I’m working on plane transformations and I’m trying to compress a sequence into a single transformation. The steps I have are:
1) reflect across the line y = 1,
2) reflect across the line y = x,
3) translate by the vector (-2, 1).
To test it, I used a couple of points. For P(4, -1) I get P’ = (1, 5). For Q(2, 3) I get Q’ = (-3, 3). I also checked R(0, 0) and get R’ = (0, 1). From plotting these, it looks orientation-preserving, so I guessed it might be a rotation. Eyeballing P → P’ and Q → Q’ makes me think the angle might be 90°, but I’m not fully certain about the direction.
My attempt: I tried to find the center by intersecting perpendicular bisectors of PP’ and QQ’. For PP’, midpoint is (2.5, 2) and I took the perpendicular slope as 1/2, so that line is y – 2 = (1/2)(x – 2.5). For QQ’, midpoint is (-0.5, 3) and the perpendicular bisector looks vertical, so x = -0.5. Intersecting these gives a center candidate c ≈ (-0.5, 0.5). When I check the angle using vectors from c to P and P’, I get tangled about whether it’s 90° clockwise or counterclockwise (sometimes I end up calling it 270° by accident). Also, I’m not sure if I’m overfitting to these sample points instead of deriving the result directly from the three given steps.
Questions:
– Is there a dependable way to tell when “reflection then reflection then translation” reduces to a rotation versus a glide reflection?
– How do I compute the exact center and angle directly from the given steps (possibly with matrices), instead of inferring from sample points?
– Finally, am I applying the operations in the correct order (left to right as written), or should they be read right to left?
Any help appreciated!
















3 Responses
Crunch it: (x,y) → (x,2−y) → (2−y,x) → (-y, x+1), so the net map is T(x,y)=(-y, x+1) = a 90° counterclockwise rotation about c = (−1/2, 1/2) (solve (I−A)c=b with A=[[0,−1],[1,0]], b=(0,1)); two reflections preserve orientation, so unless the linear part is the identity it’s a rotation, not a glide.
Apply the steps in the written order (left→right); matrices multiply right→left-want to verify by checking T(c)=c and that vectors from c are turned 90°?
You’re right to think “orientation-preserving” narrows it down. Two reflections are orientation-preserving, and adding a translation keeps it that way, so the result can only be a rotation or a translation (not a glide reflection – those come from an odd number of reflections). More specifically: reflecting in two intersecting lines is a rotation by twice the angle between them about their intersection point; reflecting in two parallel lines is a translation. Here the lines y = 1 and y = x intersect at (1, 1) with a 45° angle, so the first two steps alone give a 90° rotation about (1, 1). I always have to squint a bit about the direction, but plotting your test points suggests it’s the “positive” (counterclockwise) one.
If you want it cleanly from the steps, do the affine algebra. Reflect across y = 1: (x, y) -> (x, 2 − y). Then reflect across y = x: (x, y) -> (y, x). Then translate by (−2, 1): (x, y) -> (x − 2, y + 1). Composing in that order gives F(x, y) = (−y, x + 1). That’s exactly a 90° counterclockwise rotation matrix applied to (x, y), followed by an upward shift by 1. A rotation about some center c has the form F(p) = R90 p + t with center c solving (I − R90) c = t. Solving gives c = (−1/2, 1/2). So the whole thing is a rotation by 90° counterclockwise about (−1/2, 1/2). Funny aside: I first wanted to just “move the old center” (1, 1) by (−2, 1) to get (−1, 2), but that’s not quite right because the translation happens after the rotation – the algebra pins it down to (−1/2, 1/2), which matches your perpendicular-bisector find. For order: apply them left to right as written; if you write matrices, the combined matrix multiplies on the left, acting on column vectors on the right. A quick refresher on composing transformations (including reflections and rotations) is here: https://www.khanacademy.org/math/linear-algebra/matrix-transformations/lin-trans-examples/a/linear-transformations-examples
Dependable rule of thumb:
– Two reflections in intersecting lines -> rotation (by twice the angle) about their intersection; adding a translation keeps it a rotation (unless that twice-angle were 0°, in which case you’d just get a translation).
– Two reflections in parallel lines -> translation; adding another translation stays a translation.
Neat pattern: the composition is F(x,y)=(-y, x+1) (since (x,y)→(x,2−y)→(2−y,x)→(-y,x+1)), which is a 90° counterclockwise rotation about c = (-1/2, 1/2); in general, two reflections in intersecting lines give a rotation (parallel → translation), and adding a translation to a rotation just shifts its center, while you apply steps to points left-to-right (matrices multiply right-to-left). See a quick overview of plane isometries at https://en.wikipedia.org/wiki/Euclidean_plane_isometry – hope this helps!