@Daminark I have never used a book for it, only a combination of videos, articles, and courses. I often learn higher level things and then infill down.
I don't know the fancy stuff. I think every math major should know the basic stuff (including continuous distributions, law of large numbers, etc.) — but not the grad level which is just measure theory and all theory.
Looks like there's no MathJax here? I saw my LaTeX code come through as LaTeX code instead of as formatted math, and it took me a minute to realize that that wasn't what I wanted. :D
@TobiasKildetoft So if we call the discriminant of the characteristic polynomial $d$, and we call the characteristic polynomial $p$, then we know $d(A)\cdot p(A)=0$. (If $A$ is diagonalizable, this is the easy case of C–H, so $p(A)=0$. If $A$ isn't diagonal, its eigenvalues aren't all distinct, so $d(A)=0$.) If the product of two polynomials is identically zero, at least one is (essentially, look at the term with the highest power); since $d$ is not identically zero, $p$ must be.
@MikeMiller The sectional curvature formula is interesting because it says $\text{sec}(X, Y) = 0$ if $X$ and $Y$ commute. Which is sort of how it should be, given the parallel transport along "$XYX^{-1}Y^{-1}$" interpretation of the Riemann curvature tensor.
Assuming you accept this (which essentially says the ring of multivariate polynomials has no zero divisors), I don't see what else needs to be made precise
@TedShifrin I think I can guess how to derive it from Gauss' equation for 3-manifolds. As a baby case I can pick $X = \partial/\partial x, Y = \partial/\partial y$ in a coordinate chart and then $R(X, Y) = \nabla_{\partial/\partial x} \nabla_{\partial/\partial y} - \nabla_{\partial /\partial y}\nabla_{\partial/\partial x}$. If I expand this in terms of Christoffel symbols, I should get one of those Gauss formulas.
@Ted Pick a small exponential chart for the surface in normal coordinates. I would expect that as you linearize this surface to the tangent plane you can show that the derivative of this process (the third order contribution to curvature) is nil.
If the surface was totally geodesic, we could use normal coordinates at the point to show this, right? Because Riemann curvature tensor becomes the Euclidean tensor at that point.