Vector Spaces

Why we need a definition (not just examples)

In geometry, vectors look like arrows. But in linear algebra, “vector” really means: an object you can add and scale in a way that behaves like ordinary arithmetic.

That’s powerful because it lets us treat many different domains with one toolkit:

• •polynomials (for curve fitting),
• •functions (for differential equations),
• •matrices (for data and transformations),
• •signals (for audio/image processing).

The key is not the shape of objects—it’s the rules.

The official definition

A vector space is a set $V$ together with:

1) Vector addition: a function $+ : V \times V \to V$

2) Scalar multiplication: a function $\cdot : F \times V \to V$ (often written just as $av$)

where $F$ is a field (the scalars). Typical choices are $\mathbb{R}$ or $\mathbb{C}$. The field matters because we need to add/multiply scalars and have identities/inverses in the scalar world.

We say: “$V$ is a vector space over $F$.”

The axioms, grouped by meaning

You can memorize 8–10 axioms, but it’s better to group them into two clusters plus “compatibility.”

A) Addition behaves like normal addition (abelian group)

For all u, v, w ∈ V:

1. 1)Closure under addition: u + v ∈ V
2. 2)Associativity: (u + v) + w = u + (v + w)
3. 3)Commutativity: u + v = v + u
4. 4)Additive identity: there exists 0 ∈ V such that v + 0 = v
5. 5)Additive inverse: for each v, there exists −v with v + (−v) = 0

This is exactly: “(V, +) is an abelian group.”

B) Scaling is allowed

For all $a \in F$ and v ∈ V:

• •$a\,\mathbf{v} \in V$ (closure under scalar multiplication)

C) Scaling and addition interact consistently (compatibility)

For all $a,b \in F$ and u, v ∈ V:

1. 1)Distribute scalar over vector addition:

2. 2)Distribute field addition over vectors:

3. 3)Associativity of scalar multiplication:

4. 4)Scalar identity acts as identity:

A quick “feel” check

If you can:

• •add two objects of the set and stay inside the set,
• •scale any object by any scalar and stay inside the set,
• •and the distributive/identity rules keep working,

then linear algebra becomes available.

In this lesson, we’ll keep returning to the same habit: test closure + distributivity early—those are frequent failure points, especially in non-geometric examples.

Why focus on addition first?

Scalar multiplication is only meaningful if “adding vectors” already forms a stable arithmetic world. The abelian-group requirements guarantee you can:

• •combine multiple vectors without ambiguity (associativity),
• •reorder sums (commutativity),
• •subtract vectors via inverses,
• •and have a neutral element (0).

These are what make expressions like

well-defined.

Concrete check: does the set contain its own sums?

Closure under addition is often the first axiom that fails.

Example A (geometric): $\mathbb{R}^2$

Take u = (1, 2), v = (3, −5). Then

(1, 2) + (3, −5) = (4, −3) ∈ $\mathbb{R}^2$.

So closure holds.

Example B (non-geometry): polynomials of degree ≤ 2

Let

• •$p(x)=1+x$
• •$q(x)=x^2-3$

Then

This is still a polynomial of degree ≤ 2. So closure holds.

Example C (common fail): “polynomials of degree exactly 2”

Let $p(x)=x^2$ and $q(x)=-x^2$. Then

But 0 is not degree exactly 2, so the set is not closed under addition.

The lesson: degree ≤ n works; degree exactly n usually fails.

The zero vector is not always (0,0)

The additive identity depends on what the objects are.

• •In $\mathbb{R}^n$, 0 is (0, 0, …, 0).
• •In polynomial spaces, 0 is the zero polynomial $0$ (all coefficients 0).
• •In function spaces, 0 is the zero function $f(x)=0$ for all x.

You should think: “the zero object under addition.”

Additive inverses: subtraction must stay inside

If v is in the set, then −v must also be in the set.

Polynomial example

If $p(x)=x^2+x$, then −$p(x)=-(x^2+x)=-x^2-x$, which is still a polynomial (and still degree ≤ 2 if we’re in that set).

Function example

If $f(x)=\sin x$, then −$f(x)=-\sin x$ is still a function of the same type.

But beware: if your set is restricted (e.g., “functions that are always nonnegative”), additive inverses typically fail.

Tiny visualization: closure under addition in a function space

Think of functions as “curves.” Adding functions adds their y-values pointwise.

Take two functions $f$ and $g$. Their sum is:

Here’s a simple ASCII snapshot at a few x-values:

If your set is “all real-valued functions,” the sum is still real-valued at every x, so closure holds.

Interactive canvas idea (guided): Plot two curves (e.g., $f(x)=\sin x$, $g(x)=0.5x$). Add a toggle “show f+g.” Let learners drag a point x₀ and observe the vertical addition of y-values. The closure message becomes: “the result is still a function in the same set.”

Why scalar multiplication matters

Scalar multiplication is what lets you talk about size and direction (or more abstractly: intensity, amplitude, coefficient scaling). It’s also what makes linear combinations possible:

But scalar multiplication must be consistent with addition; otherwise linear combinations become ambiguous and algebra breaks.

Scalar multiplication closure

For every $a \in F$ and v ∈ V, we need $a\mathbf{v} \in V$.

Non-geometry closure example: polynomials

Let $F=\mathbb{R}$ and $V=\{\text{polynomials of degree ≤ 2}\}$.

If $a=3$ and $p(x)=x^2-1$, then

Still degree ≤ 2. Closure holds.

Common fail: integer-only scalars with real vectors

If you try to treat $\mathbb{R}^2$ as a vector space over $\mathbb{Z}$, you run into a deeper issue: $\mathbb{Z}$ is not a field (no multiplicative inverses for most integers). Many theorems break, and it’s not a vector space by definition.

So the “F (field)” requirement is not decoration—it’s structural.

Distributivity: the rule that keeps scaling honest

There are two distributive laws and they fail in many “almost vector spaces.”

1) Scalar over vector sum

Function visualization (pointwise):

Let $a=2$, $f(x)=x$, $g(x)=1$.

• •Left: $2(f+g)(x)=2(x+1)=2x+2$
• •Right: $(2f+2g)(x)=2x+2$

Same function.

2) Field addition over vectors

Polynomial visualization (coefficient scaling):

Let $p(x)=x^2+x$ and $a=2$, $b=-5$.

Compute the left side:

Compute the right side:

They match.

Associativity with field multiplication

This says: it doesn’t matter whether you scale by $b$ then $a$, or scale once by $ab$.

In $\mathbb{R}^n$ this is obvious; in function spaces it is still pointwise obvious:

If $(b\mathbf{v})(x) = b\cdot v(x)$, then

Identity scalar

This is the “do nothing” scale factor.

In polynomials: $1\cdot p(x)=p(x)$.

In matrices: $1\cdot A=A$.

Mini-checklist animation (guided pass/fail)

When testing “is this a vector space?”, use a consistent order:

1) What is V? (the set)

2) What is F? (the scalars)

3) How are + and scalar multiplication defined?

4) Closure tests (fast failure):

• •u+v ∈ V?
• •$a\mathbf{v}$ ∈ V?

5) Zero and inverse (often fails in restricted sets):

• •Is there a zero object in V?
• •Is −v always in V?

6) Distributivity + identity (consistency)

Interactive canvas idea: Provide toggles for each axiom. For a candidate set (e.g., “nonnegative functions”), clicking “additive inverse” highlights a counterexample: pick $f(x)=1$ then show −f(x)=−1 not in the set.

This gives learners a reliable mental procedure instead of a memorized list.

Why vector spaces are the stage for linear algebra

Most linear algebra concepts are really about linear combinations:

To even talk about this expression, you need:

• •addition to combine vectors,
• •scalar multiplication to apply coefficients,
• •closure so the result stays in your world,
• •distributivity so algebraic manipulation is valid.

That’s exactly what the vector space axioms guarantee.

Linear combinations are “mixing recipes”

Think of v₁, v₂ as ingredients and scalars as amounts.

If V is a vector space, then every recipe output is still a valid element of V.

Example: polynomial mixing

Let $V=P_2$ (polynomials degree ≤ 2 over $\mathbb{R}$). Take

$p_1(x)=1$, $p_2(x)=x$, $p_3(x)=x^2$.

A linear combination is

which is exactly “an arbitrary quadratic.”

So the vector space structure explains why $

\{1, x, x^2\}$can generate all of$P_2$.

Connection to linear independence (the next node)

The next concept—linear independence—asks whether a set of vectors contains redundancy.

A set $\{\mathbf{v}_1,\dots,\mathbf{v}_k\}$ is linearly independent if the only way to get the zero vector from a linear combination is the trivial way.

Formally (over field F):

This definition relies on:

• •the existence of 0,
• •the ability to add and scale vectors,
• •distributive properties to rearrange equations.

So: vector spaces are the rules of the game; linear independence is one of the key strategies.

Two non-geometry spaces you’ll see again

1) Function space (signals)

Let $V$ be the set of continuous functions on [0,1], scalars $F=\mathbb{R}$.

• •Adding signals adds amplitudes.
• •Scaling changes volume/brightness.

Vector-space thinking leads directly to Fourier series, least squares, and projections.

2) Matrix space (data)

Let $V$ be the set of all $m\times n$ real matrices, scalars $\mathbb{R}$.

• •Addition combines datasets.
• •Scaling changes units.

This becomes central in machine learning: datasets, parameter matrices, and gradients live in vector spaces.

A final habit to carry forward

Whenever you meet a new “vector-like” object (polynomials, functions, sequences, matrices), ask:

• •What is the zero element?
• •What does it mean to negate something?
• •Is closure obvious or subtle?

That habit will make the next nodes (linear independence, span, basis) feel natural instead of magical.