Dimension — basis cardinality: Visual walkthrough

Before line one, three words we will use constantly. Each is anchored to a picture in Step 1.

We work in the flat plane (arrows on paper). Everything transfers to higher dimensions unchanged — the pictures just get harder to draw, not harder to think about.

Step 1 — Meet the two casts of characters

WHAT. We have two teams of arrows living in the same plane.

The spanning team $S = {w_{1}, \dots, w_{n}}$ (gray): enough arrows to build everything. They may overlap or be redundant — we do not require them to be independent.
The independent team $L = {v_{1}, \dots, v_{m}}$ (blue): no waste — each points somewhere genuinely new.

WHY. The whole lemma is a claim comparing these two numbers, so we must pin down exactly what each team promises. $S$ promises coverage. $L$ promises no redundancy. Coverage is a resource; no-redundancy is a demand on that resource.

PICTURE.

In the figure, the gray $w$ -arrows overlap — that is fine, a spanning set is allowed to be sloppy. The blue $v$ -arrows point in clearly distinct directions — that is what independent looks like.

Step 2 — Every arrow is a recipe (why spanning lets us write $v_{1}$ )

WHAT. Take the first independent arrow $v_{1}$ . Because the gray team spans, $v_{1}$ is a combination of gray arrows: $v_{1} = how much of w_{1} a_{1} w_{1} + of w_{2} a_{2} w_{2} + \dots + of w_{n} a_{n} w_{n} .$

WHY THIS TOOL — a linear combination. We reach for a weighted sum because that is the only operation a vector space gives us: scale arrows and add them. The word "spans" literally means "reachable by such a sum," so writing $v_{1}$ this way is just cashing in the spanning promise. The numbers $a_{1}, \dots, a_{n}$ are the amounts of each ingredient — a recipe for $v_{1}$ .

PICTURE.

The dashed arrows show $a_{1} w_{1}$ , $a_{2} w_{2}$ , … laid head-to-tail; their sum lands exactly on the blue tip of $v_{1}$ .

Step 3 — At least one ingredient is real (the first swap)

WHAT. Not all the $a_{i}$ can be zero. If they were, the recipe would read $v_{1} = 0$ — but $v_{1}$ comes from an independent team, and independent arrows are never the zero arrow. So some $a_{i} \neq = 0$ ; relabel so that $a_{1} \neq = 0$ . Now solve the recipe backwards for $w_{1}$ : $w_{1} = \frac{1}{a _{1}} (the new arrow v_{1} - the other grays a_{2} w_{2} - \dots - a_{n} w_{n}) .$

WHY. This line says $w_{1}$ is now redundant: anything you could build with $w_{1}$ you can rebuild using $v_{1}$ and the remaining grays. The tool here is just rearranging the equation — legal because $a_{1} \neq = 0$ means we may divide by it. Division by $a_{1}$ is where " $a_{1} \neq = 0$ " earns its keep; had $a_{1}$ been $0$ we could not isolate $w_{1}$ .

PICTURE.

We kick out $w_{1}$ (faded gray, crossed) and swap in $v_{1}$ (solid blue). The new team ${v_{1}, w_{2}, \dots, w_{n}}$ still has $n$ arrows and — crucially — still spans everything the old team did.

Step 4 — Keep swapping, and watch for the trap

WHAT. Suppose after $k$ swaps our spanning team is a mix: ${k blues already in v_{1}, \dots, v_{k}, n - k grays left w_{k + 1}, \dots, w_{n}} .$ Bring in the next blue $v_{k + 1}$ . Since this mixed team still spans, write: $v_{k + 1} = amounts of old blues b_{1} v_{1} + \dots + b_{k} v_{k} + amounts of remaining grays c_{k + 1} w_{k + 1} + \dots + c_{n} w_{n} .$

WHY. Same spanning promise, applied to the current team. The recipe now has two flavours of ingredient: blues we already placed ( $b$ 's) and grays still standing ( $c$ 's). We split them on purpose, because the two flavours play opposite roles in the next step.

PICTURE.

Blue contributions in blue, gray contributions in gray, summing to $v_{k + 1}$ .

Step 5 — Why we never run out of grays to kick (the heart)

WHAT. Look at the gray amounts $c_{k + 1}, \dots, c_{n}$ . At least one is nonzero.

WHY — this is exactly where independence is used. Suppose every $c_{j} = 0$ . Then the recipe collapses to $v_{k + 1} = b_{1} v_{1} + \dots + b_{k} v_{k},$ i.e. $v_{k + 1}$ is built entirely from earlier blues. But the blue team is independent — no blue is a combination of the others. Contradiction. So some $c_{j} \neq = 0$ : a gray is still genuinely present, and we can kick it out and swap $v_{k + 1}$ in. The team size stays $n$ , and it still spans.

PICTURE.

Left panel — the forbidden world where all $c_{j} = 0$ : $v_{k + 1}$ (red) lands inside the blue span, which independence outlaws. Right panel — the real world: a nonzero $c_{j}$ keeps a gray in play (green arrow), so the swap goes through.

Step 6 — Count the swaps: the inequality falls out

WHAT. Each swap uses up one blue (added) and one gray (removed). To place all $m$ blues we need $m$ grays to remove. There are only $n$ grays. So: $m \leq n .$

WHY. This is pure bookkeeping. Independence (Step 5) guaranteed a gray is always available to remove, so the swapping never jams before the blues run out. If $m$ were bigger than $n$ we would try an $(n + 1)$ -th swap with no gray left — impossible. Hence blues cannot outnumber grays.

PICTURE.

A tally: blues-consumed on one axis, grays-remaining on the other, marching down until grays hit zero — the wall that caps $m$ .

Step 7 — Edge cases: don't skip these

WHAT & WHY. We must show every input behaves.

$m = 0$ (empty independent set). The empty set is vacuously independent, and $0 \leq n$ always. No swaps needed; claim trivially holds.
$m = n$ (blues exactly fill the grays). The final swap removes the last gray. The team is now all blue, size $n$ , still spanning — this is precisely the case that will force the Dimension Theorem in Step 8.
Redundant / dependent spanning team. The grays were never required to be independent. Two grays might be identical; a $w_{j}$ might be $0$ . None of that breaks a single step — we only ever used that the grays span. This generality is why the lemma is so powerful.
$v_{k + 1} = 0$ can't happen. A zero arrow would make the blue team dependent, contradicting our hypothesis — so Step 5's contradiction machinery is never even tested against a zero blue.

PICTURE.

Three mini-panels: empty $L$ (nothing to do), a duplicated gray (swap still fine), and $m = n$ (all grays consumed, all-blue team survives).

Step 8 — Two bases, applied both ways ⇒ equal size

WHAT. Let $B_{1}$ (size $p$ ) and $B_{2}$ (size $q$ ) each be a basis of the same space $V$ . A basis is both independent and spanning, so each can play either role in the lemma.

Treat $B_{1}$ as the blue team (independent) and $B_{2}$ as the gray team (spanning): $p \leq q .$
Now swap the roles: $B_{2}$ blue, $B_{1}$ gray: $q \leq p .$

Two arrows squeeze $p$ and $q$ from both sides: $p \leq q and q \leq p ⟹ p = q .$

WHY. The lemma alone gives only one inequality. It is the double life of a basis — independent and spanning at once — that lets us fire the lemma twice, in opposite directions, trapping the two counts into equality. That equal number is the dimension of $V$ .

PICTURE.

A see-saw: $p \leq q$ pushing down one side, $q \leq p$ the other, balancing at $p = q$ .

The one-picture summary

Read it left to right: a gray spanning pile → blues march in one at a time, each swap kicking a gray out (independence guarantees a gray is always available) → the tally caps blues at grays, giving $m \leq n$ → fire it both ways on two bases → the see-saw balances at $p = q = dim V$ .

Recall Feynman: the whole walk in plain words

You've got a messy pile of gray sticks that's enough to build any shape (that's spanning — the pile may have duplicates, doesn't matter). You've also got a neat set of blue sticks, no two pointing the same way (that's independence). Now play a game: one at a time, feed a blue stick into the pile and throw out a gray one, always keeping the pile big enough to build everything. The magic move is this: whenever you add a blue, could you be forced to remove another blue instead of a gray? No — because if only blues were left doing the work, your new blue would be a copy of old blues, and blues are never copies. So there's always a gray to throw. Since each round eats one blue and one gray, you can't have more blues than you started with grays: independent ≤ spanning. Finally, a basis is a stick set that's both neat and enough — so it can be the blue team or the gray team. Point two bases at each other both ways and you get "this one ≤ that one" and "that one ≤ this one." The only way both can be true is if they're equal. That equal number is the dimension — the fixed brick-count of the space, no matter which colours you use.

Connections

Steinitz Exchange Lemma — the engine we watched swap-by-swap.
Linear Independence — the "some gray coefficient ≠ 0" step (Step 5).
Spanning Sets — supplies the recipe in Steps 2 and 4.
Basis of a Vector Space — the both-at-once object measured in Step 8.
Dimension — basis cardinality — the parent result this page derives.
Rank-Nullity Theorem — dimension counting applied to linear maps.
Coordinates and Change of Basis — different bases, same size, in action.