The phantom point technique: how to prove a point you cannot reach

Some geometry problems stall for a reason that has nothing to do with how much geometry you know. The point the problem is about is defined in a way you cannot use. It is "where line $\ell$ meets circle $\omega$", or "the intersection of these two cevians", and every angle you chase from it leads nowhere, because the definition gives you no angles to chase.

The phantom point technique is the way out. You stop trying to reach the point you were given, build a different point that you can describe, and then prove the two are the same point. It sounds like a trick. It is really a change of which direction you prove things in, and once you see it you will notice it inside half the classical proofs you already know.

The method in three moves

1. Guess the answer and name it. Draw an accurate figure and decide what the mystery point really is: a midpoint, an arc midpoint, a point on a known circle. Call that candidate $P'$ and define it by the property you want, not by the construction you were given.
2. Prove the phantom satisfies the original construction. If the problem's point $P$ is "where line $\ell$ meets line $m$", show that $P'$ lies on $\ell$ and on $m$. This is usually the easy direction, because $P'$ was defined by something concrete.
3. Argue uniqueness. The construction pins down exactly one point, and $P'$ satisfies it, so $P = P'$. Everything true of $P'$ is now true of $P$.

The whole value is in step 2 running backwards. Proving "the intersection has property X" is hard; proving "the point with property X lies on both lines" is often two lines of angle chasing.

Watching it work

Here is the example to learn it from. In triangle $ABC$ with $AB \ne AC$, the bisector of angle $A$ and the perpendicular bisector of $BC$ meet at a point. Show that this point lies on the circumcircle.

Chasing forward from the intersection gets you nothing: you know two lines, and the circle is not in either definition. So guess. On an accurate figure the intersection sits on the circle, exactly halfway round the arc $BC$ that does not contain $A$. Call that arc midpoint $M$. That is the phantom.

Now run it backwards. $M$ is the midpoint of arc $BC$, so the chords $MB$ and $MC$ are equal, which puts $M$ on the perpendicular bisector of $BC$. The arcs $BM$ and $MC$ are equal, so the inscribed angles standing on them are equal: $\angle BAM = \angle MAC$, and $AM$ is the angle bisector. So $M$ lies on both lines.

Last move: two distinct lines share at most one point, and $AB \ne AC$ keeps these two distinct, so the intersection is $M$, and $M$ is on the circle by definition. Done, and nothing in it was harder than inscribed angles.

The step people drop

Step 3 is where solutions lose marks, because it feels too obvious to write. It is not a formality. It is the one place the argument can actually break.

Look at the condition $AB \ne AC$ in the example. If $AB = AC$, the angle bisector from $A$ is the perpendicular bisector of $BC$: the two lines coincide, they share every point, and "the point where they meet" does not exist. The phantom argument still shows $M$ is on both lines, but so is $A$, and so is every other point on that line. Uniqueness is exactly what fails.

So when you write step 3, name the reason the construction determines a single point:

• Two lines meet in one point if they are not parallel and not the same line. Say why yours are distinct.
• A line and a circle meet in at most two points. Say which of the two you mean, usually "the second intersection, other than $A$".
• Two circles meet in at most two points. Same care.

If you cannot finish the sentence "the construction determines a unique point because …", your proof has a hole, however clean the rest of it is.

You have already used it

The reason the technique feels unfamiliar is that textbooks rarely name it. Three proofs in the problem archive are phantom-point arguments in disguise.

The perpendicular bisectors of a triangle are concurrent. You let $O$ be the point where two of them meet, then prove $O$ is on the third, using that the third bisector is exactly the set of points equidistant from $A$ and $C$. That is the phantom shape: the third line is described by a property, and you show your point has it.

The medians are concurrent and cut each other $2:1$. The vector proof does not look for the intersection at all. It writes down $G = \tfrac13(A + B + C)$, a point defined by a formula, then checks that $G$ sits on each median two thirds of the way from the vertex. The point was built first and placed afterwards.

The Euler line is the sharpest version. Put the circumcentre at the origin and define $H' = A + B + C$. Because $|B| = |C|$, the vector $H' - A = B + C$ is perpendicular to $C - B$, so $H'$ lies on the altitude from $A$, and likewise on the other two. Altitudes of a triangle meet in one point, so $H'$ is the orthocentre. Then $G = \tfrac13 H'$ lies on the segment from $O$ to $H$ with $HG = 2\,GO$ for free.

When to reach for it

Signals that a phantom point is the move:

• The problem's point is an intersection, and the conclusion is about a different kind of object. Two lines meet, and you must show the point is on a circle. Two circles meet, and you must show a length. The mismatch is the tell.
• Your figure suggests a clean identity for the point. If it looks like an arc midpoint, a reflection, or the foot of a perpendicular, define that and test it.
• A forward chase keeps producing angles you cannot relate. That is usually because the point's definition carries no angle information, and the phantom's does.
• Concurrency and collinearity questions. "Three lines meet at a point" is, almost by definition, "the point where two of them meet lies on the third".

The discipline is to draw the figure accurately before guessing. A phantom built from a sloppy sketch is a confident proof of something false.

Where it sits

Lemma teaches this as a named technique in its geometry ladder: the phantom points page has the statement and where it is used, alongside the other methods on the techniques index. The problem archive holds the three proofs above in full with hints and marking points. And the daily problem posts a fresh one every day, no account needed, which is the cheapest way to meet a stalled configuration and practise building the point you need.