Let S have rows (2, 0) and (0, 1). Applied to a column vector (x, y), it produces (2x, y): the horizontal component doubles while the vertical component stays fixed.

Let R have rows (0, −1) and (1, 0). It sends (x, y) to (−y, x), a quarter-turn anticlockwise. Try the vector (1, 0) and then (0, 1). These two simple checks make the action visible.

Did you know? The columns of a two-dimensional transformation matrix tell you where the two standard basis vectors go. That is why testing those basis vectors reveals so much. You are examining the transformation’s building blocks rather than guessing from its name.

Begin at v = (1, 2). Stretch first: Sv = (2, 2). Then rotate: R(Sv) = (−2, 2). Now reverse the order. Rv = (−2, 1), then S(Rv) = (−4, 1).

The two answers differ. For column-vector convention, RS means apply S first and R second. Reading the product from left to right as a timetable is a common source of mistakes.

Calculate the products to check: RS has rows (0, −1) and (2, 0); SR has rows (0, −2) and (1, 0). Applying them to v reproduces the two results. One numerical example disproves the claim that all matrix products commute. It does not show that no pair ever commutes.

The same order issue occurs with ordinary functions. If f(x) = 2x and g(x) = x + 3, then f(g(x)) = 2x + 6, while g(f(x)) = 2x + 3. “Apply both” is incomplete information; we also need the sequence.

Matrix composition packages this sequence for linear transformations. The output of the first action becomes the input of the second. Dimensions must match, just as a process needs an output that the next process can accept.

An imagined digital map of a Punggol walking route offers a useful example: scaling its horizontal coordinates and rotating the map generally produce different displays depending on order. This is a teaching model, not a description of any particular mapping app’s software.

S is reversible because doubling x can be undone by halving it. Its inverse has rows (1/2, 0) and (0, 1). R is reversible through a clockwise quarter-turn, with inverse rows (0, 1) and (−1, 0).

To undo RS, undo R first and S second. The inverse is S⁻¹R⁻¹. Recover the original vector from (−2, 2): undo the rotation to get (2, 2), then undo the stretch to get (1, 2).

The pattern connects to dressing and unpacking a bag: undo the last operation first. The analogy helps with order, but an algebraic check still matters. Multiply the proposed inverse by the original transformation and verify the identity matrix.

Consider P with rows (1, 0) and (0, 0). It maps (x, y) to (x, 0). The vertical coordinate disappears. Both (3, 1) and (3, 8) become (3, 0), so a later operation cannot identify which input was used.

P has no inverse on the whole plane. For a 2 × 2 matrix with rows (a, b) and (c, d), a nonzero determinant ad − bc permits an inverse; a zero determinant does not. Here the determinant is zero.

This is different from merely finding a calculation difficult. The missing information makes unique recovery impossible. It is also why an average or a compressed summary cannot always reconstruct all the original data.

Challenge yourself with a reflection F that sends (x, y) to (−x, y). Compare RF and FR on (2, 3). RF gives (−3, −2); FR gives (3, 2). Draw the two routes to see the distinction.

Then ask whether uniform scaling U, which multiplies both coordinates by 2, commutes with R. It does: both orders send (x, y) to (−2y, 2x). This contrast prevents the false shortcut “matrix order always changes the result”.

In a 3-pax activity, rotate the roles of calculator, diagram reader and checker. Each learner should eventually justify the operation order and identify whether information is preserved. Continue to the existing matrices guides for foundational practice, and to functions, vectors and inverse thinking for the larger connection.