https://journals.sagepub.com/doi/full/10.1177/20416695209584...

Because the daylight spectrum defines an infinitely dimensional cuboid in the space of spectra, this region is a convex, centrally symmetric volume in color space. Its structure has been described by Schrödinger (1920). Colors on the boundary of this “color solid” are proper “parts of daylight” in the sense that their spectra are characteristic functions of connected spectral ranges or complements thereof.

This can be used to find the nature of spectral sampling by the human visual system. Split the spectrum into three parts by way of two cuts. Place the cuts thus that the resulting RGB space claims the largest possible volume fraction of the full Schrödinger color solid. This is a well-defined optimization problem because volume ratios are invariant against arbitrary colorimetric transformations. One finds (numerically, using the CIE color matching functions shown in Figure 3 right) that there is a unique solution, and the cuts should be at wavelengths of 482.65 nm and 565.43 nm (Figure 3 left). This yields a unique RGB basis for color space. The convex hull of the basis vectors is the parallelepided of largest volume that can be inscribed in the color solid, making it the optimum RGB basis (Figure 4 right). The corresponding color matching functions (Figure 4 left) are predominantly nonnegative and are mutually only weakly correlated.

- https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5624368/