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/Circumscription.dvi

����;� TeX output 2001.01.10:1505������o#f����ܚ�5#f����ܚ��RYu���N�ffcmbx12�A�ffSecond-order�Theorem�Pro���v�er�ffapplied�to������Circumscription��"�[��ʶ��K�`y

cmr10�Mic���hael�UUBeeson���i���a>�o���		cmr9�Departmen��9t�Tof�Mathematics�and�Computer�Science������ZSan�TJose�State�Univ��9ersit�y��)�����[s��t�:		cmbx9�Abstract.����&��Circumscription��eis�naturally�expressed�in�second-order�logic,����[s�but���previous�implemen��9tations�all�w�ork�b�y�handling�cases�that�can�b�A�e�re-����[s�duced�9to�rst-order�logic.�Making�use�of�a�new�second-order�unication����[s�algorithm�Džin��9tro�A�duced�in�[2],�w�e�sho�w�ho�w�a�theorem�pro�v�er�can�b�A�e�made����[s�to��nd�pro�A�ofs�in�second-order�logic,�in�particular�pro�ofs�b��9y�circumscrip-����[s�tion.�P8W��:�e�w��9ork�out�a�blo�A�c�ks-w�orld�example�in�complete�detail�and�giv�e�the����[s�output�Tof�an�implemen��9tation,�demonstrating�that�it�w�orks�as�claimed.��"FA���?���N�cmbx12�1��S@In��tro�`duction��e��?�Circumscription�D�w���as�in�tro�Gduced�b�y�John�McCarth�y�[10]�as�a�means�of�formalizing����?\common-sense�Grreasoning"�for�articial�in���telligence.�It�serv�ed�as�the�foundation����?of���his�theory�of�non-monotonic�reasoning.�The�essen���tial�idea�is�to�in�tro�Gduce,����?when�9Laxiomatizing�a�situation,�a�predicate��
�b>

cmmi10�ab��for�\abnormalit���y",�and�to�axiom-����?atize��the��ab��predicate�b���y�sa�ying�it�is�the�least�predicate�suc�h�that�the�other����?axioms��+are�v��q�alid.�Some�other�predicates�ma���y�b�Ge�allo�w�ed�to�\v��q�ary"�in�the�mini-����?mization��^as�w���ell.�There�are�sev�eral�tec�hnical�diculties�with�McCarth�y's�idea:����?First,��Ithe�circumscription�principle�is�most�naturally�expressed�in�second-order����?logic,�	where�w���e�ha�v�e�v��q�ariables�o�v�er�predicates�of�ob��8jects.�Second,�unless�the�rest����?of�[�the�axioms�con���tain��ab��only�p�Gositiv�ely��*�,�the�circumscription�principle�is�not�an����?ordinary��?inductiv���e�denition,�and�there�ma�y�not�ev�en�b�Ge�a�(unique)�least�solu-����?tion��for�the��ab��predicate,�so�the�circumscription�principle�can�b�Ge�inconsisten���t.����?McCarth���y's���ultimate�goal�w�as�implemen�tation�of�soft�w�are�using�the�circumscrip-����?tion��_principle�to�construct�articial�in���telligence.�Believing�that�implemen�tation����?of�O%second-order�logic�w���as�not�a�practical�approac�h,�man�y�researc�hers�ha�v�e�tried����?v��q�arious��^metho�Gds�of�reducing�sp�ecial�cases�of�the�circumscription�principle�to����?rst-order�,�logic;�see�[6]�for�a�summary�of�these�eorts.�Some�of�these�reductions����?w���ere�UUin�turn�implemen�ted.��6��NIn��this�pap�Ger�w���e�tak�e�the�other�path,�and�exhibit�a�direct�implemen�tation�of����?second-order�Vulogic�whic���h�is�capable�of�handling�some�circumscription�problems.����?The��;k���ey�to�making�this�w�ork�is�a�new�notion�of�second-order�unication.�This����?notion��of�unication�w���as�in�tro�Gduced�in�[2],�where�some�theorems�ab�out�it�are����?pro���v�ed.��UIn�that�pap�Ger,�I��7p�oin���ted�out�the�p�ossibilit���y�of�con�v�erting�y�our�fa�v�orite����?rst-order��	theorem�pro���v�er��	to�a�second-order�theorem�pro���v�er��	b�y�adding�second-����?order��3unication.�This�pap�Ger�sho���ws�explicitly�ho�w�this�can�b�Ge�done,�and�that�the�����*�o#f����ܚ����?�I�A�I����5#f����ܚ��?�resulting�exsecond-order�pro���v�er�excan�indeed�nd�circumscription�pro�Gofs.�Note�that����?it���w���ould�already�b�Ge�in�teresting�if�the�resulting�pro�Gof-c�hec�k�er�could�accept�and����?v���erify��<circumscription�pro�Gofs,�but�the�essen�tial�p�Goin�t�of�this�pap�Ger�is�that�the�use����?of��6the�new�unication�algorithm�of�[2]�enables�a�simple�theorem-pro���v�er��6to�nd����?circumscription���pro�Gofs�b���y�itself.�The�hard�part�of�this,�of�course,�is�nding�the����?correct��<v��q�alues�of�the�second-order�predicates�in���v�olv�ed.��<These�are�generally�giv���e����?b���y�3����terms�in�v�olving�an�op�Gerator�for�denition�b�y�cases.�It�is�therefore�essen�tial����?to�8�use�a�formalization�of�second-order�logic�whic���h�has�terms�for�denition�b�y����?cases.�����NAlthough�,w���e�ha�v�e�used�a�sp�Gecic�rst-order�pro�v�er�in�this�exercise,�w�e�mak�e����?no���claims�ab�Gout�the�v��q�alue�either�of�this�particular�pro���v�er���or�of�the�bac���kw�ards-����?Gen���tzen�@approac�h�that�it�uses.�W��*�e�b�Geliev�e�the�same�results�could�b�Ge�attained����?with�[
an���y�medium-to-go�Go�d�[
qualit�y�rst-order�theorem�pro�v�er,�suitably�extended����?to�!�second-order�b���y�implemen�ting�the�new�unication�algorithm.�W��*�e�just�used����?the�UUtheorem-pro���v�er�w�e�ha�v�e,�in�order�to�demonstrate�that�this�approac�h�w�orks.�����NOn�Otthe�other�hand,�it�ma���y�not�b�Ge�completely�trivial�to�add�the�new�unication����?algorithm��to�an�existing�resolution-st���yle�pro�v�er�suc�h�as�Otter.�Note�that�in�suc�h����?pro���v�ers,�h\v��q�ariable"�means�what�w���e�here�call�\meta�v��q�ariable",�and�our�\ob��8ject����?v��q�ariables"��Xare�just�constan���ts.�In�suc�h�pro�v�ers,�there�is�no�notion�of�a�\restriction"����?that�2prev���en�ts�unication�from�making�the�v��q�alue�of�a�v�ariable�dep�Gend�on�certain����?constan���ts.���In�a�Gen�tzen-st�yle�pro�v�er,�when�w�e�pro�v�e��
!",�

cmsy10�9�x�8�y�[�P�c��(�x;���y��),���w�e�try��P�c��(�X�:�;���y�[ٲ)����?after�COrestricting��X�1�to�not�dep�Gend�on��y�[ٲ.�In�resolution�pro���v�ers,�COthis�is�handled�b���y����?Sk���olemization,�1Jso�that�when�w�e�pro�v�e��9�x�8�y�[�P�c��(�x;���y��),�1Jw�e�replace��y��#�b�y��g�[ٲ(�x�)�and����?try�g�to�refute��:�P�c��(�x;���g�[ٲ(�x�)).�Then�the�o�Gccurs�c���hec�k�g�prev�en�ts�the�ev�en�tual�v��q�alue�of����?�x����from�dep�Gending�on��y�[ٲ.�The�expression��g��(�x�)�functions�essen���tially�as�a�v��q�ariable����?�y����whose���presence�causes��x��to�not�dep�Gend�on��y�[ٲ.�Our�new�unication�algorithm,����?ho���w�ev�er,�T�calls�for�in���tro�Gducing�new�v��q�ariables�dynamically�and�restricting�their����?ev���en�tual�Mv��q�alues.�Th���us�it�is�not�just�a�matter�of�adding�a�few�lines�to�Otter's����?unication�(�co�Gde�to�mak���e�this�w�ork�with�Otter.�Ho�w�ev�er,�it�is�certainly�p�Gossible.�� :F���?�2��S@Denitions��:F���?��"V

cmbx10�2.1��Y1�Syn��9tax��Tof�second-order�logic��:F��?�Second-order��Vlogic�refers�to�a�system�in�whic���h�w�e�ha�v�e�t�w�o�kinds�of�v��q�ariables,����?ob��8ject�hv��q�ariables�and�predicate�v�ariables.�W��*�e�write�lo���w�er-case�hletters��x;���y�[�;��:�:�:���R�for����?ob��8ject���v��q�ariables�and�upp�Ger-case�letters��X�:�;���Y��9;��:�:�:����for���predicate�v�ariables.�Similarly��*�,����?w���e��pha�v�e�ob��8ject�terms�and�predicate�terms.�The�term�formation�rules�are�as����?follo���ws:��"���?���<x

cmtt10�Variable:=�?�ObjectVariable�|�PredicateVariable�����?BinaryConnective�?�:-��_��|��^��|��!����?�ListTerm�?�:=�[ObjectTerm]�|�[ObjectTerm�|�ListTerm]����?VarList�?�:=�[ObjectVariable]�|�[ObjectVariable�|�VarList]����?ObjectTerm�?�:=�FirstOrderTerm�|�ApplicationTerm����?ApplicationTerm�?�:=�j��ap�(PredicateTerm,�ListTerm)�����
��o#f����ܚ������־I�A�I�I����5#f����ܚ��?�PredicateTerm�?�:=�LambdaTerm�|�PredicateVariable�|�PredicateConstant����?LambdaTerm�?�:=�(���VarList.�Formula)����?Formula�?�:=�AtomicFormula�|�(Formula�BinaryConnective�Formula)�|����?(�:�?��Formula)�|�CaseTerm�|��8��Variable�.�Formula����?CaseTerm:=�?��d�(ObjectTerm�,�ObjectTerm�,�PredicateTerm,�PredicateTerm)��C���N�Note�R�that�function�sym���b�Gols�taking�predicate�argumen�ts�are�not�legal.�There����?can��Qb�Ge�function�sym���b�ols,�as�usual�in�rst-order�logic,�but�they�tak���e�only�rst-����?order�UUargumen���ts.����NAs�T�usual,�[�x����ٓ�Rcmr7�1��|s�;����:�:�:����;���x����	0e�rcmmi7�n��q~�]�abbreviates�[�x����1���j�[�x����2���;����:�:�:����;���x����n��q~�]].�Eac���h��ListTerm��has�a����?unique�"length��n�.�An��AtomicFormula��is�an�expression�of�the�form��P�c��(�z�p��),�where��P����?�is���a�predicate�term�of�arit���y��n��and��z�(m�is�a�list�term�of�length��n�.�In�case��P�e�is�a�predi-����?cate��6constan���t,�w�e�consider��P�c��([�x����1��|s�;����:�:�:����;���x����n��q~�)])�to�b�Ge�the�same�as�the�usual�rst-order����?term��ߵP�c��(�x����1��|s�;����:�:�:����;���x����n��q~�).�F��*�or�the�preceding�to�mak���e�sense,�w�e�m�ust�dene�the�arit�y����?of�eac���h�predicate�term.�This�is�simply�the�n�um�b�Ger�of�its�free�ob��8ject�v��q�ariables.�In����?the�Nusual�w���a�y��*�,�Nw�e�can�asso�Gciate�to�eac�h�predicate�or�ob��8ject�term�a�list�of�its�free����?v��q�ariables�)�(in�a�sp�Gecied�order).�The�syn���tax�rule�for��ApplicationTerm��is�then����?sub��8jected��to�the�restriction�that���ap��έ�(�P�G;���z�p��)�can�b�Ge�formed�only�when�the�length����?of�Y�the�list��z��I�is�equal�to�the�arit���y�of��P�c��.�v��q�ariables�are�b�Gound�b�y�the����op�Gerator�in����?the�֋usual�w���a�y��*�.�֋Th�us�for�example��x:P�c��(�x;���y�[ٲ)�is�a�predicate�term�of�arit�y�1.�W��*�e����?can�UUtherefore�form�the�ob��8ject�term���ap��O��(�x:P�c��(�x;���y�[ٲ)�;�c�).����NW��*�e��`include�t���w�o��`predicate�constan���ts�of�arit�y�0,�namely���true��&�and���false���\�.�The����?ab�Go���v�e�vprules�then�mak���e�these�t�w�o�in�to�atomic�form�ulae�as�w�ell.�The�use�of���true�����?�and��݉�false����is�݉primarily�a�matter�of�notational�con���v�enience;��݉�false���in�݉the�succeden���t����?is��utraditionally�written�as�an�empt���y�succeden�t.�W��*�e�could�regard��:�A��as�an�abbre-����?viation��
for��A����!���false��̛�,��
but�it�is�con���v�enien�t��
to�retain�b�Goth�notations.�Ho���w�ev�er,����?w���e�UUallo�w��:�A��and��A���!���false��7i�to�unify��*�.�����N�ap��Y���(�X�:�;���z�p��)��can�b�Ge�abbreviated�in�writing�as��X���(�z��),�although�in�implemen���tations����?it�sis�main���tained,�and�is�prin�ted�in�the�output�of�our�pro�v�er.�Note,�though,�that����?predicate�ֳsym���b�Gols��P�:B�(constan�ts)�of�the�language�are�used�in�the�usual�rst-order����?syn���tax�UU�P�c��(�x�),�rather�than���ap��O��(�P�G;���x�).����NSecond-order��Dunication,�and�its�application�to�circumscription,�b�Goth�dep�end����?on�UUthe�use�of�conditional�terms,�or�case-terms.�These�are�terms�of�the�form���Ǎ�����_ğ��\���u

cmex10�(����������%�P�c��(�x�)������if������ӵx���=��y���������Q�(�x�)�����~�o���w��������ȍ�?There��$are�sev���eral�dieren�t�notations�for�suc�h�terms�in�use,�including�the�form����?used�UUin�the�C�and�Ja���v��q�a�programming�languages:��)�����V�x���=��y��.�?�UU�P�c��(�x�)�m:��Q�(�x�)����?and�UUthe�form�used�in�[2]�and�in�the�theories�of�F��*�eferman�[1]:����•��d�(�x;���y�[�;�P�c��(�x�)�;�Q�(�x�))�:����?�The�$1form�with��d��is�the�one�that�has�b�Geen�giv���en�in�the�ocial�syn�tax�ab�Go�v�e,�but����?the�g�other�t���w�o�g�forms�are�b�Goth�more�readable.�The�syn���tax�used�b�y�our�computer�����8�o#f����ܚ����?�IV����5#f����ܚ��?�implemen���tation��vallo�ws�a�more�general�kind�of�case�term�in�whic�h�there�can�b�Ge����?sev���eral�U�cases,�instead�of�just�one,�b�Gefore�the�\otherwise"�term.�F��*�or�represen�ting����?suc���h�~/terms�the�notation�with�a�brace�is�more�readable,�so�the�output�of�the����?pro���v�er�k�is�presen���ted�in�that�notation.�F��*�or�writing�pap�Gers,�the�notation�with�a����?question�/mark�is�more�compact�and�equally�readable,�so�w���e�will�use�it�in�the����?pap�Ger.�����NW��*�e�"�note�that�ev���erything�in�this�pap�Ger�applies�equally�w�ell�to�higher-order����?logic.���This�system�w���ould�allo�w�terms�of�ev�ery�nite�t�yp�Ge.�W��*�e�can�then�use�the����?device�B�of�\currying"�to�eliminate�the�need�for�list�terms:�instead�of���ap��=��(�P�G;����[�x;�y�[ٲ])����?w���e�vOw�ould�use���ap��p�(��ap�����(�P�G;���x�)�;�y�[ٲ).�W��*�e�then�ha���v�e�to�dene�the�rules�for�assigning����?t���yp�Ges�}+to�terms,�and�extend�the�denition�of�case�terms.�In�case�terms,�the�rst����?t���w�o���argumen�ts�are�still�restricted�to�t�yp�Ge�0.�This�is�one�of�the�systems�that����?has�}+b�Geen�used�in�[2].�(The�other�is�a�more�general�un���t�yp�ed�}+system�called���-D.)����?Second-order���logic�is�essen���tially�the�t�yp�Ge-2�subsystem�of�higher-order�logic.�Since����?second-order�UUlogic�suces�for�circumscription,�w���e�w�ork�in�that�system.��wЍ��?�2.2��Y1�Axioms��Tand�rules�of�second-order�logic��wЍ�?�W��*�e��xuse�a�Gen���tzen-sequen�t��xform�ulation�of�second-order�logic.�W��*�e�simply�tak�e�the����?usual�RGen���tzen�rules�(e.g.�G3�as�in�[8])�for�b�Goth�predicate�and�ob��8ject�quan�tiers.����?The��AG3�rules�need�to�b�Ge�supplemen���ted�with�rules�corresp�onding�to�the�formation����?of��8��-terms�and���ap���ݲ-terms,�as�w���ell�as�with�rules�corresp�Gonding�to�the�in�tro�Gduction����?of�V�case�terms�in�b�Goth�an���teceden�t�V�and�succeden���t.�W��*�e�do�not�rep�eat�the�G3�rules����?here,�UUbut�here�are�the�other�rules:��{�����\��N�t���=��s;���A��)��
�C�(�;��)c�t���6�=��s;���B��q;���c��)��c��C�������N���g����቎��\p���v��d�(�t;���s;�A;�B��q�))�;���c��)��c��C�����Û�������^׏t���=��s��)��
�A���׍��N���g���H���\p���N��c��)��c��d�(�t;���s;�A;�B��q�))������������]��^�s�t���6�=��s��)��
�B�������N���g���H���\p���N��c��)��c��d�(�t;���s;�A;�B��q�))�����X񍍟��\���Pꫵ��c��)��c��A�[�t=x�]���B׍��N���g���2�̎��\p����N���c��)��c��(�x:A�)�t�������*�����\���P��G;���A�[�t=x�]��)��
����B׍��N���g���<t���\p����N�G;����(�x:A�)�t��)��
�������	M��N�These�aqlast�rules�allo���w�us�to�rewrite�terms�b�y�b�Geta-reduction�when�searc�hing�in����?\bac���kw�ards-Gen�tzen"�FCst�yle�for�a�pro�Gof,�b�oth�in�the�\assumptions"�(an���teceden�t)����?and�ȭthe�\goal"�(consequen���t).�With�regard�to�equalit�y�reasoning,�w�e�can�either����?include��Gen���tzen-st�yle�equalit�y�rules,�or�w�e�can�simply�regard�the�equalit�y�ax-����?ioms�k�as�part�of�the�axioms����3�of�the�theory�in�question.�In�our�implemen���tation,����?there�*�are�certain�metho�Gds�that�nd�pro�ofs�in���v�olving�*�equalit�y�whic�h�could�b�Ge�in-����?terpreted��in�either�system,�but�actually�do�not�follo���w�either�one�v�ery�closely��*�.�As����?is��w���ell-kno�wn,�equalit�y�reasoning�oers�diculties�for�automated�deduction,�but����?these��7diculties�are�not�directly�relev��q�an���t�to�the�topics�discussed�in�this�pap�Ger,�����,֠o#f����ܚ�����0�V����5#f����ܚ��?�except�Pof�course�that�the�rst-order�asp�Gects�of�the�pro���v�er�Pm�ust�b�Ge�go�o�d�enough����?to�UUdeal�with�the�equalit���y�reasoning�required�in�the�circumscription�examples.��$ ���?�2.3��Y1�Circumscription�� ��?�If��͵U���and��V�ӱ�are�predicate�expressions�of�the�same�arit���y��*�,�then��U�Q���:�V��stands�for����?�8�x�(�U��(�x�)��˸!��V�8�(�x�)).�ƎIf��U���=��˵U����1��|s�;����:�:�:����;���U����n��	8�and��V����=��V����1��|s�;����:�:�:����;���V����n��	8�are�similar�tuples�of����?predicate��Lexpressions,�i.e.��U����i����and��V����i���are�of�the�same�arit���y��*�,�1�����i����n�,��Lthen��U��3����V����?�is���an�abbreviation�for��^���^���n��;Z�i�=0���
tO�U����i���X��U�V����i��TL�.�W��*�e�write��U�l'�=��V��e�for��U�l'���V����^�q��V�����U��,�and����?�U��3<��V��9�for�UU�U����V�qĸ^�8�:�V������U��.�� ����?�Denition��T1���sz(Second-Order��Circumscription).���':

cmti10�L��}'et����P�ZN�b�e�a�tuple�of�distinct����?pr��}'e�dic�ate��ec�onstants,��S���b�e�a�tuple�of�distinct�function�and/or�pr�e�dic�ate�c�onstants����?disjoint�ѕfr��}'om��P�c��,�and�let��T��(�P��;����S����)��b��}'e�a�sentenc�e.�The�se�c�ond-or�der�cir�cumscription����?of���P��v�in��T�c��(�P��;����S����)����with�variable��S��,�written��C��ir�Gc�(�T�c��;����P��;��S����)�,���is�given�in�[6]�as���@�����T�c��(�P��;����S����)�8�^�8�	���:�[�T�c��(�;�	��)�8�^��	��<��P�c��]����?�wher��}'e�Ua���and��	�n��ar�e�tuples�of�variables�similar�to��P����and��S����,�r�esp�e�ctively.�This�c�an����?e��}'quivalently���b�e�state�d�in�the�form�������T�c��(�P��;����S����)�8�^�8�	���[�T�c��(�;�	��)�8�^��	�ಸ���P�*��!��P����	���]�;����?�which���is�the�form�our�pr��}'over�uses.��& ���?�3��S@Unication�� ��?�In�UUthis�section,�w���e�recall�the�notion�of�unication�in�tro�Gduced�in�[2].��$ ���?�3.1��Y1�Meta��9v��\rariables�� ��?�A����metavariable���is�a�v��q�ariable�(not�part�of�the�formal�language)�ranging�o���v�er���terms����?of���the�formal�language.�Meta���v��q�ariables�are�used�in�a�theorem�pro�v�er�to�stand�tem-����?p�Gorarily��for�terms�whose�v��q�alues�will�ev���en�tually��b�e�determined.�Unication�is�the����?means��b���y�whic�h�the�v��q�alues�are�determined.�F��*�or�example,�when�the�pro�v�er�tries�to����?pro���v�e���9�Y�8�A�(�Y��),�a�new�meta���v��q�ariable���Z��
H�is�in�tro�Gduced�and�the�goal�b�ecomes��A�(��Z���).����?W��*�e�:�use�b�Goldface�letters�for�meta���v��q�ariables,�since�the�distinction�b�et���w�een�:�lo�w�er����?and�%mupp�Ger�case�is�already�used�for�something�else.�In�comparing�[2]�with�this����?pap�Ger,�K�it�should�b�e�understo�o�d�that�the�\v��q�ariables"�of�[2]�are�the�meta���v�ariables����?of��Uthis�pap�Ger,�and�the�\constan���ts"�of�[2]�are�the�ob��8ject�and�predicate�v��q�ariables����?(and�UUthe�constan���ts)�of�second-order�logic.�����;�o#f����ܚ����?�VI����5#f����ܚ���?�3.2��Y1�Restrictions��Tand�En��9vironmen�ts��߷��?�A��\�r��}'estriction��|�is�a�pair�consisting�of�a�meta���v��q�ariable�and�a�(p�Gossibly�empt�y)�list����?of�!�(ob��8ject�or�predicate)�v��q�ariables.�(In���tuitiv�ely��*�,�!�the�ev���en�tual�!�v�alue�of�the�v�ariable����?is��not�allo���w�ed��to�dep�Gend�on�the�mem���b�ers�of�the�list.)�An��envir��}'onment��is�a����?nite�^�list�of�restrictions.�(In���tuitiv�ely��*�,�^�an�en���vironmen�t�^�lists�all�the�v��q�ariables�in����?use��so�far,�whether�or�not�their�ev���en�tual��v��q�alues�are�restricted,�together�with�an���y����?restrictions�r�so�far�imp�Gosed.)�If��h��Z���;���r��i��is�a�mem���b�er�of�the�en���vironmen�t�r��E��w�e�sa�y����?that�3�the�v��q�ariable���Z��
nT�o��}'c�curs�t�in�3��E��*�or��is�mentione��}'d�in��E����,�and�that�all�the�mem���b�Gers����?of��>the�list��r�F[�are��forbidden�08to���Z��g��in��E����.��>W��*�e�sa���y�a�comp�Gound�term��t��is��forbidden����?to�����Z����in����E��f�if�D�it�con���tains�a�free�o�Gccurrence�of�an�y�constan�t�that�is�forbidden�to���Z�����?�in�صE����.�A���substitution��is�a�function�from�meta���v��q�ariables�to�terms.�The�substitution����?�����is�Z��le��}'gal���for�envir�onment�E�Z��pro���vided�Z޵�[ٲ(��Z���)�is�dened�for�all���Z���ֲthat�o�Gccur�in����?�E�I޲and��Qthat���[ٲ(��Z���)�do�Ges�not�con���tain�free�o�ccurrences�of�an���y�v��q�ariable�or�constan�t����?forbidden�jto���Z��
�in��E����.�The�substitution���cC�unies��terms��t��and��s��r��}'elative�L9to��E����if�jfor����?some�3_substitution����whose�restriction�to��E���is�the�iden���tit�y��*�,�3_w�e�ha�v�e��t�[����=��s��.����^��1����߷���?�3.3��Y1�Denition��Tof�unication��߷��?�The�B�inputs�to�the�unication�algorithm�are�t���w�o�B�terms��t��and��s��to�b�Ge�unied�and����?an�ˢen���vironmen�t��E����.�W��*�e�sa�y�that��t��and��s��are�to�b�Ge�unied�\relativ�e�to"�the�en-����?vironmen���t�?��E����.�One�output�of�the�unication�algorithm�is�a�substitution����o�whic�h����?is�R�legal�for��E����,�suc���h�that��t�"�=���s�[ٲ.�The�usual�notion�of�unication�is�obtained�b�y����?taking�q�an�en���vironmen�t�q��E�H�with�no�restrictions�on�an���y�of�the�v��q�ariables�o�Gccurring����?in�ޗ�E����.�But�note�that�the�use�of�restrictions,�ev���en�in�rst-order�unication,�cor-����?resp�Gonds�2
to�the�actual�use�of�unication�in�theorem-pro���ving,�where�for�example����?when�?�w���e�try�to�pro�v�e����c��)��c�9�Y�8�A�,�w�e�in�tro�Gduce�a�new�meta�v��q�ariable���Z����with�the����?restrictions�UUthat�its�ultimate�v��q�alue�cannot�dep�Gend�on�v�ariables�b�Gound�in��A;����c��.��_�NThe�g�unication�algorithm�has�a�second�output,�whic���h�is�a�new�en�vironmen�t����?(p�Gossibly)���enlarging�the�input�en���vironmen�t����E����.�Here�\enlarging"�means�simply����?that�UUnew�v��q�ariables�ma���y�ha�v�e�b�Geen�added.����NThe�UUk���ey�new�clauses�in�the�denition�of�unication�are�these:����NT��*�o�+unify���Z��&E�(�t�)�and��S����,�where��t��and��S����are�not�forbidden�to���Z���,�w���e�tak�e���Z��=��=����?�x�d�(�x;���t;�S�T;���Z���µx�).���Here���Z��T�is�a�new�meta���v��q�ariable.�The�output�en�vironmen�t�includes����?�Y�8�.�UUThe�v��q�ariables�forbidden�to���Z��
�IJare�the�ones�forbidden�to���Z��
\o�.����NT��*�o� unify��X���(�t����1��|s�;���t����2���)� and��S����,�where��t����1��|s�;�t����2�����and��S����are�not�forbidden�to��X���,�w���e�tak�e���㍒��.�X����=���x����1��|s�x����2���(�d�(�x����1���;���t����1���;��d�(�x����2���;�t����2���;�s;���Z����Ÿ��2��
.5�x�)�;���Z����Ÿ��1���x�)����?where���T�Z���	�n���1��X5�and���T�Z���	�n���2���are��Tnew�meta���v��q�ariables,�and�similarly�for�unifying��X���(�t����1��|s�;����:�:�:����;���t����n��q~�)����?and�UU�S����.����NIf����S��K�is�a�term�con���taining�a�v��q�ariable��z�nU�forbidden�to��X���,�to�unify��X�z�nU�and��S����,����?w���e�˿tak�e���Z��_L�=��s�z�p�:�(�S�^�_�����Z��	��z��).�The�v��q�ariable��z�<V�will�b�Ge�forbidden�to���Z�����in�the�output����?en���vironmen�t,�UUalong�with�an���y�other�v��q�ariables�forbidden�to��X���.��?�	��ff8�ϟ
L͍��������-=��Aa�cmr6�1�����
�That�Tis,��5��"		cmmi9�t��p�=����s�g��for�some�v��|ralues�of�the�the�v�ariables�not�in��E��2�.�����G0�o#f����ܚ�����@��VI�A�I����5#f����ܚ��N�The���other�clauses�in�the�denition�of�unication�are�of�t���w�o���kinds:�First,�there����?are�i#clauses�similar�to�those�for�Robinson's�original�unication,�but�it�should�b�Ge����?noted��[that�in�the�main�recursiv���e�clause,�where�unication�is�applied�successiv�ely����?to�@7the�argumen���ts�of�a�term,�the�output�en�vironmen�t�from�eac�h�recursiv�e�call�is�the����?input�~�en���vironmen�t�when�unifying�the�next�argumen�t.�Second,�there�are�clauses����?designed��to�ensure�that�terms�whic���h�can�b�Ge�reduced�(either�b�y�����-reduction�or����?�d�-reduction)�6�are�reduced�b�Gefore�the�ab�o���v�e�6�rules�are�applied.�F��*�or�the�details,�see����?[2].��-The�rules�giv���en�here�should�b�Ge�sucien�t�to�understand�the�applications�to����?circumscription.��+؍�NF��*�or�UUexample,�if�w���e�w�an�t�to�unify���Z��
\o�(�c�)�with���false��pQ�,�w�e�will�get��W�����"8�Z����j�=���x:�(�x��=��c�UU�?���false�����:��m�Z��
#��(�x�))����?whic���h�I�in�tuitiv�ely�sa�ys�that���Z��
PͲ(�c�)�should�b�Ge���false��d��,�but�on�all�v��q�alues�of��x��dieren�t����?from�o��c�,��X���(�x�)�is�undetermined.�The�use�of�a�new�meta���v��q�ariable�expresses�\unde-����?termined".����?�R��}'emark���.���In�[2],�it�is�pro���v�ed���that�with�resp�Gect�to�this�notion�of�unication,�unique����?most-general-uniers��exist,�just�as�they�do�for�Robinson�unication�in�rst-order����?logic.�z�W��*�e�w���an�t�z�to�tak���e�this�opp�Gortunit�y�to�explain�ho�w�this�result�reconciles����?with�v�the�w���ell-kno�wn�v�fact�that�there�is�no�unique�unier�for�Huet's�notion�of����?��-unication.�`MHere�is�the�statemen���t�of�the�most-general�unier�theorem�from����?[2]:���:����?�Theorem��T1���z�D(Most���general�unier).��L��}'et�pT�E���b�e�an�envir�onment.�Supp�ose�that����?�p����and��q����ar��}'e�normal�terms�in��D�G�.�Supp�ose�that�for�some�substitution������le�gal�for����?�E����,���p��and��q�[���ar��}'e�identic�al.�Then��p��and��q�1��unify,�and�the�answer�substitution�is����?le��}'gal���for��E����,�and�mor�e�gener�al�than���G�.���b��N�The���p�Goin���t�is,�that�the�conclusion�w�ould�not�b�Ge�v��q�alid�if�w�e�replaced�the�h�y-����?p�Gothesis�4\�p�VQ�and��q�[���are�iden���tical"�b�y�the�h�yp�Gothesis�\�p�VQ�and��q�[���ha���v�e�4a�common����?reduct."���Huet's�terms�with�man���y�uniers�do�not�form�a�coun�terexample�to�the����?theorem.�� �:���?�4��S@Blo�`c��ks��W���orld�Example���:��?�W��*�e�UUtreat�the�rst�example�from�[6]�as�a�t���ypical�circumscription�problem.����NLet�UU��c��(�Ab;���O�Gn�)�b�e�the�theory��W����Cm�c���6�=��b�8�^�:�O�Gn�(�c�)��^�8�x�(�:�ab�(�x�)���!��O�n�(�x�))����?where��the�v��q�ariables�range�o���v�er��\blo�Gc�ks"�and��O�Gn�(�x�)�means�\�x��is�on�the�table".����?Circumscription��wenables�us�to�conclude�that��a��is�the�only�blo�Gc���k�not�on�the�table.����?F��*�or�M�simplicit���y�,�w���e�rst�consider�the�problem�without�the�predicate��B��q�,�i.e.�w�e����?assume��?all�v��q�ariables�range�only�o���v�er��?blo�Gc�ks.�The�idea�is�that�normal�blo�Gc�ks�are����?on�
�the�table,�and�since��c��is�the�only�abnormal�blo�Gc���k,��b��is�a�normal�blo�c���k�and����?hence�UUis�on�the�table.�Circumscription�should�enable�us�to�pro���v�e�UU�O�Gn�(�b�).�����X��o#f����ܚ����?�VI�A�I�I����5#f����ܚ��N�Circumscription���in�this�example�is�tak���en�to�minimize��ab��with�v��q�ariable��O�Gn�,����?so�UUin�the�general�sc���hema�ab�Go�v�e,�w�e�tak�e��P���to�b�Ge��ab��and��S���to�b�e��O�n�.���ۍ���|9�c���6�=��b������X��(1)��������|9�8�x�(�:�ab�(�x�)���!��O�Gn�(�x�))������X�(2)��������|9�:�O�Gn�(�c�)������X�(3)��������|9�8�	���[�8�x�(�:�	��(�x�)���!���(�x�))�8�^�:��(�c�)��^��	�ಸ���ab��!��ab����	���]������X�(4)������NW��*�e�}�rst�presen���t�a�h�uman-pro�Gduced�pro�of,�for�later�comparison�to�the�pro�of����?found��%b���y�our�program.�W��*�e�tak�e�as�the�goal�to�pro�v�e��O�Gn�(�b�).�Bac�k�c�haining�from����?(2)�UUpro�Gduces�the�new�goal��:�ab�(�b�).�The�h���uman�then�suggests�the�v��q�alues�������el�	�ಲ=���x:�(�x��=��c�UU�?���true�����:����false����)������X�(5)�������������=��x:�(�x��=��c�UU�?���false��7i�:����true���K�)������X�(6)������?With��Cthese�v��q�alues�of����and��	���,�w���e�w�an�t�to�pro�v�e��ab�(�b�)��O�!���false���K�,��Cso�w�e�need�to����?v���erify��B� �[ٲ(�b�)���=���false����.�But�� �[ٲ(�b�)�=�(�b��=��c��?���true��*m�:���false����),�and��b��=��c��ev��q�aluates�to�����?�false��Y5b�since�f�b���6�=��c��is�in�the�an���teceden�t,�so�� �[ٲ(�b�)�ev��q�aluates�to���false��5b�.�It�therefore����?suces�UUto�v���erify�the�h�yp�Gothesis�of�(4),�namely�����3��8�x�(�:�	���(�x�)���!���(�x�))�8�^�:��(�c�)��^��	�ಸ���ab:����?�Fix��can��x�,�and�supp�Gose��:�	���(�x�).�Then��x���6�=��c�,��cfrom�whic���h���(�x�)�follo�ws,�whic�h�pro�v�es����?the��Qrst�conjunct.�The�second�conjunct,��:��(�c�),�follo���ws�immediately�b�y�reduction����?to���f�true�����.��fThe�third�conjunct,��	�f#��L��ab�,�is�pro���v�ed��fas�follo���ws:�supp�Gose��	���(�x�).�Then����?�x���=��c�L�and�so�w���e�m�ust�pro�v�e��ab�(�c�).�But�b�y�(3)�w�e�ha�v�e��:�O�Gn�(�c�),�and�so�b�y�(2)�w�e����?ha���v�e�UU�ab�(�c�).�That�completes�the�pro�Gof.����NW��*�e���no���w�explain�ho�w�the�pro�v�er�attac�ks�this�problem.�W��*�e�w�an�t�to�pro�v�e����?�:�ab�(�b�).���(Ocially�that�goal�is�the�succeden���t�of�a�sequen�t�whose�an�teceden�t�is�the����?list�o�of�axioms.)�So�the�pro���v�er�o�assumes��ab�(�b�),�and�the�new�goal�is��ab�(�b�)��)���
�false�� ��.�(Of����?course�ocially�the�axioms�should�app�Gear�in�the�an���teceden�t�of�the�goal�sequen���t,����?to�Go,�&�but�w���e�do�not�write�them.)�This�causes�(3)�to�b�e�\op�ened�up",�in���tro�ducing����?meta���v��q�ariables����P��and��Q�.�The�form�ula��ab�и��Q����is�really��8�w�D�(�ab�(�w��)�и!��Q�(�w�D�)),����?so���a�meta���v��q�ariable��W�
*�is�in�tro�Gduced�for��w��~�as�w�ell,�but�so�Gon�it�is�instan�tiated����?to�eŵb��to�unify��ab�(�b�)�with��ab�(�W�c��),�in�the�hop�Ges�of�pro���ving��ab�(�W��)��)��
�Q�(�W��)�from����?�ab�(�b�)��)���
�false�� ��.�UUTh���us�the�pro�v�er�tries�to�unify��Q�(�b�)�with���false��pQ�.�This�giv�es������"�Q���=��y�[�:�(�y�"�=��b�UU�?���false��7i�:���Y�8�(�y��))����?where��e�Y��I�is�a�new�v��q�ariable.�The�next�goal�is�the�conjunction�of�the�three�form���ulae����?on�}�the�left�of�the�implication�in�4.�These�are�tak���en�in�order;�the�rst�one�is����?�8�v�[ٲ(�:�Q�(�v��)���!��P�(�v�[ٲ)).�@Fixing��v�{�the�goal�is��:�Q�(�v��)���!��P�(�v��);�@writing�out�the�curren���t����?v��q�alue�UUof��Q��and�����-reducing,�the�goal�is�����M�:�(�v�"�=���b�UU�?���false��7i�:��Y�8�(�v�[ٲ))��!��P�(�v��)����?There�UUis�a�simplication�rule�for�pushing�a�negation�in���to�a�cases�term,�namely�����}��:�(�v�"�=���p�UU�?��q�xF�:��r�G�)�=�(�v�"�=��p��?��:�q�xF�:��:�r�G�)�:�����	f��o#f����ܚ������A�IX����5#f����ܚ��?�So�UUthe�goal�b�Gecomes���������(�v�"�=���b�UU�?���true�����:��:�Y�8�(�v�[ٲ))��!��P�(�v��)�:����?�This�UUis�solv���ed�b�y�second-order�unication,�taking�������P���=��v�[�:�((�v�"�=��b�?��UU�true�����:��:�Y�8�(�v��))�8�_��Z��v��)�:����?�The�UUnext�goal�is��:�P�(�c�).�That�is,�after�a�b�Geta�reduction,�����-�:�((�c���=��b�UU�?���true�����:���:�Y�8�(�c�))�8�_��Z���(�c�)))�:����?�No���w���w�e�can�apply�a�simplication�rule�using�the�axiom��c����6�=��b�,�reducing�the����?cases���term�to��:�Y�8�(�z�p��)�and�hence�the�whole�goal�to��:�(�:�Y��(�c�)�9۸_��Z���(�c�)).���Using�rewrite����?rules��6appropriate�to�classical�logic�w���e�simplify�this�to��Y�8�(�z�p��)���^�:�Z���(�c�).��6Splitting�the����?conjunction�]Fin���to�t�w�o�subgoals,�the�rst�one�to�b�Ge�pro�v�ed�is��Y�8�(�c�).�This�is�solv�ed����?b���y�UUsecond-order�unication,�taking��������ݵY����=���u:�(�u��=��c�?�UU�ab�(�b�)�:��A�(�u�))����?where�=>�A��is�a�new�meta���v��q�ariable.�Y��*�ou�migh�t�think�w�e�should�get���true��X��in�place�of����?�ab�(�b�)��uin�the�v��q�alue�of��Y�8�,�but�when�the�pro���v�er��uhas�to�pro���v�e��ua�goal�of�the�form��Y��(�c�),����?it�/do�Ges�not�try�to�unify��Y�8�(�c�)�with���true��
Q�,�but�rather�with�one�of�the�assumptions����?(form���ulas���in�the�an�teceden�t).�It�tries�the�most�recen�tly-added�ones�rst,�and�it����?nds�UU�ab�(�b�)�there,�whic���h�explains�the�v��q�alue�giv�en�for��Y�8�.��@���NThe�8�second�goal�is��:�Z���(�c�).�Then��Z��(�c�)�is�assumed,�leading�to�a�goal��Z��(�c�)��)���
�false�� ��.����?Unifying�UU�Z���(�c�)�with���false��Ŧ�giv���es��Z�q�the�v��q�alue��������Z�~4�=���r���:�(�r�5�=��c�UU�?��O�Gn�(�c�)��:��B��q�(�r��))�;����?�where�Wy�B���is�a�new�meta���v��q�ariable.�Again,�y�ou�migh�t�exp�Gect���false����to�o�ccur�in�place�of����?�O�Gn�(�c�)��<in�the�v��q�alue�of��Z���,�but�the�pro���v�er��<nds�the�v�alue�giv���en,�whic�h�is�equiv��q�alen�t����?since�UU�:�O�Gn�(�c�)�is�an�axiom.����NA���t�UUthis�p�Goin�t,�the�v��q�alues�of��P��has�b�Gecome���������P���=��v�[�:�(�v�"�=��b�UU�?���true�����:���:�(�v��=��c�?��UU�true�����:��A�(�z�p��)����?whic���h�UUsimplies�to������a�P���=��v�[�:�(�v�"�=��b�UU�?���true�����:���v��=��c�UU�?��O�Gn�(�c�)��:��:�A�(�z�p��))����?The�UUv��q�alue�of��Q��is�no���w�giv�en�b�y����y�H�Q���=��y�[�:�(�y�"�=��b�UU�?���false��7i�:��(�u:�(�u��=��c�UU�?��ab�(�b�)��:��A�(�u�)))�y��)����?whic���h�UUreduces�to������a�Q���=��y�[�:�(�y�"�=��b�UU�?���false��7i�:���y��=��c�UU�?��ab�(�b�)��:��A�(�y�[ٲ))�����
ug�o#f����ܚ����?�X����5#f����ܚ��?�The��next�goal�is��Q�����ab�,��that�is��8�z�p��(�Q�(�z��)���!��ab�(�z�p��)).��Fixing��z��,�the�goal�is��Q�(�z��)���!����?�ab�(�z�p��).��Using�the�Gen���tzen�rule�for�in�tro�Gducing��!��on�the�righ�t,�and�writing�out����?the�UUcurren���t�v��q�alue�of��Q�,�our�goal�is�the�sequen�t��z퍒�I�z�7��=���b�UU�?���false��7i�:��z�7��=��c��?��a�(�b�)�:��W�c��(�x�)��)��
�ab�(�x�)�:����?�This�UUis�pro���v�ed�UUb�y�cases,�sp�Gecically�b�y�the�cases-left�rule.����NCase�UU1,��z�7��=���b�.�The�goal�reduces�to���false��7i�!��ab�(�z�p��)�whic���h�is�immediate.����?Case�UU2,��z�7��6�=���b��and��z��=���c�.�Then��Q�(�a�)�reduces�to��z퍒���z�7��=���c�8�^��z��6�=���b��^��ab�(�b�)����?so�UUthe�goal�b�Gecomes�������z�7��=���c;���z��6�=��b;���ab�(�b�)��)��
�ab�(�z�p��)�:���(��?�The�P�h���uman�can�note�that��ab�(�c�)�follo�ws�from��8�x�(�:�ab�(�x�)���!��O�Gn�(�x�))�P�and��:�O�n�(�c�),����?and���from��ab�(�c�)�the�goal�follo���ws�quic�kly��*�.�This�is�a�relativ�ely�simple�problem�in����?rst-order��|logic�with�equalit���y��*�,�the�diculties�of�whic�h�are�irrelev��q�an�t�to�circum-����?scription�UUand�second-order�logic.�W��*�eierstrass�is�able�to�pro���v�e�UUthe�goal.����NCase�UU3,��z�7��6�=���b��and��z��6�=���c�.�Then��Q�(�z�p��)�reduces�to��W�c��(�z��),�so�the�goal�b�Gecomes��z퍒�m�W�c��(�z�p��)��)��
�ab�(�z��)�:����?�This�UUgoal�is�pro���v�ed�UUb�y�instan�tiating�the�meta�v��q�ariable��W�c��:�������W�*��=���z�p�:�(�ab�(�z��)�8�_��T�c��(�z�p��))����?where�UU�T���is�a�new�meta���v��q�ariable.�The�nal�v�alues�of��P��and��Q��are�th���us����o���P���=��v�[�:�(�v�"�=��b�UU�?���true�����:���v��=��c�UU�?��O�Gn�(�c�)��:�(�:�ab�(�v�[ٲ)�8�^�:�T�c��(�v��)))����y�2�Q���=��y�[�:�(�y�"�=��b�UU�?���false��7i�:���y��=��c�UU�?��a�(�b�)��:�(�ab�(�y�[ٲ)�8�_��T�c��(�y��)))���(��?T��*�o���ac���hiev�e�the�stated�goal��O�Gn�(�b�),�the�pro�v�er�has�only�needed�to�deduce�that��b����?�is��not�abnormal.�Unlik���e�the�h�uman,�it�has�not�gone�ahead�to�deduce�an�ything����?ab�Gout���other�ob��8jects�than��a��and��b�{�the�uninstan���tiated�meta�v��q�ariable��T�Ht�remains�as����?\undetermined".���Of�course,�the�constan���t��b��migh�t�as�w�ell�ha�v�e�b�Geen�a�v��q�ariable;����?the���pro���v�er�can�pro�v�e��8�x�(�x���6�=��a��!��O�Gn�(�x�))�just�as�w���ell�as�it�can�pro�v�e��O�Gn�(�b�).�But����?that�Q�pro�Gof,�lik���e�the�one�ab�o���v�e,�Q�will�still�use�instan���tiations�of��P��and��Q��in�v�olving����?free�UUmeta���v��q�ariables.�������?�5��S@The��automatically-pro�`duced�pro�of������?�Here���w���e�presen�t�the�pro�Gof�as�pro�duced�(and�t���yp�eset)�b���y�our�pro�v�er.�In�this�pro�Gof,����?�a�(�x�)��stands�for��ab�(�x�),�meaning�\�x��is�abnormal".��O�Gn�(�x�),�meaning�\�x��is�on�the����?table",�UUis�represen���ted�b�y��o�(�x�).���ƍ�NThe�UUgoal�is��o�(�b�)����XT��*�rying�UU�:�(��a�(�X���)��")������L�o#f����ܚ������A�XI����5#f����ܚ��b�Assuming�UU�a�(�X���)����bT��*�rying�UUfor�a�con���tradiction����lT��*�rying�UUleft-arro���w����lAssume�������8�w��D�(�a�(�w��)���!���ap��
US(�Q;���w�D�))��������lStill�UUtrying���false�����l�trying�UUsecond-order�unication�(clause�1)�on���M�����ap��У�(�Q;���W�c��)��=���false������l�The�UUmeta���v��q�ariable�gets�the�v�alue������#�Q���=��y�[�:�������\�(����������ު�false�����C-��if�����Yf��y�"�=��W�����������ap��G(�Y��9;���y��)�����@mo���w��������W���vT��*�rying�UU�a�(�W�c��)����vAha!�UUw���e�ha�v�e��a�(�W�c��)����vSuccess����lSuccess����lDisc���harging�������8�w��D�(�a�(�w��)���!���ap��
US(�Q;���w�D�))������lT��*�rying���M��h|#�8�v��[ٲ(�:�(��ap��
�;(�Q;���v��)��#�)���!���ap��
US(�P�G;���v�[ٲ))��l�$�;����:�(��ap��
�;(�P�G;�c�)��!ͣ)�;��8�z��p��(�ap��
�;(�Q;�z��)���!��a�(�z��))������vT��*�rying�������8�v��[ٲ(�:�(��ap��
�;(�Q;���v��)��#�)���!���ap��
US(�P�G;���v�[ٲ))�������T��*�rying�����F��:�(��ap��
�;(�Q;���v�[ٲ)��#�)���!���ap��
US(�P�G;���v�[ٲ)������That�UUreduces�to:��V������L����\�(��������������true��������if������Եv�"�=���b�����������:�(��ap��
�;(�Y��9;���v�[ٲ)��"XM)������϶o���w������]I�!����ap��
US(�P�G;���v�[ٲ)��W����Assuming��������k���\�(�������������true�������I�if�����5/�v�"�=���b���������#��:�(��ap��
�;(�Y��9;���v�[ٲ)��"XM)�������o���w�����������v�o#f����ܚ����?�XI�A�I����5#f����ܚ������T��*�rying��UUap��
�(�P�G;���v�[ٲ)�����trying�UUsecond-order�unication�(clause�2)�on��5�����i+ap����f(�P�G;���v�[ٲ)��=�������\�(����������$u��true�����TXL�if�����j�2�v�"�=��b����������:�(��ap��
�;(�Y��9;�v�[ٲ)��"XM)�����QJo���w�������������The�UUmeta���v��q�ariable�gets�the�v�alue���H��{Y�P�*��=���v�[�:��_��������8��	�>���>���>���>���:���������㏟��\(����������*�e�true�����Ztòif�����p���v�"�=��b���������s�:�(��ap��
�;(�Y��9;���v��)��"XM)�����Wf�o���w��������-�;������ap��8�(�Z�(�;���v��)���������$���9��	��$��>����$��>����$��>����$��>����$��;������������Aha!�UUw���e�ha�v�e��ap��
�(�P�G;���v�[ٲ)�����Success�����Disc���harging�����Success����vSuccess����vT��*�rying��P3����Ը:�(��_�(��:�(��ap��
�;(�Y��9;���c�)��!wk)��/�5�;������ap��8�(�Z�(�;���c�)��#�)��d�/)��P3���vClassically��*�,�UUit�w���ould�suce�to�pro�v�e:��5w����V�ap�����(�Y��9;���c�)�;��:�(��ap��
�;(�Z�(�;�c�)��"k)�������T��*�rying��UUap��
�(�Y��9;���c�)�����trying�UUsecond-order�unication�(clause�1)�on��5w����jap��Ԧ�(�Y��9;���c�)��=��a�(�X���)�������The�UUmeta���v��q�ariable�gets�the�v�alue���G������Y����=���u:�������\�(������������a�(�b�)�����DǺif�����[��u��=��c�����������ap��G(�A;���u�)�����A��o���w������������Aha!�UUw���e�ha�v�e��ap��
�(�Y��9;���c�)�����Success�����T��*�rying�UU�:�(��ap��
�;(�Z�(�;���c�)��"k)�����trying�UUsecond-order�unication�(clause�1)�on��ap��
�(�Z�(�;���c�)��=��o�(�c�)�����The�UUmeta���v��q�ariable�gets�the�v�alue���G���Q�Z�~4�=���r���:�������\�(������������o�(�c�)�����Dn�if�����Z�n�r�5�=��c�����������ap��G(�B��q;���r�G�)�����A`Po���w�����������
���o#f����ܚ������5�XI�A�I�I����5#f����ܚ�����Aha!�UUw���e�ha�v�e��:�(��ap��
�;(�Z�(�;���c�)��"k)�����Success����vThat�UUw���as�the�last�conjunct,�so�the�conjunction�is�pro�v�ed.����vSuccess����vT��*�rying��#Hލ�C)E�8�z�����p����^��������ap��[F(��y�[�:�������\�(������G����W�a�false������5�if������m��y�"�=���b����qЍ������ap��G(��u:�������\�(������������a�(�b�)�����DǺif�����[��u���=��c�����������ap��(�A;���u�)�����A��o���w���������c�;���y��)������&�o���w����
�������;���z�p��)���!��a�(�z��)������^�������W�����T��*�rying�����O�eap��Zf�(��y�[�:�������\�(������G����W�a�false������5�if������m��y�"�=���b����qЍ������ap��G(��u:�������\�(������������a�(�b�)�����DǺif�����[��u���=��c�����������ap��(�A;���u�)�����A��o���w���������c�;���y��)������&�o���w����
�������;���z�p��)���!��a�(�z��)��W����That�UUreduces�to:���*������Pj���\�(��������������false������uIJif�������B�z�7��=���b���������
�a�(�b�)������u�if���������z�7��=���c����������	N�ap�����(�A;���z�p��)������g�o���w������౸!���a�(�z�p��)���*����Assuming�� �񍍍���+9���\�(����������ͅ��false������P��if�����
��z�7��=���b����������a�(�b�)������P�if�����
�y�z�7��=���c������������ap���rX(�A;���z�p��)������B[o���w�����������T��*�rying�UU�a�(�z�p��)�����T��*�rying�UUleft-arro���w�����Assume��*���ƙ�8�w��D�(�a�(�w��)���!���ap��
US(�D�G;���w�D�))�������Still�UUtrying��a�(�z�p��)�����F��*�ailure�����Disc���harging�����Disc���harging�����Pro�Gceed�UUb���y�cases:�����Case�UU1:�z�7��=���b������Then�UUw���e�ha�v�e���false�������T��*�rying�UU�a�(�z�p��)�����That�UUcase�cannot�o�Gccur�����Case�UUsucceeded������Šo#f����ܚ����?�XIV����5#f����ܚ����Case�UU2:�z�7��=���c��������Then�UUw���e�ha�v�e��a�(�b�)�����T��*�rying�UU�a�(�z�p��)�����	T��*�rying�UUleft-arro���w�����	Assume��5g���ƙ�8�w��D�(�a�(�w��)���!���ap��
US(�D�G;���w�D�))���������	Still�UUtrying��a�(�z�p��)�����	F��*�ailure�����	Disc���harging�����	Disc���harging�����T��*�rying�UUpro�Gof�b���y�con�tradiction�of��a�(�z�p��)�����Assume�UU�:�(��a�(�z�p��)��'l)�����	By�UUaxiom�c2,�it�w���ould�suce�to�pro�v�e��:�(��a�(�C���)���)�����
T��*�rying�UU�:�(��a�(�C���)���)�����
Aha!�UUw���e�ha�v�e��:�(��a�(�C���)���)�����
Success�����Case�UUsucceeded�����Case�UU3:otherwise�����T��*�rying�UU�a�(�z�p��)�����trying�UUsecond-order�unication�(clause�2)�on��5g����"Uap��հ�(�A;���z�p��)��=��a�(�z��)��������The�UUmeta���v��q�ariable�gets�the�v�alue������z�A���=��z�p�:��_�(��a�(�z��)��'l�;������ap��8�(�D�G;���z��)��&x)�������Aha!�UUw���e�ha�v�e��a�(�z�p��)�����Case�UUsucceeded�����Pro�Gof�UUb���y�cases�succeeded�����Success�����Disc���harging�����Success����vSuccess����lThat�UUw���as�the�last�conjunct,�so�the�conjunction�is�pro�v�ed.����lSuccess����bSuccess����XSuccess����NSuccess.�UUThat�completes�the�pro�Gof.��������o#f����ܚ�����@_�XV����5#f����ܚ��?�References������C���1.���O��Beeson,�7OM.,��"�j��		cmti9�F��J�oundations�X�of�Constructive�Mathematics�,�7OSpringer-V��:�erlag,�Berlin/����O��Heidelb�A�erg/�TNew�Y��:�ork�(1985).������C��2.���O��Beeson,�.*M.,�Unication�in�Lam��9b�A�da�Calculus�with�if-then-else,�in:�Kirc�hner,�C.,����O��and�՛Kirc��9hner,�H.�(eds.),��A�Îutomate��d��De�duction-CADE-15.�15th�International�Con-����O��fer��enc�e�4on�A�Îutomate��d�De�duction,�Lindau,�Germany,�July�1998�Pr�o�c�e�e�dings�,�_pp.����O��96-111,�TLecture�Notes�in�Articial�In��9telligence��1421�,�Springer-V��:�erlag�(1998).������C��3.���O��Beeson,���M.,�Automatic�generation�of�epsilon-delta�pro�A�ofs�of�con��9tin�uit�y��:�,���in:�Calmet,����O��Jacques,���and�Plaza,�Jan�(eds.)��A�Îrticial��Intel�x�ligenc��e�and�Symb�olic�Computation:����O��International�XConfer��enc�e�AISC-98,�Plattsbur�gh,�New�Y��J�ork,�USA,�Septemb�er�1998����O��Pr��o�c�e�e�dings�,�Tpp.�67-83.�Springer-V��:�erlag�(1998).������C��4.���O��Beeson,��RM.,�Automatic�generation�of�a�pro�A�of�of�the�irrationalit��9y�of�e,�in�Armando,����O��A.,�aand�Jeb�A�elean,�T.�(eds.):��Pr��o�c�e�e�dings�M\of�the�Calculumus�Workshop,�1999�,�a�Ele��c-����O��tr��onic��Notes�in�The�or�etic�al�Computer�Scienc�e��/�23��3,�2000.�Elsevier.�Av��|railable�at����O��h��9ttp://www.elsevier.nl/lo�A�cate/en�tcs.�?This�pap�er�has�also�b�een�accepted�for�publi-����O��cation�	in�a�sp�A�ecial�issue�of��Journal�?Cof�Symb��olic�Computation�	�whic��9h�should�app�ear����O��in�Tthe�v��9ery�near�future.������C��5.���O��Beeson,�كM.,�Some�applications�of�Gen��9tzen's�pro�A�of�theory�to�automated�deduction,����O��in�-P��:�.�Sc��9hro�A�eder-Heister�(ed.),��Extensions�dof�L��o�gic�dPr�o�gr�amming�,�-Lecture�Notes�in����O��Computer�TScience��475��101-156,�Springer-V��:�erlag�(1991).������C��6.���O��Dohert��9y��:�,��JP�.,�Luk��|raszewicz,�W.,�And�Szalas,�A.,�Computing�circumscription�revis-����O��ited:�Ta�reduction�algorithm,��J.�N<A�Îutomate��d�R�e�asoning�T�18�,�297-334�(1997).������C��7.���O��Ginsb�A�erg,�~�M.�L.,�A�~zcircumscriptiv��9e�theorem�pro�v�er,��A�Îrticial���Intel�x�ligenc��e��39��pp.����O��209-230,�T1989.������C��8.���O���Intr��o�duction�N<to�Metamathetics�,�Tv��|ran�Nostrand,�Princeton,�N.J.�(1950).������C��9.���O��Lifsc��9hitz,�EbV.,�Computing�circumscription,�in:��Pr��o�c�e�e�dings�zdof�the�9th�International����O��Joint�N<Confer��enc�e�on�A�Îrticial�Intel�x�ligenc�e�,�Tv��9olume�1,�pages�121-127,�1985.������?10.���O��McCarth��9y��:�,�g9J.,�Circumscription,�a�form�of�non-monotonic�reasoning,��A�Îrticial��~In-����O��tel�x�ligenc��e�,�T�13��(1-2),�pp.�27-39,�1980.������?11.���O��Przym��9usinski,�[T.,�An�algorithm�to�compute�circumscription,��A�Îrticial��Intel�x�ligenc��e�,����O�¿38�,�Tpp.�49-73,�1991.������H���;�o#f���"�j��		cmti9�5��"		cmmi9��Aa�cmr6��':

cmti10���<x

cmtt10��"V

cmbx10���N�cmbx12�t�:		cmbx9�o���		cmr9���N�ffcmbx12�
!",�

cmsy10�
�b>

cmmi10�	0e�rcmmi7�K�`y

cmr10�ٓ�Rcmr7���u

cmex10���������

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