Movatterモバイル変換


[0]ホーム

URL:


Jump to content
WikipediaThe Free Encyclopedia
Search

Non-analytic smooth function

From Wikipedia, the free encyclopedia
Mathematical functions which are smooth but not analytic

Inmathematics,smooth functions (also called infinitelydifferentiable functions) andanalytic functions are two very important types offunctions. One can easily prove that any analytic function of arealargument is smooth. Theconverse is not true, as demonstrated with thecounterexample below.

One of the most important applications of smooth functions withcompact support is the construction of so-calledmollifiers, which are important in theories ofgeneralized functions, such asLaurent Schwartz's theory ofdistributions.

The existence of smooth but non-analytic functions represents one of the main differences betweendifferential geometry andanalytic geometry. In terms ofsheaf theory, this difference can be stated as follows: the sheaf of differentiable functions on adifferentiable manifold isfine, in contrast with the analytic case.

The functions below are generally used to build uppartitions of unity on differentiable manifolds.

An example function

[edit]

Definition of the function

[edit]
The non-analytic smooth functionf(x) considered in the article.

Consider the function

f(x)={e1xif x>0,0if x0,{\displaystyle f(x)={\begin{cases}e^{-{\frac {1}{x}}}&{\text{if }}x>0,\\0&{\text{if }}x\leq 0,\end{cases}}}

defined for everyreal numberx.

The function is smooth

[edit]

The functionf hascontinuousderivatives of all orders at every pointx of thereal line. The formula for these derivatives is

f(n)(x)={pn(x)x2nf(x)if x>0,0if x0,{\displaystyle f^{(n)}(x)={\begin{cases}\displaystyle {\frac {p_{n}(x)}{x^{2n}}}\,f(x)&{\text{if }}x>0,\\0&{\text{if }}x\leq 0,\end{cases}}}

wherepn(x) is apolynomial ofdegreen − 1 givenrecursively byp1(x) = 1 and

pn+1(x)=x2pn(x)(2nx1)pn(x){\displaystyle p_{n+1}(x)=x^{2}p_{n}'(x)-(2nx-1)p_{n}(x)}

for any positiveintegern. From this formula, it is not completely clear that the derivatives are continuous at 0; this follows from theone-sided limit

limx0e1xxm=0{\displaystyle \lim _{x\searrow 0}{\frac {e^{-{\frac {1}{x}}}}{x^{m}}}=0}

for anynonnegative integerm.

Detailed proof of smoothness

By thepower series representation of the exponential function, we have for everynatural numberm{\displaystyle m} (including zero)

1xm=x(1x)m+1(m+1)!xn=01n!(1x)n=(m+1)!xe1x,x>0,{\displaystyle {\frac {1}{x^{m}}}=x{\Bigl (}{\frac {1}{x}}{\Bigr )}^{m+1}\leq (m+1)!\,x\sum _{n=0}^{\infty }{\frac {1}{n!}}{\Bigl (}{\frac {1}{x}}{\Bigr )}^{n}=(m+1)!\,xe^{\frac {1}{x}},\qquad x>0,}

because all the positive terms fornm+1{\displaystyle n\neq m+1} are added. Therefore, dividing this inequality bye1x{\displaystyle e^{\frac {1}{x}}} and taking thelimit from above,

limx0e1xxm(m+1)!limx0x=0.{\displaystyle \lim _{x\searrow 0}{\frac {e^{-{\frac {1}{x}}}}{x^{m}}}\leq (m+1)!\lim _{x\searrow 0}x=0.}

We now prove the formula for thenth derivative off bymathematical induction. Using thechain rule, thereciprocal rule, and the fact that the derivative of the exponential function is again the exponential function, we see that the formula is correct for the first derivative off for allx > 0 and thatp1(x) is a polynomial of degree 0. Of course, the derivative off is zero forx < 0.It remains to show that the right-hand side derivative off atx = 0 is zero. Using the above limit, we see that

f(0)=limx0f(x)f(0)x0=limx0e1xx=0.{\displaystyle f'(0)=\lim _{x\searrow 0}{\frac {f(x)-f(0)}{x-0}}=\lim _{x\searrow 0}{\frac {e^{-{\frac {1}{x}}}}{x}}=0.}

The induction step fromn ton + 1 is similar. Forx > 0 we get for the derivative

f(n+1)(x)=(pn(x)x2n2npn(x)x2n+1+pn(x)x2n+2)f(x)=x2pn(x)(2nx1)pn(x)x2n+2f(x)=pn+1(x)x2(n+1)f(x),{\displaystyle {\begin{aligned}f^{(n+1)}(x)&={\biggl (}{\frac {p'_{n}(x)}{x^{2n}}}-2n{\frac {p_{n}(x)}{x^{2n+1}}}+{\frac {p_{n}(x)}{x^{2n+2}}}{\biggr )}f(x)\\&={\frac {x^{2}p'_{n}(x)-(2nx-1)p_{n}(x)}{x^{2n+2}}}f(x)\\&={\frac {p_{n+1}(x)}{x^{2(n+1)}}}f(x),\end{aligned}}}

wherepn+1(x) is a polynomial of degreen = (n + 1) − 1. Of course, the (n + 1)st derivative off is zero forx < 0. For the right-hand side derivative off (n) atx = 0 we obtain with the above limit

limx0f(n)(x)f(n)(0)x0=limx0pn(x)x2n+1e1/x=0.{\displaystyle \lim _{x\searrow 0}{\frac {f^{(n)}(x)-f^{(n)}(0)}{x-0}}=\lim _{x\searrow 0}{\frac {p_{n}(x)}{x^{2n+1}}}\,e^{-1/x}=0.}

The function is not analytic

[edit]

As seen earlier, the functionf is smooth, and all its derivatives at theorigin are 0. Therefore, theTaylor series off at the origin converges everywhere to thezero function,

n=0f(n)(0)n!xn=n=00n!xn=0,xR,{\displaystyle \sum _{n=0}^{\infty }{\frac {f^{(n)}(0)}{n!}}x^{n}=\sum _{n=0}^{\infty }{\frac {0}{n!}}x^{n}=0,\qquad x\in \mathbb {R} ,}

and so the Taylor series does not equalf(x) forx > 0. Consequently,f is notanalytic at the origin.

Smooth transition functions

[edit]
The smooth transitiong from 0 to 1 defined here.

The function

g(x)=f(x)f(x)+f(1x),xR,{\displaystyle g(x)={\frac {f(x)}{f(x)+f(1-x)}},\qquad x\in \mathbb {R} ,}

has a strictly positive denominator everywhere on the real line, henceg is also smooth. Furthermore,g(x) = 0 forx ≤ 0 andg(x) = 1 forx ≥ 1, hence it provides a smooth transition from the level 0 to the level 1 in theunit interval [0, 1]. To have the smooth transition in the real interval [a,b] witha < b, consider the function

Rxg(xaba).{\displaystyle \mathbb {R} \ni x\mapsto g{\Bigl (}{\frac {x-a}{b-a}}{\Bigr )}.}

For real numbersa <b <c <d, the smooth function

Rxg(xaba)g(dxdc){\displaystyle \mathbb {R} \ni x\mapsto g{\Bigl (}{\frac {x-a}{b-a}}{\Bigr )}\,g{\Bigl (}{\frac {d-x}{d-c}}{\Bigr )}}

equals 1 on the closed interval [b,c] and vanishes outside the open interval (a,d), hence it can serve as abump function.

A smooth function that is nowhere real analytic

[edit]
Approximation of the smooth-everywhere, but nowhere-analytic function mentioned here. This partial sum is taken fromk = 20 to 2500.

A morepathological example is an infinitely differentiable function which is not analyticat any point. It can be constructed by means of aFourier series as follows. Define for allxR{\displaystyle x\in \mathbb {R} }

F(x):=kNe2kcos(2kx) .{\displaystyle F(x):=\sum _{k\in \mathbb {N} }e^{-{\sqrt {2^{k}}}}\cos(2^{k}x)\ .}

Since the serieskNe2k(2k)n{\displaystyle \sum _{k\in \mathbb {N} }e^{-{\sqrt {2^{k}}}}{(2^{k})}^{n}} converges for allnN{\displaystyle n\in \mathbb {N} }, this function is easily seen to be of class C, by a standard inductive application of theWeierstrass M-test to demonstrateuniform convergence of each series of derivatives.

We now show thatF(x){\displaystyle F(x)} is not analytic at anydyadic rational multiple of π, that is, at anyx:=πp2q{\displaystyle x:=\pi \cdot p\cdot 2^{-q}} withpZ{\displaystyle p\in \mathbb {Z} } andqN{\displaystyle q\in \mathbb {N} }. Since the sum of the firstq{\displaystyle q} terms is analytic, we need only considerF>q(x){\displaystyle F_{>q}(x)}, the sum of the terms withk>q{\displaystyle k>q}. For all orders of derivationn=2m{\displaystyle n=2^{m}} withmN{\displaystyle m\in \mathbb {N} },m2{\displaystyle m\geq 2} andm>q/2{\displaystyle m>q/2} we have

F>q(n)(x):=kNk>qe2k(2k)ncos(2kx)=kNk>qe2k(2k)nenn2n(asn){\displaystyle F_{>q}^{(n)}(x):=\sum _{k\in \mathbb {N} \atop k>q}e^{-{\sqrt {2^{k}}}}{(2^{k})}^{n}\cos(2^{k}x)=\sum _{k\in \mathbb {N} \atop k>q}e^{-{\sqrt {2^{k}}}}{(2^{k})}^{n}\geq e^{-n}n^{2n}\quad (\mathrm {as} \;n\to \infty )}

where we used the fact thatcos(2kx)=1{\displaystyle \cos(2^{k}x)=1} for all2k>2q{\displaystyle 2^{k}>2^{q}}, and we bounded the first sum from below by the term with2k=22m=n2{\displaystyle 2^{k}=2^{2m}=n^{2}}. As a consequence, at any suchxR{\displaystyle x\in \mathbb {R} }

lim supn(|F>q(n)(x)|n!)1/n=+,{\displaystyle \limsup _{n\to \infty }\left({\frac {|F_{>q}^{(n)}(x)|}{n!}}\right)^{1/n}=+\infty \,,}

so that theradius of convergence of theTaylor series ofF>q{\displaystyle F_{>q}} atx{\displaystyle x} is 0 by theCauchy-Hadamard formula. Since the set of analyticity of a function is an open set, and since dyadic rationals aredense, we conclude thatF>q{\displaystyle F_{>q}}, and henceF{\displaystyle F}, is nowhere analytic inR{\displaystyle \mathbb {R} }.

Application to Taylor series

[edit]
Main article:Borel's lemma

For every sequence α0, α1, α2, . . . of real orcomplex numbers, the following construction shows the existence of a smooth functionF on the real line which has these numbers as derivatives at the origin.[1] In particular, every sequence of numbers can appear as the coefficients of theTaylor series of a smooth function. This result is known asBorel's lemma, afterÉmile Borel.

With the smooth transition functiong as above, define

h(x)=g(2+x)g(2x),xR.{\displaystyle h(x)=g(2+x)\,g(2-x),\qquad x\in \mathbb {R} .}

This functionh is also smooth; it equals 1 on the closed interval [−1,1] and vanishes outside the open interval (−2,2). Usingh, define for every natural numbern (including zero) the smooth function

ψn(x)=xnh(x),xR,{\displaystyle \psi _{n}(x)=x^{n}\,h(x),\qquad x\in \mathbb {R} ,}

which agrees with themonomialxn on [−1,1] and vanishes outside the interval (−2,2). Hence, thek-th derivative ofψn at the origin satisfies

ψn(k)(0)={n!if k=n,0otherwise,k,nN0,{\displaystyle \psi _{n}^{(k)}(0)={\begin{cases}n!&{\text{if }}k=n,\\0&{\text{otherwise,}}\end{cases}}\quad k,n\in \mathbb {N} _{0},}

and theboundedness theorem implies thatψn and every derivative ofψn is bounded. Therefore, the constants

λn=max{1,|αn|,ψn,ψn(1),,ψn(n)},nN0,{\displaystyle \lambda _{n}=\max {\bigl \{}1,|\alpha _{n}|,\|\psi _{n}\|_{\infty },\|\psi _{n}^{(1)}\|_{\infty },\ldots ,\|\psi _{n}^{(n)}\|_{\infty }{\bigr \}},\qquad n\in \mathbb {N} _{0},}

involving thesupremum norm ofψn and its firstn derivatives, are well-defined real numbers. Define the scaled functions

fn(x)=αnn!λnnψn(λnx),nN0,xR.{\displaystyle f_{n}(x)={\frac {\alpha _{n}}{n!\,\lambda _{n}^{n}}}\psi _{n}(\lambda _{n}x),\qquad n\in \mathbb {N} _{0},\;x\in \mathbb {R} .}

By repeated application of thechain rule,

fn(k)(x)=αnn!λnnkψn(k)(λnx),k,nN0,xR,{\displaystyle f_{n}^{(k)}(x)={\frac {\alpha _{n}}{n!\,\lambda _{n}^{n-k}}}\psi _{n}^{(k)}(\lambda _{n}x),\qquad k,n\in \mathbb {N} _{0},\;x\in \mathbb {R} ,}

and, using the previous result for thek-th derivative ofψn at zero,

fn(k)(0)={αnif k=n,0otherwise,k,nN0.{\displaystyle f_{n}^{(k)}(0)={\begin{cases}\alpha _{n}&{\text{if }}k=n,\\0&{\text{otherwise,}}\end{cases}}\qquad k,n\in \mathbb {N} _{0}.}

It remains to show that the function

F(x)=n=0fn(x),xR,{\displaystyle F(x)=\sum _{n=0}^{\infty }f_{n}(x),\qquad x\in \mathbb {R} ,}

is well defined and can be differentiated term-by-term infinitely many times.[2] To this end, observe that for everyk

n=0fn(k)n=0k+1|αn|n!λnnkψn(k)+n=k+21n!1λnnk21|αn|λn1ψn(k)λn1<,{\displaystyle \sum _{n=0}^{\infty }\|f_{n}^{(k)}\|_{\infty }\leq \sum _{n=0}^{k+1}{\frac {|\alpha _{n}|}{n!\,\lambda _{n}^{n-k}}}\|\psi _{n}^{(k)}\|_{\infty }+\sum _{n=k+2}^{\infty }{\frac {1}{n!}}\underbrace {\frac {1}{\lambda _{n}^{n-k-2}}} _{\leq \,1}\underbrace {\frac {|\alpha _{n}|}{\lambda _{n}}} _{\leq \,1}\underbrace {\frac {\|\psi _{n}^{(k)}\|_{\infty }}{\lambda _{n}}} _{\leq \,1}<\infty ,}

where the remaining infinite series converges by theratio test.

Application to higher dimensions

[edit]
The function Ψ1(x) in one dimension.

For every radiusr > 0,

RnxΨr(x)=f(r2x2){\displaystyle \mathbb {R} ^{n}\ni x\mapsto \Psi _{r}(x)=f(r^{2}-\|x\|^{2})}

withEuclidean norm ||x|| defines a smooth function onn-dimensionalEuclidean space withsupport in theball of radiusr, butΨr(0)>0{\displaystyle \Psi _{r}(0)>0}.

Complex analysis

[edit]

This pathology cannot occur with differentiablefunctions of a complex variable rather than of a real variable. Indeed, allholomorphic functions are analytic, so that the failure of the functionf defined in this article to be analytic in spite of its being infinitely differentiable is an indication of one of the most dramatic differences between real-variable and complex-variable analysis.

Note that although the functionf has derivatives of all orders over the real line, theanalytic continuation off from the positive half-linex > 0 to thecomplex plane, that is, the function

C{0}ze1zC,{\displaystyle \mathbb {C} \setminus \{0\}\ni z\mapsto e^{-{\frac {1}{z}}}\in \mathbb {C} ,}

has anessential singularity at the origin, and hence is not even continuous, much less analytic. By thegreat Picard theorem, it attains every complex value (with the exception of zero) infinitely many times in every neighbourhood of the origin.

See also

[edit]

Notes

[edit]
  1. ^Exercise 12 on page 418 inWalter Rudin,Real and Complex Analysis. McGraw-Hill, New Delhi 1980,ISBN 0-07-099557-5
  2. ^See e.g. Chapter V, Section 2, Theorem 2.8 and Corollary 2.9 about the differentiability of the limits of sequences of functions inAmann, Herbert; Escher, Joachim (2005),Analysis I, Basel:Birkhäuser Verlag, pp. 373–374,ISBN 3-7643-7153-6

External links

[edit]
Retrieved from "https://en.wikipedia.org/w/index.php?title=Non-analytic_smooth_function&oldid=1264871065"
Category:
Hidden categories:

[8]ページ先頭

©2009-2025 Movatter.jp