Sindbad~EG File Manager

Current Path : /usr/home/beeson/public_html/michaelbeeson/research/papers/
Upload File :
Current File : /usr/home/beeson/public_html/michaelbeeson/research/papers/JSCVersion.dvi

����;� TeX output 2000.09.15:1226�����������`��������&�':

cmti10�A���rticle���Submitte��}'d�to�Journal�of�Symb�olic�Computation����������Ӏ��i8V���N�G�cmbx12�Automatic�z�Deriv���ation�of�the�������w
Irrationalit��u�y�z�of�e��#�Ѝ��xd��-�
cmcsc10�Michael���Beeson�����2��K�cmsy8�������r][�Dep��}'artment���of�Mathematics�and�Computer�Scienc�e�������San���Jose�State�University��,����3��+��N�cmbx12�Abstract�����4�,K�`y
�3
cmr10�As��4part�of�a�pro���ject�on�automatic�generation�of�pro�M�ofs�in��!v�olving��4b�oth��
����4logic�xand�computation,�w��!e�ha�v�e�automatically�generated�a�pro�M�of�of�the�ir-����4rationalit��!y�x
of��-�b>
�3
cmmi10�e�.�The�pro�M�of�in�v�olv�es�inequalities,�b�M�ounds�on�innite�series,����4t��!yp�M�e��distinctions�(b�et��!w�een��real�n��!um�b�ers��and�natural�n��!um�b�ers),��a�subpro�of����4b��!y���mathematical�induction,�and�signican�t�mathematical�steps,�includ-����4ing�S�correct�simplication�of�expressions�in��!v�olving�S�factorials�and�summing����4an�}�innite�geometrical�series.�Meta��!v��dDariables�are�instan�tiated�b�y�inference����4rules��bem��!b�M�o�dying�mathematical�kno��!wledge,�rather�than�only�b�y�unica-����4tion.�L~The�pro�M�of�is�generated�completely�automatically��e,�without�an��!y�in�ter-����4activ��!e��fcomp�M�onen�t.����$����2��N�ffcmbx12�1.��*�1In���tro�s3duction������X�Qcmr12�The��irrationalit��ry�of����g�cmmi12�e��can�b�S�e�stated�as��q��K>�[�0��!",�
cmsy10�!�j�p=q����y��e�j��>��0,��where��p��and������q�U�are��Fv��X�ariables�o��rv�er��Fnatural�n��rum�b�S�ers.��FThis�theorem�w��ras�rst�pro�v�ed�b�y�Euler����in�̱1737.�It�has�no��rw�b�S�een�pro�v�ed�b�y�a�mac�hine�(or�p�S�erhaps�I��vshould�sa�y��V,�b�y�a����program).�BT��Vo�pro��rv�e�Bthis�theorem�automatically�,�the�program�m��rust�start�with����a��sligh��rtly�dieren�t�form�of�the�goal:��q���>����0��!�9�C�ܞ�(�j�p=q�`����`�e�j���C��=q�n9�!��>��0).����Without��the�existen��rtial�v��X�ariable��C�ܞ�,�the�program�cannot�get�started.�Without����the�S�denominator��q�n9�!�it�also�cannot�succeed.�But�giv��ren�this�form�of�the�theorem,�it����do�S�es��
pro�ceed�completely�automatically�to�nd�a�pro�of.�(The�denominator��q�n9�!�giv��res����it���the�\hin��rt"�to�m�ultip�y�b�y��q�n9�!.)�The�pro�S�of�it�nds�is�similar�to�the�usual�pro�of�(see����for��example�(�13����),�Chapter�I)���but�emplo��rys�a�sligh�tly�dieren�t�estimate�to�b�S�ound����a���certain�innite�series.�A���certain�inequalit��ry�in�v�olving�factorials�is�needed�in�the����course�Fzof�the�pro�S�of;�the�program�nds�a�pro�of�of�this�inequalit��ry�b�y�mathematical����induction.�>The�inductiv��re�pro�S�of�requires�some�not-quite-straigh�tforw�ard�algebraic��UT�����#g��^��O!�cmsy7�����(�K�`y

cmr10�Researc���h�UUpartially�supp�Gorted�b�y�NSF�gran�t�n�um�b�Ger�CCR-9528913.�������1����*������`����U��M.��fBeeson:�Automatic�Deriv��dDation�of�the�Irrationalit��!y�of�e�?�2�������������manipulations�LNto�mak��re�use�of�the�induction�h�yp�S�othesis;�the�program�also�nds�����these���steps�automatically��V.�The�heart�of�the�pro�S�of�in��rv�olv�es���expressing�a�certain����quan��rtit�y�ʩ(in�this�case�expressed�b��ry�an�innite�series)�as�a�sum�of�t�w�o�parts,�one����of�^whic��rh�is�an�an�in�teger�and�the�other�of�whic�h�can�b�S�e�sho�wn�to�b�S�e�b�et��rw�een�^0����and�g�1.�It�follo��rws�that�the�quan�tit�y�is�not�zero,�and�indeed�can�b�S�e�estimated�from����b�S�elo��rw.���This�principle�is�a�v��X�ariation�on�the�basic�principle�that�an�in�teger�kno�wn����to��b�S�e�p�ositiv��re�m�ust�b�S�e�at�least�1.����!��The���instan��rtiation�that�the�program�nds�for�the�v��X�ariable��C��n�in�v�olv�es�the�denom-����inator����q�n9�.�A���h��ruman�mathematician�can�\rene�the�estimates"�to�nd�a�constan�t�����B��indep�S�enden��rt�X�of��p��and��q��ۼ(�B����=��1�=�2�will�do)�suc�h�that��j�p=q�c������e�j���>�B��=�(�q��+���1)!.����This��is�b�S�ey��rond�the�presen�t�capacit�y�of�the�program.�(It�also�is�not�presen�ted�in����b�S�o�oks��that�presen��rt�the�pro�S�of�of�the�irrationalit�y�of��e�.)����!��This��pap�S�er�explains�the�rules�of�inference�used�b��ry�the�pro�v�er,�and�then�ex-����plain���ho��rw�the�pro�v�er�constructs�this�particular�pro�S�of.�An�app�endix�con��rtains�the����actual��prin��rtout�of�the�pro�S�of�as�pro�duced�b��ry�the�program.�The�reader�wishing����more�0pinformation�ab�S�out�the�program�can�consult�(�1���),�where�the�rst�results�with����this�)�theorem-pro��rving�program�are�describ�S�ed.�Those�results�w�ere�automatically-����generated��#epsilon-delta�pro�S�ofs�of�con��rtin�uit�y��#of�sp�ecic�functions.�It�w��ras�b�ecause����of��the�application�to�epsilon-delta�pro�S�ofs�that�the�program�w��ras�named��3���@cmti12�Weier-����str��ffass�,��6after�one�of�the�pioneers�of�the�epsilon-delta�metho�S�d.�Ho��rw�ev�er,��6the�name����seems��Nless�appropriate�no��rw,�as�an�expanded�v�ersion�of�the�pro�v�er�demonstrates����its��capabilities�in�n��rum�b�S�er��theory�as�w��rell�as�analysis.��$����2.��*�1Con���text�ffof�this�Researc�h������Mathematics��pconsists�of�logic�and�computation,�in��rterw�o�v�en��pin�tap�S�estries�of�pro�ofs.����\Logic"�+�is�represen��rted�b�y�the�manipulation�of�phrases�(or�sym�b�S�ols)�suc�h�as��for����al���l���x�,���ther��ffe�exists�an��x�,��implies�,�etc.�\Computation"�refers�to�c��rhains�of�form�ulas����progressing�9Vto��rw�ards�an�\answ�er",�suc�h�as�one�mak�es�when�ev��X�aluating�an�in�tegral����or�еsolving�an�equation.�T��rypically�computational�steps�mo�v�e�\forw�ards"�(from�the����kno��rwn�Z�facts�further�facts�are�deriv�ed)�and�logical�steps�mo�v�e�\bac�kw�ards"�(from����the�.�goal�to��rw�ards�.�the�h��ryp�S�othesis,�as�in��it�]would�suc��ffe�to�pr�ove�.�.�The�mixture����of��#logic�and�computation�giv��res�mathematics�a�ric�h�structure�that�has�not�y�et����b�S�een�T�captured,�either�in�the�formal�systems�of�logic,�or�in�computer�programs.����The���computer�program�that�pro�S�duced�the�pro�of�rep�orted�on�here�demonstrates����an���approac��rh�to�this�problem.�The�general�features�of�this�approac�h�ha�v�e�b�S�een����describ�S�ed��Vin�(�1���).�The�researc��rh�in�v�olv�es�t�w�o�computer�programs,��Mathp��ffert��and�����Weierstr��ffass�.���The�former�has�b�S�een�rep�orted�on�elsewhere�in�detail�(�3���;��4���;��5��):�it�con-����tains�6oimplemen��rtations�of�o�v�er�t�w�o�thousand�mathematical�op�S�erations,�together����with�,�logical�apparatus�to�k��reep�trac�k�of�assumptions�that�ma�y�b�S�e�required�or����generated���b��ry�those�op�S�erations.��Mathp��ffert��(as�in�\Math�Exp�ert")�uses�these�op�er-����ations��^to�pro��rvide�a�computerized�en�vironmen�t�for�learning�algebra,�trigonometry��V,����and��calculus.����!��The�Qsecond�program,��Weierstr��ffass�,�is�built�up�S�on�a�simple�bac��rkw�ards-Gen�tzen�����E������`����U��M.��fBeeson:�Automatic�Deriv��dDation�of�the�Irrationalit��!y�of�e�?�3�������������theorem�%Lpro��rv�er,�describ�S�ed�in�(�6���)�(but�re-implemen�ted�in�C).�T��Vo�this�logical�bac�k-�����b�S�one��phas�b�een�added�a�set�of�con��rtrol�structures,�or�if�y�ou�lik�e,�implemen�tations����of�P�sp�S�ecial�inference�rules,�to�facilitate�the�prop�er�con��rtrol�of�logical�and�compu-����tational� asteps.�These�con��rtrol�structures�op�S�erate�at�the�top�lev�el�of��Weierstr��ffass�,����but�Õthe�computational�steps�themselv��res�can�use,�in�principle,�an�ything�that�has����b�S�een�G�implemen��rted�in��Mathp��ffert�,�whic�h�is�all�of�high-sc�ho�S�ol�algebra,�trigonometry��V,����and�Tone-v��X�ariable�calculus�including�limits,�dieren��rtiation,�and�(textb�S�o�ok-st�yle)����in��rtegration,�-as�w�ell�as�a�go�S�o�d�-man�y�tec�hniques�for�rewriting�inequalities,�and�a����few� Qadv��X�anced�algorithms,�suc��rh�as�the�Coste-Ro�y�algorithm�(�9���),�based�on�Sturm's����theorem,�G[for�determining�whether�p�S�olynomials�ha��rv�e�G[ro�ots�in�giv��ren�in�terv��X�als.�The����implemen��rtations��of�these�op�S�erations�in��Mathp��ffert��are�logically�correct,�so�that����they��%can�b�S�e�used�in��Weierstr��ffass��without�the�risk�of�inconsistency�that�w��rould����accompan��ry���the�similar�use�of��Mathematic��ffa�,��Maple�,�or��Macsyma�.�����2�y���"�Simplication����in�^��Mathp��ffert��in��rteracts�with�the�list�of�curren�t�assumptions;�assumptions�can�b�S�e����used�t�in�simplication,�and�simplication�can�generate�new�assumptions.�A�tzmore����detailed��description�of��Weierstr��ffass��can�b�S�e�found�in�(�1���).����!��The���driving�idea�of�this�researc��rh�program�is�that�nding�mathematical�pro�S�ofs����requires��exp�S�ert�kno��rwledge�of�h�undreds�of�sp�S�ecial�inference�rules.�An�inference�rule����t��rypically��encapsulates�kno�wledge�of�ho�w�to�pro�v�e�theorems�of�a�certain�form�or�in����a��certain�con��rtext.�F��Vor�example,�in�(�1���)�w�e�ga�v�e�sp�S�ecial�inference�rules�that�are�used����for�F:b�S�ounding�giv��ren�quan�tities�in�terms�of�other�quan�tities.�These�rules�enabled�����Weierstr��ffass��K�to�automatically�generate�epsilon-delta�pro�S�ofs�of�the�con��rtin�uit�y��Kof����certain��sp�S�ecic�functions,�but�they�are�not�sp�ecial-purp�ose�rules{b�ounding�or����\estimating"���expressions�is�a�fundamen��rtal�activit�y�in�analysis.�No�w,�in�this�pap�S�er����w��re�`oer�further�evidence�for�this�view�of�the�nature�of�mathematics:�w�e�add�to�����Weierstr��ffass���a�few�simple�inference�rules�concerning�estimates�for�innite�series,����and�o{the�principle�that�a�p�S�ositiv��re�quan�tit�y�whic�h�is�an�in�teger�m�ust�b�S�e�at�least����one.�TfThe�underlying�logic�already�p�S�ossessed�the�abilit��ry�to�distinguish�v��X�ariables����of�O�t��ryp�S�e��in��teger��and��real�.�W��Ve�also�add�mathematical�induction�and�another�new����inference�%�rule�to�aid�in�con��rtrolling�the�transitivit�y�la�w�for�inequalities.�These����few�,dinference�rules,�together�with�the�elemen��rtary�mathematics�a�v��X�ailable�from�����Mathp��ffert�,��(enable�the�general�logical�mec��rhanism�of��Weierstr�ass��to�nd�a�pro�S�of����of�+the�irrationalit��ry�of��e�,�whic�h�is�a�non-trivial�theorem,�usually�presen�ted�in�an����upp�S�er-division��n��rum�b�er��theory�course.����!��There�`�are�at�presen��rt�some�fteen�mathematical�inference�rules�in��Weierstr��ffass�,����in�#�addition�to�the�logical�(Gen��rtzen)�rules�and�mathematical�induction.�It�is�note-����w��rorth�y�K�that�all�mathematical�kno��rwledge�needed�for�the�pro�S�of�of�the�irrationalit�y����of���e��is�em��rb�S�edded�in�these�rules�and�in�the�simplication�la�ws�implemen�ted�in�����Mathp��ffert�.���That�is,�the�input�le�for�this�pro�S�of�con��rtains�only�the�goal;�no�axioms����sp�S�ecial�A�to�this�pro�of�need�to�b�e�pro��rvided.�This�strategy�of�em�b�S�o�dying�A�math-��UT�����#�X��^��y����(�T��*�o��1assert�that�the�implemen���tations�are�logically�correct�is�not�to�sa�y�that�this�logical����correctness��5has�b�Geen�pro���v�ed.��5What�w���e�mean�is�simply�that�an�y�failure�of�logical�correctness����w���ould�UUb�Ge�a�programming�error,�whic�h�could�b�Ge�xed�without�altering�the�program�design.������������`����U��M.��fBeeson:�Automatic�Deriv��dDation�of�the�Irrationalit��!y�of�e�?�4�������������ematical�9Ckno��rwledge�in�inference�rules�is�the�k�ey�to�our�success:�the�inference�����rules���can�use�mathematical�kno��rwledge�to�instan�tiate�meta�v��X�ariables,�rather�than����relying��ion�unication.�Unication�(ev��ren�the�more�sophisticated�v�ersions�of�it)�is����a���v��rery�primitiv�e�w�a�y�to�instan�tiate�a�meta�v��X�ariable.�It�is�hop�S�elessly�inadequate����for�V�dealing�with�inequalities:�try�pro��rving��9�x�(0�
l�<�x�<��1)�V�b�y�a�unication-based����algorithm.����!��Related�3�w��rork�includes��A��2nalytic��ffa��(�8���),�whic�h�is�a�theorem-pro�v�er�link�ed�to����the�7�computational�facilities�of��Mathematic��ffa�,�but�deals�only�with�quan��rtier-free����pro�S�ofs.���This�means�also�meta��rv��X�ariable-free�pro�ofs;�ev��ren�though�most�of�the�w�ork����on�{Gthe�irrationalit��ry�of��e��is�quan�tier-free,�the�use�of�meta�v��X�ariables�and�con�trol����mec��rhanisms�μfor�instan�tiating�them�is�crucial,�so�it�will�not�b�S�e�p�ossible�to�repro-����duce��Ethis�w��rork�in��A��2nalytic��ffa�.�The�pro�v�er�Nqthm�(�7���)�has�pro�v�ed�some�impressiv�e����theorems�LNof�n��rum�b�S�er�LNtheory��V,�including�the�la��rw�of�quadratic�recipro�cit��ry��V,�but�lik�e�����A��2nalytic��ffa�,�xw��rorks�b�S�est�with�free-v��X�ariable�pro�ofs,�and�in�particular,�lac��rks�the�rules����of��inference�in��rtro�S�duced�here,�so�it�cannot�pro�v�e�the�irrationalit�y�of��e�.�����2�z������!���Some�ʑw��rork�on�the�v�erication�in�HOL��of�pro�S�ofs�in�v�olving�computation�is����presen��rted��in�(�11����),�but�this�do�S�es�not�in�v�olv�e�pro�S�of�generation.�Otter�(�12����)�can�do����some�t�kinds�of�computation�using�rewrite�rules�and�A��rC�t�unication,�but�these�are����no��rwhere���near�adequate�for�the�kinds�of�simplication�needed�in�this�pro�S�of.�The����strength���of�Otter�lies�in�its�pro�S�of-searc��rh�capabilities.�The�rules�of�inference�used����here�)�are�sophisticated�enough�that�there�is�almost�no�searc��rhing�in�v�olv�ed�in�the����pro�S�of.���Eac��rh�step�is�either�simplication�(forw�ard�reasoning)�or�is�an�inference����step��dictated�b��ry�the�form�of�the�goal,�whic�h�turns�out�to�succeed.����!��A��sc��rheme�for�regarding�computation�as�dening�an�equiv��X�alence�relation�on����form��rulas�f�and�dening�deduction�on�the�equiv��X�alence�classes�is�set�forth�in�(�10����).����The�(�pro��rv�er�discussed�in�this�pap�S�er�do�es�not�t�in��rto�this�framew�ork,�since�com-����putations��can�b�S�oth�use�the�assumptions�and�generate�new�ones.��$����3.��*�1Nature�ffof�the�Pro�s3ofs�Pro�duced�b���y��4F
C�ff
cmbxti10�Weierstr��&�ass������T��Vo���a��rv�oid�confusion,�some�discussion�of�nature�and�purp�S�ose�of�the�pro�ofs�gen-����erated�ob��ry��Weierstr��ffass��is�necessary��V.��Weierstr�ass��pro�S�duces�(in��rternally)�a�pro�of-����ob��ject,���whic��rh�can�b�S�e�displa�y�ed�or�sa�v�ed�in�more�than�one�form.�The�in�ten�tion����is�/to�pro�S�duce�a�pro�of�that�can�b�e�read�and�c��rhec�k�ed�/for�correctness�b��ry�a�h�uman����mathematician;�Xythe�standard�to�b�S�e�met�is�\p�eer�review",�just�as�for�journal�pub-����lication.�[The�pro�S�of�ob��ject�pro�duced�do�es�not�represen��rt�a�completely�formal�pro�of����in�I!a�sp�S�ecied�formal�system�of�the�t��ryp�e�traditionally�studied�b��ry�logicians.�It�fails����to���meet�this�standard�b�S�ecause�some�of�the�steps�are�calculations�carried�out�b��ry����algorithms,��Gfor�example,�factoring�p�S�olynomials�or�\simplifying".�One�could,�of����course,�μeasily�sp�S�ecify�a�formal�system�allo��rwing�suc�h�steps,�and�this�is�in�essence����what�bDthe�co�S�de�for��Weierstr��ffass��do�es.�But�the�steps�lab�eled�\simplify"�are�hard�to��UT�����#�X��^��z����(�Presumably�vone�could��add��these�rules�to�Nqthm.�W��*�e�are�not�sp�Geaking�here�of�some�limi-����tation�UUon�all�p�Gossible�extensions�of�Nqthm.�����.������`����U��M.��fBeeson:�Automatic�Deriv��dDation�of�the�Irrationalit��!y�of�e�?�5�������������translate�W�in��rto�traditional�logical�systems,�b�S�ecause�what�seems�to�a�h�uman�math-�����ematician��.a�v��rery�simple�\simplication"�can�b�S�e�hard�to�pro�v�e�in�a�traditional����logical��csystem.�This�is�part�of�the�reason�that�no�previous�computer�program����has� pro�S�duced�suc��rh�pro�ofs:�most�suc��rh�programs�insist�on�pro�ducing�formal�pro�ofs.����If�ga�mac��rhine�to�pro�S�duce�h�umanly-readable�pro�S�ofs,�simplication�steps��must��b�e����allo��rw�ed.���Then�one�has�three�c��rhoices:�(1)�Settle�for�computer-generated�pro�S�ofs����whose�0�correctness�is�judged�b��ry�h�uman�b�S�eings,�as�in�W��Veierstrass.�(2)�Supply�,�once����and��for�all,�a�formal�pro�S�of�of�the��r��ffesult��of�eac��rh�of�the�(1800�or�so)�p�ossible�sim-����plication��%steps,�e.g.�a�pro�S�of�that��x����2�|{Ycmr8�2���)�+��%2�xy�A^�+��y��n9���2�2�����=�UR(�x��+��y�n9�)����2�2����,��%and�instan��rtiate�these����pro�S�ofs��for�eac��rh�simplication�actually�used.�This�w�ould�turn�W��Veierstrass'�pro�S�ofs����in��rto�|aformal�ones.�(3)�Demand�in�addition�that�all�algorithms�in�v�olv�ed�ha�v�e�their����correctness�a
pro��rv�ed�within�the�system.�This�is�a�higher�standard,�whic�h��Weier-����str��ffass��N�cannot�meet,�and�w��rould�not�b�S�e�met�ev�en�b�y�(2).�Human�mathematicians����do�U8not�meet�standards�(2)�and�(3)�either,�so�it�seems�reasonable�to�demand�of����a���mac��rhine�in�tended�as�a�protot�yp�S�e�\mathematician's�assistan�t"�that�it�should����meet��the�standards�required�for�journal�publication,�instead�of�a�higher�standard.��$����4.��*�1Inference�ffrules�used�in��Weierstr��&�ass������In���this�section�w��re�describ�S�e�the�new�rules,�and�some�v��X�ariations�on�old�rules,�that����ha��rv�e���b�S�een�added�to��Weierstr��ffass��since�the�publication�of�(�1���).�The�discussion�will����mak��re�k�it�apparen�t�that�these�rules�are�general�in�nature,�rather�than��ad��2ho��ffc�,����and���that�they�em��rb�S�o�dy���mathematical�ideas�and�tec�hniques�that�are�in�tuitiv�e����and���can�b�S�e�applied�to�mathematical�argumen��rts�in�dieren�t�areas.�That�is,�they����are��Qnot�sp�S�ecial�to�the�example�of�the�irrationalit��ry�of��e�.�Of�course,�there�are����probably�Ísev��reral�h�undred�suc�h�rules�encapsulating�the�mathematical�kno�wledge����of��a�b�S�eginning�graduate�studen��rt,�and�the�selection�of�these�particular�rules�w�as����motiv��X�ated���b��ry�the�exp�S�erimen�t�of�seeing�what�w�as�required�for�the�irrationalit�y�of�����e�.��������5�"V
�3
cmbx10�4.1.��/�hT��typ�Y�es������Weierstr��ffass��m�accepts�t��ryp�S�ed�v��X�ariables,�as�in��8�x�}�:���N��S�(�x����0).��mIt�also�accepts�t�yp�S�e����\judgemen��rts"�Ξsuc�h�as��x���:���N����as�atomic�form��rulae.�The�latter�is�necessary�to����supp�S�ort��the�\v��X�ariable�t��ryp�e"�of�F��Veferman's�applicativ��re�theories�(whic�h�are�not����used�k�in�the�presen��rt�example),�and�the�former�is�v�ery�useful�for�ecien�t�theorem-����pro��rving.�nThe�t�w�o�are�connected�b�y�means�of�an�inference�rule�that�p�S�ermits�the����inference�\�of��n��)�:��A�\��(from�no�premises)�when��n��is�a�v��X�ariable�of�t��ryp�S�e��A�.�T�yp�S�e����information���for�the�atomic�t��ryp�S�es�(in�this�case,�just���N���,���Z��ȳ�,�and���R��
�{�)�is�immediately����accessible�in�the�(implemen��rted)�v��X�ariable�itself.�T�yp�S�e�information�for�comp�ound����t��ryp�S�es��y(suc�h�as���N����!��UR�N���or�the�t��ryp�S�es�of�F��Veferman's�theories)�is�k�ept�in�a�separate����table.����!��An�(,imp�S�ortan��rt�feature�of�this�pro�v�er�is�that�simplication�is�applied�to�prop�S�osi-����tions�
�as�w��rell�as�to�mathematical�expressions.�Prop�S�ositions�expressing�t�yp�S�e�judge-�����>0������`����U��M.��fBeeson:�Automatic�Deriv��dDation�of�the�Irrationalit��!y�of�e�?�6�������������men��rts�r�are�no�exception.�F��Vor�example,�the�t�yp�S�e�expression��q�n9�!�=n�!�=>:���Z���'�(whic�h�r�ex-�����presses���that��q�n9�!�=n�!�is�an�in��rteger)�reduces�to��n������q��.���Another�example�of�a�t��ryp�S�e����expression���that�o�S�ccurs�in�the�pro�of�of�irrationalit��ry�of��e��is��p�(�q�����J��1)!�UR:���Z���R�.���This�will����reduce�cKto��true��if��p�UR�:��Z�?�and�cK(�q�c���*�1)!�:���Z�����reduce�to��true�.�The�former�will�reduce�to�����true��m�since��p��is�a�v��X�ariable�of�t��ryp�S�e���N����.�The�latter�will�reduce�to��true��if�(�q�J����1)!�UR:���Z������do�S�es,�V�but�this�reduces�to�1�UR���q�n9�.�V�Ho��rw�ev�er,�since�the�assumptions�can�also�b�S�e�used����in��simplication,�if�0�UR�<�q��9�is��in�the�assumption�list,��Weierstr��ffass��is�able�to�use�the����t��ryp�S�e��6information�that��q�8o�is�of�t�yp�S�e���Z���l�to�reduce�1�UR���q�8o�to��6�true�.�Th�us��p�(�q�֛��hb�1)!�UR:���Z������can��b�S�e�reduced�to��true��b��ry�simplication�alone.�No�logical�inference�is�required.����When�jdebugging�this�pro�S�of,�this�simplication�failed�at�rst,�b�ecause�un��rtil�then,����I��had�forgotten�to�put�the�h��ryp�S�othesis��q�Ë>�UR�0�in�the�statemen�t�of�the�theorem.����!��Reducing��:t��ryp�S�e�expressions�in�v�olving�series:���������u
cmex10�P����*���d���2cmmi8�q��	U_��d�n���
5�q�n9�!�=n�!�is�an�in�teger,�and�this�also����can�ϼb�S�e�deduced�without�inference�rules,�b��ry�simplication�alone.�T��Vo�deduce�this,����w��re�h�simplify�the�expression��q�n9�!�=n�!�+�:��Z�ܞ�;�h�but�to�simplify�this�to��true�,�w�e�need�the����assumption��c�n�UR���q�n9�.�In�previous�publications�the�tec��rhnique�has�b�S�een�describ�ed�b��ry����whic��rh��~�Mathp��ffert��mak�es�temp�S�orary�assumptions�while�simplifying�an�expression����suc��rh�y�as�an�indexed�sum�or�denite�in�tegral,�so�that�suc�h�assumptions�are�indeed����a��rv��X�ailable���when�required.�This�tec�hnique�is�called�the�\binders"�tec�hnique,�since����it��applies�when�expressions�in��rv�olving��v��X�ariable-binding�op�S�erators�are�simplied.����!��Subt��ryp�S�es�nK(t�yp�e�nKem�b�edding)�nKare�supp�orted�b��ry��Weierstr��ffass�.�F��Vor�example,�one����can��,infer��a��-�:���Z��bY�from��,�a��:���N��g�.�The�system�of�atomic�t��ryp�S�es�includes���N��+f�,���Z���,�,���R����,���Q�����,����and��T��C����,�T�although�only���N����,���Z�����,�and���R���:�are�in��rv�olv�ed�T�in�the�irrationalit��ry�of��e�.�More����complicated�ɨt��ryp�S�es,�suc�h�as�function�t�yp�S�es�or�comprehension�t�yp�S�es,�are�not�used����in��this�example.��������4.2.��/�hSimplication�2�of�factorials�and�innite�series������Certain��bsimplications�needed�in�the�pro�S�of�w��rere�not�pro�vided�b�y�calls�to��Mathp��ffert�����co�S�de,��|and�new�simplication�rules�w��rere�in�tro�S�duced�in��Weierstr��ffass��to�meet�these����needs.�_#F��Vor�example,�the�la��rws�of�the�factorial�function�are�not�considered�elemen-����tary��'enough�that��Mathp��ffert��should�use�them�when�a�studen��rt�c�ho�S�oses�\simplify".����I�K�ha��rv�e�K�in�mind�here�suc��rh�la�ws�as�(�n�f��+�1)!�UR=�(�n�f��+�1)�n�!�K�and�the�corresp�S�onding�\fac-����torial�y4cancellation�la��rws",�suc�h�as�(�n���+�1)!�=�(�n��+�1)�UR=��n�!�y4and�(�n���+�1)!�=n�!�UR=�(�n���+�1).����These���la��rws�need�to�b�S�e�applied�to�fractions�that�con�tain�other�factors�in�n�umer-����ator���and�denominator,�in�an�arbitrary�order�(as�with�man��ry�simpler�rules�used����in���Mathp��ffert�).�But�for�p�S�edagogical�considerations�they�could�just�as�w��rell�ha�v�e����b�S�een�{included�with�other�more�\algebraic"�la��rws.�An�example�of�their�use�is�in����v��rerifying�X�the�inequalit�y�2�n�!����(�n����+�1)!,�X�whic�h�simplies�to�2����n����+�1�X�and�then����to��1�UR���n�.����!��Another���\simplication"�needed�in�the�pro�S�of�is�\recognizing"�and�summing����an��innite�geometric�series.�Again,�there�are�only�p�S�edagogical�reasons�wh��ry�this����step���is�not�p�S�erformed�b��ry�the�\simplify"�command�of��Mathp��ffert�,�and�so�had�to�b�e����added��sp�S�ecially�to��Weierstr��ffass�.����!��On��gthe�other�hand,�it�is�essen��rtial�in�this�example�that��e��b�S�e�expanded�in�to�an�����M͠�����`����U��M.��fBeeson:�Automatic�Deriv��dDation�of�the�Irrationalit��!y�of�e�?�7�������������innite�~�series.�Clearly�w��re�do�not�w�an�t�to�expand��e��in�to�a�series�ev�ery�time�w�e�����see����e��in�a�problem;�it�is�done�only�as�a�last�resort�when�w��re�cannot�get�an�ywhere����b��ry��other�means.��������4.3.��/�hLo��tw�er�2�b�Y�ounds�on�innite�series������The�EMonly�rule�of�this�kind�presen��rtly�implemen�ted�in��Weierstr��ffass��is�this:�T��Vo�pro�v�e����a��series�is�p�S�ositiv��re,�pro�v�e�its�general�term�is�p�S�ositiv�e.�Sp�S�ecically��V,�this�rule�tak�es����the��follo��rwing�form:��%�]����K��!���c�UR���k�g;������)��0��<�a�����k����-���!����xR��[�s��������(���UR�)��0��<�������1��������X���'؍����k�6��=�c������a�����k������%�\��!���There��is�a�similar�rule�for�nite�series,�and�a�similar�rule�for����in�place�of�strict����inequalit��ry��V.�ՁOf�course�one�could�also�form�ulate�a�v�ersion�of�the�rule�in�whic�h�the����h��ryp�S�othesis�-requires�only�that�the�general�term�b�e�nonnegativ��re�and�at�least�one����term���is�p�S�ositiv��re,�but�this�is�not�required�in�the�example�of�the�irrationalit�y�of��e�.����!��This��rule�could�ha��rv�e��b�S�een�implemen��rted�just�as�w�ell�as�a�simplication�rule����(applying�k'to�the�inequalit��ry�in�the�conclusion)�as�an�inference�rule.�F��Vor�that�mat-����ter,�k�it�could�ha��rv�e�k�b�S�een�included�as�an�axiom�and�treated�as�sp�ecial�to�this�pro�of;����in��that�case�the�logical�mec��rhanisms�w�ould�ha�v�e�handled�this�t�yp�S�e�of�inference.��������4.4.��/�hUpp�Y�er�2�b�ounds�on�innite�series������T��Vo��b�S�ound�a�series�from�ab�o��rv�e,��use�the�comparison�test�to�estimate�the�series�from����ab�S�o��rv�e�a�b�y�a�series�whose�sum�is�kno�wn.�The�simplest�form�of�this�rule�w�ould�tak�e����the��follo��rwing�form:��&�	����K��!���c�UR���k�g;������)�j�a�����k��#��j���b�����k����ٍ��!����xR��f�n��������"*ȼ�UR�)�������1��������X���'؍����k�6��=�c������a�����k��x���������1��������X���'؍����k�6��=�c����b�����k������&���!���Ho��rw�ev�er,�l
this�simple�form�of�the�rule�w��rould�b�S�e�useless.�W��Ve�w�an�t�to�use�this����principle,�ғfor�example,�to�pro��rv�e�ғthat�������P����*���}@�1��	U_��}@�k�6��=�c���"��a�����n��
�c�<���1,�b��ry�nding�a�suitable�com-����parison�a�series�whose�sum�is�less�than�1.�Therefore�this�rule�m��rust�b�S�e�used�in����com��rbination�A�with�the�transitivit�y�of�equalit�y��V.�Con�trol�of�transitivit�y�is�a�recur-����ring��theme�in�automated�deduction,�and�here�is�one�con��rtribution�to�it:��2
�����E��!���c�UR���k�g;������)�j�a�����k��#��j���b�����k�������)�������1��������X���'؍����k�6��=�c������b�����k��x��<�d���䍍�!����xR����w��������`J���UR�)�������1��������X���'؍����k�6��=�c������a�����k��x����d�������!���This��rule�is�m��ruc�h��more�in��rteresting,�since�the�question�arises�as�to�ho�w,�when�����trying�l(to�apply�this�rule�in�rev��rerse,�the�comparison�series��b�����k�����is�to�b�S�e�determined.�����_�������`����U��M.��fBeeson:�Automatic�Deriv��dDation�of�the�Irrationalit��!y�of�e�?�8�������������There���is�an�ev��ren�more�basic�question:�ho�w�can�this�rule�b�S�e�prop�erly�expressed?�����In��Fthe�rst�v��rersion�of�the�rule,��a�����k��	�ؼand��b�����k���are�just�some�terms�of�the�language����con��rtaining��Cthe�v��X�ariable��k�g�.�In�the�second�form,�ho�w�ev�er,�if��b��is�to�b�S�e�determined,����w��re���w�ould�app�S�ear�to�need�a�v��X�ariable��b�,�and�that�v�ariable�w��rould�ha�v�e�to�b�S�e�of����t��ryp�S�e���J�N��x&�!��a��R��sj�.��JThe�language�of��Weierstr��ffass��w�ould�supp�S�ort�suc�h�a�form�ulation,����but�W�that�w��rould�lea�v�e�it�to�unication�to�nd�a�series�with�a�kno�wn�upp�S�er�b�ound.����Ev��ren��if�higher-order�unication�w�ere�implemen�ted�in��Weierstr��ffass��it�w�ould�not�b�S�e����helpful��unless�there�w��rere�a�list�of�axioms�giving�upp�S�er�b�ounds�on�kno��rwn�series.����!��Instead,�,this�rule�is�included�sp�S�ecically�for�geometric�series�������P����*�����1��	U_����k�6��=�c���"EB�b�����k��#��.�A�,geo-����metric�.series�is�constructed�with�ratio�1�=�2�and�rst�term��b�����c�����=��a�����c��.y�.�In�the�future,����this���can�b�S�e�extended�to�other�t��ryp�es�of�kno��rwn-summable�series.�This�seems�to����corresp�S�ond���fairly�w��rell�to�our�in�tuitiv�e�idea�of�a�mathematician's�\bag�of�tric�ks"����for��b�S�ounding�a�series�from�ab�o��rv�e.����!��A�3dicult��ry�3Barises�when�w�e�w�an�t�to�pro�v�e�a�strict�inequalit�y��V.�Of�course�the�rule����w��rorks��if�w�e�replace����b�y��<��in�b�S�oth�premise�and�conclusion;�but�since�in�the�cases����of���in��rterest,�w�e�are�using�a�comparison�series�with�the�same�rst�term,�w�e�will����not�!�b�S�e�able�to�establish�a�strict�inequalit��ry�in�the�premise.�The�rule�is�of�course����not�x�correct�with��<��in�the�conclusion�and����in�the�premise,�but�it�is�correct�if�w��re����add���the�additional�premise�that�the�second�term�of�the�series�satises�a�strict����inequalit��ry��V.��This�form�of�the�rule�has�b�S�een�put�in�to�the�program.���卍��4.5.��/�hCon��ttrolling�2�transitivit�y��-č��It�מis�a�constan��rt�struggle�in�automated�deduction�to�p�S�ermit�all�desired�uses�of����transitivit��ry�cp(of�equalit�y�or�inequalit�y)�while�prev�en�ting�innite�regress�or�com-����binatorial���explosion.�Here�is�another�rule�along�these�lines.�W��Ve�call�it�the�\tran-����sitiv��recancel"��rule.�(In�this�rule,��Av�X�means��A��m�ultiplied�b�y��v�n9�.)��������@��!���;���A�UR���B��X�)��C�1�=��Av��)v�Ë�:���N��%c|��;�A����B��X�)��B��v�Ë���D���-���!����xR���َ��	;����~����;���A�UR���B��X�)��C�1����D�����V��!���In��5the�implemen��rtation�of�this�rule,�unication�is�not�sucien�t�to�nd��v�n9�.�In-����stead,���algebraic�cancellation�is�applied�to��C��=��X�A��for�the�v�arious��A��o�S�ccurring�in����inequalities�6Bin�the�an��rteceden�t,�6Bun�til�a�cancellation�is�found�pro�S�ducing�a�term�����v��`�=�t'�C��=��X�A��żwhic��rh�is�\ob�viously"�a�nonnegativ�e�in�teger�(for�example,�a�v��X�ariable�of����t��ryp�S�e��M��N����,�M�or�a�n�um�b�S�er,�or�a�pro�duct�of�in��rtegers).�Without�the�restriction�that��v�����should��b�S�e�an�in��rteger,�the�rule�is�of�course�still�v��X�alid�(if��v�_ռis�nonnegativ�e),�but�it����led���to�fruitless�lines�of�attac��rk.�As�implemen�ted,�then,�only�one�new�subgoal�is����generated,��since�the�rst�t��rw�o��h�yp�S�otheses�will�b�e�v��reried�b�y�simplication.����!��This��:rule�is�v��rery�useful�in�pro�S�ofs�b�y�induction,�where��A�v����B��@�is��:the�induction����h��ryp�S�othesis�
��(�n�),�and��C���>�D�]��is�the�induction�goal���(�n�nS�+�1).�
If�the�induction����v��X�ariable��)�n��o�S�ccurs�linearly�in�an�exp�onen��rt,�then��C��=��X�A��ma�y�cancel,�pro�S�ducing�a����quotien��rt���without��n��in�the�exp�S�onen�t,�so��B��v�Ë��UR�D�k�ma�y�b�S�e�simple�to�pro�v�e.�Suc�h�a����situation��	o�S�ccurs�in�the�pro�of�of�the�irrationalit��ry�of��e�,�when�pro�ving�an�estimate����on���the�terms�of�an�innite�series�b��ry�induction.�A��simple�example�of�its�use����w��rould�u*b�S�e�in�the�pro�of�b��ry�induction�of�2����2�n���#���Ӽ(�n��B�+�1)!.�u*Here�w�e�try�to�deriv�e�����	l-������`����U��M.��fBeeson:�Automatic�Deriv��dDation�of�the�Irrationalit��!y�of�e�?�9�������������2����2�n��	�����]�(�n��߼+�1)!��)��2����2�n�+1��s)���(�n��+�1�+�1)!.�D�Cancelling�2����2�n�+1�����=�2����2�n��	��w��re�nd��v�\��=�2,�so�the�����new�?goal�is�2(�n���+�1)!�����(�n���+�1�+�1)!.�?Using�factorial�simplication�as�discussed����ab�S�o��rv�e�W�this�reduces�to�2�UR���(�n�~��+�1�+�1),�W�whic��rh�is�easily�v�eried.�Indeed��Weierstr��ffass�����can��pro��rv�e�2����2�n������UR�(�n����+�1)!�b��ry�induction�in�just�this�w�a�y��V.��������4.6.��/�hMathematical�2�induction������The���principle�of�mathematical�induction�is�easily�added�to�almost�an��ry�theorem-����pro��rv�er.�:�The�delicate�issues�are�the�selection�of�an�induction�v��X�ariable�and�the����decision���to�try�mathematical�induction.��Weierstr��ffass��has�rather�crude�rules�for����b�S�oth��of�these,�but�they�are�sucien��rt�for�the�estimates�needed�in�the�course�of�the����pro�S�of���of�the�irrationalit��ry�of��e�.�The�rule�of�induction�in��Weierstr��ffass��is�as�follo�ws:������;P��!���UR�)���(�c�)�;��c����n;����(�n�)�;���UR�)���(�n����+�1)���ٍ��!����xR�������
����`��c�UR���n;������)���(�n�)�������where���n��do�S�es�not�o�ccur�in��or�in��c�.�That�is,�the�basis�case�is��n�UR�=��c�铼since��c�UR���n������is��in�the�assumption�list.�Here��n��is�a�v��X�ariable�of�t��ryp�S�e���N��c��or���Z��*��.����!��This�m�form��rulation�of�the�rule�implies�an�answ�er�to�the�\delicate�issues"�men-����tioned�1�ab�S�o��rv�e.�Namely��V,�induction�on��n��will�not�b�S�e�tried�unless�there�is�an�inequal-����it��ry��gof�the�form��c�UR���n��g�in�the�assumption�list,�with��n��of�t�yp�S�e���N��/�or���Z��g�,�and�in�that����case,��the�rst�suc��rh��n��will�b�S�e�selected�as�the�induction�v��X�ariable.����!��The��usual�form��rulation�without��c�UR���n�,��namely������;P��!���UR�)���(0)����(�n�)�;������)���(�n����+�1)���ٍ��!����xR���N��
����]ش�UR�)���(�n�)�������is�uLused�in��Weierstr��ffass��when�there�is�exactly�one�v��X�ariable�of�t��ryp�S�e���N��xҼor���Z��*��in���.�����Ho��rw�ev�er,��this�pla��rys�no�role�in�the�pro�S�of�of�irrationalit�y�of��e�.��������4.7.��/�hThe�2�Zb�Y�ound�rule������The���logical,�as�opp�S�osed�to�mathematical,�heart�of�the�pro�of�is�the�in��rterpla�y����b�S�et��rw�een�E�the�t��ryp�e���Z���N�and�the�t��ryp�e���R��
Wo�.�The�simplest�principle�connecting�these�t��rw�o����t��ryp�S�es�,fis�the�principle�that�if��n��9�:���Z��1��and�,f0��<�n��then�1����n�.�That�is,�there�are�no����in��rtegers���b�S�et�w�een�0�and�1.�A���more�quan�titativ�e�form�ulation�of�this�principle�is��`؍�the��^follo��rwing�rule,���08���m�UR�:���Z��#B�0��<����h�<��1���ō����xR�������
������	l�min���4(���;����1��������)�UR��j�m����+����j������c�whic�h�w�e�call�the�\Zb�S�ound�rule".��`؍�W��Ve�j�ha��rv�e�omitted�to�write�the�an�teceden�t��since�it�is�the�same�in�b�S�oth�the����h��ryp�S�otheses��and�conclusion.����!��This�{krule�is�only�useful�once�the�quan��rtit�y�{kw�e�are�trying�to�b�S�ound�b�elo��rw�has����b�S�een�+�written�in�the�form��m���+���?B�with�+��m���:���Z���.�F��Vor�this�purp�S�ose�there�is�a�sp�ecial����simplication�Ҭrule�whic��rh�tries�to�write�its�input�in�that�form.�In�particular,�this����rule��will�break�an�innite�series�in��rto�a�nite�sum�and�an�innite�\tail"�if�it�can�b�S�e����calculated�%�(b��ry�simplication)�that�the�terms�up�to�a�certain�p�S�oin�t�are�in�tegers.����In�g�the�application�to�the�irrationalit��ry�of��e�,�the�series�has�the�form�������P����*���p�1��	U_��p�n�=0���#�:�q�n9�!�=n�!,�����
}j������`����R�9�M.��fBeeson:�Automatic�Deriv��dDation�of�the�Irrationalit��!y�of�e�<�910�������������so��!the�natural�place�to�break�the�series�is�at��n���=��q�n9�,��!since�for��n�����q��,��!w��re�ha�v�e������q�n9�!�=n�!�UR:���Z���R�.����!��Of��sin��rterest�for�the�future�are�v��X�arious�renemen�ts�of�this�rule.�If�one�w�an�ts�the����program�:kto�b�S�e�able�to�\rene�the�estimates"�as�men��rtioned�in�the�in�tro�S�duction,����one��ma��ry�wish�to�consider�rules�suc�h�as��%������T��!���m�UR�:���Z��#B�0��<�A�������h����B��X<��1��"����!����xR���B���
������=�Imin���Q��(�A;����1������B���)�UR��j�m����+�����j�������!���Here,���in�the�application�of�the�rule,��A��and��B�e�w��rould�usually�b�S�e�new�meta�v��X�ari-�����ables,��with�certain�v��X�ariables�\forbidden"�to�their�v�alues,�so�that�w��re�w�ould�b�S�e����searc��rhing�m&e.g.�for�b�S�ounds�indep�enden��rt�of��q�n9�.�This�rule�has�b�een�implemen��rted,�but����there�*�are�diculties�to�b�S�e�o��rv�ercome�*�b�efore�suc��rh�a�rule�can�b�e�made�to�w��rork����prop�S�erly��V,��and�no�suc��rh�rule�is�curren�tly�used.��"�/����5.��*�1P���oin�ts�ffof�in���terest�in�the�pro�s3of�of�irrationalit�y�of�e��e���In�$vthis�section�w��re�go�through�the�pro�S�of�step-b�y-step,�bringing�out�the�p�S�oin�ts����ab�S�out��Zthe�pro�of�that�are�of�sp�ecial�in��rterest�for�automated�deduction,�and�sho�wing����ho��rw��the�inference�rules�giv�en�ab�S�o�v�e�are�used�to�nd�the�pro�S�of.����!��The��starting�p�S�oin��rt�is�����Y���8�p�UR�:���N��
��8�q�Ë�:���N���(�q�Ë>��0��!�9�C�ܞ�(�j�e������p=q�n9�j�UR��C��=q��!����^��C�1�>�UR�0)�:���/��!���The�<logical�apparatus�strips�o�the�quan��rtiers�and�assumes��q���>�@��0�(puts�it����in�\uthe�an��rteceden�t).�\uThe�existen��rtial�v��X�ariable��C�9�b�S�ecomes�a�meta�v��X�ariable�whic�h����m��rust�8pev�en�tually�b�S�e�instan�tiated.�Putting�the�clause��C��T>�ٶ�0�last�sa�v�es�us�from�a����fruitless�w�attempt�to�pro��rv�e�w�the�inequalit��ry�with��C�1�=�UR�q�n9�,�since��C�>�UR�0�will�unify�with����the��_h��ryp�S�othesis��q��>�rʼ0.�W��Ve�then�start�to�v�erify��j�e�����p=q�n9�j�r���C��=q��!.��_Here�is�where����the�ޒ\hin��rt"�giv�en�b�y�the��q�n9�!�term�is�used:�the�inequalit�y�simplies�b�y�clearing�the����denominator.�E�The�pro��rv�er�E�is�able�to�deduce�the�side�condition�that��q�n9�!���>��0�E�since����this�~Rinequalit��ry�reduces�to��true�.�The��q�n9�!�is�then�m�ultiplied�in�to�the�absolute�v��X�alue����(for��reasons�discussed�b�S�elo��rw),�yielding�the�goal��������j�q�n9�!�e������pq��!�=q��j�UR��C�5�:���/��!���F��Vactorial�"hsimplication�is�then�used�to�reduce��q�n9�!�=q����to�(�q�>���Н�1)!.�This�is�not�as����trivial���as�it�seems,�since�this�simplication�requires��q��8>�~��0,�so�the�assumptions����m��rust���b�S�e�used.�Next�the�pro�v�er�expands��e��in�an�innite�series{but�only�after�trying����ev��rerything�Uyelse!�(Since,�as�discussed�ab�S�o�v�e,�expanding��e��in�a�series��should��b�S�e�a����last�"uresort.)�The�form�of�this�expression,�ho��rw�ev�er,�"uis�suc��rh�that�nothing�else�in�the����rep�S�ertoire���generates�a�noticeable�false�start.�(One�has�to�w��ronder,�though,�if�that����w��rould���still�b�S�e�true�if��Weierstr��ffass��con�tained�a�couple�of�h�undred�widely-v��X�aried����mathematical��inference�rules.)�This�leads�to�����������=�j�����Z��q�n9�!����j��1����������X���
�ҍ��D�n�=0���UT�1�=n�!������p�(�q�����1)!����j�������UR�C�5�:������w������`����R�9�M.��fBeeson:�Automatic�Deriv��dDation�of�the�Irrationalit��!y�of�e�<�911������������!���No��rw��6it�is�a�problem�to�get�the��q�n9�!�m�ultiplied�in�to�the�series,�since�generally�����simplication���will�pull�constan��rts��out��of�series,�rather�than�push�them�in.�This����is�b�therefore�not�p�S�erformed�b��ry�the�\simplify"�op�eration�of��Mathp��ffert�.�P��rerhaps�it����should��&b�S�e,�on�the�grounds�that�it�is�a�go�o�d�idea�to�m��rultiply�constan�ts�in�v�olving����factorial��~in��rto�series�whose�general�term�con�tains�factorial;�but�at�presen�t,�it�is�not����done.��]The�same�applies�to�the�step�men��rtioned�ab�S�o�v�e,�of�m�ultiplying�the��q�n9�!�term����in��rto���the�absolute�v��X�alue.�Both�these�steps�are�p�S�erformed�under�the�direct�con�trol�of�����Weierstr��ffass��ݼrather�than�b��ry��Mathp�ert�'s�simplication.�The�general�principle�here����seems��_to�b�S�e�that�if�y��rou�cannot�get�y�our�theorem�pro�v�ed�b�y�simplifying�as�usual,����y��rou��Mprobably�should�expand�things�that�w�ere�con�tracted,�m�ultiply�in�things����that�Uw��rere�factored�out,�etc.�A�t�an�y�rate�that�is�what��Weierstr��ffass��do�S�es;�there�is����no�Y�ne-tuning�ab�S�out�\m��rultiply�factorials�in�to�series�that�con�tain�factorials".�It����is���simply��V,�if�nothing�else�w��rorks,�push�constan�ts�in�to�absolute�v��X�alues�and�series����instead��of�factoring�them�out.�W��Ve�arriv��re�at������������;�j���������7�1������J����X���
�ҍ�����n�=0�������q�n9�!�=n�!������p�(�q�����1)!����j�������UR�C�5�:���Y��!���No��rw�Bthe�pro�v�er�starts�to�prepare�for�the�ev�en�tual�use�of�the�Zb�S�ound�rule,�b�y����breaking��the�sum�in��rto�t�w�o�parts,�separating�o�the�initial�terms�that�it�can�see����are��in��rtegers:���"&�����y�n�j������ԍ�����q��},���+Ÿ��X���
�ҍ���n�=0������q�n9�!�=n�!���+�������1�����[���X���
�ҍ��n�=�q�I{�+1���`��q��!�=n�!����p�(�q�����1)!����j�������UR�C�5�:�� c���!���Since���simplication�is�able�to�simplify�the�t��ryp�S�e�expression�that�sa�ys�the�nite����sum�Z�is�of�t��ryp�S�e���Z���P�to��true�,�the�Zb�ound�rule�is�no��rw�applied,�writing�this�inequalit�y����in��6the�form��j�m�z�+�����j�UR�>�C�ܞ�,��6where��m�UR�=�������P����*������q��	U_����n�=0���"���q�n9�!�=n�!�z���p�(�q��P���1)!��6and���h�=��UR�����P����*������1��	U_�����n�=�q�I{�+1���-6?�q��!�=n�!.����This��runies�the�meta��rv��X�ariable��C���with��min��l�(���;����1��i�����).��r(Of�course,�the�pro��rv�er��rdo�S�es����not���in��rtro�S�duce�a�new�letter����e�and�a�denition�as�w�e�do�here�for�con�v�enience�and����legibilit��ry��V.)�a�It�remains�to�v�erify�0���<��u�and�a���3-<��1.�0��<��u�is�easily�v��reried,�since����the��general�term�of�the�series�is�p�S�ositiv��re.�The�hard�part�is���h�<�UR�1.����!��The�P(pro��rv�er�attac�ks����<��1�using�the�comparison�test.�The�pro�S�of�presen�ted�in����(�13����)�w�do�S�es�not�use�a�geometric�series,�but�rather�the�series�for��e�.��Weierstr��ffass��uses����a�Rgeometric�series�as�discussed�ab�S�o��rv�e.�RThis�seems�to�b�e�an�impro��rv�emen�t�Ro�v�er�the����original��!pro�S�of��8!�Namely��V,�using�the�series�for��e��causes�t��rw�o��!problems:�First,�it�is����necessary���to�factor�out�the�rst�term��q�n9�!�=�(�q��¼+�2�1)!�UR=�1�=�(�q��+�2�1),���and�nothing�in�the����program��gwill�cause�that�to�happ�S�en;�and�it�seems�that�only�an��ad�*�ho��ffc��rule�could����force��sit�to�happ�S�en.�Second,�and�more�imp�ortan��rt,�the�estimate�that�comes�out�is�����e=�(�q�K�+�ݺ1),��Sand�it�is�not�ev��ren�true�that�this�is�b�S�ounded�b�y�1�for�all��q�n9�.�Instead,�one����has�j\to�separately�pro��rv�e�j\�e�UR<��3�and�in��rtro�S�duce�a�case�split,�treating�the�cases��q�Ë�=�1����and���q�Ë�=�UR2�separately��V.�All�this�is�left�implicit�in�(�13����),�but�nothing�in��Weierstr��ffass�����w��rould�]b�S�e�able�to�mak�e�this�case�split.�In�other�w�ords,�if�w�e�added�the�series�for��e�����to�	the�comparison�test�inference�rule�to�b�S�e�tried�b�efore�the�geometric�series,�and������ɠ�����`����R�9�M.��fBeeson:�Automatic�Deriv��dDation�of�the�Irrationalit��!y�of�e�<�912�������������also���forced�it�to�factor�out�the�rst�term�follo��rwing�(�13����),�it�w�ould�fail�to�pro�v�e������e=�(�q��+���1)�UR�<��1��and�hence�fail�to�nd�the�estimate�in�(�13����).����!��Not�y=to�w��rorry:�the�geometric�series�estimate�do�S�es�succeed.�The�geometric�com-����parison�8�series�with�ratio�2�and�rst�term�(when��n���=��q�M��+���1)�8�matc��rhing�that�of�������has��general�term�2����2��n�+�q�I{�+1��!Һ�=�(�q��+���1).�This�generates�the�subgoal����܍��%�q��+���1�UR���n��)������ō�8V�q�n9�!�����[��z�
G�
�΍�n�!�����X�������ō����2����2��n�+�q�I{�+1������[��z�'���
�΍���q��+���1�����-nn�:��e捑!���After��simplication�this�inequalit��ry�b�S�ecomes������r2�����n��q�I{��1��6B�(�q��+���1)!�UR���n�!����!��The�/%pro��rv�er�then�pro�v�es�this�b�y�induction.�Since�the�assumption��q�G��+��I1������n��is����in�cthe�an��rteceden�t,�cthe�induction�v��X�ariable�is�selected�as��n�,�and�the�basis�case�is�����n��=��q�S��+��Q1.�@�The�basis�case�then�simplies�to��true�.�The�induction�step�yields�to����the���\transitiv��recancel"�rule�discussed�ab�S�o�v�e.�Namely��V,�the�induction�step's�goal�is�������2�����n�+1��q�I{��1��&��(�q��+���1)!�UR���(�n��+�1)!����!��Regarding�
this�as�the��C�1���UR�D�[��in�the�conclusion�of�the�transitiv��recancel�rule,�and����the��induction�h��ryp�S�othesis�as�the��A�UR���B�;�term��in�that�rule,�w�e�ha�v�e��C��=��X�A�UR�=��v�Ë�=�2.����The�t>new�goal�(generated�b��ry�the�transitiv�ecancel�rule)�is�2�n�!�?����(�n�U�+�1)!,�t>whic�h����reduces��using�factorial�simplication�to�2�UR���n�瀼+�1,��hence�to�1�UR���n�,��whic��rh�follo�ws����immediately��from��q��+���1�UR���n�.����!��Since�c�w��re�w�an�t�to�use�the�comparison�test�to�establish�a�strict�inequalit�y��V,�w�e����m��rust���also�v�erify�that�strict�inequalit�y�holds�in�the�comparison�for�at�least�one����term;�Ʊ�Weierstr��ffass��c��rho�S�oses�the�second�term,�where��n�UR�=��q��k�+�a22,�Ʊand�the�inequalit�y����to��b�S�e�pro��rv�ed��is���]�������ō��e��q�n9�!������[��z�&cΟ
�΍(�q��+���2)!�����ԓ��<������ō����2����2��(�q�I{�+2)+�q��+1������[��z�8X�
�΍�
��q��+���1���������whic��rh��Vsimplies�to��true�,�completing�the�comparison�test.�Note�that�factorial����simplication,��not�just�p�S�olynomial�algebra,�is�needed�here.����!��This�completes�the�v��rerication�of�the�h�yp�S�otheses�of�the�Zb�ound�rule.�A��rt�this����p�S�oin��rt���the�meta�v��X�ariable��C�ij�is�unied�with��min��"(���;����1��������),���and�the�pro��rv�er���mo�v�es�on����to��the�last�goal,�namely�0�UR�<�C�ܞ�.��Originally�I�had�added�the�inference�rule��\�����+�!���X�F�<�URA��X�<�B���ō��!����xR��^�7���
����)��X�F�<���UR�min����(�A;���B���)�����\���But�5this�turned�out�to�b�S�e�sup�er
uous�since��Mathp��ffert��already�simplies��X�y!<������min��)�H(�A;���B���)�cto��X�+z<�9�A�l��^��X�<�9�B���.�cHere�is�another�case�where�the�line�b�S�et��rw�een����calculation��and�inference�is�blurred.����!��The�eTnal�in��rteresting�p�S�oin�t�is�that�w�e�ha�v�e�already�v�eried�b�S�oth�of�these�goals.����Y��Vet���originally�,�there�w��ras�nothing�in��Weierstr��ffass��to�enable�it�to�re-use�those����deductions.�AxIt�w��rould�simply�rep�S�eat�them!�The�general�problem�here�is,�ho�w�do����w��re�k�recognize�a�lemma�when�w�e�see�one?�What�subpro�S�ofs�should�b�e�accorded�the�����
�a������`����R�9�M.��fBeeson:�Automatic�Deriv��dDation�of�the�Irrationalit��!y�of�e�<�913�������������status�q�of�lemmas,�and�recorded,�and�referred�to�when�the�same�goal�comes�up�����again�later?�F��Vor�the�presen��rt,��Weierstr��ffass��accords�that�status�to�the�h�yp�S�otheses����of��:the�Zb�S�ound�rule,�on�the�grounds�that�almost�alw��ra�ys��:w�e�will�b�S�e�w�an�ting�to����pro��rv�e���that�the�b�S�ound�pro�duced�is�p�ositiv��re.�The�pro�v�er�k�eeps�a�list�of�lemmas;����it�records�sequen��rts�in�this�list�on�command,�and�the�rst�thing�it�do�S�es�when����trying�%�to�pro��rv�e�%�a�goal�is�to�compare�the�goal�to�the�list�of�lemmas.�A��rt�presen�t,����only��0the�subgoals�generated�b��ry�the�Zb�S�ound�rule�are�recorded�as�lemmas.�This����mec��rhanism�z�p�S�ermits�the�pro�of�to�complete�in�a�few�lines�after�the�Zb�ound�rule����inference��completes,�rather�than�nearly�doubling�its�length.��#�4���References��34������[1]���.'Beeson,�E�M.:�Automatic�Generation�of�epsilon-delta�Pro�S�ofs�of�Con��rtin�uit�y��V,�E�in:����.'Calmen��rt,���J.,�and�Plaza,�J.�(eds.)��A��2rticial�A�Intel���ligenc��ffe�and�Symb�olic�Com-����.'putation�,�r�Lecture�Notes�in�Articial�In��rtelligence��1476�,�pp.�67{83,�Springer-����.'V��Verlag,��Berlin/Heidelb�S�erg/New�Y�ork�(1998).��L͍����[2]���.'Beeson,�r5M.:�Unication�in�lam��rb�S�da-calculus�with�if-then-else,�in:�Kirc�h-����.'ner�L
and�Kirc��rhner�(eds.):��A��2utomate��ffd�x	De�duction{CADE-15,�Pr�o�c�e�e�dings�of����.'the��15th�International�Confer��ffenc�e��on�A��2utomate��ffd�De�duction,�Lindau,�Ger-����.'many,�xPJuly�1988�,�Spp.�103{118.�Lecture�Notes�in�Articial�In��rtelligence��1421�,����.'Springer-V��Verlag,��Berlin/Heidelb�S�erg/New�Y�ork�(1998).�������[3]���.'Beeson,��?M.:�Logic�and�computation�in��Mathp��ffert�$D�:�an�exp�S�ert�system�for�learn-����.'ing�5Qmathematics,�in:�Kaltofen,�E.,�and�W��Vatt,�S.�M.�(eds.),��Computers�w�and����.'Mathematics�,�lppp.�202{214,�Springer-V��Verlag,�Berlin/Heidelb�S�erg/New�Y�ork����.'(1989).�������[4]���.'Beeson,��M.:�Design�principles�of�Mathp�S�ert:�soft��rw�are��to�supp�ort�education����.'in��algebra�and�calculus,�in:�Ka��jler,�N.�(ed.)��Computer-Human���Inter��ffaction����.'in�35Symb��ffolic�Computation�,��pp.�89-115,�Springer-V��Verlag,�Wien�(1996).�������[5]���.'Beeson,�n�M.:�Using�nonstandard�analysis�to�ensure�the�correctness�of�sym-����.'b�S�olic�ULcomputations,��International���Journal�of�F���oundations�of�Computer�Sci-����.'enc��ffe���6�(3)�(1995)�299{338.�������[6]���.'Beeson,�)M.:�Some�applications�of�Gen��rtzen's�pro�S�of�theory�in�automated����.'deduction,�r�in:�Shro�S�eder-Heister,�P��V.,��Extensions�rYof�L��ffo�gic�rYPr�o�gr�amming�,����.'Lecture�LNotes�in�Computer�Science��475�,�pp.�101{156,�Springer-V��Verlag,����.'Berlin/Heidelb�S�erg/New��Y��Vork�(1991).�������[7]���.'Bo��ry�er,�)�R.,�and�Mo�S�ore,�J.:��A�CpComputational�C�L��ffo�gic�Handb�o�ok�,�)�Academic����.'Press,��San�Diego�(1988).�������[8]���.'Clark��re,��'E.,�and�Zhao,�X.:�Analytica:�a�theorem�pro�v�er�in�Mathematica,����.'in:�9Kapur,�D.�(ed.),��A��2utomate��ffd�IQDe�duction:�CADE-11�-�Pr�o�c.�of�the�11th������������`����R�9�M.��fBeeson:�Automatic�Deriv��dDation�of�the�Irrationalit��!y�of�e�<�914������������.'�International�o�Confer��ffenc�e�on�A��2utomate�d�De�duction�,�,�pp.�761{765,�Springer-�����.'V��Verlag,��Berlin/Heidelb�S�erg�(1992).��������[9]���.'Coste,�g{M.,�and�Ro��ry��V,�M.�F.:�Thom's�lemma,�the�co�S�ding�of�real�algebraic����.'n��rum�b�S�ers,��*and�the�computation�of�the�top�ology�of�semi-algebraic�sets,�in:����.'Arnon,���D.�S.,�and�Buc��rh�b�S�erger,���B.,��A��2lgorithms�5+in�R��ffe�al�5+A�lgebr�aic�5+Ge�ometry�,����.'Academic��Press,�London�(1988).������[10]���.'Do��rw�ek,��G.,�Hardin,�Th.,�and�Kirc��rhner,�C.,�Theorem�pro�ving�mo�S�dulo,�Rap-����.'p�S�ort��de�Rec��rherc�he��3400,�INRIA�(1998).������[11]���.'Harrison,��)J.,�and�Thery��V,�L.:�Extending�the�HOL���theorem�pro��rv�er��)with�a����.'computer�=�algebra�system�to�reason�ab�S�out�the�reals,�in��Higher�\Or��ffder�L�o�gic����.'The��ffor�em�ϹPr�oving�and�its�Applic�ations:�6th�International�Workshop,�HUG����.''93�,��Vpp.�174{184,�Lecture�Notes�in�Computer�Science��780�,�Springer-V��Verlag,����.'Berlin/Heidelb�S�erg/New��Y��Vork�(1993).������[12]���.'McCune,�
��W.:�Otter�2.0,�in:�Stic��rk�el,�
��M.�E.�(ed.),��10th�
a�Interna-����.'tional�+�Confer��ffenc�e�on�A��2utomate�d�De�duction��:�pp.�663{664,�Springer-V��Verlag,����.'Berlin/Heidelb�S�erg/New��Y��Vork�(1990).������[13]���.'Siegel,��HC.�L.,��T���r��ffansc�endental�.DNumb�ers�,��HAnnals�of�Mathematics�Studies��16�,����.'Princeton��Univ��rersit�y�Press,�Princeton,�New�Jersey�(1949).��,����A.��.�TApp�s3endix:�ffV���ferbatim�cop���y�of�the�output������Weierstr��ffass�:O�pro�S�duces�an�in��rternal�pro�of�ob��ject,�whic��rh�can�b�e�view��red�and�sa�v�ed�in����either��\trace�view"�or�\pro�S�of�tree�view".�There�is�an�option�to�sa��rv�e��the�trace�view����as�5�a�T���
���UE���!X�le.�Here�is�the�trace-view�le,�as�pro�S�duced�and�t��ryp�eset�b��ry�W��Veierstrass:����!��The��goal�is����a!1�8��p�UR�:��N��}s;�����8��q�Ë�:�UR�N��;�;���9�C�5�;�������
�����	*��0�UR�<�q�Ë�!�����3,�����38�����������UR�e����������ō����p���۟[��z�孟
�΍� �q���������3,��
�����38��
������
�������L
���������C�����[��z�	D��
�΍�.�q�n9�!�����c�;����0��<�C����ܞ��
����������ō�-�T��Vrying����tW�8��q�Ë�:�UR�N��;�;�����9�C�5�;�������
�����	*��0�UR�<�q�Ë�!�����3,�����38�����������UR�e����������ō����p���۟[��z�孟
�΍� �q���������3,��
�����38��
������
�������L
���������C�����[��z�	D��
�΍�.�q�n9�!�����c�;����0��<�C����ܞ��
��������Cō�9?�T��Vrying�����k��9�C�5�;���������
�����	*��0�UR�<�q�Ë�!�����3,�����38�����������UR�e����������ō����p���۟[��z�孟
�΍� �q���������3,��
�����38��
������
�������L
���������C�����[��z�	D��
�΍�.�q�n9�!�����c�;����0��<�C����ܞ��
���������D��T��Vrying��*P8��0�UR�<�q�Ë�!�����3,�����38�����������UR�e�������ō�33�p��33�[��z�孟
�΍� �q���������3,��L���38��L����L������e������������min�����͟�
�������Hy����P����*���+�&�1��	U_��+�&�k�6��=�q�I{�+1��������w��J%�q�I{�!��I�۟���z�����k�6��!������R�;�����1�����������P����*���UU�1��	U_��UU�k�6��=�q�I{�+1��������w��-�T�q�I{�!��-:
����z�����k�6��!���������IM<��
���������	�T�z��>�
�΍�L+H�q�n9�!���������;����0��<����min������ꚟ�
���������&��1�����"�\���X���'؍�F�k�6��=�q�I{�+1��������ō�;���q�n9�!��;y��[��z�	��
�΍�k�g�!������FtE�;���1���������f�1���������X���'؍��k�6��=�q�I{�+1��������ō� /�q�n9�!�� ]�[��z�	��
�΍�k�g�!���������>頟�
����������������`����R�9�M.��fBeeson:�Automatic�Deriv��dDation�of�the�Irrationalit��!y�of�e�<�915������������P�ؼAssuming��0�UR�<�q�����P�ؼT��Vrying��'AH������3,��"E#���38��"E#����"E#�����&E#�e����������ō����p���۟[��z�孟
�΍� �q���������3,��
�����38��
������
�������L
������������min�����͟�
�������Hy����P����*���+�&�1��	U_��+�&�k�6��=�q�I{�+1��������w��J%�q�I{�!��I�۟���z�����k�6��!������R�;�����1���������P����*���UU�1��	U_��UU�k�6��=�q�I{�+1��������w��-�T�q�I{�!��-:
����z�����k�6��!���������IM<��
���������	�T�z��>�
�΍�L+H�q�n9�!���������;����0�UR�<����min������ꚟ�
���������&��1�����"�\���X���'؍�F�k�6��=�q�I{�+1��������ō�;���q�n9�!��;y��[��z�	��
�΍�k�g�!������FtE�;���1���������f�1���������X���'؍��k�6��=�q�I{�+1��������ō� /�q�n9�!�� ]�[��z�	��
�΍�k�g�!���������>頟�
������#
���\мT��Vrying��'AH������3,��u�b���38��u�b����u�b�����y�b�e����������ō����p���۟[��z�孟
�΍� �q���������3,��
�����38��
������
�������L
������������min�����͟�
�������Hy����P����*���+�&�1��	U_��+�&�k�6��=�q�I{�+1��������w��J%�q�I{�!��I�۟���z�����k�6��!������R�;�����1���������P����*���UU�1��	U_��UU�k�6��=�q�I{�+1��������w��-�T�q�I{�!��-:
����z�����k�6��!���������IM<��
���������	�T�z��>�
�΍�L+H�q�n9�!���������\�Simplifying,��it�w��rould�suce�to�pro�v�e���b����}xmin�����������
�����������*�1������J����X���'؍���l�k�6��=�q�I{�+1��������ō�����q�n9�!���A!�[��z�	��
�΍�k�g�!�������;k�;�����1���������f�1���������X���'؍��k�6��=�q�I{�+1��������ō� /�q�n9�!�� ]�[��z�	��
�΍�k�g�!���������>頟�
�����Ii�������3,��UR���38��UR����UR�����UR�e����������ō����p���۟[��z�孟
�΍� �q���������3,��
�����38��
������
����������q�n9�!��#/���\�Since��0�UR���q�n9�!�����\�it��w��rould�suce�to�pro�v�e������u��min����������
�����������~�1������֟��X���'؍�����k�6��=�q�I{�+1��������ō���G�q�n9�!���u�[��z�	��
�΍�k�g�!���������;�����1���������f�1���������X���'؍��k�6��=�q�I{�+1��������ō� /�q�n9�!�� ]�[��z�	��
�΍�k�g�!���������>頟�
�����Ii�������3,��UR���38��UR����UR��������UR��
������e����������ō����p���۟[��z�孟
�΍� �q��������
����
�����!g�q�n9�!�����3,����38������������\мBy��the�distributiv��re�la�w�����\�it��w��rould�suce�to�pro�v�e������x��min�������>��
����������Aq�1�������ɟ��X���'؍��h��k�6��=�q�I{�+1��������ō��=:�q�n9�!����h�[��z�	��
�΍�k�g�!�������Dz�;�����1���������f�1���������X���'؍��k�6��=�q�I{�+1��������ō� /�q�n9�!�� ]�[��z�	��
�΍�k�g�!���������>頟�
�����Ii�������3,��UR���38��UR����UR�����UR�eq�n9�!����������ō����pq��!���۟[��z���
�΍����q���������3,���-���38���-�����-������#
���\мUsing���㍍�����ō��t��pq�n9�!���t��[��z���
�΍����q�������Z�=�UR�p�(�q�������1)!��[��\�it��w��rould�suce�to�pro�v�e������k��min������DΟ�
����������H8�1������ݐ���X���'؍��oz�k�6��=�q�I{�+1��������ō��D�q�n9�!����/�[��z�	��
�΍�k�g�!��������y�;�����1���������f�1���������X���'؍��k�6��=�q�I{�+1��������ō� /�q�n9�!�� ]�[��z�	��
�΍�k�g�!���������>頟�
�����Ii���URj�eq�n9�!������p�(�q�����1)!�j����\мUsing��the�series�denition�of�e�����\�it��w��rould�suce�to�pro�v�e������V��min������jg��
���������zj��1�����u����X���'؍�q���k�6��=�q�I{�+1��������ō��fQ�q�n9�!�����[��z�	��
�΍�k�g�!����������;�����1���������f�1���������X���'؍��k�6��=�q�I{�+1��������ō� /�q�n9�!�� ]�[��z�	��
�΍�k�g�!���������>頟�
�����Ii�������3,��UR���38��UR����UR��������UR��
���������1���������X���'؍�j��k�6��=0��������ō�$��1��#��[��z�	��
�΍�k�g�!��������.џ�
�����5-}�q�n9�!������p�(�q�����1)!�����3,����38������������\мSimplifying,��it�w��rould�suce�to�pro�v�e������\��min������p�˟�
�����������5�1�����|*����X���'؍�w�w�k�6��=�q�I{�+1��������ō�����q�n9�!���!,�[��z�	��
�΍�k�g�!�������v�;�����1���������f�1���������X���'؍��k�6��=�q�I{�+1��������ō� /�q�n9�!�� ]�[��z�	��
�΍�k�g�!���������>頟�
�����Ii�������3,��UR���38��UR����UR�����UR�q�n9�!����j��1����������X���'؍���k�6��=0��������ō�|�1�����[��z�	��
�΍�k�g�!�����$-y�����p�(�q�����1)!�����3,����38�������������ժ������`����R�9�M.��fBeeson:�Automatic�Deriv��dDation�of�the�Irrationalit��!y�of�e�<�916������������\мPushing��a�constan��rt�in�to�the�innite�series,�����\�it��w��rould�suce�to�pro�v�e�� �>����bp;min������v���
������������1�������E���X���'؍�}0/�k�6��=�q�I{�+1��������ō����q�n9�!�����[��z�	��
�΍�k�g�!��������.�;�����1���������f�1���������X���'؍��k�6��=�q�I{�+1��������ō� /�q�n9�!�� ]�[��z�	��
�΍�k�g�!���������>頟�
�����Ii�������3,��UR���38��UR����UR�����������1�����UR���X���'؍�?��k�6��=0��������ō�M��q�n9�!���۟[��z�	��
�΍�k�g�!�����)�������p�(�q�����1)!�����3,����38����������$c��\мSince��s����h�k��o��UR�q�Ë�!������ō��W�q�n9�!�����[��z�	��
�΍�k�g�!������!:��Z��2��\мw��re��can�separate�the�initial�in�teger-v��X�alued�terms;�����\�it��w��rould�suce�to�pro�v�e��!pf����GPmin������Z����
���������j��1�����f2Z���X���'؍�a�D�k�6��=�q�I{�+1��������ō����q�n9�!��(��[��z�	��
�΍�k�g�!�������#C�;�����1���������f�1���������X���'؍��k�6��=�q�I{�+1��������ō� /�q�n9�!�� ]�[��z�	��
�΍�k�g�!���������>頟�
�����Ii�������3,��UR���38��UR����UR�������ԍ�
��q��},���UR���X���'؍�?��k�6��=0��������ō�M��q�n9�!���۟[��z�	��
�΍�k�g�!�����)��+�����f�1���������X���'؍����k�6��=�q�I{�+1��������ō� /�q�n9�!�� ]�[��z�	��
�΍�k�g�!�����-�O�����p�(�q�����1)!�����3,����38����������$7���\мSimplifying,��it�w��rould�suce�to�pro�v�e��!pf����Br�min������Vԟ�
���������f>�1�����a�����X���'؍�]2��k�6��=�q�I{�+1��������ō�{�q�n9�!��z�5�[��z�	��
�΍�k�g�!���������;�����1���������f�1���������X���'؍��k�6��=�q�I{�+1��������ō� /�q�n9�!�� ]�[��z�	��
�΍�k�g�!���������>頟�
�����Ii�������3,��UR���38��UR����UR�����UR��p�(�q�������1)!�+�����f�1���������X���'؍��k�6��=�q�I{�+1��������ō� /�q�n9�!�� ]�[��z�	��
�΍�k�g�!�����-�O+���ԍ�	J��q��},������X���'؍��M�k�6��=0��������ō���q�n9�!��31�[��z�	��
�΍�k�g�!���������3,��"-{���38��"-{����"-{������$c��\мSince��"&������p�(�q�������1)!�+���ԍ�	J��q��},������X���'؍��M�k�6��=0��������ō���q�n9�!��31�[��z�	��
�΍�k�g�!�������r��\�is��an�in��rteger�����\�it��suces�to�pro��rv�e�� �>���m0�UR�<����.�1������h���X���'؍��k�6��=�q�I{�+1��������ō�!)��q�n9�!�� ��[��z�	��
�΍�k�g�!������$7���\�and�����������+�1������g����X���'؍��m�k�6��=�q�I{�+1��������ō�����q�n9�!���^"�[��z�	��
�΍�k�g�!������<�UR�1�����h?�T��Vrying���R���m0�UR�<����.�1������h���X���'؍��k�6��=�q�I{�+1��������ō�!)��q�n9�!�� ��[��z�	��
�΍�k�g�!�������"��h?�It��w��rould�suce�to�pro�v�e�the�general�term�is�p�S�ositiv�e.����s��T��Vrying���ύ��.�8�k�g;���������
�����	*��q��+���1�UR���k��o�!��0��<������ō��Wq�n9�!�����[��z�	��
�΍�k��!���������ϟ�
�������g]�����T��Vrying�����|��q��+���1�UR���k��o�!��0��<������ō��Wq�n9�!�����[��z�	��
�΍�k�g�!���������笠�����`����R�9�M.��fBeeson:�Automatic�Deriv��dDation�of�the�Irrationalit��!y�of�e�<�917���������������Assuming���q��+���1�UR���k��������T��Vrying���㍒�.0�UR�<������ō��Wq�n9�!�����[��z�	��
�΍�k�g�!�����������Hey��V,��that�simplies�to�true!������Success������Disc��rharging������Success����s��Success����h?�Success����h?�Success:��that�completes�the�pro�S�of�of��"�R���m0�UR�<����.�1������h���X���'؍��k�6��=�q�I{�+1��������ō�!)��q�n9�!�� ��[��z�	��
�΍�k�g�!������&>���h?�Recording��this�result�as�Lemma�1����h?�T��Vrying���R�������+�1������g����X���'؍��m�k�6��=�q�I{�+1��������ō�����q�n9�!���^"�[��z�	��
�΍�k�g�!������<�UR�1�� ����h?�T��Vry��comparison�with�an�easier�series.����s��T��Vrying�� 8⍑gy��8��k��o�:�UR�Z��Y-;���������
�����	*��q��+���1�UR���k��o�!�����3,�����38����������������ō��W�q�n9�!�����[��z�	��
�΍�k�g�!���������3,�������38����������������!�������ō�F��q�n9�!�����[��z�&cΟ
�΍(�q��+���1)!��������,���
����������ō�4}e�1��4}e�[��z���
�΍2��������;����
������B�@���k�6���q�I{��1�����]lğ�
������� nq�����T��Vrying�����H�q��+���1�UR���k��o�!�����3,�����38����������������ō��W�q�n9�!�����[��z�	��
�΍�k�g�!���������3,�������38����������������!�������ō�F��q�n9�!�����[��z�&cΟ
�΍(�q��+���1)!��������,���
����������ō�4}e�1��4}e�[��z���
�΍2��������;����
������B�@���k�6���q�I{��1��� �������Assuming���q��+���1�UR���k��������T��Vrying��������3,���݋���38���݋�����݋����������ō�����q�n9�!�����[��z�	��
�΍�k�g�!���������3,������38��������������`Z�������ō�F��q�n9�!�����[��z�&cΟ
�΍(�q��+���1)!��������,���
����������ō�4}e�1��4}e�[��z���
�΍2��������;����
������B�@���k�6���q�I{��1��������Simplifying,��it�w��rould�suce�to�pro�v�e��������2�����k�6���q�I{��1�����(�q��+���1)!�UR���k�g�!������T��Vrying��mathematical�induction�on��k������?��The��basis�case�is�when��k��o�=�UR�q��+���1�������Hey��V,��that�simplies�to�true!�������That��completes�the�basis�case.�����?�No��rw��for�the�induction�step.������3������`����R�9�M.��fBeeson:�Automatic�Deriv��dDation�of�the�Irrationalit��!y�of�e�<�918�������������?��Assume��the�induction�h��ryp�S�othesis��J퍒���2�����k�6���q�I{��1�����(�q��+���1)!�UR���k�g�!�����?�W��Ve��m��rust�pro�v�e�����ud2�����k�6��+1��q�I{��1��%��(�q��+���1)!�UR���(�k�ż+�1)!�������Simplifying,��it�w��rould�suce�to�pro�v�e�����Q�2�����k�6���q����(�q��+���1)!�UR���(�k�ż+�1)!�������In��view�of�������2�����k�6���q����(�q��+���1)!�UR=�2�����k�6���q�I{��1�����(�q��+���1)!2�������and��the�assumption�������2�����k�6���q�I{��1�����(�q��+���1)!�UR���k�g�!�������it��w��rould�suce�to�pro�v�e�2�k�g�!�UR���(�k�ż+���1)!��������Hey��V,��that�simplies�to�true!�������Success�������That��completes�the�induction�step.�����?�Induction��completed�successfully��V.������Success������Disc��rharging������Success����s��Success����s��T��Vrying�� �?�������I�1������Y����X���'؍����k�6��=�q�I{�+1��������ō��n�q�n9�!���P@�[��z�&cΟ
�΍(�q��+���1)!����������A��
����������ō��E �1���E �[��z���
�΍2���������XO��
������������k�6���q�I{��1������UR�1��$	���s��Summing��the�geometric�series,����s��it��w��rould�suce�to�pro�v�e��*M�������������w���ٵ�q�I{�!���;����z��f���(�q�I{�+1)!���������:ԟ�
���������Fu��Ϙ��1��Ϙ�����z�@���2�����������
�������6����q�I{�+1��q���1������	�T�z�TK��Q����v�1���������Fu����1���۟���z�@���2�������������UR�1���	��s��Hey��V,��that�simplies�to�true!����s��Success����s��W��Ve��m��rust�pro�v�e�strict�inequalit�y�for�at�least�one�term.����s��W��Ve��try�the�term�with��k��o�=�UR�q��+���1�+�1����s��T��Vrying���i������3,���{���38���{�����{����������ō��F�q�n9�!�����[��z�:���
�΍(�q��+���1�+�1)!���������3,���	����38���	������	�������^��<������ō�F�q�n9�!�����[��z�&cΟ
�΍(�q��+���1)!��������,���
����������ō�4}e�1��4}e�[��z���
�΍2��������;����
������B�@���q�I{�+1+1��q���1�������8������`����R�9�M.��fBeeson:�Automatic�Deriv��dDation�of�the�Irrationalit��!y�of�e�<�919������������s���Hey��V,��that�simplies�to�true!�����s��Success��of�comparison�test����h?�Success:��that�completes�the�pro�S�of�of��"�䍍�����+�1������g����X���'؍��m�k�6��=�q�I{�+1��������ō�����q�n9�!���^"�[��z�	��
�΍�k�g�!������<�UR�1��&/*��h?�Recording��this�result�as�Lemma�2����\�Success����\�Success����\�T��Vrying������0�UR�<����min������ꚟ�
���������&��1�����"�\���X���'؍�F�k�6��=�q�I{�+1��������ō�;���q�n9�!��;y��[��z�	��
�΍�k�g�!������FtE�;�����1���������f�1���������X���'؍��k�6��=�q�I{�+1��������ō� /�q�n9�!�� ]�[��z�	��
�΍�k�g�!���������>頟�
������&/*��\мSimplifying,��it�w��rould�suce�to�pro�v�e������0�UR�<����.�1������h���X���'؍��k�6��=�q�I{�+1��������ō�!)��q�n9�!�� ��[��z�	��
�΍�k�g�!�����+�Q�;����0��<��1���������f�1���������X���'؍��k�6��=�q�I{�+1��������ō� /�q�n9�!�� ]�[��z�	��
�΍�k�g�!������&
	��h?�T��Vrying���R���m0�UR�<����.�1������h���X���'؍��k�6��=�q�I{�+1��������ō�!)��q�n9�!�� ��[��z�	��
�΍�k�g�!������ ����h?�This��has�b�S�een�pro��rv�ed��in�Lemma�1����h?�Success����h?�T��Vrying������0�UR�<��1���������f�1���������X���'؍��k�6��=�q�I{�+1��������ō� /�q�n9�!�� ]�[��z�	��
�΍�k�g�!������ ����h?�Simplifying,��it�w��rould�suce�to�pro�v�e��"�䍍�����+�1������g����X���'؍��m�k�6��=�q�I{�+1��������ō�����q�n9�!���^"�[��z�	��
�΍�k�g�!������<�UR�1��&/*��h?�This��has�b�S�een�pro��rv�ed��in�Lemma�2����h?�Success����\�Success����P��Success����P��Disc��rharging����D��Success����9?�Success����-�Success����!��Success.��That�completes�the�pro�S�of.���������;�����5�"V
�3
cmbx10�4F
C�ff
cmbxti10�3���@cmti12�2��N�ffcmbx12�-�b>
�3
cmmi10�,K�`y
�3
cmr10�+��N�cmbx12�&�':

cmti10���u
cmex10��K�cmsy8�!",�
cmsy10��2cmmi8���g�cmmi12�|{Ycmr8��-�
cmcsc10���N�G�cmbx12�X�Qcmr12�O!�cmsy7�K�`y

cmr10�������

Sindbad File Manager Version 1.0, Coded By Sindbad EG ~ The Terrorists