A point P has coordinates:
(1,2).
Write it as a column vector:
[[1],[2]].
Now apply the matrix:
R=[[0,−1],[1,0]].
Multiply:
R[[1],[2]]=[[−2],[1]].
The point has rotated 90° anticlockwise about the origin.
A transformation matrix is a compact rule for moving every coordinate vector in the same structured way.
The quick answer: columns carry the input
For a 2D transformation:
[[a,b],[c,d]][[x],[y]]=[[ax+by],[cx+dy]].
The original point enters as a 2×1 column vector.
The 2×2 matrix acts on it.
The output is another 2×1 coordinate vector.
Identity transformation
I=[[1,0],[0,1]].
For any vector v:
Iv=v.
Nothing changes.
Reflection in the x-axis
Matrix:
[[1,0],[0,−1]].
Point (3,2) becomes:
(3,−2).
The x-coordinate stays fixed.
The y-coordinate changes sign.
Reflection in the y-axis
Matrix:
[[-1,0],[0,1]].
Point (3,2) becomes:
(−3,2).
Rotation through 90° anticlockwise
Matrix:
[[0,−1],[1,0]].
Input (x,y) becomes:
(−y,x).
That is exactly the coordinate rule for a 90° anticlockwise rotation about the origin.
Rotation through 180°
Matrix:
[[-1,0],[0,−1]].
Input (x,y) becomes:
(−x,−y).
Enlargement about the origin
Scale factor k:
[[k,0],[0,k]].
Point (x,y) becomes:
(kx,ky).
The same scalar stretches both coordinate directions.
Why transformation order matters
Let R be 90° anticlockwise rotation:
R=[[0,−1],[1,0]].
Let S be reflection in the x-axis:
S=[[1,0],[0,−1]].
Start with v=[[1],[2]].
Rotate first:
Rv=[[-2],[1]].
Then reflect:
S(Rv)=[[-2],[-1]].
Now reverse the order.
Reflect first:
Sv=[[1],[-2]].
Then rotate:
R(Sv)=[[2],[1]].
The outputs differ.
Transformation matrices are read right-to-left when acting on column vectors: in BA v, A acts first and B acts second.
Composition creates one new matrix
If A acts first and B acts second:
v → Av → B(Av).
By associativity:
B(Av)=(BA)v.
So BA is the single matrix representing the combined transformation.
This is one of the deepest reasons ordinary matrix multiplication is designed the way it is.
Transform a whole shape point by point
Suppose a triangle has vertices:
- A(1,1);
- B(3,1);
- C(2,4).
Under reflection in the y-axis:
- A′(−1,1);
- B′(−3,1);
- C′(−2,4).
The same matrix acts on every vertex.
Because a linear transformation preserves the rule globally, we do not invent a new rule for each point.
Matrices can encode data as well as geometry
Suppose two shops sell three products.
Quantity matrix:
Q=[[10,5,8],[7,6,4]].
Rows represent shops.
Columns represent Products A, B and C.
Price column:
p=[[2],[3],[5]].
Then Qp is defined:
(2×3)(3×1)→2×1.
Shop 1 revenue:
10(2)+5(3)+8(5)=75.
Shop 2 revenue:
7(2)+6(3)+4(5)=52.
So:
Qp=[[75],[52]].
The inner dimension must have matching meaning
Q has three columns corresponding to Products A, B and C.
p has three rows corresponding to prices for A, B and C in the same order.
If the price vector were accidentally ordered C, A, B, the matrix multiplication would still be numerically defined but the model would be wrong.
Dimension compatibility is a mathematical requirement. Label compatibility is a modelling requirement. Good matrix work checks both.
Transpose can change row/column orientation
The transpose Aᵀ is formed by turning rows into columns.
If:
A=[[1,2,3],[4,5,6]],
then:
Aᵀ=[[1,4],[2,5],[3,6]].
A is 2×3.
Aᵀ is 3×2.
Transpose is useful when data orientation must be changed while keeping the same entries.
Do not transpose merely to force multiplication
Changing orientation changes the model’s meaning.
A transpose is valid only when rows and columns are intentionally being reinterpreted.
It should not be used as a mechanical trick to make inner dimensions match.
Transformation matrices preserve the origin
A pure 2×2 linear transformation sends:
[[0],[0]]
to:
[[0],[0]].
This is why translations are not represented by an ordinary 2×2 matrix acting on standard 2D coordinate vectors alone.
More advanced systems can incorporate translations using augmented or homogeneous coordinates.
Determinant as a later structural clue
For a 2×2 transformation matrix, the determinant carries information about area scaling and orientation.
This is a higher-level extension, but it reinforces the idea that a matrix is more than a calculation table: its structure encodes geometry.
Common misconception 1: transformation order can be reversed freely
In general, BA and AB represent different composed transformations.
Common misconception 2: row and column labels do not matter in data matrices
A numerically valid multiplication can still be a meaningless model if categories are misaligned.
Common misconception 3: a 90° rotation matrix can be applied in either row-vector or column-vector convention without change
The convention matters. This article uses column vectors with matrices acting on the left.
Common misconception 4: every geometric transformation is a 2×2 linear transformation
Translations require an extension beyond ordinary 2×2 matrices acting on standard coordinate vectors.
Common misconception 5: transpose is only cosmetic
Transpose changes matrix order and the row–column roles of entries.
A transformation-and-data diagnostic ladder
- Can the learner write a coordinate as a column vector?
- Can the learner apply a 2×2 matrix to a point?
- Can the learner recognise identity, reflections and simple rotations?
- Can the learner apply one matrix to every vertex of a shape?
- Can the learner explain why BA means A then B for column vectors?
- Can the learner show that reversing transformations changes the result?
- Can the learner label rows and columns in a data matrix?
- Can the learner align an inner dimension by meaning as well as size?
- Can the learner interpret the product in context?
- Can the learner distinguish transpose from arbitrary rearrangement?
How this fits Secondary Mathematics
Matrix applications connect algebra, coordinate geometry, transformations and structured data. The topic teaches a powerful general habit: mathematical operations are meaningful only when both their dimensions and their semantic labels line up correctly.
The deeper lesson: order encodes causality
In a matrix product, the right-hand transformation acts first.
That order records which state is handed to which operation next.
Row–column order matters because matrices do not merely store numbers. They store pathways: what feeds into what, which transformation happens first, which categories correspond, and how one structured system becomes the input to the next.
