12

Describe all curves in $\mathbb{R}^3$ which have constant curvature $κ > 0$ and constant torsion $τ$.

Any ideas what we can do to describe all such curves?

Do we have to use the formulas of the curvature and of the torsion?

Daniel Fischer
  • 206,697
  • 4
    You can integrate the Frenet–Serret equations directly, which gives you the equation of the tangent vector up to a rotation, and therefore the curve up to an initial position (mind you, I didn't say you should do this...). – Chappers Oct 07 '15 at 23:29
  • 1
    The Frenet-Serret equations are:

    $$\textbf{$\dot t$}=k\textbf{n} \ \textbf{$\dot n$}=-k\textbf{t}+\tau \textbf{b} \ \textbf{$\dot b$}=-\tau \textbf{n}$$

    How can we integrate these equations? @Chappers

    –  Oct 08 '15 at 17:48
  • You can write them as a matrix differential equation for the matrix $( \mathbf{t,n,b} )^T$, and then integrate in the same way as a matrix DE acting on a vector: you end up with arbitrary vectors at $t=0$ instead of just one arbitrary vector (although you have to fulfil orthogonality, of course). You need the matrix exponential, but constant curvature and torsion make the DE much easier to solve than usual. – Chappers Oct 08 '15 at 18:48
  • 2
    Why was my three-year-old question marked as a duplicate of this two-month-old one? – Shay Guy Dec 20 '15 at 18:49

3 Answers3

10

As had been explained, you may choose to either solve the differential equation given by Frenet's identities in the space of matrices, or do the smart trick suggested by Ted Shiffrin; I shall choose the latter approach.

For my typing convenience, I shall also omit all the "mathhb" formatting, therefore the normal will just be $n$. Also, since $t$ is the tangent, I shall use $s$ for the time parameter on the curve.

Starting with $\ddot n = - (k^2 + \tau ^2)n$, we shall approach it in the usual way: first we consider its associated algebraic equation $\lambda ^2 = -(k^2 + \tau ^2)$, which has the roots $\pm \Bbb i \sqrt{k^2 + \tau ^2}$. Next, we know from the general theory of linear equations that this gives two fundamental solutions to the above equation, namely $u \sin rs$ and $v \cos rs$ where $r^2 = k^2 + \tau ^2$ and $u,v \in \Bbb R ^3$ (note that $r \ne 0$ because $k>0$ by assumption). Therefore, $n(s) = u \sin rs + v \cos rs$ for some constant vectors $u,v \in \Bbb R ^3$.

May $u,v \in \Bbb R ^3$ be really arbitrary? Not quite: note that $n(0) = v$ and since $\| n (s) \| = 1 \ \forall s$ (the curve is parametrized by arc-length and the Frenet frame is taken to be orthonormal), then $\| v \| = 1$. Next, if you derive this once, you get $\dot n (0) = ru$; on the other hand, $\dot n (0) = -k t(0) + \tau b (0)$, so $r u = -k t(0) + \tau b (0)$, so $r^2 \| u \| ^2 = \| -k t(0) + \tau b (0) \| ^2 = k^2 \| t (0) \| ^2 - 2 k \tau \langle t(0), b(0) \rangle + \tau ^2 \| v \| ^2$. Since $\| t \| = \| b \| = 1$ and $\langle t, b \rangle = 0$, the last expression is exactly $k^2 + \tau ^2 = r^2$, which implies $\| u \| = 1$. Finally, $\| n \| ^2 = 1$ implies $\| u \| ^2 \sin ^2 rs + 2 \sin rs \cos rs \langle u, v \rangle + \| v \| ^2 \cos ^2 rs = 1$ which, taking into consideration all of the above, implies $\langle u, v \rangle = 0$. To conclude, $u$ and $v$ must be orthogonal, and of length $1$.

Next, if your curve is $s \mapsto x(s) \in \Bbb R ^3$ ($x$ is a vector, not the first coordinate of a vector!), then using the equation $\dot t = k n$ and the fact that $t = \dot x$, you will get $\ddot x = kn = k (u \sin rs + v \cos rs)$, that you will have to integrate twice. Integrating once gives $\dot x = -\dfrac k r u \cos rs + \dfrac k r v \sin rs + w$, with $w \in \Bbb R ^3$, and integrating once more gives $x = - \dfrac k {r^2} u \sin rs - \dfrac k {r^2} v \cos rs + ws + x_0$, with $x_0 \in \Bbb R^3$.

Similarly to the discussion about $u$ and $v$, may $w$ be arbitrary? No, it may not: first, $\langle t, n \rangle = 0$ implies $\langle w, u \sin rs + v \cos rs \rangle = 0 \ \forall s$, which implies that $w \perp \text{span} \{u, v \}$. Second, using this and the fact that $\| \dot x \| ^2 = \| t \| ^2 = 1$ implies that $\| w \| = \dfrac {| \tau |} r$. There is no constraint, on the other hand, on $x_0$.

If $\tau = 0$ (so your curve is a plane curve), after a possible orthogonal transformation you may take $u = (-1, 0, 0)$ and $v = (0, -1, 0)$, so your curve will look like $x(s) = (\dfrac k {r^2} \sin rs, \dfrac k {r^2} \cos rs, 0) + x_0$ which, after a translation by $x_0$ will have the final nice form $x(s) = (\dfrac k {r^2} \sin rs, \dfrac k {r^2} \cos rs, 0)$, so it will be a circle.

If $\tau \ne 0$, since $\{ -u, -v, w \frac r {\ |\tau| }\}$ have been shown to form an orthonormal frame (in fact, we have worked with $u$ and $v$, but absorbing a minus sign into them doesn't change orthonormality), after a possible orthogonal transformation you may take them to form the familiar frame $\{ (1, 0, 0), (0, 1, 0), (0, 0, 1) \}$, in which case you curve looks like $x(s) = (\dfrac k {r^2} \sin rs, \dfrac k {r^2} \cos rs, \frac {| \tau |} r s) + x_0$. If you also translate your curve by $x_0$ (again, this doesn't change your curve), it will have the final nice form $x(s) = (\dfrac k {r^2} \sin rs, \dfrac k {r^2} \cos rs, s\dfrac { |\tau | } r)$, which is a helix. Note that this equation reduces to the one of the circle for $\tau \to 0$.

Therefore, the only curves that satisfy your problem are circles and helices. (Had you allowed $k=0$ too, you would have also obtained (segments of) straight lines).

Alex M.
  • 35,207
  • We have $$t=\dot x = -\dfrac k r u \cos rs + \dfrac k r v \sin rs + w $$

    $$n(s) = u \sin rs + v \cos rs $$

    Which is the formula of $$\langle t, n \rangle $$ ?

    –  Oct 25 '15 at 15:52
  • @user159870: I am not sure that I understand your question properly: $\langle t, n \rangle = 0$ trivially by the definition of $n$. – Alex M. Oct 25 '15 at 20:43
  • I understood this part. But I am facing some difficulties understanding the following part: $$$$ "If $\tau = 0$ (so your curve is a plane curve), after a possible orthogonal transformation you may take $u = (-1, 0, 0)$ and $v = (0, -1, 0)$, so your curve will look like $x(s) = (\dfrac k {r^2} \sin rs, \dfrac k {r^2} \cos rs, 0) + x_0$ which, after a translation by $x_0$ will have the final nice form $x(s) = (\dfrac k {r^2} \sin rs, \dfrac k {r^2} \cos rs, 0)$, so it will be a circle." $$$$ What do you mean by "after a translation by $x_0$" ? Do we set $x_0=0$ ? –  Oct 26 '15 at 21:47
  • @user159870: Exactly, if you want your result to look nicer you may define a new (translated) curve $\bar x = x - x_0$. Note that this wouldn't change anything in the computations above, because derivatives would just kill $x_0$, so $\dot {\bar x} = \dot x$. – Alex M. Oct 26 '15 at 21:52
  • I understand. Do we take these $u$ and $v$ so that they satisfy the condition that they are orthonormal? –  Oct 26 '15 at 21:57
  • @user159870: Yes, we have already proven that they are arthogonal and of length $1$, so what better choice than that one? (The minus sign is there only to get a nice final result, otherwise it plays no role.) – Alex M. Oct 26 '15 at 22:00
  • I understand. Why does $\langle w, u \sin rs + v \cos rs \rangle = 0 \ \forall s $ imply that $w \perp \text{span} {u, v } $? –  Oct 26 '15 at 23:02
  • Also do we have to determine at the end which the curvature and the torsion is at each case? –  Oct 26 '15 at 23:38
  • @user159870: Take $s=0$ in that equality; this will produce $\langle w, v \rangle = 0$, so $w \perp v$. Next, using this, that equality simplifies to $\langle w, u \rangle \sin rs = 0 \ \forall s$ so now choose some $s$ with $\sin rs \ne 0$ in order to get $\langle w, u \rangle = 0$, so $w \perp u$. Since $w \perp v$ and $w \perp u$, it is immediate that $w \perp \text{span} {u, v}$. It is not necessary to compute the curvature and the torsion at the end, but you may do it using the standard formulae if you feel like. – Alex M. Oct 27 '15 at 10:32
  • To compute the curvature do we use the formula $$\kappa=|\ddot x|$$ ? But how can we do that when at the formula of $x$ we have $\kappa$ ? –  Nov 04 '15 at 22:37
  • @AlexM. Shouldn't it be $|w| = \frac{|k|}{r}$? – Aaron Maroja Nov 18 '15 at 14:56
  • @AaronMaroja: No, it's $\tau$, but your comment made me reread my answer and find some typographical errors in the 4th paragraph. Anyway, norm-squaring $\dot x = - \dfrac k r u \cos rs + \dfrac k r v \sin rs + w$ gives $1 = \dfrac {k^2} {r^2} \cos^2 rs + \dfrac {k^2} {r^2} \sin^2 rs + | w | ^2 = \dfrac {k^2} {r^2} + | w | ^2$ and now use $r^2 = k^2 + \tau ^2$. – Alex M. Nov 18 '15 at 16:51
  • a small unimportant question: why do you write $|\tau|$ and not $\tau$? Isn't $\tau$ already a number? – Whyka Nov 12 '16 at 21:10
  • @Whyka: $|\tau|$ is the absolute value of $\tau$. – Alex M. Nov 12 '16 at 21:12
  • @AlexM. Ah, okay. So you want to take the positive value of $\tau$. Thanks! – Whyka Nov 12 '16 at 21:51
8

HINT: If you don't want to solve the matrix differential equation, I recommend that you get a second-order differential equation for the principal normal $n$. You should know solutions of this by sight (at least, component by component).

Ted Shifrin
  • 115,160
  • 1
    We have $$\textbf{$\dot t$}=k\textbf{n} \ \textbf{$\dot n$}=-k\textbf{t}+\tau \textbf{b} \ \textbf{$\dot b$}=-\tau \textbf{n}$$

    Differentiating the second equations we get $$\textbf{$\ddot n$}=-k\textbf{$\dot t$}+\tau \textbf{$\dot b$}\Rightarrow \textbf{$\ddot n$}=-k^2\textbf{n}-\tau^2 \textbf{n} \Rightarrow \textbf{$\ddot n$}=-(k^2+\tau^2) \textbf{n}$$ What do we do next?

    –  Oct 13 '15 at 16:08
  • 6
    @user159870, don't you know the general solution of $y''+\lambda^2 y=0$? – Ted Shifrin Oct 13 '15 at 20:56
3

From the Fresnet-Serret formulas, we can derive

$$\ddddot\gamma(s)=\kappa\ddot n(s)=-\kappa(\kappa^2+\tau^2)n(s)=-a^2\ddot\gamma(s),$$ with $a=\sqrt{\kappa^2+\tau^2}$.

The general solution for this fourh order linear equation is $$\color{green}{\gamma(s)=C\cos(as)+S\sin(as)+C_1s+C_0},$$ which describes an elliptic helix.

Anyway, computing the curvature for any $s$ (such as $s=0,s=\dfrac\pi{2a}$),

$$\kappa=\|\dot t(s)\|=\|\ddot\gamma(s)\|=a^2\|C\cos(as)+S\sin(as)\|,$$

we must have $$\|C\|=\|S\|=\frac\kappa{a^2}\text{, and }C\cdot S=0$$ for the curvature to be constant (the normal vector describes a circle).

Similarly,

$$\kappa\tau=\|\kappa\dot n(s)+\kappa^2t(s)\|=\|\ddot t(s)+\kappa^2t(s)\|=\|-a\tau^2(C\cos(as)+S\sin(as))+\kappa^2C_1\|$$ requires $$C\cdot C_1=S\cdot C_1=0\text{, and }\|C_1\|=\frac{\tau\kappa}a$$ (the tangent vectors describes a circular cone).

Hence the curve is a circular helix.

  • Why do we require the following? $$$$

    $$\kappa=|\dot t(s)|=|\ddot\gamma(s)|=a^2|C\cos(as)+S\sin(as)|,$$

    we must have $$C\cdot S=0$$ for the curvature to be constant.

    Why does this stand: "the normal vector describes a circle" ?

    $$\kappa\tau=|\kappa\dot n(s)+\kappa^2t(s)|=|\ddot t(s)+\kappa^2t(s)|=|-a\tau^2(C\cos(as)+S\sin(as))+\kappa^2C_1|$$ requires $$C\cdot C_1=S\cdot C_1=0\text{, and }|C_1|=\frac{\tau\kappa}a$$

    Why does this stand: "the tangent vectors describes a circular cone" ?

    –  Oct 23 '15 at 00:05
  • @user159870 We must ensure that the curvature and torsion are constant ! (The geometric interpretations are just hints, not part of the proof.) –  Oct 23 '15 at 07:52
  • The curve is a circular helix or a circle (the problem allows for null torsion). (My remark is moot if you consider the circle to be a particular case of a helix.) – Alex M. Oct 25 '15 at 20:40
  • 1
    @AlexM: yes, a circle is an helix in flatland. –  Oct 26 '15 at 07:17