An example of non-differentiability in triangle geometry (Q2425371)

Given a triangle, use the feet of its altitudes as vertices in a new triangle. Iterate this to get a sequence of triangles that converges to a point. The authors consider how that point depends on the original triangle. They place the triangle's vertices on a circle centered at the origin in the complex plane, with an orientation that makes the sum of the arguments of the vertices equals \(0\) modulo \(2\pi\). Then the position of the limit point depends only on the first two of those arguments, \(\alpha\) and \(\beta\) say. The authors show that the resulting function of two variables is continuous but nowhere differentiable, even though it satisfies very strong Hölder conditions. They recall that, in each step of the iteration, the nine-point Euler circle for the current triangle is the circumscribed circle for the next triangle. Its radius is half the radius of the previous circumcircle. And the arguments of the new vertices relative to the center of the new circle come from multiplying the arguments of the old vertices by \(-2\). This allows the authors to represent the position of the limit point as a sum of a lacunary Fourier series in two variables. They provide an enlightening analysis of that series. In the special case where~\(\beta = \alpha\), the position of the limit point becomes a function of one variable. The Fourier series of that function is again lacunary, in a form given by Weierstrass. Hardy showed that the function is nowhere differentiable. The current authors' proof of this is simpler, and they go on to give detailed information about the function. This includes facts about local maxima and minima, and slow and rapid points. They return to their function of two variables, and prove various things about its range. Finally, they comment on the history of their methods for lacunary series.