User:Sintel/Generator optimization: Difference between revisions

Sintel (talk | contribs)
Created page with "Some derivations on optimal generators for 2-norm (Euclidean) tunings. In what follows, <math>M</math> is an <math>n \times k</math> integer matrix that represents the temper..."
 
Sintel (talk | contribs)
No edit summary
Line 50: Line 50:
\end{gather}
\end{gather}
$$
$$
That was fun. Now if we don't know which intervals to optimize for, a simple choice is just to optimize for each prime in the subgroup. This is equivalent to picking the identity matrix <math>V = I_{k \times k}</math>, so that the above expression simplifies to:
Now if we don't know which intervals to optimize for, a simple choice is just to optimize for each prime in the subgroup. This is equivalent to picking the identity matrix <math>V = I_{k \times k}</math>, so that the above expression simplifies to:
$$
$$
g  = jM^{\mathsf T} (MM^{\mathsf T})^{-1}
g  = jM^{\mathsf T} (MM^{\mathsf T})^{-1}
Line 111: Line 111:
$$
$$


We can write the lagrangian as:
We can write the Lagrangian as:
$$
$$
\begin{gather}
\begin{gather}
Line 119: Line 119:
where we introduced a vector of Lagrange multipliers <math>\lambda</math>.
where we introduced a vector of Lagrange multipliers <math>\lambda</math>.


Taking derivatives with respect to <math>g</math> and <math>\lambda</math> we get:
Taking derivatives with respect to <math>g</math> and <math>\lambda</math> we get a linear system:
$$
$$
\begin{bmatrix}
\begin{bmatrix}