3
$\begingroup$

A surface, parametrized by $U$ and $V$, has a tangent bivector given by the wedge product

$$I = \vec x_U\wedge\vec x_V$$

where subscripts represent partial derivatives. The First Fundamental Form coefficients are defined as usual, but the Second Fundamental Form coefficients are vectors perpendicular to the surface:

$$E=\vec x_U\cdot\vec x_U=\lVert\vec x_U\rVert^2$$ $$F=\vec x_U\cdot\vec x_V$$ $$G=\vec x_V\cdot\vec x_V$$ $$\lVert I\rVert^2 = -I^2 = EG-F^2$$

$$\vec L=\vec x_{UU}^\perp=(\vec x_{UU}\wedge I)I^{-1}$$ $$\vec M=\vec x_{UV}^\perp$$ $$\vec N=\vec x_{VV}^\perp$$

If the surface is embedded in 3D space, then $\vec L$, $\vec M$, and $\vec N$ are all parallel to each other. In higher dimensions, they can be independent.

In terms of the above quantities, the Gaussian and mean curvatures are

$$K=\frac{\vec L\cdot\vec N-\vec M\cdot\vec M}{\lVert I\rVert^2}$$ $$\vec H=\frac{G\vec L-2F\vec M+E\vec N}{2\lVert I\rVert^2}$$

which are both invariant with respect to re-parametrization. But I found another such invariant while deriving a different formula* for $K$, and it has the same units as $K$ (inverse area). This quadvector quantity $Q$ seems to be another type of curvature, which is always $0$ in 3D.

$$Q = I\frac{(\vec L\wedge\vec M)G+(\vec M\wedge\vec N)E+(\vec N\wedge\vec L)F}{\lVert I\rVert^4}$$

I also discovered that $Q$ is proportional to the area of an ellipse traced by the normal curvature vector $(\frac{d^2\vec x}{ds^2}=k\hat n=\vec k)$ of geodesics with varying direction passing through the given point on the surface. (This ellipse also has neat relationships with $\vec H$ and $K$ .)

Has $Q$ ever been studied before? Is there a general method for finding or classifying such invariants of embedded manifolds?

* $$i = \frac I{\lVert I\rVert}$$ $$i\frac{\langle i_Ui_V\rangle_2}{\lVert\langle\vec x_U\vec x_V\rangle_2\rVert}=I\frac{\langle i_Ui_V\rangle_2}{\lVert I\rVert^2}=K-Q$$

(The angle brackets denote grade-projection in Geometric Algebra, and inside the brackets is a product of partial derivatives.)

$\endgroup$
0
$\begingroup$

See Hestenes' Tutorial, page 10 (Differential Geometry). (He uses "$I$" for the unit tangent bivector, where I use "$i$". Also, his multiplication order is $a\cdot\partial I = (a\cdot\partial)I \neq a\cdot(\partial I)$, which confused me at first.)


The quantity $K-Q$ is simply the coefficient of the curvature for a given bivector:

$$C(a\wedge b) = -(K-Q)(a\wedge b)$$

where $a$ and $b$ are tangent vectors (so their wedge product is proportional to $I$). The intrinsic curvature is $-K(a\wedge b)$, and the extrinsic curvature is $Q(a\wedge b)$, both of which are bivectors.


The curvature, $C(a\wedge b) = \frac{S(a)S(b)-S(b)S(a)}{2}$, is defined in terms of the "shape" $S$, which is the derivative of the projection operator:

$$P(a) = (a\cdot I)I^{-1} = (a\cdot i)(-i) = \frac{ai-ia}{2}(-i)$$ $$\nabla = \vec x^U\frac{\partial}{\partial U} + \vec x^V\frac{\partial}{\partial V} = \big(\vec x_VI^{-1}\big)\frac{\partial}{\partial U} + \big(-\vec x_UI^{-1}\big)\frac{\partial}{\partial V}$$ $$S(a) = \nabla P(a) = \nabla\big((a\cdot i)(-i)\big)$$ $$= \vec x^U\big((a\cdot i_U)(-i) + (a\cdot i)(-i_U)\big) + \vec x^V\big((a\cdot i_V)(-i) + (a\cdot i)(-i_V)\big)$$

(The superscript $\vec x^U$ denotes the reciprocal basis vector.) This works for any vector $a$, tangent or not. The scalar part of $S(a)$ turns out to be $2\vec H\cdot a$; the mean curvature $\vec H$ is orthogonal to the surface, so if $a$ is tangent to the surface, this scalar part disappears. Also, for tangent $a$, the shape can be calculated as a directional derivative (I might prove this in a later update):

$$S(a^\parallel) = -i(a\cdot\nabla)i$$

(only differentiating the right $i$, not the left $i$). In particular, $S(\vec x_U) = -i\frac{\partial i}{\partial U} = -ii_U$. We can differentiate the equation $ii = -1$ to get $i_Ui + ii_U = 0$; two bivectors' geometric product is anticommutative if and only if it is another bivector (no scalar or quadvector parts), so this implies that $S(\vec x_U)$ is a bivector. Explicitly,

$$i_U = \frac{\partial}{\partial U}\frac{I}{\sqrt{-I^2}} = \frac{I_U}{\sqrt{-I^2}} + I\frac{(-1/2)}{\sqrt{-I^2}^3}\big(-I_UI-II_U\big)$$ $$= \frac{I_U}{\sqrt{-I^2}} + \frac{I}{\sqrt{-I^2}^3}\big(\frac{I_UI+II_U}{2}\big)$$ $$= \frac{-I^2I_U + \frac12\big(II_UI+I^2I_U\big)}{\sqrt{-I^2}^3}$$ $$= \frac{\frac12\big(-I^2I_U+II_UI\big)}{-I^2\sqrt{-I^2}}$$ $$= \frac{\frac12I\big(II_U-I_UI\big)}{I^2\sqrt{-I^2}}$$ $$= \frac{\frac12I^{-1}\big(II_U-I_UI\big)}{\sqrt{-I^2}}$$ $$= \frac{I^{-1}\langle II_U\rangle_2}{\lVert I\rVert}$$ $$ii_U = \frac{I}{\lVert I\rVert}\frac{I^{-1}\langle II_U\rangle_2}{\lVert I\rVert} = \frac{\langle II_U\rangle_2}{\lVert I\rVert^2}$$ $$S(\vec x_U) = -ii_U = \frac{\langle I_UI\rangle_2}{\lVert I\rVert^2}$$

We also have $I = \vec x_U\wedge\vec x_V$, so

$$I_U = \vec x_{UU}\wedge\vec x_V + \vec x_U\wedge\vec x_{VU}$$ $$= (\vec x_{UU}^\parallel+\vec x_{UU}^\perp)\wedge\vec x_V + \vec x_U\wedge(\vec x_{VU}^\parallel+\vec x_{VU}^\perp)$$ $$= (\vec L\vec x_V - \vec M\vec x_U) + (\vec x_{UU}^\parallel\wedge\vec x_V + \vec x_U\wedge\vec x_{VU}^\parallel)$$

The terms on the right are tangent bivectors; their product with $I$ is a scalar, so it disappears in the grade-projection:

$$\langle I_UI\rangle_2 = \langle(\vec L\vec x_V - \vec M\vec x_U)I\rangle_2 + 0$$ $$\frac{\langle I_UI\rangle_2}{\lVert I\rVert^2} = \langle(\vec L\vec x_V - \vec M\vec x_U)\Big(\frac{I}{-I^2}\Big)\rangle_2$$ $$= \langle(\vec L\vec x_V - \vec M\vec x_U)(-I^{-1})\rangle_2$$ $$= \langle-\vec L(\vec x_VI^{-1}) - \vec M(-\vec x_UI^{-1})\rangle_2$$ $$= \langle-\vec L\vec x^U - \vec M\vec x^V\rangle_2$$ $$= -\vec L\vec x^U - \vec M\vec x^V = \vec x^U\vec L + \vec x^V\vec M$$

(The vectors are orthogonal, so their product is already a bivector.) This gives the formulas

$$S(\vec x_U) = \vec x^U\vec L + \vec x^V\vec M$$ $$S(\vec x_V) = \vec x^U\vec M + \vec x^V\vec N$$

(The second equation comes from swapping $U$ and $V$.) The shape is a linear function, so $S(a)$ for any tangent vector $a$ can be calculated from these. Now, the curvature $C$:

$$C(\vec x_U\wedge\vec x_V) = \frac{S(\vec x_U)S(\vec x_V)-S(\vec x_V)S(\vec x_U)}{2}$$ $$= \frac{\Big(-\vec x^U\vec x^U\Big)(\vec L\vec M-\vec M\vec L)+\Big(-\vec x^U\vec x^V\Big)(\vec L\vec N-\vec M\vec M)+\Big(-\vec x^V\vec x^U\Big)(\vec M\vec M-\vec N\vec L)+\Big(-\vec x^V\vec x^V\Big)(\vec M\vec N-\vec N\vec M)}{2}$$ $$= \Big(-\vec x^U\cdot\vec x^U\Big)(\vec L\wedge\vec M)+\frac12\Big(-\vec x^U\vec x^V\Big)(\vec L\cdot\vec N+\vec L\wedge\vec N-\vec M\cdot\vec M)+\frac12\Big(-\vec x^V\vec x^U\Big)(\vec M\cdot\vec M-\vec N\cdot\vec L-\vec N\wedge\vec L)+\Big(-\vec x^V\cdot\vec x^V\Big)(\vec M\wedge\vec N)$$ $$= \frac{-G}{\lVert I\rVert^2}\vec L\wedge\vec M-\frac{\vec x^U\vec x^V-\vec x^V\vec x^U}{2}(\vec L\cdot\vec N-\vec M\cdot\vec M)-\frac{\vec x^U\vec x^V+\vec x^V\vec x^U}{2}(\vec L\wedge\vec N)+\frac{-E}{\lVert I\rVert^2}\vec M\wedge\vec N$$ $$= \frac{-G}{\lVert I\rVert^2}\vec L\wedge\vec M-\vec x^U\wedge\vec x^V(\vec L\cdot\vec N-\vec M\cdot\vec M)-\vec x^U\cdot\vec x^V(\vec L\wedge\vec N)+\frac{-E}{\lVert I\rVert^2}\vec M\wedge\vec N$$ $$= \frac{-G}{\lVert I\rVert^2}\vec L\wedge\vec M-\frac{I}{\lVert I\rVert^2}(\vec L\cdot\vec N-\vec M\cdot\vec M)-\frac{-F}{\lVert I\rVert^2}(\vec L\wedge\vec N)+\frac{-E}{\lVert I\rVert^2}\vec M\wedge\vec N$$ $$= -\frac{G}{\lVert I\rVert^2}\vec L\wedge\vec M-IK-\frac{F}{\lVert I\rVert^2}(\vec N\wedge\vec L)-\frac{E}{\lVert I\rVert^2}\vec M\wedge\vec N$$ $$= -IK-\frac{\vec L\wedge\vec MG+\vec M\wedge\vec NE+\vec N\wedge\vec LF}{\lVert I\rVert^2}$$ $$= -IK+I^2\frac{\vec L\wedge\vec MG+\vec M\wedge\vec NE+\vec N\wedge\vec LF}{\lVert I\rVert^4}$$ $$= -IK+IQ = -I(K-Q) = -(K-Q)I = -(K-Q)(\vec x_U\wedge\vec x_V)$$

This, along with linearity of $S$ and $C$, proves the claim at the beginning of this answer.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.