On the Aitchison Geometry

157 Views Asked by At

This question is for people who know the Aitchison Geometry - I'm working on a (more mathematical and not statistical) paper on the Aitchison Geometry and I try to understand how ellipses (or any other geometrical objects) are defined in the Aitchison Geometry and how to present them. I've seen in many papers that for any $\mathbf{x} \in \mathcal{S}^{3}$, which is the part simplex, a possible pair of coordinates is given by $$ \frac{2}{\sqrt{6}} \ln \frac{x_{1}}{\left(x_{2} x_{3}\right)^{1 / 2}}, \quad \frac{1}{\sqrt{2}} \ln \frac{x_{2}}{x_{3}}. $$ What does this exactly mean? and how to set up graphics like this one (can also be by using R): Ellipses in the space of coordinates (left) and in the simplex (right), where discontinuous lines represent orthogonal axes.

Thanks in advance!

1

There are 1 best solutions below

3
On BEST ANSWER

Let us give maybe the most natural way to consider simplex $S_3$ (with coordinates $(x_1,x_2,x_3)$), i.e., as a 3D object ; it is indeed the equilateral triangle obtained as the intersection of the plane with equation:

$$x_1+x_2+x_3=1$$

with the coordinate planes $x_1Ox_2, x_2Ox_3, x_3Ox_1$.

This the convention used in this excellent Question here.

enter image description here

Fig. 1.: The simplex triangle in a 3D setting.

In a second step, let us understand the genesis of the Aitchinson coordinates in this 3D representation. Let $c_1$ and $c_2$ be some positive constants ; let us consider the isolevel surfaces with the following equations (please note that we have got rid of the logarithms):

$$\dfrac{x_1}{\sqrt{x_2x_3}}=c_1 \ \ \iff \ \ x_1^2=c_1^2x_2x_3\tag{1}$$

which, when $c_1$ varies, is a variable cone (see explanation below) and

$$\dfrac{x_2}{x_3}=c_2 \ \ \iff \ \ x_2=c_2x_3 \tag{2}$$

which is a plane, rotating about the $x_1$ axis when $c_2$ varies.

The trace of the (increasingly squeezed) cones on the triangle will be (increasingly thin) ellipses sharing two common points.

The traces of the planes are straight lines passing through point $A$ (with coordinates $(x_1=1,x_2=0,x_3=0)$) .

These curves correspond to what is shown in your reference document. They can indeed be used as a system giving a new system of coordinates through coefficients $c_1$ and $c_2$ given in (1) and (2).

enter image description here

Fig. 2: Simplex triangle (red) with its plane (light blue). Plane $x_1Ox_2$ in grey. Variable plane with equation (2) (lighter grey). Variable cone (purple) always containing $x_2$ and $x_3$ axes.

enter image description here

Fig. 3: When $c_1$ and $c_2$ are varying... one gets curvilinear coordinates inside the triangle. It is to be noted that not all of the curves are ellipses. The leftmost are branches of hyperbolas.

Remark: coefficients $2/\sqrt{6},\ 1/\sqrt{2}$ are surely there for the benefit of some normalization. To be checked.

Explanation about the fact that (1) is the equation of a cone: the reason is homogeneity ; if a point $(x_1,x_2,x_3)$ belongs to the surface defined by (1), then any multiple $(ax_1,ax_2,ax_3)$ belongs to it as well.