\[
% general purpose
\newcommand{\ctext}[1]{\raise0.2ex\hbox{\textcircled{\scriptsize{#1}}}}
% mathematics
% general purpose
\DeclarePairedDelimiterX{\parens}[1]{\lparen}{\rparen}{#1}
\DeclarePairedDelimiterX{\braces}[1]{\lbrace}{\rbrace}{#1}
\DeclarePairedDelimiterX{\bracks}[1]{\lbrack}{\rbrack}{#1}
\DeclarePairedDelimiterX{\verts}[1]{|}{|}{#1}
\DeclarePairedDelimiterX{\Verts}[1]{\|}{\|}{#1}
\newcommand{\as}{{\quad\textrm{as}\quad}}
\newcommand{\st}{{\textrm{ s.t. }}}
\DeclarePairedDelimiterX{\setComprehension}[2]{\lbrace}{\rbrace}{#1\,\delimsize\vert\,#2}
\newcommand{\naturalNumbers}{\mathbb{N}}
\newcommand{\integers}{\mathbb{Z}}
\newcommand{\rationalNumbers}{\mathbb{Q}}
\newcommand{\realNumbers}{\mathbb{R}}
\newcommand{\complexNumbers}{\mathbb{C}}
\newcommand{\field}{\mathbb{F}}
\newcommand{\func}[2]{{#1}\parens*{#2}}
\newcommand*{\argmax}{\operatorname*{arg~max}}
\newcommand*{\argmin}{\operatorname*{arg~min}}
% set theory
\newcommand{\range}[2]{\braces*{#1,\dotsc,#2}}
\providecommand{\complement}{}\renewcommand{\complement}{\mathrm{c}}
\newcommand{\ind}[2]{\mathbbm{1}_{#1}\parens*{#2}}
\newcommand{\indII}[1]{\mathbbm{1}\braces*{#1}}
% number theory
\newcommand{\abs}[1]{\verts*{#1}}
\newcommand{\combi}[2]{{_{#1}\mathrm{C}_{#2}}}
\newcommand{\perm}[2]{{_{#1}\mathrm{P}_{#2}}}
\newcommand{\GaloisField}[1]{\mathrm{GF}\parens*{#1}}
% analysis
\newcommand{\NapierE}{\mathrm{e}}
\newcommand{\sgn}[1]{\operatorname{sgn}\parens*{#1}}
\newcommand*{\rect}{\operatorname{rect}}
\newcommand{\cl}[1]{\operatorname{cl}#1}
\newcommand{\Img}[1]{\operatorname{Img}\parens*{#1}}
\newcommand{\dom}[1]{\operatorname{dom}\parens*{#1}}
\newcommand{\norm}[1]{\Verts*{#1}}
\newcommand{\floor}[1]{\left\lfloor#1\right\rfloor}
\newcommand{\ceil}[1]{\left\lceil#1\right\rceil}
\newcommand{\expo}[1]{\exp\parens*{#1}}
\newcommand*{\sinc}{\operatorname{sinc}}
\newcommand*{\nsinc}{\operatorname{nsinc}}
\newcommand{\GammaFunc}[1]{\Gamma\parens*{#1}}
\newcommand*{\erf}{\operatorname{erf}}
% inverse trigonometric functions
\newcommand{\asin}[1]{\operatorname{Sin}^{-1}{#1}}
\newcommand{\acos}[1]{\operatorname{Cos}^{-1}{#1}}
\newcommand{\atan}[1]{\operatorname{{Tan}^{-1}}{#1}}
\newcommand{\atanEx}[2]{\atan{\parens*{#1,#2}}}
% convolution
\newcommand{\cycConv}[2]{{#1}\underset{\text{cyc}}{*}{#2}}
% derivative
\newcommand{\deriv}[3]{\frac{\operatorname{d}^{#3}#1}{\operatorname{d}{#2}^{#3}}}
\newcommand{\derivLong}[3]{\frac{\operatorname{d}^{#3}}{\operatorname{d}{#2}^{#3}}#1}
\newcommand{\partDeriv}[3]{\frac{\operatorname{\partial}^{#3}#1}{\operatorname{\partial}{#2}^{#3}}}
\newcommand{\partDerivLong}[3]{\frac{\operatorname{\partial}^{#3}}{\operatorname{\partial}{#2}^{#3}}#1}
\newcommand{\partDerivIIHetero}[3]{\frac{\operatorname{\partial}^2#1}{\partial#2\operatorname{\partial}#3}}
\newcommand{\partDerivIIHeteroLong}[3]{{\frac{\operatorname{\partial}^2}{\partial#2\operatorname{\partial}#3}#1}}
% integral
\newcommand{\integrate}[5]{\int_{#1}^{#2}{#3}{\mathrm{d}^{#4}}#5}
\newcommand{\LebInteg}[4]{\int_{#1} {#2} {#3}\parens*{\mathrm{d}#4}}
% complex analysis
\newcommand{\conj}[1]{\overline{#1}}
\providecommand{\Re}{}\renewcommand{\Re}[1]{{\operatorname{Re}{\parens*{#1}}}}
\providecommand{\Im}{}\renewcommand{\Im}[1]{{\operatorname{Im}{\parens*{#1}}}}
\newcommand*{\Arg}{\operatorname{Arg}}
\newcommand*{\Log}{\operatorname{Log}}
% Laplace transform
\newcommand{\LPLC}[1]{\operatorname{\mathcal{L}}\parens*{#1}}
\newcommand{\ILPLC}[1]{\operatorname{\mathcal{L}}^{-1}\parens*{#1}}
% Discrete Fourier Transform
\newcommand{\DFT}[1]{\mathrm{DFT}\parens*{#1}}
% Z transform
\newcommand{\ZTrans}[1]{\operatorname{\mathcal{Z}}\parens*{#1}}
\newcommand{\IZTrans}[1]{\operatorname{\mathcal{Z}}^{-1}\parens*{#1}}
% linear algebra
\newcommand{\bm}[1]{{\boldsymbol{#1}}}
\newcommand{\matEntry}[3]{#1\bracks*{#2}\bracks*{#3}}
\newcommand{\matPart}[5]{\matEntry{#1}{#2:#3}{#4:#5}}
\newcommand{\diag}[1]{\operatorname{diag}\parens*{#1}}
\newcommand{\tr}[1]{\operatorname{tr}{\parens*{#1}}}
\newcommand{\inprod}[2]{\left\langle#1,#2\right\rangle}
\newcommand{\HadamardProd}{\odot}
\newcommand{\HadamardDiv}{\oslash}
\newcommand{\Span}[1]{\operatorname{span}\bracks*{#1}}
\newcommand{\Ker}[1]{\operatorname{Ker}\parens*{#1}}
\newcommand{\rank}[1]{\operatorname{rank}\parens*{#1}}
% vector
% unit vector
\newcommand{\vix}{\bm{i}_x}
\newcommand{\viy}{\bm{i}_y}
\newcommand{\viz}{\bm{i}_z}
% probability theory
\newcommand{\PDF}[2]{\operatorname{PDF}\bracks*{#1,\;#2}}
\newcommand{\Ber}[1]{\operatorname{Ber}\parens*{#1}}
\newcommand{\Beta}[2]{\operatorname{Beta}\parens*{#1,#2}}
\newcommand{\ExpDist}[1]{\operatorname{ExpDist}\parens*{#1}}
\newcommand{\ErlangDist}[2]{\operatorname{ErlangDist}\parens*{#1,#2}}
\newcommand{\PoissonDist}[1]{\operatorname{PoissonDist}\parens*{#1}}
\newcommand{\GammaDist}[2]{\operatorname{Gamma}\parens*{#1,#2}}
\newcommand{\cind}[2]{\ind{#1\left| #2\right.}} % conditional indicator function
\providecommand{\Pr}{}\renewcommand{\Pr}[1]{\operatorname{Pr}\parens*{#1}}
\DeclarePairedDelimiterX{\cPrParens}[2]{(}{)}{#1\,\delimsize\vert\,#2}
\newcommand{\cPr}[2]{\operatorname{Pr}\cPrParens{#1}{#2}}
\newcommand{\E}[2]{\operatorname{E}_{#1}\bracks*{#2}}
\newcommand{\cE}[3]{\E{#1}{\left.#2\right|#3}}
\newcommand{\Var}[2]{\operatorname{Var}_{#1}\bracks*{#2}}
\newcommand{\Cov}[2]{\operatorname{Cov}\bracks*{#1,#2}}
\newcommand{\CovMat}[1]{\operatorname{Cov}\bracks*{#1}}
% graph theory
\newcommand{\neighborhood}{\mathcal{N}}
% programming
\newcommand{\plpl}{\mathrel{++}}
\newcommand{\pleq}{\mathrel{+}=}
\newcommand{\asteq}{\mathrel{*}=}
\]
はじめに
3次splineを勉強していたらB-splineも見つけたので寄り道してみた。Wikipediaの記事で紹介されている性質の証明を与える。
定義
$\dotsb < t_{-1} < t_0 < t_1 < t_2 < \dotsb$に対して次の漸化式で定められる関数$B_{i,n}$を$n$次のB-spline関数と呼ぶ。
\[ B_{i,0}(t) \coloneqq \begin{cases} 1 & (t_i \leq t < t_{i+1}) \\ 0 & (\text{otherwise}) \end{cases} \] \[ B_{i,n+1}(t) \coloneqq \omega_{i,n}(t)B_{i,n}(t) + (1-\omega_{i+1,n}(t))B_{i+1,n}(t),\quad \omega_{i,n}(t) \coloneqq \frac{t – t_i}{t_{i+n+1} – t_i} \]
この定義から、$B_{i,n}$の台は$[t_i,t_{i+n+1})$であることが判る。
性質
和
次式が成り立つ。
\[\forall t \in [t_j,t_{j+1}), \sum_{i=j-n}^j B_{i,n}(t) = 1\]
$\textit{Proof}$
$n = 0$のときは明らかに成り立つ。$n = m \in \integers$のときに成り立つと仮定して$n = m+1$のときに成り立つことを示す。
\begin{align*}
\sum_{i=j-m-1}^j B_{i,m+1}(t) &= \sum_{i=j-m-1}^j \left[ \omega_{i,m}(t)B_{i,m}(t) + (1-\omega_{i+1,m}(t))B_{i+1,m}(t) \right] \\
&= \omega_{j-m-1,m}(t)B_{j-m-1,m}(t) + \sum_{i=j-m-1}^{j-1} B_{i+1,m}(t) + (1-\omega_{j+1,m}(t))B_{j+1,m}(t)
\end{align*}
$B_{i,n}$の台が$[t_i,t_{i+n+1})$であることから、上式の$B_{j-m-1,m}(t)$と$B_{j+1,m}(t)$は$t \in [t_j,t_{j+1})$に対して0となるので
\[ \sum_{i=j-m-1}^j B_{i,m+1}(t) = \sum_{i=j-m-1}^{j-1} B_{i+1,m}(t) = \sum_{k=j-m}^j B_{k,m}(t) = 0 \]
$\square$
微分
次式が成り立つ。
\[\derivLong{B_{i,k}(t)}{t}{} = k\left(\frac{B_{i,k-1}(t)}{t_{i+k}-t_i} – \frac{B_{i+1,k-1}(t)}{t_{i+1+k} – t_{i+1}}\right)\]
$\textit{Proof}$
$k=1$のときに成り立つことは容易に解る。$k=l \in \naturalNumbers$のときに成り立つと仮定して$k=l+1$のときに成り立つことを示す。
\begin{align*}
\derivLong{B_{i,l+1}(t)}{t}{} &= \derivLong{\left(\frac{t-t_i}{t_{i+l+1}-t_i}B_{i,l}(t) + \frac{t_{i+l+2}-t}{t_{i+l+2}-t_{i+1}}B_{i+1,l}(t)\right)}{t}{} \\
&= \frac{B_{i,l}(t)}{t_{i+l+1}-t_i} + \frac{t-t_i}{t_{i+l+1}-t_i}\derivLong{B_{i,l}(t)}{t}{} – \frac{B_{i+1,l}(t)}{t_{i+l+2} – t_{i+1}} + \frac{t_{i+l+2}-t}{t_{i+l+2} – t_{i+1}}\derivLong{B_{i+1,l}(t)}{t}{} \\
&= \frac{B_{i,l}(t)}{t_{i+l+1}-t_i} – \frac{B_{i+1,l}(t)}{t_{i+l+2} – t_{i+1}} + \underbrace{\frac{t-t_i}{t_{i+l+1}-t_i}l\left(\frac{B_{i,l-1}(t)}{t_{i+l}-t_i} – \frac{B_{i+1,l-1}(t)}{t_{i+1+l} – t_{i+1}}\right)}_{(1)} \\
&\phantom{=} + \underbrace{\frac{t_{i+l+2}-t}{t_{i+l+2} – t_{i+1}}l\left(\frac{B_{i+1,l-1}(t)}{t_{i+1+l}-t_{i+1}} – \frac{B_{i+2,l-1}(t)}{t_{i+2+l} – t_{i+2}}\right)}_{(2)}
\end{align*}
ここで
\begin{align*}
(1) &= \frac{l}{t_{i+l+1}-t_i}\left[\frac{t-t_i}{t_{i+l}-t_i}B_{i,l-1}(t) + \frac{t_{i+1+l} – t}{t_{i+1+l} – t_{i+1}}B_{i+1,l-1}(t) + \frac{t_i – t_{i+1+l}}{t_{i+1+l} – t_{i+1}}B_{i+1,l-1}(t)\right] \\
&= \frac{l}{t_{i+l+1}-t_i}B_{i,l}(t) – \frac{l}{t_{i+1+l} – t_{i+1}}B_{i+1,l-1}(t)
\end{align*}
また
\begin{align*}
(2) &= -\frac{l}{t_{i+l+2} – t_{i+1}}\left[\frac{t – t_{i+1}}{t_{i+1+l}-t_{i+1}}B_{i+1,l-1}(t) + \frac{t_{i+l+2}-t}{t_{i+2+l} – t_{i+2}}B_{i+2,l-1}(t) + \frac{t_{i+1} – t_{i+l+2}}{t_{i+1+l}-t_{i+1}}B_{i+1,l-1}(t)\right] \\
&= -\frac{l}{t_{i+l+2}-t_{i+1}}B_{i+1,l}(t) + \frac{l}{t_{i+1+l} – t_{i+1}}B_{i+1,l-1}(t)
\end{align*}
以上より
\[ \derivLong{B_{i,l+1}(t)}{t}{} = (l+1)\left[\frac{B_{i,l}(t)}{t_{i+l+1}-t_i} – \frac{B_{i+1,l}(t)}{t_{i+l+2} – t_{i+1}}\right] \]
$\square$
投稿者: motchy
An embedded software and FPGA engineer for measuring instrument.
motchy のすべての投稿を表示
