Getting started

Useful resources:

You need the "scmutils" package, here are installation instructions.

I also converted the html chapters and combined them into an unofficial pdf, which I find easier to read.

Comments about scmutils

(square v) works for vector v, giving \(|v|^2\). This is useful to simplify some code.

scmutils actually bundles a let& which can deconstruct vectors. Here is an example use case from the code

(define ((L m g) state)
  (let& (((_ (x y) v) state))
    (- (* 1/2 m (square v))
       (* m g y))))

Comments about sections

Section 1.9 - I still don’t understand the value of \(\bar \Gamma\) (Gamma-bar in scheme). The examples seem pretty contrived and although Sussman and Wisdom point out that their F->C function avoids derivatives, it feels rather that they get implicitly shoved into \(\Gamma\). It seems to only be used once in the book (in section 2.2), and even there it seems that you could rewrite those functions without \(\bar \Gamma\).

Section 1.10.1 - I don't follow the reasoning in the "Alternatively" section. It shows that if \(\lambda\) can be written as a state-dependent function, then a Lagrangian with \(\lambda\) added has the same Lagrange equations as an augmented Lagrangian. However, the last paragraph makes it seem like the converse was proven (if you have an augmented Lagrangian, then \(\lambda\) can be a state-dependent function).

Comments about the exercises

Exercise 1.25 - I setup the Foucault pendulum using the recommended transformations, but finding the Lagrange equations ran out of memory. The problem doesn't ask for Lagrange equations, so perhaps this is intended??

Exercise 1.30 - This may be trivial, but I got tripped up plotting time vs theta, whereas I should have been plotting \((r \cos(\theta), r \sin(\theta))\)

Exercise 1.31 - As specified for 1.25, finding Lagrange equations ran out of memory. Finding the state derivative also ran out of memory. I tried various settings and using numbers but could not get it to work.

Exercise 1.35 - The Lagrange equations obtained are really messy, largely due to the sqrt factors which are really \(r_0\) and \(r_1\). I found I needed to simplify the equations manually, and tune the simplify parameters to get (somewhat) readable equations.

If the equations are stored in the variable equations, then the code I used to simplify them was:

(sqrt-factor-simplify false)
(divide-out-terms-simplify #t)
(set! *factoring* #t)

(show-expression
 (simplify
  (substitute
   'r_0 '(sqrt (+ (expt a_0 2) (* 2 a_0 (x t))
                  (expt (y t) 2) (expt (x t) 2)))
   (substitute
    'r_1 '(sqrt (+ (expt a_1 2) (* -2 a_1 (x t))
                   (expt (y t) 2) (expt (x t) 2)))
    (simplify equations)
   ))
 )
)

When I did it, a lot of terms actually simplified to just \(m D^2 x\) or \(m D^2 y\) (for \(x\) or \(y\) equation, respectively).

Exercise 1.36 - This problem is a lot easier (and makes more sense) if you specify the angular coordinates as \((\theta, \phi)\) where \(\theta\) is the colatitude for the x-axis and \(\phi\) is the longitude measured from the y-axis. As rotating \(\phi\) is much easier than rotating \(\theta\) is spherical coordinates.

Exercise 1.38 - Part d should really be \(\mathscr{E}[F \circ C] = D_t (DF \circ C) \partial_2 C + (DF \circ C) \mathscr{E}[C]\)