We study a variant of Napoleon’s construction on the unit sphere, in which the triangles erected on the sides of a spherical triangle ABC are required to have similar isosceles chord triangles with apex angle \beta\in(0,\pi/2). For a non-equilateral ABC, we prove that the triangle XYZ of spherical circumcentres is equilateral if and only if \cos(\theta_a+\theta_b+\theta_c\pm\Omega/2)=0, where \theta_a,\theta_b,\theta_c are the base angles of the isosceles spherical triangles XBC, YCA, ZAB, \Omega is the area of ABC, and the sign + (resp. -) corresponds to the outward (resp. inward) construction. We derive from this an explicit classification in terms of two quadrics and a cubic inequality. It recovers the theorem of Dipierro, Noakes and Valdinoci for \beta=\pi/3, and exhibits three families of solutions for \beta>\pi/3, together with a fourth one when \cos\beta<1/4.
Let ABC be a non-equilateral spherical triangle with sides shorter than 2\beta. The triangle XYZ of the outward construction is equilateral if and only if \cos(\Theta+\Omega/2)=0; for the inward construction the condition is \cos(\Theta-\Omega/2)=0. In this case XYZ is nondegenerate, and its side \ell satisfies \cos\ell=\sigma\cos\beta/(1-\sigma\cos\beta), where \sigma=\pm1 is the sine of \Theta\pm\Omega/2.
Let ABC be as in Theorem A, and let \varepsilon=1 for the outward and \varepsilon=-1 for the inward construction.
The construction with sign \varepsilon gives an equilateral XYZ if and only if, for some \kappa\in\{\cos\beta,-\cos\beta\}, we have (x,y,z)\in\mathcal{Q}_\kappa and \varepsilon\kappa(x+y+z-xyz)>0. Conversely, every point of \mathcal{Q}_\kappa in (0,\tan\beta)^3, outside the diagonal x=y=z and the surface x+y+z=xyz, determines a spherical triangle whose construction with sign \varepsilon=\operatorname{sign}(\kappa(x+y+z-xyz)) gives an equilateral XYZ.
If \beta\le\pi/3, non-equilateral solutions only occur in the outward construction. They form a two-parameter family, parametrized by the part of the ellipsoid \mathcal{Q}_{\cos\beta} in the open positive octant with its diagonal point removed. For \beta=\pi/3 this is the family of [DNV24].
If \beta>\pi/3, there are two outward families and one inward family, and a second inward family exists if and only if \cos\beta<1/4.
For every spherical triangle,
For all real p,q,r we have the polynomial identity
Together with (?) and 2\cos^2(a/2)=1+p, it gives \alpha^2+\delta^2=16g^2. On the other hand, the solid-angle formula of van Oosterom and Strackee [vOS83] reads
the argument lying in (0,\pi) since \delta>0. For \alpha\ne0 this is the relation \tan(\Omega/2)=\delta/\alpha, and for \alpha=0 it gives \Omega=\pi; the latter case does occur, e.g. for p=q=-3/5 and r=1/5. Combining the argument and the modulus, we obtain (2.1). Note that its real and imaginary parts, \cos(\Omega/2)=\alpha/(4g) and \sin(\Omega/2)=\delta/(4g), are the classical half-area formulas of Euler and Cagnoli [Tod86].
Let B,C\in\mathbb{S}^2 be distinct and not antipodal. A point D\in\mathbb{S}^2 is a \beta-point of BC if |DB|=|DC| and the Euclidean angle \angle BDC equals \beta.
The circumcentre X is
and \langle X,B\rangle=\langle X,C\rangle=x\cot\beta.
The spherical triangle XBC is isosceles with apex angle 2\beta. Its base angle \theta_a is acute and satisfies
The \beta-point is D=(1+\sec\beta)\,x\cot\beta\,X-\tfrac12\sec\beta\,(B+C), and it lies on the same side of BC as X.
The analogous statements hold for Y,E and Z,F.
We first define a candidate X by the right-hand side of (?), and then show that it is the circumcentre of the chosen D,B,C. The vectors B+C and B\times C are orthogonal, with squared lengths 2(1+p) and 1-p^2, so that (?) gives
Thus |X|=1, and \langle X,B\rangle=\langle X,C\rangle=x\cot\beta=\mathrel{\mathop:} t\in(0,1). Let \Gamma=\{\langle X,U\rangle=t\}\cap\mathbb{S}^2, with centre G=tX and radius R=\sqrt{1-t^2}. Using (?), we get
so that the spherical angle BXC equals 2\beta.
The component of X in the plane spanned by B,C is a positive multiple of B+C. Hence the perpendicular from X to the great circle BC meets the shorter arc at its midpoint (B+C)/|B+C|, and divides XBC into two right triangles, each with hypotenuse d(X,B), angle \beta at X, and leg a/2 opposite to X. If \theta_a denotes the angle at B, the second rule in (?) gives \cot\beta\cot\theta_a=\cos d(X,B)=x\cot\beta. Hence \cot\theta_a=x>0, and \theta_a is acute. The first rule gives \cos\beta=\cos(a/2)\sin\theta_a.
Let M=(B+C)/2. Since the central angle is 2\beta<\pi, we have |G-M|=R\cos\beta, so that the point D'=G+\sec\beta\,(G-M) lies on \Gamma. It lies on the perpendicular bisector of BC, beyond G as seen from M, hence on the major arc determined by the chord BC. The inscribed angle theorem gives \angle BD'C=\beta, and |D'B|=|D'C| by symmetry. Finally, \langle X,B\times C\rangle=-\varepsilon(1-p)/\tan\beta and \langle D',B\times C\rangle=(1+\sec\beta)\,t\,\langle X,B\times C\rangle both have the sign of -\varepsilon. We conclude that D' is the chosen \beta-point D, and that X is its spherical circumcentre. Expanding the expression for D' gives (iii).
Lemma 3.2 provides an intrinsic description of the construction: on each side of ABC, erect the isosceles spherical triangle with apex angle 2\beta and acute base angles, the three apices being taken all outwards or all inwards. We stress that the condition on the base angles cannot be dropped: on each side of BC there is also an isosceles triangle with the same apex angle and obtuse base angles, whose apex is (-x(B+C)-\varepsilon B\times C)/(\tan\beta\,(1+p)). In the plane the base angle equals \pi/2-\beta. On the sphere, (?) instead gives \sin\theta_a>\cos\beta, and therefore
For \beta=\pi/3 we have \sec\beta=2, and Lemma 3.2(iii) gives D+B+C=3x\cot\beta\,X. Hence X=(D+B+C)/|D+B+C| is the centroid of the equilateral spherical triangle DBC, which is the centre used in [DNV24].
Let ABC be a non-equilateral spherical triangle satisfying (?), let X,Y,Z be obtained by the construction with sign \varepsilon, and put \Phi=\Theta+\varepsilon\Omega/2. Then XYZ is equilateral if and only if \cos\Phi=0. In this case, let \sigma=\sin\Phi\in\{1,-1\}. If \sigma=1, then \beta>\pi/3. The triangle XYZ is nondegenerate, and its side \ell satisfies
By Lemmas 5.1 and 5.2, the equality \langle X,Y\rangle=\langle X,Z\rangle is equivalent to (y-z)J=0, and the other two equalities are equivalent to (z-x)J=0 and (x-y)J=0. Since ABC is not equilateral, the coordinates x,y,z are not all equal. We deduce that the three pairwise distances are equal if and only if J=0, i.e. by Lemma 5.3, if and only if \cos\Phi=0.
We now assume that \cos\Phi=0, and let \sigma=\sin\Phi. By Lemma 5.3, the product (\alpha-i\varepsilon\delta)((1-e_2)+i(e_1-e_3))=-4\sigma g/S is real. Since the second factor has squared modulus (1+x^2)(1+y^2)(1+z^2)=S^{-2}, we get
By (?), we have gS=\cos^3\beta. Letting \kappa=-\sigma\cos\beta, this reads
By (5.4), we have Q_\kappa=0, and (5.4) and (5.1) then give
and the same value for the other two pairs. This is (4.1).
It remains to show that XYZ is nondegenerate. If \sigma=1, then \rho=\cos\beta/(1-\cos\beta)\le1 forces \cos\beta\le1/2. Equality would give X=Y=Z; since \langle X,C\rangle=x\cot\beta and \langle Y,C\rangle=y\cot\beta, this would force x=y, and similarly y=z, which is impossible. Hence \beta>\pi/3 and 0<\rho<1. If \sigma=-1, then -1/2<\rho<0. In both cases the Gram matrix of X,Y,Z, whose eigenvalues are 1-\rho,1-\rho,1+2\rho, is positive definite, and we conclude that XYZ is nondegenerate.
Since 0<\theta_a,\theta_b,\theta_c<\pi/2 and 0<\Omega<2\pi, the condition \cos\Phi=0 means that
The value \pi/2 corresponds to \sigma=1, and the other value in each case to \sigma=-1. Under this condition, one reads off from (4.2) that \Theta>\pi/2 if and only if \varepsilon\sigma=-1.
Let ABC be a non-equilateral spherical triangle satisfying (?). The construction with sign \varepsilon gives an equilateral triangle XYZ if and only if, for some \kappa\in\{\cos\beta,-\cos\beta\},
In this case \kappa=-\sigma\cos\beta and \cos\ell=-\kappa/(1+\kappa).
Conversely, let (x,y,z)\in\mathcal{Q}_\kappa with 0<x,y,z<\tan\beta, not all equal, and x+y+z\ne xyz. Then there is a spherical triangle, unique up to isometry, with side cosines 2\cos^2\beta\,(1+x^2)-1, 2\cos^2\beta\,(1+y^2)-1, 2\cos^2\beta\,(1+z^2)-1. It satisfies (?), and its construction with \varepsilon=\operatorname{sign}(\kappa(x+y+z-xyz)) gives an equilateral XYZ.
(a) If XYZ is equilateral, the proof of Theorem 4.1 gives Q_\kappa=0 and \varepsilon\delta=4\kappa^3(e_1-e_3) with \kappa=-\sigma\cos\beta. Since \delta>0, the numbers \varepsilon and \kappa(e_1-e_3) have the same sign.
Conversely, assume that Q_\kappa=0 and \varepsilon\kappa(e_1-e_3)>0. By (5.4), we have \alpha=4\kappa^3(1-e_2). On the other hand, Lemma 2.1, (?) and (?) give
Hence \delta=4|\kappa^3(e_1-e_3)|, and the sign condition gives \varepsilon\delta=4\kappa^3(e_1-e_3). It follows that J=0, and XYZ is equilateral by the proof of Theorem 4.1.
(b) We define p,q,r by the stated formulas. The bounds 1<1+x^2<\sec^2\beta give -1<p<1, and similarly for q,r. To establish the existence of a triangle, we combine (Proof) and (5.4) with the algebraic identity (1+x^2)(1+y^2)(1+z^2)=(1-e_2)^2+(e_1-e_3)^2 from (?). This gives
The Gram criterion of §? now provides a triangle with these side cosines, unique up to isometry. The positivity of x,y,z gives (?), and the coordinates computed from its sides are the original x,y,z. Since they are not all equal, the triangle is not equilateral, and we conclude by applying (a) with \varepsilon=\operatorname{sign}(\kappa(e_1-e_3)).
Let 0<\beta\le\pi/3, and let ABC be non-equilateral and satisfy (?).
The inward construction never gives an equilateral XYZ.
The outward construction gives one if and only if \Theta+\Omega/2=3\pi/2, i.e. if and only if (x,y,z)\in\mathcal{Q}_{\cos\beta}. In this case \cos\ell=-\cos\beta/(1+\cos\beta).
The quadric \mathcal{Q}_{\cos\beta} is an ellipsoid, and each of its points in the open positive octant, except the diagonal one, comes from a triangle as in (ii). In particular, these triangles form a two-parameter family.
By (3.3) we have \Theta>\pi/2, and \Omega/2\in(0,\pi). For the outward construction we get \Phi>\pi/2, so that by (4.2) only \Phi=3\pi/2 is possible, with \sigma=-1, i.e. \kappa=\cos\beta; this gives (ii). For the inward construction we get \Phi>-\pi/2, so that only \Phi=\pi/2 is possible, with \sigma=1, which is excluded by Theorem 4.1 since \beta\le\pi/3; this gives (i).
For (iii), let \kappa=\cos\beta\in[1/2,1). Both eigenvalues and K_\kappa are positive, so that \mathcal{Q}_\kappa is an ellipsoid. A point of it in the open positive octant lies in the cube by (?). As \tan\beta\le\sqrt3, each base angle exceeds \pi/6, so that \Theta>\pi/2 and e_1>e_3. If the point is not diagonal, Theorem 4.3(b) provides a triangle, and the construction is the outward one since \kappa(e_1-e_3)>0. Finally, near (0,0,\sqrt{K_\kappa}) the quadric \mathcal{Q}_\kappa is the graph of
and z(x,y)>0 for small x,y\ge0. Taking small x,y>0 with x\ne y gives a two-parameter set of non-diagonal points of \mathcal{Q}_\kappa in the open octant.
Let \pi/3<\beta<\pi/2. Within the cube (0,\tan\beta)^3, consider the four sets
which correspond to the outward, outward, inward and inward construction respectively. The first three contain a two-dimensional set of non-diagonal points, and the last one contains non-diagonal points if and only if \cos\beta<1/4. In particular, the outward construction gives equilateral triangles of two different sizes, and so does the inward one when \cos\beta<1/4.
In this proof we write c=\cos\beta\in(0,1/2). For \kappa=\pm c, the two eigenvalues and K_\kappa are positive, so that both quadrics are ellipsoids. The strict inequalities defining the four parts give relatively open subsets of these surfaces. We show that the first three parts are nonempty in the admissible cube, and determine when the fourth one is.
\mathcal{Q}_c\cap\{e_1>e_3\}: by (?) we have 0<\sqrt{K_c}<\tan\beta. For small x,y>0 with x\ne y, the point (x,y,z(x,y)) given by (?) with \kappa=c lies on \mathcal{Q}_c and in the open cube, and e_1-e_3 is close to \sqrt{K_c}>0.
\mathcal{Q}_{-c}\cap\{e_1>e_3\}: here K_{-c}=(1+c)^2(1-2c)/c^2>0, and K_{-c}<\tan^2\beta is equivalent to (1+c)(1-2c)<1-c, i.e. to 2c^2>0. We then use (?) with \kappa=-c near (0,0,\sqrt{K_{-c}}) as before.
\mathcal{Q}_{-c}\cap\{e_1<e_3\}: the positive diagonal point of \mathcal{Q}_{-c} is (t,t,t), where t=(1+c)/(\sqrt3\,c). Since c<1/2, we have t>\sqrt3 and e_1-e_3=t(3-t^2)<0. Moreover t<\tan\beta, because (1+c)^2<3(1-c^2). A sufficiently small neighbourhood of this point on the ellipsoid remains in the open cube and satisfies e_1<e_3, and removing the diagonal point leaves a two-dimensional set of admissible points.
\mathcal{Q}_c\cap\{e_1<e_3\}: if c\ge1/4, then (?) gives e_2\le(1-c)^2/c^2\le9, and the arithmetic–geometric mean inequality e_1e_2\ge9e_3, with equality only for x=y=z, gives e_3\le e_1e_2/9\le e_1; hence there are no non-diagonal points. If c<1/4, the diagonal point t=(1-c)/(\sqrt3\,c)>\sqrt3 of \mathcal{Q}_c satisfies e_1<e_3 and lies in the open cube by (?), and the argument of the previous case applies.
The statement about the two sizes follows from Theorem 4.3(a).
A non-equilateral ABC gives equilateral triangles of centres in both constructions if and only if \Theta=\Omega=\pi. Indeed, by (4.2), the outward and inward values of \Phi lie in \{\pi/2,3\pi/2\} and \{-\pi/2,\pi/2\} respectively. Their difference is \Omega\in(0,2\pi) and their sum is 2\Theta>0, so that they must be 3\pi/2 and \pi/2; the converse follows from Theorem 4.1. By Lemma 2.1 and (?), this condition is equivalent to \alpha=0 and e_2=1. To express it on the quadrics, put c_\beta=\cos\beta. The algebraic identity (5.4), proved independently below, gives
so that the condition is also equivalent to (x,y,z)\in\mathcal{Q}_{\cos\beta}\cap\mathcal{Q}_{-\cos\beta}. Since the inward value of \Phi gives \sigma=1, this forces \beta>\pi/3. Example 7.2 provides an explicit instance.
Let ABC be non-equilateral and satisfy (?). Assume that the chosen construction gives an equilateral triangle XYZ whose side \ell is greater than \pi/2. Then DEF is a nondegenerate equilateral triangle, whose side m satisfies
and 0<m<\beta.
The inequality \cos\ell<0 forces \sigma=-1 in (4.1). We may therefore take \kappa=\cos\beta in (?), and \langle X,Y\rangle=-\kappa/(1+\kappa). Formula (?) gives \langle X,C+A\rangle=(x\alpha-\varepsilon\delta)/(\tan\beta\,(1+p)). Using (?) and the identity x(1-e_2)-(e_1-e_3)=-(1+x^2)(y+z), we obtain x\alpha-\varepsilon\delta=-4\kappa^3(1+x^2)(y+z), and hence
Using Lemma 3.2(iii), \sec\beta=1/\kappa and \cot^2\beta=\kappa^2/(1-\kappa^2), we get
This expression is symmetric in x,y,z, so that the three pairwise inner products of D,E,F are equal, with the value given in (4.7). Since e_2>0, their common value \rho_D is greater than \cos\beta. Since ABC is not equilateral, the inequality (?) is strict, and it gives \rho_D<1. Thus \cos\beta<\rho_D<1, and the Gram eigenvalues 1-\rho_D,1-\rho_D,1+2\rho_D are positive. We conclude that DEF is nondegenerate, and that its side satisfies 0<m<\beta.
The hypothesis that ABC is not equilateral cannot be omitted. Take \beta=\pi/3 and \cos a=\cos b=\cos c=-1/3, so that ABC is a face of a regular tetrahedron. Then x=y=z=1/\sqrt3 lies on \mathcal{Q}_{1/2}, and the outward construction gives X=-A, Y=-B, Z=-C. The triangle XYZ is equilateral with side greater than \pi/2, but D=E=F=-(A+B+C) and m=0.
The converse fails, even for a non-equilateral input. For \beta=\pi/3, take \cos a=\cos b=-1/5, \cos c=7/10, and construct outwards. This is a valid triangle with (x,y,z)=\sqrt{3/5}\,(1,1,2). We find \langle X,Y\rangle=-1/2 and \langle Y,Z\rangle=\langle Z,X\rangle=0, so that the sides of XYZ are 120^\circ,90^\circ,90^\circ. Nevertheless, DEF is equilateral, with \langle D,E\rangle=\langle E,F\rangle=\langle F,D\rangle=-1/5.
By Theorem 4.1, the condition \ell>\pi/2 is equivalent to \sigma=-1, i.e. the chosen construction selects \kappa=\cos\beta in Theorem 4.3. We stress that mere membership in \mathcal{Q}_{\cos\beta} would not suffice. In Example 7.2 the parameter point lies on both quadrics, but the inward construction has \ell<\pi/2 and a non-equilateral DEF; Example 7.3 gives another equilateral XYZ with \ell<\pi/2 and non-equilateral DEF. Note also that, contrary to \ell, the side m varies within each family covered by the proposition.
Let N_{xy}=\alpha xy-\varepsilon\delta(x+y)+pq-r, and define N_{yz}, N_{zx} cyclically; for instance N_{zx}=\alpha zx-\varepsilon\delta(z+x)+rp-q. Then
and similarly for the other pairs.
We take the inner product of the numerator in (?) with its cyclic counterpart. The four terms are given by \langle B+C,C+A\rangle=\alpha, \langle B+C,C\times A\rangle=\langle B\times C,C+A\rangle=\delta and \langle B\times C,C\times A\rangle=pq-r, which gives the first expression in (5.1); the second one follows from (?).
Let J=\alpha(e_1-e_3)-\varepsilon\delta(1-e_2). Then
and the same holds after cyclically permuting x,y,z together with a,b,c.
We compare the terms of N_{xy} and N_{zx}. The terms involving \alpha give xy(1+z^2)-xz(1+y^2)=x(y-z)(1-yz), and those involving \delta give (x+y)(1+z^2)-(x+z)(1+y^2)=(y-z)(1-e_2). For the remaining terms, recall from (?) that 1+z^2=(1+r)/(2\cos^2\beta) and 1+y^2=(1+q)/(2\cos^2\beta), and note the identity
Since q-r=2\cos^2\beta\,(y^2-z^2), these terms contribute \alpha(y-z)(y+z). The total is (y-z)[\alpha(x(1-yz)+y+z)-\varepsilon\delta(1-e_2)]=(y-z)J.
Let S=\sin\theta_a\sin\theta_b\sin\theta_c and \Phi=\Theta+\varepsilon\Omega/2. Then
In particular, J=-(4g/S)\cos\Phi.
For \kappa\in\{\cos\beta,-\cos\beta\}, let Q_\kappa=\kappa^2(x^2+y^2+z^2)+2\kappa^3e_2-(1-\kappa)^2(1+2\kappa). Then
If moreover \varepsilon\delta=4\kappa^3(e_1-e_3), then
By (?), we have \alpha=2\kappa^2(3+x^2+y^2+z^2)-2, and (5.4) reduces to -2+6\kappa^2-4\kappa^3=-2(1-\kappa)^2(1+2\kappa).
For (5.4), we substitute \alpha=4\kappa^3(1-e_2)+2Q_\kappa and \varepsilon\delta=4\kappa^3(e_1-e_3). Since (1-e_2)xy-(e_1-e_3)(x+y)=xy-x^2y^2-(x+y)^2-z(x+y), we get
The variable z also occurs in -r=1-2\kappa^2(1+z^2), and
After this substitution, N_{xy}-2(xy-1)Q_\kappa no longer involves z, and equals
Substituting p=2\kappa^2(1+x^2)-1 and q=2\kappa^2(1+y^2)-1, this expression reduces to -4\kappa^3(1-\kappa)(1+x^2)(1+y^2), which proves (5.4).
Take (x,y,z)=\frac{\sqrt2}{5}(1,2,3) on the ellipsoid x^2+y^2+z^2+xy+yz+zx=2. Then e_2=22/25, and the side cosines are -23/50, -17/50, -7/50, so that the sides are approximately 117.39^\circ, 109.88^\circ, 98.05^\circ. This is the triangle of [DNV24, Example 1.6]. Its base angles are approximately 74.21^\circ, 60.50^\circ, 49.68^\circ, so that \Theta\approx184.40^\circ, and \Omega\approx171.21^\circ; the exact relation is \Theta+\Omega/2=270^\circ. The outward centres have common inner product -1/3, and \cos m=(1+e_2)/2=47/50. We refer to Figure ? for a picture.
Let \cos^2\beta=11/60, so that \beta\approx64.65^\circ, and take \cos a=\cos b=-3/5 and \cos c=1/5. The sides of ABC are approximately 126.87^\circ, 126.87^\circ, 78.46^\circ. We have (x,y,z)=(1,1,5)/\sqrt{11}, e_2=1, e_1-e_3=72/(11\sqrt{11})>0 and \alpha=0. Hence \Theta=\Omega=180^\circ, and the parameter point lies on both quadrics. The outward triangle of centres has \cos\ell=-\cos\beta/(1+\cos\beta), i.e. \ell\approx107.45^\circ, while the inward one has \cos\ell=\cos\beta/(1-\cos\beta), i.e. \ell\approx41.51^\circ. The sign condition of Theorem 4.3 selects \mathcal{Q}_{\cos\beta} for the outward construction and \mathcal{Q}_{-\cos\beta} for the inward one. Accordingly, the outward DEF is equilateral, with \cos m=\cos\beta/(1-\cos\beta), whereas the inward DEF is not: its pairwise inner products are approximately -0.6468, -0.2391, -0.2391.
The point (2,\frac83,\frac{10}9) lies on \mathcal{Q}_{-1/3}, and e_1-e_3=-4/27<0. Its side cosines are 1/9, 65/81, -367/729, so that the sides are approximately 83.62^\circ, 36.63^\circ, 120.23^\circ. Here \Theta\approx89.11^\circ and \Omega\approx1.78^\circ, and the exact relation is \Theta+\Omega/2=90^\circ. Thus the outward XYZ is equilateral with side 60^\circ. The triangle DEF has sides approximately 54.53^\circ, 127.22^\circ, 167.24^\circ, so it is not equilateral. For comparison, the point (1,2,\frac{2\sqrt3-3}{3})\in\mathcal{Q}_{1/3} satisfies e_1-e_3>0 and gives an outward equilateral XYZ with \cos\ell=-1/4. These two inputs realize the two outward sizes for the same \beta.
We now take \cos\beta=1/5<1/4. The point (\frac43,2,\frac{10}3) lies on \mathcal{Q}_{1/5}, whose equation is 5(x^2+y^2+z^2)+2e_2=112, and e_1-e_3=-20/9<0. The side cosines are -7/9, -3/5, -7/225, so that the sides are approximately 141.06^\circ, 126.87^\circ, 91.78^\circ. The area is \Omega\approx340.27^\circ, while the base angles \operatorname{arccot}(4/3), \operatorname{arccot}2, \operatorname{arccot}(10/3) have sum \Theta\approx80.13^\circ. Thus \Theta-\Omega/2=-90^\circ, and the inward XYZ is equilateral with \cos\ell=-1/6. Proposition 4.7 shows that DEF is equilateral as well, with \cos m=8/9.
We keep \beta fixed and consider triangles with a=\rho a_0+o(\rho), b=\rho b_0+o(\rho), c=\rho c_0+o(\rho) as \rho\to0, where a_0,b_0,c_0 are the sides of a nondegenerate Euclidean triangle. We put \operatorname{diam}=\max(a,b,c). Then \Omega\to0 and each base angle tends to \pi/2-\beta. For either construction, \cos\Phi\to\cos(3\pi/2-3\beta)=-\sin3\beta, which is nonzero unless \beta=\pi/3. For \beta=\pi/3, the next term explains the relation with the planar theorem. From \sin\theta_a=1/(2\cos(a/2)) we obtain \theta_a=\pi/6+a^2/(8\sqrt3)+O(a^4). Let \Delta be the area of the Euclidean triangle with sides a,b,c, and put s_0=(a+b+c)/2. L’Huilier’s formula
and Heron’s formula give \Omega=\Delta+O(\operatorname{diam}^4). Therefore, for the inward construction,
The quadratic term is nonnegative by Weitzenböck’s inequality [BDJ+69], and vanishes if and only if a=b=c. In other words, to leading order it measures how far the inward construction is from the equation \Theta-\Omega/2=\pi/2. For the outward construction, \Theta+\Omega/2 tends to \pi/2, whereas its only admissible value is 3\pi/2. We stress that the expansion only describes this limiting behaviour; the exact exclusion of non-equilateral inward solutions is Corollary 4.4.
The appendix proves the angle criterion by composing the rotations about X,Y,Z, and uses quaternions to compute the composition. Can this argument be replaced by a synthetic proof?
What is the corresponding classification in the hyperbolic plane? The Euclidean chord-angle definition of a \beta-point has no direct intrinsic analogue, but Remark 3.3 suggests erecting isosceles hyperbolic triangles with apex angle 2\beta. Their base angles are automatically smaller than \pi/2-\beta, so that there is no ambiguity between acute and obtuse base angles. For \beta=\pi/3, this construction should be compared with [DNV25].
The Petr–Douglas–Neumann theorem extends Napoleon’s theorem to polygons (see the references in [DNV25]). Is there a spherical counterpart obtained by imposing similarity on chord polygons?
We have q_X=h_{BC}\,r_B\bigl(-\varepsilon(\pi/2-\theta_a)\bigr), and cyclically.
We write h_{BC}=(\cos(a/2),\sin(a/2)\,n), where n=(B\times C)/\sin a, and put \mu=-\varepsilon(\pi/2-\theta_a), so that \cos\mu=\sin\theta_a and \sin\mu=-\varepsilon\cos\theta_a. As n\perp B, the real part of h_{BC}r_B(\mu) is \cos(a/2)\sin\theta_a=\cos\beta. For its vector part, we use n\times B=(C-pB)/\sin a and 2\cos^2(a/2)=1+p to obtain
By (?), we have \sin\theta_a=\cos\beta/\cos\frac a2 and \cos\theta_a=x\sin\theta_a. Hence the vector part equals \cos\beta\,(-\varepsilon x(B+C)+B\times C)/(1+p)=-\varepsilon\sin\beta\,X by (?).
We have q_Zq_Yq_X=(-\sin\Phi,\varepsilon\cos\Phi\,B). In particular, R_Z\circ R_Y\circ R_X is the rotation about B through the angle \varepsilon(2\Phi-\pi) modulo 2\pi, and it is the identity if and only if \cos\Phi=0.
We write \mu_a=-\varepsilon(\pi/2-\theta_a), and similarly \mu_b,\mu_c. By Lemma A.1 and the rule r_V(\lambda)q=q\,r_U(\lambda), applied to h_{CA} (which carries C to A) and to h_{BC} (which carries B to C), we have
We now compute H. Using \langle C\times A,B\times C\rangle=pq-r and (C\times A)\times(B\times C)=-\delta C, we find
Multiplying by h_{AB}=(1+r,A\times B)/(2\cos\frac c2), and using \langle A\times B,B\times C\rangle=rp-q, \langle A\times B,C\times A\rangle=qr-p, \langle A\times B,C\rangle=\delta and (?), we see that the real part of 8g\,H is
Applied from right to left, the three factors of H take B to C, then to A, and finally back to B. Thus H fixes B, and its vector part H_{\mathrm{vec}} is a multiple of B. The identities
give \langle 8g\,H_{\mathrm{vec}},B\rangle=(1+r)\delta+\delta(1-r)=2\delta. We conclude that H=(\alpha,\delta B)/(4g), which equals r_B(\Omega/2) by Lemma 2.1. Since \mu_a+\mu_b+\mu_c=-\varepsilon(3\pi/2-\Theta) and \Omega/2+\varepsilon\Theta=\varepsilon\Phi, we deduce that
The rotation angle is 2\varepsilon(\Phi-3\pi/2)\equiv\varepsilon(2\Phi-\pi) modulo 2\pi, and it vanishes modulo 2\pi if and only if \cos\Phi=0.
The text on this page is from arXiv:2610.21408v1 by Quang Hung Tran, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.
The record of each item is in analysis/objects/2610/2610.21408.jsonl.