Why “Certainly on the Circumference” (and on the Sphere)
Category: Debate
Most ways to “draw a circle” are approximations: they split the arc into chords, use regular polygons, or compute points from arc lengths (where \(\pi\) eventually appears). In GRG we do the opposite: we do not measure the arc, we enforce the distance. We fix a radius \(R\) and choose a geometric direction; the generated point is the endpoint of a segment of length \(R\) from a center \(O\). By definition, that point is at distance \(R\) from \(O\): therefore it is certainly on the circumference. No approximation, no \(\pi\) needed.
What we prove, in plain words
In 2D, every point built as \((x,y)=(R\cos\varphi,\,R\sin\varphi)\) identically satisfies
$$x^{2}+y^{2}=R^{2}.$$
It does not “tend” to the circle: it is on the circle.
In 3D, take a unit direction \(\mathbf{u}\) and set \(\mathbf{P}=R\,\mathbf{u}\). By construction \(\|\mathbf{P}\|=R\): the point is certainly on the sphere
$$x^{2}+y^{2}+z^{2}=R^{2}.$$
The heart of it is: distance first. If the distance from the center is enforced to be \(R\), every generated point belongs to the locus of points at distance \(R\). That is the very definition of circle/sphere.
“Then what are sine and cosine for?”
Here they are not approximate numbers: they are ratios of sides in similar right triangles. They serve as a language to express the Cartesian projections of the chosen direction:
\[
x=R\cos\varphi,\qquad y=R\sin\varphi.
\]
Even if, in a digital controller, you compute them with finite precision, the geometric law you implement remains exact: the commanded points are those of the ideal object defined by \(x^{2}+y^{2}=R^{2}\) (or \(x^{2}+y^{2}+z^{2}=R^{2}\)).
If you wish to avoid direct computation of \(\sin\) and \(\cos\) altogether, there is an equivalent incremental form: rotate the vector \((x,y)\) by multiplying it with an orthogonal matrix. Orthogonal matrices preserve the norm: every step remains on the circle by construction.
“What about error accumulating at each step?”
Distinguish carefully:
- Geometric law (GRG): exact. The norm is preserved at each step (rotations), therefore there is no geometric drift.
- Physical implementation: it can introduce errors (backlash, compliance, friction, encoder resolution, numerical rounding). These are real-world errors, not flaws of the idea.
In digital control, the small loss/gain of norm due to rounding is removed with three simple techniques:
\(\text{(i) adequate precision}\quad \text{(ii) occasional renormalization enforcing }x^{2}+y^{2}=R^{2}\quad \text{(iii) periodic recomputation from }\varphi.\)
In all cases, the reference object remains the perfect circle/sphere.
“But the machine moves in straight lines, not curves!”
What you often see is the visualization (a polyline between samples). It is not the definition of the path. The GRG law commands instantaneous poses that lie on the circle/sphere. The drive’s interpolation between samples may be linear or spline-based: that is a control choice. If you want a “fuller” trajectory, increase the sampling rate or use an arc interpolator; in any case, the nodes remain exactly on the geometric locus.
Why \(\pi\) is not needed
\(\pi\) is needed when you measure the arc (lengths, areas, times along an arc). In GRG we do not measure the arc: we enforce the distance from the center and the direction. Therefore \(\pi\) is irrelevant to guarantee that points lie on a circle/sphere. You may use \(\pi\) later if you care about arc length, but not to construct the point.
Natural extension to the sphere (including the \(Z\) axis)
What holds in \(X\!-\!Y\) holds in \(X\!-\!Y\!-\!Z\). A unit direction in 3D (obtained via Cartesian projections, without “walking on the surface”) is scaled by \(R\). The result has norm \(R\). That is all. Azimuth–elevation formulas are just a convenient way to obtain \((x,y,z)\). The key property remains the preservation of distance.
Practical consequences for robotics, CNC, and 3D printers
If your machine moves on Cartesian axes and you command \((x,y)\) (or \((x,y,z)\)) according to the GRG law, the ideal path is exactly the circumference (or the sphere).
Any deviations you might measure on a real part come from mechanics and numerics: backlash, compliance, calibration, friction, quantization. Improve the machine, the controller, and the metrology, and you approach the ideal without changing the GRG law.
Typical objections, short answers
- “Without \(\pi\) I cannot make a circle.” False: a circle is the locus of points at distance \(R\) from a center. No \(\pi\) is needed to guarantee that distance.
- “Sine and cosine are approximate numeric functions.” In geometry they are exact ratios; in control you can use incremental orthogonal rotations that preserve the norm by construction.
- “Steps accumulate error.” Rotations do not change the norm. In digital control, rounding errors are contained with renormalization or periodic recomputation.
- “Points on the sphere are not uniformly distributed.” Different topic: sampling uniformity does not affect the fact that each point lies on the sphere.
The key takeaway
GRG does not approximate the circle or the sphere: it imposes them. With a radius \(R\) fixed and a direction chosen, the distance from the center is \(R\) by construction: therefore the point is certainly on the circumference (or on the sphere). In a Cartesian system, the geometric precision of the trajectory does not depend on \(\pi\) nor on arc approximations: it depends only on your machine and numerics.
GRG · Operative summary (turns & quarter)
Params. Pick \(N\). Full turn: \(u_j=j/N\) with \(\Delta u=1/N\). Generate \(P_j=(x_j,y_j)\) with \(x_j^2+y_j^2=R^2\). Quarter: \(\Delta x=R/N\).
Perimeter — full turn (radial)
Inscribed (lower)
\[
L_{\mathrm{in}}(N)=\sum_{j=0}^{N-1}\sqrt{(x_{j+1}-x_j)^2+(y_{j+1}-y_j)^2}.
\]
Circumscribed (upper) — tangents at \(P_j\): \(x\,x_j+y\,y_j=R^2\);
intersections \(T_j\) with \(\det_j=x_jy_{j+1}-y_jx_{j+1}\),
\(T_j^x=\frac{R^2(y_{j+1}-y_j)}{\det_j}\),
\(T_j^y=\frac{R^2(x_j-x_{j+1})}{\det_j}\).
\[
L_{\mathrm{out}}(N)=\sum_{j=0}^{N-1}\lVert T_{j+1}-T_j\rVert.
\]
Radial area — full turn
\[
A_{\mathrm{in}}(N)=\tfrac12\sum_{j=0}^{N-1}(x_jy_{j+1}-x_{j+1}y_j),\qquad
A_{\mathrm{out}}(N)=\tfrac12\sum_{j=0}^{N-1}(T_j^xT_{j+1}^y-T_{j+1}^xT_j^y).
\]
Cartesian area — quarter (Δx) → ×4
Abscissas \(x_i=i\Delta x\), \(\Delta x=R/N\); heights \(y_i=\sqrt{R^2-(i\Delta x)^2}\).
\[
A_{\mathrm{right,q}}=\sum_{i=1}^{N}y_i\Delta x,\quad
A_{\mathrm{mid,q}}=\sum_{i=1}^{N}y\!\left(x_{i-\tfrac12}\right)\Delta x,\quad
A_{\mathrm{trap,q}}=\sum_{i=1}^{N}\tfrac{y_{i-1}+y_i}{2}\Delta x.
\]
Bounds: \(A_{\mathrm{right,q}}\le A_{\mathrm{quarter}}\le A_{\mathrm{mid,q}}\); disk via ×4. π remains outside the computation; trig only as cross-check.
