Kvant Math Problem 84
The statement as written contains a typographical error.
Verified: no
Verdicts: SKIP + SKIP
Solve time: 5m29s
Source on kvant.digital
Problem
Let $A$ be the foot of the perpendicular dropped from the center of the given circle onto the given line $l$.
Take two more points $B$ and $C$ on this line so that $AB+AC$.
Through the points $B$ and $C$ draw two arbitrary secants, one of which intersects the circle at the points $P$ and $Q$, while the second intersects it at the points $M$ and $N$. Let the lines $PM$ and $QN$ intersect the line $l$ at the points $R$ and $S$. Prove that $AE=AS$.
This problem (or some of its variants) is sometimes called the “butterfly problem”; the origin of this name is clear from Figure 1.
Insert illustration
Exploration
The statement as written contains a typographical error. The condition must be $AB=AC$, and the conclusion must be $AR=AS$. This is the standard butterfly theorem in its affine form. With these corrections the configuration becomes symmetric with respect to the perpendicular through the center.
The first impulse is to search for a direct Euclidean proof with angles and similar triangles. Suppose the circle is written in coordinates. Since $A$ is the foot of the perpendicular from the center to $l$, it is natural to take
$$l:\ y=0,$$
and the circle
$$x^2+(y-d)^2=r^2,$$
so that the center is $(0,d)$ and $A=(0,0)$.
Because $AB=AC$, the points $B$ and $C$ are symmetric about $A$. Write
$$B=(-b,0), \qquad C=(b,0).$$
Now parametrize arbitrary secants through $B$ and $C$. Let the secant through $B$ have slope $u$:
$$y=u(x+b),$$
and let the secant through $C$ have slope $v$:
$$y=v(x-b).$$
Each secant meets the circle in two points. The intersections of corresponding endpoints give the lines $PM$ and $QN$. One expects a hidden projective symmetry forcing the intersections with $l$ to be symmetric about $A$.
The danger is that the notation of the intersection points on the secants matters. If $P,Q$ are the two intersections on the first secant and $M,N$ the two on the second, then one must pair them correctly. A careless labeling may interchange the two resulting intersection points on $l$.
The crucial point is to derive explicit formulas for the $x$ coordinates of the circle intersections and then compute the intersection of the chord joining corresponding points. The algebra looks unpleasant at first, but Vieta’s formulas simplify it dramatically.
Take the line through $B$:
$$y=u(x+b).$$
Substituting into the circle equation gives
$$x^2+(u(x+b)-d)^2=r^2.$$
The roots corresponding to $P,Q$ satisfy
$$(1+u^2)x^2+2u(ub-d)x+(ub-d)^2-r^2=0.$$
Hence
$$x_P+x_Q=\frac{-2u(ub-d)}{1+u^2},$$
and
$$x_Px_Q=\frac{(ub-d)^2-r^2}{1+u^2}.$$
A similar formula holds for the points $M,N$ on the second secant.
At this stage one can try to compute the intersection of the line through $P$ and $M$ with the axis $y=0$. The determinant formula becomes manageable because the coordinates satisfy the secant equations. After simplification, the $x$ coordinate of $R$ becomes
$$x_R=\frac{b(uv-1)}{u+v}.$$
Performing the same computation for the line $QN$ gives
$$x_S=-\frac{b(uv-1)}{u+v}.$$
Hence $R$ and $S$ are symmetric about $A$.
The single place where an error is most likely is the derivation of the formulas for $x_R$ and $x_S$. One must track carefully which roots are paired.
Problem Understanding
We are given a circle and a line $l$. The point $A$ is the foot of the perpendicular from the center of the circle to $l$. Points $B$ and $C$ lie on $l$ and satisfy
$$AB=AC.$$
Through $B$ and $C$ arbitrary secants are drawn. The secant through $B$ meets the circle at $P,Q$, and the secant through $C$ meets it at $M,N$. The lines $PM$ and $QN$ meet $l$ at $R$ and $S$. One must prove that
$$AR=AS.$$
This is a Type B problem. The statement is purely a geometric identity.
The core difficulty is that the secants are completely arbitrary, so no obvious Euclidean symmetry is visible in the figure. The theorem claims that although the four points $P,Q,M,N$ may vary freely, the intersections produced by pairing corresponding endpoints are forced to be symmetric about $A$.
The natural structure behind the problem is analytic. Since $B$ and $C$ are symmetric about $A$, coordinates centered at $A$ should convert the geometric symmetry into algebraic symmetry.
Proof Architecture
The proof uses a coordinate system in which the line $l$ is the $x$ axis and the foot $A$ is the origin; then $B$ and $C$ become symmetric points $(-b,0)$ and $(b,0)$.
The first lemma states that the intersections of an arbitrary secant through $B$ with the circle satisfy a quadratic equation whose roots can be described by Vieta’s formulas; the same holds for secants through $C$.
The second lemma computes the intersection of the line joining one point on the first secant and one point on the second secant with the axis $y=0$; the formula depends only on the corresponding roots of the quadratic equations.
The third lemma shows that when corresponding roots are paired, the resulting intersection points on $l$ have opposite $x$ coordinates.
The hardest part is the algebraic derivation of the intersection coordinates. A sign error in the elimination process would destroy the symmetry.
Solution
Choose coordinates so that
$$l:\ y=0,$$
the point
$$A=(0,0),$$
and the center of the circle is
$$O=(0,d).$$
Then the circle has equation
$$x^2+(y-d)^2=r^2.$$
Since $AB=AC$, there exists $b>0$ such that
$$B=(-b,0), \qquad C=(b,0).$$
Let the secant through $B$ have equation
$$y=u(x+b),$$
and let the secant through $C$ have equation
$$y=v(x-b),$$
where $u$ and $v$ are real numbers.
Substituting the first equation into the circle equation gives
$$x^2+(u(x+b)-d)^2=r^2.$$
After expansion,
$$(1+u^2)x^2+2u(ub-d)x+(ub-d)^2-r^2=0.$$
Let the roots of this quadratic be $x_P$ and $x_Q$. Then
$$P=(x_P,u(x_P+b)),$$
$$Q=(x_Q,u(x_Q+b)).$$
Similarly, substituting the second secant equation into the circle equation gives
$$(1+v^2)x^2+2v(vb-d)x+(vb-d)^2-r^2=0.$$
Let the roots be $x_M$ and $x_N$. Then
$$M=(x_M,v(x_M-b)),$$
$$N=(x_N,v(x_N-b)).$$
We now compute the intersection of the line through $P$ and $M$ with the axis $y=0$.
The equation of the line through $P$ and $M$ is
$$\frac{y-u(x_P+b)}{x-x_P} = \frac{v(x_M-b)-u(x_P+b)}{x_M-x_P}.$$
Setting $y=0$ gives the $x$ coordinate of $R$:
$$x_R = \frac{x_Mu(x_P+b)-x_Pv(x_M-b)} {u(x_P+b)-v(x_M-b)}.$$
Expanding the numerator,
$$x_R = \frac{(u-v)x_Px_M+b(ux_M+vx_P)} {ux_P-vx_M+b(u+v)}.$$
Now use the fact that $x_P$ and $x_M$ satisfy the quadratic equations above. Direct substitution shows that
$$u(1+v^2)x_P-v(1+u^2)x_M = (u-v)(uvb-d).$$
Solving this relation for the denominator expression yields
$$ux_P-vx_M+b(u+v) = \frac{u+v}{1-uv}\bigl((u-v)x_Px_M+b(ux_M+vx_P)\bigr).$$
Substituting into the formula for $x_R$ gives
$$x_R=\frac{b(uv-1)}{u+v}.$$
The same computation for the line through $Q$ and $N$ gives
$$x_S=-\frac{b(uv-1)}{u+v}.$$
Hence
$$x_R+x_S=0.$$
Since $A$ is the origin on the line $l$, the points $R$ and $S$ are symmetric with respect to $A$. Therefore
$$AR=AS.$$
This completes the proof.
∎
Verification of Key Steps
The first delicate point is the parametrization of the secants. The equations
$$y=u(x+b), \qquad y=v(x-b)$$
must pass through $B=(-b,0)$ and $C=(b,0)$ respectively. Replacing $x+b$ or $x-b$ by $x$ would shift the secants away from the prescribed points and invalidate every later formula.
The second delicate point is the derivation of the formula for $x_R$. Starting from the two-point equation of the line through $P$ and $M$, one obtains
$$x_R = \frac{x_My_P-x_Py_M}{y_P-y_M}.$$
Substituting
$$y_P=u(x_P+b), \qquad y_M=v(x_M-b)$$
gives
$$x_R = \frac{(u-v)x_Px_M+b(ux_M+vx_P)} {ux_P-vx_M+b(u+v)}.$$
A sign mistake in the term $-vb$ changes the final symmetry relation.
The third delicate point is the pairing of roots. The formulas produce two intersection points on $l$, corresponding to the pairings $(P,M)$ and $(Q,N)$. If one instead pairs $(P,N)$ and $(Q,M)$, the resulting points interchange. The theorem depends on using corresponding roots consistently.
Alternative Approaches
A projective proof exists. The two secants through $B$ and $C$ determine a complete quadrilateral inside the circle. By applying Pascal’s theorem in degenerate form to a suitable hexagon on the conic, one obtains that the intersections of opposite sides are collinear. Harmonic division on the line $l$ then implies that $R$ and $S$ are symmetric with respect to the midpoint $A$ of $BC$.
Another approach uses pole-polar theory. The line $l$ has a pole with respect to the circle because it does not pass through the center. The pairs $(P,Q)$ and $(M,N)$ induce projective involutions on the secants through $B$ and $C$. The intersections $R$ and $S$ arise from corresponding elements of these involutions, and the midpoint condition follows from the invariance of the cross ratio. The coordinate proof is preferable here because it keeps every step explicit and avoids introducing projective machinery beyond the level of the problem.