ 0 & 0 & \boxed{8} & 9 \\ 0 & 0 & 0 & 0 \end{bmatrix}\]

Pivots are boxed. Three non-zero rows β†’ rank = 3.


1.7 Partial Pivoting & Numerical Stability

Problem: If the pivot $a_{kk}$ is very small (near zero), dividing by it amplifies rounding errors.

Solution β€” Partial Pivoting: Before eliminating column $k$, scan the entries below (and including) the pivot. Swap the current row with the row that has the largest absolute value in that column.

\[\text{Select row } p \text{ such that } \lvert a_{pk} \rvert = \max_{i \geq k} \lvert a_{ik} \rvert, \text{ then } R_k \leftrightarrow R_p\]

Why it matters β€” a concrete example:

System: $0.0001x + y = 1$ and $x + y = 2$.

Exact solution (for reference): $x = \dfrac{10000}{9999} \approx 1.0001$, $\;y = \dfrac{19998}{9999} \approx 1.9999$.


Without Pivoting (4-digit arithmetic)

Augmented matrix:

\[\left[\begin{array}{cc|c} 0.0001 & 1 & 1 \\ 1 & 1 & 2 \end{array}\right]\]

Step 1 β€” Eliminate below pivot $a_{11} = 0.0001$:

Multiplier: $m = \dfrac{1}{0.0001} = 10000$

\[R_2 \leftarrow R_2 - 10000 \cdot R_1\] \[\left[\begin{array}{cc|c} 0.0001 & 1 & 1 \\ 0 & 1 - 10000 & 2 - 10000 \end{array}\right] = \left[\begin{array}{cc|c} 0.0001 & 1 & 1 \\ 0 & -9999 & -9998 \end{array}\right]\]

In 4-digit floating point, $-9999$ rounds to $-10000$ and $-9998$ rounds to $-10000$:

\[\left[\begin{array}{cc|c} 0.0001 & 1 & 1 \\ 0 & -10000 & -10000 \end{array}\right]\]

Step 2 β€” Back substitution:

\[y = \frac{-10000}{-10000} = 1.000\] \[0.0001x + 1 = 1 \implies 0.0001x = 0 \implies x = 0\]

Result without pivoting: $x = 0, \; y = 1$ β€” wildly wrong! (true $x \approx 1$)

The tiny pivot caused the multiplier $10000$ to amplify the rounding error and obliterate the true value of $x$.


With Partial Pivoting (4-digit arithmetic)

Before elimination, compare $\lvert a_{11} \rvert = 0.0001$ vs $\lvert a_{21} \rvert = 1$. Since $1 > 0.0001$, swap $R_1 \leftrightarrow R_2$:

\[\left[\begin{array}{cc|c} 1 & 1 & 2 \\ 0.0001 & 1 & 1 \end{array}\right]\]

Step 1 β€” Eliminate below pivot $a_{11} = 1$:

Multiplier: $m = \dfrac{0.0001}{1} = 0.0001$ (small β€” no error amplification!)

\[R_2 \leftarrow R_2 - 0.0001 \cdot R_1\] \[\left[\begin{array}{cc|c} 1 & 1 & 2 \\ 0 & 1 - 0.0001 & 1 - 0.0002 \end{array}\right] = \left[\begin{array}{cc|c} 1 & 1 & 2 \\ 0 & 0.9999 & 0.9998 \end{array}\right]\]

In 4-digit floating point, $0.9999$ and $0.9998$ are stored exactly (4 significant digits):

\[\left[\begin{array}{cc|c} 1 & 1 & 2 \\ 0 & 0.9999 & 0.9998 \end{array}\right]\]

Step 2 β€” Back substitution:

\[y = \frac{0.9998}{0.9999} = 0.9999 \approx 1.000\] \[x + 1.000 = 2 \implies x = 1.000\]

Result with pivoting: $x = 1.000, \; y = 1.000$ β€” excellent accuracy! βœ“


Takeaway: Partial pivoting keeps multipliers $\lvert m \rvert \leq 1$, preventing catastrophic amplification of rounding errors. Always use the largest available pivot.


1.8 Types of Solutions β€” Unique, Infinite, No Solution

After Gauss elimination, examine the REF:

REF Pattern Type Condition
All pivots in every column, no zero rows Unique solution $\text{rank}(A) = \text{rank}([A\mid b]) = n$
A zero row with zero on RHS too ($0 = 0$) Infinitely many solutions $\text{rank}(A) = \text{rank}([A\mid b]) < n$
A zero row with non-zero RHS ($0 = c, \; c \neq 0$) No solution (inconsistent) $\text{rank}(A) < \text{rank}([A\mid b])$

Key Takeaway: After reducing to REF, if any row reads $[0 \; 0 \; \cdots \; 0 \mid c]$ where $c \neq 0$, the system is inconsistent β€” no solution exists!


1.9 Worked Examples β€” Gauss Elimination

Example 1: 2Γ—2 System (Unique Solution)

Problem: Solve

\[\begin{cases} x + 2y = 8 \\ 3x + 4y = 18 \end{cases}\]

Step 1 β€” Augmented Matrix:

\[\begin{bmatrix} 1 & 2 & 8 \\ 3 & 4 & 18 \end{bmatrix}\]

Step 2 β€” Forward Elimination: Pivot = $a_{11} = 1$. Multiplier: $m_{21} = 3/1 = 3$. Operation: $R_2 \leftarrow R_2 - 3R_1$

\[\begin{bmatrix} 1 & 2 & 8 \\ 0 & -2 & -6 \end{bmatrix}\]

Step 3 β€” Back Substitution:

  • Row 2: $-2y = -6 \Rightarrow y = 3$
  • Row 1: $x + 2(3) = 8 \Rightarrow x = 2$

Solution: $x = 2, \; y = 3$ βœ“

Verification: $2 + 2(3) = 8$ βœ“, $3(2) + 4(3) = 18$ βœ“

Example 2: 3Γ—3 System (Unique Solution)

Problem: Solve

\[\begin{cases} x + y + z = 6 \\ 2x + 3y + z = 13 \\ x + 2y + 3z = 16 \end{cases}\]

Augmented Matrix:

\[\begin{bmatrix} 1 & 1 & 1 & 6 \\ 2 & 3 & 1 & 13 \\ 1 & 2 & 3 & 16 \end{bmatrix}\]

Eliminate column 1: $R_2 \leftarrow R_2 - 2R_1$, $R_3 \leftarrow R_3 - R_1$

\[\begin{bmatrix} 1 & 1 & 1 & 6 \\ 0 & 1 & -1 & 1 \\ 0 & 1 & 2 & 10 \end{bmatrix}\]

Eliminate column 2: $R_3 \leftarrow R_3 - R_2$

\[\begin{bmatrix} 1 & 1 & 1 & 6 \\ 0 & 1 & -1 & 1 \\ 0 & 0 & 3 & 9 \end{bmatrix}\]

Back Substitution:

  • Row 3: $3z = 9 \Rightarrow z = 3$
  • Row 2: $y - 3 = 1 \Rightarrow y = 4$
  • Row 1: $x + 4 + 3 = 6 \Rightarrow x = -1$

Solution: $x = -1, \; y = 4, \; z = 3$ βœ“

Example 3: Infinitely Many Solutions

Problem: Solve

\[\begin{cases} x + 2y + 3z = 14 \\ 2x + 4y + 6z = 28 \\ x + y + z = 6 \end{cases}\]

Augmented Matrix:

\[\begin{bmatrix} 1 & 2 & 3 & 14 \\ 2 & 4 & 6 & 28 \\ 1 & 1 & 1 & 6 \end{bmatrix}\]

Eliminate column 1: $R_2 \leftarrow R_2 - 2R_1$, $R_3 \leftarrow R_3 - R_1$

\[\begin{bmatrix} 1 & 2 & 3 & 14 \\ 0 & 0 & 0 & 0 \\ 0 & -1 & -2 & -8 \end{bmatrix}\]

Row 2 is all zeros β€” equations are dependent. Swap $R_2 \leftrightarrow R_3$:

\[\begin{bmatrix} 1 & 2 & 3 & 14 \\ 0 & -1 & -2 & -8 \\ 0 & 0 & 0 & 0 \end{bmatrix}\]

2 equations, 3 unknowns β€” there are more unknowns than pivot rows.

Pivot variables vs Free variables: After row reduction, columns that contain a pivot (leading 1 or leading non-zero entry) correspond to pivot variables β€” they are determined by back-substitution. Columns without a pivot correspond to free variables β€” they can take any real value, and the pivot variables are expressed in terms of them. Free variables are the source of infinitely many solutions.

Here column 3 ($z$) has no pivot, so $z$ is a free variable. Let $z = t$ (free parameter):

  • Row 2: $-y - 2t = -8 \Rightarrow y = 8 - 2t$
  • Row 1: $x + 2(8 - 2t) + 3t = 14 \Rightarrow x = -2 + t$

Solution:

\[\begin{pmatrix} x \\ y \\ z \end{pmatrix} = \begin{pmatrix} -2 \\ 8 \\ 0 \end{pmatrix} + t \begin{pmatrix} 1 \\ -2 \\ 1 \end{pmatrix}, \quad t \in \mathbb{R}\]

Example 4: Partial Pivoting

Problem: Solve

\[\begin{cases} 0.0001x + y = 1 \\ x + y = 2 \end{cases}\]

The pivot $0.0001$ is very small β€” swap rows first!

\[\begin{bmatrix} 0.0001 & 1 & 1 \\ 1 & 1 & 2 \end{bmatrix} \xrightarrow{R_1 \leftrightarrow R_2} \begin{bmatrix} 1 & 1 & 2 \\ 0.0001 & 1 & 1 \end{bmatrix}\]

$R_2 \leftarrow R_2 - 0.0001 R_1$:

\[\begin{bmatrix} 1 & 1 & 2 \\ 0 & 0.9999 & 0.9998 \end{bmatrix}\]

Back substitution: $y \approx 1$, $x = 2 - 1 = 1$.

Solution: $x = 1, \; y = 1$ (with improved numerical accuracy!)


1.10 Parametric Consistency Analysis

Sometimes a system contains parameters (like $\lambda, \mu, k$) and we must determine which values lead to unique solutions, infinitely many solutions, or no solution. The strategy is:

  1. Form the augmented matrix $[A \mid B]$
  2. Row reduce β€” the parameters will appear in the entries
  3. Analyze the last row: if it reads $[0 \; 0 \; \cdots \; c \mid d]$ with $c$ and $d$ involving parameters, then:
    • $c \neq 0$: unique solution (full rank)
    • $c = 0$ and $d = 0$: infinitely many solutions
    • $c = 0$ and $d \neq 0$: no solution (inconsistent)

Example 5: Parametric System with $\lambda, \mu$

Problem: Investigate for what values of $\lambda, \mu$ the system

\[\begin{cases} x + 2y + z = 8 \\ 2x + 2y + 2z = 13 \\ 3x + 4y + \lambda z = \mu \end{cases}\]

has (i) no solution, (ii) a unique solution, (iii) infinitely many solutions.

Augmented matrix:

\[\left[\begin{array}{ccc|c} 1 & 2 & 1 & 8 \\ 2 & 2 & 2 & 13 \\ 3 & 4 & \lambda & \mu \end{array}\right]\]

$R_2 \leftarrow R_2 - 2R_1$, $R_3 \leftarrow R_3 - 3R_1$:

\[\left[\begin{array}{ccc|c} 1 & 2 & 1 & 8 \\ 0 & -2 & 0 & -3 \\ 0 & -2 & \lambda - 3 & \mu - 24 \end{array}\right]\]

$R_3 \leftarrow R_3 - R_2$:

\[\left[\begin{array}{ccc|c} 1 & 2 & 1 & 8 \\ 0 & -2 & 0 & -3 \\ 0 & 0 & \lambda - 3 & \mu - 21 \end{array}\right]\]

Analysis of Row 3:

(i) Unique solution: $\lambda - 3 \neq 0$, i.e., $\lambda \neq 3$ (for any $\mu$). Then $\rho(A) = \rho(A B) = 3$.
(ii) Infinitely many solutions: $\lambda = 3$ and $\mu = 21$. Then Row 3 becomes $[0\;0\;0\;\vert\;0]$, giving $\rho(A) = \rho(A B) = 2 < 3$.
(iii) No solution: $\lambda = 3$ and $\mu \neq 21$. Then Row 3 reads $[0\;0\;0\;\vert\;\mu - 21]$ with $\mu - 21 \neq 0$, giving $\rho(A) = 2 \neq 3 = \rho(A B)$.

1.11 Homogeneous Systems ($AX = 0$)

A homogeneous linear system $AX = 0$ always has the trivial solution $\vec{x} = \vec{0}$.

  • If $\rho(A) = n$ (number of unknowns): only the trivial solution exists.
  • If $\rho(A) < n$: infinitely many non-trivial solutions exist. The number of free parameters equals $n - \rho(A)$.

For a square homogeneous system $AX = 0$ (where $A$ is $n \times n$):

  • $\det(A) \neq 0 \Rightarrow$ only trivial solution ($\rho(A) = n$)
  • $\det(A) = 0 \Rightarrow$ non-trivial solutions exist ($\rho(A) < n$)

Example 6: Homogeneous System β€” Non-Trivial Solutions

Problem: Solve the homogeneous system:

\[\begin{cases} 3x + 4y - z - 9w = 0 \\ 2x + 3y + 2z - 3w = 0 \\ 2x + y - 14z - 12w = 0 \\ x + 3y + 13z + 3w = 0 \end{cases}\]

Coefficient matrix:

\[A = \begin{bmatrix} 3 & 4 & -1 & -9 \\ 2 & 3 & 2 & -3 \\ 2 & 1 & -14 & -12 \\ 1 & 3 & 13 & 3 \end{bmatrix}\]

Swap $R_1 \leftrightarrow R_4$ for unit pivot, then eliminate:

\[\xrightarrow{\text{row reduce}} \begin{bmatrix} 1 & 3 & 13 & 3 \\ 0 & -3 & -24 & -9 \\ 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 \end{bmatrix}\]

$\rho(A) = 2 < 4$ (unknowns) $\Rightarrow$ non-trivial solutions exist.

Number of free variables $= 4 - 2 = 2$. Let $z = s$, $w = t$ (free).

From Row 2: $-3y - 24s - 9t = 0 \Rightarrow y = -8s - 3t$

From Row 1: $x + 3(-8s - 3t) + 13s + 3t = 0 \Rightarrow x = 11s + 6t$

Solution:

\[(x, y, z, w) = s(11, -8, 1, 0) + t(6, -3, 0, 1), \quad s, t \in \mathbb{R}\]

Example 7: Parametric Homogeneous System β€” Values of $k$

Problem: Discuss for all values of $k$, whether the system has only the trivial solution or also non-trivial solutions:

\[\begin{cases} 2x + 3ky + (3k+4)z = 0 \\ x + (k+1)y + (4k+2)z = 0 \\ x + 2(k+1)y + (3k+4)z = 0 \end{cases}\]

Coefficient matrix and determinant:

\[A = \begin{bmatrix} 2 & 3k & 3k+4 \\ 1 & k+1 & 4k+2 \\ 1 & 2(k+1) & 3k+4 \end{bmatrix}\]

Computing $\det(A)$:

\[\det(A) = 2[(k+1)(3k+4) - (4k+2) \cdot 2(k+1)] - 3k[(3k+4) - (4k+2)] + (3k+4)[2(k+1) - (k+1)]\]

After expansion and simplification:

\[\det(A) = -4(k+2)(k-2) = -4(k^2 - 4)\]

Analysis:

  • Trivial solution only: $\det(A) \neq 0 \Rightarrow k \neq \pm 2$
  • Non-trivial solutions: $\det(A) = 0 \Rightarrow k = 2$ or $k = -2$

Example 8: Eigenvalue Problem β€” $|A - \lambda I| = 0$

Problem: Find values of $\lambda$ for which the following homogeneous system has non-trivial solutions:

\[\begin{cases} (-1-\lambda)x_1 + 2x_2 + 3x_3 = 0 \\ 3x_1 + (-\lambda)x_2 + 2x_3 = 0 \\ 2x_1 + 3x_2 + (1-\lambda)x_3 = 0 \end{cases}\]

This system can be written as $(A - \lambda I)\vec{x} = \vec{0}$ where:

\[A = \begin{bmatrix} -1 & 2 & 3 \\ 3 & 0 & 2 \\ 2 & 3 & 1 \end{bmatrix}\]

For non-trivial solutions, we need $\det(A - \lambda I) = 0$:

\[\det \begin{bmatrix} -1-\lambda & 2 & 3 \\ 3 & -\lambda & 2 \\ 2 & 3 & 1-\lambda \end{bmatrix} = 0\]

Expanding the determinant:

\[(-1-\lambda)[-\lambda(1-\lambda) - 6] - 2[3(1-\lambda) - 4] + 3[9 + 2\lambda] = 0\] \[(-1-\lambda)(\lambda^2 - \lambda - 6) - 2(-1 - 3\lambda) + 3(9 + 2\lambda) = 0\] \[-\lambda^3 + 15\lambda + 18 = 0\]

This is the characteristic equation. The values of $\lambda$ satisfying this equation are called eigenvalues β€” a topic explored fully in Unit II.


Practice Questions β€” Systems & Gauss Elimination

Q1. Solve using Gauss elimination:

\[\begin{cases} 2x + y = 7 \\ x - y = 2 \end{cases}\]

(Ans: x = 3, y = 1)

Click for solution

Augmented matrix:

\[\left[\begin{array}{cc|c} 2 & 1 & 7 \\ 1 & -1 & 2 \end{array}\right]\]

Step 1 β€” Swap $R_1 \leftrightarrow R_2$ (unit pivot):

\[\left[\begin{array}{cc|c} 1 & -1 & 2 \\ 2 & 1 & 7 \end{array}\right]\]

Step 2 β€” $R_2 \leftarrow R_2 - 2R_1$:

\[\left[\begin{array}{cc|c} 1 & -1 & 2 \\ 0 & 3 & 3 \end{array}\right]\]

Back substitution: $3y = 3 \Rightarrow y = 1$. Then $x - 1 = 2 \Rightarrow x = 3$.

Solution: $x = 3, \; y = 1$ βœ“

Q2. Solve:

\[\begin{cases} 3x - 2y = 5 \\ 6x + 4y = 3 \end{cases}\]

Determine the type of solution.

Click for solution

Augmented matrix:

\[\left[\begin{array}{cc|c} 3 & -2 & 5 \\ 6 & 4 & 3 \end{array}\right]\]

Step 1 β€” $R_2 \leftarrow R_2 - 2R_1$:

\[\left[\begin{array}{cc|c} 3 & -2 & 5 \\ 0 & 8 & -7 \end{array}\right]\]

Back substitution: $8y = -7 \Rightarrow y = -\dfrac{7}{8}$.

\[3x = 5 + 2\left(-\frac{7}{8}\right) = 5 - \frac{7}{4} = \frac{13}{4} \Rightarrow x = \frac{13}{12}\]

Type: Unique solution. Coefficient determinant: $\Delta = 3(4) - (-2)(6) = 24 \neq 0$.

Solution: $x = \dfrac{13}{12}, \; y = -\dfrac{7}{8}$

Q3. Solve:

\[\begin{cases} x + 3y = 9 \\ 2x + 6y = 18 \end{cases}\]

(Ans: Infinitely many β€” $(x,y) = (9-3t, \; t)$)

Click for solution

Augmented matrix:

\[\left[\begin{array}{cc|c} 1 & 3 & 9 \\ 2 & 6 & 18 \end{array}\right]\]

Step 1 β€” $R_2 \leftarrow R_2 - 2R_1$:

\[\left[\begin{array}{cc|c} 1 & 3 & 9 \\ 0 & 0 & 0 \end{array}\right]\]

Row 2 is all zeros β€” the two equations are dependent ($R_2 = 2 R_1$).

From Row 1: $x = 9 - 3y$. Set $y = t$ (free parameter).

Solution: $(x, y) = (9 - 3t, \; t)$ for all $t \in \mathbb{R}$.

Examples: $t = 0 \Rightarrow (9, 0)$; $t = 1 \Rightarrow (6, 1)$; $t = 3 \Rightarrow (0, 3)$.

Q4. Solve the 3Γ—3 system:

\[\begin{cases} x + 2y - z = -1 \\ 2x - y + 3z = 8 \\ 3x + y + z = 6 \end{cases}\]

(Ans: x = 2, y = -1, z = 1)

Click for solution

Augmented matrix:

\[\left[\begin{array}{ccc|c} 1 & 2 & -1 & -1 \\ 2 & -1 & 3 & 8 \\ 3 & 1 & 1 & 6 \end{array}\right]\]

Step 1 β€” Eliminate column 1:

$R_2 \leftarrow R_2 - 2R_1$: $[0, \; -5, \; 5 \;\vert\; 10]$

$R_3 \leftarrow R_3 - 3R_1$: $[0, \; -5, \; 4 \;\vert\; 9]$

\[\left[\begin{array}{ccc|c} 1 & 2 & -1 & -1 \\ 0 & -5 & 5 & 10 \\ 0 & -5 & 4 & 9 \end{array}\right]\]

Step 2 β€” Eliminate column 2:

$R_3 \leftarrow R_3 - R_2$: $[0, \; 0, \; -1 \;\vert\; -1]$

Back substitution:

\[z = \frac{-1}{-1} = 1, \quad -5y + 5(1) = 10 \Rightarrow y = -1, \quad x + 2(-1) - 1 = -1 \Rightarrow x = 2\]

Solution: $x = 2, \; y = -1, \; z = 1$ βœ“

Q5. Solve:

\[\begin{cases} 2x + 3y + z = 11 \\ x - y + 2z = 5 \\ 3x + y - z = 2 \end{cases}\]

(Ans: x = 1, y = 2, z = 3)

Click for solution

Augmented matrix:

\[\left[\begin{array}{ccc|c} 2 & 3 & 1 & 11 \\ 1 & -1 & 2 & 5 \\ 3 & 1 & -1 & 2 \end{array}\right]\]

Step 1 β€” Swap $R_1 \leftrightarrow R_2$ (unit pivot):

\[\left[\begin{array}{ccc|c} 1 & -1 & 2 & 5 \\ 2 & 3 & 1 & 11 \\ 3 & 1 & -1 & 2 \end{array}\right]\]

Step 2 β€” Eliminate column 1:

$R_2 \leftarrow R_2 - 2R_1$: $[0, \; 5, \; -3 \;\vert\; 1]$

$R_3 \leftarrow R_3 - 3R_1$: $[0, \; 4, \; -7 \;\vert\; -13]$

Step 3 β€” Eliminate column 2:

$R_3 \leftarrow R_3 - \frac{4}{5}R_2$: $\left[0, \; 0, \; -7 + \frac{12}{5} \;\middle\vert\; -13 - \frac{4}{5}\right] = \left[0, \; 0, \; -\frac{23}{5} \;\middle\vert\; -\frac{69}{5}\right]$

Back substitution:

\[z = \frac{69/5}{23/5} = 3, \quad 5y - 9 = 1 \Rightarrow y = 2, \quad x - 2 + 6 = 5 \Rightarrow x = 1\]

Solution: $x = 1, \; y = 2, \; z = 3$ βœ“

Q6. Solve with partial pivoting:

\[\begin{cases} 0.001x + 2y + z = 3.001 \\ x + y + z = 3 \\ 2x + y - z = 2 \end{cases}\]
Click for full solution

Answer: $x = 1, \; y = 1, \; z = 1$

Step 0 β€” Write the augmented matrix:

\[\left[\begin{array}{ccc|c} 0.001 & 2 & 1 & 3.001 \\ 1 & 1 & 1 & 3 \\ 2 & 1 & -1 & 2 \end{array}\right]\]

Step 1 β€” Partial pivoting on column 1:

Scan column 1: $\lvert 0.001 \rvert, \lvert 1 \rvert, \lvert 2 \rvert$. Largest is $2$ (row 3). Swap $R_1 \leftrightarrow R_3$:

\[\left[\begin{array}{ccc|c} 2 & 1 & -1 & 2 \\ 1 & 1 & 1 & 3 \\ 0.001 & 2 & 1 & 3.001 \end{array}\right]\]

Step 2 β€” Eliminate column 1 below pivot:

$R_2 \leftarrow R_2 - \frac{1}{2} R_1$:

\[R_2: \left[0, \; 1 - 0.5, \; 1 + 0.5, \; 3 - 1\right] = \left[0, \; 0.5, \; 1.5, \; 2\right]\]

$R_3 \leftarrow R_3 - \frac{0.001}{2} R_1 = R_3 - 0.0005 \, R_1$:

\[R_3: \left[0, \; 2 - 0.0005, \; 1 + 0.0005, \; 3.001 - 0.001\right] = \left[0, \; 1.9995, \; 1.0005, \; 3.0\right]\] \[\left[\begin{array}{ccc|c} 2 & 1 & -1 & 2 \\ 0 & 0.5 & 1.5 & 2 \\ 0 & 1.9995 & 1.0005 & 3.0 \end{array}\right]\]

Step 3 β€” Partial pivoting on column 2:

Scan column 2 (rows 2–3): $\lvert 0.5 \rvert, \lvert 1.9995 \rvert$. Largest is $1.9995$ (row 3). Swap $R_2 \leftrightarrow R_3$:

\[\left[\begin{array}{ccc|c} 2 & 1 & -1 & 2 \\ 0 & 1.9995 & 1.0005 & 3.0 \\ 0 & 0.5 & 1.5 & 2 \end{array}\right]\]

Step 4 β€” Eliminate column 2 below pivot:

$m = \dfrac{0.5}{1.9995} \approx 0.25006$

$R_3 \leftarrow R_3 - 0.25006 \, R_2$:

\[R_3: \left[0, \; 0, \; 1.5 - 0.25006(1.0005), \; 2 - 0.25006(3.0)\right]\] \[= \left[0, \; 0, \; 1.5 - 0.25009, \; 2 - 0.75019\right] = \left[0, \; 0, \; 1.24991, \; 1.24981\right]\]

REF achieved:

\[\left[\begin{array}{ccc|c} 2 & 1 & -1 & 2 \\ 0 & 1.9995 & 1.0005 & 3.0 \\ 0 & 0 & 1.24991 & 1.24981 \end{array}\right]\]

Step 5 β€” Back substitution:

\[z = \frac{1.24981}{1.24991} \approx 0.99992 \approx 1\] \[y = \frac{3.0 - 1.0005(1)}{1.9995} = \frac{1.9995}{1.9995} = 1\] \[x = \frac{2 - 1(1) - (-1)(1)}{2} = \frac{2 - 1 + 1}{2} = \frac{2}{2} = 1\]

Solution: $x = 1, \; y = 1, \; z = 1$ βœ“

Verification: $0.001(1) + 2(1) + 1 = 3.001$ βœ“, $\;1 + 1 + 1 = 3$ βœ“, $\;2(1) + 1 - 1 = 2$ βœ“

Q7. Solve the 4Γ—4 system:

\[\begin{cases} x + y + z + w = 10 \\ 2x - y + z - w = 2 \\ 3x + 2y - z + 2w = 13 \\ x - 2y + 2z - w = 2 \end{cases}\]

(Ans: x = 2, y = 1, z = 3, w = 4)

Click for solution

Augmented matrix:

\[\left[\begin{array}{cccc|c} 1 & 1 & 1 & 1 & 10 \\ 2 & -1 & 1 & -1 & 2 \\ 3 & 2 & -1 & 2 & 13 \\ 1 & -2 & 2 & -1 & 2 \end{array}\right]\]

Step 1 β€” Eliminate column 1:

$R_2 \leftarrow R_2 - 2R_1$: $[0, \; -3, \; -1, \; -3 \;\vert\; -18]$

$R_3 \leftarrow R_3 - 3R_1$: $[0, \; -1, \; -4, \; -1 \;\vert\; -17]$

$R_4 \leftarrow R_4 - R_1$: $[0, \; -3, \; 1, \; -2 \;\vert\; -8]$

Step 2 β€” Eliminate column 2 (pivot = $-3$ in $R_2$):

$R_3 \leftarrow R_3 - \frac{1}{3}R_2$: $\left[0, \; 0, \; -\frac{11}{3}, \; 0 \;\middle\vert\; -11\right]$

$R_4 \leftarrow R_4 - R_2$: $[0, \; 0, \; 2, \; 1 \;\vert\; 10]$

Step 3 β€” Eliminate column 3:

From $R_3$: $z = \dfrac{-11}{-11/3} = 3$.

From $R_4$: $2(3) + w = 10 \Rightarrow w = 4$.

Back substitution:

\[-3y - 3 - 12 = -18 \Rightarrow y = 1, \quad x + 1 + 3 + 4 = 10 \Rightarrow x = 2\]

Solution: $x = 2, \; y = 1, \; z = 3, \; w = 4$ βœ“

Q8 (Application β€” Circuit Analysis). Using Kirchhoff’s laws, currents $I_1, I_2, I_3$ satisfy:

\[\begin{cases} 5I_1 - 2I_2 - I_3 = 12 \\ -2I_1 + 6I_2 - 3I_3 = 4 \\ -I_1 - 3I_2 + 4I_3 = -5 \end{cases}\]

Solve for the currents. (Ans: $I_1 = 4$ A, $I_2 = 3$ A, $I_3 = 2$ A)

Click for solution

Augmented matrix:

\[\left[\begin{array}{ccc|c} 5 & -2 & -1 & 12 \\ -2 & 6 & -3 & 4 \\ -1 & -3 & 4 & -5 \end{array}\right]\]

Step 1 β€” Eliminate column 1 (pivot = 5):

$R_2 \leftarrow R_2 + \frac{2}{5}R_1$: $\left[0, \; \frac{26}{5}, \; -\frac{17}{5} \;\middle\vert\; \frac{44}{5}\right]$

$R_3 \leftarrow R_3 + \frac{1}{5}R_1$: $\left[0, \; -\frac{17}{5}, \; \frac{19}{5} \;\middle\vert\; -\frac{13}{5}\right]$

Step 2 β€” Eliminate column 2:

$R_3 \leftarrow R_3 + \frac{17}{26}R_2$: $\left[0, \; 0, \; \frac{41}{26} \;\middle\vert\; \frac{41}{13}\right]$

Back substitution:

\[I_3 = \frac{41/13}{41/26} = 2, \quad \frac{26}{5}I_2 = \frac{44}{5} + \frac{34}{5} = \frac{78}{5} \Rightarrow I_2 = 3\] \[5I_1 = 12 + 2(3) + 2 = 20 \Rightarrow I_1 = 4\]

Solution: $I_1 = 4$ A, $I_2 = 3$ A, $I_3 = 2$ A βœ“

Q9. For what value of $k$ does the following system have no unique solution?

\[\begin{cases} x + 2y + 3z = 4 \\ 2x + 4y + 6z = 8 \\ x + y + kz = 5 \end{cases}\]
Click for solution

Observation: $R_2 = 2R_1$, so $R_2$ is dependent. After $R_2 - 2R_1$, the system reduces to:

\[\left[\begin{array}{ccc|c} 1 & 2 & 3 & 4 \\ 0 & 0 & 0 & 0 \\ 0 & -1 & k-3 & 1 \end{array}\right]\]

Only 2 independent equations remain for 3 unknowns.

Coefficient determinant (for any $k$):

\[\det \begin{pmatrix} 1 & 2 & 3 \\ 2 & 4 & 6 \\ 1 & 1 & k \end{pmatrix} = 1(4k-6) - 2(2k-6) + 3(2-4) = 4k - 6 - 4k + 12 - 6 = 0\]

The determinant is zero for all values of $k$ β€” the system never has a unique solution.

Answer: No unique solution for any $k$. (Always infinitely many solutions since $R_2 = 2R_1$ ensures consistency.)

Q10. Express the solution in parametric form:

\[\begin{cases} x + 2y - z + w = 5 \\ 2x + 4y - 2z + 2w = 10 \\ x + y + z + w = 8 \end{cases}\]
Click for solution

Augmented matrix:

\[\left[\begin{array}{cccc|c} 1 & 2 & -1 & 1 & 5 \\ 2 & 4 & -2 & 2 & 10 \\ 1 & 1 & 1 & 1 & 8 \end{array}\right]\]

$R_2 = 2R_1$, so $R_2 \leftarrow R_2 - 2R_1 = [0\;0\;0\;0\;\vert\;0]$.

$R_3 \leftarrow R_3 - R_1$: $[0, \; -1, \; 2, \; 0 \;\vert\; 3]$ β†’ $y = 2z - 3$.

From $R_1$: $x = 5 - 2y + z - w = 5 - 2(2z - 3) + z - w = 11 - 3z - w$.

Set $z = s, \; w = t$ (free parameters):

\[\boxed{(x, y, z, w) = (11 - 3s - t, \; 2s - 3, \; s, \; t)}, \quad s, t \in \mathbb{R}\]

Example: $s = 2, t = 1 \Rightarrow (4, 1, 2, 1)$. Check: $4 + 2 - 2 + 1 = 5$ βœ“, $4 + 1 + 2 + 1 = 8$ βœ“.

Q11. Write the augmented matrix and reduce to REF:

\[\begin{cases} 2x_1 + x_2 - x_3 = 1 \\ -3x_1 - x_2 + 2x_3 = -1 \\ x_1 + x_2 + x_3 = 6 \end{cases}\]
Click for solution

Augmented matrix:

\[\left[\begin{array}{ccc|c} 2 & 1 & -1 & 1 \\ -3 & -1 & 2 & -1 \\ 1 & 1 & 1 & 6 \end{array}\right]\]

Step 1 β€” Swap $R_1 \leftrightarrow R_3$:

\[\left[\begin{array}{ccc|c} 1 & 1 & 1 & 6 \\ -3 & -1 & 2 & -1 \\ 2 & 1 & -1 & 1 \end{array}\right]\]

Step 2 β€” Eliminate column 1:

$R_2 \leftarrow R_2 + 3R_1$: $[0, \; 2, \; 5 \;\vert\; 17]$

$R_3 \leftarrow R_3 - 2R_1$: $[0, \; -1, \; -3 \;\vert\; -11]$

Step 3 β€” Eliminate column 2:

$R_3 \leftarrow 2R_3 + R_2$: $[0, \; 0, \; -1 \;\vert\; -5]$

REF:

\[\left[\begin{array}{ccc|c} 1 & 1 & 1 & 6 \\ 0 & 2 & 5 & 17 \\ 0 & 0 & -1 & -5 \end{array}\right]\]

Back substitution:

\[x_3 = 5, \quad 2x_2 + 25 = 17 \Rightarrow x_2 = -4, \quad x_1 - 4 + 5 = 6 \Rightarrow x_1 = 5\]

Solution: $x_1 = 5, \; x_2 = -4, \; x_3 = 5$

Q12 (Application β€” Force Equilibrium). Forces $F_1, F_2, F_3$ satisfy:

\[\begin{cases} 2F_1 + F_2 = 100 \\ F_1 + 2F_2 - F_3 = 50 \\ F_2 + 3F_3 = 75 \end{cases}\]

Find all forces.

Click for solution

Augmented matrix (swap $R_1 \leftrightarrow R_2$ for unit pivot):

\[\left[\begin{array}{ccc|c} 1 & 2 & -1 & 50 \\ 2 & 1 & 0 & 100 \\ 0 & 1 & 3 & 75 \end{array}\right]\]

$R_2 \leftarrow R_2 - 2R_1$: $[0, \; -3, \; 2 \;\vert\; 0]$

$R_3$: $[0, \; 1, \; 3 \;\vert\; 75]$

$R_3 \leftarrow 3R_3 + R_2$: $[0, \; 0, \; 11 \;\vert\; 225]$

Back substitution:

\[F_3 = \frac{225}{11}, \quad -3F_2 + \frac{450}{11} = 0 \Rightarrow F_2 = \frac{150}{11}, \quad F_1 = 50 - \frac{300}{11} + \frac{225}{11} = \frac{475}{11}\]

Solution: $F_1 = \dfrac{475}{11} \approx 43.2$ N, $F_2 = \dfrac{150}{11} \approx 13.6$ N, $F_3 = \dfrac{225}{11} \approx 20.5$ N


2 β€” Vector Spaces

Why study vector spaces? In school you learned about 2D and 3D vectors β€” arrows with magnitude and direction. But in engineering and science, β€œdata” often has far more than 3 components:

  • A colour pixel has 3 values (R, G, B) β€” that’s a vector in $\mathbb{R}^3$.
  • An audio sample at 44100 Hz for 1 second is a vector with 44100 components.
  • A neural network with 1 million weights lives in $\mathbb{R}^{1000000}$.

Vector spaces give us a single, unified framework to study all of these β€” regardless of dimension. Every theorem you prove in this chapter applies equally well to 2D arrows on paper and to million-dimensional machine-learning models.


2.1 Vectors in ℝⁿ

2.1.1 What Exactly Is a Vector?

In physics class you drew arrows on a plane. An arrow has a length (magnitude) and a direction. That’s a perfectly fine mental picture for 2D and 3D, but it breaks once we go to 4 or higher dimensions β€” we can’t draw those arrows.

A more powerful way to think about a vector is simply as an ordered list of numbers. The number of entries in the list is the dimension.

Definition: A vector in $\mathbb{R}^n$ is an ordered $n$-tuple of real numbers:

\[\vec{v} = (v_1, v_2, \ldots, v_n) = \begin{pmatrix} v_1 \\ v_2 \\ \vdots \\ v_n \end{pmatrix}\]

where each $v_i \in \mathbb{R}$.

  • The horizontal notation $(v_1, v_2, \ldots, v_n)$ is a row vector.
  • The vertical (column) notation is a column vector.
  • $\mathbb{R}^n$ means β€œthe set of all $n$-tuples of real numbers.”

2.1.2 Building Intuition β€” Dimension by Dimension

Space Vectors look like Geometric picture
$\mathbb{R}^1$ $(x)$ Points on the number line
$\mathbb{R}^2$ $(x, y)$ Points in the plane (familiar from graphs)
$\mathbb{R}^3$ $(x, y, z)$ Points in 3D space (the world around you)
$\mathbb{R}^4$ $(x_1, x_2, x_3, x_4)$ Can’t draw it, but the algebra is identical
$\mathbb{R}^n$ $(x_1, x_2, \ldots, x_n)$ General $n$-dimensional space

Key insight: Once you stop thinking of vectors as β€œarrows” and start thinking of them as β€œordered lists,” everything generalises instantly. All the rules (addition, scaling, length, angle) that work in 2D/3D carry over to $n$ dimensions through the exact same formulas.

Examples:

  • $\vec{v} = (3, 4) \in \mathbb{R}^2$ β€” a point 3 units along the $x$-axis, 4 along the $y$-axis. Also an arrow from the origin to that point.
  • $\vec{v} = (1, 2, 3) \in \mathbb{R}^3$ β€” a point or displacement in 3D space.
  • $\vec{v} = (1.5, -2.3, 0, 4.2, -1) \in \mathbb{R}^5$ β€” we can’t visualize this, but all operations work identically!

2.1.3 Two Special Vectors You Should Know

Zero vector: $\vec{0} = (0, 0, \ldots, 0)$. Adding it to any vector changes nothing: $\vec{v} + \vec{0} = \vec{v}$. Think of it as β€œno displacement.”

Standard basis vectors (unit vectors along each axis):

In $\mathbb{R}^3$:

\[\vec{e}_1 = (1, 0, 0), \quad \vec{e}_2 = (0, 1, 0), \quad \vec{e}_3 = (0, 0, 1)\]

Any vector $(a, b, c) = a\vec{e}_1 + b\vec{e}_2 + c\vec{e}_3$. The components simply tell you β€œhow much of each basis vector” to use.


2.2 Vector Operations β€” Addition & Scalar Multiplication

There are only two fundamental operations on vectors. Every concept in linear algebra β€” span, basis, dimension, linear transformations β€” is built from these two bricks.

2.2.1 Vector Addition β€” β€œComponent-wise add”

Vector Addition: For $\vec{u} = (u_1, \ldots, u_n)$ and $\vec{v} = (v_1, \ldots, v_n)$ in $\mathbb{R}^n$:

\[\vec{u} + \vec{v} = (u_1 + v_1, \; u_2 + v_2, \; \ldots, \; u_n + v_n)\]

You simply add matching components. Both vectors must be the same size β€” you can’t add a vector in $\mathbb{R}^2$ to one in $\mathbb{R}^3$.

Geometric picture (Parallelogram Law): Place the tail of $\vec{v}$ at the head of $\vec{u}$. The sum $\vec{u} + \vec{v}$ is the arrow from the tail of $\vec{u}$ to the head of $\vec{v}$. This creates a parallelogram β€” which is why addition is commutative ($\vec{u} + \vec{v} = \vec{v} + \vec{u}$).

2.2.2 Scalar Multiplication β€” β€œStretch or squash”

Scalar Multiplication: For a scalar $c \in \mathbb{R}$ and vector $\vec{v} = (v_1, \ldots, v_n)$:

\[c\vec{v} = (cv_1, \; cv_2, \; \ldots, \; cv_n)\]

Multiply every component by the same scalar.

What does the scalar do geometrically?

Scalar $c$ Effect on the vector
$c > 1$ Stretches (makes longer), same direction
$0 < c < 1$ Compresses (makes shorter), same direction
$c = 1$ No change
$c = 0$ Collapses to the zero vector $\vec{0}$
$c = -1$ Same length, opposite direction
$c < 0$ Reverses direction and scales by $\lvert c \rvert$

2.2.3 Worked Examples

Example 1: Let $\vec{u} = (1,2,3)$ and $\vec{v} = (4,5,6)$.

  • Addition: $\vec{u} + \vec{v} = (1+4, \; 2+5, \; 3+6) = (5, 7, 9)$
  • Scalar multiplication: $2\vec{u} = (2 \cdot 1, \; 2 \cdot 2, \; 2 \cdot 3) = (2, 4, 6)$
  • Subtraction: $\vec{u} - \vec{v} = \vec{u} + (-1)\vec{v} = (1-4, \; 2-5, \; 3-6) = (-3, -3, -3)$
  • Linear combination: $3\vec{u} + 2\vec{v} = (3,6,9) + (8,10,12) = (11, 16, 21)$

Example 2 (Real-world): You walk 3 blocks east and 4 blocks north: $\vec{d}_1 = (3, 4)$. Then 1 block west and 2 blocks north: $\vec{d}_2 = (-1, 2)$.

Total displacement: $\vec{d}_1 + \vec{d}_2 = (3 + (-1), \; 4 + 2) = (2, 6)$ β€” 2 blocks east, 6 blocks north.


2.3 Linear Combinations

This is arguably the most important idea in all of linear algebra. If you understand linear combinations, every subsequent concept (span, basis, dimension, linear transformations) becomes much easier.

2.3.1 The Core Idea

Imagine you’re at a paint store and you have three base colours: Red, Blue, and Yellow. Any colour you mix is some amount of Red + some amount of Blue + some amount of Yellow. That mixture is a β€œlinear combination” of the three base colours.

Similarly, in linear algebra, a linear combination of vectors means β€œtake some amount of vector 1, plus some amount of vector 2, plus … and add them all up.”

Definition: A linear combination of vectors $\vec{v}_1, \vec{v}_2, \ldots, \vec{v}_k$ is any expression of the form:

\[c_1\vec{v}_1 + c_2\vec{v}_2 + \cdots + c_k\vec{v}_k\]

where $c_1, c_2, \ldots, c_k$ are scalars (real numbers). The $c_i$ are called weights or coefficients.

Why it matters:

  • The span of a set of vectors = the set of all their linear combinations.
  • Solving $A\vec{x} = \vec{b}$ is the same as asking β€œIs $\vec{b}$ a linear combination of the columns of $A$?”
  • A basis is a minimal set of vectors whose linear combinations produce the entire space.

2.3.2 How to Check If a Vector Is a Linear Combination

Given: Vectors $\vec{v}_1, \vec{v}_2, \ldots, \vec{v}_k$ and a target vector $\vec{w}$.

Question: Is $\vec{w}$ a linear combination of the $\vec{v}_i$?

Method:

  1. Write $c_1\vec{v}_1 + c_2\vec{v}_2 + \cdots + c_k\vec{v}_k = \vec{w}$
  2. This gives a system of linear equations in the unknowns $c_1, c_2, \ldots, c_k$
  3. Solve the system (using Gauss elimination from Section 1!)
  4. If a solution exists β†’ yes, $\vec{w}$ is a linear combination. If no solution β†’ no.

Example 1: Is $\vec{w} = (5, 8)$ a linear combination of $\vec{v}_1 = (1, 2)$ and $\vec{v}_2 = (2, 3)$?

We need scalars $c_1, c_2$ such that: $c_1(1,2) + c_2(2,3) = (5,8)$

Writing component-wise:

\[\begin{cases} c_1 \cdot 1 + c_2 \cdot 2 = 5 \\ c_1 \cdot 2 + c_2 \cdot 3 = 8 \end{cases} \quad\Rightarrow\quad \begin{cases} c_1 + 2c_2 = 5 \\ 2c_1 + 3c_2 = 8 \end{cases}\]

From the first equation: $c_1 = 5 - 2c_2$. Substituting into the second:

\[2(5 - 2c_2) + 3c_2 = 8 \;\Rightarrow\; 10 - 4c_2 + 3c_2 = 8 \;\Rightarrow\; c_2 = 2, \; c_1 = 1\]

Verification: $1 \cdot (1,2) + 2 \cdot (2,3) = (1,2) + (4,6) = (5,8)$ βœ“

So $\vec{w} = \vec{v}_1 + 2\vec{v}_2$. βœ“

Example 2: Is $\vec{w} = (1, 0, 0)$ a linear combination of $\vec{v}_1 = (1, 1, 0)$ and $\vec{v}_2 = (0, 1, 1)$?

We need $c_1(1,1,0) + c_2(0,1,1) = (1, 0, 0)$:

\[\begin{cases} c_1 = 1 \\ c_1 + c_2 = 0 \\ c_2 = 0 \end{cases}\]

From Eq.1: $c_1 = 1$. From Eq.3: $c_2 = 0$. But Eq.2: $1 + 0 = 1 \neq 0$. Contradiction!

No β€” $\vec{w}$ is NOT a linear combination of $\vec{v}_1$ and $\vec{v}_2$. The system is inconsistent.


2.4 Fields & Scalars

Before we define a β€œvector space” formally, we need to answer: what kind of numbers are we allowed to scale by?

When we write $c\vec{v}$, what set does $c$ come from? Real numbers? Complex numbers? Integers? Not all sets of numbers behave the same way, and the answer matters.

2.4.1 What Is a Field?

A field is a set of numbers where you can add, subtract, multiply, and divide (except by zero) and all the usual rules of arithmetic work. Think of it as a β€œwell-behaved number system.”

Definition: A field $F$ is a set with two operations ($+$ and $\cdot$) satisfying:

  • Associativity: $(a+b)+c = a+(b+c)$ and $(ab)c = a(bc)$
  • Commutativity: $a+b = b+a$ and $ab = ba$
  • Distributivity: $a(b+c) = ab + ac$
  • Identity elements: $0, 1 \in F$ with $a + 0 = a$ and $1 \cdot a = a$
  • Inverses: Every $a$ has an additive inverse $-a$; every $a \neq 0$ has a multiplicative inverse $a^{-1}$

2.4.2 Examples of Fields (and Non-Fields)

Set Field? Why?
$\mathbb{R}$ (real numbers) βœ“ Yes Can add, subtract, multiply, divide freely
$\mathbb{C}$ (complex numbers) βœ“ Yes Same properties; $(a+bi)^{-1}$ always exists
$\mathbb{Q}$ (rational numbers) βœ“ Yes Closed under $+, -, \times, \div$
$\mathbb{Z}$ (integers) βœ— No $\frac{1}{2} \notin \mathbb{Z}$ β€” no multiplicative inverse for $2$
$\mathbb{N}$ (natural numbers) βœ— No No additive inverses ($-3 \notin \mathbb{N}$), no division
$\mathbb{Z}_p$ (integers mod prime $p$) βœ“ Yes Used in cryptography and coding theory

Notation Primer β€” Symbols used below:

Symbol Name Meaning
$\mathbb{Z}$ β€œBlackboard bold Z” The set of all integers $\lbrace\dots, -2, -1, 0, 1, 2, \dots\rbrace$
$\mathbb{Z}_p$ Integers mod $p$ The set $\lbrace0, 1, 2, \dots, p-1\rbrace$ where all arithmetic wraps around at $p$
$\equiv$ β€œCongruent to” $a \equiv b \pmod{p}$ means $p$ divides $(a - b)$, i.e. $a$ and $b$ have the same remainder when divided by $p$
$\pmod{p}$ β€œModulo $p$” Specifies we are working with remainders after dividing by $p$
$\gcd(a, b)$ Greatest common divisor The largest positive integer that divides both $a$ and $b$
$a^{-1}$ Multiplicative inverse The element satisfying $a \cdot a^{-1} \equiv 1 \pmod{p}$ β€” the β€œreciprocal” in modular arithmetic

Quick example of $\equiv$: We write $17 \equiv 2 \pmod{5}$ because $17 - 2 = 15$ is divisible by $5$. Equivalently, both $17$ and $2$ leave remainder $2$ when divided by $5$.

BΓ©zout's Identity (click to expand)

For any integers $a$ and $b$ (not both zero), there exist integers $x$ and $y$ such that:

\[ax + by = \gcd(a, b)\]

These $x, y$ are called BΓ©zout coefficients and can be found using the Extended Euclidean Algorithm.

Why it matters here: If $p$ is prime and $0 < a < p$, then $\gcd(a, p) = 1$ (since $p$ has no divisors other than $1$ and itself). BΓ©zout’s identity guarantees integers $x, y$ with:

\[ax + py = 1\]

Taking this equation modulo $p$: since $py \equiv 0 \pmod{p}$, we get $ax \equiv 1 \pmod{p}$, so $x$ is the multiplicative inverse of $a$ in $\mathbb{Z}_p$.

Worked example β€” finding $3^{-1}$ in $\mathbb{Z}_7$:

Apply the Extended Euclidean Algorithm to $a = 3$, $b = 7$:

\(7 = 2 \times 3 + 1\) \(\Rightarrow 1 = 7 - 2 \times 3\) \(\Rightarrow 1 = (-2) \times 3 + (1) \times 7\)

So $x = -2$. Taking $\pmod{7}$: $x \equiv -2 \equiv 5 \pmod{7}$.

Check: $3 \times 5 = 15 = 2 \times 7 + 1 \equiv 1 \pmod{7}$. βœ“

So $3^{-1} \equiv 5$ in $\mathbb{Z}_7$.

Insight β€” Why $\mathbb{Z}_p$ is a field (and $\mathbb{Z}_n$ for composite $n$ is not):

$\mathbb{Z}_p = \lbrace0, 1, 2, \dots, p-1\rbrace$ with addition and multiplication done modulo $p$.

The key property: When $p$ is prime, every nonzero element $a$ has a multiplicative inverse $a^{-1}$ such that $a \cdot a^{-1} \equiv 1 \pmod{p}$. This follows from BΓ©zout’s identity β€” since $\gcd(a, p) = 1$, there exist integers $x, y$ with $ax + py = 1$, so $ax \equiv 1 \pmod{p}$ and $x$ is the inverse.

Example with $\mathbb{Z}_5 = \lbrace0, 1, 2, 3, 4\rbrace$:

$a$ $a^{-1}$ Check
$1$ $1$ $1 \times 1 = 1 \equiv 1$
$2$ $3$ $2 \times 3 = 6 \equiv 1$
$3$ $2$ $3 \times 2 = 6 \equiv 1$
$4$ $4$ $4 \times 4 = 16 \equiv 1$

Every nonzero element has an inverse β€” so $\mathbb{Z}_5$ is a field. βœ“

Why composite moduli fail: In $\mathbb{Z}_6$, the element $2$ has no inverse because $\gcd(2, 6) = 2 \neq 1$. No integer $x$ satisfies $2x \equiv 1 \pmod{6}$. Even worse, $2 \times 3 = 6 \equiv 0 \pmod{6}$ β€” two nonzero elements multiply to zero (called zero divisors). Fields cannot have zero divisors, so $\mathbb{Z}_6$ is not a field. βœ—

Why this matters beyond math:

  • RSA encryption relies on arithmetic in $\mathbb{Z}_n$ where $n = pq$ (product of two primes) β€” the difficulty of factoring $n$ provides security.
  • Elliptic curve cryptography works over $\mathbb{Z}_p$ because field properties guarantee that points on the curve form a well-behaved group.
  • Error-correcting codes (Reed–Solomon) use $\mathbb{Z}_p$ or its extensions to detect and correct transmission errors.

For this course: We almost always work over $\mathbb{R}$ (real numbers). The β€œscalars” in our vector spaces will be real numbers. When you see β€œa vector space $V$” without further qualification, assume scalars are from $\mathbb{R}$.

Why integers don’t work: If $V$ were a β€œvector space over $\mathbb{Z}$”, we couldn’t divide a vector by 2. That breaks many theorems (e.g., the midpoint $\frac{1}{2}(\vec{u} + \vec{v})$ wouldn’t exist). Fields guarantee all the arithmetic we need.


2.5 Formal Definition of a Vector Space

2.5.1 The Big Picture First

We’ve seen that $\mathbb{R}^n$ has two operations (addition and scalar multiplication) with nice properties. But $\mathbb{R}^n$ isn’t the only set with those properties β€” polynomials, matrices, and even functions share them!

A vector space is the abstract name for any set where these two operations work β€œnicely.” By proving a theorem for general vector spaces, we instantly prove it for $\mathbb{R}^n$, polynomial spaces, matrix spaces, function spaces, and any other example that fits the definition.

Definition: A vector space $V$ over a field $F$ consists of:

  1. A set $V$ (whose elements are called vectors)
  2. A field $F$ (whose elements are called scalars β€” usually $\mathbb{R}$)
  3. Two operations:
    • Vector addition: $\vec{u}, \vec{v} \in V \Rightarrow \vec{u} + \vec{v} \in V$
    • Scalar multiplication: $c \in F, \vec{v} \in V \Rightarrow c\vec{v} \in V$
  4. These operations satisfy the 10 axioms (see next section).

Intuition β€” The β€œClosed Playground” Analogy:

A vector space is a closed playground with two rules:

  1. You can add any two things inside β†’ the result stays inside.
  2. You can scale any thing inside by a number β†’ the result stays inside.

You can never β€œfall out” of the playground by adding or scaling. This property is called closure, and it’s the most important property to check when determining if something is a vector space.

2.5.2 What Can Be a β€œVector”?

The word β€œvector” in linear algebra is much broader than β€œarrow in 2D/3D.” Anything that can be added together and scaled β€” while satisfying the 10 axioms β€” qualifies:

What are the β€œvectors”? What are the β€œscalars”? Vector space?
Arrows in 2D/3D Real numbers βœ“ $\mathbb{R}^2$, $\mathbb{R}^3$
Lists of $n$ numbers Real numbers βœ“ $\mathbb{R}^n$
Polynomials of degree $\leq n$ Real numbers βœ“ $P_n$
$m \times n$ matrices Real numbers βœ“ $M_{m \times n}$
Continuous functions $\mathbb{R} \to \mathbb{R}$ Real numbers βœ“ $C(\mathbb{R})$
Audio signals (sequences of samples) Real numbers βœ“

2.6 The Ten Vector Space Axioms

2.6.1 Why Do We Need Axioms?

Not every set with β€œaddition” and β€œscaling” is a vector space. For example, if you define a weird β€œaddition” on $\mathbb{R}^2$ as $\vec{u} \oplus \vec{v} = (u_1 v_1, \; u_2 v_2)$ (component-wise multiplication), it looks like addition but breaks many expected properties. Axioms are the checklist that guarantees the operations behave β€œnormally.”

2.6.2 The Complete List

The 10 Axioms: A set $V$ with operations of addition and scalar multiplication over field $F$ is a vector space if and only if the following hold for all $\vec{u}, \vec{v}, \vec{w} \in V$ and all $c, d \in F$:

Closure (can’t escape the set):

  1. Closure under Addition: $\vec{u} + \vec{v} \in V$
  2. Closure under Scalar Multiplication: $c\vec{v} \in V$

Addition behaves normally:

  1. Commutativity: $\vec{u} + \vec{v} = \vec{v} + \vec{u}$
  2. Associativity: $(\vec{u} + \vec{v}) + \vec{w} = \vec{u} + (\vec{v} + \vec{w})$
  3. Additive Identity (zero vector exists): There exists $\vec{0} \in V$ such that $\vec{v} + \vec{0} = \vec{v}$
  4. Additive Inverse (negatives exist): For each $\vec{v} \in V$, there exists $-\vec{v} \in V$ such that $\vec{v} + (-\vec{v}) = \vec{0}$

Scalar multiplication behaves normally:

  1. Distributivity over vector addition: $c(\vec{u} + \vec{v}) = c\vec{u} + c\vec{v}$
  2. Distributivity over scalar addition: $(c + d)\vec{v} = c\vec{v} + d\vec{v}$
  3. Compatibility (associativity): $c(d\vec{v}) = (cd)\vec{v}$
  4. Scalar Identity: $1 \cdot \vec{v} = \vec{v}$

2.6.3 How to Remember Them

Memory aid β€” group them by theme:

Theme Axioms Plain English
Closure (2) A1, A2 β€œAdding or scaling keeps you inside”
Addition rules (4) A3–A6 β€œAddition works like you’d expect: commutative, associative, has a zero, has negatives”
Scaling rules (4) A7–A10 β€œScaling distributes, associates, and 1 is the identity”

In exams, the most commonly tested axioms are:

  • Closure (A1, A2) β€” this is where most β€œIs this a vector space?” questions fail
  • Zero vector (A5) β€” quick disqualifier: if $\vec{0}$ isn’t in the set, stop immediately

Common exam mistake: Students forget to check closure axioms. Always verify that addition and scalar multiplication don’t take you outside the set!

Another pitfall: Don’t confuse β€œthe zero vector exists in $V$” with β€œzero is a scalar.” They are different objects: $\vec{0} \in V$ is a vector, while $0 \in F$ is a scalar.

2.6.4 What Happens When an Axiom Fails?

To show something is NOT a vector space, you only need one counterexample violating one axiom.

Quick example: Is $S = \lbrace(x, y) \in \mathbb{R}^2 : x \geq 0, y \geq 0\rbrace$ (the first quadrant) a vector space?

  • Check closure under scalar multiplication: Take $\vec{v} = (1, 2) \in S$ and $c = -1$. Then $c\vec{v} = (-1, -2) \notin S$ (negative components!).
  • Fails axiom 2 β†’ NOT a vector space. Done!

We didn’t need to check the other 9 axioms. One failure is enough.


2.7 Examples of Vector Spaces

The best way to truly understand the definition is to see it applied to many different examples. We’ll verify the axioms carefully for the first example, then focus on the key axioms (closure and zero vector) for the rest.

Example 1: $\mathbb{R}^n$ β€” The Fundamental Example

This is the space you’ll use 90% of the time in this course.

Claim: $\mathbb{R}^3$ is a vector space over $\mathbb{R}$.

What is the set? All triples of real numbers: $(a, b, c)$ where $a, b, c \in \mathbb{R}$.

What are the operations?

  • Addition: $(a_1, b_1, c_1) + (a_2, b_2, c_2) = (a_1+a_2, \; b_1+b_2, \; c_1+c_2)$
  • Scalar multiplication: $k(a, b, c) = (ka, kb, kc)$

Verification of all 10 axioms:

  • Axiom 1 (Closure +): If $\vec{u}, \vec{v} \in \mathbb{R}^3$, then $\vec{u} + \vec{v} = (u_1+v_1, u_2+v_2, u_3+v_3)$. Since the sum of two real numbers is real, every component is in $\mathbb{R}$, so the result is in $\mathbb{R}^3$ βœ“
  • Axiom 2 (Closure Β·): $c\vec{v} = (cv_1, cv_2, cv_3)$. Product of reals is real, so the result is in $\mathbb{R}^3$ βœ“
  • Axiom 3 (Commutativity): $(u_1+v_1, u_2+v_2, u_3+v_3) = (v_1+u_1, v_2+u_2, v_3+u_3)$ because real number addition is commutative βœ“
  • Axiom 4 (Associativity): follows from associativity of real addition βœ“
  • Axiom 5 (Zero vector): $\vec{0} = (0, 0, 0)$ works because $(a,b,c) + (0,0,0) = (a,b,c)$ βœ“
  • Axiom 6 (Inverse): $-\vec{v} = (-v_1, -v_2, -v_3)$, and $\vec{v} + (-\vec{v}) = (0,0,0) = \vec{0}$ βœ“
  • Axioms 7–10: Follow from properties of real number arithmetic (distribution, associativity of multiplication, $1 \cdot a = a$). βœ“

Conclusion: All 10 axioms hold. $\mathbb{R}^3$ is a vector space. βœ“

General principle: $\mathbb{R}^n$ is a vector space for any positive integer $n$. The proof is identical β€” it always reduces to properties of real number arithmetic.

Example 2: Polynomial Space $P_n$ β€” β€œVectors” That Are Polynomials

Here’s where it gets interesting β€” the β€œvectors” are no longer lists of numbers!

Claim: $P_n = \lbrace a_0 + a_1x + a_2x^2 + \cdots + a_nx^n : a_i \in \mathbb{R}\rbrace$ is a vector space over $\mathbb{R}$.

What is the set? All polynomials with real coefficients and degree at most $n$.

What are the operations?

  • Addition: add polynomials in the usual way (add corresponding coefficients)
  • Scalar multiplication: multiply every coefficient by the scalar

Example with $P_2$: Let $p(x) = 3 + 2x - x^2$ and $q(x) = 1 - 4x + 5x^2$.

  • $p + q = 4 - 2x + 4x^2$ (degree $\leq 2$ βœ“, still in $P_2$)
  • $2p = 6 + 4x - 2x^2$ (degree $\leq 2$ βœ“, still in $P_2$)

Key axioms:

  • Closure +: Sum of two polynomials of degree $\leq n$ has degree $\leq n$ βœ“
  • Closure Β·: Scalar times a polynomial of degree $\leq n$ has degree $\leq n$ βœ“
  • Zero vector: The zero polynomial $0 + 0x + \cdots + 0x^n = 0$ βœ“
  • Inverse: $-(a_0 + a_1 x + \cdots) = (-a_0) + (-a_1)x + \cdots$ βœ“
  • Remaining axioms follow from polynomial arithmetic βœ“

Caution: $P_n$ is polynomials of degree at most $n$, not exactly $n$. Why? Because if $p(x) = 1 + x^2$ and $q(x) = 1 - x^2$, then $p + q = 2$ (degree 0, not 2). If we required degree exactly 2, we’d lose closure under addition!

Example 3: Matrix Space $M_{m \times n}(\mathbb{R})$ β€” β€œVectors” That Are Matrices

Claim: The set of all $2 \times 2$ real matrices is a vector space.

What is the set? All matrices of the form:

\[\begin{bmatrix} a & b \\ c & d \end{bmatrix} \quad \text{where } a, b, c, d \in \mathbb{R}\]

Operations: The usual matrix addition and scalar multiplication:

  • Closure +:
\[\begin{bmatrix} a_1 & b_1 \\ c_1 & d_1 \end{bmatrix} + \begin{bmatrix} a_2 & b_2 \\ c_2 & d_2 \end{bmatrix} = \begin{bmatrix} a_1+a_2 & b_1+b_2 \\ c_1+c_2 & d_1+d_2 \end{bmatrix} \in M_{2 \times 2}\]

βœ“ (sum of reals is real, so each entry is real)

  • Zero vector:
\[\begin{bmatrix} 0 & 0 \\ 0 & 0 \end{bmatrix}\]

βœ“ (the zero matrix plays the role of $\vec{0}$)

Connection to $\mathbb{R}^n$: A $2 \times 2$ matrix has 4 entries, so $M_{2 \times 2}(\mathbb{R})$ is essentially the same as $\mathbb{R}^4$ β€” just with the 4 numbers arranged in a square instead of a row. More generally, $M_{m \times n}$ behaves like $\mathbb{R}^{mn}$.

Example 4: Function Space $C(\mathbb{R})$ β€” β€œVectors” That Are Functions

This is the most abstract example, but it shows the true power of the vector space concept.

Claim: $C(\mathbb{R}) = \lbrace f : \mathbb{R} \to \mathbb{R} \mid f \text{ is continuous}\rbrace$ is a vector space.

What is the set? All continuous functions from $\mathbb{R}$ to $\mathbb{R}$ (e.g., $\sin x$, $e^x$, $x^3 + 1$, …).

Operations:

  • Addition: $(f + g)(x) = f(x) + g(x)$ β€” add the function values at each point
  • Scalar multiplication: $(cf)(x) = c \cdot f(x)$ β€” scale every output by $c$

Key axioms:

  • Closure +: Sum of continuous functions is continuous (from calculus!) βœ“
  • Closure Β·: Scalar multiple of a continuous function is continuous βœ“
  • Zero vector: $f(x) = 0$ for all $x$ (the function that is always zero) βœ“

Note: This is an infinite-dimensional vector space! Unlike $\mathbb{R}^n$, you can’t describe every continuous function with finitely many numbers.

In $\mathbb{R}^3$, any vector is fully described by 3 numbers: $(a, b, c)$. But a continuous function like $\sin x$ has an output for every real number β€” infinitely many β€œcoordinates.” No finite list of basis functions can span all of $C(\mathbb{R})$. That’s what β€œinfinite-dimensional” means: you’d need infinitely many basis vectors to express every element.

Analogy: Think of $\mathbb{R}^n$ as a box with $n$ dials β€” turn each dial to build any vector. For $C(\mathbb{R})$, you’d need a box with infinitely many dials β€” one for each β€œdirection” in function space.

Example 5 (Non-Example): Sets That Are NOT Vector Spaces

Not a vector space: The set of all polynomials of degree exactly 2.

Why? Let $p(x) = x^2$ and $q(x) = -x^2 + 1$. Then $p + q = 1$ (degree 0, not 2). Fails closure under addition.

Not a vector space: $\mathbb{Z}^n$ (vectors with integer components) over $\mathbb{R}$.

Why? $\frac{1}{2} \cdot (1, 0) = (0.5, 0)$ has non-integer component. Fails closure under scalar multiplication.

Not a vector space: $\lbrace(x, y) : x + y = 1\rbrace$ (a line not through origin).

Why? $\vec{0} = (0,0)$ doesn’t satisfy $0 + 0 = 1$. Fails zero vector axiom.

Not a vector space: The set of all invertible matrices of order $n \times n$.

Why? The zero matrix $O$ is not invertible (it has no inverse), so the set has no additive identity. Fails the zero vector axiom.


2.8 Important Properties of Vector Spaces

The following properties hold in every vector space $V$ over a field $F$:

1. For any scalar $\alpha \in F$ and the zero vector $\vec{0} \in V$:

\[\alpha \cdot \vec{0} = \vec{0}\]

Proof: $\alpha \cdot \vec{0} = \alpha \cdot (\vec{0} + \vec{0}) = \alpha \vec{0} + \alpha \vec{0}$. Subtracting $\alpha \vec{0}$ from both sides gives $\vec{0} = \alpha \vec{0}$.

2. For the zero scalar $0 \in F$ and any vector $\vec{u} \in V$:

\[0 \cdot \vec{u} = \vec{0}\]

Proof: $0 \cdot \vec{u} = (0 + 0)\vec{u} = 0\vec{u} + 0\vec{u}$. Subtracting $0\vec{u}$ from both sides gives $\vec{0} = 0\vec{u}$.

3. If $\alpha \vec{u} = \vec{0}$, then either $\alpha = 0$ or $\vec{u} = \vec{0}$ (or both).

Proof: If $\alpha \neq 0$, then $\alpha^{-1}$ exists. Multiplying both sides: $\vec{u} = \alpha^{-1}(\alpha \vec{u}) = \alpha^{-1}\vec{0} = \vec{0}$.


Practice Questions β€” Vector Spaces

Q1. Show that $\mathbb{R}^2$ is a vector space over $\mathbb{R}$ by verifying all 10 axioms.

Click for solution

Let $\vec{u} = (u_1, u_2), \vec{v} = (v_1, v_2), \vec{w} = (w_1, w_2) \in \mathbb{R}^2$ and $c, d \in \mathbb{R}$.

# Axiom Verification
1 Closure (+) $(u_1+v_1, u_2+v_2) \in \mathbb{R}^2$ βœ“
2 Commutativity $\vec{u} + \vec{v} = (u_1+v_1, u_2+v_2) = \vec{v} + \vec{u}$ βœ“
3 Associativity (+) $(\vec{u}+\vec{v})+\vec{w} = \vec{u}+(\vec{v}+\vec{w})$ βœ“
4 Zero vector $\vec{0} = (0,0)$; $\vec{u} + \vec{0} = \vec{u}$ βœ“
5 Additive inverse $-\vec{u} = (-u_1, -u_2)$; $\vec{u} + (-\vec{u}) = \vec{0}$ βœ“
6 Closure (Β·) $c\vec{u} = (cu_1, cu_2) \in \mathbb{R}^2$ βœ“
7 Distributive (vector) $c(\vec{u}+\vec{v}) = c\vec{u} + c\vec{v}$ βœ“
8 Distributive (scalar) $(c+d)\vec{u} = c\vec{u} + d\vec{u}$ βœ“
9 Associativity (Β·) $c(d\vec{u}) = (cd)\vec{u}$ βœ“
10 Identity (Β·) $1 \cdot \vec{u} = \vec{u}$ βœ“

All 10 axioms hold, so $\mathbb{R}^2$ is a vector space over $\mathbb{R}$.

Q2. Is the set $\lbrace(x, y) : x + y = 1\rbrace$ a vector space under standard operations? Why or why not? (Hint: Check if $(0,0)$ is in the set.)

Click for solution

No. The zero vector $(0,0)$ is NOT in the set since $0 + 0 = 0 \neq 1$. Any vector space must contain $\vec{0}$, so this set fails immediately.

Also fails closure: $(0.5, 0.5) + (0.5, 0.5) = (1, 1)$, but $1 + 1 = 2 \neq 1$.

Q3. Prove that $P_2$ (polynomials of degree $\leq 2$) is a vector space.

Click for solution

$P_2 = \lbrace a_0 + a_1x + a_2x^2 : a_0, a_1, a_2 \in \mathbb{R}\rbrace$ with standard polynomial addition and scalar multiplication.

Key axioms:

  • Zero: $0 + 0x + 0x^2 = 0 \in P_2$ βœ“
  • Closure (+): $(a_0+b_0) + (a_1+b_1)x + (a_2+b_2)x^2 \in P_2$ (degree $\leq 2$) βœ“
  • Closure (Β·): $ca_0 + ca_1 x + ca_2 x^2 \in P_2$ βœ“
  • Inverse: $-(a_0 + a_1x + a_2x^2) = (-a_0) + (-a_1)x + (-a_2)x^2 \in P_2$ βœ“
  • Remaining axioms inherited from properties of real number arithmetic βœ“

Q4. Is the set of all $2 \times 2$ matrices with determinant $0$ a vector space? Prove or disprove. (Hint: Check closure under addition.)

Click for solution

No β€” fails closure under addition.

\[A = \begin{pmatrix} 1 & 0 \\ 0 & 0 \end{pmatrix}, \quad B = \begin{pmatrix} 0 & 0 \\ 0 & 1 \end{pmatrix}\]

Both have $\det(A) = 0$ and $\det(B) = 0$, but:

\[A + B = \begin{pmatrix} 1 & 0 \\ 0 & 1 \end{pmatrix} = I, \quad \det(A+B) = 1 \neq 0\]

So $A + B$ is NOT in the set. Not a vector space.

Q5. Determine if $\vec{w} = (7, 11)$ is a linear combination of $\vec{v}_1 = (1, 3)$ and $\vec{v}_2 = (2, 1)$.

Click for solution

Solve $c_1(1, 3) + c_2(2, 1) = (7, 11)$:

\[\begin{cases} c_1 + 2c_2 = 7 \\ 3c_1 + c_2 = 11 \end{cases}\]

From Eq1: $c_1 = 7 - 2c_2$. Substitute into Eq2: $3(7 - 2c_2) + c_2 = 11 \Rightarrow 21 - 5c_2 = 11 \Rightarrow c_2 = 2$.

$c_1 = 7 - 4 = 3$.

Yes: $(7, 11) = 3(1, 3) + 2(2, 1)$. Verify: $(3+4, 9+2) = (7, 11)$ βœ“

Q6. Show that $\mathbb{C}^n$ (n-tuples of complex numbers) is a vector space over $\mathbb{C}$.

Click for solution

The proof follows the same structure as for $\mathbb{R}^n$, since $\mathbb{C}$ is a field. All 10 axioms hold because complex addition and multiplication satisfy the same algebraic properties (commutativity, associativity, distributivity, identity elements) as real numbers.

  • Zero vector: $(0, 0, \ldots, 0) \in \mathbb{C}^n$ βœ“
  • Closure: sum or scalar multiple of complex $n$-tuples remains a complex $n$-tuple βœ“
  • All other axioms follow from $\mathbb{C}$ being a field βœ“

Q7. Is $\lbrace(x, y, z) \in \mathbb{R}^3 : x^2 + y^2 + z^2 \leq 1\rbrace$ (the unit ball) a vector space? Why or why not?

Click for solution

No β€” fails closure under scalar multiplication.

$(1, 0, 0)$ is in the set (since $1^2 + 0^2 + 0^2 = 1 \leq 1$), but $2(1, 0, 0) = (2, 0, 0)$ with $4 + 0 + 0 = 4 > 1$, so $(2, 0, 0) \notin$ unit ball.

Also fails closure under addition: $(0.8, 0.6, 0)$ and $(0.8, -0.6, 0)$ are in the ball, but their sum $(1.6, 0, 0)$ is not.

Q8. Write $(5, 3)$ as a linear combination of $(1, 0)$ and $(0, 1)$.

Click for solution

$(5, 3) = 5(1, 0) + 3(0, 1)$. These are just the standard basis vectors $\vec{e}_1, \vec{e}_2$, so the coefficients are simply the components of the vector.


3 β€” Subspaces & Spanning Sets


3.1 What is a Subspace?

Definition: A subset $W$ of a vector space $V$ is a subspace of $V$ if $W$ is itself a vector space under the same operations as $V$.

Geometric picture:

  • In $\mathbb{R}^2$: subspaces are $\lbrace\vec{0}\rbrace$, lines through the origin, or all of $\mathbb{R}^2$
  • In $\mathbb{R}^3$: subspaces are $\lbrace\vec{0}\rbrace$, lines through the origin, planes through the origin, or all of $\mathbb{R}^3$

Critical: ALL subspaces must contain the origin $\vec{0}$. If a subset doesn’t contain $\vec{0}$, it is NOT a subspace β€” stop checking immediately!


3.2 The Subspace Test (3-Condition Shortcut)

Subspace Test Theorem: Let $V$ be a vector space and $W \subseteq V$. Then $W$ is a subspace if and only if:

  1. Contains Zero Vector: $\vec{0} \in W$
  2. Closed under Addition: $\vec{u}, \vec{v} \in W \Rightarrow \vec{u} + \vec{v} \in W$
  3. Closed under Scalar Multiplication: $\vec{v} \in W, c \in F \Rightarrow c\vec{v} \in W$

Why only 3 conditions? The other 7 axioms (commutativity, associativity, etc.) are inherited from the parent space $V$. We only need to check that $W$ is non-empty, closed, and contains $\vec{0}$.

Even shorter β€” single condition: Conditions 2 and 3 can be combined: If $\vec{u}, \vec{v} \in W$ and $c, d \in F$, then $c\vec{u} + d\vec{v} \in W$.


3.3 Examples β€” Subspace & Non-Subspace

Example 1: $W = \lbrace(x, y) : y = 2x\rbrace$ β€” Line through origin βœ“

Claim: $W$ is a subspace of $\mathbb{R}^2$.

  1. Zero vector: $(0, 0)$? Check: $0 = 2(0)$? Yes. βœ“
  2. Closed under +: If $(x_1, 2x_1), (x_2, 2x_2) \in W$, then sum = $(x_1+x_2, 2x_1+2x_2) = (x_1+x_2, 2(x_1+x_2)) \in W$. βœ“
  3. Closed under Β·: $c(x, 2x) = (cx, 2cx) \in W$. βœ“

Conclusion: YES, $W$ is a subspace. Geometrically, it’s the line $y = 2x$ through the origin.

Example 2: $W = \lbrace(x, y) : y = 2x + 1\rbrace$ β€” Line NOT through origin βœ—

Claim: $W$ is NOT a subspace.

  1. Zero vector: $(0, 0) \in W$? Check: $0 = 2(0) + 1 = 1$? NO!

Conclusion: NOT a subspace β€” fails immediately at condition 1. The line $y = 2x + 1$ doesn’t pass through the origin.

Example 3: $W = \lbrace(x, y, z) : x + y - z = 0\rbrace$ β€” Plane through origin βœ“

  1. Zero: $0 + 0 - 0 = 0$ βœ“
  2. Closed +: $(x_1+x_2) + (y_1+y_2) - (z_1+z_2) = (x_1+y_1-z_1) + (x_2+y_2-z_2) = 0 + 0 = 0$ βœ“
  3. Closed Β·: $c(x+y-z) = c \cdot 0 = 0$ βœ“

Conclusion: Subspace β€” it’s the plane $x + y - z = 0$ through the origin.

Example 4: Null Space $N(A) = \lbrace\vec{x} : A\vec{x} = \vec{0}\rbrace$ β€” ALWAYS a subspace

  1. $A\vec{0} = \vec{0}$, so $\vec{0} \in N(A)$. βœ“
  2. If $A\vec{u} = \vec{0}$ and $A\vec{v} = \vec{0}$, then $A(\vec{u} + \vec{v}) = A\vec{u} + A\vec{v} = \vec{0}$. βœ“
  3. $A(c\vec{v}) = cA\vec{v} = c\vec{0} = \vec{0}$. βœ“

This is a very important result in linear algebra! The solution space of any homogeneous system is always a subspace.

Example 5: $W = \lbrace p(x) \in P_3 : p(0) = 0\rbrace$ β€” Polynomial subspace βœ“

  1. Zero polynomial: $p(x) = 0$ satisfies $p(0) = 0$. βœ“
  2. If $p(0) = 0$ and $q(0) = 0$, then $(p+q)(0) = p(0)+q(0) = 0$. βœ“
  3. $(cp)(0) = c \cdot p(0) = c \cdot 0 = 0$. βœ“

These are polynomials with no constant term: $p(x) = a_1x + a_2x^2 + a_3x^3$.


3.3a Important Subspace Theorems

Theorem (Intersection of Subspaces): If $W_1$ and $W_2$ are subspaces of a vector space $V$, then $W_1 \cap W_2$ is also a subspace of $V$.

Proof:

Let $\vec{u}, \vec{v} \in W_1 \cap W_2$ and $\alpha, \beta \in \mathbb{R}$.

Then $\vec{u}, \vec{v} \in W_1$ and $\vec{u}, \vec{v} \in W_2$.

Since $W_1$ is a subspace: $\alpha\vec{u} + \beta\vec{v} \in W_1$.

Since $W_2$ is a subspace: $\alpha\vec{u} + \beta\vec{v} \in W_2$.

Therefore $\alpha\vec{u} + \beta\vec{v} \in W_1 \cap W_2$.

Also, $\vec{0} \in W_1$ and $\vec{0} \in W_2$, so $\vec{0} \in W_1 \cap W_2$.

$\therefore W_1 \cap W_2$ is a subspace of $V$. $\blacksquare$

Union of Subspaces is NOT a Subspace (in general):

Let $V = \mathbb{R}^2$, $W_1 = \text{x-axis}$, $W_2 = \text{y-axis}$. Both are subspaces of $\mathbb{R}^2$.

$(1, 0) \in W_1 \subseteq W_1 \cup W_2$ and $(0, 1) \in W_2 \subseteq W_1 \cup W_2$.

But $(1, 0) + (0, 1) = (1, 1) \notin W_1 \cup W_2$ (since $(1,1)$ is on neither axis).

$\therefore W_1 \cup W_2$ is not closed under addition and hence not a subspace.

Theorem (Sum of Subspaces): If $W_1$ and $W_2$ are subspaces of a vector space $V$, then

\[W_1 + W_2 = \lbrace \vec{w}_1 + \vec{w}_2 : \vec{w}_1 \in W_1, \vec{w}_2 \in W_2 \rbrace\]

is also a subspace of $V$.

Proof:

Let $\vec{u}, \vec{v} \in W_1 + W_2$ and $\alpha, \beta \in \mathbb{R}$.

Write $\vec{u} = \vec{u}_1 + \vec{u}_2$ and $\vec{v} = \vec{v}_1 + \vec{v}_2$ where $\vec{u}_1, \vec{v}_1 \in W_1$ and $\vec{u}_2, \vec{v}_2 \in W_2$.

\[\alpha\vec{u} + \beta\vec{v} = \alpha(\vec{u}_1 + \vec{u}_2) + \beta(\vec{v}_1 + \vec{v}_2) = \underbrace{(\alpha\vec{u}_1 + \beta\vec{v}_1)}_{\in W_1} + \underbrace{(\alpha\vec{u}_2 + \beta\vec{v}_2)}_{\in W_2}\]

So $\alpha\vec{u} + \beta\vec{v} \in W_1 + W_2$.

Also, $\vec{0} = \vec{0} + \vec{0} \in W_1 + W_2$.

$\therefore W_1 + W_2$ is a subspace of $V$. $\blacksquare$


3.4 Definition of Span

Definition: The span of a set $S = \lbrace\vec{v}_1, \vec{v}_2, \ldots, \vec{v}_k\rbrace$ is the set of all possible linear combinations:

\[\text{Span}(S) = \lbrace c_1\vec{v}_1 + c_2\vec{v}_2 + \cdots + c_k\vec{v}_k : c_i \in \mathbb{R} \rbrace\]

If $S$ spans $V$ (i.e., $\text{Span}(S) = V$), we say β€œ$S$ is a spanning set of $V$.”


3.5 Span is Always a Subspace

Theorem: For any set of vectors $S$ in a vector space $V$, $\text{Span}(S)$ is a subspace of $V$.

Proof Sketch:

  1. $\vec{0} \in \text{Span}(S)$ (choose all $c_i = 0$)
  2. If $\vec{u}, \vec{v} \in \text{Span}(S)$, their sum is also a linear combination of vectors in $S$
  3. If $\vec{v} \in \text{Span}(S)$ and $c \in \mathbb{R}$, then $c\vec{v}$ is also a linear combination

3.6 Examples of Spanning Sets

Example 1: Parallel vectors β†’ span is a LINE

$\text{Span}\lbrace(1,2), (2,4)\rbrace$

Since $(2,4) = 2(1,2)$, both point in the same direction:

\[c_1(1,2) + c_2(2,4) = (c_1 + 2c_2)(1,2) = k(1,2)\]

Result: A line through origin with direction $(1,2)$ β€” a 1D subspace.

Example 2: Standard basis vectors β†’ span is ALL of $\mathbb{R}^2$

$\text{Span}\lbrace(1,0), (0,1)\rbrace$

\[c_1(1,0) + c_2(0,1) = (c_1, c_2)\]

For any $(x, y) \in \mathbb{R}^2$, choose $c_1 = x, c_2 = y$.

Result: $\text{Span}\lbrace(1,0), (0,1)\rbrace = \mathbb{R}^2$ β€” the entire plane!

Example 3: Three vectors in $\mathbb{R}^3$ β€” redundant vector

$\text{Span}\lbrace(1,0,0), (0,1,0), (1,1,0)\rbrace$

Since $(1,1,0) = (1,0,0) + (0,1,0)$, the third vector is redundant.

All combinations give $(c_1+c_3, c_2+c_3, 0)$ β€” always has $z = 0$.

Result: The $xy$-plane β€” a 2D subspace of $\mathbb{R}^3$.

Example 4: Spanning polynomials

$\text{Span}\lbrace 1, x, x^2\rbrace$

\[c_1(1) + c_2(x) + c_3(x^2) = c_1 + c_2x + c_3x^2\]

Result: $\text{Span}\lbrace 1, x, x^2\rbrace = P_2$ β€” all polynomials of degree $\leq 2$.


3.7 Properties of Spanning Sets

Note 1: If $\lbrace \vec{u}_1, \vec{u}_2, \ldots, \vec{u}_n \rbrace$ spans $V$, then for any vector $\vec{w} \in V$, the set $\lbrace \vec{w}, \vec{u}_1, \vec{u}_2, \ldots, \vec{u}_n \rbrace$ also spans $V$.

(Adding a vector to a spanning set preserves spanning.)

Note 2: If $\lbrace \vec{u}1, \vec{u}_2, \ldots, \vec{u}_n \rbrace$ spans $V$ and $\vec{u}_n$ is a linear combination of the other vectors, then $\lbrace \vec{u}_1, \vec{u}_2, \ldots, \vec{u}{n-1} \rbrace$ also spans $V$.

(Removing a redundant vector from a spanning set preserves spanning.)

Note 3: If $\lbrace \vec{u}_1, \vec{u}_2, \ldots, \vec{u}_n \rbrace$ spans $V$ and one of the vectors is the zero vector, then the remaining vectors also span $V$.

(Removing the zero vector from a spanning set preserves spanning.)


3.8 Worked Examples β€” Spanning & Linear Combination

Example 5: Express $(3, 7, -4)$ as a linear combination

Problem: Express $(3, 7, -4)$ as a linear combination of $\vec{u}_1 = (1, 2, 3)$, $\vec{u}_2 = (2, 3, 7)$, $\vec{u}_3 = (3, 5, 6)$.

Set up $\alpha_1 \vec{u}_1 + \alpha_2 \vec{u}_2 + \alpha_3 \vec{u}_3 = (3, 7, -4)$:

\[\alpha_1 + 2\alpha_2 + 3\alpha_3 = 3\] \[2\alpha_1 + 3\alpha_2 + 5\alpha_3 = 7\] \[3\alpha_1 + 7\alpha_2 + 6\alpha_3 = -4\]

Form the augmented matrix and row-reduce:

\[\left[\begin{array}{ccc|c} 1 & 2 & 3 & 3 \\ 2 & 3 & 5 & 7 \\ 3 & 7 & 6 & -4 \end{array}\right] \xrightarrow{R_2 - 2R_1,\; R_3 - 3R_1} \left[\begin{array}{ccc|c} 1 & 2 & 3 & 3 \\ 0 & -1 & -1 & 1 \\ 0 & 1 & -3 & -13 \end{array}\right]\] \[\xrightarrow{R_3 + R_2} \left[\begin{array}{ccc|c} 1 & 2 & 3 & 3 \\ 0 & -1 & -1 & 1 \\ 0 & 0 & -4 & -12 \end{array}\right]\]

$\rho(A) = \rho(A \mid B) = 3$ β€” unique solution exists.

Back-substitution: $\alpha_3 = 3$, $\alpha_2 = -4$, $\alpha_1 = 2$.

\[(3, 7, -4) = 2(1, 2, 3) - 4(2, 3, 7) + 3(3, 5, 6) \checkmark\]

Remark: This problem is equivalent to solving a system $A\vec{x} = \vec{b}$ where the given vectors are the columns of $A$ and the target vector is $\vec{b}$. No solution means $\vec{b}$ is not a linear combination of the columns.

Example 6: Show $\lbrace(1, 1, 1), (1, 1, 0), (1, 0, 0)\rbrace$ spans $\mathbb{R}^3$

Problem: Show that $\lbrace\vec{w}_1, \vec{w}_2, \vec{w}_3\rbrace = \lbrace(1, 1, 1), (1, 1, 0), (1, 0, 0)\rbrace$ spans $\mathbb{R}^3$.

For any $(a, b, c) \in \mathbb{R}^3$, we need to find $x, y, z$ such that:

\[x(1, 1, 1) + y(1, 1, 0) + z(1, 0, 0) = (a, b, c)\]

This gives:

\[x + y + z = a, \quad x + y = b, \quad x = c\]

Solving: $x = c$, $y = b - c$, $z = a - b$.

Since $x, y, z$ exist for every $(a, b, c) \in \mathbb{R}^3$, the set spans $\mathbb{R}^3$. $\checkmark$


Practice Questions β€” Subspaces & Spanning Sets

Q1. Is $W = \lbrace(x, y, z) : z = x + y\rbrace$ a subspace of $\mathbb{R}^3$? Justify using the 3-condition test.

Click for solution

Yes β€” subspace.

  1. Zero: $(0,0,0)$: $0 = 0+0$ βœ“
  2. Closed +: If $z_1 = x_1+y_1$ and $z_2 = x_2+y_2$, then $z_1+z_2 = (x_1+x_2)+(y_1+y_2)$ βœ“
  3. Closed Β·: $cz = c(x+y) = cx+cy$ βœ“

Geometrically, this is the plane $z = x + y$ through the origin.

Q2. Is $W = \lbrace(x, y) : xy = 0\rbrace$ a subspace of $\mathbb{R}^2$? (Hint: Try adding $(1,0)$ and $(0,1)$.)

Click for solution

No. $(1,0) \in W$ (since $1 \cdot 0 = 0$) and $(0,1) \in W$ (since $0 \cdot 1 = 0$), but $(1,0) + (0,1) = (1,1)$ and $1 \cdot 1 = 1 \neq 0$. Fails closure under addition.

Geometrically, $W$ is the union of the two coordinate axes β€” not a subspace (looks like a cross, not a line/plane through origin).

Q3. Is the set of all differentiable functions a subspace of $C(\mathbb{R})$?

Click for solution

Yes. Let $f, g$ be differentiable functions and $c \in \mathbb{R}$.

  1. Zero: $f(x) = 0$ is differentiable βœ“
  2. Closed +: $f + g$ is differentiable (sum rule) βœ“
  3. Closed Β·: $cf$ is differentiable (constant multiple rule) βœ“

Q4. Is $W = \lbrace A \in M_{2 \times 2}(\mathbb{R}) : \det(A) = 0\rbrace$ a subspace? (Hint: Find two matrices with det = 0 whose sum has det β‰  0.)

Click for solution

No β€” same counterexample as Vector Spaces Q4.

$A = \begin{pmatrix} 1 & 0 \ 0 & 0 \end{pmatrix}$, $B = \begin{pmatrix} 0 & 0 \ 0 & 1 \end{pmatrix}$: $\det(A) = \det(B) = 0$, but $\det(A+B) = 1 \neq 0$.

Q5. Prove that the solution space of $A\vec{x} = \vec{0}$ is always a subspace.

Click for solution

Let $N(A) = \lbrace\vec{x} : A\vec{x} = \vec{0}\rbrace$.

  1. Zero: $A\vec{0} = \vec{0}$, so $\vec{0} \in N(A)$ βœ“
  2. Closed +: If $A\vec{u} = \vec{0}$ and $A\vec{v} = \vec{0}$, then $A(\vec{u}+\vec{v}) = A\vec{u} + A\vec{v} = \vec{0} + \vec{0} = \vec{0}$ βœ“
  3. Closed Β·: $A(c\vec{v}) = cA\vec{v} = c\vec{0} = \vec{0}$ βœ“

Note: This does NOT work for $A\vec{x} = \vec{b}$ with $\vec{b} \neq \vec{0}$ (non-homogeneous systems), since $\vec{0}$ is not a solution.

Q6. Find the following span β€” describe it geometrically.

\[\text{Span}\left\lbrace\begin{pmatrix}1\\1\\1\end{pmatrix}, \begin{pmatrix}1\\2\\3\end{pmatrix}\right\rbrace\]
Click for solution
\[c_1\begin{pmatrix}1\\1\\1\end{pmatrix} + c_2\begin{pmatrix}1\\2\\3\end{pmatrix} = \begin{pmatrix}c_1+c_2\\c_1+2c_2\\c_1+3c_2\end{pmatrix}\]

Why is their span a plane through the origin?

Key concept: In $\mathbb{R}^n$, the span of $k$ linearly independent vectors is a $k$-dimensional subspace:

# linearly independent vectors in $\mathbb{R}^3$ Span is Why?
0 The origin (a point) No vectors to combine β†’ only $\vec{0}$ is reachable
1 A line through the origin $\lbrace c\vec{v}\rbrace$ β€” scaling one vector forward/backward traces a line
2 A plane through the origin Two non-parallel directions β†’ you can move freely in a flat 2D surface
3 All of $\mathbb{R}^3$ Three non-coplanar directions β†’ every point in 3D space is reachable

Intuition row by row:

  • 0 vectors: You have nothing to combine. The only linear combination is $\vec{0}$ itself β€” just the origin.
  • 1 vector $\vec{v}$: $\text{Span}\lbrace\vec{v}\rbrace = \lbrace c\vec{v} : c \in \mathbb{R}\rbrace$. Varying $c$ stretches $\vec{v}$ forward ($c>0$), backward ($c<0$), or to zero ($c=0$). This traces a line through the origin along $\vec{v}$.
  • 2 vectors $\vec{v}_1, \vec{v}_2$ (independent, so not parallel): $\text{Span} = \lbrace c_1\vec{v}_1 + c_2\vec{v}_2\rbrace$. You can slide along $\vec{v}_1$ and independently along $\vec{v}_2$. Since they point in genuinely different directions, you sweep out a flat 2D surface β€” a plane through the origin.
  • 3 vectors (independent, so not coplanar): Now you can move in three independent directions. You can reach every point in $\mathbb{R}^3$.

Step 1 β€” Check linear independence: If $\vec{v}_2 = k\,\vec{v}_1$, then from the first component $k = 1$, but from the second $k = 2$ β€” contradiction. So the two vectors are not scalar multiples of each other, i.e., they are linearly independent.

Step 2 β€” Conclude dimension: Two linearly independent vectors in $\mathbb{R}^3$ span a 2-dimensional subspace, which is geometrically a plane.

Step 3 β€” Through the origin: Every span is a subspace, and every subspace contains $\vec{0}$ (set $c_1 = c_2 = 0$). So the plane passes through the origin.

Finding the plane equation: The normal $\vec{n} = (1,1,1) \times (1,2,3) = (3-2, 1-3, 2-1) = (1, -2, 1)$.

Plane: $x - 2y + z = 0$.

Q7. Does $(3, 1)$ lie in $\text{Span}\lbrace(1, 0), (0, 1), (1, 1)\rbrace$?

Click for solution

Yes. $\text{Span}\lbrace(1,0), (0,1), (1,1)\rbrace = \mathbb{R}^2$ since $\lbrace(1,0), (0,1)\rbrace$ already spans $\mathbb{R}^2$. Every vector in $\mathbb{R}^2$ is in this span.

Explicitly: $(3,1) = 3(1,0) + 1(0,1) + 0(1,1)$.

Q8. Find $\text{Span}\lbrace x^2, x^2 + 1\rbrace$ in $P_2$.

Click for solution
\[c_1 x^2 + c_2(x^2 + 1) = (c_1 + c_2)x^2 + c_2\]

This produces all polynomials of the form $ax^2 + b$ (no $x$ term). Setting $a = c_1+c_2$ and $b = c_2$, both $a$ and $b$ range freely over $\mathbb{R}$.

Answer: $\text{Span}\lbrace x^2, x^2+1\rbrace = \lbrace ax^2 + b : a, b \in \mathbb{R}\rbrace$ β€” all polynomials with no linear term.

Q9. Show that $\lbrace(1,1), (1,-1)\rbrace$ spans $\mathbb{R}^2$. (Express any $(a,b)$ as a linear combination.)

Click for solution

Solve $c_1(1,1) + c_2(1,-1) = (a,b)$:

\[c_1 + c_2 = a, \quad c_1 - c_2 = b\]

Adding: $2c_1 = a + b \Rightarrow c_1 = \frac{a+b}{2}$. Subtracting: $2c_2 = a - b \Rightarrow c_2 = \frac{a-b}{2}$.

Since $c_1, c_2$ exist for any $(a,b) \in \mathbb{R}^2$, the set spans $\mathbb{R}^2$. βœ“

Q10. Is $W = \lbrace(x,y,z) : x + y + z = 1\rbrace$ a subspace? If not, what geometric object is it?

Click for solution

Not a subspace. $(0,0,0) \notin W$ since $0+0+0 = 0 \neq 1$.

Geometrically, $W$ is a plane in $\mathbb{R}^3$ that does NOT pass through the origin (it has intercepts $(1,0,0)$, $(0,1,0)$, $(0,0,1)$). It’s an affine subspace (translated subspace).

What is an affine subspace?

An affine subspace is a subspace that has been shifted (translated) away from the origin:

\[\text{Affine subspace} = \text{subspace} + \text{a fixed point}\]
  • The subspace version of $W$ would be $\lbrace(x,y,z) : x+y+z = 0\rbrace$ β€” a plane through the origin.
  • $W$ is that same plane, but shifted so it passes through $(1,0,0)$ instead of $(0,0,0)$.
  • We can write: $W = \lbrace(1,0,0) + \vec{v} : \vec{v} \in \text{plane } x+y+z=0\rbrace$.

Analogy: A subspace is a flat surface pinned at the origin. An affine subspace is the same flat surface, but slid to a different position. That’s why it fails the subspace test β€” it no longer contains $\vec{0}$.

Q11. Show that $\text{Span}\lbrace(1,0,0), (0,1,0), (0,0,1)\rbrace = \mathbb{R}^3$.

Click for solution

For any $(a,b,c) \in \mathbb{R}^3$: $(a,b,c) = a(1,0,0) + b(0,1,0) + c(0,0,1)$.

The coefficients are simply the components of the vector. Since every $(a,b,c)$ can be expressed this way, $\text{Span}\lbrace\vec{e}_1, \vec{e}_2, \vec{e}_3\rbrace = \mathbb{R}^3$. βœ“

Q12. Is $W = \lbrace(x,y) : x \geq 0\rbrace$ (the right half-plane) a subspace? Why or why not?

Click for solution

No β€” fails closure under scalar multiplication.

$(1, 0) \in W$ (since $1 \geq 0$), but $(-1)(1, 0) = (-1, 0) \notin W$ (since $-1 < 0$).


4 β€” Linear Independence & Dependence


4.1 Definitions

Linear Dependence: Vectors $\lbrace\vec{v}_1, \ldots, \vec{v}_k\rbrace$ are linearly dependent if there exist scalars $c_1, \ldots, c_k$, not all zero, such that:

\[c_1\vec{v}_1 + c_2\vec{v}_2 + \cdots + c_k\vec{v}_k = \vec{0}\]

Linear Independence: Vectors $\lbrace\vec{v}_1, \ldots, \vec{v}_k\rbrace$ are linearly independent if the only solution to:

\[c_1\vec{v}_1 + c_2\vec{v}_2 + \cdots + c_k\vec{v}_k = \vec{0}\]

is $c_1 = c_2 = \cdots = c_k = 0$ (the trivial solution).

Plain English:

  • Dependent = at least one vector is β€œredundant” (expressible from the others)
  • Independent = every vector contributes something new β€” no redundancy

4.2 Geometric Interpretation

In $\mathbb{R}^n$ Independent Dependent
1 vector Non-zero vector Zero vector
2 vectors in $\mathbb{R}^2$ Different directions One is a scalar multiple of the other (collinear)
3 vectors in $\mathbb{R}^2$ IMPOSSIBLE β€” always dependent Always β€” can’t have 3 independent in 2D
3 vectors in $\mathbb{R}^3$ Not coplanar (point in 3 different β€œdirections”) Coplanar (all lie in one plane)
$n+1$ or more in $\mathbb{R}^n$ IMPOSSIBLE β€” always dependent Always

4.3 Method 1 β€” Direct Equation Method

Procedure:

  1. Set up: $c_1\vec{v}_1 + c_2\vec{v}_2 + \cdots + c_k\vec{v}_k = \vec{0}$
  2. Write as a system of linear equations
  3. Solve for $c_1, c_2, \ldots, c_k$
  4. Only trivial solution $(0, 0, \ldots, 0)$ β†’ Independent
  5. Non-trivial solution exists β†’ Dependent

Example 1: Dependent β€” $(1,2)$ and $(2,4)$

$c_1(1,2) + c_2(2,4) = (0,0)$

System: $c_1 + 2c_2 = 0$ and $2c_1 + 4c_2 = 0$

From first: $c_1 = -2c_2$. Second equation: $2(-2c_2) + 4c_2 = 0$ βœ“ for all $c_2$.

Non-trivial solution: Choose $c_2 = 1 \Rightarrow c_1 = -2$.

Check: $-2(1,2) + 1(2,4) = (-2,-4)+(2,4) = (0,0)$ βœ“

Linearly dependent β€” indeed $(2,4) = 2(1,2)$.

Example 2: Independent β€” $(1,0)$ and $(0,1)$

$c_1(1,0) + c_2(0,1) = (0,0) \Rightarrow c_1 = 0, c_2 = 0$.

Only the trivial solution. Linearly independent!

Example 3: Three vectors in $\mathbb{R}^3$ β€” Dependent

Test $\lbrace(1,0,0), (0,1,0), (1,1,0)\rbrace$:

$c_1(1,0,0) + c_2(0,1,0) + c_3(1,1,0) = (0,0,0)$

System: $c_1 + c_3 = 0$, $c_2 + c_3 = 0$, $0 = 0$

Two equations, three unknowns β€” $c_3$ is free. Let $c_3 = 1$: $c_1 = -1, c_2 = -1$.

Dependent: $(1,1,0) = (1,0,0) + (0,1,0)$ β€” third vector is sum of first two!

Example 4: Standard basis β€” Independent

Test $\lbrace(1,0,0), (0,1,0), (0,0,1)\rbrace$:

$c_1 = 0, c_2 = 0, c_3 = 0$ β€” only trivial solution.

Linearly independent β€” these are the standard basis vectors of $\mathbb{R}^3$.

Example 5: Polynomial dependence

Test $\lbrace1, x, 2+2x\rbrace$ in $P_1$:

$c_1(1) + c_2(x) + c_3(2+2x) = 0$

$(c_1 + 2c_3) + (c_2 + 2c_3)x = 0 + 0 \cdot x$

Comparing: $c_1 + 2c_3 = 0$ and $c_2 + 2c_3 = 0$.

Choose $c_3 = 1$: $c_1 = -2, c_2 = -2$.

Dependent: $2 + 2x = 2(1) + 2(x)$


4.4 Method 2 β€” Matrix Rank Method

For vectors in $\mathbb{R}^n$: form a matrix with vectors as columns and row-reduce.

  • rank = number of vectors β†’ Independent
  • rank < number of vectors β†’ Dependent

Example: Test columns of

\[A = \begin{bmatrix} 1 & 2 & 3 \\ 0 & 1 & 1 \\ 0 & 0 & 0 \end{bmatrix}\]

Rank = 2 (two pivots), but 3 vectors β†’ $2 < 3$ β†’ Dependent.


4.5 Key Theorems

Theorem 1 (Maximum Independent Vectors): In $\mathbb{R}^n$, any set of $n+1$ or more vectors is linearly dependent.

Intuition: You can’t have more than $n$ independent directions in $n$-dimensional space.

Example: In $\mathbb{R}^2$ ($n = 2$), take any 3 vectors, say $\lbrace(1,0), (0,1), (2,3)\rbrace$. These must be dependent:

$2(1,0) + 3(0,1) - 1(2,3) = (0,0)$. Non-trivial combination β†’ dependent. βœ“

No matter which 3 vectors you pick in $\mathbb{R}^2$, they will always be dependent.


Theorem 2 (Dependence β†’ Redundancy): If $\lbrace\vec{v}_1, \ldots, \vec{v}_k\rbrace$ is dependent, then at least one vector is a linear combination of the others. Removing it doesn’t change the span.

Example: $\lbrace(1,0), (0,1), (3,5)\rbrace$ is dependent (3 vectors in $\mathbb{R}^2$).

$(3,5) = 3(1,0) + 5(0,1)$ β€” the third vector is a combination of the first two.

Removing it: $\text{Span}\lbrace(1,0), (0,1), (3,5)\rbrace = \text{Span}\lbrace(1,0), (0,1)\rbrace = \mathbb{R}^2$. The span didn’t change!


Theorem 3 (Independence β†’ Unique Representation): If $\lbrace\vec{v}_1, \ldots, \vec{v}_k\rbrace$ is independent and $\vec{w} \in \text{Span}\lbrace\vec{v}_1, \ldots, \vec{v}_k\rbrace$, then the representation $\vec{w} = c_1\vec{v}_1 + \cdots + c_k\vec{v}_k$ is unique.

Example: $\lbrace(1,0), (0,1)\rbrace$ is independent. Express $(3,5)$:

$(3,5) = 3(1,0) + 5(0,1)$ β€” the only way to do it. You can’t find different $c_1, c_2$.

Contrast with dependent set: $\lbrace(1,0), (0,1), (1,1)\rbrace$ β€” now $(3,5) = 3(1,0) + 5(0,1) + 0(1,1)$, but also $(3,5) = 2(1,0) + 4(0,1) + 1(1,1)$. Multiple representations! Independence guarantees uniqueness.


Theorem 4 (Zero Vector): Any set containing $\vec{0}$ is linearly dependent. (Proof: assign coefficient 1 to $\vec{0}$ and 0 to everything else.)

Example: $\lbrace(1,2), (0,0), (3,4)\rbrace$ β€” is this independent?

$0(1,2) + 1(0,0) + 0(3,4) = (0,0)$. We found a non-trivial solution (coefficient of $\vec{0}$ is 1).

Always dependent. The zero vector is always β€œuseless” β€” it adds no new direction.


4.6 Important Remarks on LI/LD

Remark 1: A single non-zero vector $\vec{v}$ is always linearly independent, because $\alpha\vec{v} = \vec{0}$ with $\vec{v} \neq \vec{0}$ forces $\alpha = 0$.

Remark 2: Two vectors $\vec{v}_1, \vec{v}_2$ are linearly dependent if and only if one is a scalar multiple of the other, i.e., $\vec{v}_1 = k\vec{v}_2$ for some $k \in \mathbb{R}$, $k \neq 0$.

Remark 3: If two of the vectors $\vec{v}_1, \vec{v}_2, \ldots, \vec{v}_n$ are equal, or one is a scalar multiple of another, then the vectors must be linearly dependent.

Remark 4: If a set $S$ is linearly independent, then any rearrangement of the vectors in $S$ is also linearly independent.

Remark 5: If a set $S$ is linearly independent, then any subset of $S$ is also linearly independent. Equivalently, if $S$ contains a linearly dependent subset, then $S$ itself is linearly dependent.

Remark 6 (LD ⟹ LC of Preceding): If two or more non-zero vectors $\lbrace\vec{v}1, \vec{v}_2, \ldots, \vec{v}_n\rbrace$ are linearly dependent, then one of the vectors can be expressed as a linear combination of the preceding vectors: $\exists\, l$ such that $\vec{v}_l = \alpha_1\vec{v}_1 + \alpha_2\vec{v}_2 + \cdots + \alpha{l-1}\vec{v}_{l-1}$.

Theorem (Echelon Rows are LI): The non-zero rows of a matrix in echelon form are linearly independent.

This is why row reduction is so powerful β€” after reducing, the non-zero rows automatically give a linearly independent set.


Practice Questions β€” Linear Independence

Q1. Test $\lbrace(1,1), (2,2), (3,3)\rbrace$ for linear independence in $\mathbb{R}^2$. (Ans: Dependent β€” 3 vectors in $\mathbb{R}^2$ are always dependent.)

Click for solution

In $\mathbb{R}^2$, $\dim = 2$, so any set of 3+ vectors is automatically linearly dependent.

Explicitly: $(2,2) = 2(1,1)$ and $(3,3) = 3(1,1)$. Dependence relation: $1 \cdot (2,2) - 2 \cdot (1,1) + 0 \cdot (3,3) = \vec{0}$.

Q2. Show that $\lbrace(1,0,0), (0,1,0), (0,0,1)\rbrace$ is linearly independent.

Click for solution

Solve $c_1(1,0,0) + c_2(0,1,0) + c_3(0,0,1) = (0,0,0)$:

$(c_1, c_2, c_3) = (0, 0, 0)$. Only the trivial solution exists, so these are linearly independent.

Equivalently:

\[\det \begin{pmatrix} 1&0&0\\0&1&0\\0&0&1 \end{pmatrix} = 1 \neq 0 \checkmark\]

Q3. Are $\lbrace x, x^2, x^3\rbrace$ linearly independent in polynomial space?

Click for solution

Yes. Suppose $c_1 x + c_2 x^2 + c_3 x^3 = 0$ for all $x$.

Setting $x = 1$: $c_1 + c_2 + c_3 = 0$ Setting $x = -1$: $-c_1 + c_2 - c_3 = 0$ Setting $x = 2$: $2c_1 + 4c_2 + 8c_3 = 0$

Adding equations 1 and 2: $2c_2 = 0 \Rightarrow c_2 = 0$. Then $c_1 + c_3 = 0$ and $2c_1 + 8c_3 = 0 \Rightarrow c_1 = c_3 = 0$.

These are monomials of different degrees β€” always independent.

Q4. Test $\lbrace(1,2,-1), (2,4,-2), (3,6,-3)\rbrace$. (They’re all multiples of $(1,2,-1)$ β€” dependent.)

Click for solution

$(2,4,-2) = 2(1,2,-1)$ and $(3,6,-3) = 3(1,2,-1)$.

All three are scalar multiples of $(1,2,-1)$ β€” they lie on the same line through the origin.

Dependent. Relation: $-5(1,2,-1) + 1(2,4,-2) + 1(3,6,-3) = \vec{0}$.

Q5. Is the following set independent in $M_{2 \times 2}$?

\[\left\lbrace\begin{bmatrix}1&0\\0&0\end{bmatrix}, \begin{bmatrix}0&1\\0&0\end{bmatrix}, \begin{bmatrix}0&0\\1&0\end{bmatrix}, \begin{bmatrix}0&0\\0&1\end{bmatrix}\right\rbrace\]

(Ans: Yes.)

Click for solution

These are the standard basis matrices $E_{11}, E_{12}, E_{21}, E_{22}$ for $M_{2 \times 2}$.

If $c_1 E_{11} + c_2 E_{12} + c_3 E_{21} + c_4 E_{22} = \mathbf{0}$, then:

\[\begin{bmatrix}c_1&c_2\\c_3&c_4\end{bmatrix} = \begin{bmatrix}0&0\\0&0\end{bmatrix}\]

So $c_1 = c_2 = c_3 = c_4 = 0$. Linearly independent. (They also form a basis for $M_{2 \times 2}$, so $\dim M_{2 \times 2} = 4$.)

Q6. If $\lbrace\vec{u}, \vec{v}\rbrace$ is independent, show that $\lbrace\vec{u}, \vec{v}, \vec{u}+\vec{v}\rbrace$ is dependent.

Click for solution

$1 \cdot \vec{u} + 1 \cdot \vec{v} + (-1)(\vec{u}+\vec{v}) = \vec{u} + \vec{v} - \vec{u} - \vec{v} = \vec{0}$.

We found a non-trivial linear combination $(c_1, c_2, c_3) = (1, 1, -1) \neq (0,0,0)$ that gives $\vec{0}$.

Therefore $\lbrace\vec{u}, \vec{v}, \vec{u}+\vec{v}\rbrace$ is linearly dependent. The third vector is redundant since $\vec{u}+\vec{v} = 1\cdot\vec{u} + 1\cdot\vec{v}$.

Q7. Find a value of $k$ such that $\lbrace(1,2), (3,k)\rbrace$ is linearly dependent. (Ans: $k = 6$)

Click for solution

Two vectors in $\mathbb{R}^2$ are dependent iff one is a scalar multiple of the other, i.e., $(3, k) = c(1, 2)$ for some $c$.

$c = 3$, so $k = 2c = 6$.

Equivalently: $\det \begin{pmatrix} 1 & 3 \ 2 & k \end{pmatrix} = k - 6 = 0 \Rightarrow k = 6$.

Q8. Test $\lbrace1+x, 1-x, 2\rbrace$ for independence in $P_1$. (Hint: $(1+x) + (1-x) = 2$.)

Click for solution

Dependent. $(1+x) + (1-x) - 1 \cdot (2) = 2 - 2 = 0$.

Non-trivial combination: $c_1 = 1, c_2 = 1, c_3 = -1$.

Also: $\dim(P_1) = 2$ (basis $\lbrace1, x\rbrace$), but we have 3 vectors β€” must be dependent.

Q9. Show that any set containing the zero vector is dependent.

Click for solution

Say $\vec{v}_1 = \vec{0}$. Then $1 \cdot \vec{0} + 0 \cdot \vec{v}_2 + \cdots + 0 \cdot \vec{v}_k = \vec{0}$.

This is a non-trivial solution (coefficient of $\vec{0}$ is 1, not 0), so the set is dependent.

Intuition: The zero vector contributes nothing β€” it’s always redundant.

Q10. If $\lbrace\vec{v}_1, \vec{v}_2, \vec{v}_3\rbrace$ is independent, is $\lbrace\vec{v}_1, \vec{v}_2, \vec{v}_3, \vec{v}_1 + \vec{v}_2 + \vec{v}_3\rbrace$ independent?

Click for solution

No β€” dependent. $1 \cdot \vec{v}_1 + 1 \cdot \vec{v}_2 + 1 \cdot \vec{v}_3 + (-1)(\vec{v}_1+\vec{v}_2+\vec{v}_3) = \vec{0}$.

The fourth vector is a linear combination of the first three, so it’s redundant. Adding a vector from $\text{Span}\lbrace\vec{v}_1, \vec{v}_2, \vec{v}_3\rbrace$ always creates dependence.

Q11. A matrix has 3 columns and rank 2. What can you say about the independence of its columns?

Click for solution

The columns are linearly dependent. Rank = number of pivot columns = 2, but there are 3 columns. So at least one column is a non-pivot (free) column β€” it can be expressed as a combination of the pivot columns.

Specifically: nullity = 3 βˆ’ 2 = 1, so there exists a non-trivial solution to $A\vec{x} = \vec{0}$.

Q12. Determine by inspection (without calculation): Are $\lbrace(1,0,0), (0,2,0), (0,0,3)\rbrace$ independent? (Ans: Yes β€” non-zero entries in different positions.)

Click for solution

Yes. Each vector has its non-zero entry in a different coordinate position. The matrix

\[\begin{pmatrix} 1&0&0\\0&2&0\\0&0&3 \end{pmatrix}\]

is diagonal with all non-zero diagonal entries, so $\det = 6 \neq 0$.

These are scalar multiples of the standard basis vectors: $\vec{e}_1, 2\vec{e}_2, 3\vec{e}_3$.


5 β€” Basis & Dimension


5.1 Definition of Basis

Definition: A set $B = \lbrace\vec{v}_1, \vec{v}_2, \ldots, \vec{v}_n\rbrace$ is a basis of vector space $V$ if:

  1. $B$ spans $V$: every vector in $V$ is a linear combination of $B$
  2. $B$ is linearly independent: no vector in $B$ is redundant

Equivalently: Every vector in $V$ can be written UNIQUELY as:

\[\vec{v} = c_1\vec{v}_1 + c_2\vec{v}_2 + \cdots + c_n\vec{v}_n\]

The scalars $(c_1, \ldots, c_n)$ are the coordinates of $\vec{v}$ relative to $B$.

Basis Intuition:

  • Spanning = β€œenough” vectors to reach every point
  • Independent = β€œno extra” vectors (no redundancy)
  • Basis = just the right number of vectors!

5.2 Examples of Bases

Example 1: Standard basis of $\mathbb{R}^2$

$B = \lbrace(1,0), (0,1)\rbrace$

Spanning: $(x, y) = x(1,0) + y(0,1)$ for any $(x,y)$. βœ“

Independent: $c_1(1,0) + c_2(0,1) = (0,0) \Rightarrow c_1 = c_2 = 0$. βœ“

Coordinates: $(3, 5)$ has coordinates $(3, 5)$ relative to $B$.

Example 2: Non-standard basis of $\mathbb{R}^2$

$B = \lbrace(1,1), (1,-1)\rbrace$

Spanning: Solve $c_1(1,1) + c_2(1,-1) = (a,b)$:

\[c_1 + c_2 = a, \quad c_1 - c_2 = b\] \[c_1 = \frac{a+b}{2}, \quad c_2 = \frac{a-b}{2}\]

Unique solution for any $(a,b)$. βœ“

Independent: Set $a = b = 0 \Rightarrow c_1 = c_2 = 0$. βœ“

Example coordinates: $(3,1)$ has coordinates $c_1 = 2, c_2 = 1$, i.e., $(3,1) = 2(1,1) + 1(1,-1)$.

Why $c_1 = 2, c_2 = 1$ and not $(3,1)$?
In the standard basis $\lbrace(1,0),(0,1)\rbrace$, the vector $(3,1)$ is $3(1,0) + 1(0,1)$ β€” its coordinates are simply $3$ and $1$. But with respect to basis $B = \lbrace(1,1),(1,-1)\rbrace$, we must express it as a combination of those vectors. Plugging into the formulas: $c_1 = \frac{3+1}{2} = 2$, $c_2 = \frac{3-1}{2} = 1$. Verify: $2(1,1) + 1(1,-1) = (2,2) + (1,-1) = (3,1)$ βœ“. The same vector has different coordinates depending on which basis you use β€” $c_1, c_2$ are the weights (scalars) you multiply each basis vector by to reconstruct the original vector.

Example 3: Basis of $P_2$

$B = \lbrace1, x, x^2\rbrace$

Any $p(x) = a_0 + a_1x + a_2x^2 = a_0(1) + a_1(x) + a_2(x^2)$. βœ“

Independent: $c_1 + c_2x + c_3x^2 = 0$ for all $x$ forces $c_1 = c_2 = c_3 = 0$. βœ“

$(3 + 2x - x^2)$ has coordinates $(3, 2, -1)$.

Example 4: Checking if vectors form a basis of $\mathbb{R}^3$

Problem: Is $B = \lbrace(1,0,1), (0,1,1), (1,1,0)\rbrace$ a basis of $\mathbb{R}^3$?

Form the matrix with these as columns and compute the determinant:

\[\det\begin{bmatrix} 1 & 0 & 1 \\ 0 & 1 & 1 \\ 1 & 1 & 0 \end{bmatrix} = 1(0-1) - 0 + 1(0-1) = -1 - 1 = -2 \neq 0\]

Non-zero determinant β†’ independent β†’ spans $\mathbb{R}^3$ (3 independent vectors in 3D).

Yes, $B$ is a basis of $\mathbb{R}^3$.

Example 5: Basis of a subspace

Problem: Find a basis for $W = \lbrace(x,y,z) \in \mathbb{R}^3 : x + y + z = 0\rbrace$.

Parametrize: $z = -x - y$, so every element is:

$(x, y, -x-y) = x(1, 0, -1) + y(0, 1, -1)$

Basis: $B = \lbrace(1, 0, -1), (0, 1, -1)\rbrace$

These are independent (not proportional). They span $W$.


5.3 Dimension of a Vector Space

Definition: The dimension of $V$, denoted $\dim(V)$, is the number of vectors in any basis of $V$.

Fundamental Theorem: All bases of a vector space have the same number of elements.

How to find dimension:

  1. Find any basis of the space
  2. Count the number of vectors
  3. That count is the dimension

5.4 Dimension of Common Spaces

Vector Space Dimension Standard Basis
$\mathbb{R}^n$ $n$ $e_1, e_2, \ldots, e_n$
$P_n$ (polynomials of degree $\leq n$) $n + 1$ $1, x, x^2, \ldots, x^n$
$M_{m \times n}(\mathbb{R})$ (matrices) $mn$ Matrices with single 1, rest 0
$\lbrace\vec{0}\rbrace$ (zero space) $0$ $\emptyset$
Line through origin in $\mathbb{R}^3$ $1$ One direction vector
Plane through origin in $\mathbb{R}^3$ $2$ Two independent vectors in the plane
$C(\mathbb{R})$ (continuous functions) $\infty$ None (infinite-dimensional)

5.5 Key Theorems on Basis & Dimension

Theorem 1: Let $V$ be a vector space of finite dimension $m$. Then:

(a) Any $m + 1$ or more vectors in $V$ are linearly dependent.

(b) Any linearly independent set with exactly $m$ elements is a basis of $V$.

(c) Any spanning set with exactly $m$ elements is a basis of $V$.

Theorem 2 (Extracting a Basis from a Spanning Set): Suppose $S$ spans a vector space $V$. Then:

(a) Any maximum linearly independent subset of $S$ forms a basis of $V$.

(b) If one deletes from $S$ every vector that is a linear combination of its preceding vectors, the remaining vectors form a basis of $V$.

Theorem 3 (Extension to a Basis): Let $V$ have finite dimension, and let $S = \lbrace\vec{u}_1, \ldots, \vec{u}_r\rbrace$ be a set of linearly independent vectors in $V$. Then $S$ can be extended to a basis of $V$.

That is, there exist vectors $\vec{u}{r+1}, \ldots, \vec{u}_n$ such that $\lbrace\vec{u}_1, \ldots, \vec{u}_r, \vec{u}{r+1}, \ldots, \vec{u}_n\rbrace$ is a basis of $V$.

Theorem 4 (Subspace Dimension Bound): Let $W$ be a subspace of an $m$-dimensional vector space $V$. Then:

\[\dim(W) \leq m\]

In particular, if $\dim(W) = m$, then $W = V$.


Practice Questions β€” Basis & Dimension

Q1. Find a basis and dimension of $\lbrace(x,y,z) : 2x - y + z = 0\rbrace$ in $\mathbb{R}^3$. (Ans: dim = 2)

Click for solution

From $2x - y + z = 0$: $y = 2x + z$.

$(x, y, z) = (x, 2x+z, z) = x(1, 2, 0) + z(0, 1, 1)$.

Basis: $\lbrace(1, 2, 0), (0, 1, 1)\rbrace$, dim = 2.

Verify: $2(1)-2+0 = 0$ βœ“, $2(0)-1+1 = 0$ βœ“. Independent (not scalar multiples) βœ“.

Q2. Is $\lbrace(1,2,1), (2,1,3), (1,-3,1)\rbrace$ a basis of $\mathbb{R}^3$? (Check det.)

Click for solution
\[\det \begin{pmatrix} 1&2&1\\2&1&3\\1&-3&1 \end{pmatrix} = 1(1+9) - 2(2-3) + 1(-6-1) = 10 + 2 - 7 = 5 \neq 0\]

Yes β€” determinant is non-zero, so the three vectors are independent and span $\mathbb{R}^3$. They form a basis.

Q3. Find $\dim(P_5)$ and give a basis. (Ans: dim = 6)

Click for solution

$P_5$ = polynomials of degree $\leq 5$: $a_0 + a_1 x + a_2 x^2 + a_3 x^3 + a_4 x^4 + a_5 x^5$.

Standard basis: $\lbrace1, x, x^2, x^3, x^4, x^5\rbrace$. These are independent (different degrees) and span all of $P_5$.

dim($P_5$) = 6. In general, $\dim(P_n) = n + 1$.

Q4. Find the dimension of the space of symmetric $2 \times 2$ matrices. (Ans: 3 β€” basis below)

\[\begin{bmatrix}1&0\\0&0\end{bmatrix}, \quad \begin{bmatrix}0&1\\1&0\end{bmatrix}, \quad \begin{bmatrix}0&0\\0&1\end{bmatrix}\]
Click for solution

A symmetric $2 \times 2$ matrix has the form (3 free parameters):

\[\begin{bmatrix} a & b \\ b & c \end{bmatrix}\] \[\begin{bmatrix} a & b \\ b & c \end{bmatrix} = a\begin{bmatrix}1&0\\0&0\end{bmatrix} + b\begin{bmatrix}0&1\\1&0\end{bmatrix} + c\begin{bmatrix}0&0\\0&1\end{bmatrix}\]

The three basis matrices are independent (non-zero in different positions) and span the space. dim = 3.

Q5. If $\dim(V) = n$ and $W$ is a subspace of $V$, what are the possible values of $\dim(W)$? (Ans: $0, 1, \ldots, n$)

Click for solution

$\dim(W) \in \lbrace0, 1, 2, \ldots, n\rbrace$.

  • $\dim(W) = 0$: $W = \lbrace\vec{0}\rbrace$
  • $\dim(W) = n$: $W = V$ (a subspace with full dimension equals the whole space)
  • $\dim(W)$ cannot exceed $n$ because a basis of $W$ is an independent set in $V$, and independent sets in $V$ have at most $n$ elements.

Recall β€” Null Space & Nullity:

The null space (or kernel) of an $m \times n$ matrix $A$ is the set of all solutions to the homogeneous system $A\vec{x} = \vec{0}$:

\[N(A) = \lbrace\vec{x} \in \mathbb{R}^n : A\vec{x} = \vec{0}\rbrace\]

The null space is always a subspace of $\mathbb{R}^n$. Its dimension is called the nullity of $A$ β€” equivalently, the number of free variables after row reduction.

Q6. Find a basis for the null space of

\[A = \begin{bmatrix} 1 & 2 & 3 \\ 2 & 4 & 6 \end{bmatrix}\]
Click for solution

Row reduce: $R_2 - 2R_1 = [0, 0, 0]$. REF:

\[\begin{bmatrix} 1 & 2 & 3 \\ 0 & 0 & 0 \end{bmatrix}\]

Only one pivot (column 1). Free variables: $x_2 = s, x_3 = t$.

$x_1 = -2s - 3t$. Solution: $(x_1, x_2, x_3) = s(-2, 1, 0) + t(-3, 0, 1)$.

Basis for $N(A)$: $\lbrace(-2, 1, 0), (-3, 0, 1)\rbrace$, nullity = 2.

Q7. Find $\dim(M_{2 \times 3}(\mathbb{R}))$. (Ans: 6)

Click for solution

$M_{2 \times 3}$ has $2 \times 3 = 6$ entries. Standard basis: the 6 matrices $E_{ij}$ with 1 in position $(i,j)$ and 0 elsewhere.

dim = 6. In general, $\dim(M_{m \times n}) = mn$.

Q8. If a subspace of $\mathbb{R}^4$ has basis $\lbrace(1,0,1,0), (0,1,0,1)\rbrace$, what is its dimension?

Click for solution

dim = 2 (the number of vectors in the basis). Two vectors are independent (not scalar multiples), so this is a valid basis for a 2-dimensional subspace of $\mathbb{R}^4$.

Q9. True or False: A vector space can have more than one basis. (Ans: True β€” but all bases have the same number of elements.)

Click for solution

True. Example: $\mathbb{R}^2$ has infinitely many bases:

  • $\lbrace(1,0), (0,1)\rbrace$ β€” standard basis
  • $\lbrace(1,1), (1,-1)\rbrace$
  • $\lbrace(2,0), (0,3)\rbrace$

All have exactly 2 elements (dim = 2). The Dimension Theorem guarantees all bases have the same cardinality.

Q10. Find a basis for $W = \lbrace(x,y,z) \in \mathbb{R}^3 : x - 2y + z = 0\rbrace$. (Ans: e.g., $\lbrace(2,1,0), (-1,0,1)\rbrace$, dim = 2)

Click for solution

From $x - 2y + z = 0$: $x = 2y - z$.

$(x, y, z) = (2y - z, y, z) = y(2, 1, 0) + z(-1, 0, 1)$.

Basis: $\lbrace(2, 1, 0), (-1, 0, 1)\rbrace$, dim = 2.

Verify: $2-2+0 = 0$ βœ“, $-1-0+1 = 0$ βœ“.


6 β€” Rank of a Matrix


6.1 Definitions of Rank

Definition (Multiple Equivalent Perspectives): The rank of an $m \times n$ matrix $A$ can be defined as:

  1. Row Rank: Dimension of row space (span of row vectors)
  2. Column Rank: Dimension of column space (span of column vectors)
  3. Number of Pivots in the row echelon form
  4. Size of largest square submatrix with non-zero determinant

Fundamental Fact: All definitions give the SAME answer!

\[\text{row rank} = \text{column rank} = \text{rank}(A) = r\]

where $0 \leq r \leq \min(m, n)$.


6.2 Computing Rank β€” Row Reduction

Procedure:

  1. Reduce matrix to Row Echelon Form (REF)
  2. Count the number of non-zero rows
  3. That count is the rank

Example 1: Full Column Rank

\[A = \begin{bmatrix} 1 & 2 \\ 3 & 4 \\ 5 & 6 \end{bmatrix} \xrightarrow{\text{row reduce}} \begin{bmatrix} 1 & 2 \\ 0 & -2 \\ 0 & 0 \end{bmatrix}\]

Two non-zero rows β†’ $\text{rank}(A) = 2$. Both columns are independent (full column rank).

Example 2: Rank-Deficient Matrix

\[B = \begin{bmatrix} 1 & 2 & 3 \\ 2 & 4 & 6 \\ 1 & 2 & 3 \end{bmatrix} \xrightarrow{\text{row reduce}} \begin{bmatrix} 1 & 2 & 3 \\ 0 & 0 & 0 \\ 0 & 0 & 0 \end{bmatrix}\]

One non-zero row β†’ $\text{rank}(B) = 1$. All rows are scalar multiples of Row 1 ($R_2 = 2R_1$, $R_3 = 1 \cdot R_1$) β€” lots of redundancy!

Example 3: Square Invertible Matrix

\[C = \begin{bmatrix} 1 & 2 \\ 0 & 3 \end{bmatrix}\]

Already in REF. Two non-zero rows β†’ $\text{rank}(C) = 2$. Since $\det(C) = 3 \neq 0$, the matrix is invertible (full rank).

Example 4: Rectangular Matrix

\[D = \begin{bmatrix} 1 & 2 & 3 & 4 \\ 0 & 1 & 2 & 3 \\ 0 & 0 & 1 & 2 \end{bmatrix}\]

Already in REF. Three non-zero rows β†’ $\text{rank}(D) = 3$. Full row rank ($3 \times 4$ matrix, max rank is 3).


6.3 The Rank-Nullity Theorem

Rank-Nullity Theorem: For an $m \times n$ matrix $A$:

\[\boxed{\text{rank}(A) + \text{nullity}(A) = n}\]

where:

  • $\text{rank}(A)$ = dimension of column space = number of pivot columns
  • $\text{nullity}(A)$ = dimension of null space = number of free variables
  • $n$ = total number of columns

Physical Meaning:

\[\underbrace{\text{Independent info (rank)}}_{\text{pivot variables}} + \underbrace{\text{Redundant info (nullity)}}_{\text{free variables}} = \underbrace{\text{Total info}}_{\text{all variables}}\]

Example: For the following matrix, row reduce:

\[A = \begin{bmatrix} 1 & 2 & 3 \\ 2 & 4 & 6 \end{bmatrix}\] \[\begin{bmatrix} 1 & 2 & 3 \\ 0 & 0 & 0 \end{bmatrix}\]
  • $\text{rank}(A) = 1$ (one pivot)
  • $\text{nullity}(A) = 3 - 1 = 2$ (two free variables)
  • Verification: $1 + 2 = 3 = n$ βœ“

6.4 Rank & System Solvability

Rank Conditions for $Ax = b$:

Condition Type of Solution
$\text{rank}(A) < \text{rank}([A \mid b])$ No solution (inconsistent)
$\text{rank}(A) = \text{rank}([A \mid b]) = n$ Unique solution
$\text{rank}(A) = \text{rank}([A \mid b]) < n$ Infinitely many solutions

Full rank properties:

  • Full Column Rank ($\text{rank} = n$, requires $m \geq n$): $Ax = b$ has at most one solution.
  • Full Row Rank ($\text{rank} = m$, requires $n \geq m$): $Ax = b$ has a solution for every $b$.
  • Full Rank (square, $\text{rank} = m = n$): $Ax = b$ has exactly one solution for every $b$; $A$ is invertible.

Full Column Rank example ($3 \times 2$, rank $= 2 = n$):

\[A = \begin{bmatrix} 1 & 0 \\ 0 & 1 \\ 0 & 0 \end{bmatrix}\]

Both columns are pivot columns β†’ no free variables β†’ at most one solution. For $b = (1, 2, 0)^T$: unique solution $x = (1, 2)^T$. For $b = (1, 2, 5)^T$: no solution (last row gives $0 = 5$).

Full Row Rank example ($2 \times 3$, rank $= 2 = m$):

\[A = \begin{bmatrix} 1 & 0 & 2 \\ 0 & 1 & 3 \end{bmatrix}\]

Every row has a pivot β†’ system is consistent for every $b$. But column 3 is free β†’ infinitely many solutions. E.g., $b = (4, 6)^T$: $x_3 = t$, $x_1 = 4 - 2t$, $x_2 = 6 - 3t$.

Full Rank example ($2 \times 2$, rank $= 2 = m = n$):

\[A = \begin{bmatrix} 1 & 3 \\ 2 & 7 \end{bmatrix}\]

$\det(A) = 7 - 6 = 1 \neq 0$, so $A$ is invertible. For any $b$, there is exactly one solution: $x = A^{-1}b$.


6.5 Row Space & Column Space

Row Space: Let $A$ be an $m \times n$ matrix with rows $R_1, R_2, \ldots, R_m \in \mathbb{R}^n$. The row space of $A$ is:

\[\text{rowsp}(A) = \text{Span}\lbrace R_1, R_2, \ldots, R_m \rbrace \subseteq \mathbb{R}^n\]

Column Space: The column space of $A$ is the span of the columns of $A$:

\[\text{colsp}(A) = \text{Span}\lbrace C_1, C_2, \ldots, C_n \rbrace \subseteq \mathbb{R}^m\]

Note: $\text{colsp}(A) = \text{rowsp}(A^T)$.

Theorem: Row-equivalent matrices have the same row space.

Theorem: $\dim(\text{rowsp}(A)) = \dim(\text{colsp}(A)) = \rho(A)$.

That is, the maximum number of linearly independent rows equals the maximum number of linearly independent columns.

How to find basis of row space: Row-reduce $A$; the non-zero rows of the echelon form are a basis.

How to find basis of column space: Row-reduce $A$; the pivot columns of the original matrix $A$ form a basis (not the echelon form columns!).

Example 5: Finding Basis of Row Space and Column Space

Problem: Find basis of row space and column space of:

\[A = \begin{bmatrix} 1 & 3 & 1 & 2 \\ 2 & 1 & -2 & 3 \\ 1 & -2 & -3 & 3 \\ 3 & 1 & 0 & -6 \end{bmatrix}\]

Row reduce:

\[A \to \begin{bmatrix} 1 & 3 & 1 & 2 \\ 0 & -5 & -4 & -1 \\ 0 & 0 & 0 & 2 \\ 0 & 0 & 0 & 0 \end{bmatrix}\]

$\rho(A) = 3$.

Basis of row space: $\lbrace(1, 3, 1, 2),\; (0, -5, -4, -1),\; (0, 0, 0, 2)\rbrace$ β€” the non-zero rows of the echelon form.

Basis of column space: Pivot columns are $C_1, C_2, C_4$ (columns 1, 2, 4). Take these columns from the original matrix $A$:

\[\text{Basis of colsp}(A) = \left\lbrace \begin{pmatrix}1\\2\\1\\3\end{pmatrix}, \begin{pmatrix}3\\1\\-2\\1\end{pmatrix}, \begin{pmatrix}2\\3\\3\\-6\end{pmatrix} \right\rbrace\]

Example 6: Finding Basis of a Subspace β€” Extend to Basis of $\mathbb{R}^4$

Problem: Let $W$ be the subspace of $\mathbb{R}^4$ spanned by $\vec{u}_1 = (1, -2, 5, -3)$, $\vec{u}_2 = (2, 3, 1, -4)$, $\vec{u}_3 = (3, 8, -3, -5)$.

(i) Find a basis and dimension of $W$.

(ii) Extend the basis of $W$ to a basis of $\mathbb{R}^4$.

Part (i): Form the matrix with these as rows and row-reduce:

\[\begin{bmatrix} 1 & -2 & 5 & -3 \\ 2 & 3 & 1 & -4 \\ 3 & 8 & -3 & -5 \end{bmatrix} \to \begin{bmatrix} 1 & -2 & 5 & -3 \\ 0 & 7 & -9 & 2 \\ 0 & 0 & 0 & 0 \end{bmatrix}\]

Two non-zero rows $\Rightarrow$ $\dim(W) = 2$.

Basis of $W$: $\lbrace(1, -2, 5, -3),\; (0, 7, -9, 2)\rbrace$.

Part (ii): To extend to a basis of $\mathbb{R}^4$, we need 2 more vectors. Append the standard basis vectors $\vec{e}_1, \vec{e}_2, \vec{e}_3, \vec{e}_4$ and check which ones are independent of $W$:

\[\begin{bmatrix} 1 & -2 & 5 & -3 \\ 0 & 7 & -9 & 2 \\ 1 & 0 & 0 & 0 \\ 0 & 0 & 1 & 0 \end{bmatrix}\]

Row-reduce to verify this has rank 4. Indeed it does.

Basis of $\mathbb{R}^4$: $\lbrace(1, -2, 5, -3),\; (0, 7, -9, 2),\; (0, 0, 1, 0),\; (0, 0, 0, 1)\rbrace$.


Practice Questions β€” Rank

Q1. Find the rank of:

\[A = \begin{bmatrix} 1 & 2 & 3 \\ 0 & 1 & 2 \\ 1 & 3 & 5 \end{bmatrix}\]

(Ans: rank = 2)

Click for solution

$R_3 \leftarrow R_3 - R_1$: $[0, 1, 2]$. Now $R_3 = R_2$, so $R_3 \leftarrow R_3 - R_2 = [0, 0, 0]$.

\[\text{REF: }\begin{bmatrix} 1 & 2 & 3 \\ 0 & 1 & 2 \\ 0 & 0 & 0 \end{bmatrix}\]

2 non-zero rows β†’ rank = 2. Nullity $= 3 - 2 = 1$.

Q2. Determine if $Ax = b$ has a unique, infinitely many, or no solution:

\[A = \begin{bmatrix} 1 & 2 \\ 2 & 4 \\ 3 & 6 \end{bmatrix}, \quad b = \begin{bmatrix} 1 \\ 2 \\ 3 \end{bmatrix}\]
Click for solution

Augmented matrix:

\[\left[\begin{array}{cc|c} 1&2&1 \\ 2&4&2 \\ 3&6&3 \end{array}\right]\]

$R_2 - 2R_1 = [0,0,0]$, $R_3 - 3R_1 = [0,0,0]$.

rank$(A) = 1 =$ rank$([A\vert b])$, so the system is consistent. Free variable $x_2 = t$, $x_1 = 1 - 2t$.

Infinitely many solutions: $(x_1, x_2) = (1 - 2t, \; t)$.

Q3. Find rank and nullity of a $3 \times 5$ matrix with 2 pivots. (Ans: rank = 2, nullity = 3)

Click for solution

Rank = number of pivots = 2. By the Rank-Nullity theorem: rank + nullity = $n$ (number of columns).

Nullity = $5 - 2 = 3$. (3 free variables in the solution to $A\vec{x} = \vec{0}$.)

Q4. Can a $4 \times 3$ matrix have rank 4? Why or why not? (Ans: No β€” max rank is $\min(4,3) = 3$.)

Click for solution

No. Rank $\leq \min(m, n) = \min(4, 3) = 3$. There are only 3 columns, so at most 3 can be independent (each pivot occupies one column).

Q5. Find the rank of:

\[B = \begin{bmatrix} 1 & 0 & 2 & 0 \\ 0 & 1 & 3 & 0 \\ 0 & 0 & 0 & 1 \end{bmatrix}\]

(Ans: rank = 3)

Click for solution

$B$ is already in REF. Pivots are in columns 1, 2, and 4. Three non-zero rows β†’ rank = 3.

Column 3 ($x_3$) is a free variable. Nullity $= 4 - 3 = 1$.

Q6. If rank$(A) = 3$ and $A$ is $5 \times 7$, find nullity$(A)$.

Click for solution

By Rank-Nullity: nullity = $n -$ rank $= 7 - 3 = 4$.

The null space is 4-dimensional β€” the homogeneous system $A\vec{x} = \vec{0}$ has 4 free variables.

Q7. True or False: If $A$ is a $3 \times 3$ matrix with $\det(A) = 0$, then $\text{rank}(A) < 3$.

Click for solution

True. $\det(A) = 0 \Leftrightarrow A$ is singular $\Leftrightarrow$ rank$(A) < n = 3$. The columns are dependent, so rank is at most 2.

Q8. Find the rank of the zero matrix $O_{3 \times 4}$. (Ans: 0)

Click for solution

No non-zero rows in REF (it’s already all zeros). Rank = 0, nullity = 4.

The only matrix with rank 0 is the zero matrix.

Q9. If $Ax = 0$ has only the trivial solution, what is the relationship between rank and $n$?

Click for solution

If $A\vec{x} = \vec{0}$ has only the trivial solution, then nullity = 0.

By Rank-Nullity: rank = $n$ (full column rank). Every column is a pivot column.

Q10. A $6 \times 6$ matrix has nullity 2. What is its rank? Is it invertible?

Click for solution

Rank $= 6 - 2 = 4$. Since rank $< 6 = n$, the matrix is not invertible ($\det = 0$).

For a square matrix to be invertible, we need rank $= n$ (equivalently, nullity $= 0$).


7 β€” Inner Product & Gram-Schmidt Process


7.1 Inner Product (Dot Product)

Definition: The standard inner product (dot product) on $\mathbb{R}^n$ is:

\[\langle \vec{u}, \vec{v} \rangle = \vec{u} \cdot \vec{v} = u_1v_1 + u_2v_2 + \cdots + u_nv_n = \sum_{i=1}^{n} u_iv_i\]

Example: $\vec{u} = (1, 2, 3), \vec{v} = (4, -1, 2)$

\[\langle \vec{u}, \vec{v} \rangle = 1(4) + 2(-1) + 3(2) = 4 - 2 + 6 = 8\]

Norm (length) of a vector:

\[\lVert\vec{v}\rVert = \sqrt{\langle \vec{v}, \vec{v} \rangle} = \sqrt{v_1^2 + v_2^2 + \cdots + v_n^2}\]

$\lVert(3, 4)\rVert = \sqrt{9 + 16} = 5$

$\lVert(1, 1, 1)\rVert = \sqrt{3}$


7.2 Properties of Inner Product

An inner product $\langle \cdot, \cdot \rangle$ on a vector space $V$ satisfies:

  1. Symmetry: $\langle \vec{u}, \vec{v} \rangle = \langle \vec{v}, \vec{u} \rangle$
  2. Linearity (first argument): $\langle a\vec{u} + b\vec{w}, \vec{v} \rangle = a\langle \vec{u}, \vec{v} \rangle + b\langle \vec{w}, \vec{v} \rangle$
  3. Positive Definiteness: $\langle \vec{v}, \vec{v} \rangle \geq 0$, with equality iff $\vec{v} = \vec{0}$

Important inequalities:

Cauchy-Schwarz Inequality:

\[\lvert\langle \vec{u}, \vec{v} \rangle\rvert \leq \lVert\vec{u}\rVert \cdot \lVert\vec{v}\rVert\]

Triangle Inequality:

\[\lVert\vec{u} + \vec{v}\rVert \leq \lVert\vec{u}\rVert + \lVert\vec{v}\rVert\]

7.3 Orthogonality & Orthonormality

Orthogonal: Two vectors are orthogonal if $\langle \vec{u}, \vec{v} \rangle = 0$ (perpendicular).

Orthogonal Set: A set $\lbrace\vec{v}_1, \ldots, \vec{v}_k\rbrace$ where $\langle \vec{v}_i, \vec{v}_j \rangle = 0$ for all $i \neq j$.

Orthonormal Set: An orthogonal set where every vector has unit norm: $\lVert\vec{v}_i\rVert = 1$.

Example:

  • $(1, 0)$ and $(0, 1)$ are orthonormal: dot product = 0, both have norm 1. βœ“
  • $(1, 1)$ and $(1, -1)$ are orthogonal: $1(1) + 1(-1) = 0$. βœ“ (But not orthonormal β€” need to normalize.)
  • To normalize $(1,1)$: $\frac{(1,1)}{\lVert(1,1)\rVert} = \frac{(1,1)}{\sqrt{2}} = \left(\frac{1}{\sqrt{2}}, \frac{1}{\sqrt{2}}\right)$

Why do we care about orthonormal bases?

  • Coordinates are easy: $c_i = \langle \vec{v}, \vec{u}_i \rangle$ (just take dot product!)
  • Computations are simpler and more numerically stable
  • Foundation of Fourier analysis, signal processing, quantum mechanics

Projection formula: The projection of $\vec{v}$ onto a unit vector $\vec{u}$:

\[\text{proj}_{\vec{u}}(\vec{v}) = \langle \vec{v}, \vec{u} \rangle \, \vec{u}\]

If $\vec{u}$ is not unit length:

\[\text{proj}_{\vec{u}}(\vec{v}) = \frac{\langle \vec{v}, \vec{u} \rangle}{\langle \vec{u}, \vec{u} \rangle} \, \vec{u}\]

7.4 The Gram-Schmidt Process β€” Algorithm

7.4.1 Why Do We Study Gram-Schmidt?

The Core Problem: You have a basis for a vector space β€” great, you can represent any vector. But that basis might be messy: the vectors could point in similar directions, overlap in complicated ways, or have different lengths. Working with such a basis makes computations painful.

What Gram-Schmidt solves: It takes any basis and converts it into an orthonormal basis β€” one where every vector is perpendicular to every other and has unit length.

Concrete Example β€” Messy vs. Orthonormal Basis

Consider $\mathbb{R}^2$. Suppose you have the basis:

\[B = \left\lbrace \vec{v}_1 = \begin{pmatrix} 2 \\ 1 \end{pmatrix}, \; \vec{v}_2 = \begin{pmatrix} 3 \\ 2 \end{pmatrix} \right\rbrace\]

These vectors are not perpendicular ($\vec{v}_1 \cdot \vec{v}_2 = 6 + 2 = 8 \neq 0$) and not unit length ($\lVert \vec{v}_1 \rVert = \sqrt{5}$, $\lVert \vec{v}_2 \rVert = \sqrt{13}$).

Task: Express $\vec{v} = (5, 3)$ in this basis.

With the messy basis $B$, you must solve the system:

\[2c_1 + 3c_2 = 5, \quad c_1 + 2c_2 = 3\]

Row reducing: $c_1 = 1, \, c_2 = 1$. That took real work (setting up equations, elimination).

Now compare with an orthonormal basis. The standard basis is:

\[E = \left\lbrace \vec{e}_1 = \begin{pmatrix} 1 \\ 0 \end{pmatrix}, \; \vec{e}_2 = \begin{pmatrix} 0 \\ 1 \end{pmatrix} \right\rbrace\]

which is orthonormal.

The coordinates are instant β€” just dot products:

\[c_1 = \vec{v} \cdot \vec{e}_1 = 5, \quad c_2 = \vec{v} \cdot \vec{e}_2 = 3\]

No equations to solve. This is the power of orthonormal bases.

Now imagine doing this not in $\mathbb{R}^2$ but in $\mathbb{R}^{100}$ or an infinite-dimensional function space β€” the messy approach requires solving 100Γ—100 systems, while the orthonormal approach is still just dot products!

Why is an orthonormal basis so valuable?

1. Coordinates become trivial. Given an orthonormal basis $\lbrace\vec{u}_1, \ldots, \vec{u}_n\rbrace$, finding the coordinate of any vector $\vec{v}$ is just a dot product:

\[c_i = \langle \vec{v}, \vec{u}_i \rangle\]

No systems of equations to solve! Compare this to a non-orthogonal basis where you must solve $A\vec{c} = \vec{v}$ every time.

2. Projection is clean. The best approximation of $\vec{v}$ in a subspace $W$ with orthonormal basis $\lbrace\vec{u}_1, \ldots, \vec{u}_k\rbrace$ is:

\[\text{proj}_W(\vec{v}) = \sum_{i=1}^{k} \langle \vec{v}, \vec{u}_i \rangle \, \vec{u}_i\]

This is the foundation of least squares approximation, Fourier series, and signal processing.

3. Matrices become simple. When a matrix is expressed in an orthonormal basis, its transpose equals its inverse ($Q^T = Q^{-1}$). This makes solving $Q\vec{x} = \vec{b}$ trivial: $\vec{x} = Q^T\vec{b}$.

4. Numerical stability. Orthonormal bases avoid the rounding errors that plague computations with nearly-parallel vectors.

Real-world applications:

  • QR factorization in numerical linear algebra (solving systems, computing eigenvalues)
  • Fourier analysis β€” Fourier series are essentially Gram-Schmidt applied to $\lbrace1, \cos x, \sin x, \cos 2x, \ldots\rbrace$
  • Legendre polynomials β€” Gram-Schmidt applied to $\lbrace1, x, x^2, \ldots\rbrace$ with appropriate inner products
  • Quantum mechanics β€” state vectors live in inner product spaces; measurements require orthonormal bases
  • Machine learning β€” PCA, dimensionality reduction, and feature decorrelation

7.4.2 The Algorithm β€” Step by Step

The Gram-Schmidt Algorithm: Given linearly independent vectors $\lbrace\vec{v}_1, \vec{v}_2, \ldots, \vec{v}_n\rbrace$, produce an orthonormal basis $\lbrace\vec{u}_1, \vec{u}_2, \ldots, \vec{u}_n\rbrace$:

Step 1 β€” Normalize first vector:

\[\vec{u}_1 = \frac{\vec{v}_1}{\lVert\vec{v}_1\rVert}\]

Step $k$ (for $k = 2, 3, \ldots, n$) β€” Remove projections, then normalize:

\[\vec{w}_k = \vec{v}_k - \sum_{i=1}^{k-1} \langle \vec{v}_k, \vec{u}_i \rangle \, \vec{u}_i\] \[\vec{u}_k = \frac{\vec{w}_k}{\lVert\vec{w}_k\rVert}\]

7.4.3 Deep Dive β€” What Does Each Step Actually Do?

The formula looks compact, but let’s unpack the logic behind every part.

Step 1 β€” Why just normalize?

The first vector $\vec{v}_1$ has no predecessors to conflict with. It already points in a legitimate direction β€” we just scale it to unit length so it satisfies the β€œnormal” part of β€œorthonormal.”

Step 2 β€” The key idea (making $\vec{v}_2$ perpendicular to $\vec{u}_1$):

Think of $\vec{v}_2$ as having two components:

  • A part along $\vec{u}_1$ (the shadow of $\vec{v}_2$ on $\vec{u}_1$)
  • A part perpendicular to $\vec{u}_1$ (the genuinely new direction)
\[\vec{v}_2 = \underbrace{\langle \vec{v}_2, \vec{u}_1 \rangle \, \vec{u}_1}_{\text{component along } \vec{u}_1} + \underbrace{\vec{w}_2}_{\text{component } \perp \text{ to } \vec{u}_1}\]

Rearranging:

\[\vec{w}_2 = \vec{v}_2 - \langle \vec{v}_2, \vec{u}_1 \rangle \, \vec{u}_1\]

This is just: original vector minus its projection onto $\vec{u}_1$. What’s left ($\vec{w}_2$) is guaranteed to be perpendicular to $\vec{u}_1$.

Geometric picture (2D): Imagine $\vec{u}_1$ points east. $\vec{v}_2$ points northeast. The projection of $\vec{v}_2$ onto $\vec{u}_1$ captures the β€œeastward” part. Subtracting it leaves only the β€œnorthward” part β€” which is perpendicular to east.

         vβ‚‚ (northeast)
        /|
       / |
      /  | ← wβ‚‚ (the perpendicular part = north)
     /   |
    /    |
   u₁────+── projection of vβ‚‚ onto u₁ (east)

Step $k$ β€” The general pattern:

By step $k$, we already have orthonormal vectors $\vec{u}_1, \ldots, \vec{u}_{k-1}$. The new vector $\vec{v}_k$ might have components along each of them. So we subtract all such components:

\[\vec{w}_k = \vec{v}_k - \underbrace{\langle \vec{v}_k, \vec{u}_1 \rangle \, \vec{u}_1}_{\text{remove } \vec{u}_1 \text{ component}} - \underbrace{\langle \vec{v}_k, \vec{u}_2 \rangle \, \vec{u}_2}_{\text{remove } \vec{u}_2 \text{ component}} - \cdots - \underbrace{\langle \vec{v}_k, \vec{u}_{k-1} \rangle \, \vec{u}_{k-1}}_{\text{remove } \vec{u}_{k-1} \text{ component}}\]

After subtracting all projections, $\vec{w}_k$ is perpendicular to every previous $\vec{u}_i$. Then normalize to get $\vec{u}_k$.

Why does $\vec{w}_k \neq \vec{0}$? Because our input vectors are linearly independent. If $\vec{v}_k$ were a linear combination of $\vec{v}_1, \ldots, \vec{v}_{k-1}$, then after removing all projections, nothing would remain. Linear independence guarantees something new survives at every step.

Verification that it works: After step $k$, we can check:

\[\langle \vec{w}_k, \vec{u}_j \rangle = \langle \vec{v}_k, \vec{u}_j \rangle - \sum_{i=1}^{k-1} \langle \vec{v}_k, \vec{u}_i \rangle \underbrace{\langle \vec{u}_i, \vec{u}_j \rangle}_{=\,\delta_{ij}} = \langle \vec{v}_k, \vec{u}_j \rangle - \langle \vec{v}_k, \vec{u}_j \rangle = 0 \quad \checkmark\]

The $\delta_{ij}$ (Kronecker delta, equals 1 when $i=j$ and 0 otherwise) appears because the $\vec{u}_i$ are already orthonormal to each other. This is why the formula is so clean.

Summary of the logic:

  1. Keep: Take the original vector $\vec{v}_k$
  2. Subtract: Remove the projection onto each previously constructed orthonormal vector
  3. What remains: The part of $\vec{v}_k$ genuinely perpendicular to all previous directions
  4. Normalize: Scale to unit length
  5. Repeat: Until all vectors are processed

The process preserves the span at every step:

\[\text{span}\lbrace\vec{u}_1, \ldots, \vec{u}_k\rbrace = \text{span}\lbrace\vec{v}_1, \ldots, \vec{v}_k\rbrace\]

7.5 Worked Examples β€” Gram-Schmidt

Example 1: Standard vectors in $\mathbb{R}^3$

Problem: Apply Gram-Schmidt to $\lbrace\vec{v}_1 = (1,0,0), \; \vec{v}_2 = (1,1,0), \; \vec{v}_3 = (1,1,1)\rbrace$.

Step 1:

\[\vec{u}_1 = \frac{(1,0,0)}{1} = (1, 0, 0)\]

Step 2:

\[\langle \vec{v}_2, \vec{u}_1 \rangle = 1 \cdot 1 + 1 \cdot 0 + 0 \cdot 0 = 1\] \[\vec{w}_2 = (1,1,0) - 1 \cdot (1,0,0) = (0, 1, 0)\] \[\vec{u}_2 = \frac{(0,1,0)}{1} = (0, 1, 0)\]

Step 3:

\[\langle \vec{v}_3, \vec{u}_1 \rangle = 1, \quad \langle \vec{v}_3, \vec{u}_2 \rangle = 1\] \[\vec{w}_3 = (1,1,1) - 1 \cdot (1,0,0) - 1 \cdot (0,1,0) = (0, 0, 1)\] \[\vec{u}_3 = (0, 0, 1)\]

Orthonormal basis: $\lbrace(1,0,0), (0,1,0), (0,0,1)\rbrace$ β€” the standard basis!

(This makes sense: the input vectors were already β€œstaircase-shaped.”)

Example 2: Non-trivial vectors in $\mathbb{R}^3$

Problem: Apply Gram-Schmidt to $\lbrace\vec{v}_1 = (1, 1, 0), \; \vec{v}_2 = (1, 0, 1)\rbrace$.

Step 1:

\[\lVert\vec{v}_1\rVert = \sqrt{1^2 + 1^2 + 0^2} = \sqrt{2}\] \[\vec{u}_1 = \frac{(1,1,0)}{\sqrt{2}} = \left(\frac{1}{\sqrt{2}}, \frac{1}{\sqrt{2}}, 0\right)\]

Step 2:

\[\langle \vec{v}_2, \vec{u}_1 \rangle = 1 \cdot \frac{1}{\sqrt{2}} + 0 \cdot \frac{1}{\sqrt{2}} + 1 \cdot 0 = \frac{1}{\sqrt{2}}\] \[\vec{w}_2 = (1, 0, 1) - \frac{1}{\sqrt{2}} \cdot \left(\frac{1}{\sqrt{2}}, \frac{1}{\sqrt{2}}, 0\right) = (1, 0, 1) - \left(\frac{1}{2}, \frac{1}{2}, 0\right) = \left(\frac{1}{2}, -\frac{1}{2}, 1\right)\] \[\lVert\vec{w}_2\rVert = \sqrt{\frac{1}{4} + \frac{1}{4} + 1} = \sqrt{\frac{3}{2}} = \frac{\sqrt{6}}{2}\] \[\vec{u}_2 = \frac{1}{\frac{\sqrt{6}}{2}} \left(\frac{1}{2}, -\frac{1}{2}, 1\right) = \left(\frac{1}{\sqrt{6}}, -\frac{1}{\sqrt{6}}, \frac{2}{\sqrt{6}}\right)\]

Verification:

  • $\langle \vec{u}_1, \vec{u}_2 \rangle = \frac{1}{\sqrt{2}} \cdot \frac{1}{\sqrt{6}} + \frac{1}{\sqrt{2}} \cdot \left(-\frac{1}{\sqrt{6}}\right) + 0 = 0$ βœ“ (orthogonal)
  • $\lVert\vec{u}_1\rVert = 1, \lVert\vec{u}_2\rVert = 1$ βœ“ (unit length)

Example 3: Gram-Schmidt for Polynomials β€” A Deep Treatment

Why apply Gram-Schmidt to polynomials?

The standard basis $\lbrace1, x, x^2, x^3, \ldots\rbrace$ for polynomial spaces is simple, but it’s not orthogonal under any natural inner product. As degree increases, the monomials $x^n$ become increasingly β€œsimilar” (they all crowd toward the same shape on $[0,1]$). This causes severe numerical problems in approximation and interpolation.

Gram-Schmidt on polynomials produces orthogonal polynomials β€” families like Legendre, Chebyshev, Hermite, and Laguerre polynomials β€” which are cornerstones of:

  • Numerical integration (Gaussian quadrature)
  • Fourier-like series for polynomial approximation
  • Solving differential equations (Sturm-Liouville theory)
  • Physics β€” quantum harmonic oscillator (Hermite), angular momentum (Legendre)

The key idea: polynomials are vectors too. The space $P_n$ of polynomials of degree $\leq n$ is a vector space (you can add polynomials and scale them). All the theory of inner products, orthogonality, and Gram-Schmidt applies β€” you just need to define what β€œinner product” means for functions.

Inner product for polynomial (function) spaces:

For polynomial spaces, the inner product is defined by integration:

\[\langle f, g \rangle = \int_a^b f(x) \, g(x) \, dx\]

This replaces the dot product’s β€œmultiply corresponding components and sum” with β€œmultiply functions pointwise and integrate.” Integration is a continuous version of summation β€” the analogy is:

Vectors in $\mathbb{R}^n$ Functions in $P_n$
$\langle \vec{u}, \vec{v} \rangle = \sum u_i v_i$ $\langle f, g \rangle = \int f(x) g(x) \, dx$
entries indexed by $i = 1, \ldots, n$ β€œentries” indexed continuously by $x \in [a,b]$
two vectors perpendicular when dot product $= 0$ two functions orthogonal when integral $= 0$

This inner product satisfies all three axioms (symmetry, linearity, positive-definiteness), so Gram-Schmidt works exactly the same way.

What Does β€œMagnitude of a Polynomial” Mean?

The norm (magnitude) of a polynomial is defined the same way as for ordinary vectors β€” through the inner product:

\[\lVert p \rVert = \sqrt{\langle p, p \rangle}\]

For vectors in $\mathbb{R}^n$, this gives the familiar formula: $\lVert \vec{v} \rVert = \sqrt{v_1^2 + v_2^2 + \cdots + v_n^2}$

For polynomials with the integral inner product, plugging in $f = g = p$:

\[\lVert p \rVert^2 = \langle p, p \rangle = \int_a^b [p(x)]^2 \, dx\] \[\boxed{\lVert p \rVert = \sqrt{\int_a^b [p(x)]^2 \, dx}}\]

Why does this make sense as a β€œsize”?

Think of what $\int_a^b [p(x)]^2 \, dx$ actually measures. The function $[p(x)]^2$ is always $\geq 0$, and the integral adds up all those squared values across $[a,b]$. So:

  • If $p(x)$ is close to zero everywhere on $[a,b]$, the integral is small β†’ small norm (small polynomial)
  • If $p(x)$ takes large values on $[a,b]$, the integral is large β†’ large norm (big polynomial)
  • If $p(x) = 0$ everywhere, the integral is exactly $0$ β†’ zero norm (only the zero polynomial has norm 0)

It’s essentially measuring the β€œtotal energy” or β€œoverall size” of the polynomial on the interval. This is the same idea as the RMS (root-mean-square) value in engineering.

Concrete examples to build intuition:

Polynomial $p(x)$ Interval $\lVert p \rVert^2 = \int [p(x)]^2 dx$ $\lVert p \rVert$
$p(x) = 1$ $[0,1]$ $\int_0^1 1^2 \, dx = 1$ $1$
$p(x) = x$ $[0,1]$ $\int_0^1 x^2 \, dx = \frac{1}{3}$ $\frac{1}{\sqrt{3}}$
$p(x) = x - \frac{1}{2}$ $[0,1]$ $\int_0^1 (x-\frac{1}{2})^2 dx = \frac{1}{12}$ $\frac{1}{2\sqrt{3}}$
$p(x) = x^2$ $[0,1]$ $\int_0^1 x^4 \, dx = \frac{1}{5}$ $\frac{1}{\sqrt{5}}$
$p(x) = 3$ $[0,1]$ $\int_0^1 9 \, dx = 9$ $3$

Key difference from vector magnitude:

For a vector $(3, 4)$: $\lVert(3,4)\rVert = \sqrt{3^2 + 4^2} = 5$ β€” you square each component and sum.

For a polynomial $p(x)$: $\lVert p \rVert = \sqrt{\int [p(x)]^2 dx}$ β€” you square the function at every point and integrate (continuous sum).

The logic is identical: square, add up, take square root. The only difference is that vectors have finitely many components while functions have infinitely many β€œcomponents” (one for each $x$), so the sum becomes an integral.


Example 3a: Gram-Schmidt on $\lbrace1, x\rbrace$ over $[0, 1]$

Problem: Apply Gram-Schmidt to $\lbrace1, x\rbrace$ in $P_1$ with inner product $\langle f, g \rangle = \int_0^1 f(x)g(x) \, dx$.

Why are $1$ and $x$ not already orthogonal on $[0,1]$?

\[\langle 1, x \rangle = \int_0^1 1 \cdot x \, dx = \frac{1}{2} \neq 0\]

So $1$ and $x$ are NOT orthogonal β€” the constant function and the linear function have β€œoverlap” on $[0,1]$.

Step 1 β€” Start with $f_1(x) = 1$, normalize:

\[\lVert 1 \rVert^2 = \langle 1, 1 \rangle = \int_0^1 1 \cdot 1 \, dx = 1 \quad \Rightarrow \quad \lVert 1 \rVert = 1\] \[p_1(x) = \frac{1}{1} = 1\]

Step 2 β€” Make $x$ orthogonal to $p_1 = 1$:

Compute the projection of $x$ onto $p_1 = 1$:

\[\langle x, p_1 \rangle = \langle x, 1 \rangle = \int_0^1 x \, dx = \frac{1}{2}\]

Subtract the projection:

\[q(x) = x - \langle x, p_1 \rangle \cdot p_1 = x - \frac{1}{2}\]

What is $q(x) = x - \frac{1}{2}$ geometrically? It’s the part of $x$ that the constant function $1$ cannot capture. It’s a line that passes through zero at $x = \frac{1}{2}$ β€” the β€œpurely varying” part of $x$ after removing its average value on $[0,1]$.

Verify orthogonality: $\langle q, p_1 \rangle = \int_0^1 (x - \frac{1}{2}) \cdot 1 \, dx = \left[\frac{x^2}{2} - \frac{x}{2}\right]_0^1 = (\frac{1}{2} - \frac{1}{2}) - 0 = 0 \; \checkmark$

Normalize:

\[\lVert q \rVert^2 = \int_0^1 \left(x - \frac{1}{2}\right)^2 dx = \int_0^1 \left(x^2 - x + \frac{1}{4}\right) dx = \frac{1}{3} - \frac{1}{2} + \frac{1}{4} = \frac{1}{12}\] \[\lVert q \rVert = \frac{1}{2\sqrt{3}}\] \[p_2(x) = \frac{q(x)}{\lVert q \rVert} = \frac{x - 1/2}{1/(2\sqrt{3})} = 2\sqrt{3}\left(x - \frac{1}{2}\right)\]

Orthonormal basis for $P_1$ on $[0,1]$: $\left\lbrace 1, \; 2\sqrt{3}\left(x - \frac{1}{2}\right) \right\rbrace$


Example 3b: Gram-Schmidt on $\lbrace1, x, x^2\rbrace$ over $[0, 1]$ β€” Building Orthogonal Polynomials

Problem: Apply Gram-Schmidt to $\lbrace1, x, x^2\rbrace$ in $P_2$ with $\langle f, g \rangle = \int_0^1 f(x)g(x) \, dx$.

This produces the (shifted) Legendre polynomials on $[0,1]$ β€” one of the most important polynomial families in mathematics.

Step 1: $p_1(x) = 1$ (same as before, $\lVert 1 \rVert = 1$)

Step 2: $p_2(x) = 2\sqrt{3}\left(x - \frac{1}{2}\right)$ (same as Example 3a)

Step 3 β€” Make $x^2$ orthogonal to both $p_1$ and $p_2$:

Compute projections of $x^2$ onto each existing basis vector:

\[\langle x^2, p_1 \rangle = \int_0^1 x^2 \cdot 1 \, dx = \frac{1}{3}\] \[\langle x^2, p_2 \rangle = \int_0^1 x^2 \cdot 2\sqrt{3}\left(x - \frac{1}{2}\right) dx = 2\sqrt{3} \int_0^1 \left(x^3 - \frac{x^2}{2}\right) dx\] \[= 2\sqrt{3}\left(\frac{1}{4} - \frac{1}{6}\right) = 2\sqrt{3} \cdot \frac{1}{12} = \frac{\sqrt{3}}{6}\]

Subtract both projections:

\[q(x) = x^2 - \frac{1}{3} \cdot 1 - \frac{\sqrt{3}}{6} \cdot 2\sqrt{3}\left(x - \frac{1}{2}\right)\] \[= x^2 - \frac{1}{3} - \frac{3}{6}\cdot\left(2x - 1\right) \cdot \frac{1}{\,1\,} = x^2 - \frac{1}{3} - (x - \frac{1}{2})\]

Wait β€” let’s be more careful:

\[\frac{\sqrt{3}}{6} \cdot 2\sqrt{3}\left(x - \frac{1}{2}\right) = \frac{2 \cdot 3}{6}\left(x - \frac{1}{2}\right) = \left(x - \frac{1}{2}\right)\] \[q(x) = x^2 - \frac{1}{3} - x + \frac{1}{2} = x^2 - x + \frac{1}{6}\]

Verify: $\langle q, p_1 \rangle = \int_0^1 (x^2 - x + \frac{1}{6}) dx = \frac{1}{3} - \frac{1}{2} + \frac{1}{6} = 0 \; \checkmark$

Normalize:

\[\lVert q \rVert^2 = \int_0^1 \left(x^2 - x + \frac{1}{6}\right)^2 dx = \frac{1}{180}\] \[\lVert q \rVert = \frac{1}{\sqrt{180}} = \frac{1}{6\sqrt{5}}\] \[p_3(x) = \frac{q(x)}{\lVert q \rVert} = 6\sqrt{5}\left(x^2 - x + \frac{1}{6}\right)\]

Orthonormal basis for $P_2$ on $[0,1]$:

\[\left\lbrace 1, \quad 2\sqrt{3}\left(x - \frac{1}{2}\right), \quad 6\sqrt{5}\left(x^2 - x + \frac{1}{6}\right) \right\rbrace\]

What did Gram-Schmidt actually accomplish here?

We started with $\lbrace1, x, x^2\rbrace$ β€” a perfectly fine basis for $P_2$, but these polynomials are far from orthogonal. Gram-Schmidt produced three polynomials that are completely orthogonal to each other under the integral inner product.

Notice the pattern:

  • $p_1 = 1$ β€” constant (degree 0)
  • $p_2 \propto x - \frac{1}{2}$ β€” linear, zero mean on $[0,1]$ (degree 1)
  • $p_3 \propto x^2 - x + \frac{1}{6}$ β€” quadratic, orthogonal to all lower-degree polynomials (degree 2)

Each orthogonal polynomial captures the variation that no lower-degree polynomial can. This is why orthogonal polynomials are so powerful for approximation β€” each term in the expansion adds genuinely new information.

If you continue this process to $\lbrace1, x, x^2, \ldots, x^n\rbrace$, you get the shifted Legendre polynomials on $[0,1]$. On $[-1,1]$, the same process yields the classical Legendre polynomials $P_0(x) = 1, \; P_1(x) = x, \; P_2(x) = \frac{1}{2}(3x^2 - 1), \ldots$

Visualizing β€œEach Orthogonal Polynomial Captures New Variation”

First β€” What Does β€œApproximating a Function” Mean?

Function approximation means replacing a complicated function $f(x)$ with a simpler one that is β€œclose enough.”

For example, $\sin(\pi x)$ is a transcendental function β€” it can’t be written as a finite polynomial. But we can find a polynomial $p(x) = a_0 + a_1 x + a_2 x^2$ that stays close to $\sin(\pi x)$ over some interval.

Why would we want this?

  • Computation: Polynomials are trivial to evaluate (just multiply and add). Sine, exponentials, and other complex functions are expensive. Calculators internally use polynomial approximations.
  • Analysis: Polynomials are easy to differentiate, integrate, and manipulate algebraically.
  • Data fitting: Given measured data points, we often want a simple formula (polynomial) that captures the overall trend.

What does β€œclose enough” mean?

We measure closeness using an inner product. On an interval $[a, b]$:

\[\text{Error} = \int_a^b \big[f(x) - p(x)\big]^2\, dx\]

A smaller integral means a better approximation. The best degree-$n$ approximation is the one that minimizes this integral β€” and orthogonal polynomials give us an elegant way to find it term by term.

Analogy: Suppose you want to describe someone’s location using only β€œnorth-south” and β€œeast-west.” That’s a 2D approximation of their actual 3D position. Adding β€œaltitude” gives a better approximation. Each new independent direction adds information that the previous ones couldn’t capture β€” just like each orthogonal polynomial adds a new β€œdirection” in function space.

To truly understand why orthogonal polynomials are special, let’s see them side-by-side with the ordinary basis and walk through the geometry step by step.


Step 1 β€” The Problem with the Standard Basis

Graph 1 β€” The original (non-orthogonal) basis $\lbrace1,\, x,\, x^2\rbrace$:

These three polynomials are a valid basis for $P_2$, but look how much they overlap β€” $x$ and $x^2$ both rise together on $[0,1]$. They share a lot of the same β€œinformation.”

What to notice in Graph 1:

  • The blue line ($x$) and green parabola ($x^2$) both start at $0$ and rise to $1$ β€” they move in the β€œsame direction” over most of $[0,1]$.
  • If you were approximating some function $f$ and first used $1$ and $x$, a big chunk of what $x^2$ does is redundant β€” it partially duplicates the upward trend that $x$ already described.
  • Mathematically: $\langle x, x^2 \rangle = \int_0^1 x \cdot x^2\, dx = \frac{1}{4} \neq 0$ β€” these are not orthogonal. Their inner product is nonzero, meaning they overlap.
  • Consequence: When you add $x^2$ to an approximation $a + bx$, the coefficients $a$ and $b$ must change to redistribute what each term is responsible for. The terms are entangled.

Step 2 β€” The Orthogonal Basis (After Gram-Schmidt)

Graph 2 β€” The orthogonal basis (unnormalized) from Gram-Schmidt:

  • $q_1 = 1$ β€” flat constant (captures the β€œoverall level”)
  • $q_2 = x - \frac{1}{2}$ β€” a line that is perfectly balanced: zero average on $[0,1]$ (captures the β€œtilt”)
  • $q_3 = x^2 - x + \frac{1}{6}$ β€” a curve that has zero average and zero linear tilt (captures pure β€œcurvature”)

Each polynomial has a completely different shape from the others. No amount of scaling/adding the red and blue curves can ever produce the green curve. That’s orthogonality!

What to notice in Graph 2:

  • Red ($q_1 = 1$): A flat horizontal line. It can only shift an approximation up or down β€” it captures the average height of a function.

  • Blue ($q_2 = x - \frac{1}{2}$): A line that goes from $-\frac{1}{2}$ to $+\frac{1}{2}$. Notice it’s perfectly balanced around zero β€” it spends equal time above and below the $x$-axis. This means $\int_0^1 q_2\, dx = 0$, so it’s orthogonal to the constant $q_1$. It captures whether a function tilts left-to-right β€” something a constant can never do.

  • Green ($q_3 = x^2 - x + \frac{1}{6}$): A U-shaped curve that dips below zero in the middle and curves up at both ends. Crucially:
    • $\int_0^1 q_3 \cdot 1\, dx = 0$ β€” it has zero average (orthogonal to $q_1$)
    • $\int_0^1 q_3 \cdot (x - \frac{1}{2})\, dx = 0$ β€” it has zero tilt (orthogonal to $q_2$)
    • It captures pure bowing/curvature β€” the U-shape that no combination of flat lines and tilted lines can produce.
  • Key observation: Compare this graph with Graph 1. In Graph 1, the blue and green curves both rise together. Here, each curve oscillates in a fundamentally different pattern. They are geometrically perpendicular in function space.

Step 3 β€” The β€œNew Information” Decomposition

Graph 3 β€” Decomposing $x^2$ into β€œalready known” + β€œgenuinely new”:

This is the key picture. When Gram-Schmidt processes $x^2$:

  1. It projects $x^2$ onto the span of $\lbrace1,\, x - \frac{1}{2}\rbrace$ β€” finding the best lower-degree approximation (red line: $x - \frac{1}{6}$)
  2. The residual $x^2 - (x - \frac{1}{6}) = x^2 - x + \frac{1}{6}$ (green curve) is the part of $x^2$ that no combination of constants and lines can capture

Look at the green curve β€” it’s pure curvature. No straight line (no matter how you tilt or shift it) can ever match that U-shape. This is the β€œvariation that no lower-degree polynomial can capture.”

How to read Graph 3 β€” point by point:

At every $x$-value:

\[\underbrace{x^2}_{\text{blue}} = \underbrace{\left(x - \tfrac{1}{6}\right)}_{\text{red}} + \underbrace{\left(x^2 - x + \tfrac{1}{6}\right)}_{\text{green}}\]

Pick a concrete point to verify β€” say $x = 0.5$:

  • Blue: $(0.5)^2 = 0.25$
  • Red: $0.5 - \frac{1}{6} \approx 0.333$
  • Green: $0.25 - 0.5 + 0.167 \approx -0.083$
  • Check: $0.333 + (-0.083) = 0.25$ βœ“

What does this decomposition mean?

  • The red line is the β€œshadow” (projection) of $x^2$ onto the space of degree-$\leq 1$ polynomials. It’s the best approximation of $x^2$ you can get using only constants and straight lines. Think of it like compressing a photo β€” you keep the rough trend but lose the fine details.

  • The green curve is the β€œerror” or β€œresidual” β€” everything about $x^2$ that a straight line simply cannot express. This is genuine degree-2 behavior: the curvature, the bowing. No matter what constants $a, b$ you choose, $a + bx$ will never look like that U-shape.

  • This is exactly the β€œvariation that no lower-degree polynomial can capture.”

The 3D Analogy:

Imagine a vector $\vec{v} = (3, 4, 5)$ in 3D space.

  • Project onto the $xy$-plane β†’ $(3, 4, 0)$ β€” this is the β€œalready known” part (like the red line)
  • Residual β†’ $(0, 0, 5)$ β€” this is the part that points purely in the $z$-direction (like the green curve)

The $z$-component is β€œvariation that no vector in the $xy$-plane can capture.” Moving around in the $xy$-plane (changing $a, b$ in $a\hat{x} + b\hat{y}$) will never give you any height in $z$.

Orthogonal polynomials work identically:

  • The β€œconstant direction” ↔ $\hat{x}$
  • The β€œlinear direction” ↔ $\hat{y}$
  • The β€œquadratic direction” ↔ $\hat{z}$

Each is perpendicular to the others. Each captures something the others geometrically cannot.


Step 4 β€” Why This Matters for Approximation

Graph 4 β€” Approximating $f(x) = \sin(\pi x)$ on $[0,1]$: orthogonal vs non-orthogonal coefficients

Suppose we want to approximate $\sin(\pi x)$ using polynomials up to degree 2. With orthogonal polynomials, each coefficient is computed independently β€” adding the quadratic term doesn’t change the constant or linear coefficients.

What Graph 4 shows:

We approximate $f(x) = \sin(\pi x)$ by building up one orthogonal polynomial at a time:

  1. Red (constant only): $c_1 q_1 = 0.637$ β€” just the average value of $\sin(\pi x)$ on $[0,1]$. A flat line at $\approx \frac{2}{\pi}$.

  2. Blue (constant + linear): Adding $c_2 q_2$ tilts the line. Since $\sin(\pi x)$ is symmetric on $[0,1]$, the tilt is nearly zero β€” the blue line barely differs from the red.

  3. Green (constant + linear + quadratic): Adding $c_3 q_3$ introduces curvature. Now the approximation starts to bend and follow the hump of the sine curve.

The crucial point: When we went from red β†’ blue, the coefficient $c_1 = 0.637$ did not change. When we went from blue β†’ green, neither $c_1$ nor $c_2$ changed. Each new term adds a layer without disturbing previous layers. This is only possible because the basis is orthogonal.

With the standard basis $\lbrace1, x, x^2\rbrace$, all coefficients would reshuffle at every step β€” making the approximation harder to interpret and compute.


7.6 QR Decomposition Connection

The Gram-Schmidt process gives a QR decomposition of a matrix:

\[A = QR\]

where:

  • $Q$ has the orthonormal vectors $\vec{u}_1, \ldots, \vec{u}_n$ as columns (orthogonal matrix)
  • $R$ is upper triangular with $R_{ij} = \langle \vec{v}_j, \vec{u}_i \rangle$

Applications of QR:

  • Solving overdetermined systems (least squares)
  • Computing eigenvalues (QR algorithm)
  • Numerical linear algebra β€” more stable than direct methods

Practice Questions β€” Inner Product & Gram-Schmidt

Q1. Compute $\langle (1,2,3), (4,-1,2) \rangle$. (Ans: 8)

Click for solution

$\langle (1,2,3), (4,-1,2) \rangle = 1(4) + 2(-1) + 3(2) = 4 - 2 + 6 = 8$ βœ“

Q2. Find $\lVert(3, 4)\rVert$ and $\lVert(1, 2, 2)\rVert$. (Ans: 5, 3)

Click for solution

$\lVert(3,4)\rVert = \sqrt{9 + 16} = \sqrt{25} = 5$ βœ“

$\lVert(1,2,2)\rVert = \sqrt{1 + 4 + 4} = \sqrt{9} = 3$ βœ“

Q3. Are $(1, 2, -1)$ and $(2, -1, 0)$ orthogonal? (Check: $2 - 2 + 0 = 0$. Yes!)

Click for solution

$\langle (1,2,-1), (2,-1,0) \rangle = 1(2) + 2(-1) + (-1)(0) = 2 - 2 + 0 = 0$.

Inner product is zero β†’ orthogonal βœ“

Q4. Normalize $(3, 4)$ to a unit vector. (Ans: $(3/5, 4/5)$)

Click for solution

$\lVert(3,4)\rVert = 5$. Unit vector: $\hat{v} = \frac{1}{5}(3,4) = \left(\frac{3}{5}, \frac{4}{5}\right)$.

Verify: $\lVert\hat{v}\rVert = \sqrt{9/25 + 16/25} = \sqrt{1} = 1$ βœ“

Q5. Apply Gram-Schmidt to $\lbrace(1, 2), (3, 4)\rbrace$ in $\mathbb{R}^2$.

Click for solution

Step 1 β€” First vector:

$\vec{w}_1 = (1, 2)$, $\;\lVert\vec{w}_1\rVert = \sqrt{5}$, $\;\vec{u}_1 = \dfrac{1}{\sqrt{5}}(1, 2)$

Step 2 β€” Orthogonalize second vector:

$\langle \vec{v}_2, \vec{u}_1 \rangle = \langle (3,4), \frac{1}{\sqrt{5}}(1,2) \rangle = \frac{3+8}{\sqrt{5}} = \frac{11}{\sqrt{5}}$

\[\vec{w}_2 = (3,4) - \frac{11}{\sqrt{5}} \cdot \frac{1}{\sqrt{5}}(1,2) = (3,4) - \frac{11}{5}(1,2) = \left(\frac{4}{5}, -\frac{2}{5}\right)\]

$\lVert\vec{w}_2\rVert = \sqrt{\frac{16+4}{25}} = \frac{2\sqrt{5}}{5}$, $\;\vec{u}_2 = \dfrac{1}{\sqrt{5}}(2, -1)$

Orthonormal basis:

\[\left\lbrace\dfrac{1}{\sqrt{5}}(1, 2), \;\dfrac{1}{\sqrt{5}}(2, -1)\right\rbrace\]

Verify: $\langle \vec{u}_1, \vec{u}_2 \rangle = \frac{1}{5}(2 - 2) = 0$ βœ“

Q6. Apply Gram-Schmidt to $\lbrace(1, 0, 1), (0, 1, 1)\rbrace$ in $\mathbb{R}^3$.

Click for solution

Step 1: $\vec{w}_1 = (1,0,1)$, $\;\lVert\vec{w}_1\rVert = \sqrt{2}$, $\;\vec{u}_1 = \dfrac{1}{\sqrt{2}}(1,0,1)$

Step 2: $\langle \vec{v}_2, \vec{u}_1 \rangle = \frac{0+0+1}{\sqrt{2}} = \frac{1}{\sqrt{2}}$

\[\vec{w}_2 = (0,1,1) - \frac{1}{\sqrt{2}} \cdot \frac{1}{\sqrt{2}}(1,0,1) = (0,1,1) - \frac{1}{2}(1,0,1) = \left(-\frac{1}{2}, 1, \frac{1}{2}\right)\]

$\lVert\vec{w}_2\rVert = \sqrt{\frac{1}{4}+1+\frac{1}{4}} = \sqrt{\frac{3}{2}} = \frac{\sqrt{6}}{2}$

\[\vec{u}_2 = \frac{1}{\sqrt{6}}(-1, 2, 1)\]

Orthonormal basis:

\[\left\lbrace\dfrac{1}{\sqrt{2}}(1,0,1), \;\dfrac{1}{\sqrt{6}}(-1, 2, 1)\right\rbrace\]

Q7. Find the projection of $(3, 4)$ onto $(1, 0)$.

Click for solution
\[\text{proj}_{\vec{u}}\vec{v} = \frac{\langle \vec{v}, \vec{u} \rangle}{\lVert\vec{u}\rVert^2}\vec{u} = \frac{3(1)+4(0)}{1^2}(1,0) = 3(1,0) = (3, 0)\]

The projection drops the $y$-component β€” geometrically obvious since $(1,0)$ is the $x$-axis direction.

Q8. Apply Gram-Schmidt to $\lbrace1, x, x^2\rbrace$ in $P_2$ with $\langle p, q \rangle = \int_0^1 p(x)q(x) \, dx$.

Click for solution

Step 1: $w_1 = 1$. $\langle 1,1 \rangle = \int_0^1 1 \, dx = 1$. $u_1 = 1$.

Step 2: $\langle x, u_1 \rangle = \int_0^1 x \, dx = \frac{1}{2}$

$w_2 = x - \frac{1}{2}$. $\;\langle w_2, w_2 \rangle = \int_0^1\left(x-\frac{1}{2}\right)^2 dx = \frac{1}{12}$. $u_2 = \frac{w_2}{\sqrt{1/12}} = 2\sqrt{3}\left(x - \frac{1}{2}\right)$.

Step 3: $\langle x^2, u_1 \rangle = \int_0^1 x^2 \, dx = \frac{1}{3}$

$\langle x^2, u_2 \rangle = 2\sqrt{3}\int_0^1 x^2\left(x-\frac{1}{2}\right) dx = 2\sqrt{3}\left(\frac{1}{4}-\frac{1}{6}\right) = \frac{\sqrt{3}}{6}$

\[w_3 = x^2 - \frac{1}{3} - \frac{\sqrt{3}}{6}\cdot 2\sqrt{3}\!\left(x-\frac{1}{2}\right) = x^2 - x + \frac{1}{6}\]

$\langle w_3, w_3 \rangle = \int_0^1\left(x^2-x+\frac{1}{6}\right)^2 dx = \frac{1}{180}$

Orthonormal basis:

\[\left\lbrace 1, \;\; 2\sqrt{3}\!\left(x-\tfrac{1}{2}\right), \;\; 6\sqrt{5}\!\left(x^2-x+\tfrac{1}{6}\right)\right\rbrace\]

These are the shifted Legendre polynomials on $[0,1]$.

Q9. Verify that the result of Q5 is orthonormal (check dot product = 0 and norms = 1).

Click for solution

From Q5: $\vec{u}_1 = \frac{1}{\sqrt{5}}(1,2)$, $\vec{u}_2 = \frac{1}{\sqrt{5}}(2,-1)$.

$\langle \vec{u}_1, \vec{u}_2 \rangle = \frac{1}{5}(1 \cdot 2 + 2 \cdot (-1)) = \frac{1}{5}(0) = 0$ βœ“ (orthogonal)

$\lVert\vec{u}_1\rVert = \frac{1}{\sqrt{5}}\sqrt{1+4} = 1$ βœ“, $\;\lVert\vec{u}_2\rVert = \frac{1}{\sqrt{5}}\sqrt{4+1} = 1$ βœ“ (unit vectors)

Q10. If $Q$ is an orthogonal matrix, what is $Q^TQ$? (Ans: Identity matrix $I$)

Click for solution

By definition, a matrix $Q$ is orthogonal if $Q^TQ = I$ (equivalently, $Q^{-1} = Q^T$).

This means the columns of $Q$ form an orthonormal set: $\langle \vec{q}i, \vec{q}_j \rangle = \delta{ij}$.

Q11. Find the QR decomposition of

\[A = \begin{bmatrix} 1 & 1 \\ 0 & 1 \\ 0 & 0 \end{bmatrix}\]
Click for solution

Gram-Schmidt on columns of $A$:

$\vec{a}_1 = (1,0,0)$. $\;\lVert\vec{a}_1\rVert = 1$. $\;\vec{q}_1 = (1,0,0)$.

$\langle \vec{a}_2, \vec{q}_1 \rangle = 1$. $\;\vec{w}_2 = (1,1,0) - 1(1,0,0) = (0,1,0)$. $\;\lVert\vec{w}_2\rVert = 1$. $\;\vec{q}_2 = (0,1,0)$.

\[Q = \begin{bmatrix} 1 & 0 \\ 0 & 1 \\ 0 & 0 \end{bmatrix}, \quad R = \begin{bmatrix} \langle \vec{a}_1,\vec{q}_1\rangle & \langle \vec{a}_2,\vec{q}_1\rangle \\ 0 & \langle \vec{a}_2,\vec{q}_2\rangle \end{bmatrix} = \begin{bmatrix} 1 & 1 \\ 0 & 1 \end{bmatrix}\]

Verify:

\[QR = \begin{bmatrix}1&1\\0&1\\0&0\end{bmatrix} = A \checkmark\]

Q12. Compute the projection of $\vec{v} = (2, 3, 1)$ onto $\vec{u} = (1, 1, 1)$. (Ans: $(2, 2, 2)$)

Click for solution
\[\text{proj}_{\vec{u}}\vec{v} = \frac{\langle \vec{v}, \vec{u} \rangle}{\lVert\vec{u}\rVert^2}\vec{u} = \frac{2+3+1}{1+1+1}(1,1,1) = \frac{6}{3}(1,1,1) = (2, 2, 2)\]

The orthogonal component: $\vec{v} - \text{proj} = (2,3,1) - (2,2,2) = (0,1,-1)$.

Check: $\langle (0,1,-1), (1,1,1) \rangle = 0+1-1 = 0$ βœ“ (perpendicular to $\vec{u}$)


8 β€” Coordinates of a Vector (with Respect to a Basis)


8.1 Motivation β€” Why Coordinates Matter

The Big Idea: In $\mathbb{R}^n$ we usually describe vectors using the standard basis β€” $(3, 4)$ implicitly means $3\vec{e}_1 + 4\vec{e}_2$. But the standard basis isn’t always the most natural. Depending on the problem, a different basis may simplify computations dramatically.

Example-driven motivation:

  • In physics, rotating coordinate systems align with the problem (polar, cylindrical, spherical).
  • In image compression (JPEG), we switch from pixel coordinates to frequency coordinates (DCT basis).
  • In quantum mechanics, we pick a basis of eigenstates.
  • In PCA (machine learning), we switch to a basis of principal component directions.

Coordinates let us translate between any two β€œlanguages” (bases) for describing the same vector space.


8.2 Definition of Coordinate Vector

Definition: Let $B = \lbrace\vec{v}_1, \vec{v}_2, \ldots, \vec{v}_n\rbrace$ be an ordered basis of a vector space $V$. For any $\vec{w} \in V$, there exist unique scalars $c_1, c_2, \ldots, c_n$ such that:

\[\vec{w} = c_1\vec{v}_1 + c_2\vec{v}_2 + \cdots + c_n\vec{v}_n\]

The coordinate vector of $\vec{w}$ relative to $B$ is:

\[[\vec{w}]_B = \begin{pmatrix} c_1 \\ c_2 \\ \vdots \\ c_n \end{pmatrix}\]

Quick Example: Let $B = \lbrace(1,0), (0,1)\rbrace$ be the standard basis of $\mathbb{R}^2$ and $\vec{w} = (3, 7)$.

We need $c_1, c_2$ such that $c_1(1,0) + c_2(0,1) = (3,7)$. Clearly $c_1 = 3, \; c_2 = 7$.

\[[\vec{w}]_B = \begin{pmatrix} 3 \\ 7 \end{pmatrix}\]

In the standard basis, the coordinate vector is just the vector itself. But with a different basis (say $B’ = \lbrace(1,1), (1,-1)\rbrace$), the same vector $(3,7)$ would have different coordinates β€” you’d solve $c_1(1,1) + c_2(1,-1) = (3,7)$ to get $c_1 = 5, \; c_2 = -2$, so $[\vec{w}]_{B’} = \begin{pmatrix}5 \ -2\end{pmatrix}$.

Same vector, different β€œaddress” depending on the basis you use β€” just like the same location has different GPS coordinates depending on the map projection.

Why uniqueness? Since $B$ is a basis, it is (1) spanning (so scalars exist) and (2) linearly independent (so the scalars are unique). This is precisely why the basis concept is so powerful β€” it gives a bijection (one-to-one correspondence) between vectors in $V$ and coordinate tuples in $\mathbb{R}^n$.


8.3 Computing Coordinates β€” Examples

Example 1: Coordinates in the Standard Basis

Basis: $B = \lbrace(1,0), (0,1)\rbrace$ (standard basis of $\mathbb{R}^2$)

Vector: $\vec{w} = (5, 3)$

Since $(5,3) = 5(1,0) + 3(0,1)$:

\[[\vec{w}]_B = \begin{pmatrix} 5 \\ 3 \end{pmatrix}\]

Observation: In the standard basis, the coordinate vector is just the vector itself! This is why we usually don’t write $[\vec{w}]_B$ when using the standard basis β€” coordinates and components coincide.

Example 2: Coordinates in a Non-Standard Basis

Basis: $B = \lbrace\vec{v}_1 = (1,1), \vec{v}_2 = (1,-1)\rbrace$

Vector: $\vec{w} = (5, 3)$

Find $c_1, c_2$ such that $c_1(1,1) + c_2(1,-1) = (5,3)$:

\[\begin{cases} c_1 + c_2 = 5 \\ c_1 - c_2 = 3 \end{cases}\]

Adding: $2c_1 = 8 \Rightarrow c_1 = 4$. Subtracting: $2c_2 = 2 \Rightarrow c_2 = 1$.

\[[\vec{w}]_B = \begin{pmatrix} 4 \\ 1 \end{pmatrix}\]

Verification: $4(1,1) + 1(1,-1) = (4,4) + (1,-1) = (5,3)$ βœ“

Key point: The same vector $(5,3)$ has coordinates $(5,3)$ in the standard basis but $(4,1)$ in basis $B$. The vector hasn’t changed β€” only the β€œlanguage” we use to describe it.

Example 3: Coordinates in $\mathbb{R}^3$

Basis: $B = \lbrace(1,0,1), (0,1,1), (1,1,0)\rbrace$

Vector: $\vec{w} = (3, 2, 4)$

Solve $c_1(1,0,1) + c_2(0,1,1) + c_3(1,1,0) = (3,2,4)$:

\[\begin{cases} c_1 + c_3 = 3 \\ c_2 + c_3 = 2 \\ c_1 + c_2 = 4 \end{cases}\]

From equations 1 and 2: $c_1 = 3 - c_3$ and $c_2 = 2 - c_3$.

Substituting into equation 3: $(3 - c_3) + (2 - c_3) = 4 \Rightarrow 5 - 2c_3 = 4 \Rightarrow c_3 = \frac{1}{2}$.

Then $c_1 = \frac{5}{2}$ and $c_2 = \frac{3}{2}$.

\[[\vec{w}]_B = \begin{pmatrix} 5/2 \\ 3/2 \\ 1/2 \end{pmatrix}\]

Verification: $\frac{5}{2}(1,0,1) + \frac{3}{2}(0,1,1) + \frac{1}{2}(1,1,0) = (\frac{5}{2},0,\frac{5}{2}) + (0,\frac{3}{2},\frac{3}{2}) + (\frac{1}{2},\frac{1}{2},0) = (3,2,4)$ βœ“

Example 4: Coordinates of Polynomials

Basis: $B = \lbrace1, x, x^2\rbrace$ (standard basis of $P_2$)

Vector: $p(x) = 3 - 2x + 5x^2$

\[[p(x)]_B = \begin{pmatrix} 3 \\ -2 \\ 5 \end{pmatrix}\]

Non-standard basis: $B’ = \lbrace1, (x-1), (x-1)^2\rbrace$

Expanding $p(x) = 3 - 2x + 5x^2$ in terms of $B’$:

\[p(x) = a_0 \cdot 1 + a_1(x-1) + a_2(x-1)^2\]

Expanding the right side: $a_0 + a_1 x - a_1 + a_2 x^2 - 2a_2 x + a_2$

\[= (a_0 - a_1 + a_2) + (a_1 - 2a_2)x + a_2 x^2\]

Comparing with $3 - 2x + 5x^2$:

  • $a_2 = 5$
  • $a_1 - 2(5) = -2 \Rightarrow a_1 = 8$
  • $a_0 - 8 + 5 = 3 \Rightarrow a_0 = 6$
\[[p(x)]_{B'} = \begin{pmatrix} 6 \\ 8 \\ 5 \end{pmatrix}\]

Verification: $6 + 8(x-1) + 5(x-1)^2 = 6 + 8x - 8 + 5x^2 - 10x + 5 = 3 - 2x + 5x^2$ βœ“


8.4 Change of Basis β€” Transition Matrix

What is an ordered basis?

Think of a locker row with numbered slots: Slot 1, Slot 2, Slot 3, …

  • A basis is like saying β€œthese 3 lockers exist” β€” you know which lockers, but nobody assigned numbers to them.
  • An ordered basis is like saying β€œLocker A is Slot 1, Locker B is Slot 2, Locker C is Slot 3.” Now every item can be described by its slot number.

A coordinate vector is just the list of slot numbers. If you renumber the slots (change the order), the same items get different slot numbers β€” even though nothing physically moved.

A regular basis is a set β€” $\lbrace \vec{v}_1, \vec{v}_2 \rbrace = \lbrace \vec{v}_2, \vec{v}_1 \rbrace$. But an ordered basis uses a fixed sequence: $(\vec{v}_1, \vec{v}_2) \neq (\vec{v}_2, \vec{v}_1)$.

This matters because the coordinate vector $[\vec{x}]_B = \begin{pmatrix} c_1 \ c_2 \end{pmatrix}$ means β€œfirst entry = coefficient of the 1st basis vector, second entry = coefficient of the 2nd.” Swap the order, and the coordinate vector flips:

\[B = (\vec{v}_1, \vec{v}_2) \;\Rightarrow\; [\vec{x}]_B = \begin{pmatrix} c_1 \\ c_2 \end{pmatrix}, \qquad B' = (\vec{v}_2, \vec{v}_1) \;\Rightarrow\; [\vec{x}]_{B'} = \begin{pmatrix} c_2 \\ c_1 \end{pmatrix}\]

Without ordering, coordinate vectors, transition matrices, and matrix representations of maps would all be undefined. (See full explanation.)

Definition: Let $B$ and $B’$ be two ordered bases of a vector space $V$:

\[B = \lbrace\vec{v}_1, \ldots, \vec{v}_n\rbrace, \quad B' = \lbrace\vec{w}_1, \ldots, \vec{w}_n\rbrace\]

The transition matrix (change-of-basis matrix) from $B$ to $B’$ is the matrix $P_{B \to B’}$ such that:

\[[\vec{x}]_{B'} = P_{B \to B'} \, [\vec{x}]_B\]

The columns of $P_{B \to B’}$ are the coordinate vectors of the old basis vectors in terms of the new basis:

\[P_{B \to B'} = \Big[ [\vec{v}_1]_{B'} \;\; [\vec{v}_2]_{B'} \;\; \cdots \;\; [\vec{v}_n]_{B'} \Big]\]

Computing the transition matrix (practical method):

Form the augmented matrix $[B’ \mid B]$ (new basis on the left, old basis on the right) and row reduce to $[I \mid P_{B \to B’}]$.

This works because row reduction simultaneously solves $n$ systems expressing each old basis vector in terms of the new basis.

The Big Picture β€” What is Change of Basis Really Doing?

The core idea: same vector, different descriptions

Imagine you’re standing in a room and someone asks β€œWhere are you?” You could answer:

  • β€œ3 meters from the north wall, 4 meters from the west wall” (one coordinate system)
  • β€œ5 meters from the corner, at angle 53°” (different coordinate system)

Both answers describe the same physical position β€” only the language (the reference frame) changed. That’s exactly what change of basis does for vectors.

Core principle: A vector $\vec{x}$ is a fixed object in space. The coordinate vector $[\vec{x}]_B$ is just a description of $\vec{x}$ using basis $B$ as the language. Change of basis translates one description into another β€” the vector itself never moves.

Why do we need different bases?

The standard basis $\lbrace \vec{e}_1, \vec{e}_2, \ldots \rbrace$ is simple, but it’s not always the best description. Different bases reveal different structure:

Situation Best basis to use Why
General computation Standard basis $\lbrace(1,0),(0,1)\rbrace$ Simple, familiar
Rotation problems Eigenvectors of rotation matrix Rotation becomes diagonal
Signal processing Fourier basis $\lbrace 1, \cos x, \sin x, \ldots \rbrace$ Reveals frequency content
Differential equations Eigenvectors of coefficient matrix Each component decouples
Quantum mechanics Energy eigenstates Measurement outcomes are clear

Change of basis is the translator that lets you move between these different β€œlanguages.”

How the transition matrix works β€” step by step

Suppose we have two bases for $\mathbb{R}^2$:

\[B = (\vec{v}_1, \vec{v}_2) \quad \text{(old basis)}, \qquad B' = (\vec{w}_1, \vec{w}_2) \quad \text{(new basis)}\]

Step 1 β€” Express each old basis vector in the new language.

Since $B’$ is a basis, we can write each old basis vector as a combination of new ones:

\[\vec{v}_1 = a_{11}\vec{w}_1 + a_{21}\vec{w}_2 \quad \Longrightarrow \quad [\vec{v}_1]_{B'} = \begin{pmatrix} a_{11} \\ a_{21} \end{pmatrix}\] \[\vec{v}_2 = a_{12}\vec{w}_1 + a_{22}\vec{w}_2 \quad \Longrightarrow \quad [\vec{v}_2]_{B'} = \begin{pmatrix} a_{12} \\ a_{22} \end{pmatrix}\]

Step 2 β€” Stack those as columns to form the transition matrix:

\[P_{B \to B'} = \begin{bmatrix} a_{11} & a_{12} \\ a_{21} & a_{22} \end{bmatrix} = \Big[ [\vec{v}_1]_{B'} \;\; [\vec{v}_2]_{B'} \Big]\]

Why do we stack them as columns?

Think about what we want the transition matrix $P$ to do: given a coordinate vector $[\vec{x}]_B = (c_1, c_2)^T$, we want $P \cdot [\vec{x}]_B$ to produce $[\vec{x}]_{B'}$.

Now recall how matrix-vector multiplication works β€” the result is a linear combination of the columns, weighted by the entries of the vector:

$$P \begin{pmatrix} c_1 \\ c_2 \end{pmatrix} = c_1 \cdot (\text{column 1 of } P) + c_2 \cdot (\text{column 2 of } P)$$

So if column 1 of $P$ is $[\vec{v}_1]_{B'}$ and column 2 is $[\vec{v}_2]_{B'}$, then:

$$P \begin{pmatrix} c_1 \\ c_2 \end{pmatrix} = c_1 [\vec{v}_1]_{B'} + c_2 [\vec{v}_2]_{B'}$$

This is exactly the $B'$-coordinates of the vector $c_1\vec{v}_1 + c_2\vec{v}_2 = \vec{x}$. The columns are the translation rules for each old basis vector, and the input weights $c_1, c_2$ tell you how much of each rule to apply. That's why it must be columns β€” it's the only arrangement that makes matrix multiplication do the right thing.

Step 3 β€” Now ANY vector’s coordinates can be translated.

If $\vec{x} = c_1\vec{v}_1 + c_2\vec{v}_2$ (i.e., $[\vec{x}]_B = (c_1, c_2)^T$), then:

\[\vec{x} = c_1\vec{v}_1 + c_2\vec{v}_2 = c_1(a_{11}\vec{w}_1 + a_{21}\vec{w}_2) + c_2(a_{12}\vec{w}_1 + a_{22}\vec{w}_2)\] \[= (c_1 a_{11} + c_2 a_{12})\vec{w}_1 + (c_1 a_{21} + c_2 a_{22})\vec{w}_2\] \[\Longrightarrow \quad [\vec{x}]_{B'} = \begin{bmatrix} a_{11} & a_{12} \\ a_{21} & a_{22} \end{bmatrix} \begin{pmatrix} c_1 \\ c_2 \end{pmatrix} = P_{B \to B'} \, [\vec{x}]_B\]

The recipe in one sentence: To build $P_{B \to B’}$, take each old basis vector, express it in the new basis, and place the resulting coordinate vectors as columns. Then multiplying by $P$ translates any vector’s coordinates from old to new.

Understanding $[B’ \mid B] \to [I \mid P]$ β€” why the augmented method works

When we row reduce $[B’ \mid B]$:

  • The left half contains the new basis vectors (as columns).
  • The right half contains the old basis vectors (as columns).
  • Row reducing the left half to $I$ is equivalent to applying $(B’)^{-1}$ β€” converting from standard coordinates to $B’$-coordinates.
  • The same operations applied to the right half give $(B’)^{-1} \cdot B$ β€” this is exactly each old basis vector expressed in the new basis.
\[[B' \mid B] \xrightarrow{\text{row reduce}} [I \mid (B')^{-1}B] = [I \mid P_{B \to B'}]\]

Common confusion: which basis goes where?

  • New basis on the LEFT of the augment bar $[\;B’\; \mid \;B\;]$
  • Old basis on the RIGHT of the augment bar
  • After row reducing: the right half gives $P_{B \to B’}$

Memory trick: we’re β€œdividing out” the new basis (reducing it to $I$), which forces the old basis vectors to express themselves in new-basis language.

Visual intuition β€” the coordinate grid transforms

Picture $\mathbb{R}^2$ with the standard grid. When you change basis:

  1. The grid lines rotate and stretch to align with the new basis vectors.
  2. Every point’s β€œaddress” (its coordinates) changes because the grid changed.
  3. But the physical point stays fixed.

The transition matrix describes how to read addresses on the old grid and produce the corresponding address on the new grid.

Chain of transitions

If you need to go through multiple basis changes $B \to B’ \to B’’$, just multiply the transition matrices:

\[P_{B \to B''} = P_{B' \to B''} \cdot P_{B \to B'}\]

This is like translating English β†’ French β†’ Japanese: compose the two translation steps. Order matters β€” matrix multiplication is not commutative!

And to go backwards, just invert:

\[P_{B' \to B} = (P_{B \to B'})^{-1}\]

This always exists because basis vectors are linearly independent, so the transition matrix is always invertible.

Example: Change of Basis in $\mathbb{R}^2$

Old basis: $B = \lbrace(1,1), (1,-1)\rbrace$

New basis: $B’ = \lbrace(1,0), (0,1)\rbrace = $ standard basis

Find $P_{B \to B’}$:

Since $B’$ is the standard basis, $[\vec{v}]_{B’} = \vec{v}$ for any $\vec{v}$. So:

\[P_{B \to B'} = \begin{bmatrix} 1 & 1 \\ 1 & -1 \end{bmatrix}\]

(Just the old basis vectors as columns!)

Check: If $[\vec{x}]_B = (4, 1)^T$, then:

\[[\vec{x}]_{B'} = \begin{bmatrix} 1 & 1 \\ 1 & -1 \end{bmatrix} \begin{pmatrix} 4 \\ 1 \end{pmatrix} = \begin{pmatrix} 5 \\ 3 \end{pmatrix}\]

This matches Example 2 above: $(5,3)$ in standard coordinates ↔ $(4,1)$ in basis $B$ βœ“

Example: Transition Between Two Non-Standard Bases

Basis $B$: $\lbrace(1,0), (1,1)\rbrace$

Basis $B’$: $\lbrace(1,2), (2,1)\rbrace$

Find $P_{B \to B’}$: Express each vector in $B$ as a linear combination of vectors in $B’$.

Set up $[B’ \mid B]$ and row reduce:

\[\left[\begin{array}{cc|cc} 1 & 2 & 1 & 1 \\ 2 & 1 & 0 & 1 \end{array}\right]\]

$R_2 \leftarrow R_2 - 2R_1$:

\[\left[\begin{array}{cc|cc} 1 & 2 & 1 & 1 \\ 0 & -3 & -2 & -1 \end{array}\right]\]

$R_2 \leftarrow -\frac{1}{3}R_2$:

\[\left[\begin{array}{cc|cc} 1 & 2 & 1 & 1 \\ 0 & 1 & 2/3 & 1/3 \end{array}\right]\]

$R_1 \leftarrow R_1 - 2R_2$:

\[\left[\begin{array}{cc|cc} 1 & 0 & -1/3 & 1/3 \\ 0 & 1 & 2/3 & 1/3 \end{array}\right]\] \[P_{B \to B'} = \begin{bmatrix} -1/3 & 1/3 \\ 2/3 & 1/3 \end{bmatrix}\]

Key Properties of Transition Matrices:

  1. $P_{B \to B’}$ is always invertible (basis vectors are independent).
  2. $P_{B’ \to B} = (P_{B \to B’})^{-1}$ β€” the reverse transition is the inverse.
  3. $P_{B \to B’’} = P_{B’ \to B’’} \cdot P_{B \to B’}$ β€” transitions compose by matrix multiplication.

8.5 Isomorphism Between $V$ and $\mathbb{R}^n$

Coordinate Isomorphism Theorem: If $V$ is a vector space of dimension $n$ with ordered basis $B$, then the coordinate mapping:

\[\phi_B : V \to \mathbb{R}^n, \quad \vec{v} \mapsto [\vec{v}]_B\]

is a linear isomorphism (a bijective linear map). That is:

  1. $[\vec{u} + \vec{v}]_B = [\vec{u}]_B + [\vec{v}]_B$ (preserves addition)
  2. $[c\vec{v}]_B = c[\vec{v}]_B$ (preserves scalar multiplication)
  3. $\phi_B$ is one-to-one and onto

Why this matters: This theorem says every $n$-dimensional vector space β€” whether it consists of polynomials, matrices, functions, or anything else β€” is β€œstructurally identical” to $\mathbb{R}^n$. By choosing a basis, we can always reduce problems in abstract vector spaces to computations in $\mathbb{R}^n$!


8.6 Worked Examples β€” Coordinates & Solution Spaces

Example: $\sin t, \cos t, t$ are Linearly Independent

Problem: Show that $f(t) = \sin t$, $g(t) = \cos t$, $h(t) = t$ are linearly independent in the vector space of functions from $\mathbb{R}$ to $\mathbb{R}$.

Suppose $x \sin t + y \cos t + z \cdot t = 0$ for all $t \in \mathbb{R}$.

Evaluate at specific values of $t$:

  • $t = 0$: $x(0) + y(1) + z(0) = 0 \Rightarrow y = 0$
  • $t = \pi/2$: $x(1) + y(0) + z(\pi/2) = 0 \Rightarrow x + z\pi/2 = 0$
  • $t = \pi$: $x(0) + y(-1) + z\pi = 0 \Rightarrow z\pi = 0 \Rightarrow z = 0$

From the second equation with $z = 0$: $x = 0$.

Therefore $x = y = z = 0$ β€” the functions are linearly independent. $\checkmark$

Example: Find Homogeneous System for a Given Solution Set

Problem: Find a homogeneous system $AX = 0$ whose solution set is spanned by:

\[\vec{u}_1 = (1, -3, 0, 3),\quad \vec{u}_2 = (1, -1, 1, 4),\quad \vec{u}_3 = (2, 0, -2, 5)\]

Approach: We need $\vec{v} = (x, y, z, t)$ to be in the solution space if and only if $\vec{v}$ is a linear combination of $\vec{u}_1, \vec{u}_2, \vec{u}_3$.

Form the matrix $M = [u_1^T \mid u_2^T \mid u_3^T \mid v^T]$ and require $\rho(M) = \rho$ of the left part:

\[\begin{bmatrix} 1 & 1 & 2 & x \\ -3 & -1 & 0 & y \\ 0 & 1 & -2 & z \\ 3 & 4 & 5 & t \end{bmatrix}\]

Row-reduce and impose consistency. The conditions give the equations of the homogeneous system for which $\lbrace\vec{u}_1, \vec{u}_2, \vec{u}_3\rbrace$ is the solution basis.

After row reduction, the constraints on $(x, y, z, t)$ yield:

\[t - 5z - y = 0, \qquad 2t + z + y = 0\]

These are the two equations of the required homogeneous system.

Example: Coordinates of a Matrix Relative to a Basis

Problem: In $M_{2 \times 2}(\mathbb{R})$, let $S = \left\lbrace E_1 = \begin{pmatrix}1&1\1&1\end{pmatrix}, E_2 = \begin{pmatrix}1&-1\1&0\end{pmatrix}, E_3 = \begin{pmatrix}1&1\0&0\end{pmatrix}, E_4 = \begin{pmatrix}1&0\0&0\end{pmatrix} \right\rbrace$ be a basis. Find $[A]_S$ where $A = \begin{pmatrix}3&-5\3&1\end{pmatrix}$.

Set $A = aE_1 + bE_2 + cE_3 + dE_4$:

\[\begin{pmatrix}3&-5\\3&1\end{pmatrix} = a\begin{pmatrix}1&1\\1&1\end{pmatrix} + b\begin{pmatrix}1&-1\\1&0\end{pmatrix} + c\begin{pmatrix}1&1\\0&0\end{pmatrix} + d\begin{pmatrix}1&0\\0&0\end{pmatrix}\]

Entry-wise:

\[a + b + c + d = 3, \quad a - b + c = -5, \quad a + b = 3, \quad a = 1\]

Solving: $a = 1$, $b = 2$, $c = -8$, $d = 8$.

\[[A]_S = (1, 2, -8, 8)\]

Example: Checking LD Using Coordinate Vectors

Problem: In $M_{2 \times 3}(\mathbb{R})$, determine whether $A = \begin{pmatrix}1&2&3\4&0&1\end{pmatrix}$, $B = \begin{pmatrix}1&3&-4\6&5&4\end{pmatrix}$, $C = \begin{pmatrix}3&8&-11\16&10&9\end{pmatrix}$ are linearly dependent.

Strategy: Convert each matrix to its coordinate vector relative to the standard basis of $M_{2 \times 3}$:

\[[A]_S = (1, 2, 3, 4, 0, 1), \quad [B]_S = (1, 3, -4, 6, 5, 4), \quad [C]_S = (3, 8, -11, 16, 10, 9)\]

Form the matrix with these as rows and row-reduce:

\[\begin{bmatrix} 1 & 2 & 3 & 4 & 0 & 1 \\ 1 & 3 & -4 & 6 & 5 & 4 \\ 3 & 8 & -11 & 16 & 10 & 9 \end{bmatrix} \to \begin{bmatrix} 1 & 2 & 3 & 4 & 0 & 1 \\ 0 & 1 & -7 & 2 & 5 & 3 \\ 0 & 0 & 0 & 0 & 0 & 0 \end{bmatrix}\]

$\rho = 2 < 3$ (number of vectors), so the coordinate vectors are LD.

Since the coordinate mapping is an isomorphism (preserves linear dependence): $A, B, C$ are linearly dependent. $\checkmark$

Example: Express a Polynomial as a Linear Combination

Problem: Express $v(t) = t^2 + 4t - 3$ as a linear combination of $p_1(t) = t^2 - 2t + 5$, $p_2(t) = 2t^2 - 3t$, $p_3(t) = t + 1$ in $P_2(t)$.

Convert to coordinate vectors: $(1, 4, -3)$ vs $(1, -2, 5)$, $(2, -3, 0)$, $(0, 1, 1)$.

Set up: $v = x \cdot p_1 + y \cdot p_2 + z \cdot p_3$, which gives:

\[x + 2y = 1, \quad -2x - 3y + z = 4, \quad 5x + z = -3\] \[\left[\begin{array}{ccc|c} 1 & 2 & 0 & 1 \\ -2 & -3 & 1 & 4 \\ 5 & 0 & 1 & -3 \end{array}\right] \to \left[\begin{array}{ccc|c} 1 & 2 & 0 & 1 \\ 0 & 1 & 1 & 6 \\ 0 & 0 & 11 & 22 \end{array}\right]\]

$z = 2$, $y = 4$, $x = -7$.

\[v(t) = -7p_1(t) + 4p_2(t) + 2p_3(t) \checkmark\]

Practice Questions β€” Coordinates

Q1. Let $B = \lbrace(1,2), (3,5)\rbrace$ be a basis of $\mathbb{R}^2$. Find $[\vec{w}]_B$ where $\vec{w} = (1, 1)$.

Click for solution

Solve $c_1(1,2) + c_2(3,5) = (1,1)$:

\[\begin{cases} c_1 + 3c_2 = 1 \\ 2c_1 + 5c_2 = 1 \end{cases}\]

From Eq1: $c_1 = 1 - 3c_2$. Substituting into Eq2: $2(1 - 3c_2) + 5c_2 = 1 \Rightarrow 2 - c_2 = 1 \Rightarrow c_2 = 1$.

$c_1 = 1 - 3 = -2$.

\[[\vec{w}]_B = \begin{pmatrix} -2 \\ 1 \end{pmatrix}\]

Verification: $-2(1,2) + 1(3,5) = (-2,-4) + (3,5) = (1,1)$ βœ“

Q2. Let $B = \lbrace(1,0,0), (1,1,0), (1,1,1)\rbrace$ be a basis of $\mathbb{R}^3$. Find $[(2,3,5)]_B$.

Click for solution

Solve $c_1(1,0,0) + c_2(1,1,0) + c_3(1,1,1) = (2,3,5)$:

\[\begin{cases} c_1 + c_2 + c_3 = 2 \\ c_2 + c_3 = 3 \\ c_3 = 5 \end{cases}\]

Back substitution: $c_3 = 5$, $c_2 = 3 - 5 = -2$, $c_1 = 2 - (-2) - 5 = -1$.

\[[(2,3,5)]_B = \begin{pmatrix} -1 \\ -2 \\ 5 \end{pmatrix}\]

Verification: $-1(1,0,0) + (-2)(1,1,0) + 5(1,1,1) = (-1,0,0) + (-2,-2,0) + (5,5,5) = (2,3,5)$ βœ“

Q3. Let $B = \lbrace1+x, 1-x\rbrace$ be a basis of $P_1$. Find the coordinate vector of $p(x) = 3 + 7x$.

Click for solution

Solve $c_1(1+x) + c_2(1-x) = 3 + 7x$:

$(c_1 + c_2) + (c_1 - c_2)x = 3 + 7x$

\[\begin{cases} c_1 + c_2 = 3 \\ c_1 - c_2 = 7 \end{cases}\]

$c_1 = 5, \; c_2 = -2$.

\[[3+7x]_B = \begin{pmatrix} 5 \\ -2 \end{pmatrix}\]

Verification: $5(1+x) + (-2)(1-x) = 5+5x - 2+2x = 3+7x$ βœ“

Q4. Given bases $B = \lbrace(1,0), (0,1)\rbrace$ and $B’ = \lbrace(2,1), (1,1)\rbrace$ of $\mathbb{R}^2$, find the transition matrix $P_{B \to B’}$.

Click for solution

Express each vector of $B$ in terms of $B’$. Set up $[B’ \mid B]$:

\[\left[\begin{array}{cc|cc} 2 & 1 & 1 & 0 \\ 1 & 1 & 0 & 1 \end{array}\right]\]

$R_1 \leftrightarrow R_2$:

\[\left[\begin{array}{cc|cc} 1 & 1 & 0 & 1 \\ 2 & 1 & 1 & 0 \end{array}\right]\]

$R_2 \leftarrow R_2 - 2R_1$:

\[\left[\begin{array}{cc|cc} 1 & 1 & 0 & 1 \\ 0 & -1 & 1 & -2 \end{array}\right]\]

$R_2 \leftarrow -R_2$, then $R_1 \leftarrow R_1 + R_2$:

\[\left[\begin{array}{cc|cc} 1 & 0 & 1 & -1 \\ 0 & 1 & -1 & 2 \end{array}\right]\] \[P_{B \to B'} = \begin{bmatrix} 1 & -1 \\ -1 & 2 \end{bmatrix}\]

Check: $[(3,4)]_B = (3,4)^T$. Then:

\[[(3,4)]_{B'} = \begin{pmatrix} 1&-1\\-1&2 \end{pmatrix}\begin{pmatrix} 3\\4 \end{pmatrix} = \begin{pmatrix} -1\\5 \end{pmatrix}\]

Verify: $-1(2,1) + 5(1,1) = (-2,-1) + (5,5) = (3,4)$ βœ“

Q5. If $[\vec{v}]_B = (2, -1, 3)^T$ and $B = \lbrace(1,0,1), (0,1,0), (1,1,1)\rbrace$, find $\vec{v}$ in standard coordinates.

Click for solution
\[\vec{v} = 2(1,0,1) + (-1)(0,1,0) + 3(1,1,1) = (2,0,2) + (0,-1,0) + (3,3,3) = (5, 2, 5)\]

Q6. Find the coordinate vector of the matrix $A = \begin{bmatrix} 3 & 1 \ 2 & 4 \end{bmatrix}$ with respect to the standard basis of $M_{2 \times 2}(\mathbb{R})$:

\[B = \left\lbrace\begin{bmatrix}1&0\\0&0\end{bmatrix}, \begin{bmatrix}0&1\\0&0\end{bmatrix}, \begin{bmatrix}0&0\\1&0\end{bmatrix}, \begin{bmatrix}0&0\\0&1\end{bmatrix}\right\rbrace\]
Click for solution
\[A = 3\begin{bmatrix}1&0\\0&0\end{bmatrix} + 1\begin{bmatrix}0&1\\0&0\end{bmatrix} + 2\begin{bmatrix}0&0\\1&0\end{bmatrix} + 4\begin{bmatrix}0&0\\0&1\end{bmatrix}\] \[[A]_B = \begin{pmatrix} 3 \\ 1 \\ 2 \\ 4 \end{pmatrix}\]

Q7. Show that the coordinate mapping $\vec{v} \mapsto [\vec{v}]_B$ preserves addition: $[\vec{u} + \vec{v}]_B = [\vec{u}]_B + [\vec{v}]_B$.

Click for solution

Let $\vec{u} = \sum_{i=1}^n a_i \vec{v}i$ and $\vec{v} = \sum{i=1}^n b_i \vec{v}_i$ where $B = \lbrace\vec{v}_1, \ldots, \vec{v}_n\rbrace$.

Then $\vec{u} + \vec{v} = \sum_{i=1}^n (a_i + b_i) \vec{v}_i$.

By definition of coordinates:

\[[\vec{u}+\vec{v}]_B = \begin{pmatrix} a_1+b_1 \\ \vdots \\ a_n+b_n \end{pmatrix} = \begin{pmatrix} a_1 \\ \vdots \\ a_n \end{pmatrix} + \begin{pmatrix} b_1 \\ \vdots \\ b_n \end{pmatrix} = [\vec{u}]_B + [\vec{v}]_B \quad \blacksquare\]

Q8. Given $B = \lbrace(1,1), (1,2)\rbrace$ and $B’ = \lbrace(1,0), (0,1)\rbrace$. Find $P_{B’ \to B}$.

Click for solution

Set up $[B \mid B’]$:

\[\left[\begin{array}{cc|cc} 1 & 1 & 1 & 0 \\ 1 & 2 & 0 & 1 \end{array}\right]\]
$R_2 \leftarrow R_2 - R_1$: $\left[\begin{array}{cc cc} 1 & 1 & 1 & 0 \ 0 & 1 & -1 & 1 \end{array}\right]$
$R_1 \leftarrow R_1 - R_2$: $\left[\begin{array}{cc cc} 1 & 0 & 2 & -1 \ 0 & 1 & -1 & 1 \end{array}\right]$
\[P_{B' \to B} = \begin{bmatrix} 2 & -1 \\ -1 & 1 \end{bmatrix}\]

Check: $(3,4)$ in standard basis ($B’$) β†’ coordinates in $B$:

\[\begin{pmatrix} 2&-1\\-1&1 \end{pmatrix}\begin{pmatrix} 3\\4 \end{pmatrix} = \begin{pmatrix} 2\\1 \end{pmatrix}\]

Verify: $2(1,1) + 1(1,2) = (2,2)+(1,2) = (3,4)$ βœ“

Q9. If the transition matrix from $B$ to $B’$ is $P = \begin{bmatrix} 2 & 1 \ 3 & 2 \end{bmatrix}$, find the transition matrix from $B’$ to $B$.

Click for solution

$P_{B’ \to B} = P^{-1}$. For a $2 \times 2$ matrix:

\[\begin{pmatrix} a&b\\c&d \end{pmatrix}^{-1} = \frac{1}{ad-bc}\begin{pmatrix} d&-b\\-c&a \end{pmatrix}\]

$\det(P) = 4 - 3 = 1$.

\[P_{B' \to B} = \begin{bmatrix} 2 & -1 \\ -3 & 2 \end{bmatrix}\]

Q10. Find the coordinates of $\vec{w} = (7, -1, 3)$ with respect to $B = \lbrace(1,1,0), (0,1,1), (1,0,1)\rbrace$.

Click for solution

Solve $c_1(1,1,0) + c_2(0,1,1) + c_3(1,0,1) = (7,-1,3)$:

\[\begin{cases} c_1 + c_3 = 7 \\ c_1 + c_2 = -1 \\ c_2 + c_3 = 3 \end{cases}\]

Adding all three: $2(c_1+c_2+c_3) = 9 \Rightarrow c_1+c_2+c_3 = \frac{9}{2}$.

$c_2 = \frac{9}{2} - 7 = -\frac{5}{2}$, $c_3 = 3 - (-\frac{5}{2}) = \frac{11}{2}$, $c_1 = 7 - \frac{11}{2} = \frac{3}{2}$.

\[[\vec{w}]_B = \begin{pmatrix} 3/2 \\ -5/2 \\ 11/2 \end{pmatrix}\]

Verification: $\frac{3}{2}(1,1,0) + (-\frac{5}{2})(0,1,1) + \frac{11}{2}(1,0,1) = (\frac{3}{2},\frac{3}{2},0)+(0,-\frac{5}{2},-\frac{5}{2})+(\frac{11}{2},0,\frac{11}{2}) = (7,-1,3)$ βœ“


9 β€” Linear Transformations (Linear Mappings)


9.1 Motivation β€” What Are Linear Transformations?

The Big Picture: In Sections 1–8, we studied the spaces (vector spaces, subspaces, bases). Now we study the functions (maps) between spaces that preserve the linear structure.

A linear transformation is a function between vector spaces that respects addition and scalar multiplication. It is the most fundamental type of function in linear algebra.

Real-world examples of linear transformations:

  • Rotation of a 2D image (every pixel moves, but lines remain lines)
  • Scaling (making an image bigger or smaller)
  • Projection (casting a 3D shadow onto a 2D plane)
  • Differentiation $D: P_n \to P_{n-1}$, $D(p) = p’$ (derivative of a polynomial)
  • Matrix multiplication $T(\vec{x}) = A\vec{x}$ (the most common form!)

9.2 Formal Definition

Definition: Let $V$ and $W$ be vector spaces over the same field $F$. A function $T: V \to W$ is a linear transformation (or linear map) if it satisfies two properties for all $\vec{u}, \vec{v} \in V$ and all $c \in F$:

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

Equivalent single condition: $T(c_1\vec{u} + c_2\vec{v}) = c_1 T(\vec{u}) + c_2 T(\vec{v})$ for all scalars $c_1, c_2$ and all vectors $\vec{u}, \vec{v}$.

More generally (superposition principle):

\[T\left(\sum_{i=1}^k c_i \vec{v}_i\right) = \sum_{i=1}^k c_i \, T(\vec{v}_i)\]

Intuition β€” β€œLines map to lines”: A linear transformation sends straight lines to straight lines (or to $\vec{0}$) and the origin to the origin. It can rotate, reflect, stretch, shear, or project β€” but it can never curve, translate, or shift the origin.

What fails linearity?

  • $T(\vec{v}) = \vec{v} + (1,0)$ β€” NOT linear (translates the origin)
  • $T(x) = x^2$ β€” NOT linear ($T(2x) = 4x^2 \neq 2x^2 = 2T(x)$)
  • $T(x) = \sin(x)$ β€” NOT linear ($\sin(a+b) \neq \sin a + \sin b$)

9.3 Immediate Consequences of the Definition

Theorem: If $T: V \to W$ is linear, then:

  1. $T(\vec{0}_V) = \vec{0}_W$ β€” the zero vector always maps to the zero vector
  2. $T(-\vec{v}) = -T(\vec{v})$ β€” negatives are preserved
  3. $T(\vec{u} - \vec{v}) = T(\vec{u}) - T(\vec{v})$ β€” differences are preserved

Proof of (1): $T(\vec{0}) = T(0 \cdot \vec{v}) = 0 \cdot T(\vec{v}) = \vec{0}_W$.

Proof of (2): $T(-\vec{v}) = T((-1)\vec{v}) = (-1)T(\vec{v}) = -T(\vec{v})$.

Quick test: If $T(\vec{0}) \neq \vec{0}$, then $T$ is NOT linear. This is the fastest way to disprove linearity!


9.4 Examples of Linear Transformations

Example 1: Matrix Transformation β€” The Prototypical Example

Every matrix defines a linear transformation. If $A$ is an $m \times n$ matrix, then:

\[T: \mathbb{R}^n \to \mathbb{R}^m, \quad T(\vec{x}) = A\vec{x}\]

is linear.

Proof:

  • $T(\vec{u} + \vec{v}) = A(\vec{u} + \vec{v}) = A\vec{u} + A\vec{v} = T(\vec{u}) + T(\vec{v})$ βœ“
  • $T(c\vec{u}) = A(c\vec{u}) = cA\vec{u} = cT(\vec{u})$ βœ“

Example: Let $A = \begin{bmatrix} 2 & -1 \ 0 & 3 \end{bmatrix}$.

\[T\begin{pmatrix} x \\ y \end{pmatrix} = \begin{bmatrix} 2 & -1 \\ 0 & 3 \end{bmatrix}\begin{pmatrix} x \\ y \end{pmatrix} = \begin{pmatrix} 2x - y \\ 3y \end{pmatrix}\]

$T(1, 0) = (2, 0)$, $\;T(0, 1) = (-1, 3)$, $\;T(1, 1) = (1, 3) = T(1,0) + T(0,1)$ βœ“

Fundamental Theorem: Every linear transformation $T: \mathbb{R}^n \to \mathbb{R}^m$ can be represented as a matrix multiplication $T(\vec{x}) = A\vec{x}$, where:

\[A = \Big[ T(\vec{e}_1) \;\; T(\vec{e}_2) \;\; \cdots \;\; T(\vec{e}_n) \Big]\]

The columns of $A$ are the images of the standard basis vectors.

Example 2: Rotation in $\mathbb{R}^2$

Counter-clockwise rotation by angle $\theta$:

\[R_\theta: \mathbb{R}^2 \to \mathbb{R}^2, \quad R_\theta\begin{pmatrix}x\\y\end{pmatrix} = \begin{bmatrix}\cos\theta & -\sin\theta \\ \sin\theta & \cos\theta\end{bmatrix}\begin{pmatrix}x\\y\end{pmatrix}\]

Derivation: $R_\theta(\vec{e}1) = R\theta(1,0) = (\cos\theta, \sin\theta)$ and $R_\theta(\vec{e}2) = R\theta(0,1) = (-\sin\theta, \cos\theta)$.

$\theta = 90^\circ$:

\[R_{90^\circ} = \begin{bmatrix} 0 & -1 \\ 1 & 0 \end{bmatrix}\]

$R_{90^\circ}(1,0) = (0,1)$ βœ“, $R_{90^\circ}(0,1) = (-1,0)$ βœ“, $R_{90^\circ}(3,4) = (-4,3)$.

Linearity check: Rotation preserves addition and scaling (rotating a sum equals the sum of rotations).

Example 3: Projection

Projection onto the $x$-axis:

\[P: \mathbb{R}^2 \to \mathbb{R}^2, \quad P(x, y) = (x, 0)\]

Matrix: $P = \begin{bmatrix} 1 & 0 \ 0 & 0 \end{bmatrix}$

Linearity:

  • $P(x_1+x_2, y_1+y_2) = (x_1+x_2, 0) = (x_1,0) + (x_2,0) = P(x_1,y_1)+P(x_2,y_2)$ βœ“
  • $P(cx, cy) = (cx, 0) = c(x, 0) = cP(x,y)$ βœ“

Interesting property: $P^2 = P$ (projecting twice gives the same result β€” called idempotent).

Example 4: Reflection

Reflection about the $x$-axis:

\[T(x,y) = (x, -y)\]

Matrix: $T = \begin{bmatrix} 1 & 0 \ 0 & -1 \end{bmatrix}$

This is linear: $T(1,0) = (1,0)$, $T(0,1) = (0,-1)$.

Reflection about the line $y = x$:

\[T(x,y) = (y, x)\]

Matrix: $T = \begin{bmatrix} 0 & 1 \ 1 & 0 \end{bmatrix}$

Example 5: Scaling (Dilation)

Uniform scaling by factor $k$:

\[T(\vec{v}) = k\vec{v}\]

Matrix: $T = kI$ (scalar multiple of the identity).

For $k = 2$ in $\mathbb{R}^2$: $T = \begin{bmatrix} 2 & 0 \ 0 & 2 \end{bmatrix}$. Every vector is doubled.

Non-uniform scaling: $T(x,y) = (2x, 3y)$ β€” stretches differently in $x$ and $y$.

Matrix: $T = \begin{bmatrix} 2 & 0 \ 0 & 3 \end{bmatrix}$.

Example 6: Differentiation β€” A Linear Transformation on Polynomial Spaces

Claim: $D: P_n \to P_{n-1}$ defined by $D(p) = p’$ (the derivative) is linear.

Proof:

  • $D(p + q) = (p+q)’ = p’ + q’ = D(p) + D(q)$ βœ“ (sum rule of differentiation)
  • $D(cp) = (cp)’ = cp’ = cD(p)$ βœ“ (constant multiple rule)

Examples:

  • $D(3x^2 + 5x + 1) = 6x + 5$
  • $D(x^3) = 3x^2$
  • $D(7) = 0$ (constants map to the zero polynomial)

Matrix representation (with respect to standard bases $\lbrace 1, x, x^2\rbrace$ β†’ $\lbrace 1, x\rbrace$):

$D(1) = 0 = 0 \cdot 1 + 0 \cdot x$, $\;D(x) = 1 = 1 \cdot 1 + 0 \cdot x$, $\;D(x^2) = 2x = 0 \cdot 1 + 2 \cdot x$.

\[[D]_B = \begin{bmatrix} 0 & 1 & 0 \\ 0 & 0 & 2 \end{bmatrix}\]

Example 7: Integration β€” Also Linear!

$I: P_2 \to P_3$ defined by $I(p) = \int_0^x p(t)\,dt$ is linear.

Proof: Integral of a sum = sum of integrals, and integral of $cp$ = $c$ times integral of $p$.

$I(1) = x$, $\;I(x) = \frac{x^2}{2}$, $\;I(x^2) = \frac{x^3}{3}$.

\[[I]_B = \begin{bmatrix} 0 & 0 & 0 \\ 1 & 0 & 0 \\ 0 & 1/2 & 0 \\ 0 & 0 & 1/3 \end{bmatrix}\]

Example 8: Non-Linear Functions (Counterexamples)

NOT linear β€” Translation: $T(x,y) = (x+1, y+2)$

$T(0,0) = (1,2) \neq (0,0)$. Fails immediately!

NOT linear β€” Quadratic: $T(x) = x^2$

$T(2) = 4$ but $2T(1) = 2$. $T(2) \neq 2T(1)$. Fails homogeneity.

NOT linear β€” Absolute value: $T(x) = x $

$T(-1 + 1) = T(0) = 0$ but $T(-1) + T(1) = 1 + 1 = 2$. Fails additivity.

NOT linear β€” Norm: $T(\vec{v}) = \lVert\vec{v}\rVert$

$T(2\vec{e}_1) = 2$ but $2T(\vec{e}_1) = 2$… wait, this works for this case. But try $T(\vec{e}_1 + \vec{e}_2) = \sqrt{2} \neq 1+1 = T(\vec{e}_1)+T(\vec{e}_2)$. Fails additivity.


9.5 Kernel (Null Space) and Image (Range)

Definition: Let $T: V \to W$ be a linear transformation.

Kernel (Null Space): The set of all vectors that map to $\vec{0}$:

\[\ker(T) = \lbrace\vec{v} \in V : T(\vec{v}) = \vec{0}_W\rbrace\]

Image (Range): The set of all possible outputs:

\[\text{Im}(T) = \lbrace T(\vec{v}) : \vec{v} \in V\rbrace = \lbrace w \in W : w = T(\vec{v}) \text{ for some } \vec{v} \in V\rbrace\]

Theorem:

  1. $\ker(T)$ is a subspace of $V$
  2. $\text{Im}(T)$ is a subspace of $W$

Proof of (1):

  • $T(\vec{0}) = \vec{0}$, so $\vec{0} \in \ker(T)$ βœ“
  • If $T(\vec{u}) = \vec{0}$ and $T(\vec{v}) = \vec{0}$, then $T(\vec{u}+\vec{v}) = T(\vec{u})+T(\vec{v}) = \vec{0}$ βœ“
  • If $T(\vec{v}) = \vec{0}$, then $T(c\vec{v}) = cT(\vec{v}) = c\vec{0} = \vec{0}$ βœ“

Proof of (2):

  • $T(\vec{0}) = \vec{0} \in \text{Im}(T)$ βœ“
  • If $\vec{w}_1 = T(\vec{u})$ and $\vec{w}_2 = T(\vec{v})$, then $\vec{w}_1+\vec{w}_2 = T(\vec{u})+T(\vec{v}) = T(\vec{u}+\vec{v}) \in \text{Im}(T)$ βœ“
  • If $\vec{w} = T(\vec{v})$, then $c\vec{w} = cT(\vec{v}) = T(c\vec{v}) \in \text{Im}(T)$ βœ“

Example: Finding Kernel and Image

Let $T: \mathbb{R}^3 \to \mathbb{R}^2$ defined by $T(x,y,z) = (x+y, y+z)$.

Matrix: $A = \begin{bmatrix} 1 & 1 & 0 \ 0 & 1 & 1 \end{bmatrix}$

Kernel: Solve $A\vec{x} = \vec{0}$:

\[\begin{cases} x + y = 0 \\ y + z = 0 \end{cases}\]

$y = -x$, $z = -y = x$. So $\ker(T) = \lbrace t(1, -1, 1) : t \in \mathbb{R}\rbrace$.

$\dim(\ker(T)) = 1$. Basis: $\lbrace(1,-1,1)\rbrace$.

Image: The column space of $A$.

Columns: $(1,0)$ and $(1,1)$ are independent β†’ $\text{Im}(T) = \mathbb{R}^2$ (full image).

$\dim(\text{Im}(T)) = 2$.

Rank-Nullity check: $\dim(\ker(T)) + \dim(\text{Im}(T)) = 1 + 2 = 3 = \dim(\mathbb{R}^3)$ βœ“


9.6 Rank-Nullity Theorem for Linear Transformations

Rank-Nullity Theorem (General Version): If $T: V \to W$ is a linear transformation and $V$ is finite-dimensional, then:

\[\boxed{\dim(\ker(T)) + \dim(\text{Im}(T)) = \dim(V)}\]

or equivalently:

\[\text{nullity}(T) + \text{rank}(T) = \dim(V)\]

Consequences:

Condition What it means
$\ker(T) = \lbrace\vec{0}\rbrace$ $T$ is injective (one-to-one): different inputs give different outputs
$\text{Im}(T) = W$ $T$ is surjective (onto): every output is achievable
Both $T$ is a bijection (isomorphism): perfect correspondence
$\dim(V) = \dim(W)$ and $T$ is injective $T$ is automatically surjective (and vice versa)

9.7 Matrix Representation of Linear Transformations

Definition: Let $T: V \to W$ be linear. Let $B$ and $B’$ be ordered bases:

\[B = \lbrace\vec{v}_1, \ldots, \vec{v}_n\rbrace \;\text{(basis of }V\text{)}, \quad B' = \lbrace\vec{w}_1, \ldots, \vec{w}_m\rbrace \;\text{(basis of }W\text{)}\]

The matrix representation of $T$ relative to $B$ and $B’$ is:

\[[T]_{B}^{B'} = \Big[ [T(\vec{v}_1)]_{B'} \;\; [T(\vec{v}_2)]_{B'} \;\; \cdots \;\; [T(\vec{v}_n)]_{B'} \Big]\]

The $j$-th column is the coordinate vector of $T(\vec{v}_j)$ with respect to the output basis $B’$.

Key relationship:

\[[T(\vec{v})]_{B'} = [T]_B^{B'} \cdot [\vec{v}]_B\]

Example: Matrix Representation

$T: \mathbb{R}^2 \to \mathbb{R}^2$ defined by $T(x,y) = (2x+y, x-y)$.

Standard bases: $B = B’ = \lbrace(1,0), (0,1)\rbrace$

$T(1,0) = (2,1)$, $\;T(0,1) = (1,-1)$

\[[T]_B^{B'} = \begin{bmatrix} 2 & 1 \\ 1 & -1 \end{bmatrix}\]

Non-standard basis example: Let $B = \lbrace(1,1), (1,-1)\rbrace$ and $B’ = \lbrace(1,0), (0,1)\rbrace$ (standard).

$T(1,1) = (2+1, 1-1) = (3, 0)$. In $B’$: $[T(1,1)]_{B’} = (3,0)^T$.

$T(1,-1) = (2-1, 1+1) = (1, 2)$. In $B’$: $[T(1,-1)]_{B’} = (1,2)^T$.

\[[T]_B^{B'} = \begin{bmatrix} 3 & 1 \\ 0 & 2 \end{bmatrix}\]

This is a different matrix representing the same transformation β€” just in different coordinates!


9.8 Properties of Linear Transformations

Theorem (Injectivity): $T$ is injective (one-to-one) if and only if $\ker(T) = \lbrace\vec{0}\rbrace$.

Proof (⟹): If $T$ is injective and $T(\vec{v}) = \vec{0} = T(\vec{0})$, then $\vec{v} = \vec{0}$.

Proof (⟸): If $T(\vec{u}) = T(\vec{v})$, then $T(\vec{u} - \vec{v}) = \vec{0}$, so $\vec{u} - \vec{v} \in \ker(T) = \lbrace\vec{0}\rbrace$, giving $\vec{u} = \vec{v}$.

Theorem (Composition): If $T_1: U \to V$ and $T_2: V \to W$ are linear, then $T_2 \circ T_1: U \to W$ is also linear, and:

\[[T_2 \circ T_1] = [T_2] \cdot [T_1]\]

Matrix multiplication corresponds to the composition of linear transformations!

Theorem (Invertibility): A linear transformation $T: V \to V$ is invertible if and only if:

  • $\ker(T) = \lbrace\vec{0}\rbrace$ (injective), and
  • $\text{Im}(T) = V$ (surjective)

When $V$ is finite-dimensional, these are equivalent (injective ⟺ surjective ⟺ bijective).

The inverse $T^{-1}$ is also linear, and $[T^{-1}] = [T]^{-1}$.


9.9 Special Types of Linear Transformations

Type Definition Example
Zero transformation $T(\vec{v}) = \vec{0}$ for all $\vec{v}$ $\ker = V$, $\text{Im} = \lbrace\vec{0}\rbrace$
Identity transformation $T(\vec{v}) = \vec{v}$ for all $\vec{v}$ $\ker = \lbrace\vec{0}\rbrace$, $\text{Im} = V$
Scalar transformation $T(\vec{v}) = c\vec{v}$ Matrix $= cI$
Projection $T^2 = T$ (idempotent) Projection onto a subspace
Isomorphism Bijective linear map Invertible matrix transformation
Endomorphism $T: V \to V$ (same domain and codomain) Rotation, reflection

9.10 Determining a Linear Transformation from Its Action on a Basis

Key Theorem: A linear transformation $T: V \to W$ is completely determined by its values on a basis of $V$.

If $B = \lbrace\vec{v}_1, \ldots, \vec{v}_n\rbrace$ is a basis and you know $T(\vec{v}_1), T(\vec{v}_2), \ldots, T(\vec{v}_n)$, then for any $\vec{v} = c_1\vec{v}_1 + \cdots + c_n\vec{v}_n$:

\[T(\vec{v}) = c_1 T(\vec{v}_1) + c_2 T(\vec{v}_2) + \cdots + c_n T(\vec{v}_n)\]

Example: Find $T: \mathbb{R}^2 \to \mathbb{R}^2$ if $T(1,0) = (3,1)$ and $T(0,1) = (2,-1)$.

Since $\lbrace(1,0), (0,1)\rbrace$ is a basis: for any $(x,y)$:

\[T(x,y) = xT(1,0) + yT(0,1) = x(3,1) + y(2,-1) = (3x+2y, x-y)\]

Matrix: $A = \begin{bmatrix} 3 & 2 \ 1 & -1 \end{bmatrix}$

Find $T(5,3)$: $T(5,3) = (15+6, 5-3) = (21, 2)$ βœ“

Example (non-standard basis): Let $B = \lbrace(1,1), (1,-1)\rbrace$ be a basis of $\mathbb{R}^2$. Suppose $T(1,1) = (2,0)$ and $T(1,-1) = (0,4)$.

Find $T(3,1)$.

First, express $(3,1)$ in basis $B$: $(3,1) = 2(1,1) + 1(1,-1)$.

Then: $T(3,1) = 2T(1,1) + 1T(1,-1) = 2(2,0) + 1(0,4) = (4,4)$.


9.11 Singular and Non-Singular Mappings

Definition: A linear map $F: V \to U$ is called singular if there exists a non-zero vector $\vec{v} \in V$ such that $F(\vec{v}) = \vec{0}$. Otherwise, $F$ is non-singular.

Equivalently, $F$ is non-singular if and only if $\ker(F) = \lbrace\vec{0}\rbrace$.

Theorem (Non-Singular Maps Preserve LI): Let $F: V \to U$ be a non-singular linear map. If $\lbrace\vec{v}_1, \ldots, \vec{v}_n\rbrace$ is linearly independent in $V$, then $\lbrace F(\vec{v}_1), \ldots, F(\vec{v}_n)\rbrace$ is linearly independent in $U$.

Non-singular maps never β€œcollapse” independent vectors β€” they preserve all the independence structure.

Theorem (Isomorphism via Non-Singularity): Suppose $\dim(V) = \dim(U)$ and $F: V \to U$ is linear. Then $F$ is an isomorphism if and only if $F$ is non-singular.

(When domain and codomain have the same dimension, injectivity automatically gives surjectivity.)


9.12 Operations on Linear Maps

Let $F, G: V \to U$ be linear maps and $k \in \mathbb{R}$. Define:

  • Sum: $(F + G)(\vec{v}) = F(\vec{v}) + G(\vec{v})$
  • Scalar Multiple: $(kF)(\vec{v}) = kF(\vec{v})$

Theorem: If $F, G: V \to U$ are linear, then $(F + G)$ and $(kF)$ are also linear.

Proof for $(kF)$: Let $\vec{u}, \vec{v} \in V$ and $\alpha, \beta \in \mathbb{R}$.

\[\text{LHS} = (kF)(\alpha\vec{u} + \beta\vec{v}) = k[F(\alpha\vec{u} + \beta\vec{v})] = k[\alpha F(\vec{u}) + \beta F(\vec{v})]\] \[= \alpha \cdot kF(\vec{u}) + \beta \cdot kF(\vec{v}) = \alpha(kF)(\vec{u}) + \beta(kF)(\vec{v}) = \text{RHS} \quad \blacksquare\]

$\text{Hom}(V, U)$ β€” The Space of All Linear Maps:

The collection of all linear maps from $V$ to $U$, denoted $\text{Hom}(V, U)$, with the above operations of addition and scalar multiplication, forms a vector space over $\mathbb{R}$.

The zero element is the zero mapping $O: V \to U$ defined by $O(\vec{v}) = \vec{0}_U$ for all $\vec{v} \in V$.

Theorem: If $\dim(V) = m$ and $\dim(U) = n$, then:

\[\dim(\text{Hom}(V, U)) = m \cdot n\]

9.13 Composition Properties

Let $V, U, W$ be vector spaces over $\mathbb{R}$. Suppose $F, F’: V \to U$ and $G, G’: U \to W$ are linear maps and $k \in \mathbb{R}$. Then:

(a) $G \circ (F + F’) = (G \circ F) + (G \circ F’)$ (left distributivity)

(b) $(G + G’) \circ F = (G \circ F) + (G’ \circ F)$ (right distributivity)

(c) $k(G \circ F) = (kG) \circ F = G \circ (kF)$


9.14 Linear Operators and Algebra

A linear map $F: V \to V$ (same domain and codomain) is called a linear operator on $V$.

The set of all linear operators on $V$ is denoted $A(V)$ (or $\text{End}(V)$). This is a vector space with $\dim(A(V)) = m^2$ where $m = \dim(V)$.

Since domain = codomain, the composition $F \circ G$ is always defined for $F, G \in A(V)$, and $F \circ G \in A(V)$. This gives $A(V)$ the structure of an algebra β€” a vector space with a multiplication operation satisfying:

  1. $F(G + H) = FG + FH$ (left distributivity)
  2. $(G + H)F = GF + HF$ (right distributivity)
  3. $k(GF) = (kG)F = G(kF)$
  4. $F(GH) = (FG)H$ (associativity)

9.15 Worked Examples β€” LT Operations

Example: Determine Singular or Non-Singular, Find Inverse

(i) $F: \mathbb{R}^2 \to \mathbb{R}^2$ defined by $F(x, y) = (x - y, x - 2y)$.

Find $\ker(F)$: Set $F(x, y) = (0, 0)$:

\[x - y = 0, \quad x - 2y = 0 \implies x = y \text{ and } x = 2y \implies x = y = 0\]

$\ker(F) = \lbrace\vec{0}\rbrace$ $\Rightarrow$ $F$ is non-singular.

Find $F^{-1}$: Set $F(x, y) = (a, b)$:

\[x - y = a, \quad x - 2y = b\]

Subtracting: $y = a - b$. Then $x = a + y = 2a - b$.

\[F^{-1}(a, b) = (2a - b, a - b)\]

(ii) $G: \mathbb{R}^2 \to \mathbb{R}^2$ defined by $G(x, y) = (2x - 4y, 3x - 6y)$.

Find $\ker(G)$: $2x = 4y$ and $3x = 6y$ $\implies$ $x = 2y$.

$\ker(G) = \lbrace(2y, y) : y \in \mathbb{R}\rbrace \neq \lbrace\vec{0}\rbrace$ $\Rightarrow$ $G$ is singular.

Non-zero vector in kernel: e.g., $(2, 1)$.

Example: Operations $F + G$, $3F$, $2F - 5G$

Let $F: \mathbb{R}^3 \to \mathbb{R}^2$ be defined by $F(x, y, z) = (2x, y + z)$.

Let $G: \mathbb{R}^3 \to \mathbb{R}^2$ be defined by $G(x, y, z) = (x + y, x - z)$.

(i) $(F + G)(x, y, z) = F(x,y,z) + G(x,y,z) = (2x, y+z) + (x+y, x-z) = (3x + y, x + y)$

(ii) $(3F)(x, y, z) = 3(2x, y+z) = (6x, 3y + 3z)$

(iii) $(2F - 5G)(x, y, z) = 2(2x, y+z) - 5(x+y, x-z) = (4x - 5x - 5y, 2y + 2z - 5x + 5z) = (-x - 5y, -5x + 2y + 7z)$

Example: Composition β€” $F \circ G$ vs $G \circ F$

Let $F: \mathbb{R}^3 \to \mathbb{R}^2$ defined by $F(x, y, z) = (2x, y + z)$.

Let $G: \mathbb{R}^2 \to \mathbb{R}^3$ defined by $G(x, y) = (y, x, x + y)$.

Is $F \circ G$ defined? $G: \mathbb{R}^2 \to \mathbb{R}^3$ and $F: \mathbb{R}^3 \to \mathbb{R}^2$. So $F \circ G: \mathbb{R}^2 \to \mathbb{R}^2$. Yes, it is defined.

$(F \circ G)(x, y) = F(G(x, y)) = F(y, x, x+y) = (2y, x + x + y) = (2y, 2x + y)$

Is $G \circ F$ defined? $F: \mathbb{R}^3 \to \mathbb{R}^2$ and $G: \mathbb{R}^2 \to \mathbb{R}^3$. So $G \circ F: \mathbb{R}^3 \to \mathbb{R}^3$. Yes, also defined.

$(G \circ F)(x, y, z) = G(F(x,y,z)) = G(2x, y+z) = (y+z, 2x, 2x + y + z)$

Note: Even though both compositions exist, $F \circ G: \mathbb{R}^2 \to \mathbb{R}^2$ and $G \circ F: \mathbb{R}^3 \to \mathbb{R}^3$ have different domain/codomain, so they are different types of maps entirely!

Example: Non-Singular Map Between Spaces of Different Dimension

$G: \mathbb{R}^2 \to \mathbb{R}^3$ defined by $G(x, y) = (x + y, x - 2y, 3x + y)$.

Find $\ker(G)$: $x + y = 0$, $x - 2y = 0$, $3x + y = 0$.

From $x + y = 0$ and $x - 2y = 0$: $y = -x$ and $x = 2y = -2x$, so $3x = 0$, hence $x = y = 0$.

$\ker(G) = \lbrace\vec{0}\rbrace$ $\Rightarrow$ $G$ is non-singular.

However, since $\dim(\mathbb{R}^2) = 2 \neq 3 = \dim(\mathbb{R}^3)$, $G$ cannot be an isomorphism (not surjective β€” image is a 2D subspace of $\mathbb{R}^3$).


Practice Questions β€” Linear Transformations

Q1. Is $T: \mathbb{R}^2 \to \mathbb{R}^2$ defined by $T(x,y) = (x+y, x-y)$ a linear transformation? Verify both conditions.

Click for solution

Additivity: Let $\vec{u} = (x_1,y_1)$, $\vec{v} = (x_2,y_2)$.

$T(\vec{u}+\vec{v}) = T(x_1+x_2, y_1+y_2) = (x_1+x_2+y_1+y_2, \; x_1+x_2-y_1-y_2)$

$T(\vec{u})+T(\vec{v}) = (x_1+y_1, x_1-y_1) + (x_2+y_2, x_2-y_2) = (x_1+x_2+y_1+y_2, \; x_1+x_2-y_1-y_2)$ βœ“

Homogeneity: $T(c\vec{u}) = T(cx_1, cy_1) = (cx_1+cy_1, cx_1-cy_1) = c(x_1+y_1, x_1-y_1) = cT(\vec{u})$ βœ“

Yes, $T$ is linear. Matrix: $A = \begin{bmatrix} 1 & 1 \ 1 & -1 \end{bmatrix}$.

Q2. Is $T: \mathbb{R}^2 \to \mathbb{R}^2$ defined by $T(x,y) = (x+1, y)$ a linear transformation?

Click for solution

No. Check: $T(0,0) = (1, 0) \neq (0,0)$.

Since $T$ does not map the zero vector to the zero vector, it is not linear.

Alternatively: $T(1,0) + T(1,0) = (2,0) + (2,0) = (4,0)$, but $T(2,0) = (3,0)$. Fails additivity.

Q3. Let $T: \mathbb{R}^3 \to \mathbb{R}^2$ be defined by $T(x,y,z) = (x+2y, y-z)$. Find: (a) $T(1,0,0)$, $T(0,1,0)$, $T(0,0,1)$ (b) The matrix of $T$ (c) $\ker(T)$ and $\text{Im}(T)$

Click for solution

(a) $T(1,0,0) = (1,0)$, $T(0,1,0) = (2,1)$, $T(0,0,1) = (0,-1)$.

(b) $A = \begin{bmatrix} 1 & 2 & 0 \ 0 & 1 & -1 \end{bmatrix}$

(c) Kernel: Solve $A\vec{x} = \vec{0}$:

$x + 2y = 0$ and $y - z = 0$. So $y = z$, $x = -2z$.

$\ker(T) = \lbrace t(-2, 1, 1) : t \in \mathbb{R}\rbrace$, $\dim(\ker T) = 1$.

Image: Columns $(1,0)$ and $(2,1)$ are independent β†’ $\text{Im}(T) = \mathbb{R}^2$ (rank 2).

Rank-Nullity: $1 + 2 = 3$ βœ“

Q4. Show that differentiation $D: P_3 \to P_2$ defined by $D(p) = p’$ is linear. Find its matrix with respect to bases $\lbrace 1, x, x^2, x^3\rbrace$ and $\lbrace 1, x, x^2\rbrace$.

Click for solution

Linearity: $D(p+q) = (p+q)’ = p’+q’ = D(p)+D(q)$ βœ“. $D(cp) = (cp)’ = cp’ = cD(p)$ βœ“.

Matrix: $D(1) = 0, \; D(x) = 1, \; D(x^2) = 2x, \; D(x^3) = 3x^2$.

Express in basis $\lbrace1, x, x^2\rbrace$:

  • $D(1) = 0 \cdot 1 + 0 \cdot x + 0 \cdot x^2$
  • $D(x) = 1 \cdot 1 + 0 \cdot x + 0 \cdot x^2$
  • $D(x^2) = 0 \cdot 1 + 2 \cdot x + 0 \cdot x^2$
  • $D(x^3) = 0 \cdot 1 + 0 \cdot x + 3 \cdot x^2$
\[[D] = \begin{bmatrix} 0 & 1 & 0 & 0 \\ 0 & 0 & 2 & 0 \\ 0 & 0 & 0 & 3 \end{bmatrix}\]

Kernel: $D(p) = 0 \Rightarrow p = c$ (constants). $\ker(D) = \text{Span}\lbrace 1\rbrace$, dim = 1.

Image: $\text{Im}(D) = P_2$ (every polynomial of degree $\leq 2$ is a derivative of something).

Rank-Nullity: $1 + 3 = 4 = \dim(P_3)$ βœ“

Q5. Let $T: \mathbb{R}^2 \to \mathbb{R}^3$ be defined by $T(x,y) = (x, x+y, y)$. Find $\ker(T)$ and determine if $T$ is injective.

Click for solution

Kernel: $T(x,y) = (0,0,0)$ means $x = 0$, $x+y = 0$, $y = 0$. So $x = y = 0$.

$\ker(T) = \lbrace(0,0)\rbrace = \lbrace\vec{0}\rbrace$.

Since $\ker(T) = \lbrace\vec{0}\rbrace$, $T$ is injective (one-to-one) βœ“.

$T$ is not surjective (can’t hit all of $\mathbb{R}^3$ from $\mathbb{R}^2$). rank = 2, nullity = 0, $\dim V = 2$ βœ“.

Q6. Find the matrix of the linear transformation $T: \mathbb{R}^2 \to \mathbb{R}^2$ that (a) reflects about the line $y = x$, and (b) then scales by factor 3.

Click for solution

Reflection about $y = x$: $R = \begin{bmatrix} 0 & 1 \ 1 & 0 \end{bmatrix}$

Scaling by 3: $S = \begin{bmatrix} 3 & 0 \ 0 & 3 \end{bmatrix}$

Composition (scaling after reflection): $T = S \circ R$, so:

\[[T] = SR = \begin{bmatrix} 3 & 0 \\ 0 & 3 \end{bmatrix}\begin{bmatrix} 0 & 1 \\ 1 & 0 \end{bmatrix} = \begin{bmatrix} 0 & 3 \\ 3 & 0 \end{bmatrix}\]

Check: $T(1,0) = (0,3)$ (reflect to $(0,1)$, then scale to $(0,3)$) βœ“

Q7. Let $T: \mathbb{R}^3 \to \mathbb{R}^3$ be defined by $T(x,y,z) = (x+y, y+z, x+z)$. Find: (a) The matrix $A$ of $T$ (b) $\ker(T)$ (c) Whether $T$ is an isomorphism

Click for solution

(a) $T(1,0,0)=(1,0,1)$, $T(0,1,0)=(1,1,0)$, $T(0,0,1)=(0,1,1)$.

\[A = \begin{bmatrix} 1&1&0\\0&1&1\\1&0&1 \end{bmatrix}\]

(b) $\det(A) = 1(1-0) - 1(0-1) + 0 = 1+1 = 2 \neq 0$.

Since $\det(A) \neq 0$, $\ker(T) = \lbrace\vec{0}\rbrace$.

(c) $T$ is injective ($\ker = \lbrace\vec{0}\rbrace$) and $\dim(V) = \dim(W) = 3$, so $T$ is also surjective.

$T$ is a bijection β†’ it is an isomorphism βœ“.

Q8. If $T(1,0) = (2,3)$ and $T(0,1) = (-1,4)$, find $T(3,5)$.

Click for solution

$(3,5) = 3(1,0) + 5(0,1)$.

$T(3,5) = 3T(1,0) + 5T(0,1) = 3(2,3) + 5(-1,4) = (6,9) + (-5,20) = (1, 29)$.

Q9. Prove that if $T: V \to W$ is linear and $S = \lbrace\vec{v}_1, \ldots, \vec{v}_k\rbrace$ is linearly dependent in $V$, then $\lbrace T(\vec{v}_1), \ldots, T(\vec{v}_k)\rbrace$ is linearly dependent in $W$.

Click for solution

Since $S$ is dependent, there exist scalars $c_1, \ldots, c_k$, not all zero, such that:

\[c_1\vec{v}_1 + c_2\vec{v}_2 + \cdots + c_k\vec{v}_k = \vec{0}\]

Apply $T$ to both sides (using linearity):

\[T(c_1\vec{v}_1 + \cdots + c_k\vec{v}_k) = T(\vec{0}) = \vec{0}\] \[c_1 T(\vec{v}_1) + c_2 T(\vec{v}_2) + \cdots + c_k T(\vec{v}_k) = \vec{0}\]

Since the same $c_i$ (not all zero) give a zero combination, $\lbrace T(\vec{v}_1), \ldots, T(\vec{v}_k)\rbrace$ is linearly dependent. $\blacksquare$

Note: The converse is NOT true β€” a linear transformation can make independent vectors dependent (e.g., projection collapses a dimension).

Q10. Let $T: \mathbb{R}^3 \to \mathbb{R}^3$ be defined by $T(x,y,z) = (x-y, y-z, z-x)$. Show that $T$ is not injective and find a basis for $\ker(T)$.

Click for solution

Matrix:

\[A = \begin{bmatrix} 1&-1&0\\0&1&-1\\-1&0&1 \end{bmatrix}\]

$\det(A) = 1(1-0) - (-1)(0-1) + 0 = 1 - 1 = 0$.

Since $\det(A) = 0$, $T$ is not injective (and not surjective either).

Finding $\ker(T)$: Solve $A\vec{x} = \vec{0}$:

$R_3 \leftarrow R_3 + R_1$:

\[\begin{bmatrix} 1&-1&0\\0&1&-1\\0&-1&1 \end{bmatrix}\]

$R_3 \leftarrow R_3 + R_2$:

\[\begin{bmatrix} 1&-1&0\\0&1&-1\\0&0&0 \end{bmatrix}\]

$z = t$ (free), $y = t$, $x = t$.

$\ker(T) = \lbrace t(1,1,1) : t \in \mathbb{R}\rbrace$. Basis: $\lbrace(1,1,1)\rbrace$, nullity = 1.

Geometric interpretation: $T$ maps every point on the line $x = y = z$ to the origin.

Q11. Let $T: P_2 \to P_1$ be defined by $T(a + bx + cx^2) = (a+b) + (b+c)x$. Find the matrix of $T$ with respect to standard bases and find $\ker(T)$.

Click for solution

Standard bases: $B = \lbrace 1, x, x^2\rbrace$ for $P_2$, $B’ = \lbrace 1, x\rbrace$ for $P_1$.

$T(1) = 1 + 0 \cdot x$, $T(x) = 1 + 1 \cdot x$, $T(x^2) = 0 + 1 \cdot x$.

\[[T] = \begin{bmatrix} 1 & 1 & 0 \\ 0 & 1 & 1 \end{bmatrix}\]

Kernel: $T(a+bx+cx^2) = 0$ means $a+b = 0$ and $b+c = 0$.

$b = -a$, $c = -b = a$. So $p(x) = a - ax + ax^2 = a(1 - x + x^2)$.

$\ker(T) = \text{Span}\lbrace 1 - x + x^2\rbrace$, dim = 1.

Rank-Nullity: $1 + 2 = 3 = \dim(P_2)$ βœ“

Q12. Determine the linear transformation $T: \mathbb{R}^3 \to \mathbb{R}^3$ such that $T(1,0,0) = (1,1,0)$, $T(0,1,0) = (0,1,1)$, $T(0,0,1) = (1,0,1)$. Compute $T(2,3,-1)$.

Click for solution

Matrix: $A = \begin{bmatrix} 1&0&1\1&1&0\0&1&1 \end{bmatrix}$

$T(2,3,-1) = A\begin{pmatrix}2\3\-1\end{pmatrix} = \begin{pmatrix} 2+0-1\2+3+0\0+3-1 \end{pmatrix} = \begin{pmatrix} 1\5\2 \end{pmatrix}$

Verification: $2T(1,0,0) + 3T(0,1,0) + (-1)T(0,0,1) = 2(1,1,0) + 3(0,1,1) + (-1)(1,0,1) = (2,2,0)+(0,3,3)+(-1,0,-1) = (1,5,2)$ βœ“

Q13. Let $T: M_{2\times 2}(\mathbb{R}) \to \mathbb{R}$ be defined by $T(A) = \text{tr}(A)$ (the trace β€” sum of diagonal entries). Show that $T$ is linear and find $\ker(T)$.

Click for solution

Linearity:

$T(A+B) = \text{tr}(A+B) = \text{tr}(A) + \text{tr}(B) = T(A) + T(B)$ βœ“

$T(cA) = \text{tr}(cA) = c\,\text{tr}(A) = cT(A)$ βœ“

Kernel:

\[\ker(T) = \lbrace A \in M_{2\times 2} : \text{tr}(A) = 0\rbrace = \left\lbrace\begin{bmatrix} a&b\\c&-a \end{bmatrix} : a,b,c \in \mathbb{R}\right\rbrace\]

This has 3 free parameters, so $\dim(\ker T) = 3$.

Basis:

\[\left\lbrace\begin{bmatrix}1&0\\0&-1\end{bmatrix}, \begin{bmatrix}0&1\\0&0\end{bmatrix}, \begin{bmatrix}0&0\\1&0\end{bmatrix}\right\rbrace\]

Rank-Nullity: $3 + 1 = 4 = \dim(M_{2\times 2})$ βœ“ ($\text{Im}(T) = \mathbb{R}$, dim 1)

Q14. Find a linear transformation $T: \mathbb{R}^3 \to \mathbb{R}^3$ whose image is $\text{Span}\lbrace(1,0,1), (0,1,1)\rbrace$ and kernel is $\text{Span}\lbrace(1,1,-1)\rbrace$.

Click for solution

Strategy: We need rank = 2, nullity = 1, with the specified kernel and image.

Complete $(1,1,-1)$ to a basis of $\mathbb{R}^3$: take $\lbrace(1,0,0), (0,1,0), (1,1,-1)\rbrace$ (verify independence: $\det = -1 \neq 0$).

Define $T$ on a basis. Let $T(1,1,-1) = (0,0,0)$ (kernel), $T(1,0,0) = (1,0,1)$ (first image basis vector), $T(0,1,0) = (0,1,1)$ (second image basis vector).

Express standard basis in terms of our basis:

$(1,0,0) = 1 \cdot (1,0,0)$ $(0,1,0) = 1 \cdot (0,1,0)$ $(0,0,1) = -1 \cdot (1,0,0) - 1 \cdot (0,1,0) + 1 \cdot (1,1,-1)$

So $T(0,0,1) = -T(1,0,0) - T(0,1,0) + T(1,1,-1) = -(1,0,1) - (0,1,1) + (0,0,0) = (-1,-1,-2)$.

Matrix: $A = \begin{bmatrix} 1&0&-1\0&1&-1\1&1&-2 \end{bmatrix}$

Verify kernel: $A(1,1,-1)^T = (1+0+1, 0+1+1, 1+1+2)^T = … $ let me recheck:

$A\begin{pmatrix}1\1\-1\end{pmatrix} = \begin{pmatrix}1+0+1\0+1+1\1+1+2\end{pmatrix} = \begin{pmatrix}2\2\4\end{pmatrix}$

That’s wrong. Let me redo. I need to express the standard basis vectors in terms of $\lbrace(1,0,0), (0,1,0), (1,1,-1)\rbrace$ properly.

Actually, let me use a different approach. Set $\vec{b}_1=(1,0,0)$, $\vec{b}_2=(0,1,0)$, $\vec{b}_3=(1,1,-1)$.

Define $T(\vec{b}_1)=(1,0,1)$, $T(\vec{b}_2)=(0,1,1)$, $T(\vec{b}_3)=(0,0,0)$.

To find matrix in standard coordinates, express $(1,0,0)$, $(0,1,0)$, $(0,0,1)$ in terms of $\vec{b}_i$:

$(1,0,0) = 1\vec{b}_1 + 0\vec{b}_2 + 0\vec{b}_3$ β†’ $T(1,0,0) = (1,0,1)$

$(0,1,0) = 0\vec{b}_1 + 1\vec{b}_2 + 0\vec{b}_3$ β†’ $T(0,1,0) = (0,1,1)$

$(0,0,1)$: solve $c_1(1,0,0) + c_2(0,1,0) + c_3(1,1,-1) = (0,0,1)$

$c_1 + c_3 = 0$, $c_2 + c_3 = 0$, $-c_3 = 1$. So $c_3 = -1$, $c_1 = 1$, $c_2 = 1$.

$T(0,0,1) = 1\cdot(1,0,1) + 1\cdot(0,1,1) + (-1)\cdot(0,0,0) = (1,1,2)$

\[A = \begin{bmatrix} 1&0&1\\0&1&1\\1&1&2 \end{bmatrix}\]

Verify: $A(1,1,-1)^T = (1+0-1, 0+1-1, 1+1-2)^T = (0,0,0)$ βœ“

Image: Columns are $(1,0,1)$, $(0,1,1)$, $(1,1,2) = (1,0,1)+(0,1,1)$. So $\text{Im}(T) = \text{Span}\lbrace(1,0,1),(0,1,1)\rbrace$ βœ“

Q15. For the rotation matrix $R_{60^\circ} = \begin{bmatrix} \cos 60^\circ & -\sin 60^\circ \ \sin 60^\circ & \cos 60^\circ \end{bmatrix} = \begin{bmatrix} 1/2 & -\sqrt{3}/2 \ \sqrt{3}/2 & 1/2 \end{bmatrix}$, compute $R_{60^\circ}(1,0)$ and $R_{60^\circ}(0,1)$.

Click for solution

$R_{60^\circ}(1,0) = \left(\frac{1}{2}, \frac{\sqrt{3}}{2}\right)$

The point $(1,0)$ is rotated $60^\circ$ counter-clockwise to land at angle $60^\circ$ on the unit circle.

$R_{60^\circ}(0,1) = \left(-\frac{\sqrt{3}}{2}, \frac{1}{2}\right)$

The point $(0,1)$ (at $90^\circ$) moves to angle $150^\circ$ on the unit circle.

Verify linearity (spot check): $R_{60^\circ}(1,1) = \left(\frac{1-\sqrt{3}}{2}, \frac{\sqrt{3}+1}{2}\right)$

$R_{60^\circ}(1,0) + R_{60^\circ}(0,1) = \left(\frac{1-\sqrt{3}}{2}, \frac{\sqrt{3}+1}{2}\right)$ βœ“


Comprehensive Unit I Practice Set

These questions cover the entire unit at varying difficulty levels. Use them for exam preparation.


Section A: Short-Answer & Conceptual (2–3 marks each)

1. What are the three types of elementary row operations?

Click for answer (i) Swap two rows ($R_i \leftrightarrow R_j$), (ii) Multiply a row by a non-zero scalar ($R_i \leftarrow cR_i$), (iii) Add a multiple of one row to another ($R_i \leftarrow R_i + cR_j$).

2. Define: (a) Row Echelon Form, (b) Augmented Matrix

Click for answer **(a) REF:** A matrix where (1) all zero rows are at the bottom, (2) the leading entry (pivot) of each non-zero row is to the right of the pivot above, (3) all entries below each pivot are zero. **(b) Augmented matrix:** The coefficient matrix $A$ combined with the RHS vector $\vec{b}$, written as $[A \mid \vec{b}]$.

3. State the Rank-Nullity Theorem.

Click for answer For an $m \times n$ matrix $A$: $\text{rank}(A) + \text{nullity}(A) = n$ (number of columns). Equivalently: number of pivot columns + number of free variables = total columns.

4. What is the maximum rank of a $4 \times 6$ matrix?

Click for answer $\min(4, 6) = 4$. Maximum rank = 4.

5. When is a system $Ax = b$ inconsistent? State the rank condition.

Click for answer $Ax = b$ is inconsistent when $\text{rank}(A) < \text{rank}([A \mid b])$. This means a row reduces to $[0 \; 0 \; \cdots \; 0 \mid c]$ with $c \neq 0$.

6. Define β€œlinearly independent” in one sentence.

Click for answer Vectors $\lbrace\vec{v}_1, \ldots, \vec{v}_k\rbrace$ are linearly independent if the only solution to $c_1\vec{v}_1 + \cdots + c_k\vec{v}_k = \vec{0}$ is the trivial one $c_1 = c_2 = \cdots = c_k = 0$.

7. True or False: Every subspace must contain the zero vector.

Click for answer **True.** If $W$ is a subspace and $\vec{v} \in W$, then $0 \cdot \vec{v} = \vec{0} \in W$ by closure under scalar multiplication.

8. What is the dimension of $P_4$?

Click for answer $\dim(P_4) = 5$. Basis: $\lbrace1, x, x^2, x^3, x^4\rbrace$. In general, $\dim(P_n) = n + 1$.

9. If $A$ is $5 \times 5$ with $\text{rank}(A) = 5$, is $A$ invertible?

Click for answer **Yes.** Full rank ($\text{rank} = n$) for a square matrix means $\det(A) \neq 0$, so $A$ is invertible. Nullity $= 5 - 5 = 0$.

10. What does the Gram-Schmidt process produce from a linearly independent set?

Click for answer An **orthonormal set** (mutually orthogonal unit vectors) that spans the same subspace as the original set. It converts any basis into an orthonormal basis.

Section B: Computation (5 marks each)

11. Solve by Gauss elimination:

\[\begin{cases} 2x + y - z = 1 \\ -3x - y + 2z = 1 \\ x + y + z = 6 \end{cases}\]
Click for solution

Swap $R_1 \leftrightarrow R_3$:

\[\left[\begin{array}{ccc|c} 1 & 1 & 1 & 6 \\ -3 & -1 & 2 & 1 \\ 2 & 1 & -1 & 1 \end{array}\right]\]

$R_2 + 3R_1$: $[0, 2, 5 \;\vert\; 19]$. $R_3 - 2R_1$: $[0, -1, -3 \;\vert\; -11]$.

$2R_3 + R_2$: $[0, 0, -1 \;\vert\; -3]$ β†’ $z = 3$.

$2y + 15 = 19 \Rightarrow y = 2$. $x + 2 + 3 = 6 \Rightarrow x = 1$.

Solution: $x = 1, \; y = 2, \; z = 3$ βœ“

12. Find the rank and nullity of: \(A = \begin{bmatrix} 1 & 2 & 1 & 3 \\ 2 & 4 & 3 & 8 \\ 3 & 6 & 4 & 11 \end{bmatrix}\)

Click for solution

$R_2 - 2R_1$: $[0, 0, 1, 2]$. $R_3 - 3R_1$: $[0, 0, 1, 2]$. $R_3 - R_2$: $[0, 0, 0, 0]$.

\[\text{REF: }\begin{bmatrix} 1 & 2 & 1 & 3 \\ 0 & 0 & 1 & 2 \\ 0 & 0 & 0 & 0 \end{bmatrix}\]

2 pivots (cols 1, 3). Rank = 2, Nullity = $4 - 2 = 2$.

Free variables: $x_2 = s, x_4 = t$. From R2: $x_3 = -2t$. From R1: $x_1 = -2s + 2t - 3t + 3 … $ wait, let me redo: $x_1 + 2s + (-2t) + 3t = 0 \Rightarrow x_1 = -2s - t$.

Null space basis: $\lbrace(-2, 1, 0, 0), (-1, 0, -2, 1)\rbrace$.

13. Determine if the following vectors are linearly independent or dependent. If dependent, express one as a linear combination of the others: $(1, 2, 3), \; (4, 5, 6), \; (7, 8, 9)$

Click for solution
\[\det \begin{pmatrix} 1&4&7\\2&5&8\\3&6&9 \end{pmatrix} = 1(45-48) - 4(18-24) + 7(12-15) = -3 + 24 - 21 = 0\]

Dependent. Relation: $(7,8,9) = 2(4,5,6) - (1,2,3)$.

Check: $(8-1, 10-2, 12-3) = (7, 8, 9)$ βœ“

14. Find a basis and dimension for the subspace: $W = \lbrace(x, y, z, w) \in \mathbb{R}^4 : x + y = 0, \; z - w = 0\rbrace$

Click for solution

From constraints: $y = -x$ and $w = z$.

$(x, y, z, w) = (x, -x, z, z) = x(1, -1, 0, 0) + z(0, 0, 1, 1)$.

Basis: $\lbrace(1, -1, 0, 0), \; (0, 0, 1, 1)\rbrace$, dim = 2.

15. Apply Gram-Schmidt to $\lbrace(1, 1, 1), (1, 1, 0), (1, 0, 0)\rbrace$ and obtain an orthonormal basis.

Click for solution

Step 1: $\vec{u}_1 = \dfrac{(1,1,1)}{\sqrt{3}} = \dfrac{1}{\sqrt{3}}(1,1,1)$

Step 2: $\langle (1,1,0), \vec{u}_1 \rangle = \frac{2}{\sqrt{3}}$

$\vec{w}_2 = (1,1,0) - \frac{2}{\sqrt{3}} \cdot \frac{1}{\sqrt{3}}(1,1,1) = (1,1,0) - \frac{2}{3}(1,1,1) = \left(\frac{1}{3}, \frac{1}{3}, -\frac{2}{3}\right)$

$\lVert\vec{w}_2\rVert = \sqrt{\frac{1+1+4}{9}} = \frac{\sqrt{6}}{3}$ β†’ $\vec{u}_2 = \dfrac{1}{\sqrt{6}}(1, 1, -2)$

Step 3: $\langle (1,0,0), \vec{u}_1 \rangle = \frac{1}{\sqrt{3}}$, $\;\langle (1,0,0), \vec{u}_2 \rangle = \frac{1}{\sqrt{6}}$

$\vec{w}_3 = (1,0,0) - \frac{1}{\sqrt{3}} \cdot \frac{1}{\sqrt{3}}(1,1,1) - \frac{1}{\sqrt{6}} \cdot \frac{1}{\sqrt{6}}(1,1,-2)$

$= (1,0,0) - \frac{1}{3}(1,1,1) - \frac{1}{6}(1,1,-2) = \left(\frac{1}{2}, -\frac{1}{2}, 0\right)$

$\lVert\vec{w}_3\rVert = \frac{1}{\sqrt{2}}$ β†’ $\vec{u}_3 = \dfrac{1}{\sqrt{2}}(1, -1, 0)$

Orthonormal basis:

\[\left\lbrace\frac{1}{\sqrt{3}}(1,1,1), \;\; \frac{1}{\sqrt{6}}(1,1,-2), \;\; \frac{1}{\sqrt{2}}(1,-1,0)\right\rbrace\]

Section C: Proof & Theory (5–8 marks each)

16. Prove that the null space $N(A) = \lbrace\vec{x} : A\vec{x} = \vec{0}\rbrace$ is a subspace of $\mathbb{R}^n$.

Click for solution

We verify the three subspace conditions for $N(A) = \lbrace\vec{x} \in \mathbb{R}^n : A\vec{x} = \vec{0}\rbrace$:

1. Zero vector: $A\vec{0} = \vec{0}$, so $\vec{0} \in N(A)$. βœ“

2. Closure under addition: Let $\vec{u}, \vec{v} \in N(A)$. Then $A\vec{u} = \vec{0}$ and $A\vec{v} = \vec{0}$.

\[A(\vec{u} + \vec{v}) = A\vec{u} + A\vec{v} = \vec{0} + \vec{0} = \vec{0}\]

So $\vec{u} + \vec{v} \in N(A)$. βœ“

3. Closure under scalar multiplication: Let $\vec{u} \in N(A)$ and $c \in \mathbb{R}$. Then $A\vec{u} = \vec{0}$.

\[A(c\vec{u}) = c(A\vec{u}) = c\vec{0} = \vec{0}\]

So $c\vec{u} \in N(A)$. βœ“

All three conditions hold, therefore $N(A)$ is a subspace of $\mathbb{R}^n$. $\blacksquare$

17. Prove: If $\lbrace\vec{v}_1, \ldots, \vec{v}_k\rbrace$ is linearly independent in $\mathbb{R}^n$, then $k \leq n$.

Click for solution

Proof (by contradiction):

Suppose $k > n$. Form the $n \times k$ matrix $A = [\vec{v}_1 \; \vec{v}_2 \; \cdots \; \vec{v}_k]$.

The homogeneous system $A\vec{c} = \vec{0}$ has $n$ equations and $k$ unknowns with $k > n$.

By the theorem on homogeneous systems, when the number of unknowns exceeds the number of equations, there exists a non-trivial solution $\vec{c} = (c_1, \ldots, c_k) \neq \vec{0}$.

This means $c_1\vec{v}_1 + c_2\vec{v}_2 + \cdots + c_k\vec{v}_k = \vec{0}$ with not all $c_i = 0$.

This contradicts the linear independence of $\lbrace\vec{v}_1, \ldots, \vec{v}_k\rbrace$.

Therefore $k \leq n$. $\blacksquare$

Corollary: $\dim(\mathbb{R}^n) = n$, and no linearly independent set in $\mathbb{R}^n$ can have more than $n$ vectors.

18. Show that $W = \lbrace(x, y) : xy = 0\rbrace$ is NOT a subspace of $\mathbb{R}^2$.

Click for solution

$W$ consists of all points on the $x$-axis or $y$-axis (since $xy = 0$ iff $x = 0$ or $y = 0$).

Counterexample (closure under addition fails):

Let $\vec{u} = (1, 0)$ and $\vec{v} = (0, 1)$. Then:

  • $\vec{u} \in W$ since $1 \cdot 0 = 0$ βœ“
  • $\vec{v} \in W$ since $0 \cdot 1 = 0$ βœ“

But $\vec{u} + \vec{v} = (1, 1)$, and $1 \cdot 1 = 1 \neq 0$.

So $\vec{u} + \vec{v} \notin W$. βœ—

Since $W$ is not closed under addition, $W$ is not a subspace of $\mathbb{R}^2$. $\blacksquare$

19. Prove the Gram-Schmidt process produces an orthogonal set (i.e., show $\langle \vec{u}_i, \vec{u}_j \rangle = 0$ for $i \neq j$).

Click for solution

Proof by induction on the step number:

The Gram-Schmidt process defines:

\[\vec{w}_k = \vec{v}_k - \sum_{j=1}^{k-1} \frac{\langle \vec{v}_k, \vec{w}_j \rangle}{\langle \vec{w}_j, \vec{w}_j \rangle} \vec{w}_j, \quad \vec{u}_k = \frac{\vec{w}_k}{\lVert\vec{w}_k\rVert}\]

Claim: For each $k$, $\langle \vec{w}_k, \vec{w}_i \rangle = 0$ for all $i < k$.

Proof: Fix any $i < k$. Then:

\[\langle \vec{w}_k, \vec{w}_i \rangle = \left\langle \vec{v}_k - \sum_{j=1}^{k-1} \frac{\langle \vec{v}_k, \vec{w}_j \rangle}{\langle \vec{w}_j, \vec{w}_j \rangle} \vec{w}_j, \;\; \vec{w}_i \right\rangle\]

By linearity of the inner product:

\[= \langle \vec{v}_k, \vec{w}_i \rangle - \sum_{j=1}^{k-1} \frac{\langle \vec{v}_k, \vec{w}_j \rangle}{\langle \vec{w}_j, \vec{w}_j \rangle} \langle \vec{w}_j, \vec{w}_i \rangle\]

Since $\lbrace\vec{w}1, \ldots, \vec{w}{k-1}\rbrace$ are mutually orthogonal (by inductive hypothesis), $\langle \vec{w}_j, \vec{w}_i \rangle = 0$ for $j \neq i$. The only surviving term in the sum is $j = i$:

\[= \langle \vec{v}_k, \vec{w}_i \rangle - \frac{\langle \vec{v}_k, \vec{w}_i \rangle}{\langle \vec{w}_i, \vec{w}_i \rangle} \langle \vec{w}_i, \vec{w}_i \rangle = \langle \vec{v}_k, \vec{w}_i \rangle - \langle \vec{v}_k, \vec{w}_i \rangle = 0\]

Since normalizing ($\vec{u}_k = \vec{w}_k / \lVert\vec{w}_k\rVert$) preserves orthogonality, $\langle \vec{u}_i, \vec{u}_j \rangle = 0$ for $i \neq j$. $\blacksquare$

20. Prove: If $A$ is an $n \times n$ matrix with $\text{rank}(A) = n$, then $Ax = b$ has a unique solution for every $b$.

Click for solution

Existence: Since $\text{rank}(A) = n$, the column space of $A$ is $\mathbb{R}^n$ (the $n$ columns span all of $\mathbb{R}^n$). Therefore for any $\vec{b} \in \mathbb{R}^n$, $\vec{b}$ is in the column space, so $A\vec{x} = \vec{b}$ has at least one solution.

Uniqueness: By the Rank-Nullity Theorem:

\[\text{nullity}(A) = n - \text{rank}(A) = n - n = 0\]

So $N(A) = \lbrace\vec{0}\rbrace$. If $\vec{x}_1$ and $\vec{x}_2$ are both solutions ($A\vec{x}_1 = \vec{b}$ and $A\vec{x}_2 = \vec{b}$), then:

\[A(\vec{x}_1 - \vec{x}_2) = A\vec{x}_1 - A\vec{x}_2 = \vec{b} - \vec{b} = \vec{0}\]

So $\vec{x}_1 - \vec{x}_2 \in N(A) = \lbrace\vec{0}\rbrace$, meaning $\vec{x}_1 = \vec{x}_2$.

Therefore the solution is unique. $\blacksquare$

Equivalent conditions: $\text{rank}(A) = n$ $\iff$ $\det(A) \neq 0$ $\iff$ $A$ is invertible $\iff$ unique solution $\vec{x} = A^{-1}\vec{b}$.

Section D: Application Problems (5–8 marks each)

21. (Circuit Analysis) Three loop currents satisfy:

\[\begin{cases} 7I_1 - 2I_2 = 12 \\ -2I_1 + 11I_2 - 3I_3 = 8 \\ -3I_2 + 7I_3 = 4 \end{cases}\]

Find $I_1, I_2, I_3$ using Gauss elimination.

Click for solution

Augmented matrix:

\[\left[\begin{array}{ccc|c} 7 & -2 & 0 & 12 \\ -2 & 11 & -3 & 8 \\ 0 & -3 & 7 & 4 \end{array}\right]\]

$R_2 \leftarrow 7R_2 + 2R_1$ (eliminate $I_1$ from Row 2):

\[\left[\begin{array}{ccc|c} 7 & -2 & 0 & 12 \\ 0 & 73 & -21 & 80 \\ 0 & -3 & 7 & 4 \end{array}\right]\]

$R_3 \leftarrow 73R_3 + 3R_2$ (eliminate $I_2$ from Row 3):

\[\left[\begin{array}{ccc|c} 7 & -2 & 0 & 12 \\ 0 & 73 & -21 & 80 \\ 0 & 0 & 448 & 532 \end{array}\right]\]

Back substitution:

$448I_3 = 532 \;\Rightarrow\; I_3 = \dfrac{532}{448} = \dfrac{19}{16}$

$73I_2 = 80 + 21 \cdot \dfrac{19}{16} = 80 + \dfrac{399}{16} = \dfrac{1679}{16} \;\Rightarrow\; I_2 = \dfrac{1679}{1168} = \dfrac{23}{16}$

$7I_1 = 12 + 2 \cdot \dfrac{23}{16} = 12 + \dfrac{23}{8} = \dfrac{119}{8} \;\Rightarrow\; I_1 = \dfrac{119}{56} = \dfrac{17}{8}$

Answer: $I_1 = \dfrac{17}{8} \approx 2.125$ A, $\; I_2 = \dfrac{23}{16} \approx 1.4375$ A, $\; I_3 = \dfrac{19}{16} \approx 1.1875$ A.

Verification: $7(\frac{17}{8}) - 2(\frac{23}{16}) = \frac{119}{8} - \frac{23}{8} = 12$ βœ“, $\;-2(\frac{17}{8}) + 11(\frac{23}{16}) - 3(\frac{19}{16}) = 8$ βœ“, $\;-3(\frac{23}{16}) + 7(\frac{19}{16}) = 4$ βœ“

22. (Chemical Balancing) Balance: $C_3H_8 + O_2 \to CO_2 + H_2O$ Set up the system of linear equations for each element and solve.

Click for solution

Let the equation be $a\,C_3H_8 + b\,O_2 \to c\,CO_2 + d\,H_2O$.

Element balance equations:

  • Carbon: $3a = c$
  • Hydrogen: $8a = 2d \;\Rightarrow\; d = 4a$
  • Oxygen: $2b = 2c + d$

Set $a = 1$ (free parameter):

$c = 3(1) = 3$, $\quad d = 4(1) = 4$, $\quad 2b = 2(3) + 4 = 10 \;\Rightarrow\; b = 5$

Balanced equation:

\[C_3H_8 + 5O_2 \to 3CO_2 + 4H_2O\]

Verification: C: $3 = 3$ βœ“, H: $8 = 8$ βœ“, O: $10 = 6 + 4 = 10$ βœ“

23. (Structural Analysis) Forces satisfy:

\[\begin{cases} 2F_1 + F_2 = 100 \\ F_1 + 2F_2 - F_3 = 100 \\ F_2 + 3F_3 = 70 \end{cases}\]

Solve for all forces.

Click for solution

Augmented matrix:

\[\left[\begin{array}{ccc|c} 2 & 1 & 0 & 100 \\ 1 & 2 & -1 & 100 \\ 0 & 1 & 3 & 70 \end{array}\right]\]

$R_1 \leftrightarrow R_2$, then $R_2 \leftarrow R_2 - 2R_1$:

\[\left[\begin{array}{ccc|c} 1 & 2 & -1 & 100 \\ 0 & -3 & 2 & -100 \\ 0 & 1 & 3 & 70 \end{array}\right]\]

$R_3 \leftarrow 3R_3 + R_2$:

\[\left[\begin{array}{ccc|c} 1 & 2 & -1 & 100 \\ 0 & -3 & 2 & -100 \\ 0 & 0 & 11 & 110 \end{array}\right]\]

Back substitution:

$11F_3 = 110 \;\Rightarrow\; F_3 = 10$

$-3F_2 + 20 = -100 \;\Rightarrow\; F_2 = 40$

$F_1 + 80 - 10 = 100 \;\Rightarrow\; F_1 = 30$

Answer: $F_1 = 30$ N, $F_2 = 40$ N, $F_3 = 10$ N.

24. (Economics β€” Input-Output) In a 3-sector economy: \(\begin{bmatrix} 0.6 & -0.2 & -0.1 \\ -0.1 & 0.7 & -0.2 \\ -0.15 & -0.1 & 0.75 \end{bmatrix} \begin{pmatrix} x_1 \\ x_2 \\ x_3 \end{pmatrix} = \begin{pmatrix} 100 \\ 150 \\ 200 \end{pmatrix}\) Find production levels $x_1, x_2, x_3$.

Click for solution

Multiply through by 20 to clear decimals:

\[\left[\begin{array}{ccc|c} 12 & -4 & -2 & 2000 \\ -2 & 14 & -4 & 3000 \\ -3 & -2 & 15 & 4000 \end{array}\right]\]

$R_2 \leftarrow 6R_2 + R_1$ and $R_3 \leftarrow 4R_3 + R_1$:

\[\left[\begin{array}{ccc|c} 12 & -4 & -2 & 2000 \\ 0 & 80 & -26 & 20000 \\ 0 & -12 & 58 & 18000 \end{array}\right]\]

$R_3 \leftarrow 20R_3 + 3R_2$:

\[\left[\begin{array}{ccc|c} 12 & -4 & -2 & 2000 \\ 0 & 80 & -26 & 20000 \\ 0 & 0 & 1082 & 420000 \end{array}\right]\]

Back substitution:

$1082\,x_3 = 420000 \;\Rightarrow\; x_3 = \dfrac{420000}{1082} = \dfrac{210000}{541} \approx 388.17$

$80\,x_2 = 20000 + 26 \cdot \dfrac{210000}{541} = \dfrac{16280000}{541} \;\Rightarrow\; x_2 = \dfrac{203500}{541} \approx 376.16$

$12\,x_1 = 2000 + 4 \cdot \dfrac{203500}{541} + 2 \cdot \dfrac{210000}{541} = \dfrac{2316000}{541} \;\Rightarrow\; x_1 = \dfrac{193000}{541} \approx 356.75$

Answer: $x_1 \approx 356.75$, $x_2 \approx 376.16$, $x_3 \approx 388.17$ units.

Note: The Leontief model $(I - A)\vec{x} = \vec{d}$ often yields non-integer solutions in practice.

25. (Heat Transfer) Temperatures at interior nodes of a composite wall satisfy:

\[\begin{cases} -2T_2 + T_3 = -100 \\ T_2 - 2T_3 + T_4 = 0 \\ T_3 - 2T_4 = -20 \end{cases}\]

Boundary: $T_1 = 100^\circ\text{C}, T_5 = 20^\circ\text{C}$. Find $T_2, T_3, T_4$.

Click for solution

Augmented matrix:

\[\left[\begin{array}{ccc|c} -2 & 1 & 0 & -100 \\ 1 & -2 & 1 & 0 \\ 0 & 1 & -2 & -20 \end{array}\right]\]

$R_1 \leftrightarrow R_2$, then $R_2 \leftarrow R_2 + 2R_1$:

\[\left[\begin{array}{ccc|c} 1 & -2 & 1 & 0 \\ 0 & -3 & 2 & -100 \\ 0 & 1 & -2 & -20 \end{array}\right]\]

$R_3 \leftarrow 3R_3 + R_2$:

\[\left[\begin{array}{ccc|c} 1 & -2 & 1 & 0 \\ 0 & -3 & 2 & -100 \\ 0 & 0 & -4 & -160 \end{array}\right]\]

Back substitution:

$-4T_4 = -160 \;\Rightarrow\; T_4 = 40^\circ\text{C}$

$-3T_3 + 80 = -100 \;\Rightarrow\; T_3 = 60^\circ\text{C}$

$T_2 - 120 + 40 = 0 \;\Rightarrow\; T_2 = 80^\circ\text{C}$

Answer: $T_2 = 80^\circ\text{C}, \; T_3 = 60^\circ\text{C}, \; T_4 = 40^\circ\text{C}$.

The temperature decreases linearly from $T_1 = 100^\circ\text{C}$ to $T_5 = 20^\circ\text{C}$, which is expected for steady-state conduction in a uniform wall.


Answers to Selected Comprehensive Questions

11. $x = 1, y = 2, z = 3$.

12. Row reduce β†’ rank = 2, nullity = $4 - 2 = 2$.

13. Dependent. $(7,8,9) = 2(4,5,6) - (1,2,3)$.

14. $W$: $y = -x$ and $w = z$, so $(x, -x, z, z) = x(1,-1,0,0) + z(0,0,1,1)$. Basis: $\lbrace(1,-1,0,0), (0,0,1,1)\rbrace$, dimension = 2.

15. $\vec{u}_1 = \frac{1}{\sqrt{3}}(1,1,1)$, $\vec{u}_2 = \frac{1}{\sqrt{6}}(1,1,-2)$, $\vec{u}_3 = \frac{1}{\sqrt{2}}(1,-1,0)$.

22. $C_3H_8 + 5O_2 \to 3CO_2 + 4H_2O$.

23. $F_1 = 30$ N, $F_2 = 40$ N, $F_3 = 10$ N.

25. $T_2 = 80^\circ\text{C}, T_3 = 60^\circ\text{C}, T_4 = 40^\circ\text{C}$.


Exam Tips:

  • Always show the augmented matrix and each row operation clearly.
  • When testing subspaces, check $\vec{0} \in W$ first β€” it’s the quickest way to disprove.
  • For Gram-Schmidt, organize your work: compute projections first, then subtract, then normalize.
  • Label the rank, nullity, and verify with Rank-Nullity Theorem.
  • Practice parametric form for infinite solutions.

Back to Engineering Mathematics-II