Visual walkthrough — Properties — row operations, multiplicativity

Before any algebra, we agree on ONE picture that everything hangs on.

---

Step 1 — What is a matrix doing to a picture?

WHAT. Take the simplest square shape: the unit square — a square one unit wide and one unit tall, sitting with its corner at the origin. Its two edges are the arrows ${\mathbf{e}}_1=(1,0)$ (pointing right) and ${\mathbf{e}}_2=(0,1)$ (pointing up).

A $2\times 2$ matrix $A=\left(\begin{array}{cc}a& b\\ c& d\end{array}\right)$ is a machine: it sends the right-arrow ${\mathbf{e}}_1$ to the first column $(a,c)$ and the up-arrow ${\mathbf{e}}_2$ to the second column $(b,d)$.

• $a,c$ — where the right edge lands (its new $x$ and $y$).
• $b,d$ — where the up edge lands.

The unit square becomes a slanted parallelogram spanned by those two new arrows.

WHY. The whole point of the determinant is a single number: $\det A=\text{signed area of that parallelogram}.$ "Signed" means: positive if the two arrows keep their original anticlockwise turn, negative if the machine flipped them into a mirror image. Everything below is just: what happens to this area when I fiddle with the rows?

PICTURE.

---

Step 2 — The three seed facts (drawn, not assumed)

Everything follows from three pictures. We give each a plain name.

WHY these three? They are the minimum honest description of "signed area." (N) fixes the unit. (A) says a squashed-flat shape has no area. (M) says area grows in proportion to how far you stretch an edge — that is what "area" means. See Multilinear Alternating Forms.

PICTURE.

Annotating (M) term by term, first slot: $\underset{\text{area of the tall combined edge}}{\underbrace{D(u+w,v)}}=\underset{\text{area from }u}{\underbrace{D(u,v)}}+\underset{\text{area from }w}{\underbrace{D(w,v)}}.$ Each piece keeps the same second arrow $v$ (the shared base), so the areas literally stack.

---

Step 3 — Swap two rows: the sign flips (derived from the seeds)

WHAT. Exchange the two arrows: $D(u,v)\to D(v,u)$. We show it must equal $-D(u,v)$ — no new axiom needed.

WHY the trick. Feed the machine an arrow that is a sum, $u+v$, in both slots. By (A) it is zero (both slots equal). Then crack it open with (M):

$$D(\stackrel{\text{slot 1}}{\overbrace{u+v}},\stackrel{\text{slot 2}}{\overbrace{u+v}})=0.$$

Expand slot 1, then slot 2:

$$\underset{=0\text{ by (A)}}{\underbrace{D(u,u)}}+D(u,v)+D(v,u)+\underset{=0\text{ by (A)}}{\underbrace{D(v,v)}}=0.$$

The two end terms vanish because each has a repeated arrow. What is left:

$$D(u,v)+D(v,u)=0⟹==D(v,u)=-D(u,v)==.$$

PICTURE. The two parallelograms are the same shape, but the turn from first arrow to second reverses — anticlockwise becomes clockwise. That mirror-flip is the minus sign.

---

Step 4 — Scale a row: the determinant scales

WHAT. Stretch one edge-arrow by a factor $k$: $D(k{R}_1,R_2)$.

WHY. This is exactly the second half of (M): pulling a scalar out of one slot.

$$\underset{\text{one edge }k\text{ times longer}}{\underbrace{D(k{R}_1,R_2)}}=\underset{\text{stretch factor}}{\underbrace{k}}\underset{\text{original area}}{\underbrace{D(R_1,R_2)}}.$$

The base $R_2$ is untouched; the height along $R_1$ is $k$ times bigger, so the area is $k$ times bigger.

PICTURE.

---

Step 5 — Replacement (shear): the determinant does NOT change

WHAT. Add a multiple of one arrow to another: $R_1\to R_1+k{R}_2$. Show the area is unchanged.

WHY. Split the new first slot with (M):

$$D(\underset{\text{sheared edge}}{\underbrace{R_1+k{R}_2}},R_2)=\underset{\text{original}}{\underbrace{D(R_1,R_2)}}+\underset{\text{= }k\cdot 0\text{ by (A)}}{\underbrace{kD(R_2,R_2)}}.$$

The extra piece has $R_2$ in both slots — a flat line, area $0$. So it dies: $==D(R_1+k{R}_2,R_2)=D(R_1,R_2).==$

PICTURE. Sliding the tip of one arrow along the direction of the other is a shear. Base stays the same, height (perpendicular distance to the base line) stays the same — same area. Like pushing the top of a deck of cards sideways: same volume.

---

Step 6 — The degenerate case: a flat parallelogram means $\det =0$

WHAT. What if the two arrows point the same way (one is a multiple of the other), $R_2=m{R}_1$?

WHY. By scaling (Step 4) then alternating (Step 2):

$$D(R_1,m{R}_1)=m\underset{=0}{\underbrace{D(R_1,R_1)}}=0.$$

The parallelogram has collapsed to a line — no area, so $\det =0$. This is the picture behind a singular matrix: the machine squashes the plane onto a line, losing a dimension.

PICTURE.

---

Step 7 — Row reduction reads the determinant off the diagonal

WHAT. Use Steps 3–5 to grind $A$ down to an upper-triangular matrix $U$ (all zeros below the diagonal), tracking what each move costs.

WHY. A triangular matrix's arrows form a parallelogram whose signed area is just the product of the diagonal entries $u_{ii}$ (each cofactor expansion has one surviving term). And we know the price of each move:

• every swap multiplies $\det$ by $-1$,
• every division of a row by $c$ multiplies $\det$ by $\frac{1}{c}$,
• every replacement is free.

Putting it together, with $s$ = number of swaps and $\prod k$ the product of scalars you divided by:

$$\det A=(-1{)}^s\cdot \frac{1}{\prod k}\cdot \underset{\text{product of pivots}}{\underbrace{\prod _i{u}_{ii}}}.$$

PICTURE (worked, Example 1 of the parent).

\xrightarrow[\text{(free)}]{R_2-2R_1,\;R_3-3R_1} \begin{pmatrix}2&3&1\\0&1&3\\0&3&6\end{pmatrix} \xrightarrow[\text{(free)}]{R_3-3R_2} \begin{pmatrix}2&3&1\\0&1&3\\0&0&-3\end{pmatrix}$$ No swaps ($s=0$), no divisions ($\prod k=1$), pivots $2,1,-3$: $$\det A=(-1)^0\cdot\frac{1}{1}\cdot(2\cdot 1\cdot(-3))=-6.$$ --- ## Step 8 — Multiplicativity: two machines multiply their areas **WHAT.** Show $\det(AB)=\det A\,\det B$. **WHY (picture-first).** Apply the machine $B$: area scales by $\det B$. Then apply $A$ on top: that already-scaled shape scales again by $\det A$. Two stretches in a row **multiply**:

\underbrace{\det(AB)}{\text{do }B\text{ then }A}=\underbrace{\det A}{\text{2nd stretch}}\cdot\underbrace{\det B}_{\text{1st stretch}}.

**WHY (algebra).** Each row operation is left-multiplication by an [[Elementary Matrices|elementary matrix]] $E$, and Steps 3–5 give its determinant exactly: swap $\det E=-1$, scale $\det E=k$, replacement $\det E=1$. In every case $\det(EA)=\det E\,\det A$. An invertible $A$ is a product of elementaries $A=E_1\cdots E_m$, so $$\det(AB)=\det(E_1)\cdots\det(E_m)\det(B)=\det A\,\det B.$$ If $A$ is singular ($\det A=0$), Step 6 says its arrows already collapse; $AB$ collapses too, so both sides are $0$. **PICTURE.** ![[deepdives/dd-maths-4.5.22-d2-s08.png]] > [!example] Numbers check > $A=\begin{pmatrix}1&2\\3&4\end{pmatrix}$ ($\det=-2$), $B=\begin{pmatrix}2&0\\1&2\end{pmatrix}$ ($\det=4$), $AB=\begin{pmatrix}4&4\\10&8\end{pmatrix}$, $\det(AB)=32-40=-8=(-2)(4)$. ✓ > [!formula] Free corollaries > $\det(A^{-1})=\dfrac{1}{\det A}$, $\;\det(A^k)=(\det A)^k$, and $\det(A^T)=\det A$ (so all row pictures work for columns too). The last one links to [[Eigenvalues — det as product of eigenvalues]]. --- ## The one-picture summary ![[deepdives/dd-maths-4.5.22-d2-s09.png]] One parallelogram, four fates: **swap** flips it (sign), **scale** grows it (factor $k$), **shear** slides it (no change), **collapse** flattens it ($\det=0$). Row reduction is just a sequence of these until the shape is a triangular box whose area you read off the diagonal — and stacking two machines multiplies their areas. > [!recall]- Feynman: tell it to a 12-year-old > A matrix is a machine that stretches a paper square into a slanted parallelogram. The determinant is just *how much bigger the area got*, plus a minus sign if the shape got flipped over like in a mirror. > - **Swap** the two edge-arrows → same shape, mirror-flipped → area gets a minus sign. > - **Stretch** one edge $k$ times longer → the area gets $k$ times bigger. Stretch *both* edges (that's what $2A$ does) → area gets $k$ for *each* edge, so $\times 4$ for a square. > - **Lean** one edge sideways along the other (a shear) → the base and the height stay the same, so the area doesn't change at all. That's why you're allowed to simplify a matrix by adding rows — it never lies about the area. > - If the two arrows point the same way, the parallelogram is squashed flat into a line — no area — and the determinant is zero. > - Do one machine, then another: the areas just multiply. That's $\det(AB)=\det A\cdot\det B$. > [!recall]- Quick self-test > - Why does a swap give a minus? ::: Same parallelogram, opposite handedness (orientation flip). > - Why is shear free? ::: Base and height unchanged; the extra term has two equal rows → area $0$. > - $\det(2A)$ for $2\times2$? ::: $4\det A$ (both edges double). > - When is $\det=0$? ::: Arrows collinear → flat parallelogram → singular. > - $\det(AB)=?$ ::: $\det A\,\det B$ (stretches multiply).