Inverse of 2×2 matrix: Visual walkthrough — Maths deep dive

Step 0 — The words we are allowed to use

Before symbols appear, let us agree on pictures for each.

In the picture: the blue arrow is your original $x$ . The machine $A$ stretches and tilts it into the orange arrow $A x$ . The green dashed arrow is $A^{- 1}$ carrying orange back to blue. If undo works perfectly, applying $A$ then $A^{- 1}$ lands you exactly where you started — that is the meaning of $A^{- 1} A = I$ .

Step 1 — Write down the goal as a picture-equation

WHAT. We want a machine $A^{- 1}$ so that doing $A$ then $A^{- 1}$ changes nothing.

WHY. "Inverse" has no meaning until we pin it down. The only honest definition is: the machine that composes with $A$ to give the do-nothing machine. So we demand

A A^{- 1} = I .

PICTURE. Below, three grids in a row: the starting grid, the grid after $A$ has warped it, and the grid after $A^{- 1}$ has un-warped it back to the start. The far-left and far-right grids are identical — that identity is what the equation $A A^{- 1} = I$ is claiming.

Here each symbol earns its place:

$A$ — the known machine, $(a c b d)$ , four given numbers.
$A^{- 1} = (p r q s)$ — the unknown machine; $p, q, r, s$ are the four numbers we must find.
$I = (1001)$ — the do-nothing target.

Step 2 — Turn one matrix equation into four number equations

WHAT. Multiply the two machines out (see Matrix multiplication) and match entry by entry.

WHY. A matrix equation is secretly four ordinary equations, one per slot. Numbers are easier to solve than matrices, so we break it apart.

(a c b d) (p r q s) = (1001)

Reading each output slot (row of $A$ times column of $A^{- 1}$ ):

(1) top-left a p + b r = 1 (2) top-right a q + b s = 0

(3) bottom-left c p + d r = 0 (4) bottom-right c q + d s = 1

PICTURE. The figure shades which row of $A$ meets which column of $A^{- 1}$ to produce each of the four target slots — so you see why equation (1) has no $q$ or $s$ in it (they live in a different column).

Notice the split: equations $(1), (3)$ contain only $p$ and $r$ ; equations $(2), (4)$ contain only $q$ and $s$ . Two independent little puzzles.

Step 3 — Solve the first little puzzle for $p$ and $r$

WHAT. Use $(1)$ and $(3)$ together.

WHY. They share exactly the two unknowns $p, r$ — a clean 2-equation, 2-unknown system.

From $(3)$ : $c p + d r = 0$ . Solve for $r$ :

r = - \frac{c}{d} p (momentarily assume d \neq = 0)

Here the symbol $r$ is being expressed in terms of $p$ , so equation $(1)$ becomes single-unknown. Substitute into $(1)$ :

a p + b (- \frac{c}{d} p) = 1 ⟹ p = \frac{a d - b c}{d} (a - \frac{b c}{d}) = 1.

The bracket collapses to $\frac{a d - b c}{d}$ . So

p = \frac{d}{a d - b c}, r = - \frac{c}{a d - b c} .

PICTURE. The figure shows the substitution as a "funnel": two equations pour into one, the messy fraction $a - \frac{b c}{d}$ simplifies, and out drops $p$ . The quantity $a d - b c$ appears for the first time — highlighted, because it will haunt every remaining line.

$a d - b c$ — the recurring combination. We will name it in Step 5.
The minus in $r$ came directly from the minus in equation $(3)$ 's rearrangement — that is the origin of the sign-flip you will see in the final formula.

Step 4 — Solve the second little puzzle for $q$ and $s$

WHAT. Repeat Step 3 with $(2)$ and $(4)$ .

WHY. Identical structure, so identical method — no new ideas needed, just relabelled letters.

From $(2)$ : $a q + b s = 0 \Rightarrow q = - \frac{b}{a} s$ . Substitute into $(4)$ :

c (- \frac{b}{a} s) + d s = 1 ⟹ s \cdot \frac{a d - b c}{a} = 1 ⟹ s = \frac{a}{a d - b c}, q = - \frac{b}{a d - b c} .

PICTURE. Side-by-side mirror of Step 4's funnel, so you can literally see the symmetry: $p \to s$ swaps $d \to a$ , and $r \to q$ swaps $c \to b$ . The swap of the diagonal is being born right here.

Collect all four answers:

p = \frac{d}{a d - b c}, q = \frac{- b}{a d - b c}, r = \frac{- c}{a d - b c}, s = \frac{a}{a d - b c} .

Step 5 — Package the answer and name the divisor

WHAT. Pull the shared $\frac{1}{a d - b c}$ out front.

WHY. All four entries divide by the same thing. Factoring it out makes the pattern — swap, negate — jump out.

A^{- 1} = \frac{1}{a d - b c} (d - c - b a)

PICTURE. The figure overlays the original $A$ and its inverse: a curved blue arrow swaps the two diagonal entries; two red minus-badges land on $b$ and $c$ ; a green ring labels the shared divisor. The matrix $(d - c - b a)$ is the adjugate (see Adjugate and cofactors).

Step 6 — The degenerate case: what if $det A = 0$ ?

WHAT. Ask what happens when $a d - b c = 0$ .

WHY. We divided by $a d - b c$ four times. If it is zero, every entry becomes $\frac{number}{0}$ — undefined. The undo machine cannot exist. We must show why geometrically, not just algebraically.

PICTURE. When $a d - b c = 0$ , the two columns of $A$ point along the same line. The machine $A$ squashes the whole 2D grid onto that single line: area is crushed to zero. Two different input arrows can land on the same output arrow — so "undo" cannot know which one to send back. Information is lost forever. This is a singular matrix (see Singular vs non-singular matrices and Linear transformations and area).

Step 7 — Confirm the machine really undoes (all four slots)

WHAT. Multiply $A$ by our $A^{- 1}$ and check we get $I$ in every slot.

WHY. We forecast the answer; now we verify — cheap insurance against a sign slip (Contract rule: cover every case, including each entry).

A A^{- 1} = \frac{1}{a d - b c} (a c b d) (d - c - b a) = \frac{1}{a d - b c} (a d - b c c d - d c - ab + ba - c b + d a)

= \frac{1}{a d - b c} (a d - b c 0 0 a d - b c) = (1001) = I ✓

Term by term: the off-diagonal slots die because $- ab + ba = 0$ and $c d - d c = 0$ ; the diagonal slots each become $a d - b c$ , which the front factor cancels to $1$ . This is exactly the do-nothing machine — the undo works.

The one-picture summary

Everything on one canvas: goal $A A^{- 1} = I$ → four equations → solve → swap/negate/divide → the degenerate wall at $det A = 0$ .

Recall Feynman retelling — the whole walkthrough in plain words

I wanted a machine that undoes what $A$ does. "Undo" only means something if doing both in a row changes nothing, so I demanded $A A^{- 1} = I$ and wrote the unknown machine as four mystery numbers $p, q, r, s$ . Multiplying the two machines gave four small number-puzzles. Two of them only knew about $p$ and $r$ , the other two only about $q$ and $s$ — so I solved each pair on its own. Every answer came out as some original entry divided by the same magic number $a d - b c$ . Pulling that magic number out front, the leftovers spelled a neat rule: swap the two diagonal numbers, stick minus signs on the other two. That leftover grid is the adjugate; the magic number is the determinant. Last question: what if the magic number is zero? Then I'd be dividing by nothing — impossible. And it makes sense: a zero determinant means $A$ flattened the whole plane onto a line, so two different arrows can end up in the same place, and no undo machine could ever tell them apart. Zero determinant, no undo. Everything else, the boxed recipe works.

Active recall

What is the defining demand we impose to derive

A^{- 1}

That doing

A

then

A^{- 1}

changes nothing, i.e.

A A^{- 1} = I

Why does the matrix equation

A A^{- 1} = I

split into four number equations?

Because each output slot equals one row of

A

times one column of

A^{- 1}

, giving four independent scalar equations.

Which two equations solve for

p

and

r

, and why?

Equations (1) and (3), because they contain only

p

and

r

(a clean 2×2 system).

Where does the sign-flip on

b, c

in the final formula come from?

From the minus signs introduced when rearranging equations (2) and (3), e.g.

r = - \frac{c}{d} p

Why can the formula divide by

a d - b c

even though Step 3 assumed

d \neq = 0

The assumption only picks one algebra path; the true obstruction is

det A = 0

, and any nonzero det admits a valid path to the same formula.

Geometrically, why is there no inverse when

det A = 0

The columns are parallel, so

A

squashes the plane onto a line; distinct inputs map to the same output and information is lost.

In the verification

A A^{- 1}

, why do the off-diagonal entries vanish?

Because

- ab + ba = 0

and

c d - d c = 0

Connections

Inverse of 2×2 matrix — the parent result this page derives visually.
Matrix multiplication — used to split $A A^{- 1} = I$ into four equations and to verify.
Identity matrix — the do-nothing target of the derivation.
Determinant of a 2×2 matrix — the divisor $a d - b c$ born in Steps 3–4.
Adjugate and cofactors — the swap/negate grid packaged in Step 5.
Singular vs non-singular matrices — the $det A = 0$ wall in Step 6.
Linear transformations and area — why zero determinant means crushed area.
Solving linear systems — the payoff: $x = A^{- 1} b$ .