This question grew out of a research project in which I am an active participant, a project which curiously enough requires a hefty dose of the theory of curves and surfaces in $\Bbb R^3$. I needn't go into more detail here.

One of my colleagues in said endeavor noticed that a previous answer of mine, given in response to a question by Helen Waters, presented results which were very illuminating to his own work. The question to which I refer is:

Representing a unit speed curve on a sphere in terms of its Frenet Frame.

In point of fact, similar questions have been posed more than once here on math.stackexchange.com; but this particular one became the object of my colleague's attention, discovered by googling around.

In the course of our work it became clear that a generalization of this question and its answer is also of significant importance to us. I pose the generalized question here, and present my answer below.

A unit speed curve $\alpha(s)$, where $s$ as usual denotes arc-length, which lies on a sphere of radius $r$ centered at a point $c \in \Bbb R^3$ satisfies an equation of the form

$(\alpha(s) - c) \cdot (\alpha(s) - c) = r^2; \tag 1$

if we differentiate this equation with respect to $s$ and recall that the unit tangent vector $T(s)$ to $\alpha(s)$ is given by

$T(s) = \dot \alpha(s), \tag 2$

we obtain

$T(s) \cdot (\alpha(s) - c) = 0; \tag 3$

successive differentiation of this formula with respect to $s$ yields the results of the answer to the cited question; the engaged reader my consult it to see the details.

As indicated in the title, I seek here to broaden the result given in the above link to more general, not-necessarily spherical surfaces. To this end we note that, just as a patch on the sphere of radius $r$ centered at $c \in \Bbb R^3$ may be represented as as a $2$-parameter vector function $\mathbf r(u, v) \in \Bbb R^3$ such that

$(\mathbf r(u, v) - c) \cdot (\mathbf r(u, v) - c) = r^2, \tag 4$

so a general surface patch $\mathcal S$ in $\Bbb R^3$ may be represented by a vector function $\mathbf r(u, v)$, but sans the constraint (4). When (4) no longer applies, we may consider the role of $\delta(s)$, where

$\delta^2(s) = \mathbf r(u(s), v(s)) \cdot \mathbf r(u(s), v(s)) = \alpha(s) \cdot \alpha(s) \tag 5$

is the squared magnitude of $\mathbf r(s)$, i.e., $\delta(s)$ is the distance of $\alpha(s)$ from the coordinate origin $O$; here $(u(s), v(s))$ is the path $\alpha(s)$ takes in terms of the patch coordinates $u$ and $v$.

Seen from this point of view, what I wish to ask becomes:

The Question: Given a surface patch $\mathcal S$ specified by a vector function $\mathbf r(u, v)$ of two parameters $u$ and $v$, and a unit speed curve $\alpha(s)$ with differentiable non-vanishing curvature and non-vanishing torsion in $\mathcal S$ , express

$\alpha(s) = \mathbf r(u(s), v(s)) \tag 6$

in terms of the Frenet Frame of $\alpha(s)$ and the function $\delta(s)$.


This looks pretty ugly to me, but here goes. We write $$\alpha = \lambda T + \mu N + \nu B \tag{$\star$}$$ for some scalar functions $\lambda, \mu, \nu$. From $\|\alpha\|^2 = \delta^2$, we get $\lambda = \alpha\cdot T = \delta\delta'$. On the other hand, as we've shown numerous times in posts to which you referred, we differentiate ($\star$) and use the Frenet equations to get $$ T = (\lambda'-\mu\kappa)T + (\lambda\kappa + \mu' - \nu\tau)N + ((\mu\tau+\nu')B,$$ and so \begin{align*} \lambda' &= 1 + \mu\kappa \\ \nu' &= -\mu\tau \\ \mu' &= \nu\tau - \lambda\kappa. \end{align*} This gives us \begin{align*} \mu &= \frac1{\kappa}(\delta\delta'-1) \\ \nu &= \frac1{\tau}(\mu'+\lambda\kappa) = \frac1{\tau}\left(\Big(\frac{(\delta\delta'-1}{\kappa}\Big)' + \delta\delta'\kappa\right). \end{align*} And the final equality gives the constraint ODE $$0=\nu'+\mu\tau = \left(\frac1{\tau}\left(\Big(\frac{(\delta\delta'-1}{\kappa}\Big)' + \delta\delta'\kappa\right)\right)' + \frac{\tau}{\kappa}(\delta\delta'-1).$$ Yuck.

  • $\begingroup$ I agree 'tis not the most elegant question ever posed, but thanks for answering, +1, endorsed! $\endgroup$ Nov 20 '19 at 17:18

My answer to the question I posted above is as follows:

We begin with the observation that the relation

$\alpha(s) \cdot \alpha(s) = \mathbf r(u(s), v(s)) \cdot \mathbf r(u(x), v(s)) = \delta^2(s) \tag 1$

may be differentiated with respect to $s$,

$2\alpha(s) \cdot \dot \alpha(s) = 2\mathbf r(s) \cdot \dot{\mathbf r}(s) = 2\delta \dot \delta, \tag 2$


$\dot \alpha(s) \cdot \alpha(s) = \dot{\mathbf r}(s) \cdot \mathbf r(s) = \delta \dot \delta; \tag 3$

here of course $\dot{\mathbf r}(s)$ is the total derivative of $\mathbf r$ with respect to $s$, i.e.

$\dot{\mathbf r}(s) = \mathbf r_u \dot u(s) + \mathbf r_v \dot v(s), \tag 4$


$\mathbf r_u = \dfrac{\partial \mathbf r}{\partial u}, \tag 5$

etc. Since

$\alpha(s) = \mathbf r(s) \tag 6$

is a unit-speed curve, we have the unit tangent vector to $\alpha(s)$,

$T(s) = \dot \alpha(s) = \dot{\mathbf r}(s); \tag 7$

in light of this we have from (3)

$T(s) \cdot \mathbf r(s) = \delta \delta_s, \tag 8$

which gives the component of $\mathbf r(s)$ along $T(s)$; we next differentiate (8) again with respect to $s$ to find

$\dot T(s) \cdot \mathbf r(s) + T(s) \cdot \dot{\mathbf r}(s) = \delta_s^2 + \delta \delta_{ss}, \tag 9$

which by virtue of (7) becomes

$\dot T(s) \cdot \mathbf r(s) + T(s) \cdot T(s) = \delta_s^2 + \delta \delta_{ss}; \tag{10}$

since $T(s)$ is a unit vector,

$\dot T(s) \cdot \mathbf r(s) + 1 = \delta_s^2 + \delta \delta_{ss}; \tag{11}$

next, the Frenet-Serret equation

$\dot T(s) = \kappa(s) N(s) \tag{12}$

may be used to transform (11) to

$\kappa(s) N(s) \cdot \mathbf r(s) + 1 = \delta_s^2 + \delta \delta_{ss}; \tag{13}$

since we have assumed that

$\kappa(s) \ne 0, \tag{14}$

we may re-arrange (13) into the form

$N(s) \cdot \mathbf r(s) = \dfrac{\delta_s^2 + \delta \delta_{ss} - 1}{\kappa(s)}, \tag{15}$

which gives the compoent of $\mathbf r(s)$ along $N(s)$; we may in fact continue in this direction, and take the $s$-derivative of (13):

$\dot \kappa(s) N(s) \cdot \mathbf r(s) + \kappa(s) \dot N(s) \cdot \mathbf r(s) + \kappa(s) N(s) \cdot \dot{\mathbf r}(s) = (\delta_s^2 + \delta \delta_{ss})_s; \tag {16}$

in accord with (7) we see that

$ N(s) \cdot \dot{\mathbf r}(s) = N(s) \cdot T(s) = 0, \tag{17}$


$\dot \kappa(s) N(s) \cdot \mathbf r(s) + \kappa(s) \dot N(s) \cdot \mathbf r(s) = (\delta_s^2 + \delta \delta_{ss})_s; \tag {18}$

as for the right-hand side of this equation,

$(\delta_s^2 + \delta \delta_{ss})_s = (\delta_s^2)_s + (\delta \delta_{ss})_s = 2\delta_s \delta_{ss} +\delta_s \delta_{ss} + \delta\delta_{sss} = 3\delta_s \delta_{ss} + \delta_s \delta_{sss}; \tag{19}$


$\dot \kappa(s) N(s) \cdot \mathbf r(s) + \kappa(s) \dot N(s) \cdot \mathbf r(s) = 3\delta_s \delta_{ss} + \delta_s \delta_{sss}; \tag {20}$

we now exploit the so-called second Frenet-Serret equation,

$\dot N(s) = -\kappa(s) T(s) + \tau(s) B(s) \tag{21}$

by substituting it into (20):

$\dot \kappa(s) N(s) \cdot \mathbf r(s) + \kappa(s) ( -\kappa(s) T(s) + \tau(s) B(s)) \cdot \mathbf r(s) = 3\delta_s \delta_{ss} + \delta_s \delta_{sss}, \tag {22}$


$\dot \kappa(s) N(s) \cdot \mathbf r(s) - \kappa^2(s) T(s) \cdot \mathbf r(s) + \kappa(s) \tau(s) B(s) \cdot \mathbf r(s) = 3\delta_s \delta_{ss} + \delta_s \delta_{sss}, \tag {23}$

in which the term containing $B(s) \cdot \mathbf r(s)$ may be isolated, thus:

$\kappa(s) \tau(s) B(s) \cdot \mathbf r(s) = \kappa^2(s) T(s) \cdot \mathbf r(s) -\dot \kappa(s) N(s) \cdot \mathbf r(s) + 3\delta_s \delta_{ss} + \delta_s \delta_{sss}, \tag {24}$

from which

$B(s) \cdot \mathbf r(s) = \dfrac{\kappa(s)}{\tau(s)} T(s) \cdot \mathbf r(s) - \dfrac{1}{\kappa(s) \tau(s)} \dot \kappa(s) N(s) \cdot \mathbf r(s) + \dfrac{1}{\kappa(s) \tau(s)} (3\delta_s \delta_{ss} + \delta_s \delta_{sss}); \tag {25}$

we can now use (8) and (15) to eliminate $T(s) \cdot \mathbf r(s)$ and $N(s) \cdot \mathbf r(s)$ from this expression, to find:

$B(s) \cdot \mathbf r(s) = \dfrac{\kappa(s)}{\tau(s)} \delta \delta_s - \dfrac{1}{\kappa(s) \tau(s)} \dot \kappa(s) (\dfrac{\delta_s^2 + \delta \delta_{ss} - 1}{\kappa(s)}) + \dfrac{1}{\kappa(s) \tau(s)} (3\delta_s \delta_{ss} + \delta_s \delta_{sss}). \tag {26}$

Now the expansion of the vector $\mathbf r(s)$ in terms of the Frenet Frame $T(s)$, $N(s)$, $B(s)$ is

$\mathbf r(s) = (\mathbf r(s) \cdot T(s)) T(s) + (\mathbf r(s) \cdot N(s))N(s) + (\mathbf r(s) \cdot B(s)) B(s); \tag{27}$

using (8), (15) and (26) we arrive at

$\mathbf r(s) = (\delta \delta_s) T(s) + \left (\dfrac{\delta_s^2 + \delta \delta_{ss} - 1}{\kappa(s)} \right )N(s) $ $+ \left ( \dfrac{\kappa(s)}{\tau(s)} \delta \delta_s - \dfrac{1}{\kappa(s) \tau(s)} \dot \kappa(s) \left (\dfrac{\delta_s^2 + \delta \delta_{ss} - 1}{\kappa(s)} \right ) + \dfrac{1}{\kappa(s) \tau(s)} (3\delta_s \delta_{ss} + \delta_s \delta_{sss}) \right ) B(s). \tag{28}$

By virtue of this equation we may express $\vert \mathbf r(s) \vert$, the distance of the point $\mathbf r(s)$ from the origin, as

$\vert \mathbf r(s) \vert^2 = \mathbf r(s) \cdot \mathbf r(s) = (\delta \delta_s)^2 + \left (\dfrac{\delta_s^2 + \delta \delta_{ss} - 1}{\kappa(s)} \right )^2$ $+ \left ( \dfrac{\kappa(s)}{\tau(s)} \delta \delta_s - \dfrac{1}{\kappa(s) \tau(s)} \dot \kappa(s) \left (\dfrac{\delta_s^2 + \delta \delta_{ss} - 1}{\kappa(s)} \right ) + \dfrac{1}{\kappa(s) \tau(s)} (3\delta_s \delta_{ss} + \delta_s \delta_{sss}) \right )^2; \tag{29}$

we observe that in the case

$\delta = \text{constant} \tag{30}$

all the derivatives

$\delta_s, \; \delta_{ss}, \; \delta_{sss} = 0, \tag{31}$

and (28) reverts to

$\mathbf r(s) = -\dfrac{1}{\kappa(s)} N(s) + \dfrac{ \dot \kappa(s)}{\kappa^2(s) \tau(s)} B(s), \tag{32}$

whilst (29) becomes

$\vert \mathbf r(s) \vert^2 = \dfrac{1}{\kappa^2(s)} + \dfrac{ (\dot \kappa(s))^2}{\kappa^4(s) \tau^2(s)}; \tag{33}$

formulas (32) and (33) agree with the answers I gave in https://math.stackexchange.com/questions/2504918/representing-a-unit-speed-curve-on-a-sphere-in-terms-of-its-frenet-frame which treats the case of constant $\delta = r$, that is, when the surface in which $\alpha(s)$ lies is a sphere.

In the above discussion, we have focused exclusively on surfaces given in parametric form $\mathbf r(u, v)$, and we relied extensively on the fact that for such surfaces the quantity $\delta = \vert \mathbf r(u, v) \vert$ is the Euclidean distance in $\Bbb R^3$ 'twixt the origin $O = (0, 0, 0)$ and the point $\mathbf r(u, v)$. However, it appears this question may also be addressed by representing surfaces implicitly, that is by specifying a (sufficiently smooth)

$f: \Omega \subset \Bbb R^3 \to \Bbb R \tag{34}$

$\Omega$ open, with

$\nabla f(x, y, z) \ne 0, \; (x, y, z) \in \Omega; \tag{35}$

since specifying a surface implicitly does not give us a direct handle on the distance of points from the origin, we might expect any formulas we derive along such lines to exhibit significant points of difference from (28)-(29). Finally, it is also possible to represent the $\mathbf r$ in polar form, that is, writing

$\mathbf r = r\mathbf e_r, r = \vert \mathbf r \vert = \delta \ne 0, \; \vert \mathbf e_r \vert = 1, \tag{36}$

and perhaps carrying out calculations similar to the above but using $\mathbf r$ in this form; e.g., (7) becomes

$T(s) = \dot \alpha(s) = \dfrac{d}{ds} (r(s)\mathbf e_r(s)) = \dot r(s) \mathbf e_r(s) + r(s) \dot{\mathbf e}_r(s); \tag{37}$

if this formulation is followed through, the contributions of changes in distance m$O$ ($r(s)$ direction $(\mathbf e_r(s)$ of $\alpha(s)$ to the coefficients of $T$, $N$, and $B$ may perhaps be quantified. But this post is quite long enough as is, so I will leave such undertakings for later.


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.