previous overview next
Trigonometric Connection I
$
\def \half {\frac{1}{2}}
\def \kwart {\frac{1}{4}}
\def \EN {\quad \mbox{and} \quad}
\def \OF {\quad \mbox{or} \quad}
\def \slechts {\quad \Longleftrightarrow \quad}
$
A bit of elementary trigonometry is needed in order to acquire further
knowledge. Start with:
$$
\cos(2.\phi) = 2.\cos^2(\phi) - 1
$$
Solve for $\cos(\phi)$, divide $\phi$ by two and take care of the signs:
$$
\cos(\half \phi) = \sqrt{ \half + \half \cos(\phi)}
\quad \mbox{for} \quad 0 \le \phi \le \pi
$$
Augmented with:
$$
\cos(\pi - \phi) = - \cos(\phi)
$$
But suppose we are rather interested in the function $D(h)=2+2.\cos(\pi.h)$,
restricted to the range $0 \le h \le 1$. Rewrite the above formula as such:
$$
2 + 2.\cos(\pi.\half h) = 2 + \sqrt{ 2 + 2.\cos(\pi.h)}
\quad \mbox{for} \quad 0 \le h \le 1
$$
It follows that:
$$
D(\half h) = 2 + \sqrt{D(h)}
\quad \mbox{for} \quad 0 \le h \le 1
$$
Augmented with:
$$
2 + 2.\cos\left[\pi(1 - h) \right] = 2 - 2.\cos(\pi.h)
\quad \OF \quad D(1 - h) = 4 - D(h)
$$
The latter formula can also be implemented in a more "symmetric" way,
as opposed to $D(h/2) = 2 + \sqrt{D(h)}$ :
$$
D(1 - \half h) = 4 - D(\half h) = 2 - \sqrt{D(h)}
\quad \mbox{for} \quad 0 \le h \le 1
$$
Elementary values are:
$$
D(0) = 4 \qquad D(\half) = 2 \qquad D(1) = 0
$$
Start with $h = 1/2$ then:
$$
D(\kwart) = 2 + \sqrt{D(\half)} = 2 + \sqrt{2}
\qquad
D(\frac{3}{4}) = 4 - D(\kwart) = 2 - \sqrt{2}
$$
Let's try now for fractions $\times 1/8$. The values D(0/8), $D(2/8)$, D(4/8),
D(6/8) and D(8/8) have already been calculated. Go for the rest:
$$
D(\frac{1}{8}) = 2 + \sqrt{D(\kwart)} = 2 + \sqrt{2 + \sqrt{2}}
\qquad
D(\frac{7}{8}) = 4 - \sqrt{D(\frac{1}{8}} = 2 - \sqrt{2 + \sqrt{2}}
$$ $$
D(\frac{3}{8}) = 2 + \sqrt{D(\frac{3}{4})} = 2 + \sqrt{2 - \sqrt{2}}
\qquad
D(\frac{5}{8}) = 4 - D(\frac{3}{8}) = 2 - \sqrt{2 - \sqrt{2}}
$$
The outcomes can be sorted, in descending order, or with the
$h$-values as a key:
\begin{eqnarray*}
D(0/8) &=& 4 \\
D(1/8) &=& 2 + \sqrt{2 + \sqrt{2}} \\
D(2/8) &=& 2 + \sqrt{2} \\
D(3/8) &=& 2 + \sqrt{2 - \sqrt{2}} \\
D(4/8) &=& 2 \\
D(5/8) &=& 2 - \sqrt{2 - \sqrt{2}} \\
D(6/8) &=& 2 - \sqrt{2} \\
D(7/8) &=& 2 - \sqrt{2 + \sqrt{2}} \\
D(8/8) &=& 0
\end{eqnarray*}
It is evident that such a procedure will work for any denominator of the form
$2^N$ where $N >0$ is an integer. The proof is by complete induction. Suppose
that the work has already been done for $2^{N-1}$, then all function values for
angles $\pi.2.k/2^N$ are known. The values for $\pi.k/2^N$ can be calculated by
using $D(k/2^N)=2+\sqrt{D(2.k/2^N)}$, giving all values $k/2^N$ in the interval
$0
The structures of square-roots-of-two are very much alike, except for the $\pm$
signs. Hence they may be abbreviated as follows and interpreted as binary codes
or "numbers":
$$
2 + \sqrt{2 - \sqrt{2 + \sqrt{2 + \sqrt{ 2 - ... }}}}
\quad \equiv \quad + - + + - \: ...
\quad \equiv \quad 0 \, 1 \, 0 \, 0 \, 1 \: ...
$$
Calculations for denominators of higher degree can be automated by a computer
program. By doing so, it is observed that the "new" numbers, with $k = $ odd,
together form a binary-reflected Gray-code. A Gray-code is characterized
by the property that only one "bit" is changed at a time, while going to
the next codeword (provided that both codewords are of equal length). Suppose
we have a $N$-bit Gray code which is represented by a $(N \times 2^N)$ matrix of
'$+$'s and '$-$'s, in such a way that the $i$'th column is the $i$'th codeword:
$$
G(N) = \left[ \begin{array}{cccccc}
G_0 & G_1 & ... & G_i & ... & G_{2^N-1}
\end{array} \right]
$$
The order of the codes belonging to $G(N)$ remains unchanged while using the
formula $D(h/2)=2+\sqrt{D(h)}$, the order is reversed while using the formula
$D(1-h/2)=2-\sqrt{D(h)}$, for obtaining the next generation of codewords. Thus,
while performing the next step of the angle-refinement procedure, the code is
augmented, for odd indices $k$, with '$+$' and '$-$' signs as follows:
$$
G(N+1) = \left[ \begin{array}{ccccccccc}
G_0 & G_1 & ... & G_{2^N-1} & G_{2^N-1} & G_{2^N-2} & ... & G_1 & G_0 \\
+ & + & ... & + & - & - & ... & - & -
\end{array} \right]
$$
If $G(N)$ is a $N$-bit Gray code, then herewith it is clear that $G(N+1)$ must
be a $(N+1)$-bit Gray code. Our proof by induction is completed by verifying
the existence of a trivial $1$-bit code:
$$
G(1) = \left[ \begin{array}{cc} + & - \end{array} \right]
$$
Let's consider again now the functions which relate the the product of the
off-diagonal coefficients at the coarser grid to the product of coefficients
at the finer grid and vice versa. Start with:
$$
x = \frac{ \sqrt{x'} }{ 2.\sqrt{x'} \pm 1} = \frac{1}{2 \pm 1/\sqrt{x'} }
$$
The accompanying process of grid refinement may also be described as follows:
$$
\frac{1}{x} = 2 \pm \sqrt{\frac{1}{x'}}
$$
Start with $x'=1/2$ or $1/x'=2$, the value for which the (inverse) coarsening
process explodes, then $1/x = 2 \pm \sqrt{2}$. Iterating again gives: $1/x :=
2 \pm \sqrt{2 \pm \sqrt{2}}$. And so on and so forth. If these values are sorted
in descending order, and plotted on a computer screen, then the resemblance with
a cosine function readily becomes apparent. Indeed, the sequence:
$$
2 \pm \sqrt{2 \pm \sqrt{2 \pm \sqrt{2 \pm \sqrt{2 \pm \sqrt{ 2 ... }}}}}
$$
is recognized to be quite the same one as has been produced by iterations with
the function $D$: start with $h = 1/2$ and do exactly the same with:
$$
D(\half h) = 2 + \sqrt{D(h)} \OF
D(1 - \half h) = 2 - \sqrt{D(h)}
$$
We may conclude that there exists an intimate, one-to-one relationship between
two seemingly quite different functions. The first one maps the catastrophic
value $1/2$ onto points in the space of products of off-diagonal coefficients,
which may then called "dangerous". The second one calculates all successive
values of $2 + 2.\cos(k.\pi/2^N)$ for all $k$ with $0 \le k \le 2^N$ and $N$ an
arbitrary large integer, starting with the outcomes $2 \pm \sqrt{2}$. In fact,
nothing prevents us from writing:
$$
\frac{1}{a'.b'} \equiv D(h) \EN \frac{1}{a.b} \equiv D(\half h)
$$
Where $a.b$ denotes the off-diagonal coefficients product, which is obtained by
refinement of the grid belonging to $a'.b'$. It is observed that grid-refinement
corresponds with halving an angle. It may be questioned if the reverse is
also true: does grid coarsening correspond with doubling an angle?
$$
x' = \left( \frac{x}{2.x-1} \right)^2 \slechts
\frac{1}{x'} = \left(2 - \frac{1}{x} \right)^2
$$
Which is equivalent with:
$$
\cos(2.\phi) = 2.\cos^2(\phi) - 1 \slechts
2 + 2.\cos(2.\phi) = 2 + 4.\cos^2(\phi) - 2 =
$$ $$
\left[ 2.\cos(\phi) \right]^2 =
\left( 2 - \left[ 2 + 2.\cos(\phi) \right] \right)^2 \slechts
D(2.h) = \left[ 2 - D(h) \right]^2
$$
Where $\phi = \pi.h$. Because $1/x \equiv h$ and $1/x' \equiv 2.h$, there is
enough solid ground now to identify:
$$
\frac{1}{a.b} = D(h) = 2 + 2.\cos(\pi.h)
$$
Herewith we find the distribution of dangerous points in $(a.b)$-space,
belonging to a grid refinement of order $N$, as has been verified by numerical
experiments too:
$$
a.b = \frac{1}{2 + 2.\cos(k.\pi/2^N)} \quad \mbox{with} \quad k = 0,1,2, ... , 2^N
$$