Consider a smooth curve defined by a continuous map f:[0,1]→Rn with n≥2 where f is supposed to have a continuous derivative. One can prove that the curve is rectifiable, its arc length being L=limn→∞n∑i=1|f(ti)–f(ti−1)|=∫10|f′(t)| dt with ti=in for 0≤i≤n.
What can happen when we consider a surface instead of a curve?
Consider a compact, smooth surface (possibly with boundary) embedded in R3. We can approximate it as a polyhedral surface composed of small triangles with all vertices on the initial surface. Will the sum of the areas of the triangles converges to the area of the surface if their size is converging to zero?
The answer is negative and we provide a counterexample named Schwarz lantern. We take a cylinder of radius r and height h. We approximate the cylinder by 4nm isosceles triangles positioned as in the picture in 2n slices. All triangles have the same base and height given by b=2rsin(πm), h=√r2[1−cos(πm)]2+(h2n)2 Hence the area of the polyhedral surface is S′(m,n)=4mnrsin(πm)√r2[1−cos(πm)]2+(h2n)2=4mnrsin(πm)√4r2sin4(π2m)+(h2n)2 From there, let’s have a look to the value of S′(m,n) as m,n→∞.
Suppose that n=am with a>0
In that case we have limm→∞S′(m,am)=S=2πrh where S is the area of the cylinder. We are in a similar case to the one of a smooth curve.
Now suppose that n=am2 with a>0
Then limm→∞S′(m,am2)=2πr√a2π2r2+h2 The limit area of the polyhedral surface is finite, but can be made arbitrarily larger than the one of the cylinder as a is increasing.
Finally, let’s take n=m3
We have S′(m,m3)≥8r2m4sin(πm)sin2(π2m) In that case limm→∞S′(m,m3)=∞.
Conclusion: the area of the cylinder cannot be measured as the limit area of a polyhedral surface composed of triangles if we don’t impose appropriate conditions on the triangles shape.