How were the RGB Color Matching Functions established from 380 to 436 nm? I have been reviewing how RGB color matching functions were formed, and I seem to be missing an aspect.  The three primary monochrome lights used to generate a color like the target color are roughly described as blue, green, and red.  The blue primary color comes from either the Hg line near 436 nm, or I have also seen reference to gelatin filters to isolate 436 nm (hence not so monochromic).
The RGB color matching functions are shown below:


If I followed the methodology correctly, the curves were established by (a) providing a target color (based on a monochromatic source), and separately (b) adjusting each of the three primary colors until an observer determines that the color produced by the mixture of primaries matches the target.  The monochromatic target is scanned from 380 to 740 nm – and the question about matching is answered at each wavelength.  The primaries used in this experiment had peak locations at 436, 546, and 700 nm.  Mixing these primaries should only be able to match color from the scanning target monochromatic light between the two extremes (436 and 700 nm).
Notice that all three functions begin at 380 nm.  I do not understand how the curve from 380 to 436 nm can be established using a primary light at 436 nm.  Can anyone help me understand this?
 A: In fact, if there are only three saturated primaries, you can only directly get three color matches for the saturated colors, and a filled triangle of (desaturated) chromaticities that correspond to the mixtures of the primaries. You won't be able to match most of the monochromatic colors at all, even if the primaries are monochromatic*.
The crucial part of the methodology is that, if a color formed by the primaries appears to be impossible to match the reference monochromatic color, some amount of a primary is added to the reference, which has the effect of subtracting corresponding spectral density from the three-color mixture (+ desaturating both colors of the pair being matched).
The principle of such desaturation is like
$$C_\text{monochr.}+R_\text{offset}=R+G+B\\
\downarrow\\
C_\text{monochr.}=\underbrace{R-R_\text{offset}}_\text{can be negative}+G+B.$$
This is what gives you negative values in the RGB color space. As you can understand, the experiment didn't emit negative radiance at the subjects :) .
Note also that one shouldn't regard the $435.8\,\mathrm{nm}$ primary of the CIE RGB color space as something that was used by the original experimenters. In particular, the free access paper by Guild[1] specifies that the primaries were obtained by the use of gelatine filters.
In fact, it'll be useful for you to read the complete description of the methodology in this paper. You can also see there the actual measurement results and see how well (or poorly) different observers' CMFs match.

References

*

*J. Guild, "The Colorimetric Properties of The Spectrum", Philos. Trans. R. Soc., 1931,
Series A, vol. 230, pp. 149-187. https://doi.org/10.1098/rsta.1932.0005

*Well, you might be able to match saturated colors in the region of about $650\,\mathrm{nmD}$ to $550\,\mathrm{nmD}$ if you used $650\,\mathrm{nm}$ and $550\,\mathrm{nm}$ primaries because of almost straight line in the chromaticity gamut border at that location, but this wasn't known a priori.
A: I think it is more a biology question than a physics question. You just need to know that there are three types of color sensors in human eyes. I think the curves you show are made according to measured eye responses. As long as you invoke the same responses of the color sensors, the color perception is the same. Thus a 436 nm light source does not need to "establish curve from 380 to 436 nm (as shown in your plot)". It only needs to excite the blue sensor in human eyes.
