$P_1, P_2, P_3, P_4$ are four arbitrary points on $xy = 1$. $P_1P_4$ intersects $P_2P_3$ at $D_1$, and similarly define $D_2, D_3$. Prove that $O, D_1, D_2, D_3$ are cyclical, where $O$ is the origin.

I have never seen a geometry problem asking to prove cyclical points on a conic section. So frankly not sure where to start. I suspect bashing coordinates would work, but that's quite a bit of efforts (and I am not sure how to go from coordinates to proving points are cyclical).

Is there some shortcut I could use by some geometric properties of $xy=1$?

enter image description here

  • $\begingroup$ With the help of Mathematica, I verified the result via coordinate bashing; there must be a better way. That said, if the points are $A=(a,a')$, $B=(b,b')$, $C=(c,c')$, $D=(d,d')$, with $aa'=bb'=cc'=dd'=1$, then $$\overleftrightarrow{AB}\cap \overleftrightarrow{CD}=\left(\frac{(a'+b')-(c'+d')}{a'b'-c'd'},\frac{(a+b)-(c+d)}{ab-cd}\right)$$ Similarly for $\overleftrightarrow{AC}\cap\overleftrightarrow{BD}$ and $\overleftrightarrow{AD}\cap\overleftrightarrow{BC}$. So, there's some interesting coordinate structure, but it doesn't seem to make proving cyclicity particularly easy. $\endgroup$
    – Blue
    Dec 22 '19 at 5:45
  • $\begingroup$ @Blue that is indeed interesting. I wonder if it makes some complex number bashing easier. $\endgroup$
    – Chen Chen
    Dec 22 '19 at 14:06
  • $\begingroup$ Note that, by construction, $D_1,D_2,D_3$ form a self-polar triangle with respect to the given conic section, i.e. the pole $D_1$ has polar $D_2D_3$ etc. In a triangle coordinate system (e.g. barycentric or trilinear) based on the self-polar $D_1D_2D_3$, the symmetric $3\times3$ coefficient matrix of that conic section is diagonal. $\endgroup$
    – ccorn
    Dec 22 '19 at 17:18
  • $\begingroup$ @ccorn Mind to elaborate a bit or point me to more materials? sorry not even sure what "$D1$ has polar $D_2 D_3$" means.. $\endgroup$
    – Chen Chen
    Dec 22 '19 at 19:49

Here's a proof without coordinates. I will relabel $D_1=P_1P_4\cap P_2P_3$, $D_2=P_2P_4\cap P_1P_3$ and $D_3=P_3P_4\cap P_1P_2$.

Lemma 1. The centre of any rectangular hyperbola $\mathcal H$ through points $A$, $B$, $C$ lies on the nine-point circle of $\triangle ABC$.

Proof. Let $(ABC)$ meet $\mathcal H$ again at $D$, and let $H_A$, $H_B$, $H_C$, $H_D$ be the orthocentres of $\triangle BCD$, $\triangle CDA$, $\triangle DAB$, $\triangle ABC$, which all lie on $\mathcal H$. There is a $180^{\circ}$ rotation mapping $ABCD$ to $H_AH_BH_CH_D$, and the centre of this rotation is both the centre of $\mathcal H$ and the midpoint of $\overline{DH_D}$, so it lies on the nine-point circle of $\triangle ABC$. $\square$

Lemma 2. The incentre $I$ and excentres $I_1, I_2, I_3$ (opposite $D_1$, $D_2$, $D_3$) of $\triangle D_1D_2D_3$ lie on the hyperbola.

Proof. Take a projective transformation sending $P_1P_2P_3P_4$ to a square. Then $D_1$ and $D_3$ go to infinity, so $I_1I_2I_3I$ becomes a rectangle with sides parallel to $P_1P_2P_3P_4$. Then $D_2$ maps to the common centre of rectangle $I_1I_2I_3I$ and square $P_1P_2P_3P_4$, so these eight points lie on a common conic. $\square$

Now $(D_1D_2D_3)$ is the nine-point circle of $\triangle I_1I_2I_3$, so our two lemmas imply the result.


The following is definitely not a shortcut, but gives a self-contained reasoning (with the help of computer algebra system). Note that rectangular hyperbolas (the one with axes perpendicular to each other, or of eccentricity $\sqrt{2}$) are well studied due to its applications in other areas such as the study of centers of a given triangle. Examples include Feuerbach, Jerabek, and Kiepert hyperbolas, among others. The argument below is based on coordinates and all the computation is straightforward. A byproduct is that this also proves the Feuerbach's Theorem as a corollary. Note that since all rectangular hyperbolas are similar (being of the same eccentricity), we can use the one with equation $xy=1$ as our model without loss of generality.

Theorem. Let $P_i,i=1,\cdots,4$, be four points on a rectangular hyperbola. Let $A_i=\{P_1,P_{1+i}\},B_i=\{P_1,\cdots,P_4\}\setminus A_i,i=1,2,3$. Denote by $\ell(A_i)$ (resp. $\ell(B_i)$) the line joining the two points in $A_i$ (resp. $B_i$). Let $D_i:=\ell(A_i)\cap\ell(B_i)$. The the four points $D_1,D_2,D_3$ and the center $O$ of the rectangular hyperbola are concyclic.

Proof. As mentioned above, we use $xy=1$ as our model. Let $P_i=(t_i,1/t_i), i=1,\cdots 4.$ Then solving linear equations yields the following coordinates for $D_i=(x_i,y_i),i=1,2,3,$ where $$x_1=\frac{t_1t_2t_3+t_1t_2t_4-t_1t_3t_4-t_2t_3t_4}{t_1t_2-t_3t_4},\qquad y_1=\frac{t_1+t_2-t_3-t_4}{t_1t_2-t_3t_4}$$ $$x_2=\frac{t_1t_2t_3+t_1t_3t_4-t_1t_2t_4-t_2t_3t_4}{t_1t_3-t_2t_4},\qquad y_1=\frac{t_1+t_3-t_2-t_4}{t_1t_3-t_2t_4}$$ and $$x_3=\frac{t_1t_2t_4+t_1t_3t_4-t_1t_2t_3-t_2t_3t_4}{t_1t_4-t_2t_3},\qquad y_1=\frac{t_1+t_4-t_2-t_3}{t_1t_4-t_2t_3}.$$ Now to prove that four points ($D_1,D_2,D_3,D_0$ with $D_0=(x_0,y_0)$ ) are concyclic, it suffices to check the vanishing of the following determinant (this can be seen by working with an equation for a circle), namely $$\left|\begin{array}{cccc}x_1^2+y_1^2&x_1&y_1&1\\ x_2^2+y_2^2&x_2&y_2&1\\ x_3^2+y_3^2&x_3&y_3&1\\ x_0^2+y_0^2&x_0&y_0&1\end{array}\right|=0,$$ where since $D_0=O=(0,0)$ (the center of the hyperbola), the result is reduced to showing $$\left|\begin{array}{cc}x_1^2+y_1^2&x_1&y_1\\ x_2^2+y_2^2&x_2&y_2\\ x_3^2+y_3^2&x_3&y_3\end{array}\right|=0,$$ which is true as can be checked directly by a computer algebra system (for example, in SAGE, use the factor command: it returns that a zero cannot be factored). QED

Lemma. For any triangle $P_1P_2P_3$ inscribed in a rectangular hyperbola $\mathcal{C}$, the orthocenter $H$ of $P_1P_2P_3$ lies on $\mathcal{C}$.

Proof. Without loss of generality, use the model $xy=1$ and the same parametrizations of $P_i, (i=1,2,3)$ as above. Then by direct computation, the orthocenter is given by $$H=(-1/(t_1t_2t_3),-t_1t_2t_3),$$ which indeed lies on $\mathcal{C}.$ QED

Corollary. (Feuerbach's Theorem) For a triangle $ABC$ inscribed in a rectangular hyperbola $\mathcal{C}$, its nine-point circle passes through the center of the hyperbola.

Proof. Given an inscribed triangle $ABC$ in $\mathcal{C}$, by the above lemma, the orthocenter $H$ of $ABC$ lies on $\mathcal{C}$. Now let $A,B,C,H$ be the four points $P_i$'s as in the Theorem. It is clear that the three points $D_i$'s correspond to the three feet of altitude from $C,B$ and $A$, therefore they lie on the nine-point circle of $ABC$. Since by the Theorem, this circle passes through $O$, the result is clear. QED

Remarks. The following gives some references without details (please look up undefined terms).

  1. In "Projective Geometry" by H.S.M. Coxeter, the author says "If $4$ points in a plane are joined in pairs by $6$ distinct lines, they are called the vertices of a complete quadrangle, and the lines are its $6$ sides." And a result on polarity induced by a conic is the following (See 8.21 of the same reference): If a quadrangle is inscribed in a conic, its diagonal triangle is self-polar.

  2. In "Introduction to Plane Geometry" by H.F. Baker (1943), it was mentioned in Ex. 12 on page 158 that "..., if $PQR$ be a self-polar triangle in regard to the rectangular hyperbola, the circumcircle of $PQR$ contains the center $C$" (of the hyperbola). The proof was given (but not mentioned here because it depends on other results).


Here is an outline of a proof that still uses coordinates, but it is high-level enough to make the core of the proof an easy rearrangement without lengthy calculation.

For brevity, I rename the triangle $D_1D_2D_3$ to $ABC$. Its side lengths are denoted $a,b,c$. The rectangular hyperbola is named $K$ here.

I use barycentric coordinates with reference triangle $ABC$. I also presume a projective geometry context; that is, I will refer to points at infinity and the line at infinity. For the latter, I use the symbol $L_\infty$ and remark that in barycentric coordinates,

  • $L_\infty$ has coefficient tuple $[1:1:1]$;
  • the incenter of $ABC$ is at $(a:b:c)$; flipping the sign of one of these coordinates yields an excenter;
  • the points $(u:v:w)$ on the circumcircle of $ABC$ are those fulfilling $a^2 v w + b^2 w u + c^2 u v = 0$.

Notation: I use round parentheses for point coordinates and square brackets for coefficients of lines and curves. Both kinds are occasionally represented in matrix form without the $:$ separator; in such cases, I use $\cong$ to indicate linear dependence between two such matrices.

Some of the reasoning steps below may be regarded as reminders or observations left as exercises.

  1. Given $P_1=(u_1:v_1:w_1)$, then (a permutation of) $P_2,P_3,P_4$ can be given as $P_2=(-u_1:v_1:w_1)$, $P_3=(u_1:-v_1:w_1)$, $P_4=(u_1:v_1:-w_1)$. This has nothing to do with $K$; it follows from the construction of $ABC$.

  2. In barycentric coordinates, $K$ has its homogeneous equation in diagonal form $K(u,v,w) = Ru^2 + Sv^2 + Tw^2 = 0$ with nonzero coefficients $R,S,T$. (Nonzero because $K$ is not degenerate.) This diagonality follows from item (1) and $K(P_i) = 0$ or alternatively from the fact that the reference triangle $ABC$ is, by construction, self-polar with respect to $K$; thus the pole $A=(1:0:0)$ has associated polar $\overline{BC}=[1:0:0]$ etc.

  3. The center $O=(u_0:v_0:w_0)$ of $K$ is the pole of the line at infinity, thus:

    $$\begin{pmatrix}u_0\\v_0\\w_0\end{pmatrix} \cong \begin{bmatrix} R & 0 & 0 \\ 0 & S & 0 \\ 0 & 0 & T \end{bmatrix}^{-1} \begin{bmatrix}1\\1\\1\end{bmatrix} = \begin{pmatrix}R^{-1}\\S^{-1}\\T^{-1}\end{pmatrix}$$

  4. As a hyperbola, $K$ has two distinct points at infinity, call them $E_1=(\hat{u}_1:\hat{v}_1:\hat{w}_1)$ and $E_2=(\hat{u}_2:\hat{v}_2:\hat{w}_2)$. Thus $\overline{E_1E_2}=L_\infty$, which in barycentric coordinates means

    $$\begin{pmatrix}\hat{u}_1\\\hat{v}_1\\\hat{w}_1\end{pmatrix} \times\begin{pmatrix}\hat{u}_2\\\hat{v}_2\\\hat{w}_2\end{pmatrix} = \begin{bmatrix}\hat{v}_1\hat{w}_2-\hat{w}_1\hat{v}_2 \\\hat{w}_1\hat{u}_2-\hat{u}_1\hat{w}_2 \\\hat{u}_1\hat{v}_2-\hat{v}_1\hat{u}_2\end{bmatrix} \cong \begin{bmatrix}1\\1\\1\end{bmatrix}$$

  5. Since both $E_i$ lie on $K$, we must have

    $$\begin{aligned} \begin{bmatrix}R\\S\\T\end{bmatrix} &\cong \begin{pmatrix}\hat{u}_1^2\\\hat{v}_1^2\\\hat{w}_1^2\end{pmatrix} \times\begin{pmatrix}\hat{u}_2^2\\\hat{v}_2^2\\\hat{w}_2^2\end{pmatrix} = \begin{bmatrix}\hat{v}_1^2\hat{w}_2^2-\hat{w}_1^2\hat{v}_2^2 \\\hat{w}_1^2\hat{u}_2^2-\hat{u}_1^2\hat{w}_2^2 \\\hat{u}_1^2\hat{v}_2^2-\hat{v}_1^2\hat{u}_2^2\end{bmatrix} \\&= \begin{bmatrix}(\hat{v}_1\hat{w}_2-\hat{w}_1\hat{v}_2)(\hat{v}_1\hat{w}_2+\hat{w}_1\hat{v}_2) \\(\hat{w}_1\hat{u}_2-\hat{u}_1\hat{w}_2)(\hat{w}_1\hat{u}_2+\hat{u}_1\hat{w}_2) \\(\hat{u}_1\hat{v}_2-\hat{v}_1\hat{u}_2)(\hat{u}_1\hat{v}_2+\hat{v}_1\hat{u}_2)\end{bmatrix} \\&\cong \begin{bmatrix}\hat{v}_1\hat{w}_2+\hat{w}_1\hat{v}_2 \\\hat{w}_1\hat{u}_2+\hat{u}_1\hat{w}_2 \\\hat{u}_1\hat{v}_2+\hat{v}_1\hat{u}_2\end{bmatrix} \text{ using item (4)} \end{aligned}$$

  6. Since points at infinity represent slopes of lines, it makes sense to ask whether two given points at infinity are orthogonal. This can be answered if the equation of a circle is given: $E_1,E_2$ (at infinity) are orthogonal if and only if $E_1$ lies on the polar of $E_2$ with respect to the circle. Concretely, given the symmetric bilinear form $U$ of any circle, then $E_1,E_2$ are orthogonal if and only if $U(E_1,E_2) = 0$.

  7. $K$ is a rectangular hyperbola, therefore $E_1,E_2$ must be orthogonal. Using item (6) with the symmetric bilinear form of the circumcircle $U$ of the reference triangle, we have

    $$\begin{align} 0 &= U(E_1,E_2) = \begin{pmatrix}\hat{u}_1\\\hat{v}_1\\\hat{w}_1\end{pmatrix}^\top \begin{bmatrix} 0 & c^2 & b^2 \\ c^2 & 0 & a^2 \\ b^2 & a^2 & 0 \end{bmatrix} \begin{pmatrix}\hat{u}_2\\\hat{v}_2\\\hat{w}_2\end{pmatrix} \\ &= \begin{pmatrix}a\\b\\c\end{pmatrix}^\top \begin{bmatrix} \hat{v}_1\hat{w}_2+\hat{w}_1\hat{v}_2\hspace{-1em} & 0 & 0\quad \\ \quad 0 & \hspace{-1em}\hat{w}_1\hat{u}_2+\hat{u}_1\hat{w}_2\hspace{-1em} & 0\quad \\ \quad 0 & 0 & \hspace{-1em}\hat{u}_1\hat{v}_2+\hat{v}_1\hat{u}_2 \end{bmatrix} \begin{pmatrix}a\\b\\c\end{pmatrix} \\ &\cong \begin{pmatrix}a\\b\\c\end{pmatrix}^\top \begin{bmatrix} R & 0 & 0 \\ 0 & S & 0 \\ 0 & 0 & T \end{bmatrix} \begin{pmatrix}a\\b\\c\end{pmatrix} \tag{I} \\ &= \frac{RST}{2} \begin{pmatrix}R^{-1}\\S^{-1}\\T^{-1}\end{pmatrix}^\top \begin{bmatrix} 0 & c^2 & b^2 \\ c^2 & 0 & a^2 \\ b^2 & a^2 & 0 \end{bmatrix} \begin{pmatrix}R^{-1}\\S^{-1}\\T^{-1}\end{pmatrix} \tag{II} \end{align}$$

Now (II) tells us that the center $O$ of $K$ lies on the circumcircle of $ABC$. Furthermore, (I) tells us that $K$ contains the incenter $(a:b:c)$ of $ABC$, and we can generalize this to $K$ containing $(\pm a:\pm b:\pm c)$ which are the in- and excenters of $ABC$. (This is fitting: The in- and excenters form an orthocentric system.)


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.