Things all start with the human eye. The retina contains two kinds of cells, color-sensitive cones and color insensitive rods. The eye can sense light wavelengths from roughly 380 nm (a nm is 10-9 meters) for blue-indigo light, to 720 nm for red light.
The tristimulus theory of color states that humans have just three kinds of color-sensitive cones. the cones can be denoted S, M and L. The S cones sense blue light (rather weakly), while the M cones sense a spectrum of colors centered on the green, and the L cones sense an even broader spectrum of colors including long wavelengths (red). Here is a plot of the cone responses. The spectra of these cones define a 3-dimensional space of colors in the infinite dimensional space of light spectra that humans can see. All other bases, like XYZ and RGB are derived from the basic cone responses.
From the cones, some other values are computed which are sent to the brain. These are:
Standard models of color perception were derived by the CIE (Comission Internationale d'Eclairage, Eclairage is lighting). They used standard wavelengths of
red = 700nm, green = 546.1nm, blue = 435.8nm.
They asked observers to match an arbitrary single wavelength of light by mixing the red, green and blue lights. Negative values for red, green, blue, were allowed and were achieved by adding the same value to the target wavelength patch (instead of the red+blue+green patch).
The resulting diagram (see the textbook) has a predictable separation into red, green and blue channels. Unfortunately, there is a large spectral region requiring negative red light for matching.
Note that this diagram allows one to duplicate the appearance for a human observer, of light with any spectrum (subject to the possibility that the red component might be negative). One would break the spectrum into separate wavelengths, then match each in turn, then add up the RGB values for all of them.
To correct for problems with the negative red component, the CIE adopted another standard with components derived from the RGB components above. This is the XYZ scheme. The response spectra for X, Y and Z channels can be seen in this figure. Note that X, Y and Z are not real colors. There is no physical way to generate them and combine them to match an arbitrary color. However, they do turn out to be very convenient for representing many other kinds of color system and serve as the standard for conversions.
The X, Y, Z components are all proportional to light intensity, and the set of visible colors represented as (x,y,z) coordinates is a cone (set of rays from the origin). To understand the color information only, we take the intersection of this cone with the
X + Y + Z = 1
plane. Then we take that curve and project it onto its X-Y coordinates. The result is the CIE chromaticity diagram, which is given here in b/w line form, and below with the colors shaded in. This figure is taken from the URL http://www.arce.ukans.edu/book/color/Cie
Note that white light appears near the center of the diagram.
The gamut of a display device is the set of visible colors that it can produce. Since physical display devices can only generate positive combinations of their basic colors, the gamut of a display device is the convex hull of its basic colors.
From the choices of the CIE color matching experiments, which were:
red = 700nm, green = 546.1nm, blue = 435.8nm
you can plot a triangle on the figure above to determine the gamut of this display. Notice that there is a large region of unachievable colors to the left of the line between the blue and the green primitive color. Matching pure colors at those wavelengths would require a substantial amount of "negative red". The explains the outcome of the CIE experiments on matching.
It also explains why high-quality display devices sometimes use more than 3 colors. The colors are spaced around the boundary of the CIE chromaticity diagram so that their convex hull more fully covers the set of achievable colors.
