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/LambdaLogic.dvi

����;� TeX output 2010.10.30:1423������y�����?�������9�D��tG�G�cmr17�Lam��qb�s�da�7tLogic����������U�X�Qcmr12�Mic��rhael��Beeson������2��K�cmsy8����������������Octob�S�er��30,�2010��.���!K�t�:		cmbx9�Abstract���э�d��o���		cmr9�Lam��9b�A�da�&�logic�is�the�union�of�&�rst�order�logic�and�lam�b�A�da�calculus.����WThe���purp�A�ose�of�this���note�is�to�giv��9e�a�complete�and�precise�denition�of����Wthe���syn��9tax���of�lam�b�A�da�logic,��and�explain�its���relations�to�rst�order�logic����Wand�Tto�simple�t��9yp�A�e�theory��:�.�pThis�is�a�revised�and�expanded�v�ersion�of�[�6����].��!č��>���N�ffcmbx12�1��VL�In���tro�s3duction�����>�K�`y

cmr10�The���t���w�en�tieth�cen�tury�sa�w�the���
o�w�ering�of�rst�order�logic,��%and�the�in�v�en�tion����>and���dev���elopmen�t���of�the�lam���b�Gda�calculus.��-When�the�lam���b�Gda�calculus�w���as�rst����>dev���elop�Ged,�{Lits�@�creator�@�(Alonzo�Ch�urc�h)�in�tended�it�as�a�foundational�system,����>i.e.,�1\one�[in�Zwhic���h�mathematics�could�b�Ge�dev���elop�ed�and�Zwith�the�aid�of�whic���h����>the�6Yfoundations�6Zof�mathematics�could�b�Ge�studied.�gsHis�rst�theories�w���ere�incon-����>sisten���t,��just��as�the�rst��set�theories�had�b�Geen�at�the�op�Gening�of�the�t���w�en�tieth����>cen���tury��*�.��QMo�Gdications�/-of�/.these�inconsisten�t�/.theories�nev�er�/.ac�hiev�ed�the�/.fame����>that�H�mo�Gdications�H�of�the�inconsisten���t�set�theories�did.�m�Instead,�K7rst�order�logic����>came�	cto�	bb�Ge�the�to�Gol�of�c���hoice�for�formalizing�mathematics,��and�lam���b�Gda�calculus����>is�UUno���w�considered�as�one�of�sev�eral�to�Gols�for�analyzing�the�notion�of�algorithm.����MThe��pp�Goin���t�of�view��qunderlying�lam�b�Gda�logic�is�that�lam�b�Gda��qcalculus�is�a�go�o�d����>to�Gol�opfor�oqrepresen���ting�the�notion�of�function,�u�not�only�the�notion�of�computable����>function.���First-order�jlogic�jcan�treat�functions�b���y�in�tro�Gducing�jfunction�sym�b�Gols����>for���particular���functions,���but�then�there�is�no�w���a�y���to�construct�other�functions�b���y����>abstraction��Ror�recursion.�<qOf��Scourse,��Sone�can�consider�set�theory�as�a�sp�Gecial�case����>of�܋rst�order�܊logic,��and�dene�functions�in�set�theory�as�univ��q�alen���t�functions,��but����>this���requires���building�up�a�lot�of�formal�mac���hinery��*�,��Oand�has�other�disadv��q�an���tages����>as�fcw���ell.�"!It�is�fbnatural�to�consider�com���bining�lam�b�Gda�fccalculus�with�logic.�"!That�w���as����>done��ulong��tago�in�the�case�of�t���yp�Ged�logics;��kfor�example�G����odel's�theory�T��Ahad�what����>amoun���ted��Zto��[the�abilit�y��[to�dene�functions�b���y�lam�b�Gda-abstraction.�X�But�t�yp�Ged����>lam���b�Gda��"calculus��#lac�ks�the�full�p�Go�w�er��#of�the�un�t�yp�Ged�(ordinary)��#lam�b�Gda�calculus,����>as�UUthere�is�no�xed-p�Goin���t�theorem�to�supp�ort�arbitrary�recursiv���e�denitions.����MIn��this�pap�Ger,�χw���e�com�bine�ordinary�rst�order�lam�b�Gda�calculus�with�ordinary����>rst��order��logic�to�obtain�systems�w���e�collectiv�ely�refer�to��as�lam�b�Gda�logic.�?�W��*�e��>�X-�ff��v�	@����
����-:�q�%cmsy6����L��|{Ycmr8�Researc�Îh��Xsupp�<rorted�b�y�NSF�gran�t�n�um�b�<rer�CCR-0204362.�������1����*�y�����?������>�are��snot�the�rst��rto�dene�or�study�similar�systems.����^��ٓ�Rcmr7�1���
���The�applicativ���e�theories����>prop�Gosed���b���y���F��*�eferman�in�[�7��]�are�similar�in�concept.��vThey�are,��ho���w�ev�er,�dier-����>en���t��Rin�some�tec�hnical�details�that�are��Qimp�Gortan�t�for�the�theorems�pro�v�ed�here.����>Lam���b�Gda��logic��is�also�related�to�the�systems�of�illativ���e�com�binatory��logic�studied����>in�$�[�2��],�X�[�3��],�but�$�these�$�are�stronger�than�lam���b�Gda�logic.��SAs�far�as�w���e�kno�w,�X�the����>systems�UUdened�in�this�pap�Ger�ha���v�e�UUnot�b�een�studied�b�efore.����MBoth��<ordinary�and�lam���b�Gda�logic�can�b�e�mo�died�to�allo���w�\undened�terms".����>In�r�the�r�con���text�of�ordinary�logic�this�has�b�Geen�studied�in�[�9��],���[�5��],�[�4��].�ɆIn�r�the����>con���text��Aof�applicativ�e��@theories,�<[�4��]�dened�and�studied�\partial�com���binatory����>algebras";�Ԇbut��vapplication�in�the��
�b>

cmmi10��-calculus�is�alw���a�ys��vtotal.�	p)Moggi�[�8��]�w���as����>apparen���tly���the��rst�to�publish�a�denition�of�partial�lam���b�Gda�calculus;��see�[�8��]�for����>a��-thorough��,discussion�of�dieren���t�v�ersions��,of�partial�lam���b�Gda�calculus�and�partial����>com���binatory�UUlogic.�� 厍��>�2��VL�Syn���tax�����>�As���is�usual�in�formalizing�rst-order���logic,���w���e�supp�Gose���giv�en�an�innite�list�of����>v��q�ariables.��]W��*�e�"�also�supp�Gose�giv���en�an�"�innite�list�of�predicate�sym�b�Gols�and�an����>innite��klist��jof�function�sym���b�Gols.�OyUnlik�e��kin�rst-order�logic�w���e�require�that�ev���ery����>predicate��]sym���b�Gol�is�also�a��^function�sym�b�Gol.�.W��*�e�do��^not�require�that�the�predicate����>and�J�function�J�sym���b�Gols�come�with�a�prescrib�Ged�arit���y�(although�seman���tically��*�,�L�if��f����>�o�Gccurs��with�dieren���t��arities,���the�o�ccurrences��with�dieren���t�arities��will�b�e�treated����>as�:�if�they�w���ere�dieren�t�:�sym�b�Gols).�h�W��*�e�supp�ose�giv���en�:�an�innite�list�of�constan�t����>sym���b�Gols,��6disjoin�t�O�from�O�the�list�of�v��q�ariables.�`�The�particular�sym���b�Gol���':

cmti10�lamb��}'da��is����>not�dYa�dXfunction�or�predicate�sym���b�Gol,�hand�the�list�of�function�sym���b�Gols�con�tains�dYa����>distinguished�UUelemen���t��Ap�.����MIf��someone��wishes�a�more�precise�sp�Gecication�of�this�basic�apparatus,�Kw���e����>could�UUtak���e�the�follo�wing�completely�precise�denition:������>���<x

cmtt10�digit:=�?�0-9.����>variable:=�?�[U-Z]digit*�|�[u-z]digit*.����>constant:=�?�[A-T]digit*�|�[a-t]digit*�|�true�|�false.����>logicalSymbol�?�:=�and�|�or�|�implies�|�neg�|�=.����>functionSymbol:-�?�Ap�|�[a-t]digit*�|�[A-T]digit*�|�logicalSymbol.����>predicateSymbol:-�?�Ap�|�[A-T]digit*�|�logicalSymbol.���J��M�Here�UUis�a�grammar�dening�the�syn���tax�of�lam�b�Gda�logic.������>�binder:=�?�lambda�|�forall�|�exists.����>functionTerm:=�?�functionSymbol(term�{,term}*).����>term:=�?�variable�|�constant�|�functionTerm�|�binder(variable,�term).����>atomicFormula:=�?�predicateSymbol(term�{,�term}*)�|�true�|�false.����>formula:=�?�atomicFormula�|�compoundFormula.��>�8��ff��v�	J=�����"5��-:��Aa�cmr6�1����LܻF��J�or�u
example,���John�u	McCarth�Îy�told�me�that�he�lectured�on�suc�Îh�systems�y�Îears�ago,���but��	��nev�Îer��Xpublished�an�ything.�������2����
ޠy�����?������>�compoundFormula:-�
�connective(formula,formula)�?�|�������neg(formula)�?�|�������quantifier(variable,formula).����>connective:-�?�and,�or,�implies.����>quantifier:-�?�forall�|�exists.����M�Note�I;that�I:b�Gecause�predicate�sym���b�Gols�are�also�function�sym���b�Gols�(here�just����>called��9\sym���b�Gols"),�2and��:�true��and��false��are�constan���ts,�2ev�ery��9form�ula�is��:also�a����>term.�K�Th���us,���for���example,����l�2`ambda�(�x;���x�V�+�V�y�"�=���y���+��x�)���is�a�term,���since��x��+�V�y�"�=���y���+��x����>�is�K�a�K�term,�M�since�=�is�a�function�sym���b�Gol.�n�Hence��Ap�(�l�2`ambda�(�x;���x�&1�+�&2�y�"�=���y��
�+��x�)�;�z�p��)����>is��6a�form���ula,���since��Ap��is�a�predicate�sym�b�Gol�and��l�2`ambda�(�x;���x�Wv�+��y�oh�=���y��O�+��x�)��6is�a����>term.�S~There��{will��zalso�b�Ge�form���ulas�that�w���ould�not�o�Gccur�in�rst-order�logic,��suc���h����>as�UU�f���(�f�or�Gal�2`l��(�x;���a�))��=��b��(where��a��and��b��are�constan���ts).����MOn�~Dthe�other�hand,��not�ev���ery�term�is�a�form�ula.��F��*�or�example,��a�v��q�ariable�is����>not��a��form���ula.�@Note�that�for��c��a�constan���t,���Ap�(�l�2`ambda�(�x;���t�)�;�c�)��is�a�form���ula,���since����>�Ap�BIJis�B�a�sym���b�Gol�and���(�x;���t�)�is�a�term.�k�As�a�term,�Fzthis�expression�����-reduces�to��t�,����>but��.that�do�Ges�not��-mak���e��t��a�form�ula.�NIndeed,���when�w�e��-get�to�the�axioms,���w�e�will����>see�+�that�+�����-reduction�applies�to�form���ulas,�aYbut�only�if�b�Goth�sides��ar��}'e��form���ulas.����>Another�U�dierence�b�Get���w�een�U�terms�and�form���ulas�is�that�lam�b�Gda�terms�do�not����>coun���t��as�form�ulas.���In�other��w�ords,�E�w�e�do�not�allo�w���pr��}'e�dic�ate�D�abstr�action�,�E�for����>example�+forming�+a�new�predicate��x:P�c��(�x;���y�[ٲ)�from�the�predicate��P�c��.��This�can����>b�Ge�D�done�D�indirectly�in�case��P��S�has�a�c���haracteristic�function,�Hbut�is�not�allo���w�ed�D�in����>general.����MAs��pis�customary��oin�logic�textb�Go�oks,��vw���e��pregard�inx�notation�as�an�abbrevi-����>ation���at�the�meta-lev���el,��pso���that�(�x�u3�+�u2�y�[ٲ)�really�means�+(�x;���y��).��:Similarly��A�u3�
!",�

cmsy10�^�u2�B����>�is�P�really�P��^�(�A;���B��q�).��c���"V

cmbx10�T���>�is�regarded�as�a�syn���tactic�v��q�arian�t�P�of��tr�Gue�.�c�That�means����>that��`�T�'�and���tr�Gue��are�t���w�o��dieren�t�names��for�the�same�sym�b�Gol.���Similarly��*�,�MC�F����>�is�ca�syn���tactic�v��q�arian�t�dof��f��al�2`se�.���The�term��l�ambda�(�x;���t�)�can�dalso�b�Ge�written�as����>�x:���t�;�k�the��latter��notation�is�just�an�informal�abbreviation�for�the�ocal�nota-����>tion�q��l�2`ambda�(�x;���t�).�ƂThe�notations�q��8�x�A��and�q�(�8�x�)��A��are�q�meta-lev���el�abbreviations����>for�غ�f��or�Gal�2`l��(�x;���A�),���and�similarly�ػfor��9�.���The�sym���b�Gols��^�,�_�,��!�,���and��:��are�syn���tac-����>tic��7v��q�arian���ts��8of��and�,����or�,����implies�,�and��7�neg�,�resp�Gectiv�ely��*�.�snOther��7t�yp�ographically����>con���v�enien�t�F�sym�b�Gols,���suc�h�F�as��<��or���,���can�similarly�b�Ge�regarded�as�meta-lev���el����>abbreviations�UUfor�certain�legal�predicate�sym���b�Gols.����MEac���h���form�ula�has�t�w�o�represen�tations:�,as�a�string�and�as���a�lab�Geled�tree.�CMOne����>and�vonly�wone�tree�represen���tation�corresp�Gonds�to�eac���h�string�represen�tation,�"=and����>vice-v���ersa.���An��X�o��}'c�curr�enc�e��of�a�v��q�ariable��Win�a�form���ula�is�represen�ted��Wb�y�a�single����>leaf�ިno�Gde�in�the�ީtree�represen���tation.�
�In�that�case�there�is�a�unique�path�from����>that���no�Gde���to�the�ro�Got�of�the�tree�represen���tation.�N�The��binding�3�op��}'er�ators��ٲare��8�,���9�,����>and��ĵ�.�oAn�o�Gccurrence�of�binding���op�erator�in�a�term�or�form���ula�has�a�v��q�ariable����>as�$�its�$�rst�argumen���t�(c�hild�$�in�the�tree).�a�The�concepts�that�a�certain�o�Gccurrence����>of�a�v��q�ariable�in�a�term�or�form���ula��A��is��fr��}'e�e�RXin�RW�A��or��b�ound�RWin��A�,�Rand�the�concept����>that�-�an�o�Gccurrence�of�a�binding�-op�erator��binds��certain�o�ccurrences�of�v��q�ariables,����>are��Jdened��Im���utually�b�y�recursion�as��Ifollo�ws.�K�\Bound"�is�a�synon�ym��Ifor�\not����>free".����MIn��Xa�form���ula��8�x���A�,���this�o�Gccurrence�of��8��binds�all�o�ccurrences�of��x��that�are������3�����y�����?������>�free��}in��~�A�.��@Those�o�Gccurrences�of��x��are�b�Gound�in��8�x���A�.��AIn�a�form���ula��9�x���A�,��Gthis����>o�Gccurrence�of��9��binds�all�o�Gccurrences�of��x��that�are�free�in��A�.�ZZThose�o�Gccurrences����>of��{�x��are�b�Gound�in��9�x���A�.�I�In�a�term��l�2`ambda�(�x;�t�),��sthis�o�Gccurrence�of��l�2`ambda��binds����>all�Z�o�Gccurrences�of��x��that�are�free�in��t�.���Those�o�ccurrences�of��x��are�b�ound�in����>�l�2`ambda�(�x;���t�).��zOccurrences��of��v��q�ariables�in�comp�Gound�terms�and�form���ulas�con-����>structed��Rb���y��Qthe�syn�tax��Qrules�not�in���v�olving��Rbinding�op�Gerators�retain�the�free�or����>b�Gound�+)status�that�+*they�had�in�the�comp�Gonen���t�form�ulas.��CNote�+*that�the�same����>v��q�ariable���can�ha���v�e���b�Goth���free�o�ccurrences�and���b�ound�o�ccurrences�in�a���giv���en�for-����>m���ula.����M�R��}'enaming�<
b�ound�variables�.��If��z�|��is�a�v��q�ariable�that�do�Ges�not�o�ccur�in��8�x���A�,����>then��?the�form���ula��>obtained�b�y�replacing�all��>o�Gccurrences�of��x��that�are�b�Gound����>b���y�սthis�binding�op�Gerator�վb�y��z�FT�is�said�to�b�Ge���	z�-c��}'onvertible�վ�to��8�x���A�;���similarly����>for���9�x���A��and��l�2`ambda�(�x;�t�).�}�The�relation�of���	z�-e��}'quivalenc�e��is��the�least�re
exiv���e����>and��3transitiv���e�relation�con�taining���	z�-con�v�ertibilit�y��4and�satisfying�in�addition�the����>conditions�UUthat����M(i)��,if��t��is���	z�-equiv��q�alen���t��+to��s��then��f���(�t�)�is����-equiv��q�alen���t�to��f���(�s�),�aand�similarly����>�f���(�r���;���t�)��is���	z�-equiv��q�alen���t��to��f��(�r���;���s�),��cand�similarly�for�eac���h�argumen�t��p�Gosition�of�a����>term�UUof�an���y�arit�y��*�.����M(ii)���if��ʵt��is���	z�-equiv��q�alen���t�to��s��then��l�2`ambda�(�x;���t�)�is���	z�-equiv��q�alen���t�to��l�2`ambda�(�x;���s�).����MIf����x��is���an���y�v��q�ariable,��and��A��is�an���y�term�or�form�ula,��then�w�e�can�nd���a�term�or����>form���ula�30�B����suc�h�that��B����is�31��	z�-equiv��q�alen�t�to��A��and�31�B����do�Ges�not�con�tain�31an�y�b�Gound����>o�Gccurrences�zof��x�.�y5W��*�e�sa���y�suc�h�a��B���is�\free�for��x�".�y5Sp�Gecically��*�,�-�an�algorithm����>�f��r�GeeF�c�or��(�A;���x�)��is��dened�b���y�recursion�on�the�term-or-form���ula��A�.��
The�crucial����>clause�gis�when�gthe�main�sym���b�Gol�of��A��is�a�binding�op�Gerator�and�the�v��q�ariable����>it�R�binds�is��x�.�j4In�that�case,��-the�algorithm�is�called�R�recursiv���ely�on�the�second����>argumen���t�Y�of�Y�the�term-or-form�ula�and�Y�then,���(i)�a�\fresh"�v��q�ariable��y����is�c���hosen����>(the�o�lexicographically�o�least�one�b�Geginning�with�the�same�letter�as��x��and�not����>o�Gccurring�[�in�[�the�result�of�the�recursiv���e�call),�]�and�then�(ii)�eac���h�o�Gccurrence�of��x����>�b�Gound�UUb���y�this�o�ccurrence�of�the�binding�op�erator�is�replaced�b���y��y�[ٲ.����M�Substitution�.��By�
��A�[�x��t�:=��s�t�]�w���e�mean�the�result�of�the�follo�wing�algorithm:����>Initialize�ɺ�P�-H�to��A�.�C>Then�for�ɹeac���h�v��q�ariable��z�:P�ha�ving�at�ɹleast�one�free�o�Gccurrence�in����>�t�,��Zreplace����P��N�b���y��f��r�GeeF�c�or��(�P�G;���z�p��).�!The���nal�v��q�alue�of����P��has�no�b�Gound�o�ccurrences����>of��Qan���y��Rv��q�ariables�free�in��t�,��Pand�is���	z�-equiv��q�alen���t�to��A�.�=�Then�replace�eac���h�free����>o�Gccurrence�ǹof��x��in��P�+H�b���y��t�.���That�is�the�v��q�alue�of��A�[�x����:=��t�].�In�ǹother�w���ords,��Rw�e����>rst��rename��b�Gound�v��q�ariables�in��A��to�a���v�oid��capture�of�the�free�v��q�ariables�of��t�,��Mand����>then�UUreplace��x��b���y��t��in��A�.����M��x��r��}'e�duction�.���The��Qterm��Ap�(�l�2`ambda�(�x;���t�)�;�q�[ٲ)��Qis��Psaid�to�����-reduce�immediately����>to�$e�t�[�x���:=��q�[ٲ].�awThe�$drelation�\�t����-reduces�to��s�"�is�dened�as�the�least�re
exiv���e�and����>transitiv���e�UUrelation�con�taining�immediate�����-reduction�and�satisfying:����M(i)��if��t����-reduces�to���s��then��f���(�t�)�reduces�to��f��(�s�),�$and�similarly��f��(�r���;���t�)�reduces����>to�UU�f���(�r���;���s�),�and�similarly�for�eac���h�argumen�t�p�Gosition�of�a�term�of�an�y�arit�y��*�.����M(ii)�UUif��t����-reduces�to��s��then��l�2`ambda�(�x;���t�)����-reduces�to��l�2`ambda�(�x;���s�).����>�Example�:��in�\7rst�order�\6logic�w���e�can�form���ulate�the�theory�of�groups,���using�a����>constan���t���e��for�the�iden�tit�y��*�,�0and�function�sym�b�Gols�for�the�group�op�eration�and����>in���v�erse.���The��use��
of�inx�notation��x��������y�(�can�either�b�Ge�regarded�as�ocial�or�as����>an��informal�abbreviation��for��m�(�x;���y�[ٲ),�I�just�as�it�can�in�rst�order�logic.��@If�w���e������4����.��y�����?������>�form���ulate�YJthe�same�theory�in�lam�b�Gda�logic,�ZGw�e�use�a�unary�predicate��G��for�the����>group,��Land��Irelativize��Jthe�group�axioms�to�that�predicate,��Kjust�as�w���e�w�ould��Ido�in����>rst�^order�logic�if�w���e�needed�to�study�a�group�and�a�subgroup.�ZuThen�in�lam�b�Gda����>logic��0w���e��1can�dene�the�comm���utator��c��ز:=��׵x;���y�[�:�((�i�(�x�)�p���i�(�y��))�p���x�)����y�[�;��1�and��0deriv���e����>the�UUfollo���wing:�������G�(�x�)�8�^��G�(�y�[ٲ)���!��c�(�x;���y��)�=�((�i�(�x�)�8���i�(�y��))����x�)����y����>�The�?Yh���yp�Gothesis��G�(�x�)��^��G�(�y�[ٲ)�is�needed�b�ecause�w���e�relativized�the�group�axioms����>to����G�.�)�W��*�e�w���an�t���to�form���ulate�the���theory�of�groups,��not�the�theory�of�mo�Gdels�of����>the�ðlam���b�Gda�ñcalculus�that�can�b�Ge�turned�in���to�groups.�A<Alternately��*�,���w�e�ñcan�replace����>�G�(�x�)�UUb���y��Ap�(�G;���x�)��=��T�,�UUand��x�8���y��.�b�y�UU�Ap�(�Ap�(��;���x�)�;�y�[ٲ).����MIf���w���e���w�an�t�to���discuss�more�than�one�group�at�a�time,��then�instead�of�using�a����>constan���t��d�e��e�for�the�iden�tit�y��*�,���w�e�can��ea�v��q�ariable;�

and�w�e��ecan�ev�en�use�a��ev��q�ariable�for����>the��+group�op�Geration�(although��*then�w���e�can�no�longer�write��x�⎸��y��in��+inx�notation;����>w���e��
m�ust�write��Ap�(�Ap�(�o;���x�)�;�y�[ٲ).��Unocially�w���e�could�abbreviate��that�as�(�xoy��)����>for�UUreadabilit���y�in�the�con�text�of�group�theory��*�.����MThat�}Jform���ulation�allo�ws�us�}Ito�discuss�subgroups,���homomorphisms,��etc.�)�quite����>naturally��*�.����MSp�Gecically��*�,���a��'homomorphism��(from�a�group�(�G;���e;���)��'to�(�H�A�;��1�;��+),���where��'here����>��q�and�+�are�rv��q�ariables�b�Geing�used�for�the�t���w�o�group�qop�Gerations,�G�w�ould�b�Ge�an����>ob��8ject�UU�u��suc���h�that�����m�Ap�(�G;���x�)���!��(�H���(�Ap�(�u;�x�))����m�̵Ap�(�G;���x�)�8�^��Ap�(�G;�y�[ٲ)���!��Ap�(�u;�x�)�8�+��Ap�(�u;�y�[ٲ)��=��Ap�(�u;�x�8���y�[ٲ)����MNo���w���one�can���quan�tify�o�v�er�homomorphisms,��for�example�to���state�that�a����>homomorphism��rfrom��s�G��to��H�yq�is�one-to-one�if�and�only�if�its�k���ernel�is�the�iden���tit�y��*�.����MNote��that��w���e�could�also�ha���v�e�used��t�w�o�predicate��sym�b�Gols��G���and��H�O��instead����>of�qv��q�ariables.���Then�qw���e�w�ould�ha�v�e�written��G�(�x�)�instead�of��Ap�(�G;���x�).���These�are����>not�Jthe�same.��Using�predicate�sym���b�Gols,���w�e�Jw�ould�not�b�Ge�able�to�quan�tify�o�v�er����>groups.����M�Example:�7Pe��}'ano's��axioms�.�r�P���eano's���axioms�for�n�um�b�Ger�theory���sa�y�that�the����>natural���n���um�b�Gers����N����constitute�the�least�set�con���taining�0�and�closed�under�suc-����>cessor,��8where��=successor��>is�a�one-to-one�function�from��N��X�to��N��Y�that�tak���es�ev�ery����>v��q�alue��Fexcept�0.�H�It�is�common�to�consider�the�recursiv���e�denitions�of�addition����>and��m���ultiplication�as��part�of�P�eano's�axioms,��to�Go.�1LAn�equiv��q�alen�t�form�ulation����>is���that�mathematical���induction�holds�for�all�sets��X��w�of�in���tegers:��Hif�0��׸2��X��and����>�8�n�(�n���2��X����!��s�(�n�)��2��X���)�Jthen�J�8�n�(�n��2��X��).�nThe�Jrst-order�Jtheory�P��*�A�Jis�obtained����>b���y��)taking��(the�induction�sc�hema,��in�whic�h��n�>͸2�>εX�f
�is�replaced�b�y��(an�y�rst-order����>form���ula�����(�n�).�_�In�lam�b�Gda���logic�w�e�can���replace��n�KE�2��X�m��b�y����Ap�(�X�:�;���n�)�KD=��T�,��wwhere����>�X�7�is�UUa�v��q�ariable.�q�Sp�Gecically��*�,�the�induction�sc���hema�in�lam�b�Gda�logic�is����HK�Ap�(�X�:�;����0)�8�^�8�x��(�N��(�x�)��^��Ap�(�X�:�;�x�)���!��Ap�(�X�:�;�s�(�x�)))��!�8�z�?�(�N��(�x�)��!��Ap�(�X�:�;�z�p��))�:����>�Here�ER�N�\m�is�a�unary�predicate�sym���b�Gol;���the�other�P�eano�axioms�w�ould�b�Ge�relativized����>to����N��.�T�In�making�pro�Gofs���b���y�induction,�w�e�w�ould���w�an�t�to�instan�tiate�X��qto���a�term������5����BN�y�����?������>�suc���h��as����(�z�p�;���x��L�+��z�7��=���z�-�+��K�x�).�]/T��*�o�construct��suc���h�a�term,�#�w���e�use�the�grammar�rule����>�term�?�:=�formula�.����MThis���is���somewhat�analogous�to�a�\w���eak�second-order�theory"�in�whic���h�there����>are��t���w�o��sorts�of�rst-order�v��q�ariables.�_2The�pro�Gof-theoretic�strength�of�suc���h�theo-����>ries��adep�Gends��`on�what�comprehension�axioms�are�included�to�allo���w�the�denition����>of���sets���to�use�in�inductiv���e�pro�Gofs.�ROThe�same�presumably�is�true�in�lam���b�Gda�logic;����>without���axioms�ev���en�to�dene�the���c�haracteristic�functions�of�arithmetic�predi-����>cates,�k�this�g9form�of�g8P���eano's�axioms�will�b�Ge�w���eak,�k�probably�only�as�strong�as����^���0��l�0������>�induction�UUin�the�language�of�P��*�A.��!�>�3��VL�Axioms�ffand�inference�rules�����>�Lam���b�Gda�{?logic�{>can�b�e�{>form���ulated�using�an�y�{>of�the�usual�approac���hes�to�pred-����>icate�~@calculus.��W��*�e�distinguish�the�sequen���t-calculus�form�ulation,��{the�Hilb�Gert-����>st���yle�3/form�ulation,�j�and�the�resolution�30form�ulation.�UF��*�or�deniteness�3/w�e�c�ho�Gose����>the��Hilb�Gert-st���yle�form�ulation�as�the��denition�(sa�y�as�form�ulated�in�[�10��
],���p.�M^20),����>for���the�standard�v���ersion.���Since�ev�ery�form�ula�is�a�term,�ȸthe�concepts�\normal����>term"�N4(a�term�admitting�no�N3reductions)�and�\strongly�normalizable�term"�(ev-����>ery�k4reduction�k3sequence�terminates)�apply�to�form���ulas�as�w���ell.��cW��*�e�note�that����>there��?do��@exist�form���ulas�that�are�not�normalizable,��suc���h�as��Ap�(�`;���`�),��where��`��is����>�x:�:����Ap�(�x;�x�).��+Indeed�+!this�term�reduces�to��:����Ap�(�`;�`�),�`�so�+!the�term��Ap�(�`;���`�)�is����>a��form���ula���A��suc�h��that��A��reduces�to��:�A�.��jThe�axioms�of�lam���b�Gda�logic�p�Germit����>us�0Nto�deriv���e��A�4�=�4
�:�A�,�gbut�they�do�not�p�Germit�us�to�deriv�e��A�4
�$�4:�A�,�gthanks����>to�'\restricting�����-reduction�on�form���ulas�to�'[apply�only�to�strongly�normalizable����>form���ulas.�q�Here�UUis�the�list�of�axioms�and�inference�rules�of�lam�b�Gda�logic:����M(�Pr��}'op�)�UUprop�Gositional�axioms�(see�[�10��
],�p.�q�20)����M(�Q�)�UUstandard�quan���tier�axioms�and�rules�(see�[�10��
],�p.�q�20)����M(��	z�)�UU�t���=��s��if��t��and��s��are�alpha-equiv��q�alen���t�terms.����M(��	z�)�UU�A���!��B��Ʋif��A��and��B��are�alpha-equiv��q�alen���t�form�ulas.����M(����)�UU�Ap�(�x:���t;�s�)��=��t�[�x��:=��s�]��where��t��is�an���y�term.�����M(����)�8�Ap�(�x:���A;�s�)���$��A�[�x��:=��s�]��UUwhere��A�[�x���:=��s�]�UUis�a�strongly�normalizable�form���ula.������M(��uDz)�8�(�we��}'ak��8extensionality�)��8�x�(�Ap�(�t;���x�)��=��Ap�(�s;�x�))��!��x:�Ap�(�t;�x�)�=��x:�Ap�(�s;�x�)�����M(�true���and�false�)�UU�T���=��xy�[�:���x��and��F��=��xy�[�:���y����M�(�non-triviality�)����T���6�=��F��$č��>�4��VL�Seman���tics�����>�There��Zis��[a�standard�denition�of���-mo�Gdel�that�is�used�in�the�seman���tics�of�the����>lam���b�Gda��calculus��(see�[�1��],�D�p.��F86,�with��details�on�p.��F93).��EThere�is�also�a�w���ell-����>kno���wn�Λnotion�Μof�a�mo�Gdel�of�a�rst�order�theory��*�.�ݚIn�this�section�our�goal�is�to������6����S�y�����?������>�dene�vjthe�concept��M��r�is��Va��Wmo��}'del�of�the�the��}'ory��T���in�lamb�da��Wlo�gic�vj�in�suc���h�a�w�a�y����>that��it��will�imply�that,�~neglecting���,��M�+�is��a�rst�order�mo�Gdel�of��T�c��,�~and�also,����>neglecting�UUthe�function�sym���b�Gols�other�than��Ap�,��M�lp�is�a���-mo�del.����MThe�?cited�?denition�of���-mo�Gdel�in���v�olv�es�?the�notion�of�terms,�C|whic���h�w�e�?shall����>call��8�M��-terms,�pbuilt�up�from��7�Ap��and�constan���ts��c����	0e�rcmmi7�a��	���for�eac�h��7elemen�t��a��of�the����>mo�Gdel.�m�It�I�requires�the�I�existence,�Lfor�eac���h�term��t��of�this�kind,�Land�eac���h�v��q�ariable����>�x�,�UUof�another��M��-term�����^��O!�cmsy7�����(�x;���t�)�suc���h�that��M�lp�will�satisfy��Kڍ��j<�Ap�(����������(�x;���t�)�;�x�)��=��t:����>�Note�0that�this�/do�Ges�not�y���et�mak�e�sense,�/�as�w�e�m�ust�rst�dene�/the�notion�of����>\the��in���terpretation��of�a�term��t��in��M��".�ЀW��*�e�cannot�simply�refer�to�[�1��]�for�the����>denition,��7since��<w���e��=need�to�extend�this�denition�to�the�situation�in�whic���h�w�e����>ha���v�e��|a�theory��}�T���with�more�function�sym���b�Gols�than�just��Ap�,���although�the�required����>generalization�UUis�not�dicult.����MW��*�e�B�rst�B�dene�a���-structure.�k�As�usual�in�rst�order�logic�w���e�sometimes�use����>\�M��"���to�denote�the�carrier�set�of�the�structure��M��;��Cand�w���e�use��f����M��rP�or���\q����������f���
βfor�the����>function�MRin�the�structure��M�dm�that�serv���es�as�the�in�terpretation�of�the�function����>sym���b�Gol�#�f���,�FWbut�$w�e�sometimes�$omit�the�bar�if�confusion�is�unlik���ely��*�.��3(�M���;�����)�is����>a��8��-structure�for��T�2Ȳif�(1)��M��S�is�a�structure�with�a��9signature�con���taining�all�the����>function�psym���b�Gols�pand�constan�ts�po�Gccurring�in��T�c��,���and�another�binary�function����>�Ap����M�����to��`serv���e��aas�the�in�terpretation�of��a�Ap�,���and�(2)��`there�is�an��aop�Geration�����^����LD�on����>pairs�@=(�x;���t�),�z�where�@<�t��is�an��M��-term�and��x��is�a�v��q�ariable,�z�pro�Gducing�an�elemen���t����>����^������(�x;���t�)�UUof��M��.����MIf��/(�M���;������^������)�is��.a���-structure�for��T�c��,��then�b���y�a��valuation��w�e�mean��.a�map��g����>�from�sfthe�sgset�of�v��q�ariables�to�(the�carrier�set�of��)��M��.��If��g��@�is�a�v��q�aluation,�z�and��v�U�=����>�v����1��|s�;����:�:�:����;���v����n��␲is�qa�qlist�(v���ector)�of�v��q�ariables,�xthen�b�y��g�[ٲ(�v��)�qw�e�qmean��g�[ٲ(�v����1��|s�)�;����:�:�:����;���g��(�v����n��q~�).����>If��9�t��8�is�a�term,��then�b���y��t�[�v�"�:=���g�[ٲ(�v��)]��9w�e�mean��8the��M��-term�resulting�from�replacing����>eac���h�jZv��q�ariable�jY�v����i�����b�y�the�constan�t�jY�c���:�g�@L�(�v���O
�\cmmi5�i��*��)��|]�for�the�elemen�t�jY�g�[ٲ(�v����i��TL�)�of��M��.���If��g��2�is�a����>v��q�aluation,�UUthen�w���e�can�then�extend��g��.�to�the�set�of�terms�b�y�dening��Kڍ����=0�g�[ٲ[�f���(�t����1��|s�;����:�:�:����;���t����n��q~�)]�������=�������\q���>���������f������(�g�[ٲ(�t����1��|s�)�;����:�:�:����;���g��(�t����n��q~�))���������l�g�[ٲ[�x:t�]�������=�����������������(�x;���t�[�v�"�:=���g�[ٲ(�v��)])������MNo���w��xw�e��wha�v�e�made��wsense�of�the�notion�\the�in���terpretation�of�term��t��under����>v��q�aluation�UU�g�[ٲ".�q�If����is�a�form���ula,�w�e�recursiv�ely�dene��M��3�j���UX�=���v���g��
�����as�follo�ws:��������F�_�M����[�	�j���UX�=���v���g������s��8�x���A�����i���ȵM��3�j���UX�=���v���h���A��UU�whenev���er��h��extends��g��.�and�is�dened�on��x��������F�_M����[�	�j���UX�=���v���g������s��9�x���A�����i���ȵM��3�j���UX�=���v���h���A��UU�for�some��h��that�extends��g��.�and�is�dened�on��x��������F�_M����[�	�j���UX�=���v���g������s��:�A�����i�����it�UUis�not�the�case�that��w�_�M��3�j���UX�=���v���g��
���A�������F�_M����[�	�j���UX�=���v���g������s��A�8�^��B��+�i��r9�M��3�j���UX�=���v���g��
���A�����and��qɵM��j���UX�=���v���g���B�������F�_M����[�	�j���UX�=���v���g������s��A�8�_��B��+�i��r9�M��3�j���UX�=���v���g��
���A�����or��ꫵM��j���UX�=���v���g���B�������F�_M����[�	�j���UX�=���v���g������s��t���=��s�����i���G�g�[ٲ(�t�)�UUand��g��(�s�)�are�the�same�elemen���t�of��M��������F�_M����[�	�j���UX�=���v���g������s��P�c��(�t����1��|s�;����:�:�:����;���t����n��q~�)����i���ȵP����M���\�(�g�[ٲ(�t����1���;��:�:�:����;�g�[ٲ(�t����n��q~�))����where�UU�P���is�not��Ap��������F�_M����[�	�j���UX�=���v���g������s��Ap�(�l�2`ambda�(�x;���B��q�)�;�t�)����i����ȵB��Ʋis�UUa�form���ula�and��k�|�M��3�j���UX�=���v���g��
���B��[�x���:=��t�]��������7����`ʠy�����?������>�W��*�e��regard�these��clauses�as�an�inductiv���e�denition,�#+in�the�follo���wing�w�a�y��*�.�_�The����>denition��denes��the�satisfaction�relation��j���UX�=���v���g��]��as�a�union�of�\lev���els",��Oand�si-����>m���ultaneously���denes���a�subset�of�the�complemen���t�of�the�relation��6j���UX�=���v���g���K�b���y�lev�els.����>F��*�or���example,���the�clause�for�negation���sp�Gecies�that�if��M�nx�j���UX�=���v���g��ڵA��at�lev���el��n�,���then����>�M�';�6j���UX�=���v���g��
Ӟ�:�A��&�at��'lev���el��n�V�+�V1,��and�vice-v�ersa,��if��M�';�6j���UX�=���v���g��
Ӟ�A��at�lev�el��n��then��&�M�'<�j���UX�=���v���g��
ӝ�:�A����>�at�Zlev���el��n�;��+�1.��The�Zrelation�Z�j���UX�=���v���g����obtained�as�the�union�of�all�the�lev���els�(indexed����>b���y�UUp�Gositiv�e�in�tegers)�is�the�\minimal�satisfaction�relation."����MIf�b�B��u�is�a�bstrongly�normalizing�form���ula,�e0then��M��W�j���UX�=�����B��u�or��M��6j���UX�=�����B��v�at�bthe��n�-th����>stage,�cwhere�`F�n��is�the�complexit���y�of�the�form�ula��B��q�.���W��*�e�note,�cho�w�ev�er,�that�`Fthe����>minimal�?satisfaction�relation�?is�not�dened�on�all�form���ulas.�jZF��*�or�example,�C�if��`��is����>�x����:�Ap�(�x;�x�),��-then�N��Ap�(�`;�`�)�is�a�form���ula,��.although�it�is�not�a�strongly�normalizing����>form���ula.��
Th�us�µAp�(�`;���`�)�̸$�:�Ap�(�`;�`�)��is��not�an�axiom,�I�since�the�righ���t�side�has����>no�-Vnormal�-Wform.���But�nev���ertheless��Ap�(�`;���`�)�is�a�form���ula,�cWand�w�e�-Wha�v�e�neither����>�M��3�j���UX�=��㎵Ap�(�`;���`�)��wnor��M��6j���UX�=��㎵Ap�(�`;���`�).�G}Hence,���using��vthe�minimal�satisfaction�relation,����>the�la���w�of�the�excluded�middle�is�not�satised.�XSimilarly��*�,���::�Ap�(�`;���`�)���!��Ap�(�`;�`�)����>is�UUnot�satised.����MW��*�e�}�dene�a�\satisfaction�relation"�on��M���to�b�Ge�a�relation��j���UX�=���v���g��AK�satisfying�the����>ab�Go���v�e�وclauses,��Jand�a�\classical�satisfaction�relation"�to�هb�e�a�satisfaction�relation����>that���also�satises�the�la���w�of���the�excluded�middle.�a�(T��*�ec�hnically�,�#�a���satisfaction����>relation�C!tak���es�C t�w�o�argumen�ts:��a�form�ula�C and�a�v��q�aluation;���our�notation�sometimes����>uses��ոj���UX�=���v���g��tR�for�the�relation�of�t���w�o���v��q�ariables,��and�sometimes�for�the�relation�of����>one�7)v��q�ariable�with��g���xed.)�CThe�existence�of�suc���h�relations�can�b�Ge�pro�v�ed�b�y����>con���tin�uing�Z�the�\lev���els"�past�the�ordinal��!�[ٲ.��`If,���for�limit�ordinal���,�there�is�a����>form���ula�r��A�r��suc�h�that�neither��M���j���UX�=���v���g��
�]�A�r��nor��M���6j���UX�=���v���g��
�^�A��at�a�lev���el�b�Gelo�w���,�y�then�let����>�A���b�Ge��the�least�suc���h�form�ula��and�sp�Gecify�that��M��3�j���UX�=���v���g��
���A��at�lev���el���.�TThen�the�next����>�!����stages�:�are�dened�as�ab�Go���v�e.�h�Since�:�there�are�only�:�as�man���y�form�ulas�as��M�Q�has����>elemen���ts,�UUthis�pro�Gcess�ev�en�tually�closes�o.����MW��*�e�Z'note�that�it�Z(is�more�or�less�arbitrary�whether,��\at�the�\c���hoice�p�Goin�ts"����>in��the�construction,�A�M��j���UX�=���v���g����A��or�not;�p5there�will�b�Ge�dieren���t�classical�satisfac-����>tion��prelations��omaking�dieren���t�c�hoices�ab�Gout��othe�satisabilit�y�of,��6for�example,����>�Ap�(�`;���`�);��but�Q�all�Q�classical�satisfaction�relations�will�agree�on�strongly�normal-����>izing�Qform���ulas,�A�and�Pthey�agree�with�the�minimal�satisfaction�relation�on�suc���h����>form���ulas.����MW��*�e��Sdene��R�M�o/�j���UX�=��	t��A��if�for�all�v��q�aluations��g�[ٲ,���M�o.�j���UX�=���v���g����A�.�v�Note�that�for�form���ulae����>not�,�in���v�olving�,ܵ�,�b�the�denition�agrees�with�the�usual�denition�of�satisfaction����>in�.rst�.order�logic.���Note�also�that�for�equalit���y�statemen�ts�.�t�07�=��s�,�d-there�is�.no����>am���biguit�y��)ab�Gout�the�satisfaction�relation.�DF��*�or�example,��޵Ap�(�`;���`�)�"y=��:�Ap�(�`;�`�)��)is����>satised,�\�since�(the�(left�side�����-reduces�to�the�righ���t�side;��Xbut�this�fact�co�Gexists����>with��4the�fact�that��3one�and�only�one�of�the�form���ulas��Ap�(�`;���`�)�and��:�Ap�(�`;���`�)�can�b�Ge����>satised,�nand��3\it��4can�go�either�w���a�y��*�."�Q�Suc�h�am�biguit�y��3do�Ges�not�arise�for�strongly����>normalizable�UUform���ulas.����MW��*�e��	are�no���w��
in�a�p�Gosition�to�dene���-mo�Gdel.�Z�This�denition�coincides�with����>that�UUin�[�1��]�in�case��T���has�no�other�function�sym���b�Gols�than��Ap�.������>�Denition��T1�(��-mo�Q�del)����f��(�M���;������^������;��j���UX�=���v���g��
�}�)���is��a���-mo��}'del�of�a�the��}'ory��T�E�in�lamb�da����>lo��}'gic��]if��j���UX�=���v���g��~��is�a�satisfaction�r�elation�on��^�(�M���;������^������)�,�:and�under��]that�satisfaction�������8����	tǠy�����?������>�r��}'elation,�)�(�M���;������^������)��N�satises��Othe�axioms���	z�,�)����,�(and���u��,�and��M�j�satises��Nthe�axioms����>of���T�c��.�� 񌍍�>�5��VL�Consistency�ffof�lam���b�s3da�logic�����>�Dene���the�relation��ӵt������s��on�terms�of��T�(a�to�mean�that��t��and��s��ha���v�e���a�common����>reduct���(using��� ��and����
�reductions).�>wThe�Ch���urc�h-Rosser���theorem���for����calculus����>([�1��],�/0p.��62)��kimplies�that�this�is�an�equiv��q�alence��jrelation,�when�the�language����>includes�}�only�}��Ap��and���.��The�follo���wing�theorem�sa���ys�that�adding�additional����>function�UUsym���b�Gols�do�es�not�destro���y�the�Ch�urc�h-Rosser�prop�Gert�y��*�.��D덍��>�Theorem��T1���y�D�The���Chur��}'ch-R�osser�the�or�em�is�valid�for�lamb�da�lo�gic.����>Pr��}'o�of�.�%�F��*�or�;�eac���h�function�sym�b�Gol�;��f�O��w�e�in�tro�Gduce�a�;�constan�t���\q��c�������f���
4�.�%�W��*�e�can�then����>eliminate��d�f���in�fa���v�or��dof���\q�����������c�f���
���as�follo���ws.���F��*�or�eac�h��cterm��t��w�e��cdene�the�term��t���^����mH�as����>follo���ws:��D덍����x������������a�=������)�x���for�UUv��q�ariables��x����������u*c������������a�=������)�c���for�UUv��q�ariables��c����������l�f���(�t�)������������a�=������)�Ap�(���\q��'!������f�����;���t���������)���������0��f���(�t;���r�G�)������������a�=������)�Ap�(�Ap�(���\q��'!������f�����;���t���������)�;�r��G��������))������������Ap�(�t;���r�G�)������������a�=������)�Ap�(�t���������;���r��G��������)���������g=(�x:���t�)������������a�=������)�x:���t����������D썑>�and�z�similarly�z�for�functions�of�more�than�t���w�o�argumen�ts.��Since�there�z�are�no����>reduction���rules���in���v�olving�the���new�constan�ts,�공t����reduces�to��q��y�if�and�only�if��t���^�������>�reduces��to��q��[ٟ�^�����.��vMoreo���v�er,�9�if��t���^�����ɲreduces��to��v�[ٲ,�then��v�g��has�the�form��u���^�����Ȳfor�some����>�u�.��T(Both�//assertions�are�/.pro���v�ed�//b�y�induction�/.on�the�length�of�the�reduction.)����>Supp�Gose�w��t�w��reduces�to��q�Ӓ�and�also�to��r�G�.���Then��t���^������reduces�to��q��[ٟ�^����lv�and�to��r��G��^������.���By�the����>Ch���urc�h-Rosser�J�theorem,��*�q��[ٟ�^����	?��and��r��G��^����	*βha���v�e�J�a�common�reduct��v�[ٲ,�and��v����is��u���^����㱲for����>some�UU�u�,�so��q��.�and��r��r�b�Goth�reduce�to��u�.�q�That�completes�the�pro�of.������>�Theorem��T2���y�D�L��}'amb�da���lo�gic�is�c�onsistent,�i.e.���do�es�not�derive��T���=��F�.����>Pr��}'o�of�.�jW��*�e���dene���a�mo�Gdel�(�M���;���Ap����M���\�;��j���UX�=���v���g��
�}�)�for���lam���b�Gda�logic.�jThe�elemen�ts���of��M����>�are�4the�equiv��q�alence�classes�[�t�]�of�terms�4�t��under�the�relation��t�����s�4�dened�ab�Go���v�e.����>W��*�e�L%dene�L$�Ap����M���\�([�t�]�;����[�s�])��:=�[�Ap�(�t;�s�)].�n�Then�L$let�L%�j���UX�=���v���g����b�Ge�a�satisfaction�relation,�M�i.e.����>a�UUxed�p�Goin���t�of�the�inductiv�e�clauses�giv�en�ab�Go�v�e.����MOne�V�pro���v�es�V�that,�Wfor�v��q�aluations��g����with�v��q�alues�in��M��,�Wand��x��in�the�domain�of����>�g�[ٲ,�UUand��r��r�an���y�term�in�the�equiv��q�alence�class��g��(�x�),�w���e�ha�v�e������ou�M��3�j���UX�=���v���g��
���A��UU�if�and�only�if��?B�M��j���UX�=��㎵A�[�x���:=��r�G�]�:�����8�(1)������>The�g�����axiom�for�g�terms�is�satised�since�[�Ap�(�l�2`ambda�(�x;���t�)�;�r�G�)]�g�is�[�t�[�x��2�:=��3�r��]].���The����>��W��axiom���for���form���ulas�is�automatically�satised�b�y�the���denition�of��M��3�j���UX�=��㎵A�.�E�The����>axiom�UU�T���6�=��F��is�satised�since��T��and��F��are�distinct�normal�terms.������9����
��y�����?������M�The��sw���eak�extensionalit�y��taxiom�(��uDz)�is�satised,��;as�follo�ws:�PSupp�Gose��M�ێ�sat-����>ises���8�x����(�Ap�(�t;�x�)��>=��=�Ap�(�s;�x�)).���Let��g���b�Ge�the��v��q�aluation�assigning��x��the�v��q�alue����>[�x�].�	x�Then�Sb���y�(1),�m��M�n�satises��Ap�(�t;���x�)��=���Ap�(�s;�x�).�	x�Hence�S�M��also�satises����>�l�2`ambda�(�x;���Ap�(�t;�x�))��&=��l�ambda�(�x;���Ap�(�s;�x�)),�ȡthe�~^conclusion�~_of�(��uDz).���That�com-����>pletes�UUthe�pro�Gof.��!č��>�6��VL�Some�#consequences�#of�the�xed�p�s3oin���t�theorem�����>�The��~xed��}p�Goin���t�theorem�of�lam�b�Gda��}calculus�sa�ys�that��}for�ev�ery��F�c��,�/Gthere�is����>an���
�suc���h�that��F�c��
�S=�R
.���Namely��*�,�Bw�e�can�tak�e�
�S=��Ap�(�!�[�;���!��),�Bwhere����!�d,�=����>�l�2`ambda�(�x;���Ap�(�F�G;�Ap�(�x;�x�)).�=Another��'form�of�the�xed�p�Goin���t�theorem�sa�ys�that����>for�w�ev���ery�w�term��H���,���w�e�w�can�nd�a�term��f��y�suc���h�that��Ap�(�f��V;���x�)��c=��H���(�f�;���x�).�م(T��*�o����>pro���v�e��4this,���apply��5the�rst�theorem�with��F�r�=���x:���H���(�f��V;�x�),���and�tak���e��4�f��òto�b�Ge�the����>
�UUpro�Gduced�b���y�the�rst�theorem.)����M�R���ussel���l's�``p��}'ar�adox�as�a�xe�d�`_p�oint�.��Since�3�ev���ery�3��F��%�has�a�xed�p�Goin���t,�k%there����>cannot�Ѻb�Ge�ѹ(in�lam���b�da�ѹcalculus�or�in�lam���b�Gda�logic)�an���y�term��not��suc���h�that�for�all����>�x�,���not�(�x�)���6�=��x�.�U�This��is��the�essence�of�Russell's�parado���x:�HSupp�Gose�there�w���as�suc�h����>a�3�not�.�V�If,�mfollo���wing�Ch�urc�h,�mw�e�2iden�tify��x���2��y�`�with�3�Ap�(�y�[�;���x�)�=��T�,�mthen�follo���wing����>the�*	pro�Gof�of�*
the�xed-p�oin���t�theorem�w�e�*
w�ould�set��!��q�=�)��x:���Ap�(�not;�Ap�(�x;�x�)),����>whic���h���w�e�recognize�as�the�Russell�set��R����=�oƸf�x�oDz:��x��62��x�g�,���since����!�[ٲ(�x�)�=��T����if�and����>only��mif��Ap�(�x;���x�)�?@�6�=�??�T�.�JThen��nconsidering�the�v��q�alue�of�
�=�?@�Ap�(�!�[�;���!��)��mamoun���ts�to����>Russell's�M�famous�question�M�whether��R��߸2���R�a��or�not,�O_and�the�xed-p�Goin���t�argumen�t����>sho���ws����not�(
)�s=�
,��Vwhic�h���in���Russell's�terms�sa�ys��R���2�s�R��P�if���and�only�if��R��62�s�R�Dz.����>Of�[course�in�[lam���b�Gda�logic,�\qthis�is�not�a�parado���x,�\qsince�no�suc�h�[term��not��exists.����>Although�.one�/ma���y�nd�it�coun���terin�tuitiv�e�that�.no�suc�h�.term�exists,���this�is�not����>new��with��lam���b�Gda�logic,�3�but�is�a�feature�of�lam���b�Gda�calculus�kno���wn�for�o���v�er����>sev���en�t�y�UUy�ears�no�w.����M�Inc��}'onsistency�YKof�A���C�.��Consider�the�follo���wing�v�ersion�of�the�axiom�of�c�hoice:����M(�A���C�)�UU(�Axiom���of�Choic��}'e�)��8�x����9�y��P�c��(�x;�y�[ٲ)���!�9�f��7�8�x���P��(�x;�Ap�(�f��V;�x�)).����>�AC�a��is��winconsisten���t��vwith�lam�b�Gda��vlogic:�
Simply�tak�e��v�P�c��(�x;���y�[ٲ)�to�b�Ge��x�T��6�=��y�[ٲ.�q+Then����>b���y�the�non-trivialit�y�axiom�of�lam�b�Gda�logic,��w�e�can�pro�v�e��8�x�9�y��x���6�=��y�\ݲ(tak�e��y��to����>b�Ge��2�T��²if��x�4�6�=�3�T�,���and��F��3�if��x��=�4�T�).��_But�then��3an���y�c�hoice��3function��f��²w�ould�satisfy����>the�UUequation�for��not�,�and�hence�there�is�no�suc���h��f���.����MIn�)M[�6��],�2the�)Llast�theorem�is�only�v��q�acuously�true,�2since�its�conclusion�sa���ys�that����>something�UUis�pro���v��q�able�in�lam�b�Gda�logic�plus�A�C.����M�F��;�ailur��}'e�uof�u~Skolemization�.�f�The�4Hsame�4Gexample�sho���ws�that�Sk���olemization�do�Ges����>not�ɮw���ork�ɯin�lam�b�Gda�logic.�C;That�is,��in�rst�order�ɯlogic,�圸8�x����9�y��A�(�x;�y�[ٲ)�is�ɮsatisable����>if��and��only�if��8�x���A�(�x;�f���(�x�))��is�satisable.�0�But�if�w���e�tak�e���x�1?�6�=��y���for���A�(�x;���y�[ٲ),���the����>former�7is�satisable�in�lam���b�Gda�logic,�B�but�the�latter�is�not,�since�then��x���f���(�x�)����>m���ust�c'ha�v�e�a�c&xed�p�Goin�t.��<W��*�e�sho�w�in�the�next�c&section�that�this�is�not�a�serious����>problem{w���e��sjust�ha�v�e�to�not�allo�w�lam�b�Gda-abstraction�when�it�w�ould�capture����>free��v��q�ariables�in�Sk���olem�terms.���That��extended�v�ersion�of�lam�b�Gda�logic�can�b�e����>used�UUto�pro���v�e�UUa�completeness�theorem.�������10�������y�����?������M�A���n��example�in�� gr��}'oup�the�ory�.�7�Here���is�a���more�mathematical�example�using�the����>xed��p�Goin���t�theorem.�"�Consider�the�theory�in�whic�h�w�e��ha�v�e�a�binary�function����>��\`�(written�in�inx�notation),��"a�constan���t��e�,��#and�the�axioms�sa�ying�that����is�a����>group�op�Geration�iden���tit�y��e�.���T��*�ak�e��H���(�f��V;���x�)�L=��c��\���[�Ap�(�f�;���x�),�B}where��c��is�an���ything.����>Applying��the�second��form�of�the�xed�p�Goin���t�theorem�w�e�get��Ap�(�f��V;���x�)�Y=�Z�c�%a�����>�Ap�(�f��V;���x�).�>�It���follo���ws�from���the�axioms�of�group�theory�that��c��is�the�group�iden���tit�y��*�.����>Since�]z�c�]{�w���as�an�ything,���w�e�]{could�ha�v�e�]{tak�en��c��to�]{b�Ge��T��or��F�,���but�this�implies����>�T�ym�=�yn�F�,��con���tradicting��Uthe��Vnon-trivialit�y�axiom��Vof�lam�b�Gda��Vlogic.���This�theory�is����>inconsisten���t.����MLo�Gok���ed�[at�Zseman�tically��*�,�1\this�is�Znot�a�surprising�result:���it�only�means�that����>it�Ȓis�imp�Gossible,��agiv���en�a�lam�b�Gda�mo�del��M��,��ato�dene�a�binary�op�eration�on��M����>�and�van�iden���tit�y�velemen�t�of�u�M�1��that�mak�e��M�1��in�to�a�group.��)If�w�e�use�a�unary����>predicate��z�G�,�*and��yput�in�axioms�sa���ying�that����is�a�group�op�Geration�on��G��with����>iden���tit�y��W�e�,��then�the�ab�Go���v�e��Wargumen�t�only�sho�ws�that��Ap�(�f��V;���x�)�is�not�in��X�G�,��i.e.����>�:�G�(�Ap�(�f��V;���x�)).����M�Extensionality�.�	�}The��follo���wing�additional�form�ula�is�sometimes�added�to����>lam���b�Gda�UUcalculus.����M(��[ٲ)�UU(�extensionality�)��x:���Ap�(�t;�x�)��=��t�.����MIt��is��not�an�axiom�of�lam���b�Gda�logic.��It�is,��ho���w�ev�er,�consisten�t��with��lam�b�Gda����>logic�cGsince�cFin�the�consistency�pro�Gof�in�the�previous�section,�f�w���e�could�ha���v�e�cGused����>��[ٲ-reduction��as��w���ell�as�����-reduction,�$�app�Gealing�to�the�Ch���urc�h-Rosser�theorem��for����>��[ٲ-reduction�UUin�lam���b�Gda�calculus.����M�T��;�erms��and�formulas.�0C�Supp�Gose���w���e���ha�v�e�deduced��ɸ:�A��for�some�form���ula��A�.�0DThe����>rules���of���lam���b�Gda�logic�do�not�allo���w�us�to�deduce��A���=��F��as���a�term.�NMSimilarly��*�,�2just����>b�Gecause��Qw���e�ha�v�e��Rpro�v�ed��A��do�Ges�not��Ren�title�us�to�deduce��R�A���=��T�.�G�Indeed,��making����>�A�Z$�equiv��q�alen���t�to�Z#�A���=��T����and�Z$�:�A��equiv�alen���t�to��A���=��F�Z#�w�ould�Z$lead�to�con�tradictions.����>F��*�reek��9Wiedijk��8ga���v�e�the��8follo�wing�example:�َlet��A��b�Ge��8�Ap�(�w��x���y�[�:��T�;�w�D�).�
rThen����>�A�灲is�a��form���ula,�Lso�w�e��ha�v�e��A�D��_�:�A�.�	(LIf��w�e�w�ere�able�to��deduce�from�this,����>�A�dU�=��T��,�_��+�A�dV�=��F�,��Dw���e�MGcould�MHobtain�a�con���tradiction,��Das�follo�ws:�a�As�MGa�term,��D�A����>�reduces�ʬto�ʫ�x���y�[�:��T�.�C�Hence��x�y�[�:��T���=��T�#��_�#��x���y�:��T���=��F�.�C�But�ʬeac���h�ʫdisjunct�can����>b�Ge���refuted,��since��Ap�(�Ap�(�T�;����F�)�;��F�)�qm=�qn�F����and����Ap�(�Ap�(�F�;��F�)�;��F�)�=�qn�F����according���to����>the���lam���b�Gda-calculus���denitions�of��T��and��F�,�8but��Ap�(�Ap�(�x���y�[�:��T�;��F�)�;��F�)��F=��E�T�,����>and�UUone�of�the�axioms�of�lam���b�Gda�logic�is��T���6�=��F�.����MNote��?that��x���y�[�:��T��@�is��?not�a�form���ula,��:so�w�e�are��@not�able�to�����-reduce��A��as�a����>form���ula,���and���the���con�tradiction�cannot�b�Ge���carried�out�in�lam���b�Gda�logic.�2�A���form�ula����>only�S"����-reduces�S#if�it�reduces�to�another�form���ula�(and�indeed,�S�that�form���ula�m�ust����>b�Ge�UUstrongly�normalizable).��!č��>�7��VL�Completeness�ffof�lam���b�s3da�logic�����>�The��follo���wing��theorem�is�to�lam���b�Gda�logic�as�G����odel's�completeness�theorem�is����>to���rst���order�logic.�oAs�in�rst�order�logic,��a�theory��T��r�in�lam���b�Gda�logic�is�called����>�c��}'onsistent�9��if�9�it�do�Ges�not�deriv���e�an�y�con�tradiction.�h�(A�9�con�tradiction�is�9�a�form�ula����>of��the�form���A��y�^�:�A�.)�y�In��this�section,�.w���e�will�pro�v�e��the�lam�b�Gda�completeness�������11�����Ϡy�����?������>�theorem:�T>an���y�Cconsisten�t�theory�has�a�D��-mo�Gdel.�^First,�&w�e�need�some�preliminar-����>ies.������>�Theorem��T3�(Lam��9b�Q�da�Completeness�Theorem)���(s��L��}'et�A�T��x�b�e�A�a�c�onsistent�the-����>ory���in�lamb��}'da�lo�gic.���Then��T��v�has�a���-mo�del.����>R��}'emark�.��FThere�(�is�a�kno���wn�\rst�(�order�equiv��q�alen�t"�of�the�(�notion�of���-mo�Gdel,����>namely�A��Sc��}'ott�mXdomain�.�6�See�[�1��](section�A�5.4).�Ho���w�ev�er,�|�w�e�A�could�not�A�use�Scott����>domains�ȶto�ȷreduce�the�lam���b�Gda�completeness�theorem�to�G����odel's�rst�order�com-����>pleteness�^�theorem,�a+b�Gecause�there�is�no�syn���tactic�in�terpretation�of�the�theory�of����>Scott�UUdomains�in�lam���b�Gda�logic.����>�Pr��}'o�of�#of�#
the�c��}'ompleteness�the�or�em�.�	�WOur�0plan�/is�to�imitate�the�usual�pro�Gof����>(Henkin's�|�metho�Gd)�of�the�completeness�theorem,���as�set�out�for�example�in�[�10��
]����>pp.�ʾ43-48,��gcan�r�b�Ge�r�imitated�for�lam���b�da�r�logic.�ʿHo�w�ev�er,��fthe�r�rst�step�in�the����>classical���pro�Gof�is���Sk���olemization,�]whic�h���w�e�ha�v�e�seen�is���problematic�in�lam�b�Gda����>logic.��DThe��reason�wh���y�it��is�problematic�is�that�lam�b�Gda�logic�p�ermits�lam���b�da����>abstraction���to�b�Ge�used�on���an���y�function�sym�b�Gol,���so���for�example�w�e�cannot�con-����>struct���a���mo�Gdel�with�a�Sk���olem�function�for��8�x�9�y��x���6�=��y�[ٲ.�C�The���remedy�is�to�enlarge����>lam���b�Gda�Vlogic�to�Wallo�w�\Sk�olem�function�sym�b�Gols"�with�a�Wrestriction�on�their����>use���in�lam���b�Gda�terms.�* This�enlarged�logic�presumes�a�list�of�\Sk�olem�sym�b�Gols"����>disjoin���t���from�the�list�of�function�sym�b�Gols�of�lam�b�Gda�logic.�YA��y\Sk�olem�term"�is����>a��tterm�whose�main��usym���b�Gol�is�a�Sk�olem�sym�b�Gol.�R|The�restriction��uis�that�the�term����>�l�2`ambda�(�x;���t�)�ƪcan�ƫonly�b�Ge�formed�when��t��do�Ges�not�con���tain�a�Sk���olem�term�with����>�x�h��free.���Otherwise�Sk���olem�sym�b�Gols�h�are�used�just�lik�e�ordinary�function�sym�b�Gols����>and�UUconstan���ts�in�forming�terms.����M�Example.����If�e�f���(�x�)�eis�a�Sk���olem�term�with�the�asso�Gciated�axiom��f���(�x�)����6�=����x�,����>then���w���e���cannot�form���(�x;���f���(�x�))�(so�w���e�cannot�deriv���e�a�con�tradiction���from�the����>existence�UUof�a�xed�p�Goin���t�of�that�term).����MThe��saxioms��rof�Sk���olem�lam�b�Gda��rlogic�are�the�same�as�those�of�lam���b�Gda�logic,����>except��(of�course)�that�in�the�(����)�and�(��uDz)�axiom��sc���hemata,�!only�those�instances����>are�UUtak���en�that�are�legal�form�ulas�in�Sk�olem�lam�b�Gda�logic.����MW��*�e��krefer�to��jthis�more�elab�Gorate�system�as�\Sk���olem�lam�b�Gda�logic".�DW��*�e�will����>use�b�it�b�only�in�the�pro�Gof�of�the�completeness�theorem.���The�seman���tics�of�Sk���olem����>lam���b�Gda�glogic�ggeneralizes�the�seman���tics�of�lam�b�Gda�glogic�dened�ab�Go���v�e,��pb�y�gin-����>terpreting�
�the�
�Sk���olem�terms�as�they�w���ould�b�Ge�in���terpreted�in�rst�order�logic.����>That�pis,���the�mo�Gdel�pm���ust�sp�ecify�pa�function���\q���&�������f�����from��M�� �to��M��to�pserv���e�as�the����>in���terpretation���of�eac�h�Sk�olem�function�sym�b�Gol��f���.�d�(If�the�same�sym�b�Gol�o�ccurs����>with���dieren���t�arities,��1it�is�in�terpreted�b�y���a�dieren�t�function�for�eac�h���arit�y��*�.)�6�The����>function�h�����^������,�mbwhic���h�tak�es�a�h�v��q�ariable��x��and�an��M��-term��t��as�argumen���ts,�mbdo�Ges�not����>need�UUto�b�Ge�dened�on�terms��t��con���taining�Sk�olem�terms�with��x��free.����MW��*�e�w�sho���w�w�that�a�single�quan���tier�can�b�Ge�eliminated.��"Consider�a�theorem����>�T�=@�con���taining�ٱan�axiom�ٰ�8�x�9�y��A�(�x;���y�[ٲ),���and�consider�the�theory��T��c���^����	�$�con���taining�a����>new�Mxa�MwSk���olem�function�sym�b�Gol�Mw�f�a�and�the�axiom��8�x���A�(�x;�f���(�x�))�Mxinstead�of�the����>axiom�{?�8�x�9�y��A�(�x;���y�[ٲ).��Note�that�this�axiom�{>is�a�form���ula�of�Sk�olem�lam�b�Gda�logic����>since��A�(�x;���f���(�x�))��means��A�[�y�"�:=���f��(�x�)]�and��an���y�b�Gound�v��q�ariables�that�o�Gccur�free�in����>�f���(�x�)�@are�Arenamed�as�part�of�the�substitution�pro�Gcess,�'�so�no�free�v��q�ariables�of�the����>Sk���olem�UUterm��f���(�x�)�are�captured�b�y�a�lam�b�Gda-binding.�������12����
�D�y�����?������M�W��*�e�s<claim��T��˲has�a�mo�Gdel�if�and�only�if��T��c���^����
o��has�a�mo�del.��|A�r�mo�del�of��T��c���^�������>�already�fsatises�e�T�c��,�&0so�only�one�direction�requires�pro�Gof.�^"Supp�ose��M�1��satises��T�c��.����>Then��8for�ev���ery��x���2��M�S�there��8is�a��9�y�"�2���M��suc���h�that��M��satises��A�(�x;���y�[ٲ).�NUsing�the����>axiom��mof�c���hoice��l(at�the�meta-lev�el)�there��lis�a�function���\q��
��������f���
�<�:���M���!��M����suc�h��mthat����>�M���satises�w̵A�(�x;���\q���ɲ���������f������(�x�))�w�for�ev���ery��x�.��*W��*�e�tak���e���\q�����������f����}�as�the�in�terpretation�w�of��f���.��*Then����>the�τaxiom��8�x���A�(�x;�f���(�x�))�υis�τsatised.��TW��*�e�ha���v�e�to�extend�the�function�υ����^����hh�of��M����>�(whic���h��tak�es��a�v��q�ariable�and�an��M��-term�and�pro�Gduces�an��M��-term)�to�b�Ge�dened����>on��̵M��-terms�in�the�language�of��T��c���^�����s�.�{-W��*�e�also�ha���v�e���to�dene�the�in���terpretations������>sD���>�r���G"²of�X�terms�X�in�Sk���olem�lam�b�Gda�logic�X�in��M��.�{�These�t���w�o�tasks�X�are�accomplished����>sim���ultaneously�d�b�y�d�m�utual�recursion;�l�the�d�function�����^�����۲is�used�just�as�for�lam���b�Gda����>calculus�+�to�dene�the�+�in���terpretation�of�lam�b�Gda-terms,�4and�no�w�+�w�e�sho�w�ho�w�to����>extend��o����^����;S�to��nthose�lam���b�Gda�terms�that�are�legal�in�Sk���olem�lam�b�Gda�logic.�YLet��t����>�b�Ge��an��M��-term��of��T��c���^�����s�,��Gand��x��a�v��q�ariable,��Gand�supp�Gose�that��t��con���tains�some�Sk�olem����>�M��-subterms,�=�but�Udo�Ges�Vnot�con���tain�an�y�VSk�olem�subterms�Vwith��x��free.���Then��t����>�can���b�Ge���written�as��t���^��0���9�[�u�K1�:=��r��],��iwhere����u����stands�for�sev���eral�v��q�ariables�and��r�봲for�the����>list���of�all�Sk���olem���subterms�of��t�,��cand��t���^��0����do�Ges�not�con�tain�an�y�Sk�olem���terms.�@�Since����>the�sterms��r�J��do�rnot�con���tain��x��free,��the�v��q�ariable��x��do�Ges�not�get�renamed�when�w���e����>substitute�UU�r��r�for��u��in���(�x;���t���^��0���9�).�q�Th���us��!d������(�x;���t�����0���9�)[�u���:=��r�G�]�=���(�x;�t�����0���9�[�u��:=��r�G�])�=���(�x;�t�)����>W��*�e�UUdene������2����������(�x;���t�)��:=����������(�x;���t�����0���9�[�u��:=��c�����fѱ���r���m��])��zp��>where���ͯ���Zk�r���~�is�Zkthe�Zjin���terpretation�in��M�q��of�the�Sk���olem�term��r����and��c�����fѱ���r����
�is�the�constan���t����>denoting�UUthat�elemen���t�of��M��.����MNo���w�gw�e�c�hec�k�the�v��q�alidit�y�of�axiom�(����).��	Let��r��3�b�Ge�a�list�of�all�Sk�olem�terms����>o�Gccurring�}in�}�t��and��u��a�corresp�Gonding�list�of�v��q�ariables�not�o�Gccurring�in��t�.���Let��t���^��0�����>�b�Ge���a���term�suc���h�that��t�Dò=��t���^��0���9�[�u�D²:=��r�G�],���and����t���^��0��n�con���tains�no�Sk�olem�terms.�S�In����M��ײw�e����>ha���v�e��!d����H���Ap�(�l�2`ambda�(�x;���t�)�;�q�[ٲ)��������[=�����l�2�Ap�(����������(�x;���t�)�;�q�[ٲ)��������[=�����l�2�Ap�(����������(�x;���t�����0���9�[�u���:=��c�����fѱ���r���m��]�;���iT�����q���}R�)��b���y�UUdenition�of�����^������(�x;�t�)���������[=�����l�2�Ap�(�l�2`ambda�(�x;���t�����0���9�)[�u���:=��c�����fѱ���r���m��])�;���iT�����q���}R�)��since�UU�x��do�Ges�not�o�ccur�free�in���ș����r�����������[�=�����l�2�Ap�(�l�2`ambda�(�x;���t�����0���9�)�;���iT�����q���}R�)[�u���:=��c�����fѱ���r���m��]��since�UU�c�����fѱ���r�����is�a�constan���t�term���������[=�����l�2�t�����0���9�[�x���:���������q����²][�u��:=��c�����fѱ���r���m��]��since�UUaxiom�(����)�holds�in��M���������[�=�����l�2�t�[�x���:=���������q����²]������MFinally��+w���e��*c�hec�k�the��*v��q�alidit�y�of��*the�w�eak��*extensionalit�y�axiom��*(��uDz).�3HRecall����>that��7the��8axiom�in�question�sa���ys�that�if�for�all��x�,�50�Ap�(�t;���x�)�F�=�F�Ap�(�s;�x�),�5/then����>��(�x;���Ap�(�t;�x�))��B=���(�x;���Ap�(�s;�x�)).�{This��<is�an��;axiom�of�Sk���olem�lam�b�Gda��;logic�only����>in�L�case�L��t��do�Ges�not�con���tain�an�y�Sk�olem�L�subterms�with��x��free.�n�In�that�case,�NGlet��t���^��0�����>�and�� �s���^��0���Y�b�Ge�terms�without�Sk���olem�subterms,���and��r�"=�a�list�of�Sk�olem�terms,���suc�h����>that������^������(�x;���t�)��=�����^�����(�x;���t���^��0���9�[�u��:=��c�����fѱ���r���m��])��and�����^�����(�x;���s�)��=������^�����(�x;�s���^��0���9�[�u���:=��c�����fѱ���r���m��])��Then�in��M��,����>b���y���h�yp�Gothesis���w�e�ha�v�e����Ap�(�t;���x�)�=�=�=��Ap�(�s;�x�)�for���all��x�,��{so�also��Ap�(�t���^��0���9�[�u�=�:=�=��r�G�]�;���x�)�=�������13����Ύ�y�����?������>�Ap�(�s���^��0���9�[�u���:=��r�G�]�;���x�).�gSince�5?the�5@in���terpretation�in��M�LZ�of��t���^��0���9�[�u���:=��r�G�]�5?is�the�same�as�that����>of�UUthe��M��-term��t���^��0���9�[�u���:=��c�����fѱ���r���m��],�UUw���e�also�ha�v�e�����߹�Ap�(�t�����0���9�[�u���:=��c�����fѱ���r���m��]�;���x�)�=��Ap�(�s�����0���[�u��:=��c�����fѱ���r���m��]�;���x�)�:����>�But��these��terms�con���tain�no�Sk���olem�subterms.�Z�Hence,�[applying�axiom�(��uDz)�in��M��,����>w���e�UUha�v�e�����`M�l�2`ambda�(�x;���t�����0���9�[�u���:=��c�����fѱ���r���m��])�=��l�ambda�(�x;���s�����0���9�[�u��:=��c�����fѱ���r���m��])�:����>�Therefore�UU�M�lp�satises������̵�(�x;���t�����0���9�[�u���:=��r�G�])�=���(�x;�s�����0���9�[�u��:=��r�G�])�;����>�whic���h��is��the�same�as�to�sa���y��M��.�satises���(�x;���t�)���=�����(�x;�s�).�'That�completes��the����>v���erication�UUof�axiom�(��uDz).����MNo���w,�7tas�
:usual�
;in�rst-order�logic,�7tw���e�can�eliminate�quan���tiers�one�b���y�one����>in�Cfa���v�or�of�Sk�olem�Cfunctions,�F�so�that�to�ev���ery�theory��T����in�lam�b�Gda�logic,�F�w�e�can����>construct�0�a�theory��T��c���^����	,��in�Sk���olem�lam�b�Gda�logic�with�quan�tier-free�axioms,�7�suc�h����>that���T�u�has�a�mo�Gdel�if��and�only�if��T��c���^����	
��has�a�mo�Gdel.�[.Hence�it�suces�to�pro���v�e��the����>completeness�UUtheorem�for�quan���tier-free�Sk�olem�lam�b�Gda�logic.����MIf�2$�T����do�Ges�not�2#con���tain�innitely�man���y�constan�t�2#sym�b�Gols,�9-w�e�b�Gegin�2#b�y�adding����>them;��this��cdo�Ges�not�destro���y�the�consistency��dof��T�+�since�in�an�y�pro�Gof�of�con�tra-����>diction,�$Vw���e�could�replace�the�new�constan���ts�b�y�v��q�ariables�not�o�Gccurring�elswhere����>in��bthe�pro�Gof.�6�W��*�e�construct�the�\canonical�structure"��M�}�for�a�theory��T�c��.�The����>elemen���ts��of��M��5�are��equiv��q�alence�classes�of�closed�terms�of��T���under�the�equiv��q�alence����>relation��[of�pro���v��q�able�equalit�y:�Aʵt�����r�<w�i��[�T�*��`��t��=��r�G�.�Q�Let��Z[�t�]�b�e�the�equiv��q�alence�class����>of����t�.�<�W��*�e���dene�the�in���terpretations�of�constan���ts,�՛function�sym�b�Gols,�՛and���predicate����>sym���b�Gols�UUin��M�lp�as�follo�ws:������֟��c�����M�������Ӽ�=��������[�c�]����������:�f�������M��	��([�t����1��|s�]�;����:�:�:����;����[�t����n��q~�])������Ӽ=��������[�f���(�t����1��|s�;����:�:�:����;���t����n��q~�)]�����������P��c�����M��
C�([�t����1��|s�]�;����:�:�:����;����[�t����n��q~�)]������Ӽ=��������[�P�c��(�t����1��|s�;����:�:�:����;���t����n��q~�)]������>In�G�this�G�denition,�JGthe�righ���t�sides�dep�Gend�only�on�the�equiv��q�alence�classes�[�t����i��TL�]�(as����>sho���wn�UUin�[�10��
],�p.�q�44).����MExactly�Œas�œin�[�10��
]�one�then�v���eries�that��M�ܮ�is�a�rst�order�mo�Gdel�of��T�c��.��T��*�o����>turn�ur�M����in���to�a�us��-mo�Gdel,��yw�e�m�ust�urdene�����^������(�x;����[�t�]),��zwhere��x�us�is�a�v��q�ariable�and����>�t�<�is�an�<�M��-term,�v�i.e.�(�a�closed�term�with�parameters�from��M��,�v�and��t��do�Ges�not����>con���tain��fSk�olem�subterms�in��gwhic�h��x��o�Gccurs�free.�PwThe��g\parameters�from��M��"�are����>constan���ts�X��c���:�[�q�@L�]��JC�for�closed�X�terms��q��i�of��T�c��.�{zIf��t��is�an��M��-term��t�,�Y`let�[�t�]���^�����u�b�Ge�the�closed����>term�Hof�H�T����obtained�from��t��b���y�replacing�eac�h�Hconstan�t��c���:�[�q�@L�]��9ʲb�y�a�Hclosed�term��q���in����>the�wequiv��q�alence�class�w[�q�[ٲ].���Then�[�t�]���^�����is�w���ell-dened.���Dene�����^������(�x;����[�t�])��;=�[�x:��[�t�]���^������].����>By��$axiom��%(��uDz),��this�is�a�w���ell-dened�op�Geration�on�equiv��q�alence�classes:�1.if��T�7��pro���v�es����>�t���=��s�UU�then��T���pro���v�es�UU[�t�]���^����_��=��[�s�]���^�����9�and�hence��x:����[�t�]���^�����=���x:��[�s�]���^������.����MW��*�e��v���erify�that�the�axioms�of�lam�b�Gda�logic�hold�in��M��.��GFirst,�2�the�rst�(����)����>axiom:��еAp�(�x:���t;�r�G�)��H=��t�[�x��:=��r��].�UIt���suces���to�consider�the�case�when��t��has����>only�>��x��free.�-lThe�in���terpretation�of�the�>�left�side�in��M�U��is�the�equiv��q�alence�class�������14�����8�y�����?������>�of�u$�Ap�(�x:���t;�r�G�).��4The�in���terpretation�of�the�righ�t�side�is�the�class�of��t�[�x��IJ:=��r�G�].����>Since��rthese��qt���w�o�terms�are��qpro�v��q�ably�equal,��ltheir��qequiv�alence�classes�are��qthe�same,����>v���erifying�UUaxiom�(����).����MConsider�UUthe�second�(����)�axiom:������B��Ap�(�x:���A;�s�)���$��A�[�x��:=��s�]��UUwhere��A�[�x���:=��s�]�UUis�a�strongly�normalizable�form���ula.�����>One�Íof�Ìthe�inductiv���e�clauses�dening��j���UX�=���v���g���
�sa���ys�that�if�the�righ���t-hand�side�is����>satised,���then���the���left-hand�side�is�satised.��Since��j���UX�=���v���g��P,�is�a�xed�p�Goin���t�of�these����>inductiv���e�$�clauses,�X]and�since�no�other�clause�in�the�denition�pro�vides�for�the����>satisfaction�M�of�a�form���ula�M�of�the�form��Ap�(�l�2`ambda�(�x;���B��q�)),���then�if�the�left-hand����>side�UUis�satised�so�is�the�righ���t-hand�side.�q�Hence�the�axiom�is�satised.����MNo���w��{for��|axiom�(��uDz).�:Supp�Gose��t��and��s��are�closed�terms�and��Ap�(�t;���x�)��=��Ap�(�s;�x�)����>is��ev��q�alid�in��d�M��.�Z�Then�for�eac���h�closed�term��r�G�,�!(w�e�ha�v�e��tr�?��pro�v��q�ably��dequal�to��sr�G�.����>Since�Į�T�(=�con���tains�innitely�man�y�constan�t�ĭsym�b�Gols,��w�e�can�select�a�constan�t��c����>�that��do�Ges��
not�o�ccur��
in��t��or��s�,��so��tc���=��sc���is�pro���v��q�able.�GYReplacing�the�constan���t��c��b���y����>a��v��q�ariable��in�the�pro�Gof,���tx���=��sx���is�pro���v��q�able.�O�Hence�b�y��axiom�(��uDz),���x:���t���=��x:�s���is����>probable,��and��hence�that�equation��
holds�in��M��.�>In�v���erifying�axiom�(��uDz),��it�suces����>to�\|consider�\}the�case�when��s��and��t��are�closed�terms.��>The�axiom�(��	z�)�holds�in��M����>�since�"�it�"�simply�asserts�the�equalit���y�of�pairs�of�pro���v��q�ably�equal�terms.�`�The�axiom����>�T�c*�6�=�c+�F�L��holds�L�since��T��#�do�Ges�not�pro���v�e��T�c*�=�c+�F�,��db�Gecause�L��T��#�is�consisten�t.�W�That����>completes�UUthe�pro�Gof.��!A���>�8��VL�The�ffrst�order�fragmen���t�of�lam�b�s3da�logic�������>�Theorem��T4�(Axiomatization�of�rst-order�theorems)���I	@�L��}'et��Z�T���b�e�a�rst�or-����>der�<the��}'ory,�6�and�let��A��b�e�a�rst�or�der�;sentenc�e.�r�Then��T����pr�oves�;�A��in�lamb�da�lo�gic����>if��and��only�if�for�some�p��}'ositive�inte�ger���n�,��D�T����plus�\ther�e��exist��n���distinct�things"����>pr��}'oves���A��in�rst�or�der�lo�gic.������>Pr��}'o�of�.�֧First�v�supp�Gose��A��is�pro���v��q�able�v�from��T�ڄ�plus�\there�exist��n��distinct�things".����>W��*�e�'2sho���w��A�'1�is�pro�v��q�able�in�lam�b�Gda�'1logic,�0mb�y�induction�'2on�the�length�'1of�the�pro�Gof����>of����A�.�N�Since��lam���b�Gda�logic�includes�rst�order�logic,��jthe�induction�step�is�rivial.����>F��*�or���the���basis�case�w���e�m�ust���sho�w�that�lam�b�Gda���logic�pro�v�es���\there�exist��n��distinct����>things"���for�eac���h�p�Gositiv�e���in�teger��n�.��zThe�classical���constructions�of�n�umerals�in����>lam���b�Gda��calculus�pro�duce��innitely�man���y�distinct�things.���Ho�w�ev�er,���it�m�ust�b�Ge����>c���hec�k�ed��,that�their�distinctness��+is�pro���v��q�able�in�lam�b�Gda�logic.��KDening�n�umerals����>as�� on�p.�U'130�of�[�1��]�w���e��v�erify�� b�y�induction�on��n���that�for�all��m�Ei<�Ejn�,���d�m�e�6�=��d�n�e����>�is�|�pro���v��q�able�in�lam�b�Gda�logic.��If�|��m�	/<�	0n�SP�+�1�|�then�either��m��=��n�,���in�whic���h�case�w�e����>are��8done�b���y�the�induction�h�yp�Gothesis,��por��m�h��=�h��n�.��pSo��7what�has�to�b�e�pro���v�ed��8is����>that��for�eac���h��n�,�!
lam�b�Gda��logic�pro�v�es��d�n�e��6�=��d�n��-�+�1�e�.�[�This��in�turn�is�v�eriable�b�y����>induction�UUon��n�.����MCon���v�ersely��*�,�OWsupp�Gose�M�that�MٵA��is�not�pro���v��q�able�in��T��g�plus�\there�exist��n��distinct����>things"���for���an���y��n�.�TBThen�b�y�the���completeness�theorem�for�rst�order�logic,�|there����>is�~an�innite�mo�Gdel��M��"�of��:�A�;�aindeed�w���e�ma�y�assume�that��M��"�has�innitely����>man���y�Eelemen�ts�not�Ddenoted�b�y�closed�terms�of��T�c��.�̖W��*�e�will�sho�w�Dthat��M�5`�can�������15�����-�y�����?������>�b�Ge�CCexpanded�CDto�a�lam���b�Gda�mo�del���x䍑��^�����CD�M���Q1�satisfying�CDthe�same�rst�order�form���ulas,�F�b�y����>dening���arbitrarily�the�required�op�Geration�����^����p��on��M��-terms,���and�then�inductiv���ely����>dening�S�relations�S�E����(�x;���y�[ٲ)�and��Ap����M��4K�to�serv���e�as�the�in�terpretations�S�of�equalit�y�and����>�Ap���in���x䍑r�^�������M������.��rT��*�o�complete�the�construction���of�a���-mo�Gdel,�Їw���e�supply�an���y�classical����>satisfaction�UUrelation�on���x䍑^������M�����.����MT��*�o�Bdo�B
this�w���e�dene�the�relation��Ap����M��
"g�and�a�binary�relation��E�՘�on��M�Y&�b���y����>sim���ultaneous�oinduction.���Ap����M��Oo�will�serv�e�oas�the�in�terpretation�of��Ap��and��E���will����>serv���e�4�as�the�in�terpretation�4�of�equalit�y��*�.����^��2�����Since��M�K��is�innite�4�w�e�can�dene�an����>elemen���t��,0��+and�a�pairing�function��h�a;���b�i��on��M��G�in�suc���h�a�w�a�y�that��+the�in�terpre-����>tations��6of��5the�losed�terms�of��T��Ųare�nev���er�pairs,��-and�0�is�not�a�pair.��iDene����>�<��a;���b;�c�>�=�<�a;�<�b;�c�>>�,�N�etc.�oDene�MSthe�MTsuccessor�of��x��to�b�Ge��s�(�x�)��=�<��0�;���x�>�,����>and��mdene��lthe�\n���umeral"����G����m���~��inductiv�ely�b�y����q���m��#����m0����=���0�and����q��[���#����m�m�8�+�1���!��=����s�(��������m���Ƿ�).�Hence-����>forth�K�w���e�K�drop�the�bars,�M�writing�for�example��h�1�;���k�P��i��instead�of��h����q����#��1����;���\q���S����������k���/��i�.�n�An�elemen���t����>of��4the��5form��<�[7�1�;���j�R;�k���>��will��4b�Ge�used�as�an�\index"�of�the��k�P��-th�function�sym���b�Gol����>of�UUarit���y��j���in��T�c��,�whic�h�w�e�denote�b�y��f���1ɍ���j���v�k���J;�.����MT��*�uples�cbare�ccdened�from�pairs�b���y��h�x����1��|s�;����:�:�:����;���x����n�+1�����i�ރ�=�ބ�hh�x����1���;��:�:�:����;�x����n��q~�i�;�x����n�+1�����i�.���The����>�memb��}'ers��c�of��da�tuple��h�x����1��|s�;����:�:�:����;���x����n��	)�are�the��x����i����for��i�l/�=�l01�;��:�:�:����;�n�.���When��y�<�and��d�w��F�are����>t���w�o�&�tuples�of�&�length�at�least��m��w���e�write��E����m�����(�y�[�;���w�D�)�for�the�conjunction�of�the����>form���ulas�UU�E����(�y����i��TL�;���w����i���)�for��i�����m�.����MT��*�o��`pro�Gduce�a��a��-mo�del�w���e�m�ust�dene�an��aop�Geration�����^������,���whic�h��atak�es�a�v��q�ariable����>and�Oan��M��-term.�^�An�O
�M��-term�O(as�explained�in�[�1��],��|p.�86�O
�.�),�is�Oa�term�with����>\parameters�#ffrom�#e�M��";�4more�precisely��*�,�-ba�closed�term�in�a�language�con���taining�a����>constan���t��c����a���k�for�eac�h�elemen�t��a��of��M��.�N�A���con�v�enien�t�notation�for��M��-terms�is��t�[�y�[ٲ],����>where��εy�E��is�a�tuple�of�elemen���ts���of��M��whose�length�is�the�n���um�b�Ger���of�free�v��q�ariables����>of�P!�t�.�pThis�means�P"the�follo���wing:�o-if��x����1��|s�;����:�:�:����;���x����n�����are�the�free�v��q�ariables�of��t�,�Q,in�order����>of���their�o�Gccurrence,���then��t�[�y�[ٲ]�is��t�[�x����i��d�:=���c����y���i����3�].�?�W��*�e�also�need���the�follo���wing�notation:����>�t�[�y�[�;���x�]��<where��;�x��is�a�v��q�ariable,���and��y���is�a�tuple�of�elemen���ts�of��M��V�whose�length�is�the����>n���um�b�Ger�
�of�
�free�v��q�ariables�of��t��dieren���t�from��x�,�means�the�follo���wing:�Nif��x����1��|s�;����:�:�:����;���x����n�����>�are��kthe��lfree�v��q�ariables�of��t��dieren���t�from��x�,��in�order�of�their�o�Gccurrence,��then��t�[�y�[ٲ]����>is�UU�t�[�x����i��d�:=���c����y���i����3�].�q�Note�that��x��ma���y�or�ma�y�not�o�Gccur�free�in��t�.����MThe�UUop�Geration�����^�����9�is�giv���en�b�y���ƍ���t����������(�x;���t�[�y�[�;�x�])��:=�<��2�;���j�R;������d���S�t�����e��*��;�y�"�>����>�where���y�k^�is�a��tuple�whose�length�is�the�n���um�b�Ger��of�free�v��q�ariables�of��x:���t�,�{and��x��is����>the��s�j����-th��rv��q�ariable,��zand����^��d����t���^��e����is�the�(n���umeral�for�the)�G����odel�n���um�b�Ger��sof�the�closed����>term�pZ�t�.���The�denitions�pYof��Ap�,�w�E����,�wand�the�in���terpretation��t�[�y�[ٲ]����M��P��of�eac�h�pZ�M��-term����>�t�[�y�[ٲ]���are���giv���en�in�one�sim���ultaneous�inductiv�e���denition.�?�The�inductiv�e���conditions����>are�UUas�follo���ws.�q�In�(iii)�and�(iv),��m��is�the�n�um�b�Ger�of�free�v��q�ariables�of��t�.��/荑M(i)�UU�E����(�x;���y�[ٲ)�if��x���=��y��.����M(ii)�UU�E����(����^������(�x;���t�[�y�[�;�x�])�;����^�����(�z�p�;�t�[�x���:=��z��][�y�[ٲ])).����M(iii)�UU�Ap����M���\�(����^������(�x;���t�[�y�[�;�x�])�;�r����M���)��=��t�[�x��:=��r�G�][�y�[ٲ]����M���.����M(iv)�{�E����(����^������(�x;���t�[�y�[�;�x�])�;����^�����(�x;�s�[�w�D�;�x�])�{if��E����(�t�[�x����:=����a�][�y�[ٲ]����M���\�;���s�[�x��:=��a�][�w����M���\�),�oand����>�E����m�����(�y�[�;���w�D�)�UUwhere��a��is�a�the�rst�constan���t�not�o�Gccurring�in��t��or��s�.��>�O�ff��v�	J=�����"5��-:�2����LܻIf�`hone�insists�on�in�Îterpreting�`gequalit�y�`has�iden�Îtit�y�`hinstead�of�b�Îy�an�equiv��alence�relation,�w�one��	��ma�Îy��Xuse�the�equiv��alence�classes�of���2cmmi8�E�H0�as�the�elemen�ts�of�the�mo�<rdel.��������16������y�����?������M�(v)�UU�E����(�Ap����M���\�(�a;���b�)�;�Ap����M���(�c;�d�))�UUif��E����(�a;���c�)�and��E��(�b;���d�).����M(vi)�UU�Ap�(�a;���b�)����M���t�=���Ap����M���\�(�a����M���;�b����M���).����M(vii)��K�E����(�f���(�t�[�y�[ٲ])����M���\�;���f��(�s�[�w�D�])����M���)��Jif��K�E����(�t�[�y�[ٲ]����M���;���s�[�w��]����M���),��Mand��Jsimilarly��Kfor�sev���eral�v��q�ari-����>ables�UU�x�.����M(viii)�UU(�x:���t�[�y�[�;�x�])����M���t�=������^������(�x;�t�[�y�;�x�]).����M(ix)�UU�f���(�x�)����M���t�=���f����M���\�(�x����M���)�and�similarly�for�sev���eral�v��q�ariables��x�.����M(x)�UU(�c����a���p�)����M���t�=���a��where��c����a��+Ųis�the�constan���t�for��a�.����M(xi)�UU�E����(�a;���c�)�if��E��(�a;���b�)�and��E��(�b;���c�).������MSince��\�E���and��Ap��o�Gccur�only��]p�ositiv���ely�in�these�clauses,�$�this�is��]a�legitimate����>inductiv���e��denition.���W��*�e��in�terpret�equalit�y�in���M�1òas�the�relation��E����.���F��*�or�eac���h����>predicate�UU�P���of�arit���y��n��in�the�language�of��T�c��,�w�e�dene�a�relation���x䍑�:^������P���yDzon��M�lp�b�y��*�����x䍒�#�^��������P������(�x����1��|s�;����:�:�:����;���x����n��q~�)��:=��M��3�j���UX�=��㎵P�c��(�y����1���;����:�:�:����;���y����n��q~�)�8�^��E����n���(�x;���y�[ٲ)��*���>and��Ww���e�dene���x䍑k^������V�M���B�j���UX�=��㎵P�c��(�x����1��|s�;����:�:�:����;���x����n��q~�)�if�and�only��Vif���x䍑�<^������P���
t�(�x����1���;����:�:�:����;���x����n��q~�).�:�Then�let��V�j���UX�=���v���g��
sԲb�Ge�a����>classical���satisfaction�relation,��"in���tro�Gduced�as�a�xed�p�oin���t�of���the�inductiv�e�clauses����>for�UUsatisfaction.�q�That�mak���es�the�second�(����)�axiom�automatically�satised.����MW��*�e�6�start�with�6�the�follo���wing�lemma:�b�if��E����(�r���;���q�[ٲ)�then��E����(�t�[�x���:=��r�G�]�;���t�[�x��:=��q�[ٲ])�6�for����>terms��t�,���q�[ٲ,��and��r�G�.�Y[This�is�pro���v�ed�b�y�induction�on�the�complexit�y�of�the��M��-term����>�t�.�?6When��ϵt��вb�Gegins�with����or��Ap�,���the�corresp�Gonding�induction�step�follo���ws�from����>(iv)���and���(v).�M.When��t��b�Gegins�with�a�function�sym���b�Gol��f���,�kw�e���use�(vii).�M.When��t����>�is���atomic,���either�it���is�a�constan���t��c����a��x;�for�an�elemen���t��a�F��=��y����1��>�of��M��,���in�whic���h���case����>there��is��nothing�to�pro���v�e,� �or��else�it�is�a�v��q�ariable,� �in�whic���h�case�w���e�ha�v�e��to�pro�v�e����>that�UU�E����(�r���;���q�[ٲ)�from�the�assumption��E��(�r���;���q�[ٲ),�whic���h�is�immediate.����MW��*�e��next��v���erify�the�substitutivit�y��of�equalit�y��*�,�4 namely:��Aif���E����(�r���;���s�)�and���x䍑�>^������M����b�j���UX�=�����>�A�[�x��۲:=��r�G�]�sethen���x䍑.^������M�����j���UX�=��
�Q�A�[�x��۲:=��s�],���where�se�A��is�a�form���ula�of�lam�b�Gda�logic�with����>constan���ts��.for��/elemen�ts�of��/�M��.�SW��*�e�pro���v�e�this��/b�y�induction��/on�the�complexit���y�of����>�A�.��hSince��5substitution��6for�free�v��q�ariables�comm���utes�with�the�logical�connectiv���es����>and�_Fquan���tiers,��}only�the�case�of�atomic��A�_G�needs�a�pro�Gof.��If��A��is�an�equalit���y��t���=��q�[ٲ,����>then���A�[�x��:�:=��;�r�G�]�is��t�[�x��:=��;�r�G�]�=��q�[ٲ[�x��:=��r�G�]��and��A�[�x��:=��;�s�]�is��t�[�x��:=��;�s�]�=��q�[ٲ[�x��:=��s�].����>By���the���lemma�w���e�ha�v�e��޵t�[�x����:=����r�G�]�=��t�[�x��:=��s�]���in��M��,��and��q�[ٲ[�x����:=��r��]�=����q�[ٲ[�x��:=��s�].����>Assume���x䍑*F^�����o��M���-%�j���UX�=��	W�A�[�x���:=��r�G�].���Then�o�b���y�o�(xi)�w�e�o�ha�v�e���x䍑*F^������M���-%�j���UX�=��	W�t�[�x���:=���s�]�=��q�[ٲ[�x��:=��s�].���The����>remaining��case�is��when��A��is�an�atomic�form���ula��P�c��(�x�)�(�x��can�b�Ge�sev���eral�v��q�ariables.)����>This�UUis�tak���en�care�of�b�y�the�denition�of���x䍑�:^������P���yDzab�Go�v�e.����MBecause�Z�of�(ii)�Z�and�(iii),��+axioms�(��	z�)�and�(����)�are�satised.��0No���w�to�v�erify����>axioms�1�(��uDz).�e�Let�1��t�[�y�[ٲ]�b�Ge�an��M��-term,�8�i.e.�e�a�term�with�constan���ts�for�elemen���ts��y����of����>�M��вsubstituted���for�its�free�v��q�ariables.�:�Supp�Gose���x䍑ka^������M���Bw�j���UX�=��㎸8�x�(�t�[�y�[ٲ]�x���=��s�[�y��]�x�).�:�Then���let��c����>�b�Ge�8�the�8�rst�constan���t�not�o�ccurring�8�in��t��or��s�;�B3w���e�ha�v�e���x䍑�N^�����8��M����d�j���UX�=��㎵t�[�x���:=��c�]�=��s�[�x��:=��c�].����>Then�UUb���y�(iv),�w�e�ha�v�e���x䍑^������M�����j���UX�=��㎵x:���t�[�y�[ٲ]��=��x:�s�[�y�[ٲ].�q�Hence�UU(��uDz)�holds�in��M��.����MW��*�e��ha���v�e�no�w�pro�v�ed��that���x䍑��^������M������is�a�mo�Gdel�of�lam���b�Gda�logic.�IW��*�e�still�m�ust�pro�v�e����>it���satises���the�theory��T�鈲in�rst�order�logic.��This�follo���ws�from�the�follo���wing����>lemma:�F�F��*�or���eac���h����M��-form�ula��A�(�x����1��|s�;����:�:�:����;���x����n��q~�)���with��n��parameters�from��M��,��w���e�ha�v�e������x䍑A��^�����>�M���K�1�j���UX�=�����A�(�x�)�`�if�`�and�only�if�there�exists��y��ײwith��E����n��q~�(�x;���y�[ٲ)�and��M��j���UX�=�����A�(�y�[ٲ).���T��*�o�pro���v�e����>the�C�lemma:��The�C�case�when��A��is�an�atomic�form���ula��P�c��(�x�)�is�true�b���y�denition�of���x䍑��^������P���
�.����>The�X�case�of�an�X�atomic�form���ula��t��o�=��q��b�follo�ws�X�from�what�X�has�b�Geen�pro�v�ed�ab�Go�v�e.����>The�5�pro�Gof�5�is�completed�b���y�induction�on�the�complexit���y�of��A�;�@]the�quan���tiers�do�������17������y�����?������>�not��Doer�an���y��Edicult�y��Dsince�the�carrier�sets�of��M��_�and���x䍑��^������M���W2�are�the�same.�ĕThat����>completes�UUthe�pro�Gof�of�the�theorem.��!č��>�9��[�1Lam���b�s3da�ffUnication�����>�Ab�Go���v�e�dw�e�dened��t�[�x�ߓ�:=�ߒ�s�]�das�the�result�of�substituting�term��s��for�free�v��q�ariable����>�x�i��in�i�t�,�oafter�renaming�b�Gound�v��q�ariables�in��t��to�a���v�oid�i�capture�of�free�v��q�ariables�of����>�s�.�F2A���t�Ҕthat�time�ҕw�e�did�ҕnot�extend�this�denition�to�sim���ultaneous�(or�\parallel")����>substitution�Bfor�sev���eral�Av��q�ariables��x����1��|s�;����:�:�:����;���x����n��q~�,�but�this�can�b�Ge�done�as�usual.�YThe����>free�2v��q�ariables�1of��t��are�renamed�to�a���v�oid�clashes�2with�free�v��q�ariables�in��s����1��|s�;����:�:�:����;���s����n��q~�,����>and��then��eac���h�v��q�ariable��x����i��Hвis�replaced�b���y��s����i��Hϲat�the�same�time�(rather�than�se-����>quen���tially).����MA�I��substitution�I��is�a�function�I����βfrom�a�set�of�v��q�ariables�to�the�set�of�terms.����>It�I�is�traditional�to�write��x��w�instead�of�I���[ٲ(�x�).�N�A�I_substitution����has�a�natural����>extension�|�to�|�a�function�from�terms�to�terms,���whic���h�w�e�|�also�denote�b���y���[ٲ,���giv�en����>b���y�&͵t��
�=�$4�t�[�x��:=�$3�x�[ٲ],�[+where��x��stands�&�for�the�list�of�all�free�v��q�ariables�in��t�,�and����>�t�[�x���:=��x�[ٲ]�UUdenotes�a�sim���ultaneous�substitution.����MIn��rst��order�logic,�E�a�substitution���q��is�said�to�b�Ge�a��unier��of�terms��t��and����>�s���if���t��'�=�EN�s�[ٲ.�T�Tw���o�terms�in�rst�order�logic�(with�equalit���y)�are�pro���v��q�ably�equal����>if���and�only�if�they�are�iden���tical,��Uso�it�do�Gesn't�matter�if�w�e�in�terpret�this�to����>mean�G�that��t�"�=���s��u�is�G�pro���v��q�able�in�rst�order�logic�with�equalit�y�G�or�just�that�the����>terms��'�t�F�and��s�F�are�iden���tical.�N
When�w�e�go�to��(generalize�this�to�lam�b�Gda�logic,���it����>do�Ges���matter,���since�terms�can�b�e�syn���tactically���dieren�t���but�pro���v��q�ably�equal�(for����>example�UUif�one�����-reduces�to�the�other).����MW��*�e��dene���z��to�b�Ge�a��lamb��}'da�a�unier��of�terms��t��and��s��if��t�"�=���s�z��is�pro���v��q�able�in����>lam���b�Gda���logic.�/ISimilarly��*�,��K���Y�is�dened�to���b�e�a�lam���b�da�unier�of���form���ulas��A��and����>�B��Ʋif�UU�A���$��B��is�UUpro���v��q�able�in�lam�b�Gda�logic.����MOur��aim�here�is�to��giv���e�another�axiomatization�of�lam�b�Gda�logic,�F�based�on����>resolution,�tfactoring,�and���paramo�Gdulation.�R�The���usual�form���ulation�of�these�rules����>in���v�olv�es�UUuniers.�q�W��*�e�extend�these�rules�to�lam���b�Gda�unication�as�follo�ws:����>�Par��}'amo�dulation��p����c���M��В�=�����P�c��[�x��:=��
��8�]���	z�"�=��
����p���M���g�����`��UW���paramo�Gdulation�������|%еP�c��[�x���:=�����[ٲ]�:�����7��>�pro���vided�[�that�the�free�v��q�ariables�of��������either�o�Gccur�in��
���or�are�not�b�ound�in����>�P�c��.�f�This�4bdiers�4afrom�the�rst�order�v���ersion�of�paramo�Gdulation�only�in�the�extra����>condition��ab�Gout�the�free��v��q�ariables�of�����[ٲ.�)�Note�that�inferences�b���y�����-reduction����>are��Uincluded�in�the�paramo�Gdulation�rule,�0Utaking���I�=�@k��Xq�to�b�e�the�axiom�of����>����-reduction�UUand��
�㍲to�b�Ge���	z�.����>�Binary���R��}'esolution��7����c���M�A�"�$���B��q�[�A�UU�j��U�O���8�B��Ƹj��V���p���M���g��������UW���binary�UUresolution��������2��U���.�j�UU�V�8������������18����2ʠy�����?������>�Here�K�U�e�and��V�9/�are�Jsets�of�literals�and��U�f�j��V�9/�is�their�union;�UƸ�B����is�a�negativ���e����>literal,�^i.e.��a�)negated�)atom;���A��j��U�@0�means�the�union�of��f�A�g��with��U��;���and��U�����>�means�UU�f�P�c��"�:���P�*��2��U��g�.����>�F��;�actoring��7����c���M�A�"�$���B��q�[�A�UU�j�B��Ƹj��U���p���M���g���zY:��UW���factoring�������r���A��.�j�UU�U��������M�W��*�e��1iden���tify�a��2clause�with�the�form�ula��2of���-logic�whic�h�is��2the�disjunction�of����>the�l;literals�l:of�the�clause.��xIf��is�a�set�of�clauses,�q�then��can�also�b�Ge�considered����>as�UUa�set�of�form���ulas�in���-logic.������>�Theorem��T5�(Soundness�of�lam��9b�Q�da�unication)���(���(i)��8Supp��}'ose�ther�e�is�a�pr�o�of����>of�fclause�e�C�҂�fr��}'om�a�set�of�clauses����using�binary�r��}'esolution,�}Efactoring,�and����>p��}'ar�amo�dulation,�]and�4�the�clause�4ӵx�ꟲ=�ꞵx�.�|]Then�ther��}'e�is�a�pr��}'o�of�4�of��C����fr��}'om����in����>lamb��}'da���lo�gic.����>R��}'emark�.�5�In���[�6��],���\demo�Gdulation"�is���men���tioned,�but�it���is�not�necessary�to�consider����>demo�Gdulation��xas��ya�separate�rule�of�inference|from�the�purely�logical�viewp�Goin���t,����>demo�Gdulation�UUis�a�sp�ecial�case�of�paramo�dulation.����>�Pr��}'o�of�.�V�W��*�e��pro�Gceed��b���y�induction�on�the�lengths�of�pro�Gofs,��In�the�base�case,��if�w���e����>ha���v�e�,�a�pro�Gof�of�,�length�0,�4�the�clause��C��ײm���ust�already�b�e�,�presen���t�,�4�in�whic�h�case����>certainly�UU���`��C�q�in�lam���b�Gda�logic.����MF��*�or�Ythe�induction�step,�Yw���e�rst�supp�Gose�the�last�Zinference�is�b�y�paramo�Gdula-����>tion.��Then�fnone�fmof�the�paren���ts�of�the�inference�is�an�equation�����=�㕵�튲(or���j��=�㖵�	z�)����>where��]the��\other�paren���t����has�the�form�� �[ٲ[�x��y�:=��
��8�]�where��]for�some�substitution����>��7��w���e���ha�v�e��ݵ
��8�&�=��N��	z�[ٲ,��~and�the�free�v��q�ariables�of�����7��either�o�Gccur�in��
�j�or�are�not����>b�Gound�gXin�gW� �[ٲ,���according�to�the�denition�of�paramo�Gdulation.���Then�the�newly����>deduced�g�form���ula�is�� �[ٲ[�x�呲:=������].���W��*�e�g�ha���v�e�to�sho�w�g�that�this�form�ula�is�deriv��q�able����>in�-lam���b�Gda�- logic�from�the�paren���ts����and���В�=������.�d`Apply�the�substitution������to�the����>deriv���ed�UUform�ula���.�q�W��*�e�get�������J�������R�=������3� �[ٲ[�x���:=��
��8�]����������R�=������3(� �[���)[�x���:=��
��8�[ٲ]���������R=������3(� �[���)[�x���:=���	z�[ٲ]������>No���w�G�using�G�the�other�deduced�equation���d�=�[n����,���w���e�can�deduce���	z��F�=�[m�����ɲand����>hence��`w���e��acan,��according�to�the�rules�of�lam���b�Gda�logic,��substitute������:�for���	z�[ٲ,����>pro���vided��cthe�free��bv��q�ariables�in�����%<�either�o�Gccur�already�in���	z�%<�or�are�not�b�Gound����>in���� �[���.���Since����
��Ų=����	z��,���this�is�exactly�the�condition���on�the�v��q�ariables�that�mak���es����>the���application�of�paramo�Gdulation�legal.��That�completes�the�induction�step�in����>the�UUcase�of�a�paramo�Gdulation�inference.����MIf��/the�last��0inference�is�b���y�factoring,��jit�has�the�form�of�applying�a�substitution����>��:��to���a���previous�clause.�JKThis�can�b�Ge�done�in�lam���b�Gda�logic.�JL(In�the�factoring�rule,����>as�Nin�lam���b�Gda�logic,�O�substitution�m�ust�b�Ge�dened�Nso�as�to�not�p�ermit�capture�of����>free�UUv��q�ariables�b���y�a�binding�con�text.)�������19����A�y�����?������M�If�c�the�c�last�inference�is�b���y�binary�resolution,��uthen�the�paren���t�clauses�ha���v�e����>the���form����P�c��j�R����and���Q�j�S�/��(using��R����and��S�/��to�stand�for�the�remaining�literals�in����>the�-�clauses),�c�and�-�substitution������unies��P��H�and��Q�.���The�newly�deduced�clause����>is�lthen��R���[ٸj�S�����.��$Since�lthe�unication�steps�lare�sound,��ѵP�c���=����Q���is�pro���v��q�able����>in�퍵�-logic.�:oBy�induction��h���yp�Gothesis,����-logic�pro�v�es��b�Goth��P�c��j�R�T�and���Q�j�S���from����>assumptions�B0,�y8and�since�the�substitution�rule�is�v��q�alid�in���-logic,�it�pro���v�es�B0�P�c��[ٸj�R������>�and��1��Q�[ٸj�S�����.�Z[But�since��2�P�c��2^�=�ֆ�Q�T
�is�pro���v��q�able,� ��-logic�pro�v�es��R���[ٸj�S����T
�from�.����>This�UUcompletes�the�induction�step,�and�hence�the�pro�Gof.�� �ލ��>�10��^fdThe�ffLogic�of�P���artial�T���ferms�����>�In���the���group�theory�example,��it�is�natural�to�ask�whether��x�	5���y�鮲needs�to���b�Ge����>dened�̳if��x�̲�or��y�(��do�Ges�not�satisfy��G�(�x�).�D<In�rst�order�logic,�����is�a�function�sym���b�Gol����>and��lhence��kin�an���y�mo�Gdel�of�our�theory�in�the�usual�sense�of�rst�order�logic,�1�����>�will�j�b�Ge�in���terpreted�as�a�j�function�dened�for�all�v��q�alues�of��x��and��y��n�in�the�mo�Gdel.����>The�>Eusual�w���a�y�>Eof�>Dhandling�this�is�to�sa���y�that�the�v��q�alues�of��x�
����y���for��x�>E�and��y���not����>satisfying���G�(�x�)�or��G�(�y�[ٲ)�are�dened�but�irrelev��q�an���t.�ɭF��*�or�example,��in�rst�order����>eld�1theory��*�,��1�=�0�is�0dened,�but�no�axiom�sa���ys�an�ything�ab�Gout�0its�v��q�alue.�YfAs�this����>example�y�sho���ws,��the�problem�of�\undened�terms"�is�y�already�of�in�terest�in�rst����>order�U�logic,�Vand�U�t���w�o�dieren�t�U�(but�related)�logics�of�undened�terms�ha���v�e�b�Geen����>dev���elop�Ged.�V]W��*�e���explain���here�one�w�a�y�to�do���this,���kno�wn�as���the��L��}'o�gic��of��Partial����>T��;�erms�UU�(LPT).�See�[�5��]�or�[�4��],�pp.�q�97-99.����MLPT��$has��a�term-formation��op�Gerator��#�,�T�and�the�rule�that�if��t��is�a�term,����>then���t�\��#��is�an�atomic��
form���ula.�~�One�migh�t,��|for�example,�form���ulate�eld�theory����>with�F]the�axiom��y����6�=�X�0�Xθ!��x=y��#�F^�(using�F]inx�notation�for�the�quotien���t�term).����>Thereb���y�(�one�(�w�ould�a�v�oid�the�(sometimes)�(�incon�v�enien�t�ction�that�1�=�0�(�is�some����>real�s�n���um�b�Ger,�{
but�s�it�do�esn't�s�matter�whic���h�one�b�Gecause�w���e�can't�pro���v�e�an�ything����>ab�Gout�
it�an���yw�a�y;�z�man�y�computerized�mathematical�systems�mak���e�use�of�this����>ction.�G�T��*�aking��this��approac���h,��\one�m�ust��then�mo�Gdify��the�quan�tier��axioms.�G�The����>t���w�o�UUmo�Gdied�axioms�are�as�follo���ws:���X�����e_�8�x���A�8�^��t���#!��A�[�x��:=��t�]���������e_�A�[�x���:=��t�]�8�^��t���#!�9�x���A�����W��>�Th���us�<�from�<�\all�men�are�mortal",�A�w���e�are�not�able�to�infer�\the�king�of�F��*�rance�is����>mortal"�p+un���til�w�e�sho�w�p,that�there��is��a�king�of�F��*�rance.��JThe�other�t�w�o�quan�tier����>axioms,�_�and�]�the�prop�Gositional�axioms,�of�rst�order�logic�are�]�not�mo�Gdied.���W��*�e����>also�UUadd�the�axioms��x���#��for�ev���ery�v��q�ariable��x�,�and��c��#��for�eac���h�constan�t��c�.����MIn�5LPT,�w���e�do�not�6assert�an�ything�in�v�olving�undened�terms,�onot�ev�en�that����>the���king�of�F��*�rance�is�equal�to�the�king�of�F�rance.�WThe�w���ord�\strict"�is�applied����>here��to�indicate��that�subterms�of�dened�terms�are�alw���a�ys��dened.�?
LPT���has�the����>follo���wing�c\strictness�caxioms",�f�for�ev�ery�catomic�form�ula��R�v�and�cfunction�sym�b�Gol����>�f���.�q�In�UUthese�axioms,�the��x����i�����are�v��q�ariables�and�the��t����i���are�terms.�������ڵR�Dz(�t����1��|s�;����:�:�:����;���t����n��q~�)���!��t����1��C��#�^���:�:�:���/�^�8�t����n��8��#���������ڵf���(�t����1��|s�;����:�:�:����;���t����n��q~�)���#!��t����1��C��#�^���:�:�:���/�^�8�t����n��8��#����������20����P�y�����?���������ڵt����1��C��#��^�����:�:�:���/�^�8�t����n��8��#�^�f���(�x����1��|s�;����:�:�:����;���x����n��q~�)��#!��f��(�t����1��|s�;����:�:�:����;���t����n��q~�)��#������>�R��}'emark�.�L-In��LPT,�while��terms�can�b�Ge�undened,��form���ulas�ha�v�e�truth��v��q�alues�just����>as��in��ordinary�logic,�!�so�one�nev���er�writes��R�Dz(�t�)���#��for�a�relation�sym���b�Gol��R�Dz.�\DThat�is����>not�UUlegal�syn���tax.����MF��*�or�UUexample,�one�of�the�strictness�axioms�is�����`��t���=��r�5�!��t��#�^�r��#��:����>�W��*�e���write��t����T͍�������+3�����=�����X��r�7�to�abbreviate��t���#��_�r�޸#!��t��=��r�G�.�A�It���follo���ws���from�the�strictness����>axiom�d�just�d�stated�that��t����T͍���~����+3����~�=�����
��r����really�means�\�t��and��r����are�b�Goth�dened�and�equal,����>or�UUb�Goth�undened."����MThe��equalit���y��axioms�of�LPT�#are�as�follo���ws�(in�addition�to�the�one�just����>men���tioned):��������n�x���=��x����������nx���=��y�"�!��y��=��x����������nt����T͍�������+3�����=�����
UN�r���^�8��[�x���:=��t�]��!���[�x��:=��r�G�]�����!č��>�11��^fdP���artial�ffLam�b�s3da�Calculus�����>�In�P�lam���b�Gda�Pcalculus,�Qxthe�issue�of�undened�terms�arises�p�Gerhaps�ev���en�more�nat-����>urally��,than�in�rst��-order�logic,�$�as�it�is�natural�to�consider�partial�recursiv���e����>functions,�UUwhic���h�are�sometimes�undened.����^��3������M�Partial�^�lamb��}'da�^�c�alculus���is�a��system�similar�to�lam���b�Gda�calculus,�'but�in�whic���h����>�Ap�Α�is�Βnot�necessarily�total.�D�There�can�then�b�Ge�\undened�terms."�D�Since�lam���b�da����>calculus�F�is�F�a�system�for�deducing�(only)�equations,�|�the�system�has�to�b�Ge�mo�died.����>W��*�e�'#no���w�'"p�Germit�t�w�o�'"forms�of�statemen�ts�'"to�b�Ge�deduced:�b�t����T͍��$¸���+3���$²=�������r�n?�and��t�$¸#�.��0The����>axioms�(��	z�),�PJ(����),�and�(��uDz)�are�mo�Gdied�b���y�c�hanging�=�to����T͍������+3����=������7,�PJand�the�rules�for����>deducing�Lc�t���#�Lb�are�sp�Gecied�as�follo���ws:��MFirst,��`w�e�Lccan�alw�a�ys�Lcinfer�(without�premise)����>�x���#�UU�when��x��is�a�v��q�ariable.�q�Second,�w���e�can�apply�the�inference�rules���j��������<$������t����T͍�������+3�����=�����
UN�s�t���#������w�fe5M�	(֍�G�s���#�������������<$��	)n�t����T͍�������+3�����=�����
UN�s�s���#��	)n�w�fe6۟	(֍�Z��t���#���������r��������<$���2�t���#���r�5�#���2�w�fe.�ҟ	(֍��9�t�[�x���:=��r�G�]��#����������;�x:���t���#�����	��>�Note�>:that�w���e�do�not�ha�v�e�strictness�of��Ap�.��As�an�example,�v
w�e�ha�v�e��Ap�(�y�[�:���a;�b�)����T͍�������+3�����=��������>�a�,�Fkwhether�B�or�not��b���#�.�k�W��*�e�could�ha���v�e�B�form�ulated�a�rule�\strict�(����)"�that�w�ould��>��f�ff��v�	J=�����"5��-:�3����LܻThere��is,��Tof�course,�an��alternativ�Îe�to���-calculus�kno�wn�as���##�f�cmti8�c���ombinatory�_lo�gic�.���Application��	��in�o�com�Îbinatory�logic�is�also�total,��but�in�[�4��@],��~the�notion�of�a��p���artial��c�ombinatory��algebr�a�oݻis���in�Îtro�<rduced��and�studied,�
�follo�wing��F��J�eferman,�who��in�[�7��@]�rst�in�Îtro�<rduced�partial�applicativ�e���theories.���See��L[�8��@]�for��Msome�relationships�b�<ret�Îw�een�partial��Lcom�binatory�logic�and��Mpartial�lam�b�<rda���calculus��������21����b-�y�����?������>�require���deducing��r�g��#����b�Gefore�concluding��Ap�(�x:���t;�r��)����T͍�� �����+3��� ��=�����$�t�[�x� ��:=� ��r��],��fbut���not�requir-����>ing�7Jstrictness�corresp�Gonds�b�etter�to�7Ithe�w���a�y�7Jfunctional�programming�languages����>ev��q�aluate�56conditional�57statemen���ts.�gNote�also�that��x:���t��is�dened,�;�whether�or�not����>�t�UU�is�dened.��!č��>�12��^fdP���artial�ffLam�b�s3da�Logic�����>�Partial��?lamb��}'da�lo�gic�V:�results�V;if�w���e�mak�e�similar�mo�Gdications�V;to�lam�b�Gda�logic����>instead��iof�to�rst�order�logic�or�lam���b�Gda�calculus.��In�particular�w�e�mo�Gdify�the����>rules�Fof�logic�and�the�equalit���y�axioms�as�Fin�LPT,�add�the�strictness�axiom�(except����>for�<�Ap�),�9�and�mo�Gdify�;the�axioms�(��	z�),�(����),�and�(��uDz)�b���y�replacing�;=�with����T͍������+3����=������Z.��{In����>LPT,����T͍��4<����+3���4<�=�����/�is�4<an�abbreviation,�:�not�an�4;ocial�sym���b�Gol;�?Din�partial�lam�b�Gda�4;calculus�it����>is��+an�ocial��*sym���b�Gol;�in�partial�lam���b�Gda�logic�w�e�could��*mak�e�either�c�hoice,�� but����>for�^�deniteness�w���e�c�ho�Gose�to�^�mak�e�it�an�ocial�sym�b�Gol,�adso�that�partial�lam�b�Gda����>logic���literally���extends�b�Goth�LPT���and�partial�lam���b�da���calculus.�n�The�rst���three����>rules��of�inference�listed�ab�Go���v�e��for�partial�lam���b�da�calculus�are�sup�er
uous�in�the����>presence�j�of�j�LPT.�The�fourth�one�(actually�an�axiom,�p$not�a�rule)�is�included�in����>partial�UUlam���b�Gda�logic.����MHere�UUfor�reference�are�the�axioms�of�partial�lam���b�Gda�logic:����M(�����terms�always�dene��}'d�)�UU�x:���t���#��m�for�eac���h�term��t��GqȲ.����M(�Pr��}'op�)�UUprop�Gositional�axioms�(see�[�10��
],�p.�q�20)����M(�Q�)�UUquan���tier�axioms�for�LPT�as�giv�en�ab�Go�v�e����M(�Strictness�)���������R�Dz(�t����1��|s�;����:�:�:����;���t����n��q~�)���!��t����1��C��#�^���:�:�:���/�^�8�t����n��8��#�����������f���(�t����1��|s�;����:�:�:����;���t����n��q~�)���#!��t����1��C��#�^���:�:�:���/�^�8�t����n��8��#����if�UU�f�h�is�not��Ap�����������t����1��C��#��^�����:�:�:���/�^�8�t����n��8��#�^�f���(�x����1��|s�;����:�:�:����;���x����n��q~�)��#!��f��(�t����1��|s�;����:�:�:����;���t����n��q~�)��#�������M�(��	z�)�UU�t����T͍�������+3�����=�����
UN�s��if��t��and��s��are�alpha-equiv��q�alen���t.����M(����)�UU�Ap�(�x:���t;�s�)����T͍�������+3�����=�����
UN�t�[�x���:=��s�]����M(����)�UU�Ap�(�x:���B��q;�s�)���$��B��[�x��:=��s�]�UUwhen��B��Ʋis�a�form���ula.����M(��uDz)�8�(�we��}'ak��8extensionality�)��8�x�(�Ap�(�t;���x�)����T͍�������+3�����=�����
UN�Ap�(�s;�x�))���!��x:���Ap�(�t;�x�)����T͍�������+3�����=������x:�Ap�(�s;�x�)�����M(�true���and�false�)�UU�T���=��xy�[�:���x��and��F��=��xy�[�:���y����M�(�non-triviality�)�UU�T���6�=��F����M�W��*�e���review���the�seman���tics�of�LPT���as�giv���en�in�[�5��],��"[�4��].�8�A���mo�Gdel���consists�of�a�set����>and�?`relations�on�that�set�?_to�in���terpret�the�predicate�sym�b�Gols;��the�function�sym�b�Gols����>are�oNin���terpreted�oMb�y�partial�oMfunctions�instead�of�total�functions.���Giv���en�suc�h�oNa����>�p��}'artial��wstructur�e��J�one��Kdenes�sim���ultaneously��*�,��b�y��Kinduction�on�the�complexit���y�of����>terms�;?�t�,�@vthe�;>t���w�o�notions�;>�M��3�j���UX�=��㎵t���#��and��t����M���\�,�@vthe�elemen���t�of��M�RY�that�is�denoted�b���y����>�t�.�������22����pY�y�����?������M�W��*�e���no���w���discuss�the�seman���tics�of�partial�lam���b�Gda�logic.�8OThe�denition�of����>partial��a��-mo�Gdel�is�similar�to�that�of���-mo�del,�Ncexcept�that�no���w��Ap��and�the����>other��4function��5sym���b�Gols�can�b�Ge�in���terpreted�b�y��4partial�functions�instead�of�total����>functions.��7The�Z%function�����^�����	�in�Z&the�denition�of���-mo�Gdel�(whic���h�tak�es�a�v��q�ariable����>and�=~an�=}�M��-term�as�argumen���ts)�is�required�to�b�Ge�total,�BBso�that�the�axiom��x:���t���#����>�will�UUb�Ge�satised.������>�Denition��T2�(P��9artial�lam�b�Q�da�mo�del)����u�(�M���;������^������)���is��a�p��}'artial���-mo�del��of�a�the-����>ory��N�T�R��in��Mp��}'artial�lamb�da�lo�gic�if��M�(�M���;������^������)��satises�the�axioms��(��	z�)�,�(�(��uDz)�,�'and��(����)�,����>and�1%�M�HA�satises�al���l�1&the�axioms�of��T����and��LP�c�T��,�D�exc��}'ept�1%that��Ap��ne��}'e�d�1%not�b�e�1%strict.����M�The��Sfollo���wing��Ttheorem�generalizes�the�completeness�theorem�for�LPT���to����>partial�UUlam���b�Gda�logic.����^��4��������>�Theorem��T6�(Lam��9b�Q�da�Completeness�Theorem�for�LPT)���T��L��}'et�?/�T����b�e�?0a�c�on-����>sistent���the��}'ory�in�p�artial�lamb�da�lo�gic.���Then��T��v�has�a�p�artial�lamb�da�mo�del.����>Pr��}'o�of�.��+As���for���(total)�lam���b�Gda�logic,�()w�e���ha�v�e�to���extend�partial�lam���b�Gda�logic����>to���allo���w�Sk�olem���functions,��4not�allo�wing���the�construction�of���lam�b�Gda�terms�that����>capture�jDfree�v��q�ariables�in�jESk���olem�terms.�#lExactly�as�for�lam���b�Gda�logic,��Hev�ery�theory����>�T�D:�in��partial�lam���b�Gda��logic�has�a�Sk�olemized�v�ersion��T��c���^������in�Sk�olem�partial�lam�b�Gda����>logic,��suc���h��jthat��k�T���has�a�mo�Gdel�if�and�only�if��T��c���^�����ݲhas�a�mo�Gdel.�=$It�therefore�suces����>to�UUpro���v�e�the�completeness�of�quan�tier-free�Sk�olem�partial�lam�b�Gda�logic.����MT��*�o���do�that,���w���e�again�imitate�the�Henkin�pro�Gof�of���completeness�for�rst�order����>logic.�n0If��x�T�
�do�Ges�not�con���tain�innitely�man�y�constan�t��ysym�b�Gols,���w�e�b�Gegin�b�y����>adding��fthem;�'nthis�do�Ges�not�destro���y��ethe�consistency�of��T�D��since�in�an���y�pro�Gof�of����>con���tradiction,���w�e�Nbcould�Nareplace�the�new�constan���ts�b�y�Nav��q�ariables�not�o�Gccurring����>elswhere���in�the���pro�Gof.���W��*�e�construct���the�\canonical�structure"��M���for�a�theory����>�T�c��.�1mThe��elemen���ts�of�ꍵM���are�equiv��q�alence�classes�of�closed�terms�of��T�N�under�the����>equiv��q�alence�]{relation�of�]zpro���v�able�equalit���y:���t�Ԭ���r����i�]{�T�8;�`��t�ԭ�=��r�G�.��8But�no���w,�_�w�e�tak�e����>only��those��closed�terms��t��for�whic���h��T�Y��pro�v�es��t���#�;��that�ensures��the�v��q�alidit�y��of�the����>axiom�Ԗ�x��.�=��-�x�,��fand�of�the�ԕaxioms��x��#��for�v��q�ariables�ԕ�x��and�of�the�axioms��c��.�#��for����>constan���ts�qp�c�.��Let�[�t�]�b�Ge�the�equiv��q�alence�qqclass�of��t�.�W��*�e�dene�the�in���terpretations����>of�UUconstan���ts,�function�sym�b�Gols,�and�predicate�sym�b�Gols�in��M�lp�as�follo�ws:������֟��c�����M�������Ӽ�=��������[�c�]����������:�f�������M��	��([�t����1��|s�]�;����:�:�:����;����[�t����n��q~�])������Ӽ=��������[�f���(�t����1��|s�;����:�:�:����;���t����n��q~�)]�����������P��c�����M��
C�([�t����1��|s�]�;����:�:�:����;����[�t����n��q~�)]������Ӽ=��������[�P�c��(�t����1��|s�;����:�:�:����;���t����n��q~�)]������>In�G�this�G�denition,�JGthe�righ���t�sides�dep�Gend�only�on�the�equiv��q�alence�classes�[�t����i��TL�]�(as����>sho���wn��in�[�10��
],��p.��44).��The�atomic�form�ula��t����#��is�satised�in��M��²if�and�only�if����>�T����pro���v�es�Hk�t���#�.�mzIt�follo���ws�Hlb�y�Hkinduction�on�the�complexit���y�of�the�closed�rst-order����>term��	�t��
�that�if��T��pro���v�es��	�t���#��then�the�in���terpretation��t���^��M��ne�of��t��in��M��$�is�the�equiv��q�alence����>class��[�t�]�of��t�.�Q�The�strictness�axiom��for�rst�order�function�sym���b�Gols�is�used�in�the��>��f�ff��v�	J=�����"5��-:�4����LܻIn���[�5��@]���there�is�a�completeness�theorem�for�LPT,�generalizing�G����odel's�completeness�theorem.��	��The��Xstrictness�axiom�is�imp�<rortan�Ît�in�the�pro�of.��������23�����y�����?������>�induction��hstep,�!-so�this��ido�Ges�not�apply�to�terms�con���taining��Ap��unless�w���e�ha�v�e����>strictness�UUfor��Ap�.����MExactly�B�as�B�in�[�10��
]�one�then�v���eries�that��M�YҲis�a�rst�order�mo�Gdel�of��T��E�in����>the�ysense�y~of�LPT.�Th���us�the�completeness�theorem�for�rst�order�LPT�y3(with����>strictness)�UUis�pro���v�ed�UU(again{it�w���as�rst�pro�v�ed�in�[�5��]).����MT��*�o��turn��M��"�in���to�a����-mo�Gdel,��tw�e�m�ust��dene�����^������(�x;���t�),��swhere��t��is�an��M��-term,����>i.e.��a�[closed�term�with�parameters�[from��M��.��The�\parameters�from��M��"�are����>constan���ts��R�c���:�[��Ï�q�[ٲ],�ؑwhere�[�q��]��Sis�the�equiv��q�alence�class�of�a�closed�term��q�+�of��T�c��.���If�[�t�]����>is�rthe�equiv��q�alence�class�of�an�r
�M��-term��t�,�y9let�[�t�]���^����
�b�Ge�a�closed�term�of��T�՚�obtained����>from�0��t��b���y�replacing�eac�h�constan�t��c���:�[��Ï�q�[ٲ]�b�y�0�some�closed�term��q��ղin�the�equiv��q�alence����>class�M.[�q�[ٲ].�oWhic���h�closed�term�w�e�select�do�Ges�not�M-aect�the�equiv��q�alence�class�[�t�]���^������.����>In�UUother�w���ords,�[�t�]���^�����9�is�a�w�ell-dened�op�Geration�on�equiv��q�alence�classes.����MW��*�e��Udene��T����^������(�x;���t�)�;j=�[�x:��[�t�]���^������].���Because��Uof�the��Taxiom��x:�t�;i�#�,�,�����^������(�x;�t�)�is����>indeed�߄dened�for��M��-terms��t�.�TW��*�e�m���ust�sho�w�that�the�v��q�alue�dep�Gends�only�on����>the�޴equiv��q�alence�class�[�t�].�J<Supp�Gose��T�BC�pro���v�es�޵�t���=��s�.�Then�޴b���y�axiom�(�xi�),��n�T�BC�pro�v�es����>�x:����[�t�]��=��x:��[�s�].����MW��*�e���v���erify�that�the�axioms��of�partial�lam�b�Gda�logic�hold�in��M��.�M�First,��ythe�b�eta����>axiom:���Ap�(�x:���t;�r�G�)����T͍�������+3�����=�����
�Z�t�[�x���:=��r��].���It�m�suces�to�m�consider�the�case�when��t��has�only����>�x�$�free.��3Supp�Gose�that�the�#left�hand�side�is�dened�in��M��.��3Its�in���terpretation�is����>the���equiv��q�alence�class���of��Ap�(�x:���t;�r�G�).�dfBy�axiom���(����),�%the�righ���t�side�is�pro���v��q�ably����>equal�1 to�1!the�left�side,�8^so�it�is�dened�in��M��,�8^and�its�in���terpretation�is�the�class�of����>�t�[�x�t=�:=�t>�r�G�].��qOn��9the�other��8hand�if�the�righ���t�hand�side�is�dened�in��M��,��2then�it�is����>pro���v��q�ably���equal���to�the�left�side�and�so�b�Goth�are�dened�and�equal�in��M��.�CThis����>v���eries�UUaxiom�(����).����MNo���w�J?for�J@axiom�(��uDz).�P�Supp�Gose��t��and��s��are�closed�terms�dened�in��M��,��zand����>�Ap�(�t;���x�)����T͍��p�����+3���p��=������M�Ap�(�s;�x�)�T�is�T�v��q�alid�in��M��.�o�Then�for�eac���h�closed�term��r����suc���h�that��T����>�pro���v�es��;�r�״�#�,��tw�e�ha�v�e��tr�X�pro�v��q�ably�equal�to��sr�G�.��zSince��T�1ʲcon�tains�innitely�man�y����>constan���t��sym�b�Gols,��w�e�can�select�a�constan�t���c��that�do�Ges�not�o�ccur�in���t��or��s�,����>so�d˵tc����=����sc��is�pro���v��q�able.��)Replacing�the�constan�t��c�d̲b�y�a�v��q�ariable�in�the�pro�Gof,����>�tx�kP�=��sx��ݲis���pro���v��q�able.��`Hence�b�y���axiom�(��uDz),�Ѐ�x:���t�kO�=�kP�x:�s��is���probable,�Ѐand�hence����>that��equation�holds�in��M��.�L�In�v���erifying�axiom�(��uDz),���it�suces�to�consider�the�case����>when���s���and��t��are�closed�terms.�@�The�axiom�(��	z�)�holds�in��M���since�it�simply�asserts����>the�{�equalit���y�{�of�pairs�of�pro���v��q�ably�equal�terms.��The�axiom��T�*�6�=��F��U�holds�since�{ƵT����>�do�Ges�UUnot�pro���v�e�UU�T���=��F�,�b�Gecause��T���is�consisten���t.�q�That�completes�the�pro�of.��!č��>�13��^fdP���artial�@�lam�b�s3da�logic�in�terms�of�resolution����^fdand�ffparamo�s3dulation�����>�In�%�the�%�resolution�and�paramo�Gdulation�axiomatization�of�lam���b�Gda�logic�(and�rst����>order��klogic��lfor�that�matter)�w���e�use�the�axiom��x���=��x�,��since�paramo�Gdulation��kdo�es����>not�Peinclude�re
exivit���y��*�.�b�In�partial�Pflam�b�Gda�logic,��(w�e�replace�Pfthe�axiom��x�i��=��x����>�b���y��8the��9clause���E����(�x�)�j�x���=���x�,�	1where��E��(�t�)��9is�a�syn���tactic�v��q�arian�t��9of��t���#�.�!rW��*�e�also����>add�p-the�p,clause��x��ո6�=��Եx�j�E����(�x�),�v�th���us�expressing�that��t��ո#��is�equiv��q�alen���t�to��t��Բ=��յt�.��NThe����>soundness�UUtheorem�tak���es�the�follo�wing�form:�������24������y�����?��������>�Theorem��T7�(Soundness�of�lam��9b�Q�da�unication�for�LPT)������M�Supp��}'ose��ther�e��is�a�pr�o�of��of�clause��C����fr�om�a�set�of��clauses����using�binary�r��}'es-����>olution,�q�factoring,�q�demo��}'dulation�E^(including�����-r�e�duction),�q�and�p�ar�amo�dulation,����>the���clauses��x�޲=��x�j�Vb��E����(�x�)����and�����E��(�x�)�j�x�ݲ=�޵x�,���and�clauses���expr��}'essing�the�strict-����>ness��axioms,�!Tal���lowing�se��}'c�ond��or�der�unic�ation�in�plac�e�of�rst�or�der�unic�ation.����>Then���ther��}'e�is�a�pr�o�of�of��C�K�fr�om����in�p�artial�lamb�da�lo�gic.����>Pr��}'o�of�.�,�The���pro�Gof�is�similar�to�the�pro�of�for�lam���b�da�logic,��8except�for�the�treat-����>men���t�{�of�{�����-reduction�steps.��When����-reduction�{�is�applied,��2w���e�only�kno�w�{�that����T͍������+3����=��������>holds�
in�partial�lam���b�Gda�logic.���But�since�in�partial�lam���b�Gda�logic,�2:w�e�ha�v�e�the����>substitutivit���y�Baxioms�for����T͍������+3����=�����
ǣas�w�ell�as�for�=,�Fit�is�still�the�Ccase�in�partial�lam�b�Gda����>logic���that�if��P�c��[�x�<�:=�<�Ap�(�z�p�:���t;�r�G�)]���is�used�to�deduce��P��[�x�<�:=��t�[�z����:=��r�G�]],��and���if�(b���y����>induction��uh���yp�Gothesis)�the��vformer�is�deriv��q�able�in�partial�lam�b�Gda��vlogic�plus�A�C,����>then�7so�7is�the�latter.�g�F��*�or�example�if��Ap�(�z�p�:���a;��
)��=��a�7�is�deriv��q�able,�=*then��a��=��a�7�is����>deriv��q�able.�q�Eac���h�UUof�these�form�ulas�is�in�fact�equiv��q�alen�t�to��a���#�.��!č�>�References�������>�[1]���M�;Barendregt,�8YH.,�8X�The�:�L��}'amb�da�Calculus:��Its�Syntax�and�Semantics�,�8YStudies����M�;in�zLogic�and�the�F��*�oundations�zof�Mathematics��103�,��GElsevier�Science�Ltd.����M�;Revised�UUedition�(Octob�Ger�1984).������>[2]���M�;Barendregt,�w+H.,�w,Bunder,�M.,�and�=4Dekk���ers,�W.,�Systems�of�illativ���e�com�bi-����M�;natory���logic�complete�for�rst�order�prop�Gositional�and�predicate�calculus����M�;�Journal���of�Symb��}'olic�L�o�gic�UU�58��(3),�89-108,�1993.������>[3]���M�;Barendregt,���H.,�Bunder,�M.,�and�rbDekk���ers,�W.,�Completeness�of�t���w�o�sys-����M�;tems�(�of�(�illativ���e�com�binatory�(�logic�for�rst�order�prop�Gositional�and�predicate����M�;calculus�UU�A���r��}'chive���f� �����ur�Mathematische�L�o�gik�UU�37�,�327{341,�1998.������>[4]���M�;Beeson,�EtM.,�Es�F��;�oundations��wof�Constructive�Mathematics�,�Springer-V��*�erlag,����M�;Berlin�UUHeidelb�Gerg�New�Y��*�ork�(1985).������>[5]���M�;Beeson,��M.,��Pro���ving���programs�and���programming�pro�Gofs,�in:�h�Barcan,�Mar-����M�;cus,�$pDorn,�and��W��*�eingartner�(eds.),�$p�L��}'o�gic,�RpMetho�dolo�gy,�Rqand�,UPhilosophy�of����M�;Scienc��}'e���VII,�pr�o�c�e�e�dings���of�the�International�Congr�ess,��@Salzbur�g,�1983�,��Fpp.����M�;51-81,�UUNorth-Holland,�Amsterdam�(1986).������>[6]���M�;Beeson,��M.,�Lam���b�Gda���logic,�in���Basin,�Da�vid;��GRusino�witc�h,�Mic�hael���(eds.)���Au-����M�;tomated�C>Reasoning:�M�Second�C=In���ternational�Join�t�C=Conference,�~�IJCAR�C2004,����M�;Cork,�!`Ireland,�!aJuly�c4-8,�2004,�Pro�Gceedings.�cLecture�Notes�din�Articial�In���tel-����M�;ligence�UU3097,�pp.�460-474,�Springer�(2004).������>[7]���M�;S.�w.F��*�eferman,��Constructiv���e�theories�of�functions�w-and�classes,�pp.�159-224�in:����M�;M.���Boa,�c'D.���v��q�an�Dalen,�and�K.���McAlo�Gon�(eds.),��L��}'o�gic��Col���lo�quium��'78:����M�;Pr��}'o�c�e�e�dings�R�of�R�the�L��}'o�gic�Col���lo�quium�R�at�Mons�,�X�North-Holland,�X�Amsterdam����M�;(1979).�������25�����o�y�����?��������>�[8]���M�;E.�bjMoggi.�biThe�P���artial�Lam�b�Gda-Calculus.�PhD�b$thesis,���Univ�ersit�y�biof�Edin-����M�;burgh,�UU1988.�h���ttp://citeseer.nj.nec.com/moggi88partial.h�tml������>[9]���M�;Scott,�\'D.,�Iden���tit�y�Z�and�existence�in�Z�in���tuitionistic�logic,�\'in:�|�F��*�ourman,�M.�Z�P�.,����M�;Mulv���ey��*�,���C.�|UJ.,�and�Scott,�D.�S.�|T(eds.),��Applic��}'ations��Dof�She�aves�,���Lecture�|UNotes����M�;in�5�Mathematics�5��753��660-696,�;�Springer{V��*�erlag,�;�Berlin�Heidelb�Gerg�New�Y��*�ork����M�;(1979).������>[10]���R�<Sho�Geneld,�p�J.�
R.,��Mathematic��}'al�!L�o�gic�,�p�Addison-W��*�esley�,�p�Reading,�Mass.����M�;(1967).�������26�����7���;�y�
�##�f�cmti8��"V

cmbx10���<x

cmtt10��':

cmti10���N�ffcmbx12�t�:		cmbx9�o���		cmr9�q�%cmsy6��K�cmsy8��2cmmi8��Aa�cmr6�|{Ycmr8�X�Qcmr12�D��tG�G�cmr17�
!",�

cmsy10�O!�cmsy7�
�b>

cmmi10�	0e�rcmmi7�O
�\cmmi5�K�`y

cmr10�ٓ�Rcmr7��5�����

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