6

Here's what I've got so far (and I'm assuming $\alpha$ is a unit speed curve):

a) The fact that $\beta(s) = \alpha(s) + r(s)N(s)$ for some scalar function $r$ follows trivially because of the fact that the normal lines of $\beta$ and $\alpha$ are equal (so $\beta(s)$ is a point on the normal line of $\alpha(s)$, which is precisely what the equality states). Then:

$$\ r(s) = (\beta(s) - \alpha(s))N(s) \Rightarrow r'(s) = (\beta'(s) - \alpha'(s))N(s) + (\beta(s) - \alpha(s))N'(s) $$

$$r'(s) = (\beta'(s) -\alpha'(s))N(s) + r(s)N(s)N'(s) = (\beta'(s) - \alpha'(s))N(s)= \beta'(s)N(s) = 0$$

as desired.

b) Let $\theta$ be the angle between the unit tangent vectors mentioned. Then: $$\cos(\theta) = \frac{\beta'(s)\alpha'(s)}{||\beta'(s)||} = \frac{1-r(s)k(s)}{||\beta'(s)||}$$

but $$||\beta'(s)||^2 = ||\alpha'(s)||^2 + r^2(s)||N(s)||^2 = 1 + r^2(s)$$ so $$||\beta'(s)|| = \sqrt{1+r^2(s)}$$ $$\cos(\theta) = \frac{1-r(s)k(s)}{\sqrt{1+r^2(s)}}$$

EDIT: The formula for $||\beta'(s)||^2$ above is wrong. Actually, $||\beta'(s)||^2 = ||\alpha'(s)||^2 + r^2(s)||N'(s)||^2$, which is not as neat as the above. But the hint in the comments is much better to solve this.

c) Assuming I had done b), then the following would be true: $$(\cos(\theta))' = \left( \frac{\beta'(s)\alpha'(s)}{||\beta'(s)||} \right)' = 0 \Rightarrow (\beta'(s)\alpha'(s))' = 0 \Rightarrow \beta''(s)\alpha'(s) + \beta'(s)\alpha''(s) = 0$$

but continuing on this path only leads to $k'(s) = 0$, which, even if I had done b), would not be useful (and I'm aware using this to prove b) is circular).

Beyond these (b) and c)), I would also like some help solving d) and $21$ (I don't know where to start on those).

Update: A lot of what I wrote above is wrong, but my doubts here are almost completely solved. I just have to finish $21$ and I'll be done.

Update 2: For $20$, see the discussion in comments. For $21$:

Since $T_\beta$ is orthogonal to $N_\alpha$, we can write it as a linear combination of $T_\alpha$ and $B_\alpha$. In particular: $$T_\beta = \pm (\cos(\theta)T_\alpha + \sin(\theta)B_\alpha)$$ where $\theta$ is the angle between the unit tangent vectors of $\alpha$ and $\beta$. Then: $$B_\beta = T_\beta \times N_\beta = \pm T_\beta \times N_\alpha = \pm (\cos(\theta)T_\alpha + \sin(\theta)B_\alpha) N_\alpha= \pm(\cos(\theta)B_\alpha - \sin(\theta)T_\alpha)$$ Differentiating and using the fact that $\cos(\theta)$ and $\sin(\theta)$ are constant, we have: $$B_\beta' = \pm(\cos(\theta)B_\alpha' -\sin(\theta) T_\alpha') = \pm(-\cos(\theta)\tau_\alpha N_\alpha - \sin(\theta)\kappa_\alpha N_\alpha)$$ $$B_\beta' = \pm(-\cos(\theta)\tau_\alpha - \sin(\theta)\kappa_\alpha)N_\alpha$$

By the Frenet frame, we also have: $$B_\beta' = -\tau_\beta N_\beta = \pm(-\tau_\beta N_\alpha) $$

So:

$$\pm(-\cos(\theta)\tau_\alpha - \sin(\theta)\kappa_\alpha)N_\alpha = \pm(-\tau_\beta N_\alpha) \Rightarrow (\cos(\theta)\tau_\alpha + \sin(\theta)\kappa_\alpha) = \pm \tau_b $$

By previous work:

$$\cot(\theta) = \frac{1-r(s)\kappa_\alpha(s)}{r(s) \tau_\alpha(s)}$$

which is equivalent to: $$\kappa_\alpha(s) + \cot(\theta) \tau_\alpha = \frac{1}{r(s)} \Rightarrow \kappa_\alpha(s) \sin(\theta) + \tau_\alpha(s) \cos(\theta) = \frac{\sin(\theta)}{r(s)}$$

Then:

$$\pm (\cos(\theta)\tau_\alpha + \sin(\theta)\kappa_\alpha) = \pm \frac{\sin(\theta)}{r(s)} = \tau_\beta(s)$$

$\bar{s}$ being the arclength paramater, we also have:

$$T_\beta = \frac{d \beta}{ds} \frac{ds}{d\bar{s}}$$

so:

$$T_\beta = \left( ( 1 - r(s) \kappa_\alpha(s))T_\alpha + r(s) \tau_\alpha(s) b_\alpha(s) \right ) \frac{ds}{d\bar{s}}$$

comparing the very first expression for the coefficients, we see that: $$\pm \sin(\theta) = r(s) \tau_\alpha(s) \frac{ds}{d \bar{s}}$$

so that:

$$\frac{d \bar{s}}{ds} = \frac{r(s)\tau_\alpha(s)}{\sin(\theta)}$$

and then, by previous work:

$$\pm \frac{\tau_\alpha(s) \tau_\beta(s) r(s)}{\sin(\theta)} = \pm \frac{\sin(\theta)}{r(s)} \Rightarrow \tau_\alpha(s) \tau_\beta(s) = \frac{\sin^2{\theta}}{r^2(s)}$$

as desired.

  • Looks familiar. You don't have the correct formula for $\beta'(s)$, of course. Your approach to (b) will work once you fix that, but then you're really doing (c) while you're at it. For (b), differentiate $T_\alpha\cdot T_\beta$ "abstractly." – Ted Shifrin Feb 19 '18 at 15:26
  • @TedShifrin how so? Using a), I get $\beta'(s) = \alpha'(s) + r(s)N'(s)$. I don't really understand how b) helps me with c) (thanks for the last hint, differentiating $T_\alpha \cdot T_\beta$ "abstractly" immediately gives the desired result). – Matheus Andrade Feb 19 '18 at 15:54
  • @TedShifrin Oh, I think you meant the formula for $||\beta'(s)||$. I edited it now. Though I still don't get how to start doing the other ones, starting with c). – Matheus Andrade Feb 19 '18 at 16:42
  • So, what is $N'(s)$? Having $|\beta'(s)|$ correct will make (c) more plausible. – Ted Shifrin Feb 19 '18 at 17:03
  • @TedShifrin You are right, as I said in my last comment I fixed that, so b) is done. $N'(s) = -k(s) \alpha'(s) + \tau(s) b(s)$. So $\beta'(s) = \alpha'(s) + r(s)(-k(s) \alpha'(s) + \tau(s) b(s))$ What I'm not getting is how this helps me solve c)... how do I interpret that result? – Matheus Andrade Feb 19 '18 at 17:10
  • Look at the correct formula for $\cos\theta$. – Ted Shifrin Feb 19 '18 at 17:37
  • $$\cos(\theta) = T_\alpha \cdot T_\beta= \frac{\beta'(s)\alpha'(s)}{||\beta'(s)||}$$, and $$||\beta'(s)||^2 = 1 + r^2(k^2 + \tau^2)$$. How do I proceed here? – Matheus Andrade Feb 19 '18 at 17:43
  • Too careless! Pay careful attention! – Ted Shifrin Feb 19 '18 at 20:24
  • @TedShifrin I've looked at this for hours now and I honestly can't see where my computations are going wrong. Could you please point out my mistakes? As far as I'm aware, if $\theta$ is the angle between two vectors $u$ and $v$, then $$cos(\theta) = \frac{u \cdot v}{||u|| ||v||}$$. In my last comment I just normalized $\beta'(s)$ to get $T_\beta$ and used that formula. For computing $||\beta'(s)||^2$ I just used the fact that $\alpha'$ and $N$ are orthogonal. I would really appreciate some help here... I'm sure it must be something simple, but I've tried and tried but can't see the mistake. – Matheus Andrade Feb 19 '18 at 21:15
  • You have $\beta'$ as a linear combination of $T$ (i.e., $\alpha'$) and $B$, but you did not collect terms!! You should end up with a rather nice right triangle corresponding to this angle $\theta$. – Ted Shifrin Feb 19 '18 at 22:27
  • @TedShifrin, I get $\beta' = \alpha'(s)(1-r(s)k(s)) + r(s)\tau(s)b(s)$. I'm not really seeing how this helps me, though. Where do I go now? – Matheus Andrade Feb 19 '18 at 23:27
  • Draw your right triangle with $\beta’$ as the hypotenuse. – Ted Shifrin Feb 20 '18 at 00:20
  • @TedShifrin Alright. Using the pythagorean theorem I get $$||\beta'(s)||^2 = ||\alpha'(s)(1 - r(s)k(s)) + r(s)\tau(s) b(s) ||^2 = (1-r(s)k(s))^2 + r^2(s)\tau^2(s) = 1 - 2r(s)k(s) + r^2(s) k^2(s) + r^2(s)\tau^2(s) = 1 - 2r(s)k(s) + r^2(s)(k^2(s) + \tau^2(s))$$. How can I use this? – Matheus Andrade Feb 20 '18 at 00:29
  • Don't expand it out. Don't you recognize the $(1-r\kappa)$ from the numerator of your cosine expression earlier in your work? Remember that $r$ is constant, too! – Ted Shifrin Feb 20 '18 at 00:36
  • @TedShifrin I see that $$\cos^2(\theta) = \frac{(1-rk)^2}{(1-rk)^2 + (r \tau)^2}$$. and I know that this is constant, so it must be the case that $(1-rk)^2 = c((1-rk)^2 + (r \tau)^2)$ for some constant $c$. Is this a right approach? – Matheus Andrade Feb 20 '18 at 00:39
  • This will, in fact, work, but think instead about $\tan\theta$. – Ted Shifrin Feb 20 '18 at 00:42
  • @TedShifrin to get $\tan(\theta)$ I can just differentiate the original expression for $\cos(\theta)$ (and multiply by $-1$, of course) to get $\sin(\theta)$ and then divide one by the other, right? – Matheus Andrade Feb 20 '18 at 00:45
  • YIKES!! No. Pretend you're in high school and draw a darned right triangle! – Ted Shifrin Feb 20 '18 at 00:46
  • @TedShifrin, thanks, that should have been obvious. Drawing the right triangle with hypotenuse $||\beta'(s)||$ and sides $1-rk$ and $r \tau$, I get $$\tan(\theta) = \frac{r(s) \tau(s)}{1-r(s)k(s)}$$, and this must be constant as well, so there exists a constant $c$ such that $r(s) \tau(s) = c(1-r(s)k(s))$, and the desired result follows. Thank you! To prove the other direction can I take the formula from a) and show that it is a Bertrand mate? – Matheus Andrade Feb 20 '18 at 01:06
  • That's the idea. I think you sometimes make things too complicated and lose the main ideas ... Most of the exercises do require attention to detail, but should be reasonably natural. – Ted Shifrin Feb 20 '18 at 01:10
  • Thanks! Can you point me on the right direction on d) and $21$? You've been of huge help already, I know, but I would appreciate that a lot as well. About the last part... (next comment) – Matheus Andrade Feb 20 '18 at 01:15
  • @TedShifrin I do have a terrible habit of complicating things, yes... I don't know why, but usually the most complicated ideas come to me first (maybe because I spend more time on them than the others?). I think I would do better if I was actually enrolled in a course on the subject and had a regular teacher, but because of the pre-requisites in my uni I won't be able to until $2019$, so in the meantime I'm just going through these books because I love the subject and it will give me a good advantage. – Matheus Andrade Feb 20 '18 at 01:16
  • As you should expect, use (c) to see what happens if you have more than one Bertrand mate. ... I'm going to let you try 21 using everything you've learned in doing 20. Just one little hint: How are the Frenet frames of $\alpha$ and $\beta$ related, explicitly? – Ted Shifrin Feb 20 '18 at 01:24
  • @TedShifrin Is it true that there will be infinitely many constants $a$ and $c$ such that $ak + c\tau = 1$ (and if so, is this a right approach)? The Frenet frames of $\alpha$ and $\beta$ are related by $N_\alpha = p N_\beta$ for some constant $p$ (since the normal lines are equal), yes? – Matheus Andrade Feb 20 '18 at 01:29
  • Wait. Different Bertrand mates certainly give different $r$ values. So what happens if you have two different Bertrand mates? ... Yes, $N_\alpha = \pm N_\beta$, but I want the whole frames!! – Ted Shifrin Feb 20 '18 at 01:37
  • @TedShifrin All I can see currently is that two different bertrand mates lead to two different values of $r$, but I can't see where to go further. For the whole frames, I get: $$T'\beta = \pm k\beta N_\alpha$$ $$N'\beta = -k\beta T_\beta + \tau_\beta b_\beta$$ $$b'\beta = \tau\beta N_\beta = \tau \cdot \pm N_\alpha $$ – Matheus Andrade Feb 20 '18 at 01:45
  • No, you're not interpreting everything you've learned in this problem. What is the meaning of $\theta$? I want you to relate the two $TNB$ frames to one another. ... Write down precisely what you know if there are two different Bertrand mates. You have two equations $r\kappa + c\tau = 1$. But what does that tell you about $\kappa$ and $\tau$? – Ted Shifrin Feb 20 '18 at 01:57
  • Interpreting the result I get from c): if $\beta$ is a bertrand mate to $\alpha$, then there exist constants $a$ and $b$ such that $a \kappa + c \tau = 1$. There being another bertrand mate suggests the existence of different constants that satisfy that equation. $\theta$ is the angle between the unit tangent vectors of both curves. At the moment I can't think of anything else that relates the two $TNB$ frames besides using the fact the normal lines are equal, which I stated already. – Matheus Andrade Feb 20 '18 at 02:35
  • @TedShifrin it's getting pretty late now (at least here) so I'm going to sleep and work on this tomorrow. Would it be too much to ask that if I haven't commented here 24 hours from now with some further insight you get back to me here as soon as you can (any time would be great, really) with some other hint (if I could tag you, that'd be awesome too)? If it is I'll keep trying. You were a huge help already, so thanks a lot for that. – Matheus Andrade Feb 20 '18 at 02:37
  • Alright, using my previous work here's where I got on d): If $\beta$ is a bertrand mate to $\alpha$, then it must be true that $1 - r(s)k(s) = a_1(r(s)\tau(s))$ and that $1 - r(s)k(s) = a_2(r(s)\tau(s))$ for some constants $a_1$ and $a_2$ (and $a_1 \neq a_2$. Differentiating we have $-r(s)k'(s) = a_1(r(s) \tau'(s)) = a_2(r(s)\tau'(s))$, which implies that both the curvature and torsion are constant, the desired result (although I think I should note here that I used the fundamental theorem of curves here). – Matheus Andrade Feb 20 '18 at 13:30
  • @TedShifrin what's still giving me trouble is $21$... I'll work on it some more, but if you could make the implication of that previous hint (relating the $TNB$ frames) a little clearer that'd be great. – Matheus Andrade Feb 20 '18 at 13:32
  • You can get 20(d) just from basic linear algebra. You don't need any differentiation. But of course you need the characterization of circular helices as the only curves with constant $\kappa$ and $\tau$. (That's actually an earlier exercise, not using the Fundamental Theorem.) ... If $\theta$ is the (constant) angle between $T_\alpha$ and $T_\beta$, you can write an explicit formula for $T_\beta$ as a linear combination of $T_\alpha$ and $B_\alpha$, and similarly for $B_\beta$. – Ted Shifrin Feb 20 '18 at 18:19
  • @TedShifrin Thank you! After some hours working on it, I'm almost sure I've solved $21$. I'm aware it's kind of a mess, but if you could check my work on the now edited post so that I could post a definitive answer to the question and get some temporary peace of mind, I'd really appreciate that a lot. – Matheus Andrade Feb 20 '18 at 19:06
  • 2
    It looks good, @Matheus. My only comment (which I've made before) is that you shouldn't keep writing $r(s)$ when $r$ is in fact constant!! :) This is particularly important when you get to the punchline and want $\tau_\alpha\tau_\beta$ to be constant. Well done. – Ted Shifrin Feb 21 '18 at 00:35
  • 1
    @TedShifrin I promise I'll remember that now!! Thanks for helping me improve! :) – Matheus Andrade Feb 21 '18 at 00:47

1 Answers1

3

Asked and answered (see updates). :)