Vector Spaces

Forget the arrow. Keep the rules.

In the Vectors and Linear Transformations nodes you learned to add arrows tip-to-tail and stretch them by scalars. Fine. Useful. But a kid who only knows "vectors are arrows" is exactly like a kid who thinks "3" is the squiggle — confusing the costume for the actor. Remember the very first lesson, What Is a Number? Same fucking move. We strip away everything incidental and keep only the structure that does the work.

So here is the deranged, beautiful question Hermann Grassmann and Giuseppe Peano asked in the 1800s while everyone else was doing sensible work: what is the minimum equipment that makes "adding and scaling" behave the way we want? Their answer is a checklist. A set $V$, an addition $+$ that takes two elements of $V$ and returns one, a scalar multiplication that takes a real number $c$ and an element $v$ and returns $cv \in V$, and ten laws. Hit all ten and you've earned the name vector space. The elements are then called vectors — not because they're arrows, but because they obey the arrow rules. That's it. That's the whole damn game.

The ten axioms (grouped so they don't look like noise)

Let $u, v, w \in V$ and $c, d \in \mathbb{R}$. A vector space (over $\mathbb{R}$) satisfies:

Closure (the operations don't escape the set):

1. $u + v \in V$.
2. $cv \in V$.

Addition behaves like a respectable group: 3. $u + v = v + u$ (commutative). 4. $(u + v) + w = u + (v + w)$ (associative). 5. There exists a zero vector $\mathbf{0} \in V$ with $v + \mathbf{0} = v$ for all $v$. 6. Every $v$ has an additive inverse $-v \in V$ with $v + (-v) = \mathbf{0}$.

Scalar multiplication plays nicely with everything: 7. $1\,v = v$. 8. $c(dv) = (cd)v$. 9. $c(u + v) = cu + cv$ (distributes over vector addition). 10. $(c + d)v = cv + dv$ (distributes over scalar addition).

That's the whole factory. Notice there is no mention of arrows, length, angle, or coordinates. Those are luxuries some vector spaces happen to have. The axioms are the bare skeleton.

The zoo: wildly different animals, identical DNA

Here's the payoff, the entire business model of abstraction. Each of these is a vector space — verify a couple in your head right now:

• $\mathbb{R}^n$ — tuples $(x_1, \dots, x_n)$, added componentwise. The arrows you know.
• $P_n$ — polynomials of degree $\le n$, like $3x^2 - x + 5$. Add them, scale them — still a polynomial of degree $\le n$. A vector space whose "vectors" are functions.
• $M_{m\times n}$ — all $m \times n$ matrices, added entrywise (straight out of the Matrices stratum). The zero vector is the zero matrix.
• $F(\mathbb{R})$ — all functions $f : \mathbb{R} \to \mathbb{R}$, with $(f+g)(x) = f(x)+g(x)$. Infinite-dimensional and gorgeous.
• The zero space $\{\mathbf{0}\}$ — a single element, which is its own zero and its own inverse. Pathetic, legal, occasionally crucial.

Now the magic: when we later prove "every vector space has a basis" or "$\mathbf{0}v = \mathbf{0}$", we prove it once, from the axioms, and it instantly holds for arrows AND polynomials AND matrices AND functions. One proof, infinite payoff. That is why we suffer the abstraction — and trust me, the payoff is worth every ounce of the suffering.

First axiom-proofs: squeezing facts from the rules

Time to do what this stratum is for. These facts look obvious, and that is exactly the trap. They are not given — they must be proved from the ten axioms, and every step gets a justification or it doesn't count. No justification, no credit, no mercy.

Theorem. For every $v \in V$, $\;0\,v = \mathbf{0}$. (Here $0$ is the scalar zero, $\mathbf{0}$ the zero vector — they are not obviously related.)

Proof. Start from a true scalar fact: $0 = 0 + 0$. Multiply $v$: $0\,v = (0 + 0)v = 0\,v + 0\,v \quad \text{(axiom 10)}.$ Now $0\,v \in V$ (axiom 2), so it has an inverse $-(0\,v)$ (axiom 6). Add it to both sides: $0\,v + \big(-(0\,v)\big) = \big(0\,v + 0\,v\big) + \big(-(0\,v)\big).$ Left side is $\mathbf{0}$ (axiom 6). Right side, regroup by associativity (axiom 4): $\mathbf{0} = 0\,v + \big(0\,v + (-(0\,v))\big) = 0\,v + \mathbf{0} = 0\,v \quad \text{(axioms 6, 5)}.$ So $0\,v = \mathbf{0}$. $\blacksquare$

Read that again — slowly. We never touched a coordinate. Not one. It holds for the zero polynomial, the zero matrix, the zero function, all at once. That's the whole business model of abstraction paying out in one shot.

Theorem. For every $v$, $\;(-1)v = -v$. (The scalar $-1$ produces the additive inverse.)

Proof. We show $(-1)v$ is an inverse of $v$; inverses are unique (a one-line argument I'll leave for the gauntlet), so it must equal $-v$. $v + (-1)v = 1\,v + (-1)v \quad\text{(axiom 7)} = (1 + (-1))v \quad\text{(axiom 10)} = 0\,v = \mathbf{0},$ using the theorem we just proved. So $(-1)v$ added to $v$ gives $\mathbf{0}$, making it the inverse of $v$, i.e. $(-1)v = -v$. $\blacksquare$

Savor that, you beautiful disaster. "Negative one times a vector flips it" is not a rule someone handed you in seventh grade — it's a consequence, dragged out of the axioms kicking and screaming like a proof-shaped cat from a proof-shaped bag.

Refuting impostors: which axiom dies?

The flip side of rigor is ruthlessness. Hand me a candidate and I'll find the axiom it murders.

Candidate: the line $L = \{(x, y) : y = x + 1\}$ in $\mathbb{R}^2$, with the usual operations. Vector space?

No. It fails closure under addition AND scaling, but the cleanest kill is the zero vector. Axiom 5 demands $\mathbf{0} = (0,0) \in L$. Is it? $0 = 0 + 1$? That's $0 = 1$, which is an outright lie. No zero vector, not a space. A line that misses the origin can never be a vector space — burn that into your skull right now, because the next node (Subspaces) is entirely about which subsets survive this ruthless test.

TL;DR before the gauntlet

The arrow was training wheels. The axioms are the bike. Everything from here — span, independence, basis, dimension, linear maps — is built on exactly these ten laws and nothing else.