You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
\item (Dreh-)Spiegelmatrix mit Spiegelung an der Geraden $y=tan(\frac{\alpha}{2})\cdot x$:\ \
333
+
$\begin{pmatrix}
334
+
\cos(\alpha) & \sin(\alpha)\\
335
+
\sin(\alpha) & -\cos(\alpha)
336
+
\end{pmatrix}$
337
+
\end{itemize}
338
+
\end{tabular}
320
339
\textbf{Orthonormalisierungsvefahren einer Basis $\{v_1,\ldots,v_n\}$ nach Gram-Schmidt}
321
340
\begin{enumerate}\itemsep0pt
322
-
\item$b_1=\frac{v_1}{\|v_1\|}$\qquad (Vektor mit vielen 0en oder 1en)
323
-
\item$b_{2}= \frac{c_2}{\norm{c_2}}$\ \ mit \ \ $c_2=v_2-\frac{\sprod{v_2}{v_1}}{\sprod{v_1}{v_1}}\cdotv_1$
324
-
\item$b_{3}= \frac{c_3}{\norm{c_3}}$\ \ mit \ \ $c_3=v_3-\frac{\sprod{v_3}{v_1}}{\sprod{v_1}{v_1}} \cdotv_1-\frac{\sprod{v_3}{c_2}}{\sprod{c_2}{c_2}}\cdot c_2$
341
+
\item$b_1=\frac{c_1}{\|c_1\|}$\ \ mit \ \ $c_1=v_1$\ \(Vektor mit vielen 0en oder 1en)
342
+
\item$b_{2}= \frac{c_2}{\norm{c_2}}$\ \ mit \ \ $c_2=v_2-\frac{\sprod{v_2}{c_1}}{\sprod{c_1}{c_1}}\cdotc_1$
343
+
\item$b_{3}= \frac{c_3}{\norm{c_3}}$\ \ mit \ \ $c_3=v_3-\frac{\sprod{v_3}{c_1}}{\sprod{c_1}{c_1}} \cdotc_1-\frac{\sprod{v_3}{c_2}}{\sprod{c_2}{c_2}}\cdot c_2$
325
344
\end{enumerate}
326
345
\textbf{Erweitern einer ONB von $V$ auf eine ONB des $\mathbb{R}^n$}
327
346
\begin{enumerate}\itemsep0pt
@@ -481,42 +500,45 @@ \subsection{QR-Zerlegung}
481
500
$A = QR$, wobei $Q$ orthogonal und R oben dreieckig.\\
482
501
\textbf{Vorgehen}
483
502
\begin{itemize}\itemsep0pt
484
-
\item$Q$ berechnen durch Gram-Schmidt mit den Spalten von $A$, beginnend bei der ersten
485
-
\item Die Koeffizienten von $R$ ergeben sich durch Umstellen der jeweiligen Gram-Schmidt Gleichungen auf die Spalten von $A$
503
+
\item$Q$ berechnen durch Gram-Schmidt mit den Spalten von $A$, \textbf{beginnend bei der ersten}
504
+
\item Die Koeffizienten von $R$ ergeben sich aus den Gram-Schmidt Gleichungen wie folgt: $r_{i,i}=||c_i||^2$ und $r_{i,j}=\frac{\langle v_j,c_i \rangle}{r_{i,i}}$
486
505
\item Alternativ gilt: $R = Q^TA$
487
506
\end{itemize}
488
507
489
508
\subsection{Kleinstes-Quadrate-Problem}
490
509
Für $Ax = b$ lautet die
491
510
\textbf{Normalengleichung}
492
-
$A^TAx = A^Tb$\\
511
+
$A^TAx^* = A^Tb$\\
512
+
\\
513
+
\textbf{Lösen} durch Gauß oder Umstellen: $x^* = (A^TA)^{-1}A^Tb$\\
514
+
Für $A=QR$ lautet die Lösung $x^* = R^{-1}Q^Tb$\\
493
515
$\Rightarrow$ optimale Lösung mit minimalem quadratischen Fehler (existiert immer).
494
516
495
517
\subsection{Singulärwertzerlegung}
496
-
Bei der Singulärwertzerlegung wird eine beliebige Matrix $A\in\mathbb{R}^{m\times n}$ als Produkt dreier Matrizen $U$, $S$ und $V$ geschrieben
518
+
Bei der Singulärwertzerlegung wird eine beliebige Matrix $A\in\mathbb{R}^{m\times n}$ als Produkt dreier Matrizen $V$, $\Sigma$ und $W$ geschrieben
497
519
\begin{equation*}
498
-
A=USV^\top
520
+
A=V\Sigma W^\top
499
521
\end{equation*}
500
-
mit $U\in\mathbb{R}^{m\times m}$, $S\in\mathbb{R}^{m\times n}$ und $V\in\mathbb{R}^{n\times n}$.\\
501
-
$U$ und $V$ sind orthogonal, $S$ ist eine Diagonalmatrix.
522
+
mit $V\in\mathbb{R}^{m\times m}$, $\Sigma\in\mathbb{R}^{m\times n}$ und $W\in\mathbb{R}^{n\times n}$.\\
523
+
$V$ und $W$ sind orthogonal, $\Sigma$ ist eine Diagonalmatrix.
502
524
\subsubsection{Rezept: Singulärwertzerlegung}
503
525
Gegeben: $A\in\mathbb{R}^{m\times n}$
504
526
\begin{enumerate}\itemsep0pt
505
-
\item Bestimme alle Eigenwerte $\lambda_j$ und Eigenvektoren $v_j$ der Matrix $A^\top A\in\mathbb{R}^{n\times n}$ und ordne sie \\$\lambda_1\ge\lambda_2\ge\dots\ge\lambda_r>\lambda_{r+1}=\dots=\lambda_n=0$ mit $r\le n$
506
-
\item Bestimme eine ONB des $\mathbb{R}^n$ aus den Eigenvektoren $v_j$ und erhalte $V=\begin{pmatrix}
507
-
v_1 &\dots & v_n
527
+
\item Bestimme alle Eigenwerte $\lambda_j$ und Eigenvektoren $w_j$ der Matrix $A^\top A\in\mathbb{R}^{n\times n}$ und ordne sie \\$\lambda_1\ge\lambda_2\ge\dots\ge\lambda_r>\lambda_{r+1}=\dots=\lambda_n=0$ mit $r\le n$
528
+
\item Bestimme eine ONB des $\mathbb{R}^n$ aus den Eigenvektoren $w_j$ und erhalte $W=\begin{pmatrix}
529
+
w_1 &\dots & w_n
508
530
\end{pmatrix} \in\mathbb{R}^{n\times n}$
509
531
\item Die Singulärwerte sind $\sigma_j=\sqrt{\lambda_j}$\qquad$j=1,\dots,\min\{m,n\}$
\item Bestimme $u_1,\dots,u_r$ aus $u_i=\frac{1}{\sigma_j}Av_j$ für alle $j=1,\dots,r$ (alle $\sigma_j\ne0$)
530
-
\item Falls $r<m$ ergänze $u_1,\dots,u_r$ zu einer ONB, bzw. zu $U=\begin{pmatrix}
531
-
u_1 & \dots & u_m
551
+
\item Bestimme $v_1,\dots,v_r$ aus $v_i=\frac{1}{\sigma_j}Aw_j$ für alle $j=1,\dots,r$ (alle $\sigma_j\ne0$)
552
+
\item Falls $r<m$ ergänze $v_1,\dots,v_r$ zu einer ONB, bzw. zu $V=\begin{pmatrix}
553
+
v_1 & \dots & v_m
532
554
\end{pmatrix}$
533
555
orthogonal.
534
-
\item$A=USV^\top$
556
+
\item$A=V\Sigma W^\top$
535
557
\end{enumerate}
536
-
537
-
\subsection{Lineare Differentialgleichungen}
558
+
\subsubsection{Kompression}
559
+
\begin{itemize}\itemsep0pt
560
+
\item Rang k Matrix $A_{(k)}=V\Sigma_{(k)}W^\top$. Die Matrix $\Sigma_{(k)}$ enthält nur die Singulärwerte $\sigma_1, ...,\sigma_k$. ($\sigma_{k+1}, ...,\sigma_{n/m}$ mit 0 ersetzen).\\
561
+
\item Frobenius-Norm $|| A ||_F = \sqrt[]{\sum_{i=1}^{m} \sum_{j=1}^{m} {a^{2}_{i,j}}}$\\
562
+
\item Es gilt: $|| A-A_{(k)} ||_F \leq || A-B ||_F$ mit $B$ als eine beliebige Matrix des $\mathbb{R}^{m\times n}$ mit $rang(B) \leq rang(A_{(k)}) = k$
563
+
\item Speicheraufwand einer Matrix $A\in\mathbb{R}^{m\times n}$: $rang(A) \cdot (m+n)$
564
+
\end{itemize}
565
+
\subsection{Lineare Differentialgleichungen und Rekursive Folgen}
538
566
\subsubsection{Lösen einer linearen Differentialgleichung}
539
567
Gegeben: $y'(t) = \lambda y(t)$, mit $y(0) = c$\\
540
568
Lösung: $y(t) = ce^{\lambda t}$ mit $c \in\mathbb R$\\
541
569
\subsubsection{Lösen eines Systems linearer Differentialgleichungen}
0 commit comments