diff options
| author | Juan Marín Noguera <juan.marinn@um.es> | 2021-02-11 09:13:15 +0100 |
|---|---|---|
| committer | Juan Marín Noguera <juan.marinn@um.es> | 2021-02-11 09:13:15 +0100 |
| commit | 5b608bb9b09971b2aceb0c2d99428e61385247fb (patch) | |
| tree | 2a02c8aed613e71270f436082028ae3e80957cb5 /mne | |
| parent | e0740fb026bb6ac2a7c86230a92e69237a6cf193 (diff) | |
Ahora sí
Diffstat (limited to 'mne')
| -rw-r--r-- | mne/n5.lyx | 672 |
1 files changed, 672 insertions, 0 deletions
diff --git a/mne/n5.lyx b/mne/n5.lyx new file mode 100644 index 0000000..98aa9a9 --- /dev/null +++ b/mne/n5.lyx @@ -0,0 +1,672 @@ +#LyX 2.3 created this file. For more info see http://www.lyx.org/ +\lyxformat 544 +\begin_document +\begin_header +\save_transient_properties true +\origin unavailable +\textclass book +\use_default_options true +\maintain_unincluded_children false +\language spanish +\language_package default +\inputencoding auto +\fontencoding global +\font_roman "default" "default" +\font_sans "default" "default" +\font_typewriter "default" "default" +\font_math "auto" "auto" +\font_default_family default +\use_non_tex_fonts false +\font_sc false +\font_osf false +\font_sf_scale 100 100 +\font_tt_scale 100 100 +\use_microtype false +\use_dash_ligatures true +\graphics default +\default_output_format default +\output_sync 0 +\bibtex_command default +\index_command default +\paperfontsize default +\spacing single +\use_hyperref false +\papersize default +\use_geometry false +\use_package amsmath 1 +\use_package amssymb 1 +\use_package cancel 1 +\use_package esint 1 +\use_package mathdots 1 +\use_package mathtools 1 +\use_package mhchem 1 +\use_package stackrel 1 +\use_package stmaryrd 1 +\use_package undertilde 1 +\cite_engine basic +\cite_engine_type default +\biblio_style plain +\use_bibtopic false +\use_indices false +\paperorientation portrait +\suppress_date false +\justification true +\use_refstyle 1 +\use_minted 0 +\index Index +\shortcut idx +\color #008000 +\end_index +\secnumdepth 3 +\tocdepth 3 +\paragraph_separation indent +\paragraph_indentation default +\is_math_indent 0 +\math_numbering_side default +\quotes_style french +\dynamic_quotes 0 +\papercolumns 1 +\papersides 1 +\paperpagestyle default +\tracking_changes false +\output_changes false +\html_math_output 0 +\html_css_as_file 0 +\html_be_strict false +\end_header + +\begin_body + +\begin_layout Standard +Cuando las derivadas de la solución están acotadas, se puede acotar e incluso + predecir el error adecuadamente, y cuando la solución y su derivada crecen + moderadamente, el error absoluto crece pero el relativo es estable. + Si la derivada crece y la función no tenemos un +\series bold +problema rígido +\series default +, en el que el error relativo se dispara. + Un caso típico es el problema +\begin_inset Formula +\[ +\left\{ \begin{aligned}\dot{x}(t) & =\lambda x(t),\\ +x(0) & =\alpha, +\end{aligned} +\right. +\] + +\end_inset + +cuya solución es claramente +\begin_inset Formula $x(t)=\alpha e^{\lambda t}$ +\end_inset + + en todo +\begin_inset Formula $\mathbb{R}$ +\end_inset + + y, si +\begin_inset Formula $\lambda\ll0$ +\end_inset + +, al iterar hacia delante, +\begin_inset Formula $x(t)\to0$ +\end_inset + + rápidamente y +\begin_inset Formula $x^{(p)}(t)=\alpha\lambda^{p}e^{\lambda t}$ +\end_inset + +. + +\end_layout + +\begin_layout Standard +En este caso, con el método de Euler, para que la solución tienda a cero + y para que el error no crezca, el paso debe ser +\begin_inset Formula $h<\frac{2}{|\lambda|}$ +\end_inset + +. + En efecto, +\begin_inset Formula $\omega_{n}\to0\iff|1+h\lambda|<1\iff-1<1+h\lambda<1$ +\end_inset + +, y como +\begin_inset Formula $h>0$ +\end_inset + + y +\begin_inset Formula $\lambda<0$ +\end_inset + +, +\begin_inset Formula $1+h\lambda<1$ +\end_inset + +, pero +\begin_inset Formula $h\lambda>-2\iff h|\lambda|<2$ +\end_inset + +, y si además hay error de redondeo +\begin_inset Formula $\omega_{0}=\alpha+\varepsilon$ +\end_inset + +, el error en +\begin_inset Formula $\omega_{n}$ +\end_inset + + es +\begin_inset Formula $(1+h\lambda)^{n}\varepsilon$ +\end_inset + + y para que el error no crezca debe ser +\begin_inset Formula $|1+h\lambda|<1\iff h<\frac{2}{|\lambda|}$ +\end_inset + +. +\end_layout + +\begin_layout Standard +En general, con los métodos de un paso fijo en +\begin_inset Formula $h$ +\end_inset + + existe un polinomio +\begin_inset Formula $Q$ +\end_inset + + tal que, para que la solución converja, debe ser +\begin_inset Formula $|Q(h\lambda)|<1$ +\end_inset + +. + Para un método de Taylor de orden +\begin_inset Formula $n$ +\end_inset + +, esto es +\begin_inset Formula $Q(x):=\sum_{i=0}^{n}\frac{x^{i}}{i!}$ +\end_inset + +. +\end_layout + +\begin_layout Standard +Sean ahora un método multipaso de paso fijo +\begin_inset Formula $h$ +\end_inset + + con parámetros +\begin_inset Formula $a_{0},\dots,a_{m-1},b_{0},\dots,b_{m}$ +\end_inset + + y la ecuación de recurrencia asociada al método +\begin_inset Formula +\[ +(1-h\lambda b_{m})\omega_{i}=(a_{m-1}+h\lambda b_{m-1})\omega_{i-1}+\dots+(a_{0}+h\lambda b_{0})\omega_{i-m}, +\] + +\end_inset + +si las raíces del polinomio característico son reales y distintas, +\begin_inset Formula $\beta_{1},\dots,\beta_{n}$ +\end_inset + +, para que el método aproxime bien a la solución del problema debe ser cada + +\begin_inset Formula $|\beta_{i}|<1$ +\end_inset + +. +\end_layout + +\begin_layout Standard +La +\series bold +región de estabilidad absoluta +\series default + de un método es +\begin_inset Formula $R\subseteq\mathbb{C}$ +\end_inset + + tal que, si +\begin_inset Formula $h\lambda\in R$ +\end_inset + +, el método converge. + Para un método de un paso que converge cuando +\begin_inset Formula $|Q(h\lambda)|<1$ +\end_inset + +, +\begin_inset Formula $R=\{z\in\mathbb{C}:|Q(z)|<1\}$ +\end_inset + +, y para uno multipaso que converge cuando cada +\begin_inset Formula $|\beta_{i}|<1$ +\end_inset + +, es +\begin_inset Formula $R=\{z\in\mathbb{C}:|\beta_{i}|<1,\forall i\}$ +\end_inset + +. + +\end_layout + +\begin_layout Standard +Hay que tener en cuenta la región de estabilidad antes de considerar un + método adaptativo, pues estos pueden aumentar +\begin_inset Formula $h$ +\end_inset + + por convergencia y evitar la estabilidad. + Para problemas rígidos queremos que +\begin_inset Formula $R$ +\end_inset + + sea lo más grande posible. + Un método es +\series bold +A-estable +\series default + si +\begin_inset Formula $\{z\in\mathbb{C}:\text{Re}z<0\}\subseteq R$ +\end_inset + +. +\end_layout + +\begin_layout Standard +Como +\series bold +teorema +\series default +, los métodos explícitos de Runge-Kutta no son A-estables, y un método multipaso + A-estable tiene orden de convergencia máximo 2. + Algunos métodos implícitos son: +\end_layout + +\begin_layout Enumerate + +\series bold +Método de Euler implícito +\series default + o +\series bold +hacia atrás: +\series default + +\begin_inset Formula $\omega_{i}=\omega_{i-1}+hf(t_{i},\omega_{i})$ +\end_inset + +. +\end_layout + +\begin_deeper +\begin_layout Enumerate +Es estable. + +\end_layout + +\begin_deeper +\begin_layout Standard +Como la relación de recurrencia es +\begin_inset Formula $\omega_{i}=\omega_{i-1}$ +\end_inset + +, cumple la condición de raíz. + +\end_layout + +\end_deeper +\begin_layout Enumerate +Es consistente cuando +\begin_inset Formula $\ddot{x}$ +\end_inset + + es acotada. +\end_layout + +\begin_deeper +\begin_layout Standard +Existe +\begin_inset Formula $\xi\in[t_{i},t_{i+1}]$ +\end_inset + + con +\begin_inset Formula $\tau_{i+1}(h)=\frac{x(t_{i}+h)-x(t_{i})}{h}-f(t_{i}+h,x(t_{i}+h))=\dot{x}(t_{i}+h)+\frac{h}{2}\ddot{x}(\xi)-\dot{x}(t_{i}+h)=\frac{h}{2}\ddot{x}(\xi)$ +\end_inset + +, y si +\begin_inset Formula $\Vert\ddot{x}\Vert$ +\end_inset + + está acotado, como ocurre en el problema, +\begin_inset Formula $\max_{i}\Vert\tau_{i+1}(h)\Vert\to0$ +\end_inset + +. +\end_layout + +\end_deeper +\begin_layout Enumerate +Es A-estable. +\end_layout + +\begin_deeper +\begin_layout Standard +El polinomio +\begin_inset Formula $(1-h\lambda)z-1$ +\end_inset + + tiene como única raíz +\begin_inset Formula $\beta:=\frac{1}{1-h\lambda}$ +\end_inset + +, y si +\begin_inset Formula $h\lambda=:a+bi$ +\end_inset + + con +\begin_inset Formula $a<0$ +\end_inset + +, +\begin_inset Formula $|\beta|=\frac{1}{\sqrt{(1-a)^{2}+b^{2}}}\leq\frac{1}{1-a}<1$ +\end_inset + +. +\end_layout + +\end_deeper +\begin_layout Standard +Para implementarlo, sea +\begin_inset Formula $F(\omega):=\omega-\omega_{i-1}-hf(t_{i},\omega)$ +\end_inset + +, se trata de resolver +\begin_inset Formula $F(\omega)=0$ +\end_inset + +, lo que podemos hacer, por ejemplo, por el método de Newton, que nos da + una sucesión +\begin_inset Formula $(\omega_{i}^{k})_{k\in\mathbb{N}}$ +\end_inset + + que converge a la raíz de +\begin_inset Formula $F(\omega)=0$ +\end_inset + +, dada por +\begin_inset Formula $\omega_{i}^{0}:=\omega_{i-1}$ +\end_inset + + y +\begin_inset Formula +\[ +\omega_{i}^{k+1}:=\omega_{i}^{k}-\frac{\omega_{i}^{k}-\omega_{i-1}-hf(t_{i},\omega_{i}^{k})}{1-h\frac{\partial f}{\partial x}(t_{i},\omega)}. +\] + +\end_inset + +Si no conocemos +\begin_inset Formula $\frac{\partial f}{\partial x}$ +\end_inset + +, podemos usar el método de la secante. + Iteramos hasta que +\begin_inset Formula $\Vert\omega_{i}^{k+1}-\omega_{i}^{k}\Vert$ +\end_inset + + sea menor que una tolerancia, o si hemos llegado un máximo de iteraciones. +\end_layout + +\end_deeper +\begin_layout Enumerate + +\series bold +Método del trapecio: +\series default + +\begin_inset Formula $\omega_{i}=\omega_{i-1}+\frac{h}{2}(f(t_{i-1},\omega_{i-1})+f(t_{i},\omega_{i}))$ +\end_inset + +. +\end_layout + +\begin_deeper +\begin_layout Enumerate +Es estable. +\end_layout + +\begin_deeper +\begin_layout Standard +Cumple la condición de raíz con una única raíz 1. +\end_layout + +\end_deeper +\begin_layout Enumerate +Es consistente y de orden 2 para +\begin_inset Formula $\dddot{x}$ +\end_inset + + acotada. +\end_layout + +\begin_deeper +\begin_layout Standard +Para cierto +\begin_inset Formula $t$ +\end_inset + + existen +\begin_inset Formula $\xi,\mu\in[t,t+h]$ +\end_inset + + con +\begin_inset Formula +\begin{multline*} +x(t+h)-x(t)-\frac{h}{2}(f(t,x(t))+f(t+h,x(t+h)))=\\ +=\left(x(t)+h\dot{x}(t)+\frac{1}{2}h^{2}\ddot{y}(t)+\frac{1}{6}h^{3}\dddot{y}(\xi)\right)-x(t)-\frac{h}{2}\left(\dot{x}(t)+\left(\dot{x}(t)+h\ddot{x}(t)+\frac{h^{2}}{2}\dddot{x}(\mu)\right)\right)=\\ +=\frac{1}{6}h^{3}\dddot{x}(\xi)-\frac{1}{4}h^{3}\dddot{x}(\mu), +\end{multline*} + +\end_inset + +y si +\begin_inset Formula $\Vert\dddot{x}\Vert$ +\end_inset + + es acotada por un cierto +\begin_inset Formula $C$ +\end_inset + +, +\begin_inset Formula $\Vert\tau_{i+1}(h)\Vert\leq h^{2}(\frac{1}{6}C+\frac{1}{4}C)$ +\end_inset + + y +\begin_inset Formula $\Vert\tau(h)\Vert\leq\frac{5}{12}Ch^{2}\to0$ +\end_inset + +. +\end_layout + +\end_deeper +\begin_layout Enumerate +Es A-estable. +\end_layout + +\begin_deeper +\begin_layout Standard +La recurrencia es +\begin_inset Formula $Q_{h\lambda}(z)=(1-\frac{h\lambda}{2})z-(1+\frac{h\lambda}{2})$ +\end_inset + + y el polinomio característico tiene una única raíz +\begin_inset Formula $\beta=\frac{1+\frac{h\lambda}{2}}{1-\frac{h\lambda}{2}}$ +\end_inset + +. + Si +\begin_inset Formula $\frac{h\lambda}{2}=:a+bi$ +\end_inset + + con +\begin_inset Formula $a<0$ +\end_inset + +, +\begin_inset Formula +\[ +|\beta|=\frac{|1+a+bi|}{|1-a-bi|}=\sqrt{\frac{(1+a)^{2}+b^{2}}{(1-a)^{2}+b^{2}}}<1, +\] + +\end_inset + +pues +\begin_inset Formula $(1+a)^{2}<(1-a)^{2}$ +\end_inset + +. +\end_layout + +\end_deeper +\end_deeper +\begin_layout Standard +Dado un método a +\begin_inset Formula $m$ +\end_inset + + pasos +\begin_inset Formula +\[ +\omega_{i}=a_{0}\omega_{i-m}+\dots+a_{m-1}\omega_{i-1}+h(b_{0}f(t_{i-m},\omega_{i-m})+\dots+b_{m}f(t_{i},\omega_{i})), +\] + +\end_inset + +llamamos +\begin_inset Formula $\rho(z):=z^{m}-a_{m-1}z^{m-1}-\dots-a_{1}z-a_{0}$ +\end_inset + + y +\begin_inset Formula $\sigma(z):=b_{m}z^{m}+\dots+b_{1}z+b_{0}$ +\end_inset + +. + Entonces el método es una +\series bold +BDF +\series default + ( +\emph on +\lang english +Backwards Differentiation Formula +\emph default +\lang spanish +) si es de orden +\begin_inset Formula $m$ +\end_inset + + y +\begin_inset Formula $\sigma(z)=\beta z^{m}$ +\end_inset + + para algún +\begin_inset Formula $\beta\neq0$ +\end_inset + +. +\end_layout + +\begin_layout Standard +Todo método BDF cumple +\begin_inset Formula $\beta=\left(\sum_{j=1}^{m}\frac{1}{j}\right)^{-1}$ +\end_inset + + y +\begin_inset Formula $\rho(z)=\beta\sum_{j=1}^{m}\frac{1}{j}z^{m-j}(z-1)^{j}$ +\end_inset + +. + Así: +\end_layout + +\begin_layout Enumerate +El BDF de orden 2 es +\begin_inset Formula $\omega_{i}=\frac{4}{3}\omega_{i-1}-\frac{1}{3}\omega_{i-2}+\frac{2}{3}hf(t_{i},\omega_{i})$ +\end_inset + +. +\end_layout + +\begin_deeper +\begin_layout Standard +Se tiene +\begin_inset Formula $\beta=\left(1+\frac{1}{2}\right)^{-1}=\frac{2}{3}$ +\end_inset + +, luego +\begin_inset Formula $\rho(z)=\frac{2}{3}\left(z(z-1)+\frac{1}{2}(z-1)^{2}\right)=\frac{2}{3}\left(\frac{3}{2}z^{2}-2z+\frac{1}{2}\right)=z^{2}-\frac{4}{3}z+\frac{1}{3}$ +\end_inset + +, con lo que +\begin_inset Formula $a_{1}=\frac{4}{3}$ +\end_inset + + y +\begin_inset Formula $a_{0}=-\frac{1}{3}$ +\end_inset + +. + Además, +\begin_inset Formula $b_{0},\dots,b_{m-1},b_{m}=0,\dots,0,\beta$ +\end_inset + +, luego +\begin_inset Formula $\omega_{i}=\frac{4}{3}\omega_{i-1}-\frac{1}{3}\omega_{i-2}+\frac{2}{3}hf(t_{i},\omega_{i})$ +\end_inset + +. +\end_layout + +\end_deeper +\begin_layout Enumerate +El BDF de orden 3 es +\begin_inset Formula $\omega_{i}=\frac{18}{11}\omega_{i-1}-\frac{9}{11}\omega_{i-2}+\frac{2}{11}\omega_{i-3}+\frac{6}{11}f(t_{i},\omega_{i})$ +\end_inset + +. +\end_layout + +\begin_deeper +\begin_layout Standard +\begin_inset Formula $\beta=\left(1+\frac{1}{2}+\frac{1}{3}\right)^{-1}=\frac{6}{11}$ +\end_inset + +, luego +\begin_inset Formula $\rho(z)=\frac{6}{11}\left(z^{2}(z-1)+\frac{1}{2}z(z-1)^{2}+\frac{1}{3}(z-1)^{3}\right)=\frac{6}{11}\left(z^{3}-z^{2}+\frac{1}{2}(z^{3}-2z^{2}+z)+\frac{1}{3}(z^{3}-3z^{2}+3z-1)\right)=z^{3}-\frac{18}{11}z^{2}+\frac{9}{11}z-\frac{2}{11}$ +\end_inset + +, de modo que +\begin_inset Formula $\omega_{i}-\frac{18}{11}\omega_{i-1}+\frac{9}{11}\omega_{i-2}-\frac{2}{11}\omega_{i-3}=\frac{6}{11}f(t_{i},\omega_{i})$ +\end_inset + +. +\end_layout + +\end_deeper +\begin_layout Standard +Los métodos BDF tienen una región de estabilidad lineal grande, y como +\series bold +teorema +\series default +, un método BDF cumple la condición de raíz y es convergente si y sólo si + es de orden entre 1 y 6. +\end_layout + +\end_body +\end_document |
