When to reach for a rotation or a reflection in a geometry proof

Most people meet geometry transformations as a punchline. Someone shows you a proof that rotates the whole figure $60°$ about a vertex, three segments line up into a straight path, and the problem is over in two lines. It is beautiful and it is useless, because nothing in it tells you how anyone knew to rotate.

That is the actual gap. Rotation, reflection and homothety are easy to apply and nearly impossible to guess, so a student who understands all three still solves nothing with them. This page is about the guessing: what you look at in a figure to decide that a transformation is the move, and which one.

The one idea underneath all of them

A transformation proof works by making two things in the figure become the same thing. You do not prove $BD = CE$ by computing both. You find a motion that carries $B$ to $C$ and $D$ to $E$, and then equality is automatic, because the motion preserved every length it touched.

So the search is never "what transformation is elegant here". It is: which two pieces of this figure am I being asked to compare, and is there a rigid motion carrying one onto the other? If there is, the motion is your proof. If there is not, no amount of rotating will help and you should be doing something else.

Reflections and rotations are also not two separate tools. Compose two reflections in lines meeting at angle $\theta$ and you get exactly a rotation by $2\theta$ about their intersection. That is worth knowing because it tells you where to look for a rotation: at a point where two lines of symmetry cross.

Five signals, and what each one names

Each signal below is something visible in the statement or the figure, and each names a specific transformation with a specific centre. The centre is the whole game; an experienced solver does not decide "rotate", they decide "rotate about $A$".

• Two equal segments sharing an endpoint. $PA = PB$ is a rotation about $P$ waiting to happen, and the angle is $\angle APB$. This is the single most common trigger in olympiad geometry. When the shared endpoint is a vertex of an equilateral triangle the angle is $60°$, and the rotation carries one side of the triangle onto another.
• A line the whole figure respects. A perpendicular bisector, an angle bisector, an axis of an isosceles shape. Reflect in it and the figure maps to itself, which turns "these two things are equal" into "these two things are images of each other".
• A midpoint, with nothing else attached to it. A lone midpoint is a half-turn: rotation by $180°$ about that point. It sends every line to a parallel line, which is why it is the right response to a problem about parallelograms.
• A ratio, a centroid, or two parallel lines. Now you want a homothety rather than a rigid motion, because lengths are changing by a fixed factor. The centre is wherever the corresponding lines all meet.
• Two tangent circles, or a circle inscribed in another configuration. The point of tangency is a homothety centre, and the ratio is the ratio of the radii. This is the fastest way into a whole class of circle problems.

Watching a reflection do the work

The base angles of an isosceles triangle are equal is the smallest honest example, and the transformation proof is genuinely different from the usual one.

Signal two fires: the triangle has an axis, the perpendicular bisector of the base. Reflect in it. The apex is on the axis, so it is fixed; the two base vertices swap. The triangle maps to itself, with the two base angles exchanged. Angles are preserved by reflection, so they are equal.

Notice what did not happen. No auxiliary point was constructed, no congruence criterion was cited, no side lengths were compared. The symmetry was already in the figure and the proof just named it.

Watching a half-turn do the work

The diagonals of a parallelogram bisect each other runs the same way with signal three.

Let $M$ be the midpoint of one diagonal and take the half-turn about $M$. It swaps the endpoints of that diagonal by construction. Because a half-turn sends every line to a parallel line, it sends each side of the parallelogram to the side parallel to it, so it maps the whole parallelogram to itself. The other two vertices therefore swap as well, which says precisely that $M$ is the midpoint of the second diagonal too.

The move to steal: the half-turn was defined by the thing you already knew, and then forced to do something you did not.

When the answer is a homothety

Rigid motions preserve length. Once the problem involves a ratio, you need a transformation that scales, and then the two classic results in triangle geometry stop being surprising.

The orthocentre, centroid and circumcentre are collinear with $HG = 2\,GO$ is a homothety at the centroid with ratio $-\tfrac12$. That map sends each vertex to the midpoint of the opposite side, so it sends the triangle to its medial triangle, and it therefore sends each triangle centre to the corresponding centre of the medial triangle. The medial triangle's orthocentre is the original circumcentre. So $H \mapsto O$ under a map centred at $G$ with ratio $-\tfrac12$, and the collinearity and the $2:1$ ratio are both just what that sentence means.

The nine-point circle is the same homothety looked at once more. A circle through nine specified points is an alarming thing to prove from scratch; as the image of the circumcircle under a scaling of ratio $\tfrac12$, it is a circle of half the radius whose centre is the midpoint of $OH$, and the nine points fall out as images of points you already had.

For tangent circles, the ratio is handed to you: two internally tangent circles of radii $3$ and $8$ are related by the homothety at the tangency point with ratio $\tfrac83$, and any point-and-circle question about the pair becomes a question about a single circle.

The one advanced case worth knowing early

A spiral similarity is a rotation and a scaling about the same centre, and it is the transformation behind almost every "two similar triangles are hiding in this figure" problem. The reason it is worth naming early is that its centre has an address: if lines $AB$ and $CD$ meet at $X$, the spiral similarity taking $AB$ to $CD$ is centred at the second intersection point of the circumcircles of $XAC$ and $XBD$.

That is a recognition rule, not a technique. When a figure hands you two segments and two circles that cross twice, the second crossing is usually the centre you want.

The same family covers inversion, which is where Ptolemy's inequality comes from: inverting about one of the four points turns the inequality into the triangle inequality, and the equality case becomes "the three image points are collinear", which is exactly the cyclic condition. A hard statement about four points became an obvious statement about three.

How to actually train this

Reading transformation proofs teaches you nothing, for the same reason reading solutions never does. The skill is the recognition step, so train only that.

Take fifteen geometry problems you have already solved by other means. For each one, write two lines: which two pieces of the figure the problem compares, and what transformation would carry one to the other. Do not write the proof. Half the time there will be no such transformation, and noticing that quickly is as valuable as finding one, because it stops you burning ten minutes on an approach the figure does not support.

Then check yourself against the signal list. If your answer named a motion but not a centre, you have not finished the thought.

Lemma teaches this as the named technique transformations, sitting at the top of the geometry path rather than appearing as a trick in one solution, and the problem archive holds every problem above with difficulty and expected time attached. The daily problem posts a fresh one each day in three tiers with no account needed, which is the cheapest way to keep the recognition habit warm between study sessions.