Vectors and Vector Spaces: The Basics

A vector space is a set of objects ("vectors") that can be added together and scaled by numbers ("scalars") so that the familiar rules of arithmetic hold. Key result in words: any vector space is completely determined (up to relabeling) by a single number, its dimension, the size of any basis. Where it applies: linear algebra, geometry, physics, computer graphics, data science, quantum mechanics, differential equations. Who/when: the abstract axioms were crystallized by Giuseppe Peano (1888), building on Grassmann (1844) and Hamilton.

A vector space over a field $F$ (usually $R$ or $C$ ) is a set $V$ equipped with vector addition and scalar multiplication satisfying eight axioms (associativity, commutativity, identities, inverses, and distributivity/compatibility laws). The concept matters because it strips geometry down to its purely linear essence: arrows in the plane, polynomials, matrices, solutions of linear differential equations, and audio signals all obey the same axioms and so share the same theorems. The surprising fact is that despite the wild diversity of examples, every finite-dimensional vector space over $F$ of dimension $n$ is isomorphic to the single concrete space $F^{n}$ — there is essentially only one vector space of each dimension.

The Core Insight

Forget what the objects "are." A vector space is defined entirely by what you are allowed to do with them: add two and scale one. Everything else — lines, planes, independence, dimension, coordinates — is forced by those two operations and their algebraic laws. Linearity is the whole subject: a vector space is the natural home of the operation $α u + β v$ .

Formal Statement

Field

A field $F$ is a set with two operations $+$ and $\cdot$ in which addition and multiplication are associative and commutative, both have identities ( $0$ and $1$ , with $0 \neq = 1$ ), every element has an additive inverse, every nonzero element has a multiplicative inverse, and multiplication distributes over addition. The canonical examples are $R$ (real numbers), $C$ (complex numbers), and $Q$ (rationals). Scalars are drawn from $F$ .

Vector Space

A vector space over a field $F$ is a set $V$ together with two maps,

+ : V \times V \to V, \cdot : F \times V \to V,

⟳closure is built into the codomain

V

— adding/scaling vectors must produce vectors.

such that for all $u, v, w \in V$ and all $α, β \in F$ the following eight axioms hold:

(u + v) + w = u + (v + w) (additive associativity)

⟳grouping of sums is irrelevant; lets us write

u + v + w

unambiguously.

u + v = v + u (commutativity)

⟳order of addition does not matter;

V

is an abelian group under

+

\exists 0 \in V : v + 0 = v (zero vector)

⟳a single neutral element; uniqueness is a theorem, not an axiom.

\forall v \exists (- v) : v + (- v) = 0 (additive inverses)

⟳every vector can be "undone"; subtraction is defined as

u - v := u + (- v)

α (u + v) = α u + α v (distributivity over vector sums)

⟳scaling a sum scales each part.

(α + β) v = α v + β v (distributivity over scalar sums)

⟳this and [6] are the two distinct distributive laws; both are needed.

\quad \alpha(\beta\mathbf{v}) = (\alpha\beta)\mathbf{v},\qquad 1\,\mathbf{v}=\mathbf{v} \quad\text{(compatibility & unital)}

⟳scalar multiplication is compatible with field multiplication, and the field's

1

acts trivially.

Elements of $V$ are vectors; elements of $F$ are scalars.

Elementary Consequences

In any vector space: (a) the zero vector is unique; (b) additive inverses are unique; (c) $0 v = 0$ for every $v$ ; (d) $α 0 = 0$ for every scalar $α$ ; (e) $(- 1) v = - v$ ; (f) if $α v = 0$ then $α = 0$ or $v = 0$ .

Subspace

A nonempty subset $W \subseteq V$ is a subspace if it is closed under addition and scalar multiplication; equivalently, $W$ is a subspace iff $0 \in W$ and $α u + β v \in W$ for all $u, v \in W$ , $α, β \in F$ . A subspace is itself a vector space.

Span, Independence, Basis, Dimension

The span of $S = {v_{1}, \dots, v_{k}}$ is the set of all linear combinations:

span (S) = {\sum_{i = 1}^{k} α_{i} v_{i} : α_{i} \in F} .

⟳closure under

+

and

\cdot

makes this the smallest subspace containing

S

The set $S$ is linearly independent if

i = 1 \sum k α_{i} v_{i} = 0 ⟹ α_{1} = \dots = α_{k} = 0.

⟳the only way to write

0

is the trivial way; no vector is redundant.

A basis is a linearly independent spanning set. The dimension $dim V$ is the cardinality of any basis (well-defined by the theorem below).

Dimension Theorem / Basis Invariance

Every vector space has a basis (assuming the axiom of choice for infinite-dimensional cases), and all bases of a given vector space have the same cardinality. Hence $dim V$ is well-defined. If $dim V = n < \infty$ , then any spanning set has $\geq n$ elements, any independent set has $\leq n$ elements, and any independent set of exactly $n$ vectors is automatically a basis.

Why It Holds / Derivation

Uniqueness of zero and that $0\mathbf{v}=\mathbf{0}$

Step 1 (zero is unique). Suppose

0

and

0^{'}

both satisfy axiom [4]. Then

0 = 0 + 0^{'} = 0^{'} + 0 = 0^{'}

, using [4] for

0^{'}

, then commutativity [3], then [4] for

0

. So

0 = 0^{'}

. Step 2 ( $0 v = 0$ ). Using distributivity [7] with

α = β = 0

0 v = (0 + 0) v = 0 v + 0 v .

Add

- (0 v)

(which exists by [5]) to both sides and use [2],[4],[5]:

0 = 0 v + (- (0 v)) = 0 v .

Step 3 ( $(- 1) v = - v$ ). By [7] and Step 2,

v + (- 1) v = 1 v + (- 1) v = (1 + (- 1)) v = 0 v = 0

. By uniqueness of inverses,

(- 1) v = - v

■

Basis invariance via the Steinitz Exchange Lemma

Lemma (Exchange). If

{v_{1}, \dots, v_{m}}

spans

V

and

{w_{1}, \dots, w_{n}}

is linearly independent, then

n \leq m

, and one may replace

n

of the

v

's by the

w

's while still spanning. Step 1. Since the

v

's span, write

w_{1} = \sum_{i} α_{i} v_{i}

. Some

α_{j} \neq = 0

(else

w_{1} = 0

, contradicting independence). Solve for

v_{j}

in terms of

w_{1}

and the other

v

's; the new set still spans. Step 2. Iterate: at step

k

express

w_{k}

using the current spanning set

{w_{1}, \dots, w_{k - 1}, remaining v ’s}

. Independence of the

w

's guarantees a nonzero coefficient on some remaining

v

, so a

v

can be swapped out. This requires there to be a remaining

v

at each step, forcing

n \leq m

. Step 3 (invariance). If

B_{1}

and

B_{2}

are both bases, apply the lemma twice:

B_{2}

independent +

B_{1}

spanning gives

∣ B_{2} ∣ \leq ∣ B_{1} ∣

; symmetrically

∣ B_{1} ∣ \leq ∣ B_{2} ∣

. Hence

∣ B_{1} ∣ = ∣ B_{2} ∣

■

Giuseppe Peano | 1858–1932

The Italian logician and mathematician gave, in his 1888 book Calcolo geometrico secondo l'Ausdehnungslehre di H. Grassmann, the first fully axiomatic definition of a vector space over the reals — essentially the eight axioms above, including the abstract notion of a "linear system" of arbitrary (even infinite) dimension. He distilled Hermann Grassmann's dense 1844 Ausdehnungslehre into clean axioms decades before the language became standard. Peano is more famous for his axioms for the natural numbers, but his vector-space definition is strikingly modern and was largely ignored until the 1920s revival by Weyl and Banach.

Building Intuition

Picture an arrow anchored at the origin. Adding two arrows is the parallelogram (tip-to-tail) rule; scaling by $α$ stretches the arrow ( $∣ α ∣ > 1$ ), shrinks it ( $∣ α ∣ < 1$ ), or flips it ( $α < 0$ ). The eight axioms are just the statement that this geometric arithmetic "behaves like numbers." A subspace is a flat object through the origin — a line, a plane — that is self-contained under these operations. A basis is a minimal set of arrows from which every other arrow can be reached by sliding and scaling; dimension counts how many independent directions you need. The deep point: even though polynomials, matrices, and audio signals look nothing like arrows, they obey the identical rules, so the picture transfers. Coordinates are nothing but the recipe (the coefficients $α_{i}$ ) that rebuilds a vector from the basis arrows.

Vector addition by the parallelogram rule and the span of two vectors

Worked Examples

Checking independence and finding a basis in $\mathbb{R}^3$

Given:

v_{1} = (1, 2, 3)

v_{2} = (2, 4, 6)

v_{3} = (0, 1, 1)

R^{3}

. Find

dim span {v_{1}, v_{2}, v_{3}}

and a basis. Step 1. Notice

v_{2} = 2 v_{1}

, so

v_{2}

is redundant. They are not independent:

2 v_{1} - v_{2} + 0 \cdot v_{3} = 0

, a nontrivial relation. Step 2. Test

{v_{1}, v_{3}}

. Suppose

a v_{1} + b v_{3} = 0

(a, 2 a + b, 3 a + b) = (0, 0, 0) .

From the first coordinate

a = 0

; then

b = 0

. Independent. Step 3. Two independent vectors span a plane, dimension 2. Answer:

dim = 2, basis {(1, 2, 3), (0, 1, 1)}

(a plane through the origin in

R^{3}

A function space: solutions of a differential equation

Given: The set

V

of all real solutions of

y^{''} + 4 y = 0

. Show

V

is a vector space and find its dimension. Step 1 (subspace of $C^{\infty} (R)$ ). If

y_{1}, y_{2}

solve the equation, then for scalars

α, β

(α y_{1} + β y_{2})^{''} + 4 (α y_{1} + β y_{2}) = α (y_{1}^{''} + 4 y_{1}) + β (y_{2}^{''} + 4 y_{2}) = 0 + 0 = 0.

V

is closed under linear combinations and contains

y \equiv 0

— it is a subspace, hence a vector space. Step 2 (basis). Standard ODE theory gives the general solution

y = c_{1} cos 2 x + c_{2} sin 2 x

. The functions

cos 2 x, sin 2 x

are independent (their Wronskian is

2 \neq = 0

) and span

V

. Answer:

dim V = 2, basis {cos 2 x, sin 2 x}

. A second-order linear ODE has a 2-dimensional solution space.

Common Misconceptions

"Vectors are arrows / lists of numbers." That is only one model. Polynomials, matrices, continuous functions, and infinite sequences are all genuine vectors. The axioms — not the appearance — decide. Also beware: the zero vector must lie in every subspace; a line not through the origin is not a subspace, because it fails closure and lacks

0

. Finally, "dimension" counts basis vectors, not the number of coordinates you happen to write down — a plane in

R^{3}

uses 3-tuples but has dimension 2.

Is $W = {(x, y) \in R^{2} : x^{2} + y^{2} \leq 1}$ (the unit disk) a subspace? No. It contains $0$ , but it fails scalar closure: $(1, 0) \in W$ yet $2 \cdot (1, 0) = (2, 0) \in / W$ . Closure under scaling forces a subspace to be unbounded (unless it is just ${0}$ ). Similarly ${(x, y) : x y = 0}$ (the union of both axes) contains $0$ and is closed under scaling but fails addition: $(1, 0) + (0, 1) = (1, 1)$ has $x y = 1 \neq = 0$ . Subspaces must pass all closure tests, not some.

Connections

Vector spaces are the objects on which Linear Transformations act; the structure-preserving maps between them are precisely linear maps, and choosing bases turns these into Matrices. Endowing a space with a Inner Product Spaces|dot product gives lengths and angles, leading to orthogonality and Gram-Schmidt Orthogonalization. The quotient of a space by a subspace yields Quotient Spaces, and the catalog of subspaces forms a lattice. Over $C$ , vector spaces underpin Eigenvalues and Eigenvectors and the spectral theorem. Infinite-dimensional vector spaces with a topology become Hilbert Spaces and Banach Spaces, the arena of functional analysis.

Real-World Applications

Computer graphics & game engines: points, normals, and transformations live in $R^{3}$ and $R^{4}$ (homogeneous coordinates); rendering pipelines are chains of linear maps.
Quantum mechanics: physical states are unit vectors in a complex Hilbert space; observables act linearly. Superposition is vector addition.
Machine learning & data science: feature vectors, word embeddings (word2vec), and principal component analysis all treat data as points in high-dimensional vector spaces.
Signal & image processing: the Fourier and wavelet transforms express signals in alternative bases of a function space; JPEG compression discards small coordinates.
Engineering / structural analysis: the finite element method solves PDEs by projecting onto finite-dimensional subspaces of solution spaces.
Cryptography & coding theory: linear error-correcting codes (Hamming, Reed–Solomon) are subspaces of $F_{q}^{n}$ over finite fields.
Economics & operations research: linear programming optimizes a linear functional over a polyhedron in $R^{n}$ .

Historical Development

1827–1844 — Möbius & Grassmann: Möbius introduces barycentric coordinates; Hermann Grassmann's Ausdehnungslehre (1844) develops abstract $n$ -dimensional linear algebra, far ahead of its time and largely unread.
1843 — Hamilton: invents quaternions, an early algebraic system mixing scalar and vector parts; coins the word "vector."
1888 — Peano: gives the first axiomatic definition of a real vector space and of linear independence/dimension.
1880s–1901 — Gibbs & Heaviside: develop the modern vector calculus (dot and cross products) used in physics, derived from quaternions.
1918 — Weyl: in Raum, Zeit, Materie, uses axiomatic vector spaces in relativity and popularizes the abstract viewpoint.
1920s — Banach, Hahn, Wiener: extend the theory to infinite-dimensional normed spaces (Banach spaces), founding functional analysis.
1930s — van der Waerden & Bourbaki: Moderne Algebra (1930) and later Bourbaki cement vector spaces over arbitrary fields as a standard chapter of algebra.
1940s–present: vector spaces become foundational across pure and applied mathematics, physics, and computer science.

Subtleties & Limits

The clean theory above is for finite-dimensional spaces over a field. Subtleties multiply otherwise. (1) Proving that every vector space has a basis requires the Axiom of Choice (it is in fact equivalent to it); for infinite-dimensional spaces a basis (a Hamel basis) is non-constructive and typically uncountable, e.g. for $R$ over $Q$ . (2) Over a general ring rather than a field, the analogue (a module) can fail to have a basis at all, and dimension need not be well-defined — the IBN (invariant basis number) property can fail. (3) In infinite dimensions, the algebraic Hamel basis is usually useless for analysis; one uses topological (Schauder) bases instead, raising open questions historically (the basis problem, resolved negatively by Enflo in 1973: there exist Banach spaces with no Schauder basis). These boundaries mark where "the basics" hand off to module theory and functional analysis.