Linear independence — formal definition, testing: Visual walkthrough

Step 1 — What is a vector, and what is an arrow doing?

WHAT. A vector in $R^{2}$ is just a pair of numbers, like $(3, 1)$ . We picture it as an arrow starting at the origin (the point $(0, 0)$ ) and ending at that pair of coordinates. The first number says how far right, the second how far up.

WHY. Everything that follows — "combining vectors", "reaching a point" — is easier to trust when you see it as arrows you can lay end to end. We need this picture locked in before any algebra.

PICTURE. The red arrow is $v_{1} = (3, 1)$ : 3 across, 1 up. The cyan arrow is $v_{2} = (1, 2)$ .

Figure — Linear independence — formal definition, testing

Step 2 — What is a linear combination? (scaling and adding arrows)

WHAT. Take a number $c$ (a scalar) and a vector $v$ . Then $c v$ stretches the arrow by factor $c$ : $2 v$ is twice as long, $- 1 v$ flips it to point backwards, $0 v$ crushes it to $0$ . Adding two vectors means putting them tip to tail.

$c_{1} v_{1} + c_{2} v_{2}$

Term by term: $c_{1}$ stretches $v_{1}$ ; $c_{2}$ stretches $v_{2}$ ; the $+$ lays the second stretched arrow onto the tip of the first. The final tip is where you land.

WHY. The definition of independence is entirely about combinations equalling $0$ . So we must first see, physically, what a combination looks like — a recipe of stretches that carries you to a destination.

PICTURE. Amber dashed arrows show $1.5 v_{1}$ and $- 0.5 v_{2}$ laid tip to tail; the black arrow is where their sum lands.

Step 3 — The special destination: landing back at the origin

WHAT. Now we ask the key question: can a combination land you back at $0$ ? That is, can you pick scalars so that $c_{1} v_{1} + c_{2} v_{2} = 0 ?$ There is always one boring way: $c_{1} = 0, c_{2} = 0$ — take no steps at all, you never leave the origin. That is the trivial solution.

WHY. Returning to $0$ with nonzero steps means the arrows secretly cancel — one arrow's stretch is undone by the others. That cancellation is exactly redundancy. So "can we hit $0$ the exciting way?" is the independence question.

PICTURE. Left: independent arrows — the only way back to $0$ is take-no-steps. Right: two parallel arrows — you can walk out along one and walk back along the other, a nonzero round trip to $0$ .

Step 4 — Why a nonzero round trip = "one vector is a copy of the others"

WHAT. Suppose an exciting solution exists, say $c_{1} \neq = 0$ . Divide the whole equation by $c_{1}$ and move terms across: $c_{1} v_{1} + c_{2} v_{2} = 0 ⟹ v_{1} = - \frac{c _{2}}{c _{1}} v_{2} .$ Reading it: $v_{1}$ equals a scaled copy of $v_{2}$ . The number $- \frac{c _{2}}{c _{1}}$ is just the stretch factor. So $v_{1}$ carried no new direction — it was a passenger.

WHY. This is the bridge between the algebra (equation $= 0$ ) and the intuition ("no vector built from the others") that the parent note promised. Seeing it geometrically makes it undeniable.

PICTURE. The red arrow $v_{1}$ sits exactly along the line through the cyan arrow $v_{2}$ — same line, different length. Redundancy made visible.

Step 5 — Packing the question into a matrix

WHAT. Stack the vectors as columns of a matrix $A$ and collect the scalars into a column $c = (c_{1}, c_{2})$ . Then the matrix–vector product is the linear combination: $A c = ∣ v_{1} ∣ ∣ v_{2} ∣ [c_{1} c_{2}] = c_{1} v_{1} + c_{2} v_{2} .$ So the independence question becomes: does $A c = 0$ have only $c = 0$ ? — a homogeneous system.

WHY. Drawing arrows works for 2 or 3 vectors, but in higher dimensions we need machinery. A matrix lets a computer (or row reduction) answer the same geometric question. Nothing new is happening — just repackaging Steps 2–4.

PICTURE. The two arrows slide in to become the two columns of $A$ ; the label shows how column $\times$ scalar rebuilds the combination.

Step 6 — Counting pivots: the actual test

WHAT. Row-reduce $A$ to staircase form and count pivots (the leading nonzero entry of each nonzero row). A column with no pivot corresponds to a free variable — a knob you can turn freely, giving infinitely many nonzero solutions.

Pivot in every column $\Rightarrow$ $rank (A) = n$ $\Rightarrow$ only $c = 0$ $\Rightarrow$ independent.
A pivotless column $\Rightarrow$ free variable $\Rightarrow$ exciting solutions exist $\Rightarrow$ dependent.

WHY. This turns "can we hit $0$ the exciting way?" into a finite, mechanical count. See rank — it is the number of independent columns.

PICTURE. Two staircases: left has a pivot in each column (independent); right has a zero row, so its last column is pivotless — a free variable, dependence.

Step 7 — The square shortcut: determinant = signed area

WHAT. When you have exactly $n$ vectors in $R^{n}$ (a square $A$ ), the determinant measures the (signed) area/volume the arrows enclose. If they're dependent they lie on a line/plane — the enclosed area collapses to $0$ . $det A \neq = 0 ⟺ independent ⟺ A is [[I n v er t ib l e ma t r i x t h eor e m ∣ in v er t ib l e]] .$

WHY. It's a one-number test only when square. Term by term for $2 \times 2$ : $det [a c b d] = a d - b c$ is the area of the parallelogram the two columns span; area $0$ means they're parallel (Step 4's picture).

PICTURE. Left: independent pair spans a real parallelogram (amber fill, area $\neq = 0$ ). Right: parallel pair — the parallelogram is squashed flat, area $0$ .

Step 8 — Every edge case, drawn

WHAT. Three scenarios the count of pivots handles automatically:

Zero vector present. If $v_{1} = 0$ , then $1 \cdot 0 = 0$ is an exciting solution ( $c_{1} = 1 \neq = 0$ ). Always dependent.
Too many vectors ( $n >$ dimension). $A$ is short and wide, so at most (dimension) pivots for more columns $\Rightarrow$ a pivotless column always exists. Always dependent — you can't have more independent directions than the space has.
A single nonzero vector. One arrow that isn't $0$ can only reach $0$ by $c = 0$ . Always independent.

WHY. The parent note listed these as rules; here you see why each is forced, so you never meet a case unprepared.

PICTURE. Three panels: the flattened $0$ arrow; three fat columns in a 2-row matrix with a guaranteed free column; a lone nonzero arrow that only trivially cancels.

The one-picture summary

One figure, the whole chain: arrows → combination → return to $0$ ? → columns of $A$ → pivots → independent or dependent, with the square-case determinant shortcut branching off.

Recall Feynman retelling — the walkthrough in plain words

Picture arrows from the middle of the page. A "combination" is a recipe: stretch this one by so much, that one by so much, lay them tip to tail, see where you land. The whole game is one question: can a recipe with real (nonzero) stretches carry you back to the exact centre? If the only way home is "take no steps", your arrows are all pulling in genuinely different directions — independent. But if you can march out along one arrow and sneak back along the others, then one of them was a copy in disguise — dependent, a passenger. To check without drawing, stand the arrows up as the columns of a box called $A$ and row-reduce it: count the "steps" (pivots) in the staircase. A pivot for every column means no free knob to turn, so only the do-nothing recipe reaches zero — independent. A missing pivot is a free knob — spin it and you find a real round trip to zero — dependent. When the box is square you get a bonus: the determinant is the area the arrows enclose, and a squashed-flat area (zero) is the same as "dependent". And the tricky freebies: a zero arrow is always a passenger; more arrows than the space has dimensions must have a passenger; a single non-zero arrow never does.

Connections

Linear independence — formal definition, testing — the parent this page visualises.
Homogeneous systems and null space — $A c = 0$ ; a nonzero solution = dependence.
Rank of a matrix — pivots counted in Step 6.
Determinant — the square-case area test in Step 7.
Invertible matrix theorem — independent columns ⇔ invertible.
Basis and dimension · Span and spanning sets — independence + spanning = a basis.