Central-difference formulae. (Q1513629)

I. Entwickelung der hyperbolischen Functionen \(\sinh^n\varphi\) nach Potenzen von \(\varphi\), \(\cosh n\varphi\), und \(\sinh n\varphi\), \(\varphi^m\) nach Potenzen von \(\sinh\varphi\), und dergleichen. II. ``Centraldifferenz''-Formeln. Es sei \(u = f(x)\), \(u_0 = f(x_0)\), \(x_n = x_0 + nh\) für \(u = 0,\pm1,\pm2,\dots\); die Operatoren \(\delta\) und \(\mu\) werden definirt: \[ \begin{aligned} \delta f(x) &= f(x+\frac12h) - f(x-\frac12h),\\ \mu f(x) &= \frac12\{f(x+\frac12h) + f(x-\frac12h)\};\\ \delta^2f(x) &= \delta\delta f(x);\,\delta^3f(x) = \delta\delta^2f(x);\,\dots;\\ \mu^2f(x) &= \mu\mu f(x);\,\mu^3f(x) = \mu\mu^2f(x),\,\dots;\end{aligned} \] dann ist \[ \begin{aligned} \delta^nf(x) &= f(x+\frac12nh) - nf(x+\frac12nh-h) +\cdots+ (-1)^nf(x-\frac12nh),\\ \mu^nf(x) &= \{f(x+\frac12nh) + nf(x+\frac12nh-h) +\cdots+ f(x-\frac12nh)\}/2^n,\\ \delta^nu_r &= \delta^nf(x+rh);\,\mu\delta^nu_r = \frac12(\delta^nu_{r+\frac12} + \delta^nu_{r-\frac12}).\end{aligned} \] In der Tabelle: \[ \begin{matrix} &\quad&\quad&\quad&\quad&\quad\\ x&u&1.\text{ Diff.}&2.\text{ Diff.}&3.\text{ Diff.}&\hdots\\ \vdots&\vdots&\vdots&\vdots&\vdots&\\ x_{-1}&u_{-1}&(\mu\delta u_{-1})&\delta^2u_{-1}&(\mu\delta^3u_{-1})&\hdots\\ &&\delta u_{-\frac12}&&\delta^3u_{-\frac12}&\\ x_0&u_0&(\mu\delta u_0)&\delta^2u_0&(\mu\delta^3u_0)&\hdots\\ &&\delta u_{\frac12}&&\delta^3u_{\frac12}&\\ x_1&u_1&(\mu\delta u_1)&\delta^2u_1&(\mu\delta^3u_1)&\hdots\\ \vdots&\vdots&\vdots&\vdots&\vdots&\end{matrix} \] heissen die einem Werte von \(u\) in einer Horizontalreihe folgenden Werte seine ``Centraldifferenzen''. Es ist: \[ \mu\{\varphi(x) + \psi(x)\} = \mu\varphi(x) + \mu\psi(x);\,\mu\delta f(x) = \delta\mu f(x). \] \(\mu_n\), \(\delta_n\) werden erklärt: \(\mu_nf(x) = \frac12\{f(x+\frac12nh) + f(x-\frac12nh)\}\), \(\delta_nf(x) = f(x+\frac12nh) - f(x-\frac12nh)\), wo \(n\) positiv oder negativ ist; dann ist \(\mu_{-1} = \mu\), \(\delta_{-1} = \delta\), dagegen \(\mu^{-1} = (1 + \frac14\delta^2)^{-\frac12}\); \[ \begin{multlined} u_\frac12 = \mu\left(1 + \frac14\delta^2\right)^{-\frac12}u_\frac12 = \left(\mu - \frac18\mu\delta^2 + \frac3{128}\mu\delta^4 \frac5{1024}\mu\delta^6 + \frac{35}{32768}\mu\delta^8 -\cdots\right)u_\frac12\\ = \frac12(u_1+u_0) - \frac1{16}(\delta^2u_1+\delta^2u_0) + \frac3{256}(\delta^4u_1+\delta^4u_0) - \frac5{2048}(\delta^6u_1+\delta^6u_0) + \frac{35}{65536}(\delta^8u_1+\delta^8u_0)\cdots.\end{multlined} \] Ist \(Df(x) = f'(x)\), \(Du = u'\), so ist nach dem Taylor'schen Satze \[ \begin{aligned} &f(x + \frac12h) = f(x) + \frac12hf'(x) + \frac1{2!}(\frac12h)^2f''(x) +\cdots= e^{\frac12hD}f(x);\\ &f(x - \frac12h) = e^{-\frac12hD}f(x),\\ &\mu f(x) = \frac12(e^{\frac12hD} + e^{-\frac12hD})f(x)\text{ oder }\mu = \cosh\frac12hD,\\ &\delta f(x) = (e^{\frac12hD} - e^{-\frac12hD})f(x),\text{ oder }\delta f(x)= \frac12\sinh\frac12hD;\end{aligned} \] die Formeln \(\delta^n = (2\sinh\frac12hD)^n\), \(\mu\delta^{n-1} = (2\sinh\frac12hD)^{n-1}\cosh\frac12hD\) werden für \[ \mu\delta u_0,\,\delta^2u_0,\,\mu\delta^3u_0,\,\delta^4u_0,\, \mu\delta^5u_0,\,\delta^6u_0,\,\dots,\,\mu u_{\frac12},\,\delta u_\frac12,\,\mu\delta^2u_\frac12,\,\delta^3u_\frac12,\, \mu\delta^4u_\frac12,\,\delta^5u_\frac12,\,\dots \] specialisirt, ähnliche Formeln für \[ hDu_0,\, h^2D^2u_0,\, h^3D^3u_0,\,\dots,\, u_\frac12,\, hDu_\frac12,\, h^2D^2u_\frac12,\,\dots \] Nimmt man die Intervalle gleich \(nh\) statt \(h\) und bezeichnet die neuen Werte von \(\mu\) und \(\delta\) durch \(\mu_n\) und \(\delta_n\), so dass \(\mu_n = \cosh\frac12nhD\), \(\delta_n = 2\sinh\frac12nhD\) wird, so wird \(\mu_{-n}=\mu_n\), \(\mu_{m\pm n}=\mu_m\mu_n\pm\frac14\delta_m\delta_n\), \(\delta_{m\pm n}=\mu_n\delta_m\pm\mu_m\delta_n\), und \[ \begin{multlined} \delta_n^m = n^m\{{}_mB_m\delta^m + {}_mB_{m+2}\delta^{m+2}/(2m+2)(2m+4)\\ + {}_mB_{m+4}\delta^{m+1}/(2m+2)\dots(2m+8) +\cdots\},\end{multlined} \] \[ \begin{multlined} \mu_n\delta_n^{m-1} = n^{m-1}\{{}_mB_m\mu\delta^{m-1} + {}_mB_{m+2}\mu\delta^{m+1}/2m(2m+2)\\ + {}_mB_{m+4}\delta^{m+3}/2m\dots(2m+6) +\cdots\},\end{multlined} \] wo die Coefficienten durch die Gleichungen \({}_mB_m = 1\) und \[ {}_mB_{m+2r+2} = \{m^2n^2 - (m+2r)^2\}{}_mB_{m+2r} + {}_{m-2}B_{m+2r} \] bestimmt sind. Schliesslich werden die Interpolationsformeln \[ u_n = u_0 + \frac n{1!}\mu\delta u_0 + \frac{n^2}{2!}\mu\delta^2u_0 + \frac{n(n^2-1^2)}{3!}\mu\delta^3u_0 + \frac{n^2(n^2-1^2)}{4!}\delta^4u_0 +\cdots, \] \[ \begin{multlined} u_{\frac12+n} = \mu u_\frac12 + \frac{2n}{2\cdot1!}\delta u_\frac12 + \frac{4n^2-1^2}{2^2\cdot2!}\mu\delta^2u_\frac12 + \frac{2n(4n^2-1^2)}{2^3\cdot3!}\delta^3u_\frac12\\ + \frac{(4n^2-1^2)(4n^2-3^2)}{2^4\cdot4!}\mu\delta^4u_\frac12 +\cdots,\end{multlined} \] \[ u_n = \left\{1 + \binom n2\delta^2 + \binom{n+1}4\delta^4 +\cdots\right\}u_0 + \left\{\binom n1\delta + \binom{n+1}3\delta^3 + \binom{n+2}5\delta^5 +\cdots\right\}u_\frac12, \] \[ u_{-n} = \left\{1 + \binom n2\delta^2 + \binom{n+1}4\delta^4 +\cdots\right\}u_0 + \left\{\binom n1\delta + \binom{n+1}3\delta^3 + \binom{n+2}5\delta^5 +\cdots\right\}u_{-\frac12} \] abgeleitet. III. ``Centralsummen''-Formeln. \(\sigma f(x)\) wird definirt durch die Functionalgleichung: \(\delta\sigma f(x) = \sigma f(x+\frac12h) - \sigma f(x-\frac12h) = f(x)\). Sie involvirt eine willkürliche Constante; wird diese passend gewählt, so ist \[ \sigma\{\varphi(x) + \psi(x)\} = \sigma\varphi(x) + \sigma\psi(x);\quad \delta\sigma u = u(u = f(x))\mu\sigma u = \sigma\mu u;\quad \sigma = \delta^{-1}. \] Auch hier wird gesetzt: \(\sigma^nu = \sigma\sigma^{n-1}u\), so dass \(\sigma^nu\) \(n\) willkürliche Constanten enthält. Die Operatoren \(\mu\), \(\delta\), \(\sigma\) und \(D\) (\(Df(x) = f'(x)\), \(Du = u'\)) können nach den Gesetzen der Algebra mit einander verbunden werden. Es ist \[ \begin{multlined} u_1 - u_0 = \delta u_\frac12 = 2\sinh\frac12 hDu_\frac12\\ = \left(1 + \frac1{24}h^2D^2 + \frac1{1920}h^4D^4 + \frac1{322560}h^6D^6 +\cdots\right)hDu_\frac12\end{multlined} \] \[ \begin{multlined} u_1 - u_0 = \mu\delta/\mu u_\frac12 = 2\tanh\frac12 hDu_\frac12\\ = \mu\left(1 - \frac1{12}h^2D^2 + \frac1{120}h^4D^4 - \frac{17}{20160}h^6D^6 +\cdots\right)hDu_\frac12.\end{multlined} \] Für \(u = \int\limits_0^x vdx\) ergeben sich die Formeln: \[ \begin{multlined} \int_0^x vdx = \left(1 + \frac1{24}h^2D^2 + \frac1{1920}h^4D^4 +\cdots\right)hv_\frac12\\ = \mu\left(1 - \frac1{12}h^2D^2 + \frac1{120}h^4D^4 -\cdots\right)hv_\frac12,\end{multlined} \] \[ \begin{multlined} \int_{x_0}^{x_n} vdx = h\sigma(v_n-v_0) + h(\delta X_n-\delta X_0)\\ = h\mu\sigma(v_n-v_0) + h(\delta X_n'-\delta X_0'),\end{multlined} \] wo \[ \begin{aligned} X &= \left(\frac1{24} - \frac{17}{5760}h^2D^2 + \frac{43}{322560}h^4D^4 -\cdots\right)hv,\\ X' &= \left(-\frac1{12} + \frac7{1440}h^2D^2 - \frac{31}{161280}h^4D^4 +\cdots\right)hv\end{aligned} \] ist. \[ \begin{aligned} u_n &= \int^{x_n} vdx = \left(\sigma + \frac1{24}\delta - \frac{17}{5760}\delta^3 + \frac{367}{967680}\delta^5 -\cdots\right)hv_n,\\ u_n &= \int^{x_n} vdx = \mu\left(\sigma - \frac1{12}\delta + \frac{11}{720}\delta^3 - \frac{191}{60480}\delta^5 +\cdots\right)hv_n.\end{aligned} \]