Sindbad~EG File Manager
����; � TeX output 2003.09.23:0001� ������� ���`� ����� � �'� ':
cmti10�A���rticle���Submitte��}'d�to�Journal�of�Symb�olic�Computation������� ���Ӏ ��N!���N� G� cmbx12�The�z�meaning�of�innit��u�y�in�calculus������`�aand�z�computer�algebra�systems��%>��p(���-�
cmcsc10�Michael���Beeson����2�&�-�
cmcsc10�1�� t��and�Freek�Wiedijk����2�2��� ��� �CM��^��ٓ�R cmr7�1�� ����San���Jos�����$�e�State�University�� ��� �S���^��2�� ���University���of�Nijme��}'gen��"� �� �3��4��N� cmbx12�Abstract�����4 �5K�`y
�3
cmr10�W��ee��use�lters�of�op�M�en�sets�to�pro��!vide�a�seman�tics�justifying�formally�the��
����4 use��&of�innit��!y�in�informal�limit�calculations�in�calculus,�and�in�the�same����4 kind��/of�calculations�in�computer�algebra.�W��ee�compare�the�b�M�eha��!vior�of����4 these�=�lters�to�the�w��!a�y�=�Mathematica�b�M�eha��!v�es�=�when�calculating�with�in-����4 nit��!y��e.�:A�prop�M�er�seman�tics�for�computer�algebra�expressions�is�necessary����4 not�H�only�for�the�correct�application�of�those�metho�M�ds,�but�also�in�order����4 to���use�results�and�metho�M�ds�from�computer�algebra�in�theorem�pro��!v�ers.����4 The�7�computer�algebra�metho�M�d�under�discussion�in�this�pap�er�is�the�use����4 of��rewrite�rules�to�ev��dDaluate�limits�in��!v�olving��innit�y��e.�Unlik�e�in�other�areas����4 of���computer�algebra,�where�the�problem�has�b�M�een�a�mismatc��!h�b�et��!w�een���a����4 kno��!wn� seman�tics�and�implemen�tations,�w�e�here�pro�vide�the�rst�precise����4 seman��!tics.��� �$ ��� �7��N� ff cmbx12�1.��*�1In���tro�s3duction��� �� �X�Q cmr12�In�K8calculus,�when�calculating�limits,�one�often�rst�uses�the�heuristic�of�`calcu-��� �� lating�^�with�innit��ry'�b�S�efore�trying�to�ev��X�aluate�the�limit�in�a�more�formal�w�a�y��V.�F�or���� instance��one�`calculates':������� ���lim��33�� �-.��2 cmmi8�x��K� cmsy8�!1��������ō� ��k�1�� �)V�[�� z� &�
����g� cmmi12�x����+�1����� ̸=������ō���1�����[�� z� X֟
��!",�
cmsy10�1����+�1�����)i�=�������1�����[�� z� �
��1������=�UR0��{:�� whic��rh�C5indeed�giv�es�the�correct�answ�er.�Ho�w�ev�er,�it�is�not�clear�what�the��8���@ cmti12�me��ffaning���� �of�5�this�use�of�the�sym��rb�S�ol�`�1�'�is,�and�wh�y�this�metho�S�d�w�orks.�This�problem�arises���� in���calculus�textb�S�o�oks,���whic��rh�usually�a�v�oid�examples�of�suc�h�calculations�for�fear���� of�v`lac��rk�of�rigor',�although�studen�ts�are�taugh�t�these�metho�S�ds�at�the�blac�kb�S�oard.���� This�uRproblem�also�arose�in�the�design�of�the�rst�author's�soft��rw�are,�uRMathX-���� p�S�ert�'(�1���;��3�� �#;��4��).�This�soft��rw�are,�'whic�h�is�designed�to�assist�a�studen�t�in�pro�S�ducing���� step-b��ry-step��solutions�to�calculus�problems,�had�to�b�S�e�able�to�pro�duce�`ideal'��� �� �1���� *��� ���`� ���s��Beeson��fand�Wiedijk:�The�meaning�of�innit��!y�]�2����� ����� �� �step-b��ry-step�4nsolutions�of�limit�problems.�Are�suc�h�`ideal�solutions'�allo�w�ed�to��� �� use���calculations�in��rv�olving���innit�y?�Or�are�those�calculations�just�priv��X�ate�prelim-���� inary�K�considerations�in��rtended�to�guide�a�rigorous�pro�S�of,�just�as�Newton�regarded���� calculus�bto�b�S�e�preliminary�computations?�MathXp�ert�do�es�allo��rw�calculations�in-���� v��rolving���innit�y��V,�but�not�the�full�system�justied�in�this�pap�S�er,�since�that�go�es���� b�S�ey��rond��what�one�nds�in�calculus�textb�o�oks.����!��In��kthe�Mathematica�system�(�12����)�the�approac��rh�of�calculating�with�innit�y�is���� used.�#Since�Mathematica�giv��res�answ�ers,�rather�than�step-b�y-step�solutions,�one���� will�!Dnot�notice�the�calculations�with�innit��ry��V,�in�cases�where�the�limit�turns�out�to���� exist��(and�b�S�e�a�nite�n��rum�b�er).��But�in�fact,�in�Mathematica�there�is�a�complete���� `calculus��of�innit��ry'�(and�some�related�sym�b�S�ols):��� �� �9߆�T cmtt12�In[1]:=�,�1/(Infinity�+�1)��� �� Out[1]=�,�0���� In[2]:=�,�Sqrt[Infinity]���� Out[2]=�,�Infinity���� In[3]:=�,�Infinity�-�Infinity���� Infinity::indet:��� ��(�gIndeterminate�,�expression�(-Infinity)�+�(Infinity)�encountered.���� Out[3]=�,�Indeterminate���� In[4]:=�,�Indeterminate�+�Infinity���� Out[4]=�,�Indeterminate���� In[5]:=�,�Sin[Infinity]���� Out[5]=�,�Interval[{-1,�1}]���� In[6]:=�,�1/Interval[{-1,�1}]���� Out[6]=�,�Interval[{-Infinity,�-1},�{1,�Infinity}]���� In[7]:=�,�Interval[{-1,�1}]*Interval[{-1,�1}]���� Out[7]=�,�Interval[{-1,�1}]���� In[8]:=�,�Interval[{-1,�1}]^2���� Out[8]=�,�Interval[{0,�1}]���� In[9]:=�,�0*Sin[Infinity]���� Out[9]=�,�Interval[{0,�0}]���� In[10]:=�,�Infinity/Sin[Infinity]���� Out[10]=�,�Interval[{-Infinity,�-Infinity},�{Infinity,�Infinity}]���� In[11]:=�,�Infinity/Sin[Infinity]^2���� Out[11]=�,�Interval[{Infinity,�Infinity}]��� �� �Other��\computer�algebra�systems�implemen��rt�similar�calculi.�F��Vor�instance,�the����� ֠�� ���`� ���s��Beeson��fand�Wiedijk:�The�meaning�of�innit��!y�]�3����� ����� �� �Maple�WQsystem�(�8���)�uses�the�sym��rb�S�ols��infinity��and��undefined��in�answ�ers�to�limit��� �� problems.�����2�|{Y cmr8�1������!���It�Oeis�w��rell�kno�wn�that�man�y�computer�algebra�pac�k��X�ages�mak�e�errors.�(See�for���� example���(�11����)�and�(�2���).)�One�of�the�reasons�for�that�is�that�they�fail�to�c��rhec�k���the���� pre-conditions�}�or�`side�conditions'�that�m��rust�b�S�e�satised�for�a�simplication�rule���� to�A�b�S�e�applicable.�F��Vor�example,�b�efore�applying��������p���A������ z� mV�
_��x������2������@�=��O�x��w��re�need�to�c�hec�k�that���� �x�UR���0.���Systematically�k��reeping�trac�k�of�suc�h�assumptions�is�dicult.�����2�2���}��The�errors���� in��computer�algebra�systems�sometimes�giv��re�the�impression�that�those�systems���� place�4�a�higher�priorit��ry�on�p�S�erforming�as�man�y�simplications�as�p�S�ossible�than�on���� ensuring��Athat�only�correct�computations�are�p�S�erformed.�Generally��V,�`ev��X�aluation���� errors'��Kwhic��rh�users�complain�ab�S�out�are�tak�en�care�of�on�an��ad��ho��ffc�h\�basis�only��V,�to���� get��rid�of�the�most�em��rbarrassing�ones.����!��Related�$Ato�these�errors�is�the�fact�that�these�systems�ha��rv�e�$Ano�unied�seman-���� tics��}for�their�expression�language.�In�this�pap�S�er�w��re�fo�cus�on�the�apparatus�for���� limits���and�oer�a�solution:�a�seman��rtics�explaining�and�supp�S�orting�the�use�of�in-���� nities���in�limit�calculations.�W��Ve�will�presen��rt�a�formal�seman�tics�of�limits,�whic�h���� not�̷only�explains�the�calculations�usually�p�S�erformed�with�innities,�but�oers���� some�39extensions�b��ry�in�tro�S�ducing�some�other�sym�b�S�ols�for�common�w�a�ys�in�whic�h�a���� function�F�can�fail�to�ha��rv�e�F�a�limit.�Th��rus,�w�e�will�b�S�e�able�to�get�an�answ�er�b�y�calcu-���� lation�P�for�suc��rh�a�limit�as��lim����a���x�!1���+�V�1�=�(�2���+��sin��Ŀx��+sB�)��o��whic�h�will�b�S�e�`oscillations�through���� the�"in��rterv��X�al�[�����Fu��33�1��33���� z� @���3������j�;����1]'.�W��Ve�then�compare�the�resulting�seman�tics�to�the�b�S�eha�vior�of���� Mathematica��Gas�illustrated�ab�S�o��rv�e.��GThere�is�a�rough�general�corresp�ondence,�and���� our�ŭseman��rtics�agrees�with�some�of�the�examples�ab�S�o�v�e,�but�in�some�instances���� Mathematica���do�S�es�giv��re�incorrect�answ�ers,�and�in�some�cases�w�e�are�able�to�dis-���� tinguish��db�S�et��rw�een�iden�tical�Mathematica�expressions�whic�h�are�dieren�t�in�our���� seman��rtics.����!��W��Ve�)�will�represen��rt��1��and�its�cousins��;��R6 cmss12�indeterminate��and��interval��b�y��lters�'��o�v�er���� some�dunderlying�top�S�ological�space.�In�calculus�textb�o�oks�this�is�the�space�of�real���� n��rum�b�S�ers,��?but�it�could�also�b�e�the�complex�n��rum�b�ers��?or�more�general�spaces.�(As���� w��re�$vshall�see,�Mathematica�is�a�bit�sc�hizophrenic�ab�S�out�whether�it�is�w�orking���� with�A�the�reals�or�the�complex�n��rum�b�S�ers,�A�but�for�most�of�our�w��rork�the�reader���� will���do�w��rell�to�k�eep�the�reals�in�mind.)�F��Vor�eac�h�p�S�oin�t�of�the�space�there�will���� b�S�e���a�lter�asso�ciated�with�it,�whic��rh�is�called�the��p�rincipal��lter�of�the�p�S�oin�t.�F��Vor���� eac��rh���function�on�the�space�there�will�b�S�e�a��lifte��ffd���v�ersion�that�w�orks�on�the�lters���� instead��of�on�the�p�S�oin��rts.����!��F��Vurthermore�vw��re�will�dene�classes�of�lters�called�the��interval�Q �lters�and�the���� �c��ffonne�cte�d��ֹlters.�TCIt�will�turn�out�that�those�t��rw�o�TCclasses�coincide�and�that�con-�� �����#����^��1����( �K�`y
cmr10�There�*�is�also�some�notion�of�in���terv��q�al�in�Maple,�written�as��:��<x
cmtt10�1�..�2�,�but�our�attempts�to�� �� calculate�Rwith�these�terms�led�only�to�error�messages.�These�terms�seem�primarily�to�b�Ge�used���� for�UUgenerating�in���teger�sequences,�although�the�answ�er�to��lim���8��� 0e�r cmmi7�x�O!� cmsy7�!1���'�<�sin��5��
�b>