Vector Spaces and Subspaces

Vector spaces are sets equipped with addition and scalar multiplication satisfying eight axioms; subspaces are subsets that are themselves vector spaces under the inherited operations. Key result: a nonempty subset is a subspace iff it is closed under addition and scalar multiplication. Applies anywhere linear structure appears — function spaces, solution sets of linear ODEs, signal processing, quantum mechanics. Axiomatized by Giuseppe Peano (1888), building on work by Hermann Grassmann (1844).

A vector space over a field $F$ is a set $V$ together with two operations — vector addition $+ : V \times V \to V$ and scalar multiplication $\cdot : F \times V \to V$ — satisfying eight axioms that abstract the familiar arithmetic of arrows in the plane. The concept matters because it isolates exactly what is needed to do "linear algebra": once you know something is a vector space, every theorem about dimension, bases, linear maps, and coordinates immediately applies, whether the "vectors" are arrows, polynomials, matrices, or solutions of differential equations. A subspace is a subset that is closed enough under the operations to form a vector space in its own right. The surprising fact: spaces that look wildly different — the plane $R^{2}$ , degree- $\leq 1$ polynomials, $2 \times 1$ matrices — are indistinguishable as vector spaces (all isomorphic), so abstract reasoning loses nothing and gains everything.

The Core Insight

Linearity is a structure, not a formula. The eight vector space axioms capture precisely the operations under which "taking linear combinations" makes sense:

a u + b v

must again live in the space. Everything in linear algebra — span, independence, basis, dimension, rank — is a consequence of closure under linear combinations. A subspace is just the smallest meaningful self-contained unit carrying this structure.

Formal Statement

Vector Space

Let $F$ be a field (typically $R$ or $C$ ). A vector space over $F$ is a set $V$ with operations $+$ (addition) and $\cdot$ (scalar multiplication) such that for all $u, v, w \in V$ and all $a, b \in F$ :

Additive axioms (making

(V, +)

an abelian group):

u + v = v + u (commutativity)

⟳This is an axiom, asserted directly; not derived.

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

⟳Axiom; guarantees unambiguous finite sums.

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

⟳Existence axiom; uniqueness is provable (see proof block).

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

⟳Existence axiom; the inverse is unique once

0

is fixed.

Scalar-multiplication axioms:

a \cdot (u + v) = a \cdot u + a \cdot v (distributivity over vector addition)

⟳Axiom linking the two operations.

(a + b) \cdot v = a \cdot v + b \cdot v (distributivity over field addition)

⟳Axiom; here

a + b

is addition in

F

, not in

V

(ab) \cdot v = a \cdot (b \cdot v) (compatibility with field multiplication)

⟳Axiom; relates scalar multiplication to multiplication in

F

1 \cdot v = v (identity scalar)

⟳Axiom;

1

is the multiplicative identity of

F

. Essential — without it

0 \cdot v = v

is possible.

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

Subspace

A subset $W \subseteq V$ of a vector space $V$ over $F$ is a subspace if $W$ is itself a vector space under the operations inherited from $V$ . Equivalently (the Subspace Test), $W$ is a subspace iff:

W \neq = \emptyset, u, v \in W \Rightarrow u + v \in W, a \in F, v \in W \Rightarrow a v \in W

⟳Closure under

+

and

\cdot

suffices; the other axioms are inherited from

V

automatically. See theorem below.

The two closure conditions combine into a single one-line test:

u, v \in W, a, b \in F \Rightarrow a u + b v \in W

⟳Set

b = 0

to get scalar closure; set

a = b = 1

to get additive closure. Equivalence with [9] is immediate.

Subspace Test

A nonempty subset $W \subseteq V$ is a subspace of $V$ if and only if it is closed under vector addition and scalar multiplication (equation [10]). In particular every subspace contains $0$ .

Why It Holds / Derivation

The closure conditions imply all eight axioms hold in $W$

We show that closure under $+$ and $\cdot$ (equation [9]) is sufficient for $W$ to be a vector space.

Step 1 — $0 \in W$ . Since

W \neq = \emptyset

, pick any

w \in W

. By scalar closure,

0 \cdot w \in W

. We prove

0 \cdot w = 0

: by [6],

0 \cdot w = (0 + 0) \cdot w = 0 \cdot w + 0 \cdot w

; adding

- (0 \cdot w)

to both sides gives

0 = 0 \cdot w

. Hence

0 \in W

. Step 2 — additive inverses lie in $W$ . For

w \in W

, scalar closure gives

(- 1) \cdot w \in W

. We show

(- 1) \cdot w = - w

: by [6] and [8],

w + (- 1) w = 1 w + (- 1) w = (1 + (- 1)) w = 0 \cdot w = 0

. So

- w \in W

. Step 3 — closure makes $+$ and $\cdot$ well-defined operations on $W$ . Equation [9] guarantees the results stay inside

W

. Step 4 — the remaining axioms [1],[2],[5],[6],[7],[8] are universally quantified equalities that hold for all elements of $V$ . Since

W \subseteq V

, they automatically hold for elements of

W

. No verification is needed.

Therefore $(W, +, \cdot)$ satisfies all eight axioms and is a vector space — a subspace of $V$ . $■$

Uniqueness of $\mathbf{0}$ and elementary consequences

Uniqueness of zero: Suppose

0

and

0^{'}

both satisfy [3]. Then

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

, using [3] twice and [1]. Uniqueness of inverses: If

v + a = 0

and

v + b = 0

, then

a = a + 0 = a + (v + b) = (a + v) + b = 0 + b = b

. $a \cdot 0 = 0$ :

a \cdot 0 = a \cdot (0 + 0) = a \cdot 0 + a \cdot 0

by [5]; cancel one copy. $a v = 0 \Rightarrow a = 0$ or $v = 0$ : If

a \neq = 0

, multiply by

a^{- 1}

and use [7],[8]:

v = 1 v = (a^{- 1} a) v = a^{- 1} (a v) = a^{- 1} 0 = 0

■

Giuseppe Peano | 1858–1932

Italian mathematician who, in his 1888 book Calcolo geometrico secondo l'Ausdehnungslehre di H. Grassmann, gave the first fully axiomatic definition of a finite-dimensional vector space over the reals — essentially the eight axioms used today, including a precise treatment of linear independence, dimension, and linear maps. Peano recognized that the algebraic essence of Grassmann's geometry could be captured in a short list of axioms, decades before this viewpoint became standard. He also worked on the foundations of arithmetic (the Peano axioms) and constructed the famous space-filling Peano curve.

Building Intuition

Think of a vector space as a playground with two legal moves: you may slide vectors together (addition) and stretch or shrink them (scalar multiplication). The axioms are the rulebook ensuring these moves never throw you off the playground and always behave predictably. A subspace is a smaller region of the same playground that you can never leave using legal moves — start anywhere inside it, add and scale to your heart's content, you stay inside.

Geometrically, in $R^{3}$ the subspaces are exactly: the origin ${0}$ , every line through the origin, every plane through the origin, and all of $R^{3}$ . The phrase "through the origin" is forced — the origin must belong to any subspace (Step 1 of the proof), so a line or plane missing the origin is not a subspace. This is the single most useful sanity check: if it doesn't contain $0$ , it isn't a subspace.

Subspaces of $\mathbb{R}^3$: only flat objects through the origin qualify

Worked Examples

Is a plane a subspace of $\mathbb{R}^3$?

Given:

W = {(x, y, z) \in R^{3} : 2 x - y + 3 z = 0}

. Determine whether

W

is a subspace. Step 1 — nonempty / contains $0$ :

(0, 0, 0)

gives

2 (0) - 0 + 3 (0) = 0

. ✓ So

0 \in W

. Step 2 — additive closure: Take

u = (x_{1}, y_{1}, z_{1})

v = (x_{2}, y_{2}, z_{2})

W

, so

2 x_{1} - y_{1} + 3 z_{1} = 0

and

2 x_{2} - y_{2} + 3 z_{2} = 0

. Then for

u + v = (x_{1} + x_{2}, y_{1} + y_{2}, z_{1} + z_{2})

2 (x_{1} + x_{2}) - (y_{1} + y_{2}) + 3 (z_{1} + z_{2}) = (2 x_{1} - y_{1} + 3 z_{1}) + (2 x_{2} - y_{2} + 3 z_{2}) = 0 + 0 = 0.

✓ Step 3 — scalar closure: For

a \in R

a u = (a x_{1}, a y_{1}, a z_{1})

2 (a x_{1}) - (a y_{1}) + 3 (a z_{1}) = a (2 x_{1} - y_{1} + 3 z_{1}) = a \cdot 0 = 0.

✓ Boxed answer:

W is a subspace of R^{3} — a plane through the origin, dim W = 2.

(The key feature: the defining equation is homogeneous — equals zero. A nonzero right-hand side, e.g. $2 x - y + 3 z = 5$ , fails Step 1.)

A subspace of a function space, with a dimension count

Given: Let

V = P_{3} (R)

, the space of real polynomials of degree

\leq 3

. Let

W = {p \in P_{3} : p (1) = 0 and p^{'} (0) = 0} .

Show

W

is a subspace and find its dimension. Step 1 — closure: Write

p (x) = a_{0} + a_{1} x + a_{2} x^{2} + a_{3} x^{3}

. The conditions are linear in the coefficients:

p (1) = a_{0} + a_{1} + a_{2} + a_{3} = 0, p^{'} (0) = a_{1} = 0.

p, q

satisfy both equations then

α p + β q

does too, since each condition is a homogeneous linear equation in the coefficients. Hence

W

is a subspace (and contains the zero polynomial). Step 2 — solve the constraints: From

a_{1} = 0

and

a_{0} + a_{2} + a_{3} = 0

, we get

a_{0} = - (a_{2} + a_{3})

. Free parameters:

a_{2}, a_{3}

. General element:

p (x) = - (a_{2} + a_{3}) + a_{2} x^{2} + a_{3} x^{3} = a_{2} (x^{2} - 1) + a_{3} (x^{3} - 1) .

Step 3 — read off a basis:

{x^{2} - 1, x^{3} - 1}

spans

W

and is linearly independent. Boxed answer:

dim W = 2, W = span {x^{2} - 1, x^{3} - 1} .

(Check: $dim P_{3} = 4$ ; two independent constraints drop the dimension by $2$ , giving $4 - 2 = 2$ . ✓)

Common Misconceptions

A subspace must contain the zero vector — always check this first. It is the fastest disqualifier. Sets defined by non-homogeneous equations (

A x = b

with

b \neq = 0

), the first quadrant of

R^{2}

(fails scalar closure under negative scalars), the unit circle, and any line not through the origin are not subspaces. Closure must hold for all scalars in the field $F$ , not just positive ones, not just integers. The set

{(x, y) : x, y \geq 0}

is closed under addition but fails

(- 1) \cdot

. Axiom [8], $1 \cdot v = v$ , cannot be dropped. Without it, pathological "spaces" exist where

a \cdot v = 0

for all

a, v

; these satisfy axioms [1]–[7] but are not what we want. The empty set is not a subspace (it fails

W \neq = \emptyset

/ lacks

0

), but

{0}

— the trivial subspace — is.

A set closed under addition but not a subspace

Let $W = {(x, y) \in R^{2} : x \geq 0, y \geq 0}$ , the closed first quadrant. It is closed under addition: summing two vectors with nonnegative components gives nonnegative components. It contains $0$ . Yet it is not a subspace, because scalar multiplication by $- 1$ fails: $(1, 1) \in W$ but $(- 1) \cdot (1, 1) = (- 1, - 1) \in / W$ . This shows additive closure plus containing $0$ is not enough — scalar closure for all scalars is essential.

Union of two subspaces usually fails

Let $U = {(x, 0) : x \in R}$ (the $x$ -axis) and $W = {(0, y) : y \in R}$ (the $y$ -axis), both subspaces of $R^{2}$ . Their union $U \cup W$ is not a subspace: $(1, 0) \in U \cup W$ and $(0, 1) \in U \cup W$ , but their sum $(1, 1)$ lies on neither axis. (In general $U \cup W$ is a subspace iff $U \subseteq W$ or $W \subseteq U$ .) The correct subspace-producing operation is the sum $U + W = {u + w}$ , which here equals all of $R^{2}$ .

Connections

Vector spaces are the stage on which the rest of linear algebra is performed. A subspace is generated by taking the Span and Linear Combinations of any set of vectors, and the size of a minimal spanning set is governed by Linear Independence and Basis, leading directly to Dimension of a Vector Space. Maps that respect the structure are Linear Transformations, whose kernel and image are themselves subspaces — connecting to the Rank-Nullity Theorem. When the field is $R$ or $C$ and an inner product is added, subspaces acquire orthogonal complements (Inner Product Spaces). The intersection of two subspaces is always a subspace, while their union rarely is; the corrective operation is the Sum and Direct Sum of Subspaces.

Real-World Applications

Quantum mechanics — physical states are unit vectors in a complex Hilbert (vector) space; superposition is literally a linear combination, and observables' eigenspaces are subspaces.
Signal processing & JPEG/MP3 compression — signals are vectors; restricting to a low-dimensional subspace (e.g. dominant Fourier or DCT components) achieves lossy compression.
Solutions of homogeneous linear ODEs — the set of solutions of $y^{''} + p y^{'} + q y = 0$ forms a 2-dimensional vector space; the general solution is a basis-spanned combination.
Machine learning — PCA — Principal Component Analysis finds the best-fitting low-dimensional subspace approximating high-dimensional data.
Computer graphics — 3D transformations act on the vector space $R^{3}$ (or $R^{4}$ with homogeneous coordinates); rotations preserve subspaces.
Error-correcting codes — linear codes (Hamming, Reed–Solomon) are subspaces of $F_{q}^{n}$ ; their dimension determines the information rate.
Control theory — controllable and unobservable subspaces of a linear system determine which states can be steered or detected.

Historical Development

1843–1844 — Hermann Grassmann publishes Die lineale Ausdehnungslehre, introducing $n$ -dimensional spaces and linear combinations far ahead of his time; the work is largely ignored due to its difficulty.
1844 — William Rowan Hamilton discovers quaternions, an early example of structure that is a real vector space (and more).
1857 — Arthur Cayley develops matrix algebra, providing concrete realizations of linear maps on $R^{n}$ .
1888 — Giuseppe Peano gives the first modern axiomatic definition of a (finite-dimensional) vector space over $R$ , including dimension and linear maps.
1898 — Cesare Burali-Forti and others extend and popularize Peano's vectorial calculus.
1918 — Hermann Weyl, in Raum, Zeit, Materie, uses axiomatic vector spaces in the foundations of relativity.
1920s — Stefan Banach and Hans Hahn develop infinite-dimensional normed vector spaces, founding functional analysis.
1932 — Banach's Théorie des opérations linéaires codifies the abstract theory of (complete normed) vector spaces.
1930s–1940s — Vector spaces over arbitrary fields become standard in abstract algebra textbooks (van der Waerden, Bourbaki), cementing the modern viewpoint.

Subtleties & Limits

Infinite dimensions break intuition. Every vector space has a basis — but proving this in infinite dimensions requires the Axiom of Choice (it is in fact equivalent to it, a result of Andreas Blass, 1984). Such a "Hamel basis" for

R

over

Q

, or for spaces like

ℓ^{\infty}

, cannot be exhibited explicitly; it is purely non-constructive. Subspaces in infinite dimensions need not be closed. In a normed infinite-dimensional space, a subspace can be dense (its closure is the whole space) yet proper — e.g. the polynomials inside

C [0, 1]

. This forces functional analysis to distinguish algebraic subspaces from closed subspaces, a distinction invisible in finite dimensions. Over finite fields, counting subspaces leads to Gaussian binomial coefficients and deep combinatorics (matroid theory, the

q

-analogue program), and questions about the lattice of subspaces connect to projective geometry and unresolved problems in combinatorial design theory.