% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % % % % The Project Gutenberg EBook of Orders of Infinity, by Godfrey Harold Hardy % % % This eBook is for the use of anyone anywhere at no cost and with % % almost no restrictions whatsoever. You may copy it, give it away or % % re-use it under the terms of the Project Gutenberg License included % % with this eBook or online at www.gutenberg.org % % % % % % Title: Orders of Infinity % % The 'Infinitärcalcül' of Paul Du Bois-Reymond % % % % Author: Godfrey Harold Hardy % % % % Release Date: November 25, 2011 [EBook #38079] % % Most recently updated: June 11, 2021 % % % % Language: English % % % % Character set encoding: UTF-8 % % % % *** START OF THIS PROJECT GUTENBERG EBOOK ORDERS OF INFINITY *** % % % % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % \def\ebook{38079} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% %% %% Packages and substitutions: %% %% %% %% book: Required. %% %% inputenc: Standard DP encoding. Required. %% %% %% %% ifthen: Logical conditionals. Required. %% %% %% %% amsmath: AMS mathematics enhancements. Required. %% %% amssymb: Additional mathematical symbols. Required. %% %% %% %% alltt: Fixed-width font environment. Required. %% %% %% %% footmisc: Extended footnote capabilities. Required. %% %% %% %% indentfirst: Indent first word of each sectional unit. Required. %% %% icomma: Make the comma a decimal separator in math. Required. %% %% %% %% calc: Length calculations. Required. %% %% %% %% fancyhdr: Enhanced running headers and footers. Required. %% %% %% %% graphicx: Standard interface for graphics inclusion. Required. %% %% caption: Caption customization. Required. %% %% %% %% geometry: Enhanced page layout package. Required. %% %% hyperref: Hypertext embellishments for pdf output. Required. %% %% %% %% %% %% Producer's Comments: %% %% %% %% Changes are noted in this file in two ways. %% %% 1. \DPtypo{}{} for typographical corrections, showing original %% %% and replacement text side-by-side. %% %% 2. [** TN: Note]s for lengthier or stylistic comments. %% %% %% %% %% %% Compilation Flags: %% %% %% %% The following behavior may be controlled by boolean flags. %% %% %% %% ForPrinting (false by default): %% %% Compile a screen-optimized PDF file (pages cropped, one-sided, %% %% blue hyperlinks). Set to true for print-optimized file. %% %% %% %% Modernize (true by default): %% %% Modernize the mathematical notation (see below for details). %% %% %% %% %% %% PDF pages: 101(if ForPrinting set to false) %% %% PDF page size: 5.25 x 7.875in (non-standard) %% %% %% %% Images: 1 png diagram (CUP device), 5 PDF diagrams %% %% %% %% Summary of log file: %% %% * Two overfull hboxes, three underfull hboxes. %% %% * Three underfull vboxes. %% %% %% %% %% %% Compile History: %% %% %% %% November, 2011: adhere (Andrew D. Hwang) %% %% texlive2007, GNU/Linux %% %% %% %% Command block: %% %% %% %% pdflatex x2 %% %% %% %% %% %% November 2011: pglatex. %% %% Compile this project with: %% %% pdflatex 38079-t.tex ..... TWO times %% %% %% %% pdfTeX, Version 3.1415926-1.40.10 (TeX Live 2009/Debian) %% %% %% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \listfiles \documentclass[12pt]{book}[2005/09/16] %%%%%%%%%%%%%%%%%%%%%%%%%%%%% PACKAGES %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \usepackage[utf8]{inputenc}[2006/05/05] \usepackage{ifthen}[2001/05/26] %% Logical conditionals \usepackage{amsmath}[2000/07/18] %% Displayed equations \usepackage{amssymb}[2002/01/22] %% and additional symbols \usepackage{alltt}[1997/06/16] %% boilerplate, credits, license %% extended footnote capabilities \usepackage[symbol,perpage]{footmisc}[2005/03/17] \usepackage{indentfirst}[1995/11/23] \usepackage{icomma}[2002/03/10] \usepackage{calc}[2005/08/06] \IfFileExists{yfonts.sty}{% \usepackage{yfonts}% }{% \providecommand{\textgoth}[1]{\textbf{#1}}% } % for running heads \usepackage{fancyhdr} \usepackage{graphicx}[1999/02/16]%% For diagrams \usepackage[labelformat=empty,textfont=small]{caption}[2007/01/07] % Modernize notation: Use square root signs instead of surds, square % brackets for closed intervals. \newboolean{Modernize} %% COMMENT the line below to revert to the original notation. %% (This switch does not affect typographical corrections.) \setboolean{Modernize}{true} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%% Interlude: Set up PRINTING (default) or SCREEN VIEWING %%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % ForPrinting=true (default) false % Asymmetric margins Symmetric margins % Black hyperlinks Blue hyperlinks % Start Preface, ToC, etc. recto No blank verso pages % \newboolean{ForPrinting} %% UNCOMMENT the next line for a PRINT-OPTIMIZED VERSION of the text %% %\setboolean{ForPrinting}{true} %% Initialize values to ForPrinting=false \newcommand{\ChapterSpace}{} \newcommand{\Margins}{hmarginratio=1:1} % Symmetric margins \newcommand{\HLinkColor}{blue} % Hyperlink color \newcommand{\PDFPageLayout}{SinglePage} \newcommand{\TransNote}{Transcriber's Note} \newcommand{\TransNoteCommon}{% Minor typographical corrections and presentational changes have been made without comment. All changes are detailed in the \LaTeX\ source file, which may be downloaded from \begin{center} \texttt{www.gutenberg.org/ebooks/\ebook}. \end{center} \bigskip } \newcommand{\TransNoteText}{% \TransNoteCommon This PDF file is optimized for screen viewing, but may easily be recompiled for printing. Please consult the preamble of the \LaTeX\ source file for instructions. } %% Re-set if ForPrinting=true \ifthenelse{\boolean{ForPrinting}}{% \renewcommand{\ChapterSpace}{\vspace*{1in}} \renewcommand{\Margins}{hmarginratio=2:3} % Asymmetric margins \renewcommand{\HLinkColor}{black} % Hyperlink color \renewcommand{\PDFPageLayout}{TwoPageRight} \renewcommand{\TransNoteText}{% \TransNoteCommon This PDF file is optimized for printing, but may easily be recompiled for screen viewing. Please consult the preamble of the \LaTeX\ source file for instructions. } }{% If ForPrinting=false, don't skip to recto \renewcommand{\cleardoublepage}{\clearpage} } %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%% End of PRINTING/SCREEN VIEWING code; back to packages %%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \ifthenelse{\boolean{ForPrinting}}{% \setlength{\paperwidth}{8.5in}% \setlength{\paperheight}{11in}% \usepackage[body={5in,8.5in},\Margins]{geometry}[2002/07/08] }{% \setlength{\paperwidth}{5.25in}% \setlength{\paperheight}{7.875in}% \usepackage[body={5in,7in},\Margins,includeheadfoot]{geometry}[2002/07/08] } \providecommand{\ebook}{00000} % Overridden during white-washing \usepackage[pdftex, hyperfootnotes=false, pdftitle={The Project Gutenberg eBook \#\ebook: Orders of Infinity}, pdfauthor={Godfrey Harold Hardy}, pdfkeywords={Andrew D. Hwang, Brenda Lewis, Project Gutenberg Online Distributed Proofreading Team, Internet Archive/American Libraries}, pdfstartview=Fit, % default value pdfstartpage=1, % default value pdfpagemode=UseNone, % default value bookmarks=true, % default value linktocpage=false, % default value pdfpagelayout=\PDFPageLayout, pdfdisplaydoctitle, pdfpagelabels=true, bookmarksopen=true, bookmarksopenlevel=1, colorlinks=true, linkcolor=\HLinkColor]{hyperref}[2007/02/07] %%%% Fixed-width environment to format PG boilerplate %%%% \newenvironment{PGtext}{% \begin{alltt} \fontsize{9.2}{10.5}\ttfamily\selectfont}% {\end{alltt}} %%%% Global style parameters %%%% % No hrule in page header \renewcommand{\headrulewidth}{0pt} \setlength{\headheight}{15pt} % Loosen horizontal spacing \setlength{\emergencystretch}{1.5em} % Local spacing coercion \newcommand{\Loosen}{\spaceskip 0.375em plus 0.75em minus 0.25em} \newcommand{\Squeeze}[2][0.9]{\scalebox{#1}[1]{#2}} \DeclareMathSizes{12}{12}{9}{7} % Misc spacing parameters \newcommand{\Medskip}{\vspace{4pt plus 0.5\baselineskip minus 2pt}} % "Scratch pad" for length calculations \newlength{\TmpLen} %% Parametrized vertical space %% \newcommand{\Strut}[1][16pt]{\rule{0pt}{#1}} %%%% Corrections and in-line transcriber's notes %%%% % Errors \newcommand{\DPtypo}[2]{#2} %%%% Notational modernizations %%%% \ifthenelse{\boolean{Modernize}}{% % Original uses interchangeable symbols for \leq; make consistent \renewcommand{\leqq}{\leqslant} \renewcommand{\geqq}{\geqslant} }{% Modernize = false % Use surd sign... \let\oldsqrt=\sqrt% \renewcommand*{\sqrt}[2][\ ]{\oldsqrt[#1]{\vphantom{#2}}#2} } %%%% Running heads %%%% \newcommand{\FlushRunningHeads}{% \clearpage \pagestyle{fancy} \fancyhf{} \cleardoublepage \thispagestyle{empty} \InitRunningHeads } \newcommand{\InitRunningHeads}{% \ifthenelse{\boolean{ForPrinting}} {\fancyhead[RO,LE]{\thepage}} {\fancyhead[R]{\thepage}} } \newcommand{\SetCenterHeads}[1]{% \fancyhead[C]{{\footnotesize\MakeUppercase{#1}}}% } \newcommand{\BookMark}[2]{\phantomsection\pdfbookmark[#1]{#2}{#2}} %%%% Major document divisions %%%% \newcommand{\FrontMatter}{% \cleardoublepage \frontmatter \BookMark{-1}{Front Matter} } \newcommand{\PGBoilerPlate}{% \pagenumbering{Alph} \pagestyle{empty} \BookMark{0}{PG Boilerplate} } \newcommand{\MainMatter}{% \FlushRunningHeads \mainmatter \BookMark{-1}{Main Matter} } \newcommand{\BackMatter}{% \FlushRunningHeads \backmatter \BookMark{-1}{Back Matter} } \newcommand{\PGLicense}{% \FlushRunningHeads \pagenumbering{Roman} \BookMark{-1}{PG License} \SetCenterHeads{License} } \newcommand{\TranscribersNote}[1]{% \begin{minipage}{0.85\textwidth} \small \BookMark{0}{Transcriber's Note} \subsection*{\centering\normalfont\scshape\normalsize\TransNote} % \raggedright #1 \end{minipage} } %%%% Table of Contents %%%% % Contents heading \newcommand{\Contents}{% \FlushRunningHeads \SetCenterHeads{CONTENTS} \BookMark{0}{Contents} \thispagestyle{empty} \section*{\centering CONTENTS} } \newcommand{\PageLine}{\null\hfill{\scriptsize PAGE}} % Chapter entries; macro discards third argument (original page number) \newcommand{\ToCChap}[3]{% \settowidth{\TmpLen}{VIII\@.\ }% \medskip\par\noindent\parbox[b]{\textwidth-1.25em}{% \Strut\small\makebox[\TmpLen][c]{#1}\hangindent2\TmpLen% #2\ \dotfill}\ToCPage{chapter:#1}% } % Appendix entries \newcommand{\ToCApp}[3]{% \settowidth{\TmpLen}{ III\@.\ }% \medskip\par\noindent\parbox[b]{\textwidth-1.25em}{% \Strut\small\textsc{Appendix}\makebox[\TmpLen][c]{#1}\hangindent2\TmpLen% #2\ \dotfill}\ToCPage{appendix:#1}% } % Page numbers \newcommand{\ToCPage}[1]{% \makebox[1.25em][r]{\small\pageref{#1}}% } %%%% Document Sectioning %%%% \newcommand{\ChapterHead}[2]{% \ChapterSpace \section*{\centering #1} \subsection*{\centering\normalsize\normalfont\MakeUppercase{#2}} } \newcommand{\Preface}{% \cleardoublepage \pagestyle{fancy} \thispagestyle{empty} \section*{\centering PREFACE}} % \Chapter[Running head]{Number}{Heading title} \newcommand{\Chapter}[3][]{% \ifthenelse{\equal{#2}{I.}}{\FlushRunningHeads\thispagestyle{empty}}{} \phantomsection \label{chapter:#2} \BookMark{0}{Chapter #2}% \ifthenelse{\equal{#1}{}}{% \SetCenterHeads{#3}% }{% \SetCenterHeads{#1}% } \ChapterHead{#2}{#3} } \newcommand{\Section}[2][]{% \ifthenelse{\equal{#1}{}}{% \subsection*{\centering\normalsize\normalfont\itshape #2} }{% \subsection*{\centering\normalsize\normalfont#1\quad \itshape #2} }% } %\Appendix{Number}{Title} \newcommand{\Appendix}[3][]{% \FlushRunningHeads \phantomsection \label{appendix:#2} \BookMark{0}{Appendix #2} \ifthenelse{\equal{#1}{}}{% \SetCenterHeads{#3}% }{% \SetCenterHeads{#1}% } \ChapterHead{APPENDIX #2}{#3} } \newcommand{\Subsection}[2]{\paragraph*{\indent\normalfont#1} \textit{#2}} \newcommand{\Paragraph}[1]{\paragraph*{\indent #1}} \newcommand{\Titlefont}[1]{\textbf{\LARGE #1}} %%%% Other semantic units %%%% % Numbered item \newcommand{\Item}[1]{\makebox[2em][l]{\normalfont\upshape#1}\Strut[14pt]} % Document-level environments % Italicized constructs \newenvironment{Result}[1][]{% \ifthenelse{\not\equal{#1}{}}{% \Paragraph{#1}}{}% \itshape\ignorespaces}{\normalfont} \newenvironment{Example}{\itshape\ignorespaces}{\normalfont\ignorespacesafterend} \newenvironment{Lemma}[1][]{\begin{Result}[Lemma.]}{\end{Result}} \newenvironment{Theorem}{\begin{Result}[Theorem.]}{\end{Result}} % Passages of small text having no special run-in heading \newenvironment{Remark}{\Medskip\par\small}{\normalsize\Medskip} \newcommand{\Signature}[3]{% \null\hfill#1\hspace*{3\parindent}\\ \hspace*{\parindent}#2 \\ \hspace*{5\parindent}#3 \\ } %%%% Misc. textual macros %%%% \newcommand{\First}[1]{\textsc{#1}} \newcommand{\continued}{{\normalfont\textit{continued}}} % Table entries \newcommand{\CCEntry}[1]{\multicolumn{1}{|c|}{#1\rule[-12pt]{0pt}{36pt}}} \newcommand{\CEntry}[1]{\multicolumn{1}{c|}{#1\rule[-12pt]{0pt}{36pt}}} \newcommand{\Dash}{\multicolumn{1}{c|}{\rule[0.5ex]{24pt}{0.5pt}}} % Bibliography \newcommand{\Author}[1]{% \noindent\textsc{#1}\hangindent\parindent% } \newcommand{\Same}{% \noindent\rule[0.5ex]{\parindent}{0.5pt}\ \hangindent\parindent% } \newcommand{\Work}{% \hspace*{\parindent}\hangindent\parindent% } % Catalog \newcommand{\Catalog}[1]{% \noindent\setlength{\TmpLen}{2\parindent}% \makebox[\TmpLen][l]{#1}\hangindent5\parindent% } \newcommand{\Inprep}{% \hspace*{\parindent}\hangindent5\parindent% } %%%% Illustrations %%%% \newcommand{\Graphic}[3][pdf]{\includegraphics[width=#2]{./images/#3.#1}} %\Figure[width]{Figure number}{File name} \newcommand{\Figure}[3][0.5\textwidth]{% \begin{figure}[hbt!] \centering \Graphic{#1}{fig#2} \caption{\textsc{Fig.}~#2.} \label{fig:#2} \end{figure}\ignorespaces% } %%%% Cross-referencing %%%% %% Anchors \newcommand{\PageLabel}[2][page]{\phantomsection\label{#1:#2}} %% Links \newcommand{\PageRef}[2]{\hyperref[page:#2]{#1~\pageref*{page:#2}}} \newcommand{\Fig}[1]{\hyperref[fig:#1]{Fig.~#1}} % Code stubs; no hyperlinking \newcommand{\Tag}[1]{\tag*{\ensuremath{#1}}} \newcommand{\LTag}[1]{\llap{\ensuremath{#1}\qquad\qquad}} \newcommand{\RTag}[2][\qquad\qquad]{\rlap{#1#2}} \newcommand{\ITag}[1]{\ensuremath{#1}\ \ } \newcommand{\Eq}[1]{\ensuremath{#1}} \newcommand{\Ref}[2]{\textsc{#1}\ifthenelse{\not\equal{#2}{}}{\upshape~#2}{}} %%%% Typographical conveniences %%%% \newcommand{\eg}{\textit{e.g.}} \newcommand{\ie}{\textit{i.e.}} \newcommand{\lc}{\textit{l.c.}} \renewcommand{\(}{{\upshape(}} \renewcommand{\)}{{\upshape)}} %%%% Misc. mathematical macros %%%% \newcommand{\ds}{\displaystyle} \newcommand{\dd}{\partial} \newcommand{\ff}{f\!f} \newcommand{\lf}{\mathit{lf}} \newcommand{\llf}{\mathit{llf}} \DeclareMathOperator{\Li}{\mathit{Li}} \renewcommand{\bar}[1]{\overline{\Strut[9pt]#1}} % Binary relations \renewcommand{\leq}{\leqslant} \renewcommand{\geq}{\geqslant} \newcommand{\eqq}{\equiv} \newcommand{\ceq}{\asymp} \newcommand{\clt}{\prec} \newcommand{\cgt}{\succ} \newcommand{\ceqq}{\mathbin{\usebox{\ceqqsymb}}} \newcommand{\cleq}{\preccurlyeq} \newcommand{\cgeq}{\succcurlyeq} % Define \ceqq \newbox{\ceqqsymb} \savebox{\ceqqsymb}{\settowidth{\TmpLen}{\ensuremath{\asymp}}% \makebox[\TmpLen][c]{\makebox[0pt][c]{$\asymp$}\makebox[0pt][c]{$-$}}% } % Handle Latin-1 center dot character \DeclareUnicodeCharacter{00B0}{{}^\circ} \DeclareUnicodeCharacter{00B1}{\pm} \DeclareUnicodeCharacter{00B7}{\cdot} \DeclareUnicodeCharacter{00D7}{\times} \DeclareUnicodeCharacter{00F7}{\div} %%%%%%%%%%%%%%%%%%%%%%%% START OF DOCUMENT %%%%%%%%%%%%%%%%%%%%%%%%%% \begin{document} \FrontMatter %%%% PG BOILERPLATE %%%% \PGBoilerPlate \begin{center} \begin{minipage}{\textwidth} \small \begin{PGtext} The Project Gutenberg EBook of Orders of Infinity, by Godfrey Harold Hardy This eBook is for the use of anyone anywhere at no cost and with almost no restrictions whatsoever. You may copy it, give it away or re-use it under the terms of the Project Gutenberg License included with this eBook or online at www.gutenberg.org Title: Orders of Infinity The 'Infinitärcalcül' of Paul Du Bois-Reymond Author: Godfrey Harold Hardy Release Date: November 25, 2011 [EBook #38079] Most recently updated: June 11, 2021 Language: English Character set encoding: UTF-8 *** START OF THIS PROJECT GUTENBERG EBOOK ORDERS OF INFINITY *** \end{PGtext} \end{minipage} \end{center} \clearpage %%%% Credits and transcriber's note %%%% \begin{center} \begin{minipage}{\textwidth} \begin{PGtext} Produced by Andrew D. Hwang, Brenda Lewis and the Online Distributed Proofreading Team at http://www.pgdp.net (This file was produced from images generously made available by The Internet Archive/Canadian Libraries) \end{PGtext} \end{minipage} \vfill \TranscribersNote{\TransNoteText} \end{center} %%%%%%%%%%%%%%%%%%%%%%%%%%% FRONT MATTER %%%%%%%%%%%%%%%%%%%%%%%%%% %% -----File: 001.png---Folio xx------- \cleardoublepage \pagenumbering{roman} \null\vfill \begin{center} \Titlefont{Cambridge Tracts in Mathematics \\[12pt] and Mathematical Physics} \bigskip \textsc{General Editors} \medskip J. G. LEATHEM, M.A. \\ E. T. WHITTAKER, M.A., F.R.S. \vfill \Titlefont{No.\ 12 \\[24pt] ORDERS OF INFINITY} \end{center} %% -----File: 002.png---Folio xx------- \clearpage \begin{center} \large CAMBRIDGE UNIVERSITY PRESS \\ \textgoth{London}: FETTER LANE, E.C. \\ C. F. CLAY, \textsc{Manager} \bigskip \Graphic[png]{1.25in}{cups} \bigskip \normalsize \textgoth{Edinburgh}: 100, PRINCES STREET \\ \textgoth{Berlin}: A. ASHER AND CO. \\ \textgoth{Leipzig}: F. A. BROCKHAUS \\ \textgoth{New York}: G. P. PUTNAM'S SONS \\ \textgoth{Bombay and Calcutta}: MACMILLAN AND CO., \textsc{Ltd.} \vfill \textit{All rights reserved} \end{center} %% -----File: 003.png---Folio xx------- \clearpage \begin{center} \Titlefont{\Huge ORDERS OF INFINITY} \bigskip {\large THE `INFINITÄRCALCÜL' OF \\[8pt] PAUL DU BOIS-REYMOND} \vfill \vfill by \bigskip G. H. HARDY, M.A., F.R.S. \\ \medskip {\small Fellow and Lecturer of Trinity College, Cambridge} \vfill \vfill \vfill {\large Cambridge: \\ at the University Press 1910} \end{center} %% -----File: 004.png---Folio xx------- \clearpage \null\vfill \begin{center} \textgoth{Cambridge}: \medskip \footnotesize PRINTED BY JOHN CLAY, M.A. \medskip AT THE UNIVERSITY PRESS \end{center} \vfill %% -----File: 005.png---Folio xx------- \Preface \First{The} ideas of Du~Bois-Reymond's \textit{Infinitärcalcül} are of great and growing importance in all branches of the theory of functions. With the particular system of notation that he invented, it is, no doubt, quite possible to dispense; but it can hardly be denied that the notation is exceedingly useful, being clear, concise, and expressive in a very high degree. In any case Du~Bois-Reymond was a mathematician of such power and originality that it would be a great pity if so much of his best work were allowed to be forgotten. There is, in Du~Bois-Reymond's original memoirs, a good deal that would not be accepted as conclusive by modern analysts. He is also at times exceedingly obscure; his work would beyond doubt have attracted much more attention had it not been for the somewhat repugnant garb in which he was unfortunately wont to clothe his most valuable ideas. I have therefore attempted, in the following pages, to bring the \textit{Infinitärcalcül} up to date, stating explicitly and proving carefully a number of general theorems the truth of which Du~Bois-Reymond seems to have tacitly assumed---I may instance in particular the theorem of~\Ref{iii.}{§\;2}. I have to thank Messrs J.~E. Littlewood and G.~N. Watson for their kindness in reading the proof-sheets, and Mr J.~Jackson for the numerical results contained in Appendix~III\@. \Signature{G. H. H.} {\textsc{Trinity College},} {\textit{April}, 1910.} %% -----File: 006.png---Folio xx------- %% -----File: 007.png---Folio xx------- \Contents \PageLine \ToCChap{I.}{Introduction}{1} \ToCChap{II.}{Scales of infinity in general}{7} \ToCChap{III.}{Logarithmico-exponential scales}{16} \ToCChap{IV.}{Special problems connected with logarithmico-exponential scales}{21} \ToCChap{V.}{Functions which do not conform to any logarithmico-exponential scale}{26} \ToCChap{VI.}{Differentiation and integration}{36} \ToCChap{VII.}{Some developments of Du Bois-Reymond's \textit{Infinitärcalcül}}{41} % Prints heading "Appendix I." \ToCApp{I.}{General Bibliography}{47} \ToCApp{II.}{A sketch of some applications, with references}{48} \ToCApp{III.}{Some numerical results}{58} %% -----File: 008.png---Folio xx------- %% -----File: 009.png---Folio 1------- \MainMatter \Chapter{I.}{Introduction.} \Paragraph{1.} \First{The} notions of the `order of greatness' or `order of smallness' of a function~$f(n)$ of a positive integral variable~$n$, when $n$~is `large,' or of a function~$f(x)$ of a continuous variable~$x$, when $x$~is `large' or `small' or `nearly equal to~$a$,' are of the greatest importance even in the most elementary stages of mathematical analysis.\footnote {See, for instance, my \textit{Course of pure mathematics}, pp.~168~\textit{et seq.}, 183~\textit{et seq.}, 344~\textit{et seq.}, 350.} The student soon learns that as $x$~tends to infinity ($x \to \infty$) then also $x^{2} \to \infty$, and moreover that $x^{2}$~tends to infinity \emph{more rapidly than~$x$}, \ie\ that the ratio~$x^{2}/x$ tends to infinity as well; and that $x^{3}$~tends to infinity more rapidly than~$x^{2}$, and so on indefinitely: and it is not long before he begins to appreciate the idea of a `scale of infinity'~$(x^{n})$ formed by the functions $x$,~$x^{2}$, $x^{3}$,~\dots, $x^{n}$,~\dots. This scale he may supplement and to some extent complete by the interpolation of fractional powers of~$x$, and, when he is familiar with the elements of the theory of the logarithmic and exponential functions, of irrational powers: and so he obtains a scale~$(x^{\alpha})$, where $\alpha$~is any positive number, formed by all possible positive powers of~$x$. He then learns that there are functions whose rates of increase cannot be measured by any of the functions of this scale: that $\log x$, for example, tends to infinity more slowly, and $e^{x}$ more rapidly, than \emph{any} power of~$x$; and that $x/(\log x)$ tends to infinity more slowly than~$x$, but more rapidly than any power of~$x$ less than the first. As we proceed further in analysis, and come into contact with its most modern developments, such as the theory of Fourier's series, the theory of integral functions, or the theory of singular points of analytic functions, the importance of these ideas becomes greater and %% -----File: 010.png---Folio 2------- greater. It is the systematic study of them, the investigation of general theorems concerning them and ready methods of handling them, that is the subject of Paul du~Bois-Reymond's \textit{Infinitärcalcül} or `calculus of infinities.' \Paragraph{2.} The notion of the `order' or the `rate of increase' of a function is essentially a relative one. If we wish to say that `the rate of increase of~$f(x)$ is so and so' all we can say is that it is greater than, equal to, or less than that of some other function~$\phi(x)$. Let us suppose that $f$~and~$\phi$ are two functions of the continuous variable~$x$, defined for all values of~$x$ greater than a given value~$x_{0}$. Let us suppose further that $f$~and~$\phi$ are positive, continuous, and steadily increasing functions which tend to infinity with~$x$; and let us consider the ratio~$f/\phi$. We must distinguish four cases: \Item{(i)} If $f/\phi \to \infty$ with~$x$, we shall say that the rate of increase, or simply the \emph{increase}, of~$f$ is greater than that of~$\phi$, and shall write \[ f \cgt \phi. \] \Item{(ii)} If $f/\phi \to 0$, we shall say that the increase of~$f$ is less than that of~$\phi$, and write \[ f \clt \phi. \] \Item{(iii)} If $f/\phi$ remains, for all values of~$x$ however large, between two fixed positive numbers $\delta$,~$\Delta$, so that $0 < \delta < f/\phi < \Delta$, we shall say that the increase of~$f$ is equal to that of~$\phi$, and write \[ f \ceq \phi. \] It may happen, in this case, that $f/\phi$ actually tends to a definite limit. If this is so, we shall write \[ f \ceqq \phi. \] Finally, if this limit is \emph{unity}, we shall write \[ f \sim \phi. \] When we can compare the increase of~$f$ with that of some standard function~$\phi$ by means of a relation of the type $f \ceq \phi$, we shall say that $\phi$~\emph{measures}, or simply \emph{is}, the increase of~$f$. Thus we shall say that the increase of~$2x^{2} + x + 3$ is~$x^{2}$. It usually happens in applications that $f/\phi$~is monotonic (\ie\ steadily increasing or steadily decreasing) as well as $f$~and~$\phi$ themselves. It is clear that in this case $f/\phi$ must tend to infinity, or zero, or to a positive limit: so that one of the three cases indicated above %% -----File: 011.png---Folio 3------- must occur, and we must have $f \cgt \phi$ or $f \clt \phi$ or $f \ceqq \phi$ (not merely $f \ceq \phi$). We shall see in a moment that this is not true in general. \Item{(iv)} It may happen that $f/\phi$ neither tends to infinity nor to zero, nor remains between fixed positive limits. \begin{Remark} Suppose, for example, that $\phi_{1}$,~$\phi_{2}$ are two continuous and increasing functions such that $\phi_{1} \cgt \phi_{2}$. A glance at the figure (\Fig{1}) will probably show with sufficient % [Illustration: Fig. 1.] \Figure[0.45\textwidth]{1}{011} clearness how we can construct, by means of a `staircase' of straight or curved lines, running backwards and forwards between the graphs of $\phi_{1}$~and~$\phi_{2}$, the graph of a steadily increasing function~$f$ such that $f = \phi_{1}$ for $x = x_{1}$, $x_{3}$,~\dots\ and $f = \phi_{2}$ for $x = x_{2}$, $x_{4}$,~\dots. Then $f/\phi_{1} = 1$ for %[** TN: Next line displayed in the original] $x = x_{1}$, $x_{3}$,~\dots, but assumes for $x = x_{2}$, $x_{4}$,~\dots\ values which decrease beyond all limit; while $f/\phi_{2} = 1$ for $x = x_{2}$, $x_{4}$,~\dots, but assumes for $x = x_{1}$, $x_{3}$,~\dots\ values which increase beyond all limit; and $f/\phi$, where $\phi$~is a function such that $\phi_{1} \cgt \phi \cgt \phi_{2}$, as \eg\ $\phi = \sqrt{\phi_{1} \phi_{2}}$, assumes both values which increase beyond all limit and values which decrease beyond all limit. Later on (\Ref{v.}{§\;3}) we shall meet with cases of this kind in which the functions are defined by explicit analytical formulae. \end{Remark} \Paragraph{3.} If a positive constant~$\delta$ can be found such that $f > \delta \phi$ for all sufficiently large values of~$x$, we shall write \[ f \cgeq \phi; \] and if a positive constant~$\Delta$ can be found such that $f < \Delta \phi$ for all sufficiently large values of~$x$, we shall write \[ f \cleq \phi. \] If $f \cgeq \phi$ and $f \cleq \phi$, then $f \ceq \phi$. It is however important to observe (i)~that $f \cgeq \phi$ is not logically equivalent to the negation of $f \clt \phi$\footnote {The relations $f \cgeq \phi$, $f \clt \phi$ are mutually exclusive but not exhaustive: $f \cgeq \phi$ implies the negation of $f \clt \phi$, but the converse is not true.} and (ii)~that it is not logically equivalent to the alternative `\emph{$f \cgt \phi$ or $f \ceq \phi$}.' Thus, in the example discussed at the end of~§\;2, $\phi_{1} \cgeq f \cgeq \phi_{2}$, but no one of the relations $\phi_{1} \cgt f$, etc.\ holds. If however we know that one of the relations $f \cgt \phi$, $f \ceq \phi$, $f \clt \phi$ \emph{must} hold, then these various assertions \emph{are} logically equivalent. %% -----File: 012.png---Folio 4------- The reader will be able to prove without difficulty that the symbols $\cgt$,~$\ceq$,~$\clt$ satisfy the following theorems. \begin{align*} &\text{If $f \cgt \phi$, $\phi \cgeq \psi$, then $f \cgt \psi$.} \\ &\text{If $f \cgeq \phi$, $\phi \cgt \psi$, then $f \cgt \psi$.} \\ &\text{If $f \cgeq \phi$, $\phi \cgeq \psi$, then $f \cgeq \psi$.} \\ &\text{If $f \ceq \phi$, $\phi \ceq \psi$, then $f \ceq \psi$.} \displaybreak[1] \\[6pt] &\text{If $f \cgeq \phi$, then $f + \phi \ceq f$.} \\ &\text{If $f \cgt \phi$, then $f - \phi \ceq f$.} \displaybreak[1] \\[6pt] &\text{If $f \cgt \phi$, $f_{1} \cgt \phi_{1}$, then $f + f_{1} \cgt \phi + \phi_{1}$.} \\ &\text{If $f \cgt \phi$, $f_{1} \ceq \phi_{1}$, then $f + f_{1} \cgeq \phi + \phi_{1}$.} \\ &\text{If $f \ceq \phi$, $f_{1} \ceq \phi_{1}$, then $f + f_{1} \ceq \phi + \phi_{1}$.} \displaybreak[1] \\[6pt] &\text{If $f \cgt \phi$, $f_{1} \cgeq \phi_{1}$, then $ff_{1} \cgt \phi \phi_{1}$.} \\ &\text{If $f \ceq \phi$, $f_{1} \ceq \phi_{1}$, then $ff_{1} \ceq \phi \phi_{1}$.} \end{align*} Many other obvious results of the same character might be stated, but these seem the most important. The reader will find it instructive to state for himself a series of similar theorems involving also the symbols $\ceqq$~and~$\sim$. \Paragraph{4.} So far we have supposed that the functions considered all tend to infinity with~$x$. There is nothing to prevent us from including also the case in which $f$~or~$\phi$ tends steadily to zero, or to a limit other than zero. Thus we may write $x \cgt 1$, or $x \cgt 1/x$, or $1/x \cgt 1/x^{2}$. Bearing this in mind the reader should frame a series of theorems similar to those of~§\;3 but having reference to \emph{quotients} instead of to sums or products. It is also convenient to extend our definitions so as to apply to \emph{negative} functions which tend steadily to~$-\infty$ or to~$0$ or to some other limit. In such cases we make no distinction, when using the symbols $\cgt$,~$\clt$, $\ceq$,~$\ceqq$, between the function and its modulus: thus we write $-x \clt -x^{2}$ or $-1/x \clt 1$, meaning thereby exactly the same as by $x \clt x^{2}$ or $1/x \clt 1$. But $f \sim \phi$ is of course to be interpreted as a statement about the actual functions and not about their moduli. It will be well to state at this point, once for all, that all functions referred to in this tract, from here onwards, are to be understood, unless the contrary is expressly stated or obviously implied, to be positive, continuous, and monotonic, increasing of course if they tend to~$\infty$, and decreasing if they tend to~$0$. But it is sometimes convenient %% -----File: 013.png---Folio 5------- to use our symbols even when this is not true of all the functions concerned; to write, for example, \[ 1 + \sin x \clt x, \qquad x^{2} \cgt x\sin x, \] meaning by the first formula simply that $|1 + \sin x|/x \to 0$. This kind of use may clearly be extended even to complex functions (\eg~$e^{ix} \clt x$). Again, we have so far confined our attention to functions of a continuous variable~$x$ which tends to~$+\infty$. This case includes that which is perhaps even more important in applications, that of functions of the positive integral variable~$n$: we have only to disregard values of~$x$ other than integral values. Thus $n! \cgt n^{2}$, $-1/n \clt n$. Finally, by putting $x = -y$, $x = 1/y$, or $x = 1/(y - a)$, we are led to consider functions of a continuous variable~$y$ which tends to~$-\infty$ or~$0$ or~$a$: the reader will find no difficulty in extending the considerations which precede to cases such as these. In what follows we shall generally state and prove our theorems only for the case with which we started, that of indefinitely increasing functions of an indefinitely increasing continuous variable, and shall leave to the reader the task of formulating the corresponding theorems for the other cases. We shall in fact always adopt this course, except on the rare occasions when there is some essential difference between different cases. \Paragraph{5.} There are some other symbols which we shall sometimes find it convenient to use in special senses. By \[ O(\phi) \] we shall denote a function~$f$, otherwise unspecified, but such that \[ |f| < K\phi, \] where $K$~is a positive constant, and $\phi$~a positive function of~$x$: this notation is due to Landau. Thus \[ x + 1 = O(x), \qquad x = O(x^{2}), \qquad \sin x = O(1). \] We shall follow Borel in using the same letter~$K$ in a whole series of inequalities to denote a positive constant, not necessarily the same in all inequalities where it occurs. Thus \[ \sin x < K, \qquad 2x + 1 < Kx, \qquad x^{m} < Ke^{x}. \] {\Loosen If we use~$K$ thus in any finite number of inequalities which (like the first two above) do not involve any variables other than~$x$, or whatever other variable we are primarily considering, then all the values of~$K$ lie %% -----File: 014.png---Folio 6------- between certain absolutely fixed limits $K_{1}$~and~$K_{2}$ (thus $K_{1}$~might be $10^{-10}$ and $K_{2}$~be~$10^{10}$). In this case all the~$K$'s satisfy $0 < K_{1} < K < K_{2}$, and every relation $f < K\phi$ might be replaced by $f < K_{2}\phi$, and every relation $f > K\phi$ by $f > K_{1}\phi$. But we shall also have occasion to use $K$ in equalities which (like the third above) involve a parameter (here~$m$). In this case $K$, though independent of~$x$, is a function of~$m$. Suppose that $\alpha$,~$\beta$,~\dots\ are all the parameters which occur in this way in this tract. Then if we give any special system of values to $\alpha$,~$\beta$,~\dots, we can determine $K_{1}$,~$K_{2}$ as above. Thus all our $K$'s satisfy} \[ 0 < K_{1}(\alpha, \beta, \dots) < K < K_{2}(\alpha, \beta, \dots), \] where $K_{1}$,~$K_{2}$ are positive functions of $\alpha$,~$\beta$,~\dots\ defined for any permissible set of values of those parameters. But $K_{1}$~has zero for its lower limit; by choosing $\alpha$,~$\beta$,~\dots\ appropriately we can make~$K_{1}$ as small as we please---and, of course, $K_{2}$~as large as we please.\footnote {I am indebted to Mr~Littlewood for the substance of these remarks.} It is clear that the three assertions \[ f = O(\phi), \qquad |f| < K\phi, \qquad f \cleq \phi \] are precisely equivalent to one another. When a function~$f$ possesses any property for all values of~$x$ greater than some definite value (this value of course depending on the nature of the particular property) we shall say that $f$~possesses the property for $x > x_{0}$. Thus \[ x > 100 \quad (x > x_{0}), \qquad e^{x} > 100 x^{2} \quad (x > x_{0}). \] We shall use $\delta$ to denote an arbitrarily small but fixed positive number, and $\Delta$~to denote an arbitrarily great but likewise fixed positive number. Thus \[ f < \delta\phi \quad (x > x_{0}) \] means `however small~$\delta$, we can find~$x_{0}$ so that $f < \delta\phi$ for $x > x_{0}$,' \ie\ means the same as $f \clt \phi$; and $\phi > \Delta f\ (x > x_{0})$ means the same: and \[ (\log x)^{\Delta} \clt x^{\delta} \] means `any power of~$\log x$, however great, tends to infinity more slowly than any positive power of~$x$, however small.' Finally, we denote by~$\epsilon$ a function (of a variable or variables indicated by the context or by a suffix) whose limit is zero when the variable or variables are made to tend to infinity or to their limits in the way we happen to be considering. Thus \[ f = \phi(1 + \epsilon), \qquad f \sim \phi \] are equivalent to one another. %% -----File: 015.png---Folio 7------- In order to become familiar with the use of the symbols defined in the preceding sections the reader is advised to verify the following relations; in them $P_{m}(x)$,~$Q_{n}(x)$ denote polynomials whose degrees are $m$~and~$n$ and whose leading coefficients are positive: \begin{gather*} P_{m}(x) \cgt Q_{n}(x) \quad (m > n), \qquad P_{m}(x) \ceqq Q_{n}(x) \quad (m = n), \\ P_{m}(x) \ceqq x^{m}, \qquad P_{m}(x)/Q_{n}(x) \ceqq x^{m-n}, \displaybreak[1] \\[6pt] \sqrt{ax^{2} + 2bx + c} \ceqq x \quad (a > 0), \qquad \sqrt{x + a} \sim \sqrt{x}, \\ \sqrt{x + a} - \sqrt{x} \sim a/2\sqrt{x}, \qquad \sqrt{x + a} - \sqrt{x} = O(1/\sqrt{x}), \displaybreak[1] \\[6pt] e^{x} \cgt x^{\Delta}, \qquad e^{x^{2}} \cgt e^{\Delta x}, \qquad e^{e^{x}} \cgt e^{x^{\Delta}}, \\ \log x \clt x^{\delta}, \quad \log P_{m}(x) \ceqq \log Q_{n}(x), \quad \log \log P_{m}(x) \sim \log \log Q_{n}(x), \displaybreak[1] \\[6pt] x + a\sin x \sim x, \qquad x(a + \sin x) \ceq x\quad (a > 1), \\ e^{a + \sin x} \ceq 1, \qquad \cosh x \sim \sinh x \ceqq e^{x}, \\ x^{m} = O(e^{\delta x}), \qquad (\log x)/x = O(x^{\delta-1}), \displaybreak[1] \\[6pt] 1 + \frac{1}{2} + \dots + \frac{1}{n} \cgt 1, \qquad 1 + \frac{1}{2^{2}} + \dots + \frac{1}{n^{2}} \ceqq 1, \\ 1 + \frac{1}{2} + \dots + \frac{1}{n} \sim \log n, \qquad 1 + \frac{1}{2} + \dots + \frac{1}{n} - \log n \ceqq 1, \\ n! \clt n^{n}, \qquad n! \cgt e^{\Delta n}, \qquad n! = n^{n^{1+\epsilon}} = n^{n(1 + \epsilon)}, \\ n! \sim n^{n + \frac{1}{2}} e^{-n} \sqrt{2\pi}, \qquad n!\, (e/n)^{n} = (1 + \epsilon) \sqrt{2\pi n}, \\ \int_{1}^{x} \frac{dt}{t} \cgt 1, \qquad \int_{1}^{x} \frac{dt}{t} \sim \log x, \qquad \int_{2}^{x} \frac{dt}{\log t} \sim \frac{x}{\log x}. \end{gather*} \Chapter{II.}{Scales of infinity in general.} \Paragraph{1.} \First{If} we start from a function~$\phi$, such that $\phi \cgt 1$, we can, in a variety of ways, form a series of functions \[ \phi_{1} = \phi,\quad \phi_{2},\quad \phi_{3},\ \dots,\quad \phi_{n},\ \dots \] such that the increase of each function is greater than that of its predecessor. Such a sequence of functions we shall denote for shortness by~$(\phi_{n})$. One obvious method is to take $\phi_{n} = \phi^{n}$. Another is as follows: If $\phi \cgt x$, it is clear that \[ \phi\{\phi(x)\} / \phi(x) \to \infty, \] %% -----File: 016.png---Folio 8------- and so $\phi_{2}(x) = \phi \phi(x) \cgt \phi(x)$; similarly $\phi_{3}(x) = \phi \phi_{2}(x) \cgt \phi_{2}(x)$, and so on.\footnote {For some results as to the increase of such iterated functions see \Ref{vii.}{§\;2~(vi)}.} Thus the first method, with $\phi = x$, gives the scale $x$,~$x^{2}$, $x^{3}$,~\dots\ or~$(x^{n})$; the second, with $\phi = x^{2}$, gives the scale $x^{2}$,~$x^{4}$, $x^{8}$,~\dots\ or~$(x^{2^{n}})$. \begin{Remark} These scales are \emph{enumerable} scales, formed by a simple progression of functions. We can also, of course, by replacing the integral parameter~$n$ by a continuous parameter~$\alpha$, define scales containing a non-enumerable multiplicity of functions: the simplest is~$(x^{\alpha})$, where $\alpha$~is any positive number. But such scales fill a subordinate \textit{rôle} in the theory. \end{Remark} It is obvious that we can always insert a new term (and therefore, of course, any number of new terms) in a scale at the beginning or between any two terms: thus $\sqrt{\phi}$ (or $\phi^{\alpha}$, where $\alpha$~is any positive number less than unity) has an increase less than that of any term of the scale, and $\sqrt{\phi_{n} \phi_{n+1}}$ or $\phi_{n}^{\alpha} \phi_{n+1}^{1-\alpha}$ has an increase intermediate between those of $\phi_{n}$~and~$\phi_{n+1}$. A less obvious and far more important theorem is the following \begin{Result}[Theorem of Paul du~Bois-Reymond.] Given any ascending scale of increasing functions~$\phi_{n}$, \ie\ a series of functions such that $\phi_{1} \clt \phi_{2} \clt \phi_{3} \clt \dots$, we can always find a function~$f$ which increases more rapidly than any function of the scale, \ie\ which satisfies the relation $\phi_{n} \clt f$ for all values of~$n$. \end{Result} In view of the fundamental importance of this theorem we shall give two entirely different proofs. \Paragraph{2.} (i)~We know that $\phi_{n+1} \cgt \phi_{n}$ for all values of~$n$, but this, of course, does not necessarily imply that $\phi_{n+1} \geq \phi_{n}$ for all values of $x$~and~$n$ in question.\footnote {$\phi_{n+1} \cgt \phi_{n}$ implies $\phi_{n+1} > \phi_{n}$ for sufficiently large values of~$x$, say for $x > x_{n}$. But $x_{n}$ may tend to~$\infty$ with~$n$. Thus if $\phi_{n} = x^{n}/n!$ we have $x_{n} = n + 1$.} We can, however, construct a new scale of functions~$\psi_{n}$ such that \Item{(\textit{a})} $\psi_{n}$ is identical with~$\phi_{n}$ for all values of~$x$ from a certain value $x_{n}$ onwards ($x_{n}$, of course, depending upon~$n$); \Item{(\textit{b})} $\psi_{n+1} \geq \psi_{n}$ for all values of $x$~and~$n$. For suppose that we have constructed such a scale up to its $n$th~term~$\psi_{n}$. Then it is easy to see how to construct~$\psi_{n+1}$. Since $\phi_{n+1} \cgt \phi_{n}$, $\phi_{n} \sim \psi_{n}$, it follows that $\phi_{n+1} \cgt \psi_{n}$, and so $\phi_{n+1} > \psi_{n}$ from a certain value of~$x$ (say~$x_{n+1}$) onwards. For $x \geq x_{n+1}$ we take $\psi_{n+1} = \phi_{n+1}$. For $x < x_{n+1}$ we give $\psi_{n+1}$ a value equal to the greater of the values of %% -----File: 017.png---Folio 9------- $\phi_{n+1}$,~$\psi_{n}$. Then it is obvious that $\psi_{n+1}$~satisfies the conditions (\textit{a})~and~(\textit{b}). Now let \[ f(n) = \psi_{n}(n). \] From $f(n)$ we can deduce a continuous and increasing function~$f(x)$, such that \[ \psi_{n}(x) < f(x) < \psi_{n+1}(x) \] for $n < x < n + 1$, by joining the points~$(n, \psi_{n}(n))$ by straight lines or suitably chosen arcs of curves. \begin{Remark} It is perhaps worth while to call attention explicitly to a small point that has sometimes been overlooked (see, \eg, Borel, \textit{Leçons sur la théorie des fonctions}, p.~114; \textit{Leçons sur les séries à termes positifs}, p.~26). It is not always the case that the use of straight lines will ensure \[ f(x) > \psi_{n}(x) \] for $x > n$ (see, for example, \Fig{2}, where the dotted line represents an appropriate arc). \end{Remark} % [Illustration: Fig. 2.] \Figure{2}{017} Then \[ f/\psi_{n} > \psi_{n+1}/\psi_{n} \] for $x > n + 1$, and so $f \cgt \psi_{n}$; therefore $f \cgt \phi_{n}$ and the theorem is proved. \begin{Remark} {\Loosen The proof which precedes may be made more general by taking $f(n) = \psi_{\lambda_{n}} (n)$, where $\lambda_{n}$~is an integer depending upon~$n$ and tending steadily to infinity with~$n$.} \end{Remark} (ii)~The second proof of Du~Bois-Reymond's Theorem proceeds on entirely different lines. We can always choose positive coefficients~$a_{n}$ so that \[ f(x) = \sum_{1}^{\infty} a_{n}\psi_{n}(x) \] is convergent for all values of~$x$. This will certainly be the case, for instance, if \[ 1/a_{n} = \psi_{1}(1) \psi_{2}(2) \dots \psi_{n}(n). \] For then, if $\nu$~is any integer greater than~$x$, $\psi_{n}(x) < \psi_{n}(n)$ for $n \geqq \nu$, and the series will certainly be convergent if \[ \sum_{\nu}^{\infty} \frac{1}{\psi_{1}(1) \psi_{2}(2) \dots \psi_{n-1}(n-1)} \] is convergent, as is obviously the case. %% -----File: 018.png---Folio 10------- Also \[ f(x)/\psi_{n}(x) > a_{n+1}\psi_{n+1}(x)/\psi_{n}(x) \to \infty, \] so that $f \cgt \phi_{n}$ for all values of~$n$. \begin{Remark} \Paragraph{3.} Suppose, \eg, that $\phi_{n} = x^{n}$. If we restrict ourselves to values of~$x$ greater than~$1$, we may take $\psi_{n} = \phi_{n} = x^{n}$. The first method of construction would naturally lead to \[ f = n^{n} = e^{n\log n}, \] or $f = (\lambda_{n})^{n}$, where $\lambda_{n}$~is defined as at the end of §\;2~(i), and each of these functions has an increase greater than that of any power of~$n$. The second method gives \[ f(x) = \sum_{1}^{\infty} \frac{x^{n}}{1^{1} 2^{2} 3^{3} \dots n^{n}}. \] It is known\footnote {\textit{Messenger of Mathematics,} vol.~34, p.~101.} that when $x$~is large the order of magnitude of this function is roughly the same as that of \[ e^{\frac{1}{2}(\log x)^{2}/\log\log x}. \] {\Loosen As a matter of fact it is by no means necessary, in general, in order to ensure the convergence of the series by which $f(x)$~is defined, to suppose that $a_{n}$~decreases so rapidly. It is very generally sufficient to suppose $1/a_{n} = \phi_{n}(n)$: this is always the case, for example, if $\phi_{n}(x) = \{\phi(x)\}^{n}$, as the series} \[ \sum \left\{\frac{\phi(x)}{\phi(n)}\right\}^{n} \] is always convergent. This choice of~$a_{n}$ would, when $\phi = x$, lead to \[ f(x) = \sum \left(\frac{x}{n}\right)^{n} \sim \sqrt{\frac{2\pi x}{e}}\, e^{x/e}.\footnote {Lindelöf, \textit{Acta Societatis Fennicae}, t.~31, p.~41; Le~Roy, \textit{Bulletin des Sciences Mathématiques}, t.~24, p.~245.\PageLabel{10}} \] But the simplest choice here is $1/a_{n} = n!$, when \[ f(x) = \sum \frac{x^{n}}{n!} = e^{x} - 1; \] it is naturally convenient to disregard the irrelevant term~$-1$. \Paragraph{4.} We can always suppose, if we please, that $f(x)$~is defined by a power series $\sum a_{n}x^{n}$ convergent for all values of~$x$, in virtue of a theorem of Poincaré's\footnote {\textit{American Journal of Mathematics}, vol.~14, p.~214.} which is of sufficient intrinsic interest to deserve a formal statement and proof. \begin{Result} Given any continuous increasing function~$\phi(x)$, we can always find an integral function~$f(x)$ \(\ie\ a function~$f(x)$ defined by a power series $\sum a_{n}x^{n}$ convergent for all values of~$x$\) such that $f(x) \cgt \phi(x)$. \end{Result} The following simple proof is due to Borel.\footnote {\textit{Leçons sur les séries à termes positifs}, p.~27.} Let $\Phi(x)$ be any function (such as the square of~$\phi$) such that $\Phi \cgt \phi$. Take %% -----File: 019.png---Folio 11------- an increasing sequence of numbers~$a_{n}$ such that $a_{n} \to \infty$, and another sequence of numbers~$b_{n}$ such that \[ a_{1} < b_{2} < a_{2} < b_{3} < a_{3} < \dots; \] and let \[ f(x) = \sum \left(\frac{x}{b_{n}}\right)^{\nu_{n}}, \] where $\nu_{n}$~is an integer and $\nu_{n+1} > \nu_{n}$. This series is convergent for all values of~$x$; for the $n$th~root of the $n$th~term is, for sufficiently large values of~$n$, not greater than~$x/b_{n}$, and so tends to zero. Now suppose $a_{n} \leqq x < a_{n+1}$; then \[ f(x) > \left(\frac{a_{n}}{b_{n}}\right)^{\nu_{n}}. \] Since $a_{n} > b_{n}$ we can suppose $\nu_{n}$~so chosen that (i)~$\nu_{n}$~is greater than any of $\nu_{1}$,~$\nu_{2}$, \dots,~$\nu_{n-1}$ and (ii) \[ \left(\frac{a_{n}}{b_{n}}\right)^{\nu_{n}} > \Phi(a_{n+1}). \] Then \[ f(x) > \Phi(a_{n+1}) > \Phi(x), \] and so $f \cgt \phi$. \end{Remark} \Paragraph{5.} So far we have confined our attention to ascending scales, such as $x$,~$x^{2}$, $x^{3}$,~\dots, $x^{n}$,~\dots\ or~$(x^{n})$; but it is obvious that we may consider in a similar manner \emph{descending} scales such as $x$,~$\sqrt{x}$, $\sqrt[3]{x}$,~\dots, $\sqrt[n]{x}$,~\dots\ or~$(\sqrt[n]{x})$. It is very generally (though not always) true that if $(\phi_{n})$~is an ascending scale, and $\psi$~denotes the function inverse to~$\phi$, then $(\psi_{n})$~is a descending scale. \begin{Remark} If $\phi > \bar{\phi}$ for all values of~$x$ (or all values greater than some definite value), then a glance at \Fig{3} is enough to show that if $\psi$~and~$\bar{\psi}$ are the functions inverse to $\phi$~and~$\bar{\phi}$, then $\psi < \bar{\psi}$ for all values of~$x$ (or all values greater than some definite value). We have only to remember that the graph of~$\psi$ may be obtained from that of~$\phi$ by looking at the latter from a different point of view (interchanging the \textit{rôles} of $x$~and~$y$). But it is not true that $\phi \cgt \bar{\phi}$ involves $\psi \clt \bar{\psi}$. Thus $e^{x} \cgt e^{x}/x$. The function inverse to~$e^{x}$ is~$\log x$: the function inverse to~$e^{x}/x$ is obtained by solving the equation $x = e^{y}/y$ with respect to~$y$. This equation gives \[ y = \log x + \log y, \] and it is easy to see that $y \sim \log x$. \end{Remark} %[Illustration: Fig. 3.] \Figure[0.4\textwidth]{3}{019} \begin{Result} Given a scale of increasing functions~$\phi_{n}$ such that \[ \phi_{1} \cgt \phi_{2} \cgt \phi_{3} \cgt \dots \cgt 1, \] %% -----File: 020.png---Folio 12------- we can find an increasing function~$f$ such that $\phi_{n} \cgt f \cgt 1$ for all values of~$n$.\end{Result} The reader will find no difficulty in modifying the argument of §\;2~(i) so as to establish this proposition. \Paragraph{6.} The following extensions of Du~Bois-Reymond's Theorem (and the corresponding theorem for descending scales) are due to Hadamard.\footnote {\textit{Acta Mathematica}, t.~18, pp.~319 \textit{et seq.}} \begin{Result} Given \[ \phi_{1} \clt \phi_{2} \clt \phi_{3} \clt \dots \clt \phi_{n} \clt \dots \clt \Phi, \] we can find $f$ so that $\phi_{n} \clt f \clt \Phi$ for all values of~$n$. \end{Result} \begin{Result} Given \[ \psi_{1} \cgt \psi_{2} \cgt \psi_{3} \cgt \dots \cgt \psi_{n} \cgt \dots \cgt \Psi, \] we can find $f$ so that $\psi_{n} \cgt f \cgt \Psi$ for all values of~$n$. \end{Result} \begin{Result} Given an ascending sequence~$(\phi_{n})$ and a descending sequence~$(\psi_{p})$ such that $\phi_{n} \clt \psi_{p}$ for all values of $n$~and~$p$, we can find $f$ so that \[ \phi_{n} \clt f \clt \psi_{p} \] for all values of $n$~and~$p$. \end{Result} To prove the first of these theorems we have only to observe that \[ \Phi/\phi_{1} \cgt \Phi/\phi_{2} \cgt \dots \cgt \Phi/\phi_{n} \cgt \dots \cgt 1, \] and to construct a function~$F$ (as we can in virtue of the theorem of~§\;5) which tends to infinity more slowly than any of the functions~$\Phi/\phi_{n}$. Then \[ f = \Phi/F \] is a function such as is required. Similarly for the second theorem. The third is rather more difficult to prove. \begin{Remark} In the first place, we may suppose that $\phi_{n+1} > \phi_{n}$ for all values of $x$~and~$n$: for if this is not so we can modify the definitions of the functions~$\phi_{n}$ as in §\;2~(i). Similarly we may suppose $\psi_{p+1} < \psi_{p}$ for all values of $x$~and~$p$. Secondly, we may suppose that, if $x$~is fixed, $\phi_{n} \to \infty$ as $n \to \infty$, and $\psi_{p} \to 0$ as $p \to \infty$. For if this is not true of the functions given, we can replace them by $H_{n}\phi_{n}$ and $K_{p}\psi_{p}$, where $(H_{n})$~is an increasing sequence of constants, tending to~$\infty$ with~$n$, and $(K_{p})$~a decreasing sequence of constants whose limit as $p \to \infty$ is zero. % [Illustration: Fig. 4.] \Figure{4}{020} Since $\psi_{p} \cgt \phi_{n}$ but, for any given~$x$, $\psi_{p} < \phi_{n}$ for sufficiently large values of~$n$, it is clear (see \Fig{4}) that the curve $y = \psi_{p}$ intersects the curve $y = \phi_{n}$ for all sufficiently large values of~$n$ (say for $n \geq n_{p}$). %% -----File: 021.png---Folio 13------- At this point we shall, in order to avoid unessential detail, introduce a restrictive hypothesis which can be avoided by a slight modification of the argument,\footnote {See Hadamard's original paper quoted above.} but which does not seriously impair the generality of the result. We shall assume that no curve $y = \psi_{p}$ intersects any curve $y = \phi_{n}$ in more than one point; let us denote this point, if it exists, by~$P_{n, p}$. If $p$ is fixed, $P_{n, p}$~exists for $n > n_{p}$; similarly, if $n$~is fixed, $P_{n, p}$~exists for $p > p_{n}$. And as either $n$~or~$p$ increases, so do both the ordinate or the abscissa of~$P_{n, p}$. The curve~$\psi_{p}$ contains all the points~$P_{n, p}$ for which $p$~has a fixed value: and $y = \phi_{n}$ contains all the points for which $n$~has a fixed value. It is clear that, in order to define a function~$f$ which tends to infinity more rapidly than any~$\phi_{n}$ and less rapidly than any~$\psi_{p}$, all that we have to do is to draw a curve, making everywhere a positive acute angle with each of the axes of coordinates, and crossing all the curves $y = \phi_{n}$ from below to above, and all the curves $y = \psi_{p}$ from above to below. Choose a positive integer~$N_{p}$, corresponding to each value of~$p$, such that (i)~$N_{p} > n_{p}$ and (ii)~$N_{p} \to \infty$ as $p \to \infty$. Then $P_{N_{p}, p}$~exists for each value of~$p$. And it is clear that we have only to join the points $P_{N_{1}, 1}$,~$P_{N_{2}, 2}$, $P_{N_{3}, 3}$,~\dots\ by straight lines or other suitably chosen arcs of curves in order to obtain a curve which fulfils our purpose. The theorem is therefore established. \end{Remark} \Paragraph{7.} Some very interesting considerations relating to scales of infinity have been developed by Pincherle.\PageLabel{13}\footnote {\textit{Memorie della Accademia delle Scienze di Bologna} (ser.~4, t.~5, p.~739).} We have defined $f \cgt \phi$ to mean $f/\phi \to \infty$, or, what is the same thing, \[ \log f - \log \phi \to \infty. \Tag{(1)} \] We might equally well have defined $f \cgt \phi$ to mean \[ F(f) - F(\phi) \to \infty, \Tag{(2)} \] where $F(x)$~is any function which tends steadily to infinity with~$x$ (\eg~$x$,~$e^{x}$). Let us say that if \Eq{(2)}~holds then \[ f \cgt \phi \quad (F), \Tag{(3)} \] so that $f \cgt \phi$ is equivalent to $f \cgt \phi\ (\log x)$. Similarly we define $f \clt \phi\ (F)$ to mean that $F(f) - F(\phi) \to -\infty$, and $f \ceq \phi\ (F)$ to mean that $F(f) - F(\phi)$ remains between certain fixed limits. Thus \begin{gather*} x + \log x \ceq x, \qquad x + \log x \cgt x \quad (x), \\ x + 1 \ceq x\quad (x), \qquad x + 1 \cgt x \quad (e^{x}), \end{gather*} since $e^{x+1} - e^{x} = (e - 1)e^{x} \to \infty$. %% -----File: 022.png---Folio 14------- It is clear that the more rapid the increase of~$F$, the more likely is it to discriminate between the rates of increase of two given functions $f$~and~$\phi$. More precisely, \begin{Result}if \[ f \cgt \phi \quad (F), \] and $\bar{F} = FF_{1}$, where $F_{1}$~is any increasing function, then will \[ f \cgt \phi \quad (\bar{F}). \] \end{Result} For \[ \bar{F}(f) - \bar{F}(\phi) = F(f) F_{1}(f) - F(\phi) F_{1}(\phi) > \{F(f) - F(\phi)\} F_{1}(\phi) \to \infty. \] \Paragraph{8.} The substance of the following theorems is due in part to Pincherle and in part to Du Bois-Reymond.\footnote {Pincherle, \lc; Du~Bois-Reymond, \textit{Math.\ Annalen}, Bd.~8, S.~390 \textit{et seq.}} \begin{Result} \Item{1.} However rapid the increase of~$f$, as compared with that of~$\phi$, we can so choose~$F$ that $f \ceq \phi\ (F)$. \end{Result} \begin{Result} \Item{2.} {\Loosen If $f - \phi$ is positive for $x > x_{0}$, we can so choose~$F$ that $f \cgt \phi\ (F)$.} \end{Result} \begin{Result} {\Loosen \Item{3.} If $f - \phi$ is monotonic and not negative for $x > x_{0}$, and $f \ceq \phi\ (F)$, however great be the increase of~$F$, then $f = \phi$ from a certain value of~$x$ onwards.} \end{Result} \Item{(1)} If $f \cgt \phi$, we may regard~$f$ as an increasing function of~$\phi$, say \[ f = \theta(\phi), \] where $\theta(x) \cgt x$. We can choose a constant~$g$ greater than~$1$, and then choose~$X$ so that $\theta(x) > gx$ for $x > X$. Let $a$~be any number greater than~$X$, and let \[ a_{1} = \theta(a), \qquad a_{2} = \theta(a_{1}), \qquad a_{3} = \theta(a_{2}),\ \dots. \] Then $(a_{n})$~is an increasing sequence, and $a_{n} \to \infty$, since $a_{n} > g^{n}a$. We can now construct an increasing function~$F$ such that \[ F(a_{n}) = \tfrac{1}{2} nK, \] where $K$~is a constant. Then if $a_{\nu-1} \leqq x \leqq a_{\nu}$, $a_{\nu} \leqq \theta(x) \leqq a_{\nu+1}$, and \[ F\{\theta(x)\} - F(x) < F(a_{\nu+1}) - F(a_{\nu-1}) < K. \] Accordingly $F(f) - F(\phi)$ remains less than a constant, and so the first theorem is established. \Item{(2)} Let $f - \phi = \lambda$, so that $\lambda > 0$. If $\lambda$, as $x$~increases, remains greater than a constant~$K$, then \[ e^{f} - e^{\phi} > (e^{K} - 1)e^{\phi} \to \infty, \] so that we may take $F(x) = e^{x}$. %% -----File: 023.png---Folio 15------- If it is not true that $\lambda \geqq K$, $\lambda$~assumes values less than any assignable positive number, as $x \to \infty$. Let $\bar{\lambda}(x)$ be defined as the lower limit of~$\lambda(\xi)$ for $\xi \leqq x$. Then $\bar{\lambda}$~tends steadily to zero as $x \to \infty$, and $\bar{\lambda} \leqq \lambda$. We may also regard $\bar{\lambda}$ as a steadily decreasing function of~$\phi$, say $\bar{\lambda} = \mu(\phi)$. Let $\varpi(\phi)$ be an increasing function of~$\phi$ such that $\varpi \cgt 1/\mu$, $\mu\varpi \cgt 1$. Then if \begin{gather*} F = \int^{\phi} \varpi(t)\, dt,\\ F(f) - F(\phi) = \int_{\phi}^{\phi + \lambda} \varpi\, dt \geqq \int_{\phi}^{\phi + \mu(\phi)} \varpi\, dt > \mu(\phi)\varpi(\phi) \cgt 1, \end{gather*} and $F(x)$~fulfils the requirement of theorem~2. The third theorem is obviously a mere corollary of the second. \begin{Remark} The reader will find it instructive to deduce or prove independently the following three theorems, which are closely analogous to those which have just been proved. \begin{Result} \Item{1.} However great be the increase of~$f$ as compared with that of~$\phi$, we can determine an increasing function~$F$ such that $F(f) \ceq F(\phi)$. \end{Result} \begin{Result} \Item{2.} If $f - \phi$ is positive for $x > x_{0}$, we can determine an increasing function~$F$ such that $F(f) \cgt F(\phi)$. \end{Result} \begin{Result} \Item{3.} If $f - \phi$ is monotonic and not negative for $x > x_{0}$, and $F(f) \ceq F(\phi)$, however great the increase of~$F$, then $f = \phi$ from a certain value of~$x$ onwards. \end{Result} {\Loosen To these he may add the theorem (analogous to that proved at the end of~§\;7) that \begin{Result}$f \cgt \phi$ involves $F(f) \cgt F(\phi)$ if $\log F(x)/\log x$ is an increasing function\end{Result} (a condition which may for practical purposes be replaced by $F \cgt x$).} \Paragraph{9.} Let us consider some examples of the theorems of the last paragraph. \Item{(i)} Let $f = x^{m}$ ($m > 1$) and $\phi = x$. Then, following the argument of §\;8~(1), we have $\theta(\phi) = \phi^{m}$. We may take \[ a = e, \qquad a_{1} = e^{m}, \qquad a_{2} = e^{m^{2}},\ \dots, \qquad a_{n} =e^{m^{n}},\ \dots, \] and we have to define~$F$ so that \[ F(e^{m^{n}}) = \tfrac{1}{2}nK. \] The most natural solution of this equation is \[ F(x) = K\log\log x/2\log m. \] And in fact \[ F(x^{m}) - F(x) = \frac{K}{2\log m}\{\log(m\log x) - \log\log x\} = \tfrac{1}{2}K, \] so that $x^{m} \ceq x\ (F)$. %% -----File: 024.png---Folio 16------- \Item{(ii)} Let $f = e^{x} + e^{-x}$, $\phi = e^{x}$. Following the argument of §\;8~(2), we find \[ \lambda = e^{-x} = \bar{\lambda}, \qquad \mu(\phi) = 1/\phi, \] and we may take $\varpi(\phi) = \phi^{1+\alpha}$ ($\alpha > 0$). This makes $F$ a constant multiple of~$x^{2+\alpha}$, and it is easy to verify that \[ (e^{x} + e^{-x})^{k} - e^{kx} \to \infty, \] if $k > 2$. \Item{(iii)} The relation $F(f) \ceq F(\phi)$ is equivalent to $f \ceq \phi\ (\log F)$. Using the result of~(i) we see that $F(x^{m}) \ceq F(x)$ if $F \cleq \log x$. Similarly, using the result of~(ii), we see that $F(e^{x} + e^{-x}) \cgt F(e^{x})$ if $F \cgeq e^{x^{k}}$ ($k > 2$). \end{Remark} \Paragraph{10.} Before leaving this part of our subject, let us observe that all of the substance of §§\;1--6 of this section may be extended to the case in which our symbols $\cgt$,~etc., are defined by reference to an arbitrary increasing function~$F$. We leave it as an exercise to the reader to effect these extensions. \Chapter{III.}{Logarithmico-Exponential Scales.} \Paragraph{1.} \First{The} only scales of infinity that are of any practical importance in analysis are those which may be constructed by means of the logarithmic and exponential functions. We have already seen (\Ref{ii.}{§\;3}) that \[ e^{x} \cgt x^{n} \] for any value of~$n$ however great. From this it follows that \[ \log x \clt x^{1/n} \] for any value of $n$.\footnote {It was pointed out above (\Ref{ii.}{§\;5}) that $\phi \cgt \bar{\phi}$ does not necessarily involve $\psi \clt \bar{\psi}$ ($\psi$,~$\bar{\psi}$ being the functions inverse to $\phi$,~$\bar{\phi}$). But it does involve $\psi < \bar{\psi}$ for sufficiently large values of~$x$, and therefore $\psi \cleq \bar{\psi}$. Hence $\phi \cgt \phi_{n}$ (for any~$n$) involves $\psi \cleq \psi_{n}$ (for any~$n$) and therefore, if $(\psi_{n})$~is a descending scale, as is in this case obvious, $\psi \clt \psi_{n}$ for any~$n$. For proofs of the relations $e^{x} \cgt x^{n}$, $\log x \clt x^{1/n}$, proceeding on different lines, see my \textit{Course of pure mathematics}, pp.~345,~350.} It is easy to deduce that \begin{gather*} e^{e^{x}} \cgt e^{x^{n}}, \qquad e^{e^{e^{x}}} \cgt e^{e^{x^{n}}},\ \dots, \\ \log\log x \clt (\log x)^{1/n}, \qquad \log\log\log x \clt (\log\log x)^{1/n},\ \dots. \end{gather*} %% -----File: 025.png---Folio 17------- The repeated logarithmic and exponential functions are so important in this subject that it is worth while to adopt a notation for them of a less cumbrous character. We shall write \begin{alignat*}{3} %[** TN: Unaligned in the original] l_{1}x &\eqq lx \eqq \log x, \qquad& l_{2}x &\eqq llx, \qquad& l_{3}x &\eqq ll_{2}x,\ \dots,\\ e_{1}x &\eqq ex \eqq e^{x}, \qquad& e_{2}x &\eqq eex, \qquad& e_{3}x &\eqq ee_{2}x,\ \dots. \end{alignat*} It is easy, with the aid of these functions, to write down any number of ascending scales, each containing only functions whose increase is greater than that of any function in any preceding scale; for example \begin{gather*} x,\quad x^{2},\ \dots,\quad x^{n},\ \dots;\qquad e^{x},\quad e^{2x},\ \dots,\quad e^{nx},\ \dots; \\ e^{x^{2}},\quad e^{x^{3}},\ \dots,\quad e^{x^{n}},\ \dots;\qquad e_{2}x,\quad e_{3}x,\ \dots,\quad e_{n}x,\ \dots. \end{gather*} In among the functions of these scales we can of course interpolate new functions as freely as we like, using, for instance, such functions as \[ x^{\alpha} e^{\beta x^{\gamma} e^{\delta x^{\epsilon}}}, \] where $\alpha$, $\beta$, $\gamma$, $\delta$, $\epsilon$ are any positive numbers; and we can of course construct non-enumerable (\Ref{ii.}{§\;1}) as well as enumerable scales. Similarly we can construct any number of descending scales, each composed of functions whose increase is less than that of any functions in any preceding scale: for example \[ lx, \quad (lx)^{1/2}, \ \dots, \quad (lx)^{1/n},\ \dots; \qquad l_{2}x, \quad l_{3}x, \ \dots, \quad l_{n}x,\ \dots. \] Two special scales are of particularly fundamental importance; the ascending scale \[ \LTag{(E)} x, \quad ex, \quad e_{2}x, \quad e_{3}x, \ \dots, \] and the descending scale \[ \LTag{(L)} x, \quad lx, \quad l_{2}x, \quad l_{3}x, \ \dots. \] These scales mark the \emph{limits} of all logarithmic and exponential scales: it is of course, in virtue of the general theorems of~\Ref{ii.}{}, possible to define functions whose increase is more rapid than that of any~$e_{n}x$ or slower than that of any~$l_{n}x$; but, as we shall see in a moment, this is \emph{not} possible if we confine ourselves to functions defined by a finite and explicit formula involving only the ordinary functional symbols of elementary analysis. \Paragraph{2.} We define a \emph{logarithmico-exponential function} (shortly, an \emph{$L$-function}) as a real one-valued function defined, for all values of~$x$ greater than some definite value, by a finite combination of the ordinary algebraical symbols (viz.\ $+$,~$-$, $×$,~$÷$,~$\sqrt[n]{}$) and the functional symbols $\log(\dots)$ and $e^{(\dots)}$, operating on the variable~$x$ and on real constants. %% -----File: 026.png---Folio 18------- \begin{Remark} It is to be observed that the result of working out the value of the function, by substituting~$x$ in the formula defining it, is to be real at all stages of the work. It is important to exclude such a function \[ \tfrac{1}{2}\{e^{\sqrt{-x^{2}}} + e^{-\sqrt{-x^{2}}}\}, \] which, with a suitable interpretation of the roots, is equal to~$\cos x$. \end{Remark} \begin{Theorem} Any $L$-function is ultimately continuous, of constant sign, and monotonic, and, as $x \to \infty$, tends to~$\infty$, or to zero or to some other definite limit. Further, if $f$~and~$\phi$ are $L$-functions, one or other of the relations \[ f \cgt \phi, \qquad f \ceqq \phi, \qquad f \clt \phi \] holds between them. \end{Theorem} We may classify $L$-functions as follows, by a method due to Liouville.\footnote {See my tract \textit{The integration of functions of a single variable} (No.~2 of this series), pp.~5 \textit{et~seq.}, where references to Liouville's original memoirs are given.} An $L$-function is of order zero if it is purely algebraical; of order~$1$ if the functional symbols $l(\dots)$ and $e(\dots)$ which occur in it bear only on algebraical functions; of order~$2$ if they bear only on algebraical functions or $L$-functions of order~$1$; and so on. Thus \[ x^{x^{x}} = e^{\log x e^{x\log x}} \] is of order~$3$. As the results stated in the theorem are true of algebraical functions, it is sufficient to prove that, if true of $L$-functions of order $n - 1$, they are true of $L$-functions of order~$n$. Let us observe first that if $f$~and~$\phi$ are $L$-functions, so is~$f/\phi$. Hence the last part of the theorem is a mere corollary of the first part. Again, the derivative of an $L$-function of order~$n$ is an $L$-function of order~$n$ (or less). Hence it is enough to prove that, if the results stated are true of $L$-functions of order~$n - 1$, then an $L$-function of order~$n$ is ultimately continuous and of constant sign, \ie\ that it is continuous and cannot vanish for a series of values of~$x$ increasing beyond limit. For, if this is true of any $L$-function of order~$n$, it is true of the derivative of any such function; and therefore the function itself is ultimately continuous and monotonic. Now any $L$-function of order~$n$ can be expressed in the form \begin{align*} f_{n} &= A\{e\phi_{n-1}^{(1)}, e\phi_{n-1}^{(2)}, \dots, e\phi_{n-1}^{(r)},\ l\psi_{n-1}^{(1)}, \dots, l\psi_{n-1}^{(s)}, \chi_{n-1}^{(1)}, \dots, \chi_{n-1}^{(t)}\}\\ &= A\{z_{1}, z_{2}, \dots, z_{q}\}, \end{align*} say, where $q = r + s + t$, the functions with suffix~$n - 1$ are $L$-functions of order~$n - 1$, and $A$~denotes an algebraical function: and there is therefore an identical relation \[ F \eqq M_{0} f_{n}^{p} + M_{1} f_{n}^{p-1} + \dots + M_{p} = 0, \] %% -----File: 027.png---Folio 19------- where the coefficients are polynomials in $z_{1}$,~$z_{2}$, \dots,~$z_{q}$. These polynomials are comprised in the class of functions \[ M = \sum \rho_{n-1} e\sigma_{n-1} (l\tau_{n-1}^{(1)})^{\kappa_{1}} (l\tau_{n-1}^{(2)})^{\kappa_{2}} \dots (l\tau_{n-1}^{(h)})^{\kappa_{h}}, \] in which the $\kappa$'s are positive integers, the number of terms in the summation is finite, and the functions with suffix~$n - 1$ are again $L$-functions of order~$n - 1$. So also are \[ \frac{dM_{0}}{dx}, \quad \frac{dM_{1}}{dx},\ \dots, \quad \frac{dM_{p}}{dx}, \] and the discriminant of~$F$ \textit{qua} function of~$f_{n}$. Let us suppose our conclusions established in so far as relates to functions of the type~$M$. Then it follows by a well known theorem\footnote {If $F(x, y)$ is a function of $x$~and~$y$ which vanishes for $x = a$, $y = b$, and has derivatives $\dfrac{\dd F}{\dd x}$,~$\dfrac{\dd F}{\dd y}$ continuous about~$(a, b)$, and if $\dfrac{\dd F}{\dd y}$~does not vanish for $x = a$, $y = b$, then there is a unique continuous function~$y$ which is equal to~$b$ when $x = a$, and satisfies the equation $F(x, y) = 0$ identically. See, \eg, W.~H.~Young, \textit{Proc.\ Lond.\ Math.\ Soc.}, vol.~7, pp.~397 \textit{et~seq.}} that $f_{n}$~is continuous, and, since $f_{n} = 0$ involves $M_{p} = 0$, that $f_{n}$~also is ultimately of constant sign. Hence it is enough to establish our conclusions for functions of the type~$M$. Let us call \[ \kappa_{1} + \kappa_{2} + \dots + \kappa_{h} \] the \emph{degree} of a term of~$M$, and let us suppose that the greatest degree of a term of~$M$ is~$\lambda$, and that there are $\mu$~terms of degree~$\lambda$, and that the term printed in the expression of~$M$ above is one of them. In the first place it is obvious, from the form of~$M$ and the fact that $ey$~and~$ly$ are ultimately continuous when $y$~is ultimately continuous and monotonic, that $M$~is ultimately continuous. Again, if $M$~vanishes for values of~$x$ surpassing all limit, the same is true of \[ M/(\rho_{n-1} e\sigma_{n-1}), \] and therefore, by Rolle's theorem,\footnote {If a function possesses a derivative for all values of its argument, the derivative must have at least one root between any two roots of the function itself.} of the derivative of the latter function. But the reader will easily verify that when we differentiate, and arrange the terms of the derivative in the same manner as those of~$M$, we obtain a function of the same form as~$M$ but containing at most $\mu - 1$~terms of order~$\lambda$. And by repeating this process we clearly arrive ultimately at a function of the form \[ N = \sum \rho_{n-1} e\sigma_{n-1}, \] %% -----File: 028.png---Folio 20------- in which there are no factors of the form~$l\tau_{n-1}$, and which must vanish for a sequence of values of~$x$ surpassing all limit. Hence it is sufficient for our purpose to prove that this is impossible. Let the number of terms in~$N$ be~$\varpi$. Then \[ \frac{d}{dx} \{N/(\rho_{n-1} e\sigma_{n-1})\} \] must (for reasons similar to those advanced above) vanish for values of~$x$ surpassing all limit. But when we differentiate, and arrange the terms of the derivative in the same manner as those of~$N$, we are left with a function of the same form as~$N$, but containing only $\varpi - 1$~terms. And it is clear that a repetition of this process leads to the conclusion that a function of the type \[ \rho_{n-1} e\sigma_{n-1} \] vanishes for values of~$x$ surpassing all limit, which is \textit{ex~hypothesi} untrue. Hence the theorem is established. \Paragraph{3.} The proof just given, it may be observed, does not in any way depend upon the fact that the symbols of algebraical functionality, admitted into the definition of $L$-functions, are of an \emph{explicit} character. We might admit such functions as \[ e_{2}\sqrt{ly}, \] where $y^{5} + y - x = 0$. But the case contemplated in the definition seems to be the only one of any interest. Another interesting theorem is: \begin{Result}if $f$~is any $L$-function, we can find an integer~$k$ such that \[ f \clt e_{k}x; \] and, if $f \cgt 1$, we can find~$k$ so that \[ f \cgt l_{k}x: \] that is to say, an $L$-function cannot increase more rapidly than any exponential, or more slowly than any logarithm. \end{Result} More precisely, an $L$-function of order~$n$ cannot satisfy $f \cgt e_{n}(x^{\Delta})$ or $1 \clt f \clt (l_{n}x)^{\delta}$. The first part of this result is easily established; the second appears to require a more elaborate proof. \Paragraph{4.} Let $f$~and~$\phi$ be any two $L$-functions which tend to infinity with~$x$, and let $\alpha$ be any positive number. Then one of the three relations \[ f \cgt \phi^{\alpha}, \qquad f \ceqq \phi^{\alpha}, \qquad f \clt \phi^{\alpha} \] must hold between $f$ and~$\phi$; and the second can hold for at most one %% -----File: 029.png---Folio 21------- value of~$\alpha$. If the first holds for any~$\alpha$ it holds for any smaller~$\alpha$; and if the last holds for any~$\alpha$ it holds for any greater~$\alpha$. Then there are three possibilities. Either the first relation holds for every~$\alpha$; then \[ f \cgt \phi^{\Delta}. \] Or the third holds for every~$\alpha$; then \[ f \clt \phi^{\delta}. \] Or the first holds for some values of~$\alpha$ and the third for others; and then there is a value a of~$\alpha$ which divides the two classes of values of~$\alpha$, and we may write \[ f = \phi^{\alpha} f_{1}, \] where $\phi^{-\delta} \clt f_{1} \clt \phi^{\delta}$. We shall find this result very useful in the sequel. \Chapter[Logarithmico-Exponential Scales.] {IV.}{Special Problems Connected with Logarithmico-Exponential Scales.} \begin{Remark} \Paragraph{1. The functions $e_{r}(l_{s}x)^{\mu}$.} We have agreed to express the fact that, however large be~$\alpha$ and however small be~$\beta$, $x^{\alpha}$~has an increase less than that of~$e^{x^{\beta}}$, by \[ \Tag{(1)} x^{\Delta} \clt e^{x^{\delta}}.\footnote {Such a relation as \[ x^{\Delta_{1}} (lx)^{\Delta_{2}} \clt e^{\delta_{1} x^{\delta_{2}} (lx)^{-\Delta_{3}}} \] might at first sight appear to afford more information than~\Eq{(1)}: but \[ x^{\Delta_{1}} (lx)^{\Delta_{2}} \clt x^{\Delta_{1}'}, \qquad \delta_{1} x^{\delta_{2}} (lx)^{-\Delta_{3}} \cgt x^{\delta_{2}'}, \] where $\Delta_{1}'$,~$\delta_{2}'$ are any positive numbers greater than~$\Delta_{1}$ and less than~$\delta_{2}$ respectively. Hence our relation really expresses no more than~\Eq{(1)}.} \] Let us endeavour to find a function~$f$ such that \[ x^{\Delta} \clt f \clt e^{x^{\delta}}. \Tag{(2)} \] If $\phi_{1} \cgt \phi_{2}$, $e^{\phi_{1}} \cgt e^{\phi_{2}}$ (\Ref{ii.}{§\;8}). Thus \Eq{(2)}~will certainly be satisfied if \[ \log x \clt \log f \clt x^{\delta}. \] Hence a solution of our problem is given by \[ f = e^{(\log x)^{1+\delta}}. \] %% -----File: 030.png---Folio 22------- Similarly we can prove that \[ f = e^{(\log x)^{1-\delta}} \] satisfies \[ (\log x)^{\Delta} \clt f \clt x^{\delta}. \] It will be convenient to write \[ e_{0}x \eqq l_{0}x \eqq x, \] and then we have the relations \[ e_{0}(l_{1}x)^{\gamma} \clt e_{1}(l_{1}x)^{1-\delta} \clt e_{0}(l_{0}x)^{\gamma} \clt e_{1}(l_{1}x)^{1+\delta} \clt e_{1}(l_{0}x)^{\gamma}, \Tag{(3)} \] where $\gamma$~denotes \emph{any} positive number.\footnote {Here $\delta$, as usual, denotes `any positive number however small.' Of course, in using the index~$1 - \delta$, it is tacitly implied that $\delta < 1$.} Let us now consider the functions \[ f = e_{r}(l_{s}x)^{\mu}, \qquad f' = e_{r'}(l_{s'}x)^{\DPtypo{\mu}{\mu'}}, \] where $\mu$,~$\mu'$ are positive and not equal to~$1$. If $r = r'$, $f \cgt f'$ or $f \clt f'$ according as $s < s'$ or $s > s'$. If $s = s'$, the same relations hold according as $r > r'$ or $r < r'$. If $r = r'$ and $s = s'$, then $f \cgt f'$ or $f \clt f'$ according as $\mu > \mu'$ or $\mu < \mu'$. Leaving these cases aside, suppose $s > s'$, $s - s' = \sigma > 0$. Putting $l_{s'}x = y$, we obtain \[ f = e_{r}(l_{\sigma}y)^{\mu}, \qquad f' = e_{r'}y^{\mu'}. \] If $r < r'$ it is clear that $f \clt \phi$. If $r > r'$, let $r - r' = \rho$; then \[ l_{r}f = (l_{\sigma}y)^{\mu}, \qquad l_{r}f' = l_{\rho}y^{\mu'} \ceqq l_{\rho}y: \] if $\rho > 1$ the symbol~$\ceqq$ may be replaced by~$\sim$. If $\sigma > \rho$, $l_{r}f \clt l_{r}f'$ and so $f \clt f'$. If $\sigma < \rho$, $f \cgt f'$. If $\sigma = \rho$, $f \cgt f'$ or $f \clt f'$ according as $\mu > 1$ or $\mu < 1$. Thus \[ f \cgt f' \quad (r - s > r' - s'), \qquad f \clt f' \quad (r - s < r' - s'), \] while if $r - s = r' - s'$, $f \cgt f'$ or $f \clt f'$ according as $\mu > 1$ or $\mu < 1$, $\mu$~being the exponent of the logarithm of higher order which occurs in $f$~or~$f'$. From this it follows that \begin{gather*} \dots e_{1}(l_{2}x)^{1-\delta} \clt e_{0}(l_{1}x)^{\gamma} \eqq (lx)^{\gamma} \clt e_{1}(l_{2}x)^{1+\delta} \clt e_{2}(l_{3}x)^{1+\delta} \clt \dots\\ \dots \clt e_{2}(l_{2}x)^{1-\delta} \clt e_{1}(l_{1}x)^{1-\delta} \clt e_{0}(l_{0}x)^{\gamma} \eqq x^{\gamma} \clt e_{1}(l_{1}x)^{1+\delta} \clt \dots\\ \dots \clt e_{3}(l_{2}x)^{1-\delta} \clt e_{2}(l_{1}x)^{1-\delta} \clt e_{1}(l_{0}x)^{\gamma} \eqq ex^{\gamma} \clt e_{2}(l_{1}x)^{1+\delta} \clt \dots\DPtypo{}{.} \end{gather*} These relations enable us to interpolate to any extent among what we may call the fundamental logarithmico-exponential orders of infinity, viz.\ $(l_{k}x)^{\gamma}$, $x^{\gamma}$, $e_{k}x^{\gamma}$. Thus \[ e^{(lx)^{1+\delta}}, \quad e^{e^{(llx)^{1+\delta}}},\ \dots, \] and \[ e^{e^{(lx)^{1-\delta}}}, \quad e^{e^{e^{(llx)^{1-\delta}}}},\ \dots, \] are two scales, the first rising from above~$x^{\gamma}$, the second falling from below~$ex^{\gamma}$, and never overlapping. These scales, and the analogous scales which can be interpolated between other pairs of the fundamental logarithmico-exponential orders, possess %% -----File: 031.png---Folio 23------- another interesting property. The two scales written above \begin{Result}cover up \emph{(to put it roughly)} the whole interval between $x^{\gamma}$ and~$ex^{\gamma}$, so far as $L$-functions \(\Ref{iii.}{§\;2}\) are concerned\end{Result}: that is to say, it is impossible that an $L$-function~$f$ should satisfy \begin{alignat*}{2} f &\cgt e_{r}(l_{r}x)^{1+\delta}, &&\RTag{(\emph{every} $r$),}\\ f &\clt e_{r+1}(l_{r}x)^{1-\delta},&&\RTag{(\emph{every} $r$);} \end{alignat*} and the corresponding pairs of scales lying between $(l_{k+1}x)^{\gamma}$ and~$(l_{k}x)^{\gamma}$, or between $e_{k}x^{\gamma}$ and~$e_{k+1}x^{\gamma}$, possess a similar property. This property is analogous to that possessed (\Ref{iii.}{§\;3}) by the scales $(l_{r}x)$,~$(e_{r}x)$; viz.\ that no $L$-function~$f$ can satisfy $f \cgt e_{r}x$, or $1 \clt f \clt l_{r}x$, for all values of~$r$. A little consideration is all that is needed to render this theorem plausible: to attempt to carry out the details of a formal proof would occupy more space than we can afford. \Paragraph{2.} \begin{Example}\Item{(i)} Compare the rates of increase of \[ f = (lx)^{(lx)^{\mu}}, \qquad \phi = x^{(lx)^{-\nu}}. \] \end{Example} These functions are the same as $e\{(lx)^{\mu}\, llx\}$,~$e\{(lx)^{1-\nu}\}$. If $\mu + \nu \geqq 1$, $f \cgt \phi$; if $\mu + \nu < 1$, $f \clt \phi$. \begin{Example}\Item{(ii)} Compare the rates of increase of \[ f = x^{a}(lx)^{b}, \qquad \phi = e^{A(lx)^{\alpha}(llx)^{\beta}}, \qquad (a,\ A,\ \alpha > 0). \] \end{Example} Here $f = e(a\, lx + b\, llx)$. If $\alpha < 1$, then $f \cgt \phi$; if $\alpha > 1$, then $f \clt \phi$. If $\alpha = 1$, $\beta < 0$, then $f \cgt \phi$; if $\alpha = 1$, $\beta > 0$, then $f \clt \phi$. If $\alpha = 1$, $\beta = 0$, $a > A$, then $f \cgt \phi$; if $\alpha = 1$, $\beta = 0$, $a < A$, then $f \clt \phi$. If $\alpha = 1$, $\beta = 0$, $a = A$, then $f \cgt \phi$ if $b > 0$ and $f \clt \phi$ if $b < 0$. Finally if $\alpha = 1$, $\beta = 0$, $a = A$, $b = 0$ the two functions are identical. \begin{Example}\Item{(iii)} Compare the increase of $f = x^{\phi/(1+\phi)}$, where $\phi$~is a function of~$x$ such that $\phi \cgt 1$, with that of~$x^{\gamma}$. \end{Example} It is clear that $f \cleq x$, but $f \cgt x^{\gamma}$ for any value of~$\gamma$ less than unity. For, if $x$~is large enough, $\phi > n$, where $n$~is any positive integer, and so \[ f > x^{n/(1+n)}. \] Again $f = xe^{-lx/(1+\phi)}$, and so, if $\phi \clt lx$, $f \clt x$: but if $\phi \ceq lx$, $f \ceq x$; while if $\phi \cgt lx$, $f \sim x$. \Paragraph{3. Successive approximations to a logarithmico-exponential function.} Consider such a function as \[ f = \sqrt{x}(lx)^{2} e^{\sqrt{lx}(l_{2}x)^{2}e^{\sqrt{l_{2}x}(l_{3}x)^{2}}}. \] If we omit one or more of the parts of the expression of~$f$ we obtain another function whose increase differs more or less widely from that of~$f$. The question arises as to which parts are of the greatest and which of the least importance; \ie\ as to which are the parts whose omission affects the increase of~$f$ most or least fundamentally. Taking logarithms we find \[ \lf = \tfrac{1}{2}lx + \sqrt{lx}(l_{2}x)^{2} e^{\sqrt{l_{2}x}(l_{3}x)^{2}} + 2l_{2}x, \Tag{(a)} \] %% -----File: 032.png---Folio 24------- the three terms being arranged in order of importance. Again \[ l_{2}f = l_{2}x - l2 + \epsilon, \qquad l_{3}f = l_{3}x + \epsilon, \] where (\Ref{i.}{§\;5}) in each of the last equations $\epsilon$~denotes a function (not the same function) which tends to zero as $x \to \infty$. If we neglect this term in each of them in turn we deduce the approximations \[ \ITag{(1)} f = x,\qquad \ITag{(2)} f = \sqrt{x}. \] By neglecting the last term in the equation~\Eq{(a)} we obtain the much closer approximation \[ \ITag{(6)} f = \sqrt{x} e^{\sqrt{lx} (l_{2}x)^{2} e^{\sqrt{l_{2}x} (l_{3}x)^{2}}}. \] In order to obtain a more complete series of approximations to~$f$ we must replace the equation~\Eq{(a)} by a series of approximate equations. Now if \[ \phi = \sqrt{lx}(l_{2}x)^{2} e^{\sqrt{l_{2}x} (l_{3}x)^{2}} \] we have \begin{gather*} l\phi = \tfrac{1}{2}l_{2}x + \sqrt{l_{2}x} (l_{3}x)^{2} + 2l_{3}x,\\ l_{2}\phi = l_{3}x - l2 + \epsilon, \qquad l_{3}\phi = l_{4}x + \epsilon. \end{gather*} Hence we obtain (0)~$\phi = lx$, (3)~$\phi = \sqrt{lx}$, and (5)~$\phi = \sqrt{lx} e^{\sqrt{l_{2}x}(l_{3}x)^{2}}$ as approximations to the increase of~$\phi$: of these, however, the first is valueless, inasmuch as it would make~$\phi$ preponderate over the first term on the right hand side of~\Eq{(a)}. A similar argument, applied to the function $e^{\sqrt{l_{2}x}(l_{3}x)^{2}}$, leads us to interpolate (4)~$\phi = \sqrt{lx} e^{\sqrt{l_{2}x}}$ between (3)~and~(5). We can now, by adopting a series of approximate forms of the equation~\Eq{(a)}, deduce a complete system of closer and closer approximations to the increase of~$f$, viz.\ \begin{gather*} \ITag{(1)} x,\qquad \ITag{(2)} \sqrt{x},\qquad \ITag{(3)} \sqrt{x} e^{\sqrt{lx}},\qquad \ITag{(4)} \sqrt{x} e^{\sqrt{lx} e^{\sqrt{l_{2}x}}},\\ \ITag{(5)} \sqrt{x} e^{\sqrt{lx} e^{\sqrt{l_{2}x}(l_{3}x)^{2}}},\qquad \ITag{(6)} \sqrt{x} e^{\sqrt{lx}(l_{2}x)^{2} e^{\sqrt{l_{2}x}(l_{3}x)^{2}}}. \end{gather*} This order corresponds exactly to the order of importance of the various parts of the expression of~$f$. \Paragraph{4. Legitimate and illegitimate forms of approximation to a logarithmico-exponential function.} In applications of this theory, such as occur, for instance, in the theory of integral functions, we are continually meeting such equations as \[ f = (1 + \epsilon)e^{x^{\alpha}}, \qquad f = e^{(1+\epsilon)x^{\alpha}}, \qquad f = e^{x^{\alpha+\epsilon}}, \qquad (\alpha > 0). \Tag{(1)} \] It is important to have clear ideas as to the degree of accuracy of such representations of~$f$. The simplest method is to take logarithms repeatedly, as in §\;3~above. In the first example the term~$\epsilon$ does not affect the increase of~$f$: we have $f \sim ex^{\alpha}$. This is not true in the second; but $\lf \sim x^{\alpha}$, so that the term~$\epsilon$ does not affect the increase of~$\lf$; while in the third this is not true, though $\llf \sim \alpha$. Of the three formulae the first gives the most, and the last the least, information as to the increase of~$f$ (see also \Ref{vii.}{§\;3}). %% -----File: 033.png---Folio 25------- Such a formula as \[ f = xe^{(1+\epsilon)x^{\alpha}} \Tag{(2)} \] would not be a legitimate form of approximation at all. For the factor~$e(\epsilon x^{\alpha})$ which is not completely specified may well be far more important than the explicitly expressed factor~$x$: we might for example have $\epsilon = x^{-\beta}$, where $0 < \beta < \alpha$, in which case $e(\epsilon x^{\alpha})$ is more important than any power of~$x$. Thus \Eq{(2)}~does not really convey more information than the second equation~\Eq{(1)}, and to use it would involve a logical error similar to that involved in saying that the sun's distance is $92,713,600$~miles, with a probable error of some $100,000$~miles. \Paragraph{5. Attempts to represent orders of infinity by symbols.} It is natural to try to devise some simple method of representing orders of infinity by symbols which can be manipulated according to laws resembling as far as possible those of ordinary algebra. Thus Thomae\footnote {\textit{Elementare Theorie der analytischen Funktionen}, S.~112.} has proposed to represent the order of infinity of $f = x^{\alpha}(lx)^{\alpha_{1}}(l_{2}x)^{\alpha_{2}} \dots$ by \[ Of = \alpha + \alpha_{1}l_{1} + \alpha_{2}l_{2} + \dots,\footnote {The reader will not confuse this use of the symbol~$O$ (which does not extend beyond this paragraph) with that explained in \Ref{i.}{§\;5}.} \] where the symbols $l_{1}$,~$l_{2}$,~\dots\ are to be regarded as new units. It is clear that these units cannot, in relation to one another, obey the Axiom of Archimedes:\footnote {`If $x > y > 0$, we can find an integer~$n$ such that $ny > x$.'} however great~$n$, $nl_{2}$~cannot be as great as~$l_{1}$, nor $nl_{1}$~as great as~$1$. The consideration of a few simple cases is enough to show that any such notation, if it is to be of any use, must obey the following laws: \begin{alignat*}{2} &\Item{(i)} && \text{if} f \cgeq \phi, \qquad O(f + \phi) = Of;\\ &\Item{(ii)} &&\quad O(f\phi) = Of + O\phi;\\ &\Item{(iii)}\quad&&\quad O\{f(\phi)\} = Of × O\phi. \end{alignat*} And Pincherle\footnote {\lc\ (see \PageRef{p.}{13} above).} has pointed out that these laws are in any case inconsistent with the maintenance of the laws of algebra in their entirety. Thus if \[ Ox = 1, \qquad O\, lx = \lambda, \] we have, by~(iii), $O\, llx = \lambda^{2}$, and by (iii)~and~(ii), \[ O\, l(x\, lx) = \lambda(1 + \lambda) = \lambda + \lambda^{2}; \] and on the other hand, by~(i), \[ O\, l(x\, lx) = O(lx + llx) = \lambda. \] Pincherle has suggested another system of notation; but the best yet formulated is Borel's.\footnote {\textit{Leçons sur les séries à termes positifs}, pp.~35 \textit{et~seq.}; for further information see his recently published \textit{Leçons sur la théorie de la croissance}, pp.~14 \textit{et~seq.}} Borel preserves the three laws (i),~(ii),~(iii), the %% -----File: 034.png---Folio 26------- commutative law of addition, and the associative law of multiplication. But multiplication is no longer commutative, and only distributive on one side.\footnote {$(a + b)c = ac + bc$, but in general $a(b + c) \neq ab + ac$.} He would denote the orders of \begin{align*} e^{x}x^{n},&& x^{n}(lx)^{p},&& e^{2x},&& e^{x^{2}},&& e^{e^{x}},&& e^{\sqrt{lx}},&& \tfrac{1}{2} x, \\ \intertext{by} \omega + n,&& n + \frac{p}{\omega},&& 2 · \omega,&& \omega · 2,&& \omega^{2},&& \omega · \frac{1}{2} · \frac{1}{\omega},&& \frac{1}{\omega} · \frac{1}{2} · \omega. \end{align*} But little application, however, has yet been found for any such system of notation; and the whole matter appears to be rather of the nature of a mathematical curiosity. \end{Remark} \Chapter[Logarithmico-Exponential Scales.] {V.}{Functions Which do not Conform to any Logarithmico-Exponential Scale.} \Paragraph{1.} \First{We} saw in \Ref{i.}{(§\;2)} that, given two increasing functions $\phi$~and~$\psi$ ($\phi \cgt \psi$), we can always construct an increasing function~$f$ which is, for an infinity of values of~$x$ increasing beyond all limit, of the order of~$\phi$, and for another infinity of values of~$x$ of the order of~$\psi$. The actual construction of such functions by means of explicit formulae we left till later. We shall now consider the matter more in detail, with special reference to the case in which $\phi$~and~$\psi$ are $L$-functions. We shall say that $f$~is an \emph{irregularly increasing} function (\textit{fonction à croissance irrégulière}) if we can find two $L$-functions $\phi$~and~$\psi$ ($\phi \cgt \psi$) such that \[ f \geq \phi \quad (x = x_{1},\ x_{2},\ \dots), \qquad f \leq \psi \quad (x = x_{1}',\ x_{2}',\ \dots), \] $x_{1}$,~$x_{2}$,~\dots\ and $x_{1}'$,~$x_{2}'$,~\dots\ being any two indefinitely increasing sequences of values of~$x$. We shall also say that `the increase of~$f$ is irregular' and that `the logarithmico-exponential scales are \emph{inapplicable} to~$f$.' \begin{Remark} The phrase `\textit{fonction à croissance irrégulière}' has been defined by various writers in various senses. Borel\footnote {\textit{Leçons sur les fonctions entières}, p.~107.} originally defined $f$ to be \textit{à croissance régulière} if \[ e^{x^{\alpha-\delta}} < f < e^{x^{\alpha+\delta}}, \RTag{$(x > x_{0})$,} \] or in other words if $\llf \sim \alpha lx$ or $\llf \ceqq lx$. This definition was of course designed to meet the particular needs of the %% -----File: 035.png---Folio 27------- theory of integral functions: and has been made more precise by Boutroux and Lindelöf,\footnote {Boutroux, \textit{Acta Mathematica}, t.~28, p.~97; Lindelöf, \textit{Acta Societatis Fennicae}, t.~31, p.~1. See also Blumenthal, \textit{Principes de la théorie des fonctions entières d'ordre infini}.} who use inequalities of the form \[ e^{x^{\alpha}(lx)^{\alpha_{1}} \dots (l_{k}x)^{\alpha_{k}-\delta}} < f < e^{x^{\alpha}(lx)^{\alpha_{1}} \dots (l_{k}x)^{\alpha_{k}+\delta}}. \] All functions which are not \textit{à croissance régulière} for these writers are included in our class of irregularly increasing functions. \end{Remark} \Paragraph{2.} The logarithmico-exponential scales may fail to give a complete account of the increase of a function in two different ways. The function may be of irregular increase, as explained above, and the scales \emph{inapplicable}: on the other hand they may be, not inapplicable, but \emph{insufficient} (\textit{en~défaut}). That is to say, although the increase of the function does not oscillate from that of one $L$-function to that of another, there may be no $L$-function capable of measuring it. That such functions exist follows at once from the general theorems of~\Ref{ii}{}. Thus we can define a function which tends to infinity more rapidly than any~$e_{r}x$, or more slowly than any~$l_{r}x$: and the increase of such a function is more rapid or slower than that of any $L$-function (\Ref{iii.}{§\;2}). Or again, we can (\Ref{ii.}{§\;6}) define a function whose increase is greater than that of~$e_{r}(l_{r}x)^{1+\delta}$ (any~$r$) and less than that of~$e_{r+1}(l_{r}x)^{1-\delta}$ (any~$r$); and the increase of such a function (\Ref{iv.}{§\;1}) cannot be equal to that of any $L$-function. We shall now discuss some actual examples of functions for which the logarithmico-exponential scales are inapplicable or insufficient. \Paragraph{3. Irregularly increasing functions.} Functions whose increase is irregular may be constructed in a variety of ways. \begin{Remark} \Item{(i)} Pringsheim\footnote {See \textit{Math.\ Annalen}, Bd.~35, S.~347 \textit{et~seq.}\ and \textit{Münchener Sitzungsberichte}, Bd.~26, S.~605 \textit{et~seq.}} has used, in connection with the theory of the convergence of series, functions of an integral variable~$n$ whose increase is irregular. A simple example of such a function is \[ f(n) = 10^{[(\log_{10} n)^{1/\tau}]^{\tau}}, \RTag{$(\tau > 1)$,} \] where $[x]$~denotes the integral part of~$x$. It is easily proved, for instance, when $\tau = 2$, that the increase of~$f(n)$ varies between that of~$n$ and that of $n · 10^{1-2\sqrt{\log_{10}n}}$. We shall not do more than mention functions of this type. They are defined, most naturally, as functions of an integral variable~$n$: if we extend the definition to the continuous variable, the resulting function is discontinuous. The definition can of course be modified so as to give a %% -----File: 036.png---Folio 28------- continuous function of~$x$ with substantially the same properties; but it is not easy to effect this by a simple, natural, and explicit formula. \Item{(ii)} A more natural type of function is given by \[ f = \phi \cos^{2} \theta + \psi \sin^{2} \theta, \] where $\phi$,~$\psi$,~$\theta$ are increasing $L$-functions. We have to consider what conditions $\phi$,~$\psi$,~$\theta$ must satisfy in order that $f$~may increase steadily with~$x$. That its increase oscillates between that of~$\phi$ and that of~$\psi$ is obvious. Differentiating, \[ f' = \phi' \cos^{2} \theta + \psi' \sin^{2} \theta + 2(\psi - \phi)\theta' \cos \theta \sin \theta. \] Suppose $\phi \cgt \psi$: and let us assume that (as will be proved in the next chapter) relations between $L$-functions involving the symbols $\cgt$,~etc.\ may be differentiated and integrated. The condition that $f'$~should always be positive is $\phi'\psi' \cgt (\phi - \psi)^{2}\theta'^{2}$ or $\phi'\psi' \cgt \phi^{2}\theta'^{2}$. \textit{A~fortiori}, since $\phi' \cgt \psi'$, we must have $\phi' \cgt \phi\theta'$, or $\log\phi \cgt \theta$. Thus $f$~is certainly monotonic if \[ \phi \cgt \psi, \qquad \log\phi \cgt \theta, \qquad \psi' \cgt \phi^{2}\theta'^{2}/\phi'. \] If, \eg, $\theta = x$, we require $\log\phi \cgt x$, which is satisfied, for example, if $\phi = x^{\alpha} e^{x^{\rho}}$ ($\rho > 1$). It is convenient to write $a + \rho - 1$ for~$\alpha$. Then, since $\phi' \sim \rho x^{\alpha+\rho-1} e^{x^{\rho}}$, we must have $\psi' \cgt x^{a} e^{x^{\rho}}$; and so \[ \psi \cgt \int^{x} t^{a} e^{t^{\rho}}\, dt = \frac{1}{\rho} \int^{x} t^{a-\rho+1} \frac{d}{dt}\, (e^{t^{\rho}})\, dt \sim \frac{1}{\rho} x^{a-\rho+1} e^{x^{\rho}}, \] as is easily seen on integrating by parts. Thus we may take $\psi = x^{\beta} e^{x^{\rho}}$, where $\alpha - 2\rho + 2 < \beta < \alpha$. Changing our notation a little we see that \[ f = (x^{\gamma+\delta} \cos^{2} x + x^{\gamma-\delta} \sin^{2} x) e^{x^{\rho}} \] is monotonic if $0 < \delta < \rho-1$; and the increase of~$f$ obviously oscillates between that of~$x^{\gamma+\delta} e^{x^{\rho}}$ and that of~$x^{\gamma-\delta} e^{x^{\rho}}$. Similarly it may be shown that \[ f = (e^{\mu x} \cos^{2} x + e^{\nu x} \sin^{2} x) e^{e^{x}} \] is monotonic if $\nu < \mu < \nu + 2$;\footnote {Cf.\ \textit{Messenger of Mathematics}, vol.~31, p.~1.} and again the increase of~$f$ is irregular. \end{Remark} \Paragraph{4. Irregularly increasing functions (\continued).} We shall now consider two more general and more important methods for the construction of irregularly increasing functions. \Item{(iii)} Borel\footnote {See Borel, \textit{Leçons sur les fonctions entières}, pp.~120~\textit{et~seq.}; \textit{Leçons sur les séries à termes positifs}, pp.~32~\textit{et~seq}. Borel considers the cases only in which $\psi = e^{x}$, $\phi = e^{x^{2}}$ or~$e^{e^{x}}$; but his method is obviously of general application. The proof here given is however more general and much simpler.} has shown how, by means of power series, we may define functions which increase steadily with~$x$, while their increase oscillates to an arbitrary extent. %% -----File: 037.png---Folio 29------- Let \[ \phi(x) = \sum a_{n}x^{n}, \qquad \psi(x) = \sum b_{n}x^{n} \] be two integral functions of~$x$ with positive coefficients; suppose also $\phi \cgt \psi$. The increase of $\phi$~and~$\psi$ may be as large as we like (\Ref{ii.}{§\;4}); but in each case it must be greater than that of any power of~$x$. Then we can define a function \[ f(x) = \sum c_{n}x^{n}, \] where every~$c_{n}$ is equal either to~$a_{n}$ or to~$b_{n}$, in such a way that, for an infinity of values~$x_{\nu}$ whose limit is infinity, we have $f \sim \phi$, and for a similar infinity of values~$x_{\nu}'$ we have $f \sim \psi$.\footnote {By `$f \sim \phi$ for an infinity of values~$x_{\nu}$' we mean of course that $f/\phi \to 1$ as $x \to \infty$ through this particular sequence of values.} Let $(\eta_{\nu})$ be a sequence of decreasing positive numbers whose limit is zero. Take a positive number~$x_{0}$ such that $\phi(x_{0}) > 1$, $\psi(x_{0}) > 1$, and a number~$x_{1}$ greater than~$x_{0}$. When $x_{1}$~is fixed we can choose~$n_{1}$ so that \[ \sum_{n_{1}}^{\infty} a_{n}x_{1}^{n} < \tfrac{1}{3} \eta_{1}, \qquad \sum_{n_{1}}^{\infty} b_{n}x_{1}^{n} < \tfrac{1}{3} \eta_{1}, \] and so, if $c_{n}$~is either of $a_{n}$,~$b_{n}$ (however the selection may be made for different values of~$n$), \[ \sum_{n_{1}}^{\infty} c_{n}x_{1}^{n} < \sum_{n_{1}}^{\infty} (a_{n} + b_{n})x_{1}^{n} < \tfrac{2}{3} \eta_{1}. \] For $0 \leq n < n_{1}$ we take $c_{n} = a_{n}$. Then \[ |f(x_{1})-\phi(x_{1})| < \sum_{n_{1}}^{\infty} (a_{n} + c_{n})x_{1}^{n} < \eta_{1}, \] and so, since $\phi(x_{1}) > 1$, \[ \left|\frac{f(x_{1})}{\phi(x_{1})} - 1\right| < \eta_{1}. \Tag{(1)} \] Now let $x_{2}$ be a number greater than~$x_{1}$; we can suppose $x_{2}$~chosen so that \[ \biggl(\,\sum_{0}^{n_{1}-1} a_{n}x_{2}^{n}\biggr) \bigg/ \psi(x_{2}) < \tfrac{1}{5} \eta_{2}, \qquad \biggl(\,\sum_{0}^{n_{1}-1} b_{n}x_{2}^{n}\biggr) \bigg/ \psi(x_{2}) < \tfrac{1}{5} \eta_{2}. \] When $x_{2}$~is fixed we can choose~$n_{2}$ ($n_{2} > n_{1}$) so that \[ \sum_{n_{2}}^{\infty} a_{n}x_{2}^{n} < \tfrac{1}{5} \eta_{2}, \qquad \sum_{n_{2}}^{\infty} b_{n}x_{2}^{n} < \tfrac{1}{5} \eta_{2}. \] For $n_{1} \leqq n < n_{2}$ we take $c_{n} = b_{n}$. And, however $c_{n}$~be chosen for $n \geqq n_{2}$, we have \[ \sum_{n_{2}}^{\infty} c_{n}x_{2}^{n} < \sum_{n_{2}}^{\infty} (a_{n} + b_{n})x_{2}^{n} < \tfrac{2}{5} \eta_{2}. \] %% -----File: 038.png---Folio 30------- Also \begin{align*} %[** TN: Not aligned in the original] |f(x_{2}) - \psi(x_{2})| &< \sum_{0}^{n_{1}-1} a_{n}x_{2}^{n} + \sum_{0}^{n_{1}-1} b_{n}x_{2}^{n} + \sum_{n_{2}}^{\infty} c_{n}x_{2}^{n} + \sum_{n_{2}}^{\infty} b_{n}x_{2}^{n} \\ &< \tfrac{2}{5} \eta_{2} \psi(x_{2}) + \tfrac{3}{5} \eta_{2} < \eta_{2}\psi(x_{2}), \end{align*} and so \[ \left|\frac{f(x_{2})}{\psi(x_{2})} - 1\DPtypo{}{\right|} < \eta_{2}. \Tag{(2)} \] It is plain that, by a repetition of this process, we can find a sequence $x_{1}$,~$x_{2}$, $x_{3}$,~\dots\ whose limit is infinity, so that \[ % [** TN: Semantic \RTags, but using \RTag entails ad hoc spacing] \left|\frac{f(x_{3})}{\phi(x_{3})} - 1\right| < \eta_{3} \quad (3),\qquad % \left|\frac{f(x_{4})}{\psi(x_{4})} - 1\right| < \eta_{4} \quad(4),\qquad \dots; \] and our conclusion is thus established. Incidentally we may remark that not only $f$~itself, but all its derivatives also, are increasing and continuous. It is clear that, if we were given any number of integral functions $\phi_{1}$,~$\phi_{2}$, \dots,~$\phi_{k}$, with positive coefficients, we could define~$f$ so that $f/\phi_{s} \to 1$, as $x \to \infty$ through a suitably chosen sequence of values, for each of the functions~$\phi_{s}$. \begin{Remark} \Item{(iv)} \textbf{Power series with gaps.} There is another method of constructing irregularly increasing functions by means of power series which, though less general theoretically than that explained above, is in some ways more interesting, inasmuch as the functions to which it leads us are of a far simpler and more natural type. We shall confine ourselves here to explaining in general terms the general principle of the method and indicating a few simple examples.\footnote {For fuller details see Hardy, \textit{Proc.\ Lond.\ Math.\ Soc.}, vol.~2, pp.~332~\textit{et~seq.}; \textit{Messenger of Mathematics}, vol.~39, p.~28; Borel, \textit{Rendiconti del Circolo Matematico di Palermo}, t.~23, p.~320; \textit{Leçons sur la théorie de la croissance}, pp.~111~\textit{et~seq.}; Blumenthal, \textit{Principes de la théorie des fonctions entières d'ordre infini}, pp.~5~\textit{et~seq.}} Let \[ \phi(x) = \sum a_{n}x^{n} \Tag{(1)} \] be an integral function with positive coefficients: suppose, to fix our ideas, that the coefficients decrease steadily as $n$~increases. Suppose also that, for a particular value of~$x$, \[ \varpi(x) = a_{\nu} x^{\nu} \] is the greatest term of the series. In general one term will be the greatest, but for certain particular values of~$x$, say $\xi_{1}$,~$\xi_{2}$,~\dots, two consecutive terms will be equal.\footnote {We leave aside the possibility, which obviously applies only to particular cases, of more than two terms being equal.} %% -----File: 039.png---Folio 31------- As $x$~increases, the index~$\nu$ of~$\varpi(x)$ increases, and tends to~$\infty$ with~$n$: it thus defines a function~$\nu(x)$ such that \[ \nu(x) = i,\quad (\xi_{i} < x < \xi_{i+1}). \] At the point of discontinuity~$\xi_{i}$, where $\nu(x)$~jumps from $i - 1$ to~$i$, we may assign to it the value~$i$. When $\nu$~is thus defined for all values of~$x$, or $\varpi(x)$~defines a function of~$x$ which tends continuously and steadily to~$\infty$ with~$x$. The increase of~$\phi$ is obviously at least as great as that of~$\varpi$; it may be expected to be greater: but it is, in ordinary cases, not so very much greater---the increase of~$\varpi$ gives a very fair approximation to that of~$\phi$. Thus, if $\phi(x) = e^{x}$, $a_{n} = 1/n!$, and $\xi_{i} = i$. And for $i < x < i +1$ we have \[ e^{i} < \phi < e^{i+1}, \qquad (1 - \epsilon_{i}) \frac{e^{i}}{\sqrt{2\pi i}} < \varpi < (1 + \epsilon_{i}) \frac{e^{i+1}}{\sqrt{2\pi i}}.\footnote {The second pair of inequalities are an immediate consequence of Stirling's theorem, that $i! \sim i^{i+\frac{1}{2}} e^{-i} \sqrt{2\pi}$.}% \] Thus $\phi \cgt \varpi$, but $\log\phi \sim \log\varpi$: the difference between the increases of $\phi$ and~$\varpi$ is small compared with the increases themselves. Now let \[ f(x) = \sum a_{\chi(n)} x^{\chi(n)}, \Tag{(2)} \] where $\chi(n) \cgt n$: and let $p(x)$ be the function related to~$f$ as $\varpi(x)$~is to~$\phi$. The laws of increase of~$\varpi(x)$ and of~$p(x)$ may be expected to be very much the same, for $p(x)$~is defined by a selection from \emph{some} of the terms from \emph{all} of which $\varpi(x)$~was selected. The increase of~$f(x)$ clearly cannot be greater, and may be expected to be less, than that of~$\phi(x)$: but it cannot be less than that of~$p(x)$. Hence we may expect relations of the type \[ p \ceq \varpi \clt f \clt \phi.\footnote {We \emph{must} have $p \cleq \varpi$, $p \cleq f$, $\varpi \cleq \phi$, $f \cleq \phi$.} \] Also it is clear that, the more rapidly we suppose $\chi(n)$~to increase, the lower in the gap between $\varpi$ and~$\phi$ will $f$~sink, and that, if we suppose $\chi$~to increase with sufficient rapidity, we may expect to find $\varpi \ceq f$, so that the increase of~$f$ is completely dominated by that of one (variable) term. We then shall have \[ f(x) \ceq a_{N(x)}x^{N(x)}, \] where $N(x)$~is a function of~$x$ which assumes successively each of a series of integral values~$N_{i}$, so that \[ N(x) = N_{i}, \RTag{$(x_{i} \leqq x < x_{i+1})$.\footnotemark} \] \footnotetext{$N_{i}$,~$x_{i}$ are, of course, not the same as $\nu_{i}$,~$\xi_{i}$ above.}% But, as $x$~increases from $x_{i}$ to~$x_{i+1}$, the order of~$a_{N_{i}}x^{N_{i}}$, considered as a function of~$x$, may vary considerably, since $N_{i}$, though depending on the %[** TN: Hardy's notation for a closed interval; inconsistent, not modernizing] interval $(x_{i}, x_{i+1})$, does not depend on the particular position of~$x$ in that interval. And so it is clear that we are in this way likely to be led to functions whose increase is irregular in the sense explained in~§\;1. %% -----File: 040.png---Folio 32------- Suppose, for example, that $a_{n} = n^{-n}$, so that \[ \phi(x) = \sum \left(\frac{x}{n}\right)^{n} \sim \sqrt{\frac{2\pi x}{e}} e^{x/e}.\footnote {See \Ref{ii.}{§\;3}, and the references given in the footnote to \PageRef{p.}{10}. We might have taken $\phi(x) = e^{x}$, but our choice of~$\phi(x)$ leads to the simplest examples.} \] Here \[ \xi_{i} = i\left(1 + \frac{1}{i}\right)^{i+1} \sim ei, \] and it is easily shown that $\varpi(x) \ceq e^{x/e}$. Now let $\chi(n) = 2^{n}$, so that \[ f(x) = \sum \frac{x^{2^{n}}}{2^{n2^{n}}} = \sum v_{n} \] say. Then $v_{i-1} = v_{i}$ if $x = 2^{i+1}$, so that $x_{i} = 2^{i+1}$ and $N_{i} = 2^{i}$ for \[ 2^{i+1} \leqq x < 2^{i+2}. \] For this range of values of~$x$, $v_{i}$~is the greatest term; when $x = 2^{i+2}$, $v_{i} = v_{i+1}$. Further, it is not difficult to show that $f(x) \ceq p(x) = v_{i}$, the behaviour of~$f(x)$ being dominated by that of its greatest term.\footnote {We may say roughly that \emph{in general} $f \sim p$---that is to say, $f/p \to 1$ as~$x \to \infty$ through any sequence of values not falling inside any of certain intervals surrounding the values~$\xi_{i}$. At a point~$\xi_{i}$, $f/p$~is nearly equal to~$2$.} If we put $x = 2^{i+1+\theta}$, where $0 < \theta < 1$, we find \[ f(x) \ceq v_{i} = 2^{(1+\theta)2^{i}} = 2^{\alpha x}, \] where $\alpha = (1 + \theta)2^{-1-\theta}$. This is a maximum when $1 + \theta = 1/(\log 2)$, when it is equal to~$1/(e\log 2) = .53\dots$. Hence the increase of~$f(x)$ oscillates (roughly) between that of~$2^{.53\dots x}$ and~$2^{\frac{1}{2}x + 1}$.\footnote {The latter function is multiplied by~$2$, as there are two equal terms when $\theta = 0$ or~$1$.} Similar considerations may be applied to the more general series \[ \sum \frac{x^{a^{n}}}{b^{na^{n}}}, \] where $a$~is an integer greater than unity. This series is derived from $\sum (x/n^{a})^{n}$, where $\alpha = (\log b)/(\log a)$, by taking $\chi(n) = a^{n}$. Another example of an irregularly increasing function defined in a similar manner is \[ f(x) = \sum \frac{x^{n^{3}}}{(n^{3})!}, \] the increase of which oscillates between the increases of~$e^{x}/\sqrt{x}$ and \[ x^{-\frac{1}{2}} e^{x-\frac{9}{8}x^{1/3}}.\footnote {\textit{Messenger of Mathematics}, vol.~39, p.~28.} \] These examples are of course typical of a large class of functions. Before we leave this subject let us call attention to a point of considerable %% -----File: 041.png---Folio 33------- interest suggested by the foregoing examples. In forming the logarithmico-exponential scales we started from the scale $x$,~$x^{2}$,~\dots\ and then formed the function~$\sum \dfrac{x^{n}}{n!}$. If we had started, as we equally well might have done, from the scale $x^{2}$,~$x^{4}$, $x^{8}$,~\dots\ (cf.~\Ref{ii.}{§\;1}), we should have been led to choose, as a function transcending this scale, not~$e^{x}$ but some such function as \[ \sum \frac{x^{2^{n}}}{(2^{n})!}. \] \emph{This is one of the irregularly increasing functions of the type just considered.} Had we proceeded thus, and completed the construction of our fundamental scales on similar lines, our fundamental functions would for the most part have been among those which do not conform to the logarithmico-exponential scale, and it would have been the functions of that scale that would have appeared as irregularly increasing functions. \end{Remark} \Paragraph{5. Functions which transcend the logarithmico-exponential scales.} We turn our attention now to functions for which the logarithmico-exponential scales are not inapplicable but \emph{insufficient} (§\;2). Of the existence of such functions we are already assured. Thus a function which assumes the values $e_{1}(1)$,~$e_{2}(2)$, \dots,~$e_{\nu}(\nu)$,~\dots\ for $x = 1$, $2$,~\dots, $\nu$,~\dots\ certainly has an increase greater than that of any logarithmico-exponential function. No such function, however, has as yet made its appearance naturally in analysis; it will be sufficient, therefore, to mention two examples of such functions which transcend the logarithmico-exponential scales in quite different manners. \Item{(i)} The series \[ \sum \frac{e_{\nu}(x)}{e_{\nu}(\nu)} \] has obviously, if it converges, an increase greater than that of any~$e_{\nu}(x)$. Suppose $k - 1 \leqq x < k$. Then \[ \frac{e_{k}(x)}{e_{k}(k)} < 1, \qquad \frac{e_{k+\nu}(x)}{e_{k+\nu}(k+\nu)} < \frac{e_{k+\nu}(k)}{e_{k+\nu}(k+\nu)} < \frac{e_{k+\nu}(k)}{e_{k+\nu}(k+1)}. \] But, by the Mean Value Theorem, \[ e_{k+\nu}(k+1) = e_{k+\nu}(k) + e_{k+\nu}(y)e_{k+\nu-1}(y) \dots e_{2}(y)e_{1}(y), \] where $y$~is some number between $k$~and~$k + 1$; and so \[ e_{k+\nu}(k+1) > e_{k+\nu}(k)e_{k+\nu-1}(k) \dots e_{1}(k). \] It follows that the terms of the series \[ \sum_{\nu=k}^{\infty} \frac{e_{\nu}(x)}{e_{\nu}(\nu)} \] are less than those of the series \[ 1 + \sum_{\nu=1}^{\infty} \frac{1}{e_{1}(k)e_{2}(k) \dots e_{k+\nu-1}(k)}, \] %% -----File: 042.png---Folio 34------- which is plainly convergent, and therefore that the original series is convergent; and it is obviously only one of a large class of series possessing similar properties. \begin{Remark} (ii) Let $\phi(x)$~be an increasing function such that $\phi(0) > 0$, $\phi \cgt x$. We can define an increasing function~$f$, which satisfies the equation \[ \ff(x) = \phi(x), \Tag{(1)} \] as follows. Draw the curves $y = x$, $y = \phi(x)$ (\Fig{5}). Take $Q_{0}$~arbitrarily on~$OP_{0}$ (see the figure); draw~$Q_{0}R_{1}$ parallel to~$OX$ and complete the rectangle~$Q_{0}Q_{1}$. Join $Q_{0}$,~$Q_{1}$ by any continuous arc everywhere inclined at an acute angle to the axes. On this arc take any point~$Q$; draw $QP$,~$QR$ parallel to the axes, and complete the rectangle~$QQ'$. As $Q$~moves from $Q_{0}$ to~$Q_{1}$, $Q'$~moves from $Q_{1}$ to~$Q_{2}$, say. As we constructed $Q'$ from~$Q$, so we can construct $Q''$ from~$Q'$: proceeding thus we define a continuous curve $Q_{0}Q_{1}Q_{2}Q_{3}\dots$ corresponding to a continuous and increasing function~$f(x)$. Then $f(x)$~satisfies~\Eq{(1)}. For if $y = f(x)$ is the ordinate of~$Q$, it is clear that $\ff(x)$~is the ordinate of~$Q'$, which is equal to~$\phi(x)$, the ordinate of~$P$. %[Illustration: Fig. 5] \Figure[0.6\textwidth]{5}{042} Let us write \[ f(x) = f_{1}(x), \qquad \phi(x) = f_{1}f_{1}(x) = f_{2}(x), \qquad f\phi(x) = \phi f(x) = f_{3}(x), \] and so on, so that $Q_{n}$~is the point $f_{n}(0)$,~$f_{n+1}(0)$. Also let $\psi$~be the function inverse to~$\phi$, and write~$\psi_{2}$ for~$\psi\psi$, and so on. Finally, let the equation of~$Q_{0}Q_{1}$ be $\theta(x, y) = 0$. Then it is easy to see that the equations of~$Q_{2n}Q_{2n+1}$ and of~$Q_{2n+1}Q_{2n+2}$ are respectively \[ \theta\{\psi_{n}(x), \psi_{n}(y)\} = 0, \qquad \theta\{\psi_{n+1}(y), \psi_{n}(x)\} = 0. \] Suppose for example that $\phi(x) = e^{x}$, $OQ_{0} = a < 1$, and that $Q_{0}Q_{1}$~is the straight line $y = a + \alpha x$, where $\alpha = (1 - a)/a$. Then the equations of~$Q_{2n}Q_{2n+1}$ and of~$Q_{2n+1}Q_{2n+2}$ are \[ l_{n}y = a + \alpha l_{n}x, \qquad l_{n}x = a + \alpha l_{n+1}y, \] or \[ y = e_{n-1}\{e^{\alpha} (l_{n-1}x)^{\alpha}\}, \qquad y = e_{n}\{e^{-a/\alpha} (l_{n-1}x)^{1/\alpha}\}. \] %% -----File: 043.png---Folio 35------- For simplicity let us take $a = \frac{1}{2}$, $\alpha = 1$. Then the equations of~$Q_{2n}Q_{2n+1}$ and of~$Q_{2n+1}Q_{2n+2}$ are respectively \begin{alignat*}{3} y &= e_{n-1}\{\sqrt{e}(l_{n-1}x)\} &&= e_{n-2}\{(l_{n-2}x)^{\sqrt{e}}\} &&= \lambda_{n}(x),\\ y &= e_{n}\{(l_{n-1}x)/ \sqrt{e}\} &&= e_{n-1}\{(l_{n-2}x)^{1/\sqrt{e}}\} &&= \mu_{n}(x), \end{alignat*} say. Now (\Ref{iv.}{§\;1}) \[ x^{\gamma} \clt \lambda_{3} \clt \dots \clt \lambda_{n} \clt \dots \clt \mu_{n} \clt \dots \clt \mu_{3} \clt e^{x^{\gamma}} \] and a function~$f$, such that $\lambda_{n} \clt f \clt \mu_{n}$ for all values of~$n$, transcends the logarithmico-exponential scales. But $f$~clearly satisfies these relations, and so its increase is incapable of exact measurement by these scales. It is easily verified that $\lambda_{n}\lambda_{n}x \clt e^{x}$ and $\mu_{n}\mu_{n}x \cgt e^{x}$ for all values of~$n$. Hence it is clear \textit{a~priori} that any increasing solution of~\Eq{(1)} must satisfy $\lambda_{n} \clt f \clt \mu_{n}$. This kind of `graphical' method may also be employed to define functions whose increase, like that of the function considered under (i) above, is slower than that of any logarithm or more rapid than that of any exponential. It can be employed, for example, to solve the equation \[ \phi(2^{x}) = 2\phi(x); \] and it can be proved that the increase of a function such that $\phi(2^{x}) \ceq \phi(x)$ is slower than that of any logarithm (\Ref{vii.}{§\;3}). \end{Remark} \Paragraph{6. The importance of the logarithmico-exponential scales.} As we have seen in the earlier paragraphs of this section, it is possible, in a variety of ways, to construct functions whose increase cannot be measured by any $L$-function. It is none the less true that no one yet has succeeded in defining a mode of increase genuinely independent of all logarithmico-exponential modes. Our irregularly increasing functions oscillate, according to a logarithmico-exponential law of oscillation, between two logarithmico-exponential functions; the functions of~§\;5 were constructed expressly to fill certain gaps in the logarithmico-exponential scales. No function has yet presented itself in analysis the laws of whose increase, in so far as they can be stated at all, cannot be stated, so to say, in logarithmico-exponential terms. It would be natural to expect that the arithmetical functions which occur in the theory of the distribution of primes might give rise to genuinely new modes of increase. But, so far as analysis has gone, the evidence is the other way. \begin{Remark} Thus if we denote by~$\varpi(x)$ the number of prime numbers less than~$x$, it is known that \[ \varpi(x) \sim \frac{x}{\log x}. \] %% -----File: 044.png---Folio 36------- More precisely \[ \varpi(x) = \int_{2}^{x} \frac{dt}{\log t} + \rho(x) = \Li(x) + \rho(x), \] where $|\rho(x)| \clt x(\log x)^{-\Delta}$. The precise order of~$\rho(x)$ has not yet been determined, but there is reason to anticipate that $\rho(x) \cleq \sqrt{x}/(\log x)$. \end{Remark} \Chapter{VI.}{Differentiation and Integration.} \Paragraph{1. Integration.} It is important to know when relations of the types $f(x) \cgt \phi(x)$, etc., can be differentiated or integrated. The results are very much what might be expected from analogy with similar results in other branches of analysis, and may therefore be discussed somewhat summarily. For brevity we denote \[ \int_{a}^{x} f(t)\, dt, \qquad \int_{a}^{x} \phi(t)\, dt \] (where $a$~is a constant) by $F(x)$ and~$\Phi(x)$. And we suppose for the moment that $f$ and~$\phi$ are positive for $x \geqq a$. It may be well to repeat (cf.~\Ref{i.}{§\;4}) that $f$ and~$\phi$ are always supposed to be (at any rate for $x > x_{0}$) positive, continuous, and monotonic, unless the contrary is stated or clearly implied. Some of our conclusions are valid under more general conditions; but the case thus defined, and the corresponding case in which $f$ or~$\phi$ or~both of them are negative, are the only cases of importance. \begin{Lemma} If $\Phi \cgt 1$, and $f > H\phi$ for $x > x_{0}$, then $x_{1}$~can be found so that $F > (H - \delta)\Phi$ for $x > x_{1}$: similarly $f < h\phi$ for $x > x_{0}$ involves $F < (h + \delta)\Phi$ for $x > x_{1}$. \end{Lemma} For if $f > H\phi$ for $x > x_{0}$, we have \[ F = \int_{a}^{x} f\, dt > \int_{a}^{x_{0}} f\, dt + H \int_{x_{0}}^{x} \phi\, dt > H\Phi + \int_{a}^{x_{0}} f\, dt - H \int_{a}^{x_{0}} \phi\, dt, \] and if we choose $x_{1}$ so that \[ \left(\int_{a}^{x_{0}} f\, dt + H \int_{a}^{x_{0}} \phi\, dt\right) \bigg/ \Phi < \epsilon \] for $x \geq x_{1}$, as we certainly can if $\Phi \cgt 1$, the result follows. Similarly in the other case. From this lemma we can at once deduce the following %% -----File: 045.png---Folio 37------- \begin{Theorem} Any one of the relations \begin{alignat*}{5} f &\cgt \phi, \qquad& f &\clt \phi, \qquad& f &\ceq \phi, \qquad& f &\ceqq \phi, \qquad& f &\sim \phi \\ \intertext{involves the corresponding one of the relations} F &\cgt \Phi, \qquad& F &\clt \Phi, \qquad& F &\ceq \Phi, \qquad& F &\ceqq \Phi, \qquad& F &\sim \Phi \end{alignat*} if either $F \cgt 1$ or $\Phi \cgt 1$. \end{Theorem} To this we may add: \begin{Result}if both $\ds\int^{\infty} f\,dt$, $\ds\int^{\infty} \phi\,dt$ are convergent, then $f \cgt \phi$, $f \clt \phi$, $f \ceq \phi$, $f \ceqq \phi$, $f \sim \phi$ involve corresponding relations between \[ \bar{F} = \int_{x}^{\infty} f\,dt, \qquad \bar{\Phi} = \int_{x}^{\infty} \phi\,dt. \] \end{Result} The proof we may leave to the reader. These results have been stated primarily for the case in which $f$~and~$\phi$ are positive; but there is no difficulty in extending them to the case in which either function or both are negative. \Paragraph{2. Differentiation.} {\Loosen It follows from~§\;1 that $f \cgt \phi$ involves $f' \cgt \phi'$ if $f \cgt 1$ or $f \clt 1$ and \emph{if any one of the relations expressed by $\cgt$,~$\clt$, $\ceq$,~$\ceqq$,~$\sim$ holds between $f'$~and~$\phi'$}.} \begin{Remark} In interpreting this statement regard must be paid to the conventions laid down in \Ref{i.}{§\;4}. Thus if $f \cgt \phi \cgt 1$, $f'$~and~$\phi'$ are positive; and $f' \cgt \phi'$. But if $f \cgt 1 \cgt \phi$, $\phi$~is a decreasing function and $\phi' < 0$. In this case $f' \cgt -\phi'$, a relation which we have agreed to denote by $f' \cgt \phi'$. If $1 \cgt f \cgt \phi$ both $f'$~and~$\phi'$ are negative: the relation $-f' \clt -\phi'$ would involve \[ -\int_{x}^{\infty} f'\,dt \clt -\int_{x}^{\infty} \phi'\,dt \] or $f \clt \phi$, and is therefore impossible; similarly for $-f' \ceq -\phi'$; so we must have $-f' \cgt -\phi'$, a relation which we have agreed also to denote by $f' \cgt \phi'$. The case in which $f \ceq 1$ is exceptional; any one of the relations $f' \cgt \phi'$,~etc.\ may then hold. Thus if $f = 1 + e^{-x}$, $f' = 1/x$, we have $f \cgt \phi$, $f' \clt \phi'$. The fact is that in this case $f$, regarded as the integral of~$f'$, is dominated by the constant of integration. \end{Remark} Similar results hold, of course, for the relations $f \clt \phi$,~etc., with similar exceptions. With regard to all of them it is to be observed that the assumption that one of the relations holds between $f'$~and~$\phi'$ is essential. We can never \emph{infer} that one of them holds. We cannot even infer that $f'$~or~$\phi'$ is a steadily increasing or decreasing function at all. Thus if $f = e^{x}$, $\phi = e^{x} + \sin e^{x}$, we have $f' = e^{x}$ and $\phi' = e^{x}(1 + \cos e^{x})$. Thus $f$~and~$\phi$ increase steadily and $f \sim \phi$, $f' \sim f$; %% -----File: 046.png---Folio 38------- but $\phi'$~does not tend to infinity (vanishing for an infinity of values of~$x$). Again if \[ \phi = e^{x}(\sqrt{2} + \sin x) + \tfrac{1}{2} x^{2}, \] we have \[ \phi' = e^{x} (\sqrt{2} + \sin x + \cos x) + x \] and $\phi \ceq e^{x}$, while $\phi'$~oscillates between the orders of $e^{x}$ and~$x$. It is possible, though less easy, to obtain examples of this character in which $\phi'$~also is monotonic. \Paragraph{3. Differentiation of $L$-functions.} If $f$~and~$\phi$ are $L$-functions, so are $f'$~and~$\phi'$, and one of the relations $f' \cgt \phi'$, $f' \ceq \phi'$, $f' \clt \phi'$ certainly holds (\Ref{iii.}{§\;2}). Thus in this case \emph{both differentiation and integration are always legitimate}\footnotemark---this statement, however, being subject to certain exceptions in the cases in which $f \ceq 1$ or $\phi \ceq 1$. \footnotetext{A tacit assumption to this effect underlies much of Du~Bois-Reymond's work.} In what follows we shall suppose that all the functions concerned are $L$-functions, or at any rate resemble $L$-functions in so far that one of the relations $f \cgt \phi$, $f \ceqq \phi$, $f \clt \phi$ is bound to hold between any pair of functions, and that differentiation and integration are permissible.\footnote {The results which follow are all in substance due to Du~Bois-Reymond.} \begin{Result}[1.] If $f$~is an increasing function, and $f' \cgt f$, then $f \cgt e^{\Delta x}$. If $f' \clt f$, then $f \clt e^{\delta x}$. Similarly if $f$~is a decreasing function, $f' \cgt f$ and $f' \clt f$ involve $f \clt e^{-\Delta x}$ and $f \cgt e^{-\delta x}$ respectively. If $f' \ceqq f$, then $e^{\delta x} \clt f \clt e^{\Delta x}$ or $e^{-\Delta x} \clt f \clt e^{-\delta x}$, and we can find a number~$\mu$ such that $f = e^{\mu x} f_{1}$, where $e^{-\delta x} \clt f_{1} \clt e^{\delta x}$. \end{Result} The proofs of these assertions are almost obvious. Thus if $f$~is an increasing function, and $f' \cgt f$, we have \[ f'/f \cgt 1, \qquad \log f \cgt x, \] and so $\log f > \Delta x$ for $x > x_{0}$, \ie\ $f > e^{\Delta x}$, or, what is the same thing, $f \cgt e^{\Delta x}$. The last clause of the theorem follows at once from~\Ref{iii.}{§\;4}. \begin{Result}[2.] More generally, if $v$~is any increasing function, $f'/f \cgt v'/v$ involves $f \cgt v^{\Delta}$ or $f \clt v^{-\Delta}$, according as $f$~is an increasing or a decreasing function; and $f'/f \clt v'/v$ involves $f \clt v^{\delta}$ or $f \cgt v^{-\delta}$. And $f'/f \ceqq v'/v$ involves $v^{\delta} \clt f \clt v^{\Delta}$ or $v^{-\Delta} \clt f \clt v^{-\delta}$; and then we can find a number~$\mu$ such that $f = v^{\mu}f_{1}$, where $v^{-\delta} \clt f_{1} \clt v^{\delta}$. \end{Result} When $f$~is an increasing function we shall call $f'/f$ the \emph{type}~$t$ of~$f$:\footnote {Du~Bois-Reymond calls $f/f'$ the type; the notation here adopted seems slightly more convenient.} it being understood that $t$~may be replaced by any simpler function~$\tau$ such that $t \ceqq \tau$. The type of a \emph{decreasing} function~$f$ we define to be %% -----File: 047.png---Folio 39------- the same as that of the increasing function~$1/f$. The following table shews the types of some standard functions: \[ \begin{array}{lcccccrlcc} \text{\textit{Function}} & 1 & llx & lx & x^{\alpha} & e^{x} & e^{\alpha x^{\beta}} & e_{2}x & e_{3}x & \dots \\ \text{\textit{Type}} & 0 & \dfrac{1}{x\, lx\, llx} & \dfrac{1}{x\, lx} & \dfrac{1}{x} & 1 & x^{\beta-1} & ex & e_{2}x\,ex & \dots \end{array} \] \begin{Remark} If $f \cgt \phi$, then $f'/f \cgeq \phi'/\phi$. By making the increase of~$f$ large enough we can make the increase of $t = f'/f$ as large as we please. The reader will find it instructive to write out formal proofs of these propositions, and also of the following. \Item{1.} As the increase of~$f$ becomes smaller and smaller, $f'/f$~tends to zero more and more rapidly, but, so long as $f \to \infty$ at all, we cannot have \[ f'/f \clt \phi, \qquad \int^{\infty} \phi\, dx \quad \text{\emph{convergent}}. \] On the other hand, if the last integral is divergent we can find~$f$ so that $f \cgt 1$, $f'/f \clt \phi$. \Item{2.} Although we can find~$f$ so that $f'/f$~shall have an increase larger than that of any given function of~$x$, we cannot have \[ f'/f \cgt \phi(f), \qquad \int^{\infty} \frac{dx}{x\phi(x)} \quad \text{\emph{convergent}}. \] On the other hand, if the last integral is divergent we can find~$f$ so that $f'/f \cgt \phi(f)$. {\Loosen[Thus we cannot find a function~$f$ which tends to infinity so slowly that $f'/f \clt 1/x^{\alpha}$ ($\alpha > 1$). But we can find~$f$ so that $f'/f \clt 1 / x\, lx\, llx$ (\eg~$f = l_{3}x$). We cannot find~$f$ so that $f'/f \cgt f^{\alpha}$ or $f' \cgt f^{1+\alpha}$ ($\alpha > 0$). But we can find~$f$ so that $f'/f \cgt \lf$ (\eg~$f = e_{3}x$).]} \Item{3.} If $f \cgt e_{k}x$ for all values of~$k$, $f'/f$~satisfies the same condition, and \[ f' \cgt f\, \lf\, l_{2}f \dots l_{k}f. \] He will also find it profitable to formulate corresponding theorems about functions of a positive variable~$x$ which tends to zero. \end{Remark} \Paragraph{4. Successive differentiation.} Du~Bois-Reymond has given the following general theorem, which enables us to write down the increase of any derivative of any logarithmico-exponential function. We write $t$ for~$f'/f$, as in the last section, and we assume that no derivative~$f^{(n)}$ satisfies $f^{(n)} \ceqq 1$: if this should be the case the results of the theorem, so far as the derivatives $f^{(n+1)}$,~\dots\ are concerned, cease to be true. \begin{Theorem} \Item{(i)} If $t \cgt 1/x$ \(so that $f \cgt x^{\Delta}$\) then \[ f \ceqq f'/t \ceqq f''/t^{2} \ceqq f'''/t^{3} \dots \ceqq f^{(n)}/t^{n} \dots. \] %% -----File: 048.png---Folio 40------- \Item{(ii)} If $t \clt 1/x$ \(so that $f \clt x^{\delta}$\) then \[ f \ceqq f'/t \ceqq xf''/t \ceqq x^{2}f'''/t \dots \ceqq x^{n-1} f^{(n)}/t \dots. \] \Item{(iii)} If $t \ceqq 1/x$ \(so that $f = x^{\mu} f_{1}$, where $x^{-\delta} \clt f_{1} \clt x^{\delta}$\), then if $\mu$~is not integral either set of formulae is valid. But if $\mu$~is integral \[ f \ceqq xf' \ceqq x^{2}f'' \dots \ceqq x^{\mu}f^{(\mu)} \ceqq x^{\mu} f^{(\mu+1)}/t_{1} \ceqq x^{\mu+1}f^{(\mu+2)}/t_{1} \dots, \] where $t_{1}$ is the type of~$f_{1}$. \end{Theorem} \Item{(i)} If $t \cgt 1/x$, $1/t \clt x$ and so $t'/t^{2} \clt 1$; hence $t'/t \clt t = f'/f$ or \[ ft' \clt f't. \] Differentiating the relation $f' \ceqq ft$, and using the relation just established, we obtain \[ f'' \ceqq f't + ft' \ceqq f't. \] Thus the type of~$f'$ is the same as that of~$f$; accordingly the argument may be repeated and the first part of the theorem follows. \Item{(ii)} If $t \clt 1/x$, $xf' \clt f$ and so \[ xf'' + f' \clt f'. \] But this cannot possibly be the case unless $xf'' \ceqq f'$. Differentiating again we infer \[ xf''' + 2f'' \clt f'', \] whence $xf''' \ceqq f''$; and so on generally.\footnote {More precisely $xf'' \sim -f'$, $xf''' \sim -2f''$, and so on.} Thus the second part follows. \Item{(iii)} If $t \ceq 1/x$, $f = x^{\mu}f_{1}$ and $t_{1}$,~the type of~$f_{1}$, satisfies $t_{1} \clt 1/x$. Then \[ f' = \mu x^{\mu-1} f_{1} + x^{\mu}f_{1}' \ceqq x^{\mu-1} f_{1}(\mu + xt_{1}) \ceqq x^{\mu-1}f_{1}; \] Similarly $f'' \ceqq x^{\mu-2}f_{1}$ and so on. We can proceed indefinitely in this way unless $\mu$~is integral: in this case we find $f^{(\mu)} \ceq f_{1}$, and from this point we proceed as in case~(ii). \begin{Remark} \textit{Examples.} \Item{(i)} If $f = e^{\sqrt{x}}$, then $t = 1/\sqrt{x} \cgt 1/x$, and $f^{(n)} \ceqq e^{\sqrt{x}}/(\sqrt{x})^{n}$. If $f = e^{(\log x)^{2}}$, then $t = (\log x)/x \cgt 1/x$, and $f^{(n)} \ceqq e^{(\log x)^{2}} (\log x)^{n}/x^{n}$. \Item{(ii)} If $f = (\log x)^{m}$, then $t = 1/(x\log x) \clt 1/x$, and \[ f^{(n)} \ceqq tx^{-(n-1)}f \ceqq (\log x)^{m-1}/x^{n}. \] \Item{(iii)} If $f = x^{2}\, llx$, $t \ceqq 1/x$. Here \[ f' \ceqq x\, llx, \qquad f'' \ceqq llx, \qquad f''' \ceqq 1/x\, lx, \qquad f'''' \ceqq 1/x^{2}\, lx,\ \dots. \] \Item{(iv)} The results of the theorem, in the first two cases, can be stated more precisely as follows: If $t \cgt 1/x$, then \[ f^{(n)} \sim (f'/f)^{n}f. \] %% -----File: 049.png---Folio 41------- If $t \clt 1/x$, then \[ f^{(n)} \sim (-1)^{n-1} (n - 1)!\, x^{-(n-1)}f'. \] If $f$~is a positive increasing function, then if $t \cgt 1/x$ all the derivatives are ultimately positive, while if $t \clt 1/x$ they are alternately ultimately positive and ultimately negative. \end{Remark} \Paragraph{5. Functions of an integral variable.} The theorems for functions of an integral variable~$n$, corresponding to those of §§\;1--4, involve sums \[ A_{n} = a_{1} + a_{2} + \dots + a_{n} \] in place of integrals, and differences \[ \Delta a_{n} = a_{n} - a_{n+1} \] instead of differential coefficients. The reader will be able to formulate and to prove for himself the theorems which correspond to those of~§\;1. Thus \begin{quote}`\begin{Result}% $a_{n} \cgt b_{n}$, $a_{n} \clt b_{n}$, $a_{n} \ceq b_{n}$, $a_{n} \ceqq b_{n}$, $a_{n} \sim b_{n}$ involve the corresponding equations for $A_{n}$,~$B_{n}$, if one at least of $A_{n}$,~$B_{n}$ tends to infinity with~$n$% \end{Result}' \end{quote} and so on.\footnote {This is of course the well known theorem of Cauchy and Stolz: see Bromwich, \textit{Infinite Series}, p.~377.} Considerations of space forbid that we should go further into the subject here. \Chapter[Developments of the Infinitärcalcül.] {VII.}{Some Developments of Du~Bois-Reymond's Infinitärcalcül.} \Paragraph{1.} \First{We} shall conclude our account of the general theory by a brief sketch of some interesting results due in the main to Du~Bois-Reymond. For further details we must refer to his memoirs catalogued in the Bibliographical Appendix. \Section{The functions $\dfrac{f(x + a)}{f(x)}$, $\dfrac{f(ax)}{f(x)}$, etc.} It is often necessary to obtain approximations to such functions as \[ f(x + a)/f(x), \] where $a$~is itself a function of~$x$, which for simplicity we suppose positive, and which may tend to infinity with~$x$. In this connection %% -----File: 050.png---Folio 42------- Du Bois-Reymond\footnote {\textit{Math.\ Annalen}, Bd.~8, S.~363 \textit{et~seq.}} has proved a whole series of theorems: it will be sufficient for our present purpose to give a few specimens of his results. In what follows it will be assumed throughout that all the functions dealt with are $L$-functions, or at any rate such that any pair of them satisfy one of the relations $f \cgt \phi$, $f \ceqq \phi$, $f \clt \phi$, and that such relations may be differentiated or integrated. This being so we have \[ \frac{f(x + a)}{f(x)} = e^{\lf(x + \alpha) - \lf(x)} = e\left\{a\frac{f'(x + \alpha)}{f(x + \alpha)}\right\}, \] where $0 < \alpha < a$. This expression has certainly the limit unity if $f' \cleq f$ and $a \clt 1$. Hence \[ f(x + a) \sim f(x) \Tag{(1)} \] if $a \clt 1$ and $e^{-\Delta x} \clt f \clt e_{\Delta x}$. If $f'/f \clt 1$, \ie\ if $e^{-\delta x} \clt f \clt e^{\delta x}$, the relation~\Eq{(1)} holds for $a \clt f/f'$: it certainly holds, for instance, if $a = x\{f(x)\}^{-\mu}$, where $\mu > 0$, since $x/f^{\mu} \clt f/f'$ whenever $f \cgt 1$.\footnote {For $\ds\int^{\infty} f^{-1-\mu} f'\,dx$ is convergent, and so $f'/f^{1+\mu} \clt 1/x$.} If $a \ceqq f/f'$ (as \eg\ if $f = e^{\mu x}f_{1}$, where $e^{-\delta x} \clt f_{1} \clt e^{\delta x}$, and $a \ceqq 1$), $f(x + a)/f(x)$ will tend to a limit different from unity. Again \[ \frac{f(x + a)}{f(x)} = e\left\{a\frac{f'(x)}{f(x)}\, \frac{t(x + \alpha)}{t(x)}\right\}, \] where $t = f'/f$. Hence \[ \frac{f(x + a)}{f(x)} = e\left\{u\frac{f'(x)}{f(x)}\right\} \quad (u \sim a) \Tag{(2)} \] {\Loosen in all cases in which $t(x + \alpha)/t(x) \sim 1$; as for example if $a \cleq 1$, $e^{-\delta x} \clt t \clt e^{\delta x}$, or, what is the same thing, if} \[ a \cleq 1, \qquad e^{-e^{\delta x}} \clt f \clt e^{e^{\delta x}}. \] The reader will find it instructive to write down conditions under which the equation~\Eq{(2)} holds when $u \ceqq a$ is substituted for $u \sim a$, and to consider in what circumstances either relation holds when $a \cgt 1$. \Paragraph{2.} The reader is also recommended to verify some of the following results: \begin{Remark} \begin{Result} \Item{(i)} If $1 \clt a \clt x$ and $x^{-\Delta} \clt f \clt x^{\Delta}$, then $f(x + a)/f(x) \sim 1$. \Item{(ii)} \Squeeze{If $f \clt x$ and $a \clt 1/f'$, or if $f \ceqq x$ and $a \clt 1$, then $f(x + a) - f(x) \clt 1$}. \Item{(iii)} If $e^{-\delta x} \clt f \clt e^{\delta x}$ and $a \clt f'/f''$, then \[ f(x + a) - f(x) \sim af'(x). \] \end{Result} %% -----File: 051.png---Folio 43------- The condition $a \clt f'/f''$ may be simplified by means of the theorem of \Ref{vi.}{§\;4}. Thus if $t \clt 1/x$ (\ie\ if $f \clt x^{\delta}$) it is equivalent to $a \clt x$. \begin{Result} \Item{(iv)} If $x^{-\delta} \clt a \clt x^{\delta}$, $(lx)^{-\Delta} \clt f \clt (lx)^{\Delta}$, then $f(ax)/f(x) \sim 1$. \Item{(v)} If $e^{-\Delta\sqrt{lx}} \clt f \clt e^{\Delta\sqrt{lx}}$, then \[ \frac{f\{xf(x)\}}{f(x)} \ceqq 1, \qquad e\left\{\frac{x\, \lf(x)f'(x)}{f(x)}\right\} \ceqq 1; \] and the limits of the two functions are the same: and if $e^{-\delta\sqrt{lx}} \clt e^{\delta\sqrt{lx}}$ this limit is unity. \end{Result} Suppose, \eg\ $f \cgt 1$, and let $f(x) = \phi(lx)$; then, if $a = f(x)$, \[ \frac{f(ax)}{f(x)} = e^{l\phi(lx + la) - l\phi(lx)} = e^{la\phi'(lx + la_{1})/\phi(lx + la_{1})}, \] where $1 < a_{1} < a$. The exponent is \[ l\phi(lx + la_{1}) \frac{\phi'(lx + la_{1})}{\phi(lx + la_{1})}\, \frac{l\phi(lx)}{l\phi(lx + la_{1})}. \] Now $a = f(x) \clt x^{\delta}$ and therefore $la_{1} \cleq la \clt lx$, and so, by~(i), \[ l\phi(lx + la_{1}) \sim l\phi(lx) \] if $l\phi \clt x^{\Delta}$ or if $f \clt e^{(lx)^{\Delta}}$, which is certainly the case. Hence the exponent is asymptotically equivalent to \[ l\phi(u) \phi'(u)/\phi(u), \] where $u = lx + la_{1}$. And $l\phi(\phi'/\phi) \cleq 1$ if $(l\phi)^{2} \cleq u$, \ie\ if $\phi \cleq e^{\Delta\sqrt{u}}$ or $f \cleq e^{\Delta\sqrt{lx}}$. In this case $f(ax) \ceqq f(x)$; and it is easy to see that if $f \cleq e^{\delta\sqrt{lx}}$ the symbol~$\ceqq$ may be replaced by~$\sim$. \Item{(vi)} \emph{If $f(x) = x\phi(x)$, and $e^{-\delta\sqrt{lx}} \clt \phi \clt e^{\delta\sqrt{lx}}$, then} \[ f_{2}(x) \eqq \ff(x) \sim x\phi^{2},\ \dots,\ f_{n} \sim x\phi^{n},\ \dots. \] The reader will easily prove this by the aid of the preceding results. He will also find it instructive to calculate the increase of~$f_{n}$ when $f = e^{\sqrt{lx}}$ and when $f = e^{(lx)^{\alpha}}$, where $\alpha > \frac{1}{2}$. \end{Remark} \Section{The accuracy of approximations.} \Paragraph{3.} We have already (\Ref{iv.}{§§\;3--4}) had occasion to use the notion of an approximation to the increase of a function, and to distinguish legitimate and illegitimate forms of approximation. Du~Bois-Reymond has given the following more precise definitions. He defines $\psi(x, u, u_{1}, \dots)$ to be an `approximate form' of~$y$ if \[ y = \psi(x, u, u_{1}, \dots), \] $\psi$~being a known function, and $u$,~$u_{1}$,~\dots\ unknown functions whose increase is, however, subject to certain limitations. It is clear that it is really useless, however, to insert more than one unknown function~$u$ %% -----File: 052.png---Folio 44------- in~$\psi$. The effect of the presence of~$u$ is to define a certain stretch within which the increase of~$y$ lies, and the presence of several~$u$'s can effect no more. We shall therefore consider only approximate forms of the type \[ y = \psi(x, u). \Tag{(1)} \] Thus \[ e^{x^{u}} \quad (u \sim 1), \qquad e^{(1+u)x} \quad (u \clt 1), \qquad x^{1+u}e^{x} \quad (u \clt 1) \Tag{(2)} \] are approximate forms of $y = xe^{x}/lx$; the second clearly closer than the first and the third than the second. The closeness of an approximation may be measured as follows. The presence of~$u$ in~\Eq{(1)} lends a certain degree of indeterminateness to the increase of~$y$: all that we can say (the increase of~$u$ being known to lie between certain limits) is that $y$~lies in a certain interval \[ \eta_{1} \cleq y \cleq \eta_{2}. \] Now (\Ref{ii.}{§\;8}) we can find an increasing function~$F$ so that $F(\eta_{1}) \ceq F(\eta_{2})$: if $F$~satisfies this condition, any more slowly increasing function will do so too. \begin{Result}The slower the increase of~$F$ must be taken, the rougher the approximation.\end{Result} {\Loosen The facts may be stated the other way round. Given~$y$, and a function~$F$, such that $1 \clt F \clt x$, we can determine an interval $\eta_{1} \cleq y \cleq \eta_{2}$ such that $F(\eta_{1}) \ceq F(\eta_{2})$. The slower the increase of~$F$, the larger this interval may be taken; if $F \ceq x$ it vanishes, if $F \ceq 1$ %[** TN: Hardy's notation for a closed interval; inconsistent, not modernizing] it may be taken as large as we please. If $F = lx$ it might be $(y^{\delta}, y^{\Delta})$; if $F = l_{2}x$ it might be} \[ e^{(ly)^{\delta}}, \qquad e^{(ly)^{\Delta}}, \] and so on. No logarithmico-exponential form of~$F$, however, can give an interval as large as~$(\log y, e^{y})$; a function~$F$ such that $F(y) \ceq F(e^{y})$ must transcend any logarithmico-exponential scale. \begin{Remark} Let us consider the approximations~\Eq{(2)} for~$xe^{x}/lx$. \Item{(i)} If $y = e^{x^{u}}$ ($u\sim l$), $y$~lies in the interval $e^{x^{1-\delta}}$, $e^{x^{1+\delta}}$. Since \[ ll(e^{x^{1-\delta}}) = (1 - \delta)lx \ceq ll(e^{x^{1+\delta}}) \] we may take $F = llx$, or even $F = (llx)^{\Delta}$: but the increase of~$F$ cannot be taken as large as~$(lx)^{\delta}$. \Item{(ii)} If $y = e^{(1+u)x}$ ($u \clt 1$), $y$~lies in the interval $e^{(1-\delta)x}$, $e^{(1+\delta)x}$. Then we may take $F = (lx)^{\Delta}$, but we cannot take $F = e^{(lx)^{\delta}}$. \Item{(iii)} {\Loosen If $y = x^{1+u}e^{x}$ we may, as the reader will easily verify, take $F = e^{(lx)^{\mu}}$, where $\mu$~is any number less than unity.} %% -----File: 053.png---Folio 45------- Another example of an approximation is given by the formula \[ \frac{f(x + a)}{f(x)} = e\left\{u\frac{f'(x)}{f(x)}\right\} \quad (u \sim a). \] If, \eg, $a$~is a constant, \[ l\left\{\frac{f(x + a)}{f(x)}\right\} \sim l\left\{e\left[\frac{f'(x)}{f(x)}\right]\right\}, \] and the degree of accuracy of the approximation is great enough to be measured by the function $F = lx$. \end{Remark} \Section{The approximate solution of equations.} \Paragraph{4.} It is often important to obtain an asymptotic solution of an equation $f(x, y) = 0$, \ie\ to find a function whose increase gives an approximation to that of~$y$. No very general methods of procedure can be given, but the kind of methods which may be pursued are worth illustrating by a few examples. \Item{(i)} Suppose that the equation is \[ x = y\kappa(y), \Tag{(1)} \] where $y^{-\delta} \clt \kappa \clt y^{\delta}$. If the increase of~$\kappa$ is so slow that $\kappa\{y\kappa(y)\} \ceq \kappa(y)$ it is clear that \[ y \ceq x/\kappa(y) \ceq x/\kappa(x): \] and if the increase of~$\kappa$ is slow enough we may have $y \sim x/\kappa(x)$. The conditions \[ e^{-\Delta\sqrt{ly}} \clt \kappa(y) \clt e^{\Delta\sqrt{ly}}, \qquad e^{-\delta\sqrt{ly}} \clt \kappa(y) \clt e^{\delta\sqrt{ly}} \] are, by the result~(v) of~§\;2, enough to ensure the truth of these hypotheses; and then $y = ux/\kappa(x)$, where $u \ceq 1$ (or $u \sim 1$) is an approximate solution of our equation. \begin{Remark} Du~Bois-Reymond has proved that the more elaborate approximations \[ y = ux/\{\kappa(x/\kappa)\}, \qquad y = ux\kappa^{-1/\{1+(x\kappa'/\kappa)\}} \] have a wider range of validity: and that more elaborate approximations still may be constructed valid within the range \[ e^{-\Delta(ly)^{1-\delta}} \clt \kappa \clt e^{\Delta (ly)^{1-\delta}}. \] \end{Remark} The more general equation \[ x = y^{m}\kappa(y) \] can clearly be reduced to the form considered above by writing~$x^{m}$ for~$x$ and $\kappa^{m}$ for~$\kappa$. %% -----File: 054.png---Folio 46------- In general, if $x = \phi(y)$, the more rapid the increase of~$\phi$ the more precisely can we determine the increase of~$y$ as a function of~$x$. Thus if \[ x = ye^{y} \] we have $lx = y + ly$ and \[ y = lx - ly = lx(1 + u), \] where $u \sim ly/lx \sim llx/lx$. This is a solution of a much more precise kind than those considered above. \Paragraph{5.} The reader will find it instructive to examine the following results: \begin{Remark} \Item{(i)} Let \[ x = ye^{(ly)^{3/8}}. \] This is an example of the work of~§\;4: and \[ y \sim xe^{-(lx)^{3/8}}. \] \Item{(ii)} Let \[ x = ye^{(ly)^{5/8}}. \] Here \begin{align*} y &\sim xe [-(lx)^{5/8} \{1-(lx)^{-3/8}\}^{5/8}]\\ &\sim xe\{-(lx)^{5/8} + \tfrac{5}{8}(lx)^{1/4}\}. \end{align*} \Item{(iii)} Let \[ x = y^{m}(ly)^{m_{1}}(l_{2}y)^{m_{2}} \dots (l_{r}y)^{m_{r}}. \] Here \[ y \sim m^{m_{1}/m} x^{1/m} (lx)^{-m_{1}/m} \dots (l_{r}x)^{-m_{r}/m}. \] \Item{(iv)} Let \[ x = e^{y^{2}}ly. \] Here \[ y = \sqrt{lx - l_{3}x} + u \quad (u \clt 1). \] \Item{(v)} As an example of another type, Du~Bois-Reymond has considered the equation \[ f(x + y) - f(x) = C, \] where $C$~is a positive constant. He finds \begin{gather*} y \sim C/f'(x) \quad (f(x) \cgt lx),\\ y = xe\{Cu/xf'(x)\} \quad (u \sim 1,\ lx \cgt f(x) \cgt llx), \end{gather*} and so on: the forms of the solution when $f \ceq lx$, $f \ceq llx$,~\dots\ are exceptional. \Item{(vi)} As an example of an approximation pushed to greater lengths let us take the following result: if \[ x = y\, ly, \] then \[ y = \frac{x}{lx} \left\{1 + \frac{llx}{lx} + \frac{(llx)^{2}}{(lx)^{2}} - \frac{llx}{(lx)^{2}}\right\} + u, \] where \[ u \ceqq \frac{x(llx)^{3}}{(lx)^{4}}. \] \end{Remark} \Paragraph{6.} Here we may bring our account of the general theory to a close. It is a theory that has found, and is finding, a large and increasing variety of applications in various branches of mathematics: the nature of some of these applications the reader may glean from Appendix~II\@. %% -----File: 055.png---Folio 47------- \Appendix{I.}{General Bibliography.} \Author{Du~Bois-Reymond}'s memoirs bearing on the subjects of this tract are: \Work Sur la grandeur relative des infinis des fonctions (\textit{Annali di Matematica}, Serie~2, t.~4, p.~338). \Work Théorème général concernant la grandeur relative des infinis des fonctions et de leurs derivées (\textit{Crelle's Journal}, Bd.~74, S.~294). \Work Eine neue Theorie der Convergenz und Divergenz von Reihen mit positiven Gliedern. \textit{Anhang}: Ueber die Tragweite der logarithmischen Kriterien (\textit{Crelle's Journal}, Bd.~76, S.~61). \Work Ueber asymptotische Werthe, infinitäre Approximationen, und infinitäre Auflösung von Gleichungen (\textit{Math.\ Annalen}, Bd.~8, S.~363). Nachtrag zur vorstehenden Abhandlung (\textit{ibid.}, S.~574). \Work Notiz über infinitäre Gleichheiten (\textit{Math.\ Annalen}, Bd.~10, S.~576). \Work Ueber die Paradoxen des Infinitärcalcüls (\textit{Math.\ Annalen}, Bd.~11, S.~149). \Work Notiz über Convergenz von Integralen mit nicht verschwindendem Argument (\textit{Math.\ Annalen}, Bd.~13, S.~251). \Work Ueber Integration und Differentiation infinitären Relationen (\textit{Math.\ Annalen}, Bd.~14, S.~498). \Work Ueber den Satz: $\lim f'(x) = \lim f(x)/x$ (\textit{Math.\ Annalen}, Bd.~16, S.~550). See also \Author{A. Pringsheim}: Ueber die sogenannte Grenze und die Grenzgebiete zwischen Convergenz und Divergenz (\textit{Münchener Sitzungsberichte}, Bd.~26, S.~605). \Same Ueber die Du~Bois-Reymond'sche Convergenz-Grenze u.s.w. (\textit{Münchener Sitzungsberichte}, Bd.~27, S.~303). \Same Allgemeine Theorie der Convergenz und Divergenz von Reihen mit positiven Gliedern (\textit{Math.\ Annalen}, Bd.~35, S.~347). \Same Zur Theorie der bestimmten Integrale und der unendlichen Reihen (\textit{Math.\ Annalen}, Bd.~37, S.~591). \Author{J. Hadamard}: Sur les caractères de convergence des séries à termes positifs et sur les fonctions indéfiniment croissantes (\textit{Acta Mathematica}, t.~18, p.~319 and p.~421). %% -----File: 056.png---Folio 48------- \Author{S. Pincherle}: Alcune osservazioni sugli ordini d'infinito delle funzioni (\textit{Memorie della Accademia delle Scienze di Bologna}, Ser.~4, t.~5, p.~739). \Author{E. Borel}: \textit{Leçons sur les fonctions entières}, pp.~111--122. \Same \textit{Leçons sur les séries à termes positifs}, pp.~1--50. \Same \textit{Leçons sur la théorie de la croissance.} \Appendix[Applications.] {II.}{A Sketch of Some Applications,\protect\footnotemark\ With References.} \footnotetext{That is to say of certain regions of mathematical theory in which the notation and the ideas of the \textit{Infinitärcalcül} may be used systematically with a great gain in clearness and simplicity.} \Section[A.]{Convergence and divergence of series and integrals.} \Subsection{(i)}{The logarithmic tests.} The series $\sum u_{n}$ ($u_{n} \geq 0$) is convergent if \begin{flalign*} &&u_{n} &\cleq n^{-1-\alpha} && \\ \RTag{\emph{or}} &&u_{n} &\cleq (n\, ln \dots l_{k-1}n)^{-1}(l_{k}n)^{-1-\alpha}, && \end{flalign*} where $\alpha > 0$, and divergent if \begin{flalign*} &&u_{n} &\cleq n^{-1} && \\ \RTag{\emph{or}} &&u_{n} &\cgeq (n\, ln \dots l_{k}n)^{-1}(l_{k}n)^{-1}. && \end{flalign*} The integral $\ds\int^{\infty} f(x)\,dx$ ($f \geqq 0$) is convergent if \begin{flalign*} &&f &\cleq x^{-1-\alpha} && \\ \RTag{\emph{or}} &&f &\cleq (x\, lx \dots l_{k-1}x)^{-1}(l_{k}x)^{-1-\alpha}, && \end{flalign*} where $\alpha > 0$, and divergent if \begin{flalign*} &&f &\cleq x^{-1} && \\ \RTag{\emph{or}} &&f &\cleq (x\, lx \dots l_{k}x)^{-1}. && \end{flalign*} The integral $\ds\int_{0} f(x)\,dx$ ($f \geqq 0$) is convergent if \begin{flalign*} &&f &\cleq (1/x)^{1-\alpha} && \\ \qquad{\emph{or}} &&f &\cleq (1/x)\{l(1/x) \dots l_{k-1}(1/x)\}^{-1} \{l_{k}(1/x)\}^{-1-\alpha}, && \end{flalign*} where $\alpha > 0$, and divergent if \begin{flalign*} &&f &\cgeq 1/x && \\ \qquad{\emph{or}} &&f &\cgeq (1/x)\{l(1/x) \dots l_{k}(1/x)\}^{-1}. && \end{flalign*} %% -----File: 057.png---Folio 49------- [The first general statement of the `logarithmic criteria,' so far as series are concerned, appears to have been made by De~Morgan: see his \textit{Differential and Integral Calculus}, 1839, p.~326. The essentials of the matter, however, appear in a posthumous memoir of Abel (\textit{\OE uvres complètes}, t.~2, p.~200; see also t.~1, p.~399). This memoir appears also to have been first published in 1839. The case of $k = 1$ had been dealt with by Cauchy (\textit{Exercices de Mathématiques}, t.~2, 1827, pp.~221 \textit{et~seq.}). Bertrand appears to have arrived at some or all of De~Morgan's results independently (see \textit{Liouville's Journal}, t.~7, 1842, p.~37) and the criteria are very commonly attributed to him. The criteria for integrals do not appear to have been stated generally before Riemann, \textit{Inaugural-Dissertation} of 1854 (\textit{Werke}, S.~229). The following references may also be useful: Bonnet, \textit{Liouville's Journal}, t.~8, p.~78. Dini, \textit{Sulle serie a termini positivi} (Pisa, 1867); also in the \textit{Annali dell' Univ.\ Tosc.}, t.~9, p.~41. Du~Bois-Reymond, \textit{Crelle's Journal}, Bd.~76, S.~619. Pringsheim, \textit{Math.\ Annalen}, Bd.~35, S.~347 and Bd.~37, S.~591; also in the \textit{Encyklopädie der Math.\ Wiss.}, Bd.~1, Th.~1, S.~77 \textit{et~seq.} Hobson, \textit{Theory of functions of a real variable}, p.~406. Bromwich, \textit{Infinite series}, pp.~29,~37. Hardy, \textit{Course of pure mathematics}, pp.~357 \textit{et~seq.} Chrystal, \textit{Algebra}, vol.~2, pp.~109 \textit{et~seq.}] \Subsection{(ii)}{General theorems analogous to Du~Bois-Reymond's Theorem \(\Ref{ii.}{§\;1}\).} Given any divergent series $\sum u_{n}$ of positive terms, we can find a function~$v_{n}$ such that $v_{n} \clt u_{n}$ and $\sum v_{n}$ is divergent; \ie\ given any divergent series we can find one more slowly divergent. Given any convergent series $\sum u_{n}$ of positive terms we can find~$v_{n}$ so that $v_{n} \cgt u_{n}$ and $\sum v_{n}$ is convergent; \ie\ given any convergent series we can find one more slowly convergent. Given any function~$\phi(n)$ tending to infinity, however slowly, we can find a convergent series~$\sum u_{n}$ and a divergent series~$\sum v_{n}$ such that $v_{n}/u_{n} = \phi(n)$. Given an infinite sequence of series, each converging (diverging) more slowly than its predecessor, we can find a series which converges (diverges) more slowly than any of them. [See Abel and Dini, \lc~\textit{supra}; Hadamard, \textit{Acta Mathematica}, t.~18, p.~319 and t.~27, p.~177; Bromwich, \textit{Infinite series}, p.~40; Littlewood, \textit{Messenger of Mathematics}, vol.~39, p.~191.] %% -----File: 058.png---Folio 50------- There is no function~$\phi(n)$ such that $u_{n}\phi(n) \cgeq 1$ is a necessary condition for the divergence of $\sum u_{n}$, and no function~$\phi(n)$ such that $\phi(n) \cgt 1$ and $u_{n}\phi(n) \cleq 1$ is a necessary condition for the convergence of~$\sum u_{n}$. If $u_{n}$~is a \emph{steadily decreasing} function of~$n$, then $nu_{n} \clt 1$ \emph{is} a necessary condition for convergence; but there is no function~$\phi(n)$ such that $\phi(n) \cgt 1$ and $n\phi(n) u_{n} \clt 1$ is a necessary condition. [Pringsheim, \textit{Math.\ Annalen}, Bd.~35, S.~343 \textit{et~seq.; ibid.}, Bd.~37, S.~591 \textit{et~seq.}] If however $nu_{n}$~decreases steadily, then $n\log nu_{n} \to 0$ is a necessary condition; and if $n\psi(n)u_{n}$, where $n\psi(n) \cgt 1$ and $\ds\int \frac{dn}{n\psi(n)} \cgt 1$, decreases steadily, then \[ \left(n\psi(n) \int \frac{dn}{n\psi(n)}\right) u_{n} \to 0 \] is a necessary condition. \Subsection{(iii)}{Special series and integrals possessing peculiarities in respect to the mode of their convergence or divergence.} For examples of series and integrals which converge or diverge so slowly as not to answer to any of the logarithmic criteria see Du~Bois-Reymond, Pringsheim, Borel (\lc~\textit{supra}), and Blumenthal, \textit{Principes de la théorie des fonctions entières d'ordre infini}, ch.~1. In these cases the logarithmic tests are insufficient (\textit{en~défaut}, \Ref{iv.}{§§\;2,~5}). For examples of series and integrals to which the logarithmic tests are \emph{inapplicable} (\Ref{v.}{§§\;3,~4}) see the writings just mentioned and also Thomae: \textit{Zeitschrift für Mathematik}, Bd.~23, S.~68. Gilbert: \textit{Bulletin des Sciences Mathématiques}, t.~12, p.~66. Goursat: \textit{Cours d'Analyse}, t.~1, p.~205. Hardy: \textit{Messenger of Mathematics}, vol.~31, p.~1; \textit{ibid.},~vol.~31, p.~177; \textit{ibid.},~vol.~39, p.~28. \Section[B.]{Asymptotic formulae for finite series and integrals.} A closely connected problem is that of the determination of asymptotic formulae for \[ A_{n} = a_{1} + a_{2} + \dots + a_{n} \] or for \[ \Phi(x) = \int_{a}^{x} \phi(t)\,dt, \] {\Loosen when the behaviour of $a_{n}$ or~$\phi(x)$ for large values of $n$ or~$x$ is known. A good deal can be accomplished in this direction by means of %% -----File: 059.png---Folio 51------- (i)~the theorem of Cauchy and Stolz, that, if $a_{n}$~and~$b_{n}$ are positive and $a_{n} \sim Cb_{n}$, then $A_{n} \sim CB_{n}$, (ii)~the theorems of~\Ref{vi.}{}\ and (iii)~the theorem of Maclaurin and Cauchy, that} \[ \phi(1) + \phi(2) + \dots + \phi(n) - \int_{1}^{n} \phi(x)\,dx, \] where $\phi(x)$~is a positive and decreasing function of~$x$, tends to a limit as $n \to \infty$. [For~(i) see Cauchy, \textit{Analyse algébrique}, p.~52; Stolz, \textit{Math.\ Annalen}, Bd.~14, S.~232, or \textit{Allgemeine Arithmetik}, Bd.~1, S.~173; Jensen, \textit{Tidskrift for Mathematik}~(5), Bd.~2, S.~81; Bromwich, \textit{Infinite series}, p.~378, and \textit{Proc.\ Lond.\ Math.\ Soc.}, ser.~2, vol.~7, p.~101. Proofs of~(iii) will be found in almost any modern treatise on analysis: \eg, Bromwich, \textit{Infinite series}, p.~29; Hardy, \textit{Course of pure mathematics}, p.~305. An important extension to \emph{slowly oscillating} series has been given recently by Bromwich (\textit{Proc.\ Lond.\ Math.\ Soc.}, ser.~2, vol.~7, p.~327).] Among the most important results which follow from these theorems are \begin{gather*} 1^{s} + 2^{s} + \dots + n^{s} \sim \frac{n^{s+1}}{s + 1} \quad (s > -1), \\ 1^{s} + 2^{s} + \dots + n^{s} - \frac{n^{s+1}}{s + 1} \sim \zeta(-s) \quad (-1 < s < 0), \\ 1 + \frac{1}{2} + \dots + \frac{1}{n} - \log n \sim \gamma, \end{gather*} \begin{multline*} 1 + \frac{\alpha·\beta}{1·\gamma} + \frac{\alpha(\alpha + 1) \beta(\beta + 1)}{1·2·\gamma(\gamma + 1)} + \dots\ \text{to $n$~terms,} \\ \begin{aligned} &\sim \frac{\Gamma(\gamma)}{\Gamma(\alpha)\, \Gamma(\beta)}\, \frac{n^{\alpha+\beta-\gamma}}{\alpha + \beta - \gamma}\quad (\alpha + \beta > \gamma), \\ \LTag{\emph{or}} &\sim \frac{\Gamma(\alpha + \beta)}{\Gamma(\alpha)\, \Gamma(\beta)}\, \log n\quad (\alpha + \beta = \gamma). \end{aligned} \end{multline*} In connection with the last result see Bromwich, \textit{Proc.\ Lond.\ Math.\ Soc.}, ser.~2, vol.~7, p.~101; in the earlier formulae $\gamma$~is Euler's constant and $\zeta$~denotes the `Riemann $\zeta$-function.' The most important of all formulae of this kind is beyond question \[ \log 1 + \log 2 + \dots + \log n - (n + \tfrac{1}{2})\log n + n \sim \tfrac{1}{2} \log(2\pi), \] which, in the form \[ n! \sim n^{n+\frac{1}{2}} e^{-n} \sqrt{2\pi}, \] constitutes \emph{Stirling's Theorem}. The literature connected with Stirling's Theorem and its extensions to the Gamma-function of a non-integral %% -----File: 060.png---Folio 52------- or complex variable is far too extensive to be summarized here. See \textit{Encykl.\ der Math.\ Wiss.}, Bd.~II.~(2), S.~165 \textit{et~seq.}; Bromwich, \textit{Infinite series}, pp.~461 \textit{et~seq.} Another formula of the same kind is \[ 1^{1}2^{2}3^{3} \dots n^{n} \sim An^{\frac{1}{2}n^{2} + \frac{1}{2}n + \frac{1}{12}}\, e^{-\frac{1}{4}n^{2}}, \] where $A$~is a constant defined by the equation \[ \log A = \tfrac{1}{12}\log 2\pi + \tfrac{1}{12} \gamma + \frac{1}{2\pi^{2}} \sum_{1}^{\infty} \frac{\log \nu}{\nu^{2}}. \] The properties of this constant have been investigated by Kinkelin and Glaisher (Kinkelin, \textit{Crelle's Journal}, Bd.~57, S.~122: Glaisher, \textit{Messenger of Mathematics}, vol.~6, p.~71; vol.~7, p.~43; vol.~23, p.~145; vol.~24, p.~1; \textit{Quarterly Journal of Mathematics}, vol.~26, p.~1: see also Barnes, \textit{ibid.},~vol.~31, pp.~264 \textit{et~seq.}). All these results are intimately bound up with the theory of the general `Euler-Maclaurin Sum Formula' \[ \sum_{1}^{n} f(n) = \int^{n} f(x)\,dx + C + \tfrac{1}{2}f(n) + \frac{B_{1}}{2!} f'(n) - \frac{B_{2}}{4!} f'''(n) + \dots \] which also possesses an extensive literature (see Schlömilch, \textit{Theorie der Differenzen und Summen}; Boole, \textit{Finite differences}; Markoff, \textit{Differenzenrechnung}; Seliwanoff, \textit{Differenzenrechnung}; \textit{Encykl.\ der Math.\ Wiss.}, Bd.~I. S.~929 \textit{et~seq.}; Bromwich, \textit{Infinite series}, p.~238 and p.~324; Barnes, \textit{Proc.\ Lond.\ Math.\ Soc.}, ser.~2, vol.~3, pp.~253 \textit{et~seq.}; where many further references are given). A simple example of the use of the general formula is afforded by the relation \[ \sum_{1}^{n} \nu^{s} - \frac{n^{s+1}}{s + 1} - \tfrac{1}{2} n^{s} - \sum_{1} (-1)^{i-1} \left(\frac{s}{2i - 1}\right) \frac{B_{i}}{2i} n^{s-2i+1} \sim \zeta(-s). \] Here $s$~is positive and not integral, and the summation with respect to~$i$ is continued until we come to a negative power of~$n$. \Section[C.]{Formulae involving prime numbers only.} Asymptotic formulae involving functions defined arithmetically, and particularly functions defined by sums of functions of prime numbers only, play a most important part in the analytical theory of numbers. Of these the most important is the formula \[ \Pi(n) \sim \frac{n}{ln}, \] where $\Pi(n)$~denotes the number of prime numbers less than~$n$. %% -----File: 061.png---Folio 53------- Similarly it is known that \[ \sum lp \sim n, \qquad \sum \frac{lp}{p} \sim ln, \qquad \sum \frac{1}{p} \sim lln \] (the summation in each case applying to all primes less than~$n$) while $\sum\limits^{\infty} \dfrac{1}{p\, lp}$ is convergent. Many more accurate results have been established by recent writers, particularly Mertens, Hadamard, Von~Mangoldt, De~la~Vallée-Poussin, and Landau; and the theory has to a considerable extent been freed from Riemann's still unproved assumption that all the roots of his Zeta-function have their real part equal to~$\frac{1}{2}$. Thus it has been shown that \[ \Pi(n) = \int_{2}^{n} \frac{dx}{\log x} + O\left\{\frac{n}{(ln)^{\Delta}}\right\}, \] or, still more accurately, \[ \Pi(n) = \int_{2}^{n} \frac{dx}{\log x} + O\{ne^{-\alpha\sqrt{ln}}\}, \] where $\alpha$~is a positive constant; but it still remains to be settled whether (as there is some reason to suppose) the last term can be replaced by~$O(\sqrt{n})$ or even by \[ O\left(\frac{\sqrt{n}}{ln}\right). \] [It would carry us too far to give detailed references to the literature of this exceedingly difficult and fascinating subject. The reader should consult Landau's exhaustive \textit{Handbuch der Lehre von der Verteilung der Primzahlen} (Teubner, 1909).] \Section[D.]{The theory of integral functions.} \Subsection{1.}{}The series $\sum c_{n}x^{n}$ will converge for all values of~$x$ (real or complex), and so define an \emph{integral function}~$f(x)$, if and only if $\sqrt[n]{|c_{n}|} \to 0$, \ie\ if $|c_{n}| \clt e^{-\Delta n}$. \Subsection{2.}{The three indices of a function of finite order.} The three most important characters of an integral function~$f(x)$ are: \Item{(i)} $\gamma_{n} = |c_{n}|$, the modulus of the $n$th~coefficient; \Item{(ii)} $\alpha_{n} = |a_{n}|$, the modulus of the $n$th (in order of absolute magnitude) zero of~$f(x)$; \Item{(iii)} $M(r)$, the maximum of~$|f(x)|$ on the circle $|x| = r$. $M(r)$~is known to be an increasing function of~$r$, and in all cases $M(r) \cgt r^{\Delta}$. %% -----File: 062.png---Folio 54------- A function such that $M(r) \clt e^{r^{\Delta}}$ is called a \emph{function of finite order}. We shall confine our remarks to such functions. The principal problem of the theory of integral functions is the determination of the relations between the increases of $\alpha_{n}$,~$1/\gamma_{n}$, and~$M(r)$. Those which subsist between the two latter functions are the simplest: when $\alpha_{n}$~is taken into account the theory is complicated by the `Picard case of exception'---the case of functions which (like~$e^{x}$) have no zeroes, or whose zeroes are scattered abnormally widely over the plane. The nature of the results of the general theory may be gathered from a statement of a few of the simplest of them. If \[ n^{-\mu-\delta} \clt \sqrt[n]{\gamma_{n}} \clt n^{-\mu+\delta}, \] \ie\ if \[ l(1/\gamma_{n}) \sim \mu n\, ln, \] we call $\mu$ the \emph{$\mu$-index}. The index may be defined in \emph{all} cases without any assumption as to the existence of a limit for $\{l(1/\gamma_{n})/(n\, ln)\}$; we confine ourselves to the simplest case. If \[ n^{(1/\lambda)-\delta} \clt \alpha_{n} \clt n^{(1/\lambda)+\delta}, \] we call $\lambda$ the \emph{$\lambda$-index}; and if \[ e^{r^{\nu-\delta}} \clt M(r) \clt e^{r^{\nu+\delta}}, \] we call $\nu$ the \emph{$\nu$-index}: thus \[ l\alpha_{n} \sim (ln)/\lambda, \qquad llM(r) \sim \nu\, lr. \] Then $\mu = 1/\nu$: and \emph{in general} $\lambda = \nu$. Thus for the function \[ \frac{\sin(\sqrt{x})}{\sqrt{x}} = 1 - \frac{x}{3!} + \frac{x^{2}}{5!} - \dots \] we have $\lambda = \nu = \frac{1}{2}$ and $\mu = 2$, as the reader will easily verify (using Stirling's Theorem to determine~$\mu$). \Subsection{3.}{Special results.} More precise results than these have been obtained in many cases. Thus if \[ \{n(ln)^{-\alpha_{1}} \dots (l_{\nu}n)^{-\alpha_{\nu}+\delta}\}^{-1/\rho} \clt \sqrt[n]{\gamma_{n}} \clt \{n(ln)^{-\alpha_{1}} \dots (l_{\nu}n)^{-\alpha_{\nu}-\delta}\}^{-1/\rho}, \] then \[ e\{r^{\rho}(lr)^{\alpha_{1}} \dots (l_{\nu}r)^{\alpha_{\nu}-\delta}\} \clt M(r) \clt e\{r^{\rho}(lr)^{\alpha_{1}} \dots (l_{\nu}r)^{\alpha_{\nu}+\delta}\}, \] and conversely. %% -----File: 063.png---Folio 55------- As examples of still more accurate, but more special results, we may quote the following: \begin{align*} &\sum \frac{x^{n}}{n^{\alpha n}} \sim \sqrt{\frac{2\pi}{e\alpha}}\, x^{1/2\alpha} e^{(\alpha/e)x^{1/\alpha}},\\ &\sum \frac{x^{n}}{(n!)^{\alpha}} \sim \frac{1}{\sqrt{\alpha}}\, (2\pi)^{(1-\alpha)/2} x^{(1-\alpha)/2\alpha} e^{\alpha x^{1/\alpha}},\\ &\sum \frac{x^{n}}{\Gamma(\alpha n + 1)} \sim (1/\alpha) e^{x^{1/\alpha}},\\ &\sum e^{-n^{p}}x^{n} % [** TN: Braces (not parentheses) in sqrt in original] \sim \sqrt{\frac{2\pi}{p(p - 1)}} \left(\frac{\log x}{p}\right)^{\frac{2-p}{2p-2}} e^{(p-1)\left(\frac{\log x}{p}\right)^{p/(p-1)}}, \end{align*} where $\alpha > 0$ and in the last formula $1 < p < 2$, and throughout $x \to \infty$ by positive values. These results may of course be used to give an upper limit for the modulus of the particular function considered when $x$~is not necessarily real, and so for~$M(r)$. Thus in the first case \[ M(r) \cleq r^{1/2\alpha} e^{(\alpha/e) x^{1/\alpha}}. \] [The reader who wishes to become familiar with the theory of integral functions should begin by reading Borel's \textit{Leçons sur les fonctions entières}. Some additions will be found in the notes at the end of the same writer's \textit{Leçons sur les fonctions méromorphes}. He should then read two memoirs by E.~Lindelöf; a short one in the \textit{Bulletin des Sciences Mathématiques}, t.~27, p.~1, and a long one in the \textit{Acta Societatis Fennicae}, t.~31, p.~1. Some of the results of this last paper were proved independently by Boutroux (\textit{Acta Mathematica}, t.~28, pp.~97 \textit{et~seq.}); but M.~Boutroux's important memoir is largely occupied by a discussion of some of the most difficult points in the theory. Much of the theory has been developed in a very simple and elementary way by Pringsheim (\textit{Math.\ Annalen}, Bd.~58, S.~257); and the reader should certainly consult a short note by Le~Roy (\textit{Bulletin des Sciences Mathématiques}, t.~24, p.~245). But, after reading the works of Borel and Lindelöf mentioned above, he will be wise to turn to Vivanti's \textit{Teoria delle funzioni analitiche} (German translation by Gutzmer), which contains by far the fullest treatment of the subject yet published, and an exhaustive list of original memoirs.] %% -----File: 064.png---Folio 56------- \Section[E.]{Power series with a finite radius of convergence.} Suppose that $a_{1} + a_{2} + \dots$ is a divergent series: for simplicity suppose that $a_{n}$~is always positive and steadily increases or decreases as $n$~increases. Further suppose $e^{-\delta n} \clt a_{n} \clt e^{\delta n}$, so that $\sum a_{n}x^{n}$ is convergent if $0 \leqq x < 1$. Then a large number of interesting results have been established connecting the increase of~$a_{n}$, as $n \to \infty$, and that of $f(x) = \sum a_{n}x^{n}$ as $x \to 1$. The fundamental result is: \emph{if $a_{n} \sim Cb_{n}$, or, more generally, if $(a_{1} + a_{2} + \dots + a_{n}) \sim C(b_{1} + b_{2} + \dots + b_{n})$, and $f(x) = \sum a_{n}x^{n}$, $g(x) = \sum b_{n}x^{n}$, then} \[ f(x) \sim Cg(x). \] From this theorem it may be deduced that \begin{align*} \sum \frac{x^{n}}{n^{p}} &\sim \frac{\Gamma(1 - p)}{(1 - x)^{1-p}}\quad (p < 1), \\ F(\alpha, \beta, \gamma, x) &\sim \frac{\Gamma(\gamma)\, \Gamma(\alpha + \beta - \gamma)}{\Gamma(\alpha)\, \Gamma(\beta)}\, \frac{1}{(1 - x)^{\alpha+\beta-\gamma}}\quad (\alpha + \beta > \gamma) \\ F(\alpha, \beta, \alpha + \beta, x) &\sim \frac{\Gamma(\alpha + \beta)}{\Gamma(\alpha)\, \Gamma(\beta)}\, l\left(\frac{1}{1 - x}\right). \end{align*} Of further results the following is typical: if \[ a_{n} \sim n^{p}/n\, ln \dots l_{m-1}n (l_{m}n)^{q} \dots (l_{m+k}n)^{q_{k}}, \] then \begin{multline*} F(x) \sim \Gamma(p) \bigg/ \bigg\{(1 - x)^{p+1} \\ × \frac{1}{1 - x}\, l\frac{1}{1 - x} \dots l_{m-1} \frac{1}{1 - x} \left(l_{m} \frac{1}{1 - x}\right)^{q}\!\! \dots \left(l_{m+k} \frac{1}{1 - x}\right)^{q_{k}}\bigg\} \end{multline*} if $p > 0$, $q \neq 1$: but \[ F(x) \sim 1 \bigg/ \biggl\{(1 - q)\left(l_{m} \frac{1}{1 - x}\right)^{q-1}\!\! \biggl(l_{m+1} \frac{1}{1 - x}\biggr)^{q_{1}}\!\! \dots \biggl(l_{m+k} \frac{1}{1 - x}\biggr)^{q_{k}}\biggl\} \] if $p = 0$, $q < 1$ (if $p < 0$ or $p = 0$, $q > 1$, then $\sum a_{n}$ is convergent). Thus, \eg \[ \sum \frac{n^{p} x^{n}}{(lx)^{q}} \sim \Gamma(p + 1) \bigg/ \left\{(1 - x)^{p+1} \left(l \frac{1}{1 - x}\right)^{q}\right\}. \] %% -----File: 065.png---Folio 57------- As specimens of further results of this character we may quote \begin{gather*} x + x^{4} + x^{9} + \dots \sim \tfrac{1}{2} \sqrt{\frac{\pi}{1 - x}}, \\ x + x^{\alpha} + x^{\alpha^{2}} + \dots \sim \frac{1}{la}\, l\left(\frac{1}{1 - x}\right)\quad (a > 1), \\ \sum a^{n} x^{n^{2}} \sim e\left\{\tfrac{1}{4}\, \frac{(la)^{2}}{l(1/x)}\right\} \quad (a > 1), \\ \sum e^{n/ln} x^{n} = e_{2}\{u/(1 - x)\} \quad (u \sim 1). \end{gather*} Many similar results have been established about series other than power series: thus \begin{align*} \sum \frac{x^{n}}{n(1 + x^{n})} &\sim \tfrac{1}{2}\, l\left(\frac{1}{1 - x}\right), \\ \sum \frac{x^{n}}{1 - x^{n}} &\sim \frac{1}{1 - x}\, l\left(\frac{1}{1 - x}\right). \end{align*} As an example of a more precise result we may quote the formula \[ \sum \frac{x^{n}}{1 + x^{2n}} = \tfrac{1}{4} \left\{\frac{\pi}{l(1/x)} - 1\right\} + O\{(1 - x)^{\Delta}\}. \] [See Bromwich, \textit{Infinite series}, pp.~131 \textit{et~seq.}, 171~\textit{et~seq.}; Le~Roy, \textit{Bulletin des Sciences Mathématiques}, t.~24, pp.~245 \textit{et~seq.}; Lasker, \textit{Phil.\ Trans.\ Roy.\ Soc.},~(A), vol.~196, p.~433; Pringsheim, \textit{Acta Mathematica}, t.~28, p.~1; Barnes, \textit{Proc.\ Lond.\ Math.\ Soc.}, vol.~4, p.~284; \textit{Quarterly Journal}, vol.~37, p.~289; Hardy, \textit{Proc.\ Lond.\ Math.\ Soc.}, vol.~3, p.~381; \textit{ibid.},~vol.~5, p.~197; \textit{ibid.},~vol.~5, p.~342; \\ where further references will be found. These writers also consider the extensions of such results to the field of the complex variable.] %% -----File: 066.png---Folio 58------- \Appendix{III.}{Some Numerical Illustrations.} Mr J.~Jackson, scholar of Trinity College, has been kind enough to calculate for me the following numerical results, which will, I think, be found instructive as comments on some of the matters dealt with in the body of this tract and in Appendix~II\@. It will of course be understood that, except in one or two instances, they are approximations and sometimes quite rough approximations. \Section[1.]{Table of the functions $\log x$, $\log\log x$, $\log\log\log x$, etc.} { \[ \begin{array}{|l|r|r|r|r|r|} \hline \CCEntry{x} & \CEntry{\log x} & \CEntry{\log_{2} x} & \CEntry{\log_{3} x} & \CEntry{\log_{4} x} & \CEntry{\log_{5} x}\\ \hline \Strut 10 & 2.30 & 0.834 & -0.182 & \Dash & \Dash\\ 10^{3} & 6.91 & 1.933 & 0.659 & -0.417 & \Dash\\ 10^{6} & 13.82 & 2.626 & 0.966 & -0.035 & \Dash\\ 10^{10} & 23.03 & 3.137 & 1.143 & 0.134 & -2.011\\ 10^{15} & 34.54 & 3.542 & 1.265 & 0.235 & -1.449\\ 10^{20} & 46.05 & 3.830 & 1.343 & 0.295 & -1.221\\ 10^{30} & 69.08 & 4.235 & 1.443 & 0.367 & -1.003\\ 10^{60} & 138.15 & 4.928 & 1.595 & 0.467 & -0.762\\ 10^{100} & 230.26 & 5.439 & 1.693 & 0.527 & -0.641\\ 10^{1000} & 2302.58 & 7.742 & 2.047 & 0.716 & -0.334\\ 10^{10^{6}} & 2303 × 10^{3} & 14.650 & 2.685 & 0.987 & -0.013\\ 10^{10^{10}} & 2303 × 10^{7} & 23.860 & 3.172 & 1.154 & 0.144\\ \hline \end{array} \] } %% -----File: 067.png---Folio 59------- \Section[2.]{Table of the functions $e^{x}$, $e^{e^{x}}$, $e^{e^{e^{x}}}$, etc.} \[ \begin{array}{|c|c|c|c|c|c|} \hline \CCEntry{x} & \CEntry{ex} & \CEntry{e_{2}x} & \CEntry{e_{3}x} & \CEntry{e_{4}x} \\ \hline \Strut 1 & 2.718 & 15.154 & 3,814,260 & 10^{1,656,510}\\ 2 & 7.389 & 1618.2 & 5.85 × 10^{702} & \Dash\\ 3 & 20.085 & 5.28 × 10^{8} & 10^{2.295 × 10^{8}} & \Dash\\ 5 & 148.413 & 2.85 × 10^{64} & 10^{1.24 × 10^{64}} & \Dash\\ 10 & 22026 & 9.44 × 10^{9565} & \Dash & \Dash\\ \hline \end{array} \] The function $\log x$ is defined only for $x > 0$, $\log_{2}x$~for $x > 1$, $\log_{3}x$~for $x > e$, $\log_{4} x$~for $x > e^{e} = e_{2}$, and so on. The values of the first few numbers $e$,~$e_{2}$, $e_{3}$,~\dots\ are given above, viz.\ $e = 2.718$, $e_{2} = 15.154$, $e_{3} = 3,814,260$, $e_{4} = 10^{1,656,510}$. \Section[3.]{Table of the functions $n!$, $n^{n}$, $n^{n^{n}}$.} \[ \begin{array}{|c|c|c|c|} \hline \CCEntry{n} & \CEntry{n!} & \CEntry{n^{n}} & \CEntry{n^{n^{n}}} \\ \hline \Strut 1 & 1 & 1 & 1\\ 2 & 2 & 4 & 16\\ 3 & 6 & 27 & 7.634 × 10^{12}\\ 4 & 24 & 256 & 1.491 × 10^{154}\\ 5 & 120 & 3,125 & 9.55 × 10^{2,184}\\ 6 & 720 & 46,656 & 2.7 × 10^{36,305}\\ 7 & 5,040 & 823,543 & 1.4 × 10^{695,974}\\ 8 & 40,320 & \DPtypo{16,827,216}{16,777,216} & 10^{15,151,345}\\ 9 & 362,880 & 3.8742 × 10^{8} & 10^{369,693,100}\\ 10 & 3,628,800 & 10^{10} & 10^{10,000,000,000}\\ 100 & 9.346 × 10^{157} & 10^{200} & \Dash\\ 10^{10} & 10^{9.57 × 10^{10}} & 10^{10^{11}} & \Dash\\ \hline \end{array} \] %% -----File: 068.png---Folio 60------- \Section[4.]{Table to illustrate the convergence of the series\DPtypo{}{.}} {\small \begin{gather*} \begin{aligned} &(1)\ \sum_{3}^{\infty} \frac{1}{n\log n (\log\log n)^{2}}. &&(2)\ \sum_{2}^{\infty} \frac{1}{n(\log n)^{2}}. &&(3)\ \sum_{1}^{\infty} \frac{1}{n^{s}}\ (s = 1.1).\\ &(4)\ \sum_{1}^{\infty} \frac{1}{n^{s}}\ (s = 1.5). &&(5)\ \sum_{1}^{\infty} \frac{1}{n^{s}}\ (s = 2). &&(6)\ \sum_{1}^{\infty} \frac{1}{n^{s}}\ (s = 10).\\ &(7)\ \sum_{1}^{\infty} \frac{1}{n^{s}}\ (s = 100). &&(8)\ \sum_{0}^{\infty} x^{n}\ (x = .9). &&(9)\ \sum_{0}^{\infty} x^{n}\ (x = .5). \end{aligned} \\ \begin{aligned} &(10)\ \sum_{0}^{\infty} x^{n}\ (x = .1). &&(11)\ 1 + \frac{1}{2!} + \frac{1}{3!} + \dots. &&(12)\ 1 + \frac{1}{2^{2}} + \frac{1}{3^{3}} + \dots.\\ &(13)\ \sum_{0}^{\infty} x^{n^{2}}\ (x = .9). &&(14)\ \sum_{0}^{\infty} x^{n^{2}}\ (x = .5) &&(15)\ \sum_{0}^{\infty} x^{n^{2}}\ (x = .1). \end{aligned} \\ (16)\ \frac{1}{1^{1^{1}}} + \frac{1}{2^{2^{2}}} + \frac{1}{3^{3^{3}}} +\dots. \end{gather*} \footnotesize\settowidth{\TmpLen}{calculate the sum correctly to}% \[ %[** TN: Force centering of slightly over-wise table] \makebox[0pt][c]{$ \begin{array}{|c|c|c|c|c|c|} \hline &&\multicolumn{4}{|c|}{% \parbox{\TmpLen}{% \centering\footnotesize\medskip Number of terms required to\\ calculate the sum correctly to}} \\ \text{Series} & \text{Sum} & 2 & 10 & 100 & 1000 \\ &&&\multicolumn{2}{|c|}{\centering\text{\footnotesize decimal places\footnotemark}} & \\ \hline \Strut 1 & 38.43 & 10^{3.14 × 10^{86}} & \Dash & \Dash & \Dash\\ 2 & 2.11 & 7.23 × 10^{86} & 10^{8.6 × 10^{9}} & \Dash & \Dash\\ 3 & 10.58 & 10^{33} & 10^{113} & 10^{1013} & 10^{10013}\\ 4 & 2.612 & 160,000 & 16 × 10^{20} & 6 ×10^{200} & 16 ×10^{2000}\\ 5 & \frac{1}{6}\pi^{2} = 1.64493 & 200 & 2 × 10^{10} & 2 × 10^{100} & 2 × 10^{1000}\\ 6 & 1.0009846 & 1 & 11 & 1.093 × 10^{11} & 1.093 × 10^{111}\\ 7 & 1 + (1.27 × 10^{-30}) & 1 & 1 & 10 & 1.213 × 10^{10}\\ 8 & 10 & 73 & 247 & 2214 & 21883\\ 9 & 2 & 9 & 36 & 336 & 3325\\ 10 & 10/9 & 3 & 11 & 101 & 1001\\ 11 & e - 1 = 1.718282 & 5 & 13 & 70 & 440\\ 12 & 1.291286 & 3 & 10 & 57 & 386\\ 13 & 3.234989 & 8 & 15 & 46 & 148\\ 14 & 1.564468 & 3 & 6 & 19 & 58\\ 15 & 1.100100 & 2 & 4 & 11 & 32\\ 16 & 1.062500 & 2 & 2 & 3 & 4\\ \hline \end{array}$} \]} \footnotetext{The phrase `calculate the sum correctly to $m$~decimal places' is used as equivalent to `calculate with an error less than $\frac{1}{2} × 10^{-m}$.' In the case of a very slowly convergent series the interpretation affects the numbers to a considerable extent. The numbers would be considerably more difficult to calculate were the phrase interpreted in its literal sense.}% %% -----File: 069.png---Folio 61------- Such a series as~(7) is of course exceedingly rapidly convergent \emph{at first}, \ie\ a very few terms suffice to give the sum correctly to a considerable number of places; but if the sums are wanted to a very large number of places, even the series~(8) proves to be far more practicable. Mr William Shanks (\textit{Proc.\ Roy.\ Soc.}, vol.~21, p.~318) calculated the value of~$\pi$ to $707$~places of decimals from Machin's formula \[ \pi = 16\left(\frac{1}{5} - \frac{1}{3·5^{3}} + \frac{1}{5·5^{5}} - \dots\right) - 4 \left(\frac{1}{239} - \frac{1}{3·239^{3}} + \dots\right). \] He does not state the number of terms he found it necessary to use, but, in a previous calculation to $530$~places, used $747$~terms of the first and $219$~terms of the second series. He also (\textit{ibid.}, vol.~6, p.~397) calculated~$e$ to $205$~places from the series~(11). \Section[5.]{Table to illustrate the divergence of the series} {\small \begin{align*} &(1)\ \frac{1}{\log \log 3} + \frac{1}{\log \log 4} + \dots. &&(2)\ \frac{1}{\log 2} + \frac{1}{\log 3} + \dots.\\ &(3)\ 1 + \frac{1}{\sqrt{2}} + \frac{1}{\sqrt{3}} + \dots. &&(4)\ 1 + \frac{1}{2} + \frac{1}{3} + \dots.\\ &(5)\ \frac{1}{2\log 2} + \frac{1}{3\log 3} + \dots. &&(6)\ \frac{1}{3\log 3\log\log 3} + \frac{1}{4\log 4 \log\log 4} \DPtypo{}{{}+ \dots}. \end{align*} \[ \begin{array}{|c|c|c|c|c|c|c|} \hline &\multicolumn{6}{|c|}{% \text{\footnotesize Number of terms required to make the sum greater than}} \\ \text{Series} & 3 & 5 & 10 & 100 & 1000 & 10^{6}\\ \hline \Strut 1 & 1 & 1 & 1 & 116 & 1800 & 2.6 × 10^{6}\\ 2 & 3 & 7 & 20 & 440 & 7600 & 1.5 × 10^{7}\\ 3 & 5 & 10 & 33 & 2500 & 2.5 × 10^{5} & 2.5 × 10^{11}\\ 4 & 11 & 82 & 12390 & 10^{43} & 10^{.43 ×10^{3}} & 10^{.43 × 10^{6}}\\ 5 & 8690 & 1.3 × 10^{29} & 10^{4300} & 10^{5 × 10^{42}} & \Dash & \Dash\\ 6 & 1 & 60 \text{ \emph{to} } 70 & 10^{10^{100}} & \Dash & \Dash & \Dash\\ \hline \end{array} \]} \Section[6.]{Roots of certain equations.} \Item{(i)} The equation $e^{x} = x^{1,000,000}$ has a root just larger than unity (the excess of the root over unity being practically~$10^{-6}$) and a large root in the neighbourhood of~$16,610,800$. The equation $e^{x} = 1,000,000 x^{1,000,000}$ has roots nearly equal to those of the above. The one near unity is practically $12.82 x 10^{-6}$ less than unity, while the large root exceeds the root of the above equation by about~$13.82$. %% -----File: 070.png---Folio 62------- \Item{(ii)} The equation $e^{x^{2}} = x^{10^{10}}$ has a root somewhere near~$357,500$. \Item{(iii)} {\Loosen The equation $e^{e^{x}} = 10^{10} x^{10} e^{10^{10} x^{10}}$ has a root near~$64.7$. The root differs by less than~$10^{-26}$ from the corresponding root of $e^{x} = 10^{10} x^{10}$. The corresponding root of $e^{x} = x^{10}$ is about~$35.8$.} \Item{(iv)} The positive roots of $x^{x^{x}} = 1,000,000$ and of $x^{x^{x}} = 10^{1,000,000}$ are approximately $2.68$~and~$7.11$. \Item{(v)} If $x^{10} = 10^{y}$, then for $x = 100$, $y = 20$; and for $x = 10^{10}$, $y = 100$. If $x^{10^{10}} = 10^{10^{y}}$, then for $x = 100$, $y = 10.30$; for $x = 10^{10}$, $y = 11$; and for $x = 10^{10^{10}}$, $y = 20$. If $x^{10^{10^{10}}} = 10^{10^{10^{y}}}$, then for $x = 10^{10}$, $y = 10 + (4.3 × 10^{-11})$; for $x = 10^{10^{10}}$, $y = 10 + (4.3 × 10^{-10})$; and for $x = 10^{10^{10^{10}}}$, $y = 10.30$. \Section[7.]{Some numbers of physics.} The distance to $\alpha$~Centauri is roughly $26,000,000,000,000$ miles or $1.65 × 10^{18}$~inches. The number of inches lies between $19!$~and~$20!$ and is approximately equal to~$e^{e^{3.74}}$ or~$16^{e^{e}}$. Again, writing $15$~letters to the inch (an average size in print) a line to the star would be sufficient for the writing at length of~$10^{2.47 × 10^{19}}$. The latter number is approximately equal to $(14 × 10^{17})!$, $e^{e^{e^{3.83}}}$, or $(10^{6.5 × 10^{12}})^{e^{e^{e}}}$. If we take the distance to the end of the visible universe to be that through which light travels in $10,000$~years, we find that this distance when expressed in wave-lengths of sodium light is measured roughly by the numbers \[ 1.6 ×10^{26},\quad 26!,\quad e^{e^{4.10}},\quad (53.6)^{e^{e}},\quad 3.29^{3.29^{3.29}}. \] If we assume the average distance between the centres of two adjacent molecules of the earth's substance to be $10^{-8}$~cm., we find that the number of molecules in the earth is roughly \[ 10.8 × 10^{50},\quad 42!,\quad e^{e^{4.77}},\quad (2333)^{e^{e}},\quad 3.56^{3.56^{3.56}}. \] \vfill \hrule \Strut[8pt] {\scriptsize CAMBRIDGE: PRINTED BY JOHN CLAY, M.A. AT THE UNIVERSITY PRESS} %% -----File: 071.png---Folio 63------- %% -----File: 072.png---Folio 64------- %% -----File: 073.png---Folio 65------- %% -----File: 074.png---Folio 66------- \clearpage \thispagestyle{empty} \begin{center} \Titlefont{Cambridge Tracts in Mathematics\\ and Mathematical Physics}\\ \rule{1.5in}{1.0pt} \end{center} {\footnotesize \Catalog{No.~1.} VOLUME AND SURFACE INTEGRALS USED IN PHYSICS, by \textsc{J.~G. Leatham}, M.A. \textit{$2$s.\ $6$d.\ net.} \Catalog{No.~2.} THE INTEGRATION OF FUNCTIONS OF A SINGLE VARIABLE, by \textsc{G.~H. Hardy}, M.A., F.R.S. \textit{$2$s.\ $6$d.\ net.} \Catalog{No.~3.} QUADRATIC FORMS AND THEIR CLASSIFICATION BY MEANS OF INVARIANT FACTORS, by \textsc{T.~J.~I'A. Bromwich}, M.A., F.R.S. \textit{$3$s.\ $6$d.\ net.} \Catalog{No.~4.} THE AXIOMS OF PROJECTIVE GEOMETRY, by \textsc{A.~N. Whitehead}, Sc.D., F.R.S. \textit{$2$s.\ $6$d.\ net.} \Catalog{No.~5.} THE AXIOMS OF DESCRIPTIVE GEOMETRY, by \textsc{A.~N. Whitehead}, Sc.D., F.R.S. \textit{$2$s.\ $6$d.\ net.} \Catalog{No.~6.} ALGEBRAIC EQUATIONS, by \textsc{G.~B. Mathews}, M.A., F.R.S. \textit{$2$s.\ $6$d.\ net.} \Catalog{No.~7.} THE THEORY OF OPTICAL INSTRUMENTS, by \textsc{E.~T. Whittaker}, M.A., F.R.S. \textit{$2$s.\ $6$d.\ net.} \Catalog{No.~8.} THE ELEMENTARY THEORY OF THE SYMMETRICAL OPTICAL INSTRUMENT, by \textsc{J.~G. Leatham}, M.A. \textit{$2$s.\ $6$d.\ net.} \Catalog{No.~9.} INVARIANTS OF QUADRATIC DIFFERENTIAL FORMS, by \textsc{J.~E. Wright}, M.A. \textit{$2$s.\ $6$d.\ net.} \Catalog{No.~10.} AN INTRODUCTION TO THE STUDY OF INTEGRAL EQUATIONS, by \textsc{Maxime Bôcher}, B.A., Ph.D. \textit{$2$s.\ $6$d.\ net.} \Catalog{No.~11.} THE FUNDAMENTAL THEOREMS OF THE DIFFERENTIAL CALCULUS, by \textsc{W.~H. Young}, Sc.D., F.R.S. \textit{$2$s.\ $6$d.\ net.} \Catalog{No.~12.} ORDERS OF INFINITY: \textsc{The 'Infinitärcalcül' of Paul du~Bois-Reymond}, by \textsc{G.~H. Hardy}, M.A., F.R.S. \textit{$2$s.\ $6$d.\ net.} \subsubsection*{\normalfont\itshape In preparation.} \Inprep THE DEFINITE INTEGRAL, ITS MEANING AND FUNDAMENTAL PROPERTIES, by \textsc{E.~W. Hobson}, Sc.D., F.R.S. \Inprep SINGULAR POINTS AND ASYMPTOTES OF PLANE CURVES, by Miss \textsc{C.~A. Scott}. \Inprep THE CIRCLE AT INFINITY, by \textsc{J.~H. Grace}, M.A., F.R.S. \Inprep AN INTRODUCTION TO THE THEORY OF ATTRACTIONS, by \textsc{T.~J.~I'A. Bromwich}, M.A., F.R.S.\par } %%%%%%%%%%%%%%%%%%%%%%%%% GUTENBERG LICENSE %%%%%%%%%%%%%%%%%%%%%%%%%% \FlushRunningHeads \vfill \PGLicense \begin{PGtext} End of Project Gutenberg's Orders of Infinity, by Godfrey Harold Hardy *** END OF THIS PROJECT GUTENBERG EBOOK ORDERS OF INFINITY *** ***** This file should be named 38079-pdf.pdf or 38079-pdf.zip ***** This and all associated files of various formats will be found in: http://www.gutenberg.org/3/8/0/7/38079/ Produced by Andrew D. Hwang, Brenda Lewis and the Online Distributed Proofreading Team at http://www.pgdp.net (This file was produced from images generously made available by The Internet Archive/Canadian Libraries) Updated editions will replace the previous one--the old editions will be renamed. Creating the works from public domain print editions means that no one owns a United States copyright in these works, so the Foundation (and you!) can copy and distribute it in the United States without permission and without paying copyright royalties. Special rules, set forth in the General Terms of Use part of this license, apply to copying and distributing Project Gutenberg-tm electronic works to protect the PROJECT GUTENBERG-tm concept and trademark. Project Gutenberg is a registered trademark, and may not be used if you charge for the eBooks, unless you receive specific permission. If you do not charge anything for copies of this eBook, complying with the rules is very easy. You may use this eBook for nearly any purpose such as creation of derivative works, reports, performances and research. They may be modified and printed and given away--you may do practically ANYTHING with public domain eBooks. Redistribution is subject to the trademark license, especially commercial redistribution. *** START: FULL LICENSE *** THE FULL PROJECT GUTENBERG LICENSE PLEASE READ THIS BEFORE YOU DISTRIBUTE OR USE THIS WORK To protect the Project Gutenberg-tm mission of promoting the free distribution of electronic works, by using or distributing this work (or any other work associated in any way with the phrase "Project Gutenberg"), you agree to comply with all the terms of the Full Project Gutenberg-tm License (available with this file or online at http://gutenberg.org/license). Section 1. General Terms of Use and Redistributing Project Gutenberg-tm electronic works 1.A. By reading or using any part of this Project Gutenberg-tm electronic work, you indicate that you have read, understand, agree to and accept all the terms of this license and intellectual property (trademark/copyright) agreement. If you do not agree to abide by all the terms of this agreement, you must cease using and return or destroy all copies of Project Gutenberg-tm electronic works in your possession. If you paid a fee for obtaining a copy of or access to a Project Gutenberg-tm electronic work and you do not agree to be bound by the terms of this agreement, you may obtain a refund from the person or entity to whom you paid the fee as set forth in paragraph 1.E.8. 1.B. "Project Gutenberg" is a registered trademark. It may only be used on or associated in any way with an electronic work by people who agree to be bound by the terms of this agreement. There are a few things that you can do with most Project Gutenberg-tm electronic works even without complying with the full terms of this agreement. See paragraph 1.C below. There are a lot of things you can do with Project Gutenberg-tm electronic works if you follow the terms of this agreement and help preserve free future access to Project Gutenberg-tm electronic works. See paragraph 1.E below. 1.C. The Project Gutenberg Literary Archive Foundation ("the Foundation" or PGLAF), owns a compilation copyright in the collection of Project Gutenberg-tm electronic works. Nearly all the individual works in the collection are in the public domain in the United States. If an individual work is in the public domain in the United States and you are located in the United States, we do not claim a right to prevent you from copying, distributing, performing, displaying or creating derivative works based on the work as long as all references to Project Gutenberg are removed. Of course, we hope that you will support the Project Gutenberg-tm mission of promoting free access to electronic works by freely sharing Project Gutenberg-tm works in compliance with the terms of this agreement for keeping the Project Gutenberg-tm name associated with the work. You can easily comply with the terms of this agreement by keeping this work in the same format with its attached full Project Gutenberg-tm License when you share it without charge with others. 1.D. The copyright laws of the place where you are located also govern what you can do with this work. Copyright laws in most countries are in a constant state of change. If you are outside the United States, check the laws of your country in addition to the terms of this agreement before downloading, copying, displaying, performing, distributing or creating derivative works based on this work or any other Project Gutenberg-tm work. The Foundation makes no representations concerning the copyright status of any work in any country outside the United States. 1.E. Unless you have removed all references to Project Gutenberg: 1.E.1. The following sentence, with active links to, or other immediate access to, the full Project Gutenberg-tm License must appear prominently whenever any copy of a Project Gutenberg-tm work (any work on which the phrase "Project Gutenberg" appears, or with which the phrase "Project Gutenberg" is associated) is accessed, displayed, performed, viewed, copied or distributed: This eBook is for the use of anyone anywhere at no cost and with almost no restrictions whatsoever. You may copy it, give it away or re-use it under the terms of the Project Gutenberg License included with this eBook or online at www.gutenberg.org 1.E.2. If an individual Project Gutenberg-tm electronic work is derived from the public domain (does not contain a notice indicating that it is posted with permission of the copyright holder), the work can be copied and distributed to anyone in the United States without paying any fees or charges. If you are redistributing or providing access to a work with the phrase "Project Gutenberg" associated with or appearing on the work, you must comply either with the requirements of paragraphs 1.E.1 through 1.E.7 or obtain permission for the use of the work and the Project Gutenberg-tm trademark as set forth in paragraphs 1.E.8 or 1.E.9. 1.E.3. If an individual Project Gutenberg-tm electronic work is posted with the permission of the copyright holder, your use and distribution must comply with both paragraphs 1.E.1 through 1.E.7 and any additional terms imposed by the copyright holder. Additional terms will be linked to the Project Gutenberg-tm License for all works posted with the permission of the copyright holder found at the beginning of this work. 1.E.4. Do not unlink or detach or remove the full Project Gutenberg-tm License terms from this work, or any files containing a part of this work or any other work associated with Project Gutenberg-tm. 1.E.5. Do not copy, display, perform, distribute or redistribute this electronic work, or any part of this electronic work, without prominently displaying the sentence set forth in paragraph 1.E.1 with active links or immediate access to the full terms of the Project Gutenberg-tm License. 1.E.6. You may convert to and distribute this work in any binary, compressed, marked up, nonproprietary or proprietary form, including any word processing or hypertext form. However, if you provide access to or distribute copies of a Project Gutenberg-tm work in a format other than "Plain Vanilla ASCII" or other format used in the official version posted on the official Project Gutenberg-tm web site (www.gutenberg.org), you must, at no additional cost, fee or expense to the user, provide a copy, a means of exporting a copy, or a means of obtaining a copy upon request, of the work in its original "Plain Vanilla ASCII" or other form. Any alternate format must include the full Project Gutenberg-tm License as specified in paragraph 1.E.1. 1.E.7. Do not charge a fee for access to, viewing, displaying, performing, copying or distributing any Project Gutenberg-tm works unless you comply with paragraph 1.E.8 or 1.E.9. 1.E.8. You may charge a reasonable fee for copies of or providing access to or distributing Project Gutenberg-tm electronic works provided that - You pay a royalty fee of 20% of the gross profits you derive from the use of Project Gutenberg-tm works calculated using the method you already use to calculate your applicable taxes. The fee is owed to the owner of the Project Gutenberg-tm trademark, but he has agreed to donate royalties under this paragraph to the Project Gutenberg Literary Archive Foundation. Royalty payments must be paid within 60 days following each date on which you prepare (or are legally required to prepare) your periodic tax returns. Royalty payments should be clearly marked as such and sent to the Project Gutenberg Literary Archive Foundation at the address specified in Section 4, "Information about donations to the Project Gutenberg Literary Archive Foundation." - You provide a full refund of any money paid by a user who notifies you in writing (or by e-mail) within 30 days of receipt that s/he does not agree to the terms of the full Project Gutenberg-tm License. You must require such a user to return or destroy all copies of the works possessed in a physical medium and discontinue all use of and all access to other copies of Project Gutenberg-tm works. - You provide, in accordance with paragraph 1.F.3, a full refund of any money paid for a work or a replacement copy, if a defect in the electronic work is discovered and reported to you within 90 days of receipt of the work. - You comply with all other terms of this agreement for free distribution of Project Gutenberg-tm works. 1.E.9. If you wish to charge a fee or distribute a Project Gutenberg-tm electronic work or group of works on different terms than are set forth in this agreement, you must obtain permission in writing from both the Project Gutenberg Literary Archive Foundation and Michael Hart, the owner of the Project Gutenberg-tm trademark. Contact the Foundation as set forth in Section 3 below. 1.F. 1.F.1. Project Gutenberg volunteers and employees expend considerable effort to identify, do copyright research on, transcribe and proofread public domain works in creating the Project Gutenberg-tm collection. Despite these efforts, Project Gutenberg-tm electronic works, and the medium on which they may be stored, may contain "Defects," such as, but not limited to, incomplete, inaccurate or corrupt data, transcription errors, a copyright or other intellectual property infringement, a defective or damaged disk or other medium, a computer virus, or computer codes that damage or cannot be read by your equipment. 1.F.2. LIMITED WARRANTY, DISCLAIMER OF DAMAGES - Except for the "Right of Replacement or Refund" described in paragraph 1.F.3, the Project Gutenberg Literary Archive Foundation, the owner of the Project Gutenberg-tm trademark, and any other party distributing a Project Gutenberg-tm electronic work under this agreement, disclaim all liability to you for damages, costs and expenses, including legal fees. YOU AGREE THAT YOU HAVE NO REMEDIES FOR NEGLIGENCE, STRICT LIABILITY, BREACH OF WARRANTY OR BREACH OF CONTRACT EXCEPT THOSE PROVIDED IN PARAGRAPH 1.F.3. YOU AGREE THAT THE FOUNDATION, THE TRADEMARK OWNER, AND ANY DISTRIBUTOR UNDER THIS AGREEMENT WILL NOT BE LIABLE TO YOU FOR ACTUAL, DIRECT, INDIRECT, CONSEQUENTIAL, PUNITIVE OR INCIDENTAL DAMAGES EVEN IF YOU GIVE NOTICE OF THE POSSIBILITY OF SUCH DAMAGE. 1.F.3. LIMITED RIGHT OF REPLACEMENT OR REFUND - If you discover a defect in this electronic work within 90 days of receiving it, you can receive a refund of the money (if any) you paid for it by sending a written explanation to the person you received the work from. If you received the work on a physical medium, you must return the medium with your written explanation. The person or entity that provided you with the defective work may elect to provide a replacement copy in lieu of a refund. If you received the work electronically, the person or entity providing it to you may choose to give you a second opportunity to receive the work electronically in lieu of a refund. If the second copy is also defective, you may demand a refund in writing without further opportunities to fix the problem. 1.F.4. Except for the limited right of replacement or refund set forth in paragraph 1.F.3, this work is provided to you 'AS-IS' WITH NO OTHER WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO WARRANTIES OF MERCHANTIBILITY OR FITNESS FOR ANY PURPOSE. 1.F.5. Some states do not allow disclaimers of certain implied warranties or the exclusion or limitation of certain types of damages. If any disclaimer or limitation set forth in this agreement violates the law of the state applicable to this agreement, the agreement shall be interpreted to make the maximum disclaimer or limitation permitted by the applicable state law. The invalidity or unenforceability of any provision of this agreement shall not void the remaining provisions. 1.F.6. INDEMNITY - You agree to indemnify and hold the Foundation, the trademark owner, any agent or employee of the Foundation, anyone providing copies of Project Gutenberg-tm electronic works in accordance with this agreement, and any volunteers associated with the production, promotion and distribution of Project Gutenberg-tm electronic works, harmless from all liability, costs and expenses, including legal fees, that arise directly or indirectly from any of the following which you do or cause to occur: (a) distribution of this or any Project Gutenberg-tm work, (b) alteration, modification, or additions or deletions to any Project Gutenberg-tm work, and (c) any Defect you cause. Section 2. Information about the Mission of Project Gutenberg-tm Project Gutenberg-tm is synonymous with the free distribution of electronic works in formats readable by the widest variety of computers including obsolete, old, middle-aged and new computers. It exists because of the efforts of hundreds of volunteers and donations from people in all walks of life. Volunteers and financial support to provide volunteers with the assistance they need, are critical to reaching Project Gutenberg-tm's goals and ensuring that the Project Gutenberg-tm collection will remain freely available for generations to come. In 2001, the Project Gutenberg Literary Archive Foundation was created to provide a secure and permanent future for Project Gutenberg-tm and future generations. To learn more about the Project Gutenberg Literary Archive Foundation and how your efforts and donations can help, see Sections 3 and 4 and the Foundation web page at http://www.pglaf.org. Section 3. Information about the Project Gutenberg Literary Archive Foundation The Project Gutenberg Literary Archive Foundation is a non profit 501(c)(3) educational corporation organized under the laws of the state of Mississippi and granted tax exempt status by the Internal Revenue Service. The Foundation's EIN or federal tax identification number is 64-6221541. Its 501(c)(3) letter is posted at http://pglaf.org/fundraising. Contributions to the Project Gutenberg Literary Archive Foundation are tax deductible to the full extent permitted by U.S. federal laws and your state's laws. The Foundation's principal office is located at 4557 Melan Dr. S. Fairbanks, AK, 99712., but its volunteers and employees are scattered throughout numerous locations. Its business office is located at 809 North 1500 West, Salt Lake City, UT 84116, (801) 596-1887, email business@pglaf.org. Email contact links and up to date contact information can be found at the Foundation's web site and official page at http://pglaf.org For additional contact information: Dr. Gregory B. Newby Chief Executive and Director gbnewby@pglaf.org Section 4. Information about Donations to the Project Gutenberg Literary Archive Foundation Project Gutenberg-tm depends upon and cannot survive without wide spread public support and donations to carry out its mission of increasing the number of public domain and licensed works that can be freely distributed in machine readable form accessible by the widest array of equipment including outdated equipment. Many small donations ($1 to $5,000) are particularly important to maintaining tax exempt status with the IRS. The Foundation is committed to complying with the laws regulating charities and charitable donations in all 50 states of the United States. Compliance requirements are not uniform and it takes a considerable effort, much paperwork and many fees to meet and keep up with these requirements. We do not solicit donations in locations where we have not received written confirmation of compliance. To SEND DONATIONS or determine the status of compliance for any particular state visit http://pglaf.org While we cannot and do not solicit contributions from states where we have not met the solicitation requirements, we know of no prohibition against accepting unsolicited donations from donors in such states who approach us with offers to donate. International donations are gratefully accepted, but we cannot make any statements concerning tax treatment of donations received from outside the United States. U.S. laws alone swamp our small staff. Please check the Project Gutenberg Web pages for current donation methods and addresses. Donations are accepted in a number of other ways including checks, online payments and credit card donations. To donate, please visit: http://pglaf.org/donate Section 5. General Information About Project Gutenberg-tm electronic works. Professor Michael S. Hart is the originator of the Project Gutenberg-tm concept of a library of electronic works that could be freely shared with anyone. For thirty years, he produced and distributed Project Gutenberg-tm eBooks with only a loose network of volunteer support. Project Gutenberg-tm eBooks are often created from several printed editions, all of which are confirmed as Public Domain in the U.S. unless a copyright notice is included. Thus, we do not necessarily keep eBooks in compliance with any particular paper edition. Most people start at our Web site which has the main PG search facility: http://www.gutenberg.org This Web site includes information about Project Gutenberg-tm, including how to make donations to the Project Gutenberg Literary Archive Foundation, how to help produce our new eBooks, and how to subscribe to our email newsletter to hear about new eBooks. \end{PGtext} % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % % % % End of Project Gutenberg's Orders of Infinity, by Godfrey Harold Hardy % % % % *** END OF THIS PROJECT GUTENBERG EBOOK ORDERS OF INFINITY *** % % % % ***** This file should be named 38079-t.tex or 38079-t.zip ***** % % This and all associated files of various formats will be found in: % % http://www.gutenberg.org/3/8/0/7/38079/ % % % % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % \end{document} ### @ControlwordReplace = ( ['\\Contents', 'CONTENTS'], ['\\Preface', 'PREFACE'], ['\\continued', 'continued'], ['\\ie', 'i.e.'], ['\\eg', 'e.g.'], ['\\\(', '('], ['\\\)', ')'], ['\\end{Theorem}', ''], ['\\end{Result}', ''], ['\\begin{Remark}', ''], ['\\end{Remark}', ''], ); @ControlwordArguments = ( ['\\ToCChap', 1, 1, '', ' ', 1, 1, '', ' '], ['\\ToCApp', 1, 1, 'Appendix ', ' ', 1, 1, '', '', 1, 1, ' ... ', ''], ['\\Chapter', 0, 0, '', '', 1, 1, 'Chapter ', ' ', 1, 1, '', ''], ['\\Section', 0, 1, '', ' ', 1, 1, '', ''], ['\\Subsection', 1, 1, '', ' ', 1, 1, '', ''], ['\\Paragraph', 0, 1, '', '', 1, 1, '', ''], ['\\Appendix', 0, 0, '', '', 1, 1, 'Appendix ', ' ', 1, 1, '', ''], ['\\Item', 1, 1, '', ''], ['\\begin{Theorem}', 0, 1, 'Theorem ', ''], ['\\begin{Result}', 0, 1, '', ''], ['\\TranscribersNote', 1, 0, '', ''], ['\\Signature', 1, 1, '', ' ', 1, 1, '', ''], ['\\Graphic', 0, 0, '', '', 1, 0, '', '', 1, 0, '', ''], ['\\Figure', 0, 0, '', '', 1, 1, '
', 1, 0, '', ''], ['\\First', 1, 1, '', ''], ['\\Strut', 0, 0, '', ''], ['\\Eq', 1, 1, '', ''], ['\\Ref', 1, 1, '', ' ', 1, 1, '', ''], ['\\PageLabel', 0, 0, '', '', 1, 0, '', ''], ['\\PageRef', 1, 1, '', ' ', 1, 1, '', ''], ['\\Fig', 1, 1, 'Fig. ', ''], ['\\DPtypo', 1, 0, '', '', 1, 1, '', ''], ['\\First', 1, 1, '', ''] ); ### This is pdfTeX, Version 3.1415926-1.40.10 (TeX Live 2009/Debian) (format=pdflatex 2011.9.6) 25 NOV 2011 11:10 entering extended mode %&-line parsing enabled. **38079-t.tex (./38079-t.tex LaTeX2e <2009/09/24> Babel and hyphenation patterns for english, usenglishmax, dumylang, noh yphenation, farsi, arabic, croatian, bulgarian, ukrainian, russian, czech, slov ak, danish, dutch, finnish, french, basque, ngerman, german, german-x-2009-06-1 9, ngerman-x-2009-06-19, ibycus, monogreek, greek, ancientgreek, hungarian, san skrit, italian, latin, latvian, lithuanian, mongolian2a, mongolian, bokmal, nyn orsk, romanian, irish, coptic, serbian, turkish, welsh, esperanto, uppersorbian , estonian, indonesian, interlingua, icelandic, kurmanji, slovenian, polish, po rtuguese, spanish, galician, catalan, swedish, ukenglish, pinyin, loaded. (/usr/share/texmf-texlive/tex/latex/base/book.cls Document Class: book 2007/10/19 v1.4h Standard LaTeX document class (/usr/share/texmf-texlive/tex/latex/base/bk12.clo File: bk12.clo 2007/10/19 v1.4h Standard LaTeX file (size option) ) \c@part=\count79 \c@chapter=\count80 \c@section=\count81 \c@subsection=\count82 \c@subsubsection=\count83 \c@paragraph=\count84 \c@subparagraph=\count85 \c@figure=\count86 \c@table=\count87 \abovecaptionskip=\skip41 \belowcaptionskip=\skip42 \bibindent=\dimen102 ) (/usr/share/texmf-texlive/tex/latex/base/inputenc.sty Package: inputenc 2008/03/30 v1.1d Input encoding file \inpenc@prehook=\toks14 \inpenc@posthook=\toks15 (/usr/share/texmf-texlive/tex/latex/base/latin1.def File: latin1.def 2008/03/30 v1.1d Input encoding file )) (/usr/share/texmf-texlive/tex/latex/base/ifthen.sty Package: ifthen 2001/05/26 v1.1c Standard LaTeX ifthen package (DPC) ) (/usr/share/texmf-texlive/tex/latex/amsmath/amsmath.sty Package: amsmath 2000/07/18 v2.13 AMS math features \@mathmargin=\skip43 For additional information on amsmath, use the `?' option. (/usr/share/texmf-texlive/tex/latex/amsmath/amstext.sty Package: amstext 2000/06/29 v2.01 (/usr/share/texmf-texlive/tex/latex/amsmath/amsgen.sty File: amsgen.sty 1999/11/30 v2.0 \@emptytoks=\toks16 \ex@=\dimen103 )) (/usr/share/texmf-texlive/tex/latex/amsmath/amsbsy.sty Package: amsbsy 1999/11/29 v1.2d \pmbraise@=\dimen104 ) (/usr/share/texmf-texlive/tex/latex/amsmath/amsopn.sty Package: amsopn 1999/12/14 v2.01 operator names ) \inf@bad=\count88 LaTeX Info: Redefining \frac on input line 211. \uproot@=\count89 \leftroot@=\count90 LaTeX Info: Redefining \overline on input line 307. \classnum@=\count91 \DOTSCASE@=\count92 LaTeX Info: Redefining \ldots on input line 379. LaTeX Info: Redefining \dots on input line 382. LaTeX Info: Redefining \cdots on input line 467. \Mathstrutbox@=\box26 \strutbox@=\box27 \big@size=\dimen105 LaTeX Font Info: Redeclaring font encoding OML on input line 567. LaTeX Font Info: Redeclaring font encoding OMS on input line 568. \macc@depth=\count93 \c@MaxMatrixCols=\count94 \dotsspace@=\muskip10 \c@parentequation=\count95 \dspbrk@lvl=\count96 \tag@help=\toks17 \row@=\count97 \column@=\count98 \maxfields@=\count99 \andhelp@=\toks18 \eqnshift@=\dimen106 \alignsep@=\dimen107 \tagshift@=\dimen108 \tagwidth@=\dimen109 \totwidth@=\dimen110 \lineht@=\dimen111 \@envbody=\toks19 \multlinegap=\skip44 \multlinetaggap=\skip45 \mathdisplay@stack=\toks20 LaTeX Info: Redefining \[ on input line 2666. LaTeX Info: Redefining \] on input line 2667. ) (/usr/share/texmf-texlive/tex/latex/amsfonts/amssymb.sty Package: amssymb 2009/06/22 v3.00 (/usr/share/texmf-texlive/tex/latex/amsfonts/amsfonts.sty Package: amsfonts 2009/06/22 v3.00 Basic AMSFonts support \symAMSa=\mathgroup4 \symAMSb=\mathgroup5 LaTeX Font Info: Overwriting math alphabet `\mathfrak' in version `bold' (Font) U/euf/m/n --> U/euf/b/n on input line 96. )) (/usr/share/texmf-texlive/tex/latex/base/alltt.sty Package: alltt 1997/06/16 v2.0g defines alltt environment ) (/usr/share/texmf-texlive/tex/latex/footmisc/footmisc.sty Package: footmisc 2009/09/15 v5.5a a miscellany of footnote facilities \FN@temptoken=\toks21 \footnotemargin=\dimen112 \c@pp@next@reset=\count100 \c@@fnserial=\count101 Package footmisc Info: Declaring symbol style bringhurst on input line 855. Package footmisc Info: Declaring symbol style chicago on input line 863. Package footmisc Info: Declaring symbol style wiley on input line 872. Package footmisc Info: Declaring symbol style lamport-robust on input line 883. Package footmisc Info: Declaring symbol style lamport* on input line 903. Package footmisc Info: Declaring symbol style lamport*-robust on input line 924 . ) (/usr/share/texmf-texlive/tex/latex/tools/indentfirst.sty Package: indentfirst 1995/11/23 v1.03 Indent first paragraph (DPC) ) (/usr/share/texmf-texlive/tex/latex/was/icomma.sty Package: icomma 2002/03/10 v2.0 (WaS) ) (/usr/share/texmf-texlive/tex/latex/tools/calc.sty Package: calc 2007/08/22 v4.3 Infix arithmetic (KKT,FJ) \calc@Acount=\count102 \calc@Bcount=\count103 \calc@Adimen=\dimen113 \calc@Bdimen=\dimen114 \calc@Askip=\skip46 \calc@Bskip=\skip47 LaTeX Info: Redefining \setlength on input line 76. LaTeX Info: Redefining \addtolength on input line 77. \calc@Ccount=\count104 \calc@Cskip=\skip48 ) (/usr/share/texmf-texlive/tex/latex/yfonts/yfonts.sty Package: yfonts 2003/01/08 v1.3 (WaS) ) (/usr/share/texmf-texlive/tex/latex/fancyhdr/fancyhdr.sty \fancy@headwidth=\skip49 \f@ncyO@elh=\skip50 \f@ncyO@erh=\skip51 \f@ncyO@olh=\skip52 \f@ncyO@orh=\skip53 \f@ncyO@elf=\skip54 \f@ncyO@erf=\skip55 \f@ncyO@olf=\skip56 \f@ncyO@orf=\skip57 ) (/usr/share/texmf-texlive/tex/latex/graphics/graphicx.sty Package: graphicx 1999/02/16 v1.0f Enhanced LaTeX Graphics (DPC,SPQR) (/usr/share/texmf-texlive/tex/latex/graphics/keyval.sty Package: keyval 1999/03/16 v1.13 key=value parser (DPC) \KV@toks@=\toks22 ) (/usr/share/texmf-texlive/tex/latex/graphics/graphics.sty Package: graphics 2009/02/05 v1.0o Standard LaTeX Graphics (DPC,SPQR) (/usr/share/texmf-texlive/tex/latex/graphics/trig.sty Package: trig 1999/03/16 v1.09 sin cos tan (DPC) ) (/etc/texmf/tex/latex/config/graphics.cfg File: graphics.cfg 2009/08/28 v1.8 graphics configuration of TeX Live ) Package graphics Info: Driver file: pdftex.def on input line 91. (/usr/share/texmf-texlive/tex/latex/pdftex-def/pdftex.def File: pdftex.def 2009/08/25 v0.04m Graphics/color for pdfTeX \Gread@gobject=\count105 )) \Gin@req@height=\dimen115 \Gin@req@width=\dimen116 ) (/usr/share/texmf-texlive/tex/latex/caption/caption.sty Package: caption 2009/10/09 v3.1k Customizing captions (AR) (/usr/share/texmf-texlive/tex/latex/caption/caption3.sty Package: caption3 2009/10/09 v3.1k caption3 kernel (AR) \captionmargin=\dimen117 \captionmargin@=\dimen118 \captionwidth=\dimen119 \caption@indent=\dimen120 \caption@parindent=\dimen121 \caption@hangindent=\dimen122 ) \c@ContinuedFloat=\count106 ) (/usr/share/texmf-texlive/tex/latex/geometry/geometry.sty Package: geometry 2008/12/21 v4.2 Page Geometry (/usr/share/texmf-texlive/tex/generic/oberdiek/ifpdf.sty Package: ifpdf 2009/04/10 v2.0 Provides the ifpdf switch (HO) Package ifpdf Info: pdfTeX in pdf mode detected. ) (/usr/share/texmf-texlive/tex/generic/oberdiek/ifvtex.sty Package: ifvtex 2008/11/04 v1.4 Switches for detecting VTeX and its modes (HO) Package ifvtex Info: VTeX not detected. ) \Gm@cnth=\count107 \Gm@cntv=\count108 \c@Gm@tempcnt=\count109 \Gm@bindingoffset=\dimen123 \Gm@wd@mp=\dimen124 \Gm@odd@mp=\dimen125 \Gm@even@mp=\dimen126 \Gm@dimlist=\toks23 (/usr/share/texmf-texlive/tex/xelatex/xetexconfig/geometry.cfg)) (/usr/share/te xmf-texlive/tex/latex/hyperref/hyperref.sty Package: hyperref 2009/10/09 v6.79a Hypertext links for LaTeX (/usr/share/texmf-texlive/tex/generic/ifxetex/ifxetex.sty Package: ifxetex 2009/01/23 v0.5 Provides ifxetex conditional ) (/usr/share/texmf-texlive/tex/latex/oberdiek/hycolor.sty Package: hycolor 2009/10/02 v1.5 Code for color options of hyperref/bookmark (H O) (/usr/share/texmf-texlive/tex/latex/oberdiek/xcolor-patch.sty Package: xcolor-patch 2009/10/02 xcolor patch )) \@linkdim=\dimen127 \Hy@linkcounter=\count110 \Hy@pagecounter=\count111 (/usr/share/texmf-texlive/tex/latex/hyperref/pd1enc.def File: pd1enc.def 2009/10/09 v6.79a Hyperref: PDFDocEncoding definition (HO) ) (/usr/share/texmf-texlive/tex/generic/oberdiek/etexcmds.sty Package: etexcmds 2007/12/12 v1.2 Prefix for e-TeX command names (HO) (/usr/share/texmf-texlive/tex/generic/oberdiek/infwarerr.sty Package: infwarerr 2007/09/09 v1.2 Providing info/warning/message (HO) ) Package etexcmds Info: Could not find \expanded. (etexcmds) That can mean that you are not using pdfTeX 1.50 or (etexcmds) that some package has redefined \expanded. (etexcmds) In the latter case, load this package earlier. ) (/etc/texmf/tex/latex/config/hyperref.cfg File: hyperref.cfg 2002/06/06 v1.2 hyperref configuration of TeXLive ) (/usr/share/texmf-texlive/tex/latex/oberdiek/kvoptions.sty Package: kvoptions 2009/08/13 v3.4 Keyval support for LaTeX options (HO) (/usr/share/texmf-texlive/tex/generic/oberdiek/kvsetkeys.sty Package: kvsetkeys 2009/07/30 v1.5 Key value parser with default handler suppor t (HO) )) Package hyperref Info: Option `hyperfootnotes' set `false' on input line 2864. Package hyperref Info: Option `bookmarks' set `true' on input line 2864. Package hyperref Info: Option `linktocpage' set `false' on input line 2864. Package hyperref Info: Option `pdfdisplaydoctitle' set `true' on input line 286 4. Package hyperref Info: Option `pdfpagelabels' set `true' on input line 2864. Package hyperref Info: Option `bookmarksopen' set `true' on input line 2864. Package hyperref Info: Option `colorlinks' set `true' on input line 2864. Package hyperref Info: Hyper figures OFF on input line 2975. Package hyperref Info: Link nesting OFF on input line 2980. Package hyperref Info: Hyper index ON on input line 2983. Package hyperref Info: Plain pages OFF on input line 2990. Package hyperref Info: Backreferencing OFF on input line 2995. Implicit mode ON; LaTeX internals redefined Package hyperref Info: Bookmarks ON on input line 3191. (/usr/share/texmf-texlive/tex/latex/ltxmisc/url.sty \Urlmuskip=\muskip11 Package: url 2006/04/12 ver 3.3 Verb mode for urls, etc. ) LaTeX Info: Redefining \url on input line 3428. (/usr/share/texmf-texlive/tex/generic/oberdiek/bitset.sty Package: bitset 2007/09/28 v1.0 Data type bit set (HO) (/usr/share/texmf-texlive/tex/generic/oberdiek/intcalc.sty Package: intcalc 2007/09/27 v1.1 Expandable integer calculations (HO) ) (/usr/share/texmf-texlive/tex/generic/oberdiek/bigintcalc.sty Package: bigintcalc 2007/11/11 v1.1 Expandable big integer calculations (HO) (/usr/share/texmf-texlive/tex/generic/oberdiek/pdftexcmds.sty Package: pdftexcmds 2009/09/23 v0.6 LuaTeX support for pdfTeX utility functions (HO) (/usr/share/texmf-texlive/tex/generic/oberdiek/ifluatex.sty Package: ifluatex 2009/04/17 v1.2 Provides the ifluatex switch (HO) Package ifluatex Info: LuaTeX not detected. ) (/usr/share/texmf-texlive/tex/generic/oberdiek/ltxcmds.sty Package: ltxcmds 2009/08/05 v1.0 Some LaTeX kernel commands for general use (HO ) ) Package pdftexcmds Info: LuaTeX not detected. Package pdftexcmds Info: \pdf@primitive is available. Package pdftexcmds Info: \pdf@ifprimitive is available. ))) \Fld@menulength=\count112 \Field@Width=\dimen128 \Fld@charsize=\dimen129 \Field@toks=\toks24 Package hyperref Info: Hyper figures OFF on input line 4377. Package hyperref Info: Link nesting OFF on input line 4382. Package hyperref Info: Hyper index ON on input line 4385. Package hyperref Info: backreferencing OFF on input line 4392. Package hyperref Info: Link coloring ON on input line 4395. Package hyperref Info: Link coloring with OCG OFF on input line 4402. Package hyperref Info: PDF/A mode OFF on input line 4407. (/usr/share/texmf-texlive/tex/generic/oberdiek/atbegshi.sty Package: atbegshi 2008/07/31 v1.9 At begin shipout hook (HO) ) \Hy@abspage=\count113 \c@Item=\count114 ) *hyperref using driver hpdftex* (/usr/share/texmf-texlive/tex/latex/hyperref/hpdftex.def File: hpdftex.def 2009/10/09 v6.79a Hyperref driver for pdfTeX \Fld@listcount=\count115 ) \TmpLen=\skip58 \ceqqsymb=\box28 LaTeX Font Info: Try loading font information for U+msa on input line 547. (/usr/share/texmf-texlive/tex/latex/amsfonts/umsa.fd File: umsa.fd 2009/06/22 v3.00 AMS symbols A ) LaTeX Font Info: Try loading font information for U+msb on input line 547. (/usr/share/texmf-texlive/tex/latex/amsfonts/umsb.fd File: umsb.fd 2009/06/22 v3.00 AMS symbols B ) (./38079-t.aux) \openout1 = `38079-t.aux'. LaTeX Font Info: Checking defaults for OML/cmm/m/it on input line 552. LaTeX Font Info: ... okay on input line 552. LaTeX Font Info: Checking defaults for T1/cmr/m/n on input line 552. LaTeX Font Info: ... okay on input line 552. LaTeX Font Info: Checking defaults for OT1/cmr/m/n on input line 552. LaTeX Font Info: ... okay on input line 552. LaTeX Font Info: Checking defaults for OMS/cmsy/m/n on input line 552. LaTeX Font Info: ... okay on input line 552. LaTeX Font Info: Checking defaults for OMX/cmex/m/n on input line 552. LaTeX Font Info: ... okay on input line 552. LaTeX Font Info: Checking defaults for U/cmr/m/n on input line 552. LaTeX Font Info: ... okay on input line 552. LaTeX Font Info: Checking defaults for LY/yfrak/m/n on input line 552. LaTeX Font Info: ... okay on input line 552. LaTeX Font Info: Checking defaults for LYG/ygoth/m/n on input line 552. LaTeX Font Info: ... okay on input line 552. LaTeX Font Info: Checking defaults for PD1/pdf/m/n on input line 552. LaTeX Font Info: ... okay on input line 552. (/usr/share/texmf/tex/context/base/supp-pdf.mkii [Loading MPS to PDF converter (version 2006.09.02).] \scratchcounter=\count116 \scratchdimen=\dimen130 \scratchbox=\box29 \nofMPsegments=\count117 \nofMParguments=\count118 \everyMPshowfont=\toks25 \MPscratchCnt=\count119 \MPscratchDim=\dimen131 \MPnumerator=\count120 \everyMPtoPDFconversion=\toks26 ) Package caption Info: Begin \AtBeginDocument code. Package caption Info: hyperref package is loaded. Package caption Info: End \AtBeginDocument code. *geometry auto-detecting driver* *geometry detected driver: pdftex* -------------------- Geometry parameters paper: class default landscape: -- twocolumn: -- twoside: true asymmetric: -- h-parts: 9.03375pt, 361.34999pt, 9.03375pt v-parts: 0.54495pt, 567.7638pt, 0.81749pt hmarginratio: 1:1 vmarginratio: 2:3 lines: -- heightrounded: -- bindingoffset: 0.0pt truedimen: -- includehead: true includefoot: true includemp: -- driver: pdftex -------------------- Page layout dimensions and switches \paperwidth 379.4175pt \paperheight 569.12624pt \textwidth 361.34999pt \textheight 505.89pt \oddsidemargin -63.23624pt \evensidemargin -63.23624pt \topmargin -71.72504pt \headheight 15.0pt \headsep 19.8738pt \footskip 30.0pt \marginparwidth 98.0pt \marginparsep 7.0pt \columnsep 10.0pt \skip\footins 10.8pt plus 4.0pt minus 2.0pt \hoffset 0.0pt \voffset 0.0pt \mag 1000 \@twosidetrue \@mparswitchtrue (1in=72.27pt, 1cm=28.45pt) ----------------------- (/usr/share/texmf-texlive/tex/latex/graphics/color.sty Package: color 2005/11/14 v1.0j Standard LaTeX Color (DPC) (/etc/texmf/tex/latex/config/color.cfg File: color.cfg 2007/01/18 v1.5 color configuration of teTeX/TeXLive ) Package color Info: Driver file: pdftex.def on input line 130. ) Package hyperref Info: Link coloring ON on input line 552. (/usr/share/texmf-texlive/tex/latex/hyperref/nameref.sty Package: nameref 2007/05/29 v2.31 Cross-referencing by name of section (/usr/share/texmf-texlive/tex/latex/oberdiek/refcount.sty Package: refcount 2008/08/11 v3.1 Data extraction from references (HO) ) \c@section@level=\count121 ) LaTeX Info: Redefining \ref on input line 552. LaTeX Info: Redefining \pageref on input line 552. (./38079-t.out) (./38079-t.out) \@outlinefile=\write3 \openout3 = `38079-t.out'. \AtBeginShipoutBox=\box30 [1 {/var/lib/texmf/fonts/map/pdftex/updmap/pdftex.map}] [2 ] [1 ] <./images/cups.png, id=91, 72.27pt x 69.1383pt> File: ./images/cups.png Graphic file (type png) [2 <./images/cups.png (PNG copy)>] [3 ] [4 ] LaTeX Font Info: Try loading font information for OMS+cmr on input line 703. (/usr/share/texmf-texlive/tex/latex/base/omscmr.fd File: omscmr.fd 1999/05/25 v2.5h Standard LaTeX font definitions ) LaTeX Font Info: Font shape `OMS/cmr/m/n' in size <12> not available (Font) Font shape `OMS/cmsy/m/n' tried instead on input line 703. [5 ] [6 ] LaTeX Font Info: Font shape `OMS/cmr/m/n' in size <9> not available (Font) Font shape `OMS/cmsy/m/n' tried instead on input line 752. LaTeX Font Info: Font shape `OMS/cmr/m/n' in size <7> not available (Font) Font shape `OMS/cmsy/m/n' tried instead on input line 753. [1 ] [2] <./images/fig1.pdf, id=160, 165.61874pt x 278.03876pt> File: ./images/fig1.pdf Graphic file (type pdf) [3 <./images/fig1.pdf>] LaTeX Font Info: Font shape `OMS/cmr/m/n' in size <10.95> not available (Font) Font shape `OMS/cmsy/m/n' tried instead on input line 862. [4] [5] [6] [7] [8] LaTeX Font Info: Font shape `OMS/cmr/m/n' in size <10> not available (Font) Font shape `OMS/cmsy/m/n' tried instead on input line 1123. [9] [10] <./images/fig2.pdf, id=221, 211.79124pt x 329.23pt> File: ./images/fig2.pdf Graphic file (type pdf) [11] [12 <./images/fig2.pdf>] LaTeX Font Info: Font shape `OMS/cmr/m/n' in size <8> not available (Font) Font shape `OMS/cmsy/m/n' tried instead on input line 1258. [13] [14] <./images/fig3.pdf, id=261, 143.53625pt x 195.73125pt> File: ./images/fig3.pdf Graphic file (type pdf) [15 <./images/fig3.pdf>] Underfull \hbox (badness 1082) in paragraph at lines 1370--1374 []\OT1/cmr/bx/n/12 6. \OT1/cmr/m/n/12 The fol-low-ing ex-ten-sions of Du Bois- Reymond's The-o-rem [] <./images/fig4.pdf, id=272, 202.7575pt x 223.83624pt> File: ./images/fig4.pdf Graphic file (type pdf) [16] [17 <./images/fig4.pdf>] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] Overfull \hbox (0.10402pt too wide) in paragraph at lines 2121--2122 []\OT1/cmr/bx/n/12 3. Suc-ces-sive ap-prox-i-ma-tions to a logarithmico-expone ntial [] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] <./i mages/fig5.pdf, id=459, 220.825pt x 242.9075pt> File: ./images/fig5.pdf Graphic file (type pdf) [46] [47 <./images/fig5.pdf>] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] Underfull \hbox (badness 1668) in paragraph at lines 3604--3606 []| \OT1/cmr/m/n/12 Ue-ber die Para-doxen des In-finit^^?arcalc^^?uls (\OT1/cmr /m/it/12 Math. An-nalen\OT1/cmr/m/n/12 , [] Underfull \hbox (badness 1881) in paragraph at lines 3622--3624 [] \OT1/cmr/m/n/12 Ue-ber die Du Bois-Reymond'sche Convergenz-Grenze u.s.w. [] [64 ] [65] [66 ] [67] [68] [69] [70] [71] [72] [73] Overfull \hbox (1.83644pt too wide) detected at line 4026 \OMS/cmsy/m/n/12 f\OML/cmm/m/it/12 n\OT1/cmr/m/n/12 (\OML/cmm/m/it/12 ln\OT1/cm r/m/n/12 )[] [] (\OML/cmm/m/it/12 l[]n\OT1/cmr/m/n/12 )[]\OMS/cmsy/m/n/12 g[] ^ ^^ [] ^^^ f\OML/cmm/m/it/12 n\OT1/cmr/m/n/12 (\OML/cmm/m/it/12 ln\OT1/cmr/m/n/1 2 )[] [] (\OML/cmm/m/it/12 l[]n\OT1/cmr/m/n/12 )[]\OMS/cmsy/m/n/12 g[]\OML/cmm/ m/it/12 ; [] [74] [75] [76] [77] [78] Underfull \vbox (badness 5878) has occurred while \output is active [] [79 ] [80] [81] Underfull \vbox (badness 4954) has occurred while \output is active [] [82] Underfull \vbox (badness 2359) has occurred while \output is active [] [83] [84] [85 ] Underfull \vbox (badness 10000) has occurred while \output is active [] [1 ] Underfull \vbox (badness 10000) has occurred while \output is active [] [2] Underfull \vbox (badness 10000) has occurred while \output is active [] [3] Underfull \vbox (badness 10000) has occurred while \output is active [] [4] Underfull \vbox (badness 10000) has occurred while \output is active [] [5] Underfull \vbox (badness 10000) has occurred while \output is active [] [6] Underfull \vbox (badness 10000) has occurred while \output is active [] [7] [8] (./38079-t.aux) *File List* book.cls 2007/10/19 v1.4h Standard LaTeX document class bk12.clo 2007/10/19 v1.4h Standard LaTeX file (size option) inputenc.sty 2008/03/30 v1.1d Input encoding file latin1.def 2008/03/30 v1.1d Input encoding file ifthen.sty 2001/05/26 v1.1c Standard LaTeX ifthen package (DPC) amsmath.sty 2000/07/18 v2.13 AMS math features amstext.sty 2000/06/29 v2.01 amsgen.sty 1999/11/30 v2.0 amsbsy.sty 1999/11/29 v1.2d amsopn.sty 1999/12/14 v2.01 operator names amssymb.sty 2009/06/22 v3.00 amsfonts.sty 2009/06/22 v3.00 Basic AMSFonts support alltt.sty 1997/06/16 v2.0g defines alltt environment footmisc.sty 2009/09/15 v5.5a a miscellany of footnote facilities indentfirst.sty 1995/11/23 v1.03 Indent first paragraph (DPC) icomma.sty 2002/03/10 v2.0 (WaS) calc.sty 2007/08/22 v4.3 Infix arithmetic (KKT,FJ) yfonts.sty 2003/01/08 v1.3 (WaS) fancyhdr.sty graphicx.sty 1999/02/16 v1.0f Enhanced LaTeX Graphics (DPC,SPQR) keyval.sty 1999/03/16 v1.13 key=value parser (DPC) graphics.sty 2009/02/05 v1.0o Standard LaTeX Graphics (DPC,SPQR) trig.sty 1999/03/16 v1.09 sin cos tan (DPC) graphics.cfg 2009/08/28 v1.8 graphics configuration of TeX Live pdftex.def 2009/08/25 v0.04m Graphics/color for pdfTeX caption.sty 2009/10/09 v3.1k Customizing captions (AR) caption3.sty 2009/10/09 v3.1k caption3 kernel (AR) geometry.sty 2008/12/21 v4.2 Page Geometry ifpdf.sty 2009/04/10 v2.0 Provides the ifpdf switch (HO) ifvtex.sty 2008/11/04 v1.4 Switches for detecting VTeX and its modes (HO) geometry.cfg hyperref.sty 2009/10/09 v6.79a Hypertext links for LaTeX ifxetex.sty 2009/01/23 v0.5 Provides ifxetex conditional hycolor.sty 2009/10/02 v1.5 Code for color options of hyperref/bookmark (HO ) xcolor-patch.sty 2009/10/02 xcolor patch pd1enc.def 2009/10/09 v6.79a Hyperref: PDFDocEncoding definition (HO) etexcmds.sty 2007/12/12 v1.2 Prefix for e-TeX command names (HO) infwarerr.sty 2007/09/09 v1.2 Providing info/warning/message (HO) hyperref.cfg 2002/06/06 v1.2 hyperref configuration of TeXLive kvoptions.sty 2009/08/13 v3.4 Keyval support for LaTeX options (HO) kvsetkeys.sty 2009/07/30 v1.5 Key value parser with default handler support (HO) url.sty 2006/04/12 ver 3.3 Verb mode for urls, etc. bitset.sty 2007/09/28 v1.0 Data type bit set (HO) intcalc.sty 2007/09/27 v1.1 Expandable integer calculations (HO) bigintcalc.sty 2007/11/11 v1.1 Expandable big integer calculations (HO) pdftexcmds.sty 2009/09/23 v0.6 LuaTeX support for pdfTeX utility functions ( HO) ifluatex.sty 2009/04/17 v1.2 Provides the ifluatex switch (HO) ltxcmds.sty 2009/08/05 v1.0 Some LaTeX kernel commands for general use (HO) atbegshi.sty 2008/07/31 v1.9 At begin shipout hook (HO) hpdftex.def 2009/10/09 v6.79a Hyperref driver for pdfTeX umsa.fd 2009/06/22 v3.00 AMS symbols A umsb.fd 2009/06/22 v3.00 AMS symbols B supp-pdf.mkii color.sty 2005/11/14 v1.0j Standard LaTeX Color (DPC) color.cfg 2007/01/18 v1.5 color configuration of teTeX/TeXLive nameref.sty 2007/05/29 v2.31 Cross-referencing by name of section refcount.sty 2008/08/11 v3.1 Data extraction from references (HO) 38079-t.out 38079-t.out ./images/cups.png omscmr.fd 1999/05/25 v2.5h Standard LaTeX font definitions ./images/fig1.pdf ./images/fig2.pdf ./images/fig3.pdf ./images/fig4.pdf ./images/fig5.pdf *********** ) Here is how much of TeX's memory you used: 7352 strings out of 493848 104698 string characters out of 1152824 206416 words of memory out of 3000000 10224 multiletter control sequences out of 15000+50000 20610 words of font info for 77 fonts, out of 3000000 for 9000 714 hyphenation exceptions out of 8191 37i,17n,46p,298b,460s stack positions out of 5000i,500n,10000p,200000b,50000s Output written on 38079-t.pdf (101 pages, 839819 bytes). PDF statistics: 890 PDF objects out of 1000 (max. 8388607) 263 named destinations out of 1000 (max. 500000) 159 words of extra memory for PDF output out of 10000 (max. 10000000)