Sharedwww / projects / kleinerman_senior_thesis.dviOpen in CoCalc
����;� TeX output 2004.04.07:0912�������7
�����V������TE��"V

cmbx10�ON���THE�TORSION�POINTS�OF�ELLIPTIC�CUR����VES�&������'MODULAR���ABELIAN�V����ARIETIES������|{Ycmr8�SETH��XKLEINERMAN��0���ҥ���-�

cmcsc10�Contents����B��K�`y

cmr10�1.�
In���tro�Gduction�UUand�Bac�kground��Õ=2�����B�1.1.�
Elliptic�UUCurv���e�Denitions��ͱ�2�����B�2.�
F��*�ormal�UUGroups��X�3�����B�2.1.�
The�UUBasics��
��3�����B�2.2.�
So�UUwhere�are�the�elemen���ts?���q�6�����B�2.3.�
More�UUTheory�of�F��*�ormal�Groups�����6�����B�3.�
Application�UUto�Elliptic�Curv���es:�q�the�Finitude�of��
�b>

cmmi10�E����(�K���)����ٓ�Rcmr7�tor���\j��7�����B�3.1.�
An�UUElliptic�Curv���e's�F��*�ormal�Group���ʓ7�����B�3.2.�
Reduction�UUon�Elliptic�Curv���es���j�8�����B�4.�
Statemen���ts�UUof�Tw�o�Theorems�Ab�Gout�T��*�orsion��|q�11�����B�5.�
T��*�orsion�UUP���oin�ts�on�Elliptic�Curv�es:�q�Algorithmic�Questions��E�12�����B�5.1.�
The�UUNagell-Lutz�Theorem��ƪ�12�����B�5.2.�
Doud's�UUAlgorithm����|13�����B�6.�
Bac���kground�UUon�Mo�Gdular�Ab�elian�V��*�arieties���8�14�����B�6.1.�
Mo�Gdular�UUGroups�&�Mo�dularit���y����14�����B�6.2.�
Hec���k�e�UUOp�Gerators�and�Mo�dular�Sym���b�ols����16�����B�7.�
Computations�UUwith�Mo�Gdular�Ab�elian�V��*�arieties��t��17�����B�7.1.�
Finding�UUthe�Quotien���ts�of��J����0��|s�(�N��)����?17�����B�7.2.�
The�UUT��*�orsion�Multiple���17�����B�8.�
Examples��J18�����B�8.1.�
Bounds�UUusing�Go�Go�d�UUReduction���_�20�����B�8.2.�
Conjectures����21�����B�9.�
Ac���kno�wledgmen�ts���d22�����B�References��#��23��������k�1����*��7
��������6ऱ2����SETH���KLEINERMAN���V����������1.�����Intr��oduction��and�Ba�ck�gr�ound����Y�IJAge�UUcannot�wither�her,�nor�custom�stale����Y��Her�UUinnite�v��q�ariet���y��*�.����Y��|�%�':

cmti10�A���ntony���and�Cle��}'op�atr�a�,�UU2.2.271-2.��%׍�B�The��study�of�ab�Gelian�v��q�arieties�is�of�great�in���terest�in�mo�dern�n���um�b�er��theory��*�.�"In����6�this�]�pap�Ger�w���e�will�try�to�understand�some�of�the�basics�of�the�v��q�arieties'�torsion����6�subgroups.�dThat�܉they�are�nite�at�all�is�a�matter�of�some�concern,��Vand�that�will����6�b�Ge�E�the�main�thrust�of�the�theory�w���e�dev�elop.�l�W��*�e�will�often�sp�Gecialize�to�results�on����6�elliptic�UUcurv���es,�whic�h�are�b�Getter�understo�o�d.��0���6�1.1.��Elliptic�Curv��9e�Denitions.��An��elliptic�curv���e��E�6�o�v�er�a�eld��K�6Ųis�the�lo�Gcus����6�of�UUp�Goin���ts�(�x;���y�[ٲ)�satisfying�the��Weierstr��}'ass���e�quation��o�����Եy��[ٟ����2��,�+�8�a����1��|s�xy����+��a����3���y�"�=���x������3���S�+��a����2���x������2���S�+��a����4���x��+��a����6���;��K���6लwith���co�Gecien���ts��a����	0e�rcmmi7�i����
!",�

cmsy10�2��6�K���.�22A��notion�of�summation�of�p�oin���ts�turns�this�set�in�to�an����6�ab�Gelian��group.�W4W��*�e�w���on't�giv�e�the�group�la�w�algorithm�here�explicitly��*�,��but�refer�the����6�reader�+to�[�17��
],�3�Ch.�c�I�GI�I�+�x�2.�The�+motiv��q�ation�for�the�comp�Gosition�la���w�is�that�Bezout's����6�theorem���indicates�that�a�line�in���tersects�this�degree�three�curv�e�in�exactly�three����6�p�Goin���ts�n6(coun�ting�m�ultiplicit�y�in�the�sp�Gecial�case�of�tangency).��iTherefore�there�is�a����6�unique��third�p�Goin���t��R���on�b�oth�the�curv���e�and�the�line�connecting�an�y�giv�en�p�Goin�ts��P����6लand�Y`�Q�.�}�Also,�Zcgiv���en�a�base�p�Goin�t��O�G�,�Zcthere�is�a�unique�p�oin���t��R��ǟ�^��O!�cmsy7�0��;`�on�the�curv�e�that�is����6�on���the�line�connecting��O��˲and��R�Dz.��Dene��R����^��0���V�=�V�P��U��YƵQ�.��This�binary�op�Geration�giv���es����6वE����the�*kstructure�of�an�ab�Gelian�group.��	(All�the�group�la���ws�are�straigh�tforw�ard�to����6�observ���e��except�asso�Gciativit�y��*�,�GDwhic�h�one�can�demonstrate�in�a�tec�hnical�argumen�t����6�with�yGthe�explicit�equations�for�the�curv���e�and�lines,��Dbut�whic�h�w�e�happily�a�v�oid����6�here.)����B�This���c���haracterization�is�already�an�abuse�of�notation,�4since�tec�hnically�the�curv�e����6�ough���t���to�b�Ge�in�pro��8jectiv�e�space�for�the�discussion�ab�Go�v�e�to�mak�e�an�y�sense�at�all.����6�Ho���w�ev�er,��
the���ane�curv���e�has�a�unique,�w���ell-dened�pro��8jectiv�e�closure,��
so�rigor�is����6�not�plost.��This�aside�is�b���y�w�a�y�of�excusing�the�fact�that�sometimes�w�e'll�b�Ge�lax����6�and�8�refer�to�the�p�Goin���t�(�x;���y�[ٲ)�on�a�curv�e�instead�of�the�p�Goin�t�[�X�
�:�B�Y�z��:��Z���]�8�on�the����6�pro��8jectivized�UUcurv���e.����B�T��*�o��further�simplify�things�for�us,��if�the�eld�is�not�of�c���haracteristic�2�or�3,�w���e����6�can�UUmak���e�a�rather�messy�substitution��s�������������x�������=#�7�������v!�����<$��zL�1���K�w�fe
�	(֍36�����-�(�x�8���3�a�������2����1����S���12�a����2��|s�)��������������6�(1.1)������������������3�y�������=#�7�������v!�����<$���L�1���K�w�fe�	(֍216�����-�(�y�����8�a����1��|s�x����a����3���)��������������6�(1.2)������K���6�to��reduce�the�equation�to�the�form��y��[ٟ�^��2���d�=���x���^��3���&�+�u��Ax��+��B��q�,�Bwhere��A��and��B�t/�hide�the�ugly����6�expressions�������������?uZ�A���=�������S�q��8�27�a�������4����1����S���216�a�������2����1���|s�a����2�����432�a�������2����2����+�648�a����1��|s�a����3���+�1296�a����4��|s�;������u�����������>�"B�G��=�������Q��54�a�������6����1����S�+�8�648�a�������4����1���|s�a����2���+�2592�a�������2����1���|s�a�������2����2������1944�a�������3����1���|s�a����3���+�3456�a�������3����2������3888�a�������2����1���|s�a����4�����7776�a����1��|s�a����2���a����3�������������������S�q�+�8�11664�a�������2����3����S���15552�a����2��|s�a����4���+�46656�a����6��|s�:�����W��6लWhen�[:w���e�need�an�explicit�W��*�eierstrass�equation,��@w�e'll�often�use�the�simplied�v�ersion����6�for�1�con���v�enience,�8�k�eeping�in�mind�that�the�argumen�ts�w�ould�need�to�b�Ge�mo�died�to����6�t�.sthe�full�six-co�Gecien���t�W��*�eierstrass�form�if�w�e�are�going�to�b�Ge�able�to�apply�them����6�to�jOcurv���es�o�v�er�elds�with�c�haracteristic�2�or�3.���In�the�simplied�case,���ho�w�ev�er,�������7
�������@v��ON���THE�TORSION�POINTS�OF�ELLIPTIC�CUR��#�VES�&�MODULAR�ABELIAN�V�ARIETIES����3����V������6लdene��,the��discriminant���of�the�W��*�eierstrass�equation�to�b�Ge��R�=���16(4�A���^��3���7�+�p�27�B���q��^��2����).����6�The�
�general�form�of�the�discriminan���t�is�not�enligh�tening�for�our�purp�Goses,��but�note����6�that���it�is�not�just��with�the�rev���erse�substitutions�for��A��and��B�``�(b�Gecause�then�it����6�w���ould�?alw�a�ys�b�Ge�a�m�ultiple�of�16).�j[The�discriminan�t�of�an�elliptic�curv�e�is�0�if�and����6�only�(�if�it�is�singular.��yThe�group�la���w�still�applies�to�the�non-singular�p�Goin�ts�of�a����6�giv���en�UUsingular�curv�e,�ho�w�ev�er.����B�Finally��*�,�UUw���e�remark�that�elliptic�curv�es�are�ab�Gelian�v��q�arieties�of�gen�us�1.��!%t����&2.����%�F��Uormal��Gr��oups����6ल2.1.��The��EBasics.��The�Ifollo���wing�mac�hinery�is�giv�en�in�Silv�erman�[�17��
],�K�c�hapter�IV.����6�W��*�e�P}try�to�b�Ge�a�little�clearer�in�the�explanations,��Gsometimes�lling�in�argumen���ts����6�that��Rhe�elides.�OThis�is�eviden���t�particularly�when�it�comes�to�facts�ab�Gout�the�in�v�erse����6�map,��Iwhic���h��Eare�assumed�in�the�source�without�an�y�men�tion�of�what�exactly�is�b�Geing����6�assumed.�
"But��they�aren't,��in�fact,�terribly�tedious�or�dicult�to�pro���v�e,�and��so�w���e����6�pro���v�e�UUthem�here.������6��Denition��=2.1.��X�A�@��formal�l�gr��}'oup�‸F�?!�o���v�er�@�a�ring��R�T��is�a�t���w�o-v��q�ariable�@�p�Go�w�er�series����6वF�c��(�X�:�;���Y�8�)���2��R�Dz[[�X�;���Y�8�]]�UUsatisfying�the�follo���wing�v�e�prop�Gerties:��T������G�(1)����Y�ĵF�c��(�X�:�;���Y�8�)��=��X�²+�8�Y��+�UU[terms�of�higher�degree].�������G�(2)����Y�ĵF�c��(�X�:�;���F��(�Y��9;�Z���))��=��F�c��(�F��(�X�:�;�Y�8�)�;�Z���).�������G�(3)����Y�ĵF�c��(�X�:�;���Y�8�)��=��F��(�Y��9;���X���).�������G�(4)����Y�ĵF�c��(�T��V;���i�(�T��))��=�0�UUhas�a�unique�solution�for��i�(�T�c��)���2��R�Dz[[�T��]].�������G�(5)����Y�ĵF�c��(�X�:�;����0)��=��X���,�UU�F��(0�;���Y�8�)�=��Y��.����6�Observ���e�ezthat�these�v�e�prop�Gerties�pro�vide�an�ab�Gelian�group�structure�without�an�y����6�explicit�0�elemen���ts.�e�(The�second,�8third,�and�0�fourth�prop�Gerties�are�asso�ciativit���y��*�,�8com-����6�m���utativit�y��*�,���and���in�v�erses,�resp�Gectiv�ely��*�,�and���the�fth�pro���vides�an�iden�tit�y��*�.)�=�Instead�of����6�elemen���ts,��)there�}�is�only�the�p�Go�w�er�series��F�፲whic�h�pro�vides�the�recip�Ge�for�the�group����6�op�Geration:�q�hence�UUthe�name�\formal�group."������6��Denition���2.2.�|�A�p��homomorphism�51�b�Get���w�een�p�t�w�o�formal�groups��f�q�:���F��!�G���dened����6�o���v�er��a�ring��R��زis�a�one-v��q�ariable�p�Go���w�er��series��f���(�T�c��)��Q�2��R�Dz[[�T��]]��with�no�constan���t�term����6�suc���h�WZthat��f���(�F�c��(�X�:�;���Y�8�))��=��G�(�f��(�X���)�;���f��(�Y�8�)).�The�WZformal�groups��F�U��and��G��+�are��isomorphic����6लif�
�there�exists�a�homomorphism��f�%�as�ab�Go���v�e�
�and�a�rev���erse�homomorphism��g�fo�suc�h����6�that�UU�f���(�g�[ٲ(�T�c��))��=��g��(�f���(�T�c��))�=��T��.����B�F��*�or��{example,���dene�the�map�[0]�X:��F��!�F����b���y��{[0](�T�c��)�=�0.��:It�is�a�homomorphism����6�b�Gecause��v����S[0](�F�c��(�X�:�;���Y�8�))��=�0�=��F��(0�;����0)�=��F��([0](�X���)�;����[0](�Y�8�))�:������6लThat���is�an�en���tirely�trivial�example,���but�it�is�imp�Gortan�t:�H�w�e�can�use�it�to�dene�a����6�map�UU[�m�]��:��F��Q!�F�S��inductiv���ely�for�an�y�in�teger��m��b�y���C�����������x�[[�m�](�T�c��)���������=���F�c��([�m�8���1](�T��)�;���T��)�������������:2�(�m��>��0)������������������x�[[�m�](�T�c��)���������=���F�c��([�m�8�+�1](�T��)�;���i�(�T��))�������������:2�(�m��<��0)�����Bߍ�6�W��*�e'll�zysho���w�brie
y�that�these�maps,���whic�h�are�(p�Gerhaps�confusingly�at�rst)�called����6��multiplic��}'ation��by�m�,�Y�are�X�all�homomorphisms�(w���e'll�sa�y�that�the�case��m���>��0�X�will�do����6�to�UUsho���w�this;�the�other�case�is�similar).��������7
��������6ऱ4����SETH���KLEINERMAN���V��������6��Pr��}'o�of.���V5�Assume�UUthat�[�m�8���1]�UUis�kno���wn�to�b�Ge�a�homomorphism.�q�Then�����������x�P[�m�](�F�c��(�X�:�;���Y�8�))��������,�=���F�c��([�m�8���1](�F��(�X�:�;���Y�8�))�;�F��(�X�:�;�Y�8�))���������������������,�=���F�c��(�F��([�m�8���1]�X�:�;����[�m����1]�Y�8�)�;�F�c��(�X�:�;�Y��))�������������������,�=���F�c��([�m�8���1]�X�:�;���F��([�m����1]�Y��9;���F��(�X�:�;�Y�8�)))�������������������,�=���F�c��([�m�8���1]�X�:�;���F��(�F��([�m����1]�Y��9;���Y�8�)�;�X���))�������������������,�=���F�c��([�m�8���1]�X�:�;���F��([�m�]�Y��9;�X���))�������������������,�=���F�c��(�F��([�m�8���1]�X�:�;���X���)�;��[�m�]�Y�8�)�������������������,�=���F�c��([�m�]�X�:�;����[�m�]�Y�8�)�����!Q��6�where��sw���e'v�e�used�the�comm�utativit�y�and�asso�Gciativit�y�prop�Gerties�of�the�formal����6�group�UUa�few�times.�q�In�particular,�[�m�]�is�a�homomorphism�as�w���e'd�w�an�ted.����B�W��*�e�D�should�also�note�that�in�the�do���wn�w�ard�D�induction�w���e�need�to�use�the�fact����6�that����i�<R�:��F�:�!�F��9�,��Athe�unique�in���v�erse�map�asso�Gciated�to�the�formal�group��F��9�,��Ais�also����6�a�UUhomomorphism.�q�T��*�o�see�this,�consider�the�expression��������F�c��(�F��(�X�:�;���iX���)�;�F�c��(�Y��9;�iY�8�))��=��F��(�F��(�X�:�;���Y�8�)�;�F��(�iX�:�;�iY�8�))����6�obtained�_Db���y�rep�Geated�application�of�comm�utativit�y�and�asso�Gciativit�y�of�the�formal����6�p�Go���w�er�Wyseries.�)The�left�hand�side�is�trivially�0.�On�the�righ���t,��?remem�b�Ger�Wythat��iF�c��(�X�:�;���Y�8�)����6�is�/Othe�unique�p�Go���w�er�/Oseries�for�whic���h��F�c��(�F��(�X�:�;���Y�8�)�;�iF�c��(�X�;�Y�8�))�2b=�0.���So�/O�iF�c��(�X�;���Y�8�)�=����6वF�c��(�iX�:�;���iY�8�)�Xand��i�vf�:��F�t�!�F�VX�is�Xa�homomorphism.�z$The�second�induction�therefore����6�follo���ws�UUexactly�as�the�rst�did.���#}�!����

msam10����C荑BलW��*�e�UUwill�no���w�pro�v�e�t�w�o�facts�ab�Gout�the�m�ultiplication-b�y-�m��map.������6��Prop�Q�osition��p2.3.�F��F��*�or��Wan���y�in�teger��m�,���the�map�[�m�](�T�c��)��=��mT��+��W[higher�order�terms].������6��Pr��}'o�of.���V5�W��*�e�~rst�induce�to�obtain�the�result�for�non-negativ���e��m�.���It�is�clear�that����6�[0](�T�c��)�@�=�0�ьw���orks.��lNo�w�assume�the�result�ab�Gout�the�map�[�m�6Y���1].��lUsing�the����6�recursiv���e�UUdenition�and�then�prop�Gert�y�1�of�formal�groups,�w�e�ha�v�e����������������[�m�](�T�c��)��������P=���F�c��([�m�8���1](�T��)�;���T��)���������������������P=��[�m�8���1](�T�c��)�+��T��o�+���[higher�UUorder�terms]��������u��������������P=��(�m�8���1)�T��o�+��T��c�mo�Gd��!�~�T��c������2������!Q��6लwhic���h�UUis�exactly�what�w�e�needed.����B�F��*�or��8the�do���wn�w�ard��8induction,���w�e'll�need�to�establish�a�minor�result�ab�Gout��i�(�T�c��)����6�again,��namely�
Nthat�it�has�no�constan���t�term.�Y�This�follo�ws�from�prop�Gert�y�5�of�formal����6�p�Go���w�er�UUseries:��⼍��{�i�(0)��=��F�c��(0�;���i�(0))�=��F��(�T��V;���i�(�T��))�j����T���=0��3Ѳ=�0�:���U��BलNo���w,�UUnotice�that�������0��=��F�c��(�T��V;���i�(�T��))��=��T��o�+�8�i�(�T�c��)�+���:�:�:�����6लwhere�UUthe�[��:���:�:�����]�are�higher�order�com���binations�of��T���and��i�(�T�c��).�q�Rearrange�to�nd��Lj��i�(�T�c��)��=���T��o�+��8�:���:�:�����6लW��*�e��[sho���w�ed�that��i�(�T�c��)�has�no�constan�t�term,��]so�ev�erything�on�the�righ�t�hand�side����6�after�UUthe���T���v��q�anishes�mo�Gdulo��T��c���^��2����.�q�No���w�the�do�wn�w�ard�induction�is�clear.��!�h����C荑BलF��*�or��5the�second�prop�Gosition,��ow���e�ha�v�e�to�rst�pro�v�e�a�standard�result�ab�Gout�formal����6�p�Go���w�er�UUseries.�����*Ơ�7
�������@v��ON���THE�TORSION�POINTS�OF�ELLIPTIC�CUR��#�VES�&�MODULAR�ABELIAN�V�ARIETIES����5����V������6��Lemma�۟2.4.�H�L��}'et���R����b�e�a�ring�and��f���(�T�c��)��6�2��R�Dz[[�T��]]����a�one-variable�p��}'ower�series,��2with����6वf���(�T�c��)�qz=��aT��+����[higher�or��}'der�terms].���If��a�qz�2��R��ǟ�^�������,�	xthen���ther�e�is�a�unique�p�ower�series����6वg�[ٲ(�T�c��)�K��2��R�Dz[[�T��]]�jZ�that�c��}'omp�oses�jZwith��f���(�T��)��so�that��f���(�g�[ٲ(�T��))�K�=��T��.��This�jZp��}'ower�series����6�happ��}'ens���to�also�satisfy��g�[ٲ(�f���(�T�c��))��=��T��.��2��6�Pr��}'o�of.���V5�The� +goal�is�to�pro�Gduce�a�p�o���w�er� +series�that�satises��f���(�g�[ٲ(�T�c��))�����T����mo�dulo� +�T��c���^��n�����6लfor��8an���y��n��>��1.�EhW��*�e'll��8do�this�b�y�explicitly�constructing�the�p�Golynomials��g����n��q~�(�T�c��)�whic�h����6�w���e��;dene�to�b�Ge��g�[ٲ(�T�c��)�mo�dulo��T��c���^��n�+1����;�-since�they'll�b�e�inductiv���ely�w�ell-dened,��4that����6�will�UUgiv���e�us�a�precise�denition�of�the�formal�p�Go�w�er�series��g�[ٲ(�T�c��).����B�First,��tak���e����g����1��|s�(�T�c��)���=��a���^���1��
�t�T��.��Clearly����f���(�g����1��|s�(�T��))�=��a�(�a���^���1��
�t�T��)���+���:���:�:���~���ܵT�G��mo�Gd����T����^��2����,��as����6�w���e'd�UUw�an�ted.����B�Next,��assume��N�f���(�g����k�+B��1��(�(�T�c��))������T�Fݲmo�Gd��T��c���^��k��O�.��W��*�e'll�ha���v�e��g����k���F�=����g����k�+B��1�����+����r�GT��c���^��k��
2m�for�some����6वr�5�2���R�Dz.�q�Observ���e�UUthat���V������������O�f���(�g����k��됲(�T�c��))��������A�=���f���(�g����k�+B��1��(�(�T�c��)�8�+��r�GT�������k��O�)�������������������A�����f���(�g����k�+B��1��(�(�T�c��))�8�+��ar�GT�������k���O�mo�d��&��T�������k�+B�+1������#d��6लsince��Uall�cross-terms�con���tributed�b�y�the��r�GT��c���^��k��	t�monomial�ha�v�e�degree��>��k�P��.�=�Remem�b�Ger����6�b���y�UUour�initial�assumption�that���������������z�f���(�g����k�+B��1��(�(�T�c��))���������Q����T��c�mo�Gd��!�~�T��c������k���������������������Q����T��o�+�8�bT��c������k���O�mo�Gd��&��T��c������k�+B�+1����������6लfor�UUsome�elemen���t��b���2��R�Dz.�q�Then�UUput�it�all�together�to�get�������f���(�g����k��됲(�T�c��))�����T��o�+�8�bT�������k�����+��ar�GT�������k���O�mo�d��&��T�������k�+B�+1��o"�:����6लSo�rthere's�a�unique�c���hoice��r�5�=����b=a��at�eac�h�stage,�)lw�ell-dened�b�Gecause��a��is�a�unit,����6�suc���h�]zthat��f���(�g����k��됲(�T�c��))�S���T��/�mo�Gd���.�T����^��k�+B�+1��o"�,���and�]zthe�formal�p�Go���w�er�]zseries��g�[ٲ(�T��)�therefore����6�exists.����B�No���w��that�w�e�kno�w��f���(�g�[ٲ(�T�c��))�Y�=��T��,�CPw���e��also�ha�v�e��g�[ٲ(�f���(�g��(�T�c��))�Y�=��g�[ٲ(�T��),�CPwhic���h��is����6�a�h8formal�iden���tit�y�h8in��R�Dz[[�g�[ٲ(�T�c��)]].��qIn�particular,���under�the�formal�map�of�v��q�ariables����6वg�[ٲ(�T�c��)���7!��T��,�UUw���e�nd�that��g�[ٲ(�f���(�T��))��=��T���as�UUw���e'd�claimed.����B�Finally��*�,�E�to�Bsho���w�uniqueness�of��g�[ٲ,�assume�there's�another�p�Go���w�er�Bseries��h�(�T�c��)�suc���h����6�that�UU(�f�Lo��8�h�)(�T�c��)��=��T��;�UUthen�����~z�g�[ٲ(�T�c��)��=��g�����8�(�f�Lo���h�)(�T��)��=�(�g�����8�f���)����h�(�T��)��=��h�(�T��)�:�����������X��BलNo���w��Rw�e�can�pro�v�e�the�second�fact,�Swhic�h�quite�naturally�follo�ws�from�the�ab�Go�v�e.��2�6��Prop�Q�osition��f2.5.��W�If�ݑthe�in���teger��m��is�a�unit�in��R��ǟ�^�������,�?�then�[�m�]�T�:��F�S!�F��ʲis�ݑan����6�isomorphism.������6��Pr��}'o�of.���V5�>F��*�rom���Prop�Gosition�2.3,���[�m�](�T�c��)�K�=��mT��w�+��m�:���:�:����;�̧kno���wing���that��m�K��2��R��ǟ�^����Q��and���ap-����6�plying�z5Lemma�2.4,��lthere�is�a�p�Go���w�er�z5series�[�m�]���^���1��
�t�(�T�c��)�suc���h�that�[�m�]���^���1���([�m�](�T�c��))��5=����6�[�m�]([�m�]���^���1��
�t�(�T�c��))��=��T��.�J	The��only�thing�remaining�is�to�sho���w�that�[�m�]���^���1��
���is�a�homomor-����6�phism�UU�F��Q!��F��9�.�q�W��*�e�kno���w�[�m�]�is�a�homomorphism,�so���/�����������w�[�m�]�������1��
�t�F�c��(�X�:�;���Y�8�)����������=��[�m�]�������1��
�t�F�c��([�m�][�m�]�������1���(�X���)�;����[�m�][�m�]�������1���(�Y�8�))������u����������������=��[�m�]�������1��
�t�[�m�]�F�c��([�m�]�������1���(�X���)�;����[�m�]�������1���(�Y�8�))���������������������=���F�c��([�m�]�������1��
�t�(�X���)�;����[�m�]�������1���(�Y�8�))���������6�giv���es�UUus�what�w�e�need.���c��������:a��7
��������6ऱ6����SETH���KLEINERMAN���V������6ल2.2.��So��%where�are�the�elemen��9ts?��The���natural�thing�to�do�after�dening�a�formal����6�group�Z�is�to�think�ab�Gout�supplying�elemen���ts�that�can�follo�w�this�recip�Ge�for�their�group����6�structure.����B�Recall���(A���tiy�ah-Macdonald�[�1��],���c�h.�.�10)�that�a�complete�lo�Gcal�ring��R��t�with�maximal����6�ideal�UU�M�lp�is�dened�as�a�ring�that�is�equal�to�its�natural�in���v�erse�UUlimit:���H���˵R����T͍���߸���+3����߲=�������
ilim����
�;� ������_�������N�R��=��q�M�������n�����:�����6लIf�g�w���e�ha�v�e�a�p�Go�w�er�series��f�{��and�an�elemen�t���2���M��,�l�w�e�can�use�these�to�pro�Gduce�a����6�unique�Telemen���t������n��8��2���M����^��n�����,�$�since�truncating�the�p�Go�w�er�series�at�the�degree��n��term�is����6�ne�an(all�higher�terms�will�b�Ge�congruen���t�to�zero�mo�dulo��M����^��n�����).��The�completeness�of����6�the�ǹring�indicates�that�this�sequence�of�elemen���ts�(�a����1��|s�;����:�:�:����;���a����n��q~�;��:�:�:��
UO�)�ǹis�under�isomor-����6�phism�1�an�elemen���t�of��R�Dz.�AIn�particular,�h�this�argumen�t�extends�to�sho�w�that�with����6वR��;���M�n�as�V�ab�Go���v�e,�WKa�p�Go�w�er�series�in�t�w�o�v��q�ariables��F�c��(�X�:�;���Y�8�)�giv�en�as�the�group�la�w�for����6�a�W�formal�group�will�con���v�erge�W�for��X�:�;���Y�ĸ2���M��.�x�The�set��M�n��is�th���us�made�in�to�a�group,����6�denoted��Z�F��9�(�M��),��with�the�group�comp�Gosition��x�t���y�"�=���F�c��(�x;���y�[ٲ)��Zand�in���v�erse��Z�	�x��=��i�(�x�).����6�The�O:closure�of�the�group�is�sho���wn�b�y�prop�Gert�y�1�of�formal�p�Go�w�er�series�(Denition����6�2.1):�>;the��=equation�tells�us�that��x�j����y�"����0��=mo�Gdulo��M��.�OjFinally��*�,�ܸF��9�(�M����^��n�����)��=is�a�subgroup����6�of�UU�F��9�(�M��),�since�it�is�closed�under�comp�Gosition�b���y�precisely�the�same�argumen�t.��Y1��6��R��}'emark�⸲2.6�.�&��No���w���w�e�see�exactly�wh�y�the�m�ultiplication-b�y-�m��map�w�as�so�called;����6�the�h�map�on�the�formal�group�induces�a�map�on�the�asso�Gciated�group��F��9�(�M��)�whic���h����6�acts�UUb���y�comp�Gosing�an�elemen�t�with�itself��m��times.����B�W��*�e�t�no���w�pro�v�e�a�ma��8jor�result�ab�Gout�torsion�in�groups�asso�ciated�to�formal�groups.����6��Prop�Q�osition�D�2.7.��˲Let����F��5�b�Ge�a�formal�group�o���v�er���a�complete�lo�cal�ring��R��òwith����6�maximal���ideal��M��,��and�let��k��H�b�Ge�the�residue�eld��R��=��q�M��.�e�Let�the�c���haracteristic�of��k����6लb�Ge�l��p�.���(Zero�c���haracteristic�is�allo�w�ed.)���Then�ev�ery�elemen�t�of�nite�order�in��F��9�(�M��)����6�has�UUorder�a�p�Go���w�er�UUof��p�.������6��Pr��}'o�of.���V5�F��*�or��p�Gositiv���e��p�,�if�there�w�ere�a�p�Goin�t��r�5�2���M�
��that�serv�ed�as�a�coun�terexample,����6�it���w���ould�ha�v�e�order��p���^��k��됵m��for�some��m��relativ�ely�prime�to��p�.���There�w�ould�then�also����6�b�Ge�-a�p�oin���t�of�order��m�,�5namely�[�p���^��k��됲]�r��.�dXTherefore�it�is�enough�to�pro���v�e�-that�there�are����6�no�w p�Goin���ts�of�order��m��with��p��and��m��relativ�ely�prime.��((F��*�or��p��j�=�0,��w�e�w ha�v�e�to�pro�v�e����6�that�UUthere�are�no�p�Goin���ts�of�order��m��>��0�UUat�all.)����B�Then�}alet��m��b�Ge�a�p�ositiv���e�in�teger�c�hosen�as�ab�Go�v�e,��dand�let��x��b�Ge�a�p�oin���t�of�order����6वm�,�Řso��$[�m�](�x�)�\�=�0.�5Observ���e�that��m��cannot�b�Ge�an�elemen�t�of�the�ideal��M��,�Řb�Gecause����6�then��5�R��=��q�M��P�w���ould�ha�v�e�c�haracteristic�dividing��m�,��nwhic�h�is�not�p�Germitted.�G�Therefore����6वm�uθ2��R��ǟ�^�������,��^and��)applying�Prop�Gosition�2.5,�the�map�[�m�]�u�:��F�t!�F��b�is��)an�isomorphism.����6�That's�Gall�w���e�need,�OCb�Gecause�it�induces�an�isomorphism�of�groups�[�m�]�U:��F��9�(�M��)��!����6�F��9�(�M��),�UUwhose�k���ernel�consists�only�of�the�p�Goin�t�0,�so��x��m�ust�b�Ge�trivial.��.���������6ल2.3.��More�}rTheory�of�F��
�ormal�Groups.��If��the�ring�is�equipp�Ged�with�a�discrete����6�v��q�aluation�UU�v�[ٲ,�w���e�can�sa�y�something�further�ab�Gout�the�torsion�in�the�group��F�c��(�M��).����6��Prop�Q�osition�]�2.8.��;�Let��$�R��b�Ge�a�complete�lo�cal�ring�with�maximal�ideal��M��,��and�with����6�a��discrete�v��q�aluation��v�[ٲ.�7sLet��p��b�Ge�the�c���haracteristic�of�the�eld��R��=�M��.�7sLet��F��Dzb�Ge�a����6�formal�Tgroup�o���v�er�T�R�Dz,��with�asso�Gciated�group��F��9�(�M��).�VqGiv���en�an�elemen�t��x���2�F��9�(�M��)�Tof����6�prime�UUp�Go���w�er�order��p���^��n��q~�,�w�e�ha�v�e��덒‘��v�[ٲ(�x�)��������<$�����v��(�p�)���K�w�fe+hN�	(֍�p���r�n���^��8�p���r�n��1������0�̵:�����MQ��7
�������@v��ON���THE�TORSION�POINTS�OF�ELLIPTIC�CUR��#�VES�&�MODULAR�ABELIAN�V�ARIETIES����7����V������BलThis�<prop�Gosition�relies�on�the�notion�of�in���v��q�arian�t�<dieren�tials�for�formal�groups,����6�the�V�in���tro�Gduction�of�whic�h�ma�y�tak�e�us�a�bit�to�Go�far�aeld.��It�is�pro�v�ed�quite�concisely����6�in�UU[�17��
],�c���h.�q�IV,��x�4�and��x�6.��[]���S.o3.��_���Applica��UTtion��to�Elliptic�Cur��rves:���the�Finitude�of��E����(�K���)����&�-�
cmcsc10�tor�����BलIn��this�section�w���e�will�demonstrate�that�torsion�subgroups�of�elliptic�curv�es�o�v�er����6�n���um�b�Ger��9elds�are�nite,���as�a�consequence�of�a�demonstrably�injectiv���e�map�in�to�a����6�nite�rPgroup.�ȸThis�result�also�follo���ws�from�the�Mordell-W��*�eil�theorem�(whic�h�states����6�that�H�the�group��E����(�K���)�of�an�elliptic�curv���e�o�v�er�a�n�um�b�Ger�eld�is�nitely�generated).����6�Ho���w�ev�er,�fLthe���argumen�t�that�w�e�will�presen�t�here�is�indep�Genden�t�and�easier�to����6�pro���v�e.�P�Additionally��*�,��it��giv�es�an�eectiv�e�b�Gound�(since�one�can�compute�the�n�um�b�Ger����6�of��p�Goin���ts�in�the�reduction�of�the�curv�e�easily),�όwhere�the�Mordell-W��*�eil�theorem�do�Ges����6�not.��ޭ��6�3.1.��An��2Elliptic�Curv��9e's�F��
�ormal�Group.��In�E�what�follo���ws,�H�w�e�E�will�use�the�sim-����6�plied���form�of�the�elliptic�curv���e,��so�the�full�generalit�y�of�c�haracteristic�2�and�3�is����6�denied���us;��ho���w�ev�er,��[�17��
],�IV����x�1,�do�Ges�that�rather�w���ell.�Q�In�c�ho�Gosing�to�w�ork�with�the����6�simplied��"curv���e,���w�e�hop�Ge�to�trade�full�generalit�y�for�some�added�in�tuition,���as�the����6�algebraic���con���tortions�with�the�v�e-parameter�form�are�not�particularly�enligh�ten-����6�ing.�cbThe�*%follo���wing�section�is�mostly�sk�etc�hes�to�get�a�grasp�of�what's�going�on;�8�the����6�form���ulas�UUthemselv�es�will�b�Ge�elided�most�of�the�time.����B�Giv���en�Fan�elliptic�curv�e��E�Z��:���y��[ٟ�^��2���d�=��x���^��3��7�+��ĵAx��+��B��q�,��p�Gerform�Fthe�c�hange�of�co�Gordinates����6वs�ކ�=�������:ñ1�����&�feN������y�����	�s�,�f�t��=���������x�����&�fe������/y�����	�в.���The�cdadditiv���e�iden�tit�y�b�Gecomes�(0�;����0)�under�the�new�(�s;�t�)-co�Gordinate�����6�system.�q�The�UUequation�of�the�curv���e�b�Gecomes�������oU�s���=��t������3���S�+�8�Ats������2���+��B��qs������3��|s�:����6लWhat�ʌif�w���e�w�an�t��s��as�a�function�en�tirely�of��t�?�C�The�natural�thing�to�do�is�recursiv�ely����6�substitute�UUthe�equation�for��s��bac���k�in�to�itself,�as�follo�ws:������������Q���s������X��=���t������3���S�+�8�Ats������2���+��B��qs������3�������u�������������X��=���t������3���S�+�8�At�(�t������3���+��Ats������2���+��B��qs������3��|s�)������2���+��B��q�(�t������3���+��Ats������2���+��B��qs������3��|s�)������3�������������������X��=���t������3���S�+�8�At������7���+��B��qt������9���+��s������2��|s�(2�A������2���t������5���S�+�3�AB��qt������7���)�+��s������3���(2�AB��qt������4���S�+�3�B�������2����t������6��|s�)������������������_�+�8�s������4��|s�(�A������3���t������3���S�+�3�A������2���B��qt������5���)�+��s������5���(2�A������2���B��qt������2���S�+�6�AB�������2����t������4��|s�)������������������_�+�8�s������6��|s�(�AB���q�����2����t��+�3�B���q�����3���t������3���S�+��A������3��|s�B��qt������3���)�8�+��s������7��|s�(�A������2���B�������2����t������2���)�8�+��s������8��|s�(3�AB�������3����t�)�+��s������9��|s�(�B�������4����)�:������썑6लThis���rep�Geated�substitution�gets�us�a�p�olynomial�in��t��follo���w�ed���b�y�a�p�Golynomial�in����6वs��IJand��t�;���b���y�con�tin�uing�inductiv�ely��*�,��`the�p�Go�w�ers�of��s��get�larger�and�larger,��`and�th�us����6�con���tribute�Plarger�and�larger�p�Go�w�ers�of��t�,��Rso�the�pro�Gcedure�feels�lik�e�it�should�con�v�erge����6�to��an�elemen���t�of�the�ring�of�formal�p�Go�w�er�series��R�Dz[[�t�]].��OIn�fact,�?Nit�do�Ges;�nthis�is�a����6�consequence�UUof�Hensel's�lemma�(see�Silv���erman�[�17��
],�IV�Lemma�1.2.).����B�In�%Wparticular,�.�there�is�a�unique�formal�p�Go���w�er�%Wseries��s�(�t�)���2��R�Dz[[�t�]]�%Wsuc���h�that��s�(�t�)��=����6वt���^��3���ڲ+�|g�ats�(�t�)���^��2���+��B��qs�(�t�)���^��3��|s�.���Letting���the�expression��f���(�s;���t�)�o�=��t���^��3���ڲ+��Ats���^��2���+��B��qs���^��3��|s�,���then����s�(�t�)����6�satises�ϗ�f���(�s�(�t�)�;���t�)���=��s�(�t�).��W��*�e�kno���w�the�rst�few�terms�as�b�Gefore,��'after�a�few�self-����6�substitutions:�����������������T�s�(�t�)����������=���t������3���S�+�8�At������7���+��B��qt������9���+�2�A������2��|s�t������11��
�Ʋ+�5�AB��qt������13���������������6ल(3.1)���������������������n+�8�(5�A������3���S�+�3�B���q�����2����)�t������15��
�Ʋ+�21�A������2��|s�B��qt������17���+���:���:�:���������`���7
��������6ऱ8����SETH���KLEINERMAN���V������6लNo���w�UUsubstituting�bac�k�for�the�standard�co�Gordinates,�w�e�ha�v�e���ʍ����������8��x�(�t�)�������L}�=�����<$��
5ڵt���K�w�fe��	(֍s�(�t�)�����&=�����<$����1���K�w�fe�	(֍�t���r�2������C��8�At������2���S���B��qt������4�����A������2��|s�t������6�����3�AB��qt������8���+�(��2�A������3�����2�B���q�����2����)�t������10��
�Ƹ��10�A������2��|s�B��qt������12���
#��:���:�:�������������������96y�[ٲ(�t�)�������L}�=�����<$��	�1���K�w�fe��	(֍�s�(�t�)�����&=�����<$����1���K�w�fe�	(֍�t���r�3������C��8�At����B��qt������3���S���A������2��|s�t������5�����3�AB��qt������7���+�(��2�A������3�����2�B���q�����2����)�t������9�����10�A������2��|s�B��qt������11���
#��:���:�:������0��BलBecause�A�of�our�discussion�ab�Go���v�e�A�on�the�con���v�ergence�A�of�formal�p�o���w�er�A�series,�E�if��K����6लis���a�complete�lo�Gcal�eld�with��R��L�and��M�נ�as�w���e'v�e���b�een�using�them�ab�o���v�e,��Hthen���taking����6वt�$M�to�b�Ge�an�elemen���t�of��M�;h�gets�us�a�p�oin���t�in��E����(�K���).�ao(The�fact�that�these�are�Lauren�t����6�series�T�do�Gesn't�c���hange�an�ything�ab�Gout�the�con�v�ergence�b�Gecause�w�e're�only�adding�a����6�nite�D�n���um�b�Ger�of�terms�to�a�series�kno�wn�to�b�Ge�con�v�ergen�t.)�lNF��*�urthermore,�H2the�map����6�is���injectiv���e,���since��t�aL�=��x=y�
��is���the�in�v�erse.��XW��*�e�can�think�of��t��as�a�parameter�for�all����6�p�Goin���ts�UUon�the�curv�e�with��t���=��x=y�"�2��M��.����B�Finally��*�,�ۚsince��+there�is�a�group�la���w�for�the�explicit�addition�of�p�Goin�ts��P����1��C��=��(�x����1��|s�;���y����1���)����6�and�`̵P����2��C��=��(�x����2��|s�;���y����2���)��2��E����,���w���e�`�can�mak�e�a�corresp�Gonding�form�ula�for�the�formal�addition����6�of���indeterminates��t����1��V>�and��t����2��|s�,��whic���h�w�e�esc�hew�here,��since�w�e�ha�v�e�made�it�a�p�Goin�t�to����6�a���v�oid�[Othe�group�la���w�calculations�ab�Go�v�e.���Ho�w�ev�er,�\�the�addition�form�ula�is�a�formal����6�p�Go���w�er���series�in�t���w�o���v��q�ariables��F�c��(�t����1��|s�;���t����2���).��kIt���is�constructed�to�ob�ey�all�the�prop�erties����6�that��+the�usual�addition�on�elliptic�curv���es�ob�Geys;���therefore�it�is�the�basis�for�a�formal����6�group.��q��6��Denition�yI3.1.�U��The���formal�group���x䍑3�^������E����"�(o���v�er���R�Dz)�asso�Gciated�to�the�elliptic�curv���e��E����6ल(o���v�er�UU�K���)�is�the�formal�p�Go���w�er�UUseries��F���dened�ab�o���v�e,�UUwhic�h�giv�es�the�group�la�w.��b��6�3.2.��Reduction��ion�Elliptic�Curv��9es.��Let�n��K�%��b�Ge�a�lo�cal�eld�with�discrete�v��q�alua-����6�tion�`G�v�[ٲ,��Jand��R�t�b�Ge�the�ring�of�in���tegers�of��K���.� Recall�that�since��R��߲=���f�x��2��K�~4�:��v��(�x�)����0�g�,����6�w���e���can�decomp�Gose��R��O�in�to�a�group�of�units��R��ǟ�^����23�with�v��q�aluation�0�and�the�set�of�ele-����6�men���ts��~with�p�Gositiv�e�v��q�aluation;���these�last�constitute�an�ideal�in��R�Dz,���whic�h�is�clearly����6�the���unique�maximal�ideal�since�it�con���tains�all�non-units,�ڄso�w�e'll�call�it��M���as�b�Gefore.����6�(The�UUring�of�in���tegers�of�a�lo�Gcal�eld�is�th�us�a�lo�Gcal�ring.)����B�No���w���let��M�˲b�Ge���[�R�w�for�some�uniformizing�elemen�t���[ٲ,�$and�for�con�v�en�tion's�sak�e����6�let's�4Zsa���y�that�the�discrete�v��q�aluation�has�b�Geen�normalized�so�that��v�[ٲ(���)��=�1.�f�Finally��*�,����6�let�(ݵk�yt�b�Ge�the�eld��R��=��q�M��.�b�In�a�bit�w���e'll�b�e�able�to�think�ab�out�mo���ving�a�curv�e's�base����6�eld�UUfrom��K�q�to��k�P��.�q�But�rst�w���e�ha�v�e�to�put�the�curv�e�in�canonical�form.����6��Denition��O3.2.�:IJGiv���en�z/an�elliptic�curv�e��E���=K�1K�with�W��*�eierstrass�equation��y��[ٟ�^��2��Z�+����a����1��|s�xy��p�+����6वa����3��|s�y�=�=��5�x���^��3���Ų+�CR�a����2���x���^��2���+�CR�a����4���x��+��a����6���,�h�w���e�esa�y�the�equation�is��minimal�m��if��a����i��5��2��5�R�xDzand��v�[ٲ()�is����6�minimized�UUo���v�er�all�p�Gossible�c�hanges�of�co�Gordinates.����6��R��}'emark����3.3�.�턲There�@�is�a�minimal�equation�represen���ting�an�y�curv�e,�D�b�Gecause�the�v��q�al-����6�uation�UU�v��.�is�discrete.����6��R��}'emark�]��3.4�.�v۲The�lexplicit�c���hange�of�co�Gordinates�allo�w�ed�is�giv�en�in�[�17��
],�L�c�h.��
I�GI�I,����6सx�1.��kObserv���e���that�w�e�can�also�minimize�a�simplied�W��*�eierstrass�equation��y��[ٟ�^��2��	�g�=����6वx���^��3�����+�,��Ax��+��B�O��if��%w���e're�not�w�orking�o�v�er�a�lo�Gcal�eld�with�residue�c�haracteristic�2�or�3:����6�certainly�͏w���e�can�minimize�the�v��q�aluation�of�the�determinan�t�in�the�v�e-parameter����6�form,�5�then�-�do�the�transformations�1.1�and�1.2�in�order�to�put�it�in�simplied�form.����6�The��8determinan���t�c�hanges�b�y�a�factor�of�1�=�6���^��12��x�,�pso�it�remains�minimal�with�our����6�v��q�aluation.�oHTherefore�M�in�what�follo���ws�w�e'll�consider�elliptic�curv�es�of�the�form��y��[ٟ�^��2���d�=����6वx���^��3���S�+�8�Ax��+��B��q�,�UUwith��A;���B�G��2���#���

msbm10�Z�.�����	tz��7
�������@v��ON���THE�TORSION�POINTS�OF�ELLIPTIC�CUR��#�VES�&�MODULAR�ABELIAN�V�ARIETIES����9����V������6��Denition�9�3.5.�4z�The���natural��r��}'e�duction��]map�Nb�is���the�map��R�l��!�X۵R��=��q�M��,�¨whic���h�action����6�on�Fthe�elemen���ts�of��R�*
�w�e�denote�b�y��r�5�7!���:\�~�����r����,�.�\�The�reduction�of�an�elliptic�curv�e�o�v�er�a����6�lo�Gcal��Reld��K�rn�is�the�curv���e�pro�duced�b���y�the�action�of�the�reduction�map�on�a�minimal����6�W��*�eierstrass�UUequation�for��E����:��#���yǍ[�E�Z��:���y��[ٟ����2���d�=��x������3���S�+�8�Ax��+��B��q�]���7�������v!��[���x䍑P~�����E���
�l�:��y��[ٟ����2���=��x������3���S�+���x䍑�q~�����8�A���	��x�8�+���x䍑�S~������B���
O�]�:������6लThe�reduction�on�a�pro��8jectiv���e�p�Goin�t��P�[E�=���[�x��:��y�S��:��z�p��]�is�the�reduction�map�on�the����6�co�Gordinates�h�of�the�p�oin���t�once�it�has�b�een�put�in���to�canonical�form,�mgthat�is,�with�all����6�co�Gordinates�pain�the�ring�of�in���tegers��R��(�and�minimized�so�that�one�or�more�is�a�unit.��&���6�This��is�required�for�the�image�of�the�p�Goin���t�to�b�e�w���ell-dened.�W�Then���x䍑C�~������P���
��is�a�p�oin���t�on������x䍑90�~�����6वE���>��=k�P��.��sP��B�Since�٭the�curv���e's�equation�w�as�c�hosen�to�b�Ge�minimal,���the�map�b�et���w�een�٭curv�es����6�is��Mw���ell-dened,�IJup�to�a�c�hange�of�co�Gordinates�o�v�er�the�residue�eld��k�P��.�	!�It�is�a����6�homomorphism�;�b�Gecause�it�maps�lines�in��P���^��2��|s�(�K���)�to�lines�in��P���^��2���(�k�P��),�@�and�therefore�the����6�group�UUla���ws�are�preserv�ed.����B�The��reduced�curv���e�is�not�guaran�teed�to�b�Ge�non-singular.�VJIf�the�discriminan�t�of�a����6�W��*�eierstrass��'equation�is�0,���then�its�elliptic�curv���e�has�singular�p�Goin�ts;�	6ho�w�ev�er,���it�is�a��&���6�fact�	�that�the�remaining�non-singular�p�Goin���ts�still�form�an�ab�elian�group.�X�Let���x䍑Y�~������E����ns�����'�(�k�P��)���ԍ�6�b�Ge���the�group�of�non-singular�p�oin���ts�of���x䍑��~������E����C�.�$�Then�its�in�v�erse�image�in�the�set��E����(�K���)����6�is��Vdenoted��E����0��|s�(�K���),��the�p�Goin���ts�of�non-singular�reduction.���Also�dene��E����1���(�K���)�to�b�Ge����6�the�*�k���ernel�of�reduction,�3lthat�is,�the�p�Goin���ts�that�map�to�the�iden�tit�y�on�the�reduced����6�curv���e.����6��Prop�Q�osition��T3.6.��The�UUsequence��#�����(0���!��E����1��|s�(�K���)��!��E����0���(�K���)��!���x䍑�~������E����ns����>��(�K��)�;������6लwhere�UUthe�second�map�is�inclusion�and�the�last�map�is�reduction,�is�exact.������6��Pr��}'o�of.���V5�This���follo���ws�from�the�denitions.�wSince�the�map��E����1��|s�(�K���)� ]�!��E����0���(�K���)���is�inclu-����6�sion,�[�as�'6observ���ed�ab�Go�v�e,�[�it�is�injectiv�e.��iThen�the�k�ernel�of�the�reduction�map�is����6�exactly�UU�E����1��|s�(�K���)�b���y�denition.��������� H��6��R��}'emark����3.7�.��ٲIn�.Ffact,�6the�reduction�map�is�surjectiv���e�(see�[�17��
],�c���h.�d�VI�GI,�prop�osition����6�2.1|the�U�pro�Gof�uses�Hensel's�lemma�for�the�lifting),���so�w���e�can�extend�the�ab�o���v�e����6�in���to�y�a�short�exact�sequence�of�groups,���but�the�surjectivit�y�is�not�essen�tial�to�what����6�follo���ws.����B�Finally��*�,��@w���e��wput�together�the�reduction�map�and�the�formal�group�asso�Gciated�to����6�the�UUcurv���e.����6��Prop�Q�osition��
3.8.��5�Let���x䍑�X^�����?W�E���t�o���v�er�?W�R�S�b�Ge�the�formal�group�asso�ciated�to��E���o���v�er�?W�K���,�C�and��j���6�let��d�s�(�t�)�b�Ge�the�p�o���w�er��dseries�as�in�equation�3.1.�>"Then�the�map���x䍑
e^������E���
���(�M��)���!��E����1��|s�(�K���)��dgiv���en����6�b���y���t�=�7�������v!��(�������6Q�t��33��&�fe
�����s�(�t�)�����q��;�������ek�1���۟�&�fe
������s�(�t�)�������)�is�an�isomorphism.�FIn�this�map,���whose�image�is�not�explicitly��
$���6�comp�Gosed�2of�pro��8jectiv���e�p�oin���ts,�i<w�e�2assert�that�the�p�oin���t��z����=�6�0�is�mapp�ed�to�the����6�additiv���e�UUiden�tit�y�(0�;����1�).������6��Pr��}'o�of.���V5�W��*�e�aialready�kno���w�that�the�image�p�Goin�ts�satisfy�the�W��*�eierstrass�equation,����6�and�K�w���e�ha�v�e�remark�ed�that��t���2��M�c	�mak�es�K�the�dening�series�con�v�erge�to�a�p�Goin�t�in����6वE����(�K���)�(it�is�in��E��(�K���)�and�not��E��(�M��)�b�Gecause�the�series�is�Lauren���t).�W]T��*�o�see�that�it�is����6�actually�&�in��E����1��|s�(�K���),�/�notice�that�after�clearing�denominators�of�elemen���ts�of��M�=��in�the����6�p�Goin���t�ʖ[�������6Q�t��33��&�fe
�����s�(�t�)������{�:�������	EF�1������&�fe
������s�(�t�)��������:���1],���the�rst�and�third�co�ordinates�ha���v�e�ʖp�ositiv�e�ʖv��q�aluation�while�����
�x��7
��������6ऱ10����SETH���KLEINERMAN���V������6लthe��}second�is�a�unit,��and�therefore�the�p�Goin���t�reduces�to�[0��:�1�:�0]��}on�the�curv�e���x䍑/~~������E���
�Ѳ(�k�P��).����6�The�UUexplanation�in�[�17��
]�(c���h.�q�VI�GI,��x�2)�of�this�fact�is�not�particularly�clear.����B�It���remains�only�to�pro�Gduce�an�in���v�erse.�*�W��*�e���men�tioned�a�putativ�e�one�in�our����6�original�d+remarks,�g�namely��t��Ӳ=��x=y�[ٲ,�so�d+it�suces�to�sho���w�that�giv�en�(�x;���y�[ٲ)��Ӹ2��E����1��|s�(�K���),����6वx=y����2�H��M��.�[	But��this�is�eviden���t�b�y�insp�Gection:�
Hif�[�x�H��:��y����:�1]��is�to�reduce�to�[0�H�:�1�:�0]����6�as��ab�Go���v�e,��w�e�m�ust�ha�v�e��x��and��y�j��with�negativ�e�v��q�aluation,��and,�further,��v�[ٲ(�y��)���<�v�[ٲ(�x�).����6�Therefore�UU�v�[ٲ(�x=y��)���>��0�and�so��x=y�"�2��M��.����V����Gԍ�BलNo���w�UUw�e�ha�v�e�the�material�needed�to�pro�v�e�the�main�result.���!��6��Theorem���3.9.�|��L��}'et�S��E���b�e�an�el���liptic�curve�over��Q�,��kand��p�"6>��3��a�prime.��lAssume��&���6�that��P�p��do��}'es�not�divide�the�discriminant���,���so�the�r�e�duc�e�d�curve���x䍑�Q�~������E���1��is�non-singular.����6�Then���the�induc��}'e�d���map�on�torsion��E����(�Q�)����(t}\�cmti7�tor��
�A�!���x䍑�~�������E���
�l�(�F����p���R�)����tor����is�inje��}'ctive.����BलThe�4
theorem�is�also�true�for��p�:I�=�3,�k�but�4
false�for��p�:I�=�2.�
�W��*�e�4
will�discuss�this����6�further�UUb�Gelo���w.�q�This�pro�of�w���orks�v�erbatim�for��p���=�3.��Gԍ���6��Pr��}'o�of.���V5�Consider���the�extension�our�curv���e��E����(�Q�)�����E��(�Q����p���R�)���to�a�lo�Gcal�eld�in�whic���h�it�is����6�em���b�Gedded.�dbW��*�e�-%can�no�w�apply�all�the�theory�w�e'v�e�b�Geen�assem�bling,�5.with��K�~4�=���Q����p���R�.����6�W��*�e�UUha���v�e�the�standard��p�-adic�v��q�aluation,�and�the�maximal�ideal�is��M��3�=��(�p�).����B�First,���with���the�aid�of�Prop�Gosition�2.8,�w���e�sho�w�that�there�cannot�b�Ge�an�y�elemen�ts��j���6�of��8order��p��in�the�formal�group���x䍑9^������E���
���(�M��).�DAssume�an�elemen���t��x���2��M��S�has��8order��p�.�Then��
����6�b���y��Prop�Gosition�2.8,�JHw�e�ha�v�e�0�c�<�v�[ٲ(�x�)�������&h���7�v�@L�(�p�)���7��ʉfe�������!�p��1�����i�.�	$Since���v�A�is�the�standard��p�-adic��
O<��6�v��q�aluation��
on�the�rational�n���um�b�Gers,�
�v�[ٲ(�p�)��=�1��
and��v��(�x�)�is�an�in���teger.�TSince��p��>��2,�
�w�e����6�ha���v�e�UU0���<�v�[ٲ(�x�)���������	)��1���K��&�fe[Ɵ����p��1�����P\�<��1,�a�con���tradiction.�����B�Next,���recall�|�that�there�cannot�b�Ge�an���y�elemen�ts�of�order�relativ�ely�prime�to��p��in������x䍑90�^�����6वE���>���(�M��),��ib���y���Prop�Gosition�2.7.�_�Hence�w�e�conclude�that�there�are�no�torsion�p�Goin�ts�of���㍑6�an���y�UUkind�in�the�group���x䍑�V^������E���J��(�M��).����B�By���Prop�Gosition�3.8,���x䍑!M^������L�E���Ơ�(�M��)����T͍��la����+3���la�=�������E����1��|s�(�Q����p���R�)�and�th���us�there�is�no�torsion�in�the�group����6वE����1��|s�(�Q����p���R�).����B�Finally��*�,��]w���e���a�v��q�ail�ourselv�es�of�the�exact�sequence�of�Prop�Gosition�3.6,��]whic�h�tells����6�us��that�the�group��E����1��|s�(�Q����p���R�)�is�injectiv���ely�mapp�Ged�in�to��E����0��|s�(�Q����p���R�),�%�th�us�has�a�torsion-����6�free�
�image�under�this�map,�<and�this�image�is�exactly�the�k���ernel�of�the�reduction��&���6�map��Fin���to���x䍑G~������E����ns����)��(�F����p���R�).���Since�the�reduced�curv�e�is�non-singular,���x䍑�~�����ɂ�E����ֲ(�F����p���R�)�a�=���x䍑��~������E����ns�����x�(�F����p���)��Fand����6वE����0��|s�(�Q����p���R�)��=��E����(�Q����p���).�q�Th���us�UUw�e�ha�v�e�an�exact�sequence��n����0���!���x䍑�^������E���
�l�(�M��)��!��E����(�Q����p���R�)��!���x䍑�~������E����(�F����p���R�)�:�����6लIn��particular,�*Yall�the�p�Goin���ts�in��E����(�Q����p���R�)�that�reduce�to�the�iden�tit�y�in��E����(�F����p���R�)�are�non-����6�torsion.� So���the�torsion�in��E����(�Q����p���R�)�and�hence�in��E��(�Q�)�m���ust�b�Ge�mapp�ed�injectiv���ely��
D��6�in���to���x䍑�V~�����UU�E���J��(�F����p���R�).�����������Gԍ�6��Corollary��3.10.����The�B�order�of�a�torsion�subgroup�of�an�elliptic�curv���e�o�v�er��Q��is����6�nite.���!����6��Pr��}'o�of.���V5�There�v1exist�innitely�man���y�primes�of�go�Go�d�v1reduction,�~hsince�an�y�prime�that����6�do�Ges���not�divide�the�discriminan���t�is�a�prime�of�go�o�d�reduction.���Cho�ose��p�`�>��3���and����6�reduce.�q�The�UUreduction�map�is�injectiv���e�on�torsion.����X������6��R��}'emark����3.11�.�嶲A�8similar�8argumen���t�w�orks�on�elliptic�curv�es�o�v�er�nite�extensions�of��
_��6��Q�.���It��?ma���y�b�Gecome�necessary�to�c�ho�Gose�a�higher�prime�so�that�����&h���r�v�@L�(�p�)���r��ʉfe�������!�p��1�����":�<�f�1,��9since�the��������7
�������@v��ON���THE�TORSION�POINTS�OF�ELLIPTIC�CUR��#�VES�&�MODULAR�ABELIAN�V�ARIETIES���11����V������6लv��q�aluation���of��p��ma���y�not�b�Ge�1�in�the�corresp�onding�extension�of��Q����p���R�,��but�once�a�prime����6�has�UUb�Geen�found�that�w���orks,�the�torsion�subgroup�is�automatically�nite.��(ȍ�B�W��*�e�8�should�observ���e�that�if�w�e�assume�the�preliminary�results�for�the�v�e-parameter����6�W��*�eierstrass��equations�(not�a�big�leap),�.WTheorem�3.9�will�also�hold�for��p��o�=�3��but����6�not�UU�p���=�2,�since��v�[ٲ(2)�=�1�do�Ges�not�satisfy�the�condition��v�[ٲ(�p�)��<�p�8���1.����B�The�`�elliptic�curv���e��E���:���y��[ٟ�^��2��Ð�+��D�xy�G�+��y���=��x���^��3��g��+��x���^��2�����10�x����10�`�is�an�example�of�a����6�curv���e��ywith�go�Go�d��yreduction�at��p�0S�=�2��y(since��0S=�3���^��4��|s�5���^��4���)��yfor�whic�h�the�corresp�Gonding����6�reduction��
map�is�not�injectiv���e�on�torsion.�#�Chec�king�the�torsion�b�y�computer,�
7w�e����6�nd��&that�it�is�equal�to��Z�=�8�Z�;��ho���w�ev�er,��0there��&are�only�5�p�Goin���ts�that�could�p�ossibly�b�e����6�in�UU�E����(�F����2��|s�).�q�This�is�not�a�con���tradiction,�ho�w�ev�er,�b�Gecause�of�the�ab�o���v�e�UUobserv��q�ation.����B�T��*�o�C�b�Ge�explicit,�GPw���e�c�hec�k�the�torsion�o�v�er��F����2���A�as�w�ell,�GPusing�the�computer�algebra����6�pac���k��q�age�UUMA�GMA:��=H��6��*��<x

cmtt10�>�?�E�:=�EllipticCurve([�1,�1,�1,�-10,�-10�]);����6�>�?�F�:=�ChangeRing(E,GF(2));����6�>�?�AbelianGroup(F);����6�Abelian�?�Group�isomorphic�to�Z/4����6�Defined�?�on�1�generator����6�Relations:����Q �4*$.1�?�=�0����BलA�c�brief�c�aside:��4ho���w�did�w�e�kno�w�that�there�could�b�Ge�at�maxim�um�v�e�p�Goin�ts�on����6वE����(�F����2��|s�)?�>kIt's���not�hard�for��F����2��kT�b�Gecause�there�are�four�p�oin���ts�in�(�F����2��|s�)���^��2��kT�together�with����6�the��p�Goin���t�at�innit�y��*�.�8�But�for�larger�primes,��w�e�can�use�the�Hasse�b�Gound,��pro�v�ed����6�in���the�1930s;�Ʊw���e�state�the�result�in�a�sligh�tly�w�eak�ened�form�here.�TThe�full�result����6�includes�UUall�nite�elds.��(ȍ�6��Theorem��T3.12��(Hasse)�.��#�E����(�F����p���R�)�����p�8�+�2�������p���UW�����fe�;��p�����7�+�1�:����BलNo���w��apply�this�to�the�prime��p�h;�=�3:�3%it��sa�ys�that�there�at�most�sev�en�p�Goin�ts�on����6�an��elliptic�curv���e�o�v�er��F����3��|s�.�_�Ho�w�ev�er,�)�w�e'v�e�seen�more�torsion�than�that�in�an�elliptic����6�curv���e���o�v�er��Q�;�just�tak�e�the�example�ab�Go�v�e.�R�Ho�w�ev�er,�
�the�reason�w�e�can't�apply�the����6�theorem��is�that�the�curv���e�do�Gesn't�ha�v�e�go�Go�d��reduction�at��p���=�3.�^>This��giv�es�us�an��a����6�priori�w�b�Gound�son�torsion.���If�the�discriminan���t�is�not�divisible�b�y�3,�zythen�the�torsion����6�subgroup�UUhas�sev���en�p�Goin�ts�or�few�er.��2M���o��4.��|��St��UTa�tements��of�Tw��o�Theorems�About�Torsion����BलIn�B�1977-8,�y{B.�Mazur�published�a�pap�Ger�[�10��
]�that�completely�answ���ered�the�question����6�of�p�what�torsion�subgroups�are�p�Gossible�for�an�elliptic�curv���e�o�v�er��Q�.�%�W��*�e�cite�his�result����6�here.��(ȍ�6��Theorem��M4.1��(Mazur)�.�T��L��}'et�䯵E���=�Q��b�e�an�el���liptic�curve.�_.Its�torsion�sub�gr�oup��E����(�Q�)����tor�����6��must��8b��}'e�one�of�the�fol���lowing�fte�en�gr�oups:���Z�=n�Z��for��n���2�f�1�;����2�;��3�;��4�;��5�;��6�;��7�;��8�;��9�;��10�;��12�g�,����6��Z�=�2�Z�['���Z�=n�Z�-��for��n���2�f�2�;����4�;��6�;��8�g�.�w�F��;�urther,�BOal���l�-�of�these�gr��}'oups�ar�e�r�epr�esente�d�as�the����6�torsion���gr��}'oups�of�some�el���liptic�curve�over��Q�.����BलIn��~1996,��L.�Merel�pro���v�ed��~a�theorem�that�resolv���ed�a�long-standing�question�ab�Gout����6�the��ytorsion�groups�of�elliptic�curv���es�o�v�er�n�um�b�Ger�elds��K���.��4The�pro�of�of�it�is�also����6�w���ell�UUb�Gey�ond�the�scop�Ge�of�this�pap�er�(see�[�11��
]).����6��Theorem��M4.2��(Merel)�.�e��L��}'et�/�E���=K����b�e�an�el���liptic�curve,�V�wher�e�the�numb�er�eld��K��
^r��6��satises���[�K�~4�:���Q�]�=��d�>��1�.���Then�any�torsion�p��}'oint�has�or�der��n��<�d���^��3�d����r���Zcmr5�2����E�.������%��7
��������6ऱ12����SETH���KLEINERMAN���V������BलIn�M|particular,�OMerel�uses�this�theorem�along�with�previous�w���ork�of�F��*�altings�and����6�F��*�rey��Kto�demonstrate�that�there�is�a�nite�n���um�b�Ger��Kof�p�ossible�torsion�groups�of����6�elliptic�-Scurv���es�o�v�er�a�degree��d��extension�of��Q�.���This�result�is�stunningly�general.����6�Ho���w�ev�er,�W�although�WQone�kno���ws�that�the�list�is�nite,�and�for�a�giv���en��d��one�can�tak�e����6�man���y��Velliptic�curv�es,��compute�their�torsion�groups,�and�compile�a�list�of�the�torsion����6�that�l�has�app�Geared,�rvit�is�not�kno���wn�ho�w�to�pro�v�e�that�an�y�suc�h�list�w�ould�actually����6�b�Ge�UUcomplete.�q�Mazur's�metho�d�do�es�not�w���ork�on�n�um�b�Ger�elds.��fd���Lt�5.��Y;��Torsion��Points�on�Elliptic�Cur��rves:���Algorithmic�Questions����6ल5.1.��The��Nagell-Lutz�Theorem.��No���w��Lsince��E����(�K���)����tor��
s��is�nite,���the�natural�follo�w-����6�up�miquestion�is:���is�it�easily�computable?�$yF��*�or�elliptic�curv���es,���the�answ�er�is�w�ell�kno�wn.����6�Tw���o�(pap�Gers�published�in�the�1930s,�\�one�b�y�E.�Lutz�and�the�other�b�y�T.�Nagell,����6�giv���e�da�clear�idea�of�what�torsion�p�Goin�ts�are�p�Gossible.���Their�conclusion�is�presen�ted����6�in�UUthe�follo���wing�theorem:��g9��6��Theorem�=5.1.�#��L��}'et���E�^��b�e�an�el���liptic�curve�over��Q��with�Weierstr�ass�e�quation��y��[ٟ�^��2��B�=����6वx���^��3��U��+��"�Ax��+��B��q�,�p�with�gܵA�,��B��M�inte��}'gers.���Then�the�c�o�or�dinates�of�a�non-zer�o�torsion�p�oint����6वP�Ԡ�=�q(�x����0��|s�;���y����0���)��2��E����(�Q�)��ar��}'e�in��Z�.�[F��;�urthermor�e,��յP���is�either�of�or�der��2�,���or�else��y���^���[ٱ2��l�0������6��divides���4�A���^��3���S�+�8�27�B���q��^��2�����.����BलThis��c���haracterization�of�torsion�p�Goin�ts�on�elliptic�curv�es�o�v�er��Q��limits�the�searc�h����6�space.�A[Practically��*�,���p�Goin���ts���of�order�2�are�easy�to�nd,�b�Gecause�in�that�case��y����0���ݲ=�:j0;����6�and��the�second�condition�ensures�that�if��P�D��is�of�higher�order�then�there�are�only����6�nitely�v�man���y�p�Gossibilities�to�c�hec�k�for�the�v��q�alue�of��y����0��|s�.�ֹTherefore�there�are�nitely����6�man���y��:pairs�(�x����0��|s�;���y����0���)��:to�c�hec�k,��ssince�eac�h�c�hoice�of��y����0��	"��pro�Gduces�a�monic�in�tegral����6�cubic�<�equation�for��x����0��|s�,�A�an���y�in�tegral�solution�of�whic�h�will�divide�the�constan�t�term,����6वB�.����)�y���^���[ٱ2��l�0����L�.�Z�W��*�e��should�remark�that�this�is�y���et�another�pro�Gof�of�niteness�of�the�torsion����6�subgroup,�UUindep�Genden���t�of�either�of�the�t�w�o�metho�Gds�previously�men�tioned.����B�The�UUrst�step�is�to�sho���w�that�the�co�Gordinates�of�a�torsion�p�oin���t�are�in�tegers.������6��Pr��}'o�of.���V5�Let��3�m��b�Ge�the�smallest�in���teger�for�whic�h��mP�*��=���O�G�.�AgIf��m��=�2,��:then��P�*��=���P��=����6�(�x����0��|s�;�����y����0���).��So����y����0����=�&�0�and�hence��x����0���2�&��Z�,���since�it�is�the�ro�Got�of�a�monic�p�olynomial����6�with�UUin���tegral�co�Gecien�ts.����B�No���w��iassume��m��>��2.�F�Our��igoal�will�b�Ge�to�sho�w�that�for�ev�ery�em�b�Gedding��E����(�Q�)���,��UX�!����6वE����(�Q����p���R�),��gthe��0co�Gordinates�of�our�p�oin���t�ha�v�e�nonnegativ�e�v��q�aluation�in�the�lo�Gcal�eld.����6�Let�UU�M�lp�b�Ge�the�maximal�ideal�(�p�)�to�a���v�oid�UUnotation�confusion.����B�First,�:�if�4,�v�[ٲ(�x����0��|s�)�����0,�then�4,the�W��*�eierstrass�equation�tells�us�that��v�[ٲ(�y����0��|s�)�����0�4,as�w���ell.����6�So�UUw���e'll�assume��v�[ٲ(�x����0��|s�)���<��0�UUand�nd�a�con�tradiction.����B�W��*�e��Wobserv���e�that�the�curv�e�has�a�minimal�mo�Gdel,�̊b�y�Remark�3.3,�̊so�the�reduction����6�map�>is�w���ell�dened.�,By�applying�the�v��q�aluation�to�b�Goth�sides�of�the�W��*�eierstrass����6�equation�p�again,�wٵv�[ٲ(�y����0��|s�)���<�v��(�x����0��|s�)��<��0,�w�and�p�so�the�pro��8jectiv���e�p�Goin�t�[�x����0��q��:���y����0���:�1]�p�reduces���ԍ�6�to���the�p�Goin���t�[0�nA:�1�:�0]���under�the�map��E����(�Q����p���R�)�nA�!���x䍑�B�~������E���c��(�F����p���)���(w���e'v�e�used�this�reasoning��
US��6�b�Gefore,���in���the�pro�of�of�Prop�osition�3.8).�3�So�(�x����0��|s�;���y����0���)���is�a�mem���b�er�of��E����1��|s�(�Q����p���R�)����T͍�������+3�����=�������x䍑�O^�����
UN�E���J��(�M��).����B�W��*�e��can�no���w�apply�our�usual�prop�Gositions�ab�out�torsion�in�formal�groups.�8�Prop�o-����6�sition��2.7�tells�us�that�if��m��is�relativ���ely�prime�to��p��w�e�already�ha�v�e�our�con�tradiction����6�(there��#can't�b�Ge�suc���h�a�torsion�p�oin���t),��Vand�Prop�osition�2.8�tells�us�that�if��m��ò=��p���^��n�����6लis�o-a�p�Go���w�er�o-of��p�,�u�then�0��+�<�v�[ٲ(�x=y��)��+���������˱1��%^��&�fe!L�����p����O
�\cmmi5�n���l��p����n���0ncmsy5��1������&wݲ.��PBut�o-for��m�>��2�this�righ���t�hand�side�is�����6�less�UUthan�1,�so�w���e�ha�v�e�our�con�tradiction.����B�Because���the�co�Gordinates,���when�though���t�of�as�elemen�ts�of��Q����p���R�,���are�in�the�ring�of����6�in���tegers�UU�Z����p���for�ev�ery��p�,�they�m�ust�b�Ge�in��Z�����Q�.�����
�M��7
�������@v��ON���THE�TORSION�POINTS�OF�ELLIPTIC�CUR��#�VES�&�MODULAR�ABELIAN�V�ARIETIES���13����V��������������G��BलIt���remains�only�to�v���erify�that��y���^���[ٱ2��l�0�����divides�4�A���^��3���1�+�q�27�B���q��^��2�����when�the�torsion�p�Goin�t�is����6�not�a�of�order�2.���This�is�essen���tially�a�computation�using�the�duplication�form�ula�for����6�a�UUp�Goin���t�under�the�formal�addition�la�w�on�the�elliptic�curv�e,�follo�wing�[�17��
].������6��Pr��}'o�of.���V5�Assume�xe2�P���6�=��1�O�G�.���W��*�rite�2�P��=��1(�x����1��|s�;���y����1���).���Since�xeit�is�also�a�torsion�p�Goin���t,����6वx����1��|s�;���y����1���۸2�&h�Z�.�YW��*�e���can�use�the�duplication�form���ula�for�a�p�Goin�t�to�express��x����1��
��in�terms����6�of�UU�x����0��|s�:������y�x����1��C��=�����<$���K�x���^���4��l�0����S��8�2�Ax���^���2��l�0������8�B��qx����0���+��A���^��2����K�w�feh���	(֍���4(�x����:��3��l�0����S�+�8�Ax����0���+��B��q�)�����nu�:��9썑6लNotice�n|that�the�denominator�is�just��y���^���[ٱ2��l�0����L�.��=No���w�w�e�pro�Gduce�a�p�olynomial�iden���tit�y�n|b�y����6�using��cthe�Euclidean�algorithm�on�the�n���umerator�and�denominator�of�the�fraction����6�ab�Go���v�e.�+�Since��Wthe�fraction�is�reduced,���w���e'll�get�some�linear�com�bination�of�the�t�w�o����6�p�Golynomials�UUequal�to�a�constan���t�term.���������������tN(3�X��������2��~5�+�8�4�A�)(�X��������4�����2�AX��������2�����8�B��qX�²+��A������2��|s�)��������ڸ������u�����������|���(3�X��������3��~5��8�5�AX�¸��27�B��q�)(�X��������3���+��AX�²+��B��q�)������� U�=��4�A������3���S�+�8�27�B���q�����2����:������`��6लObserv���e���that�the�constan�t�term�w�orks�out�to�exactly�the�v��q�alue�w�e�had�ab�Go�v�e.�}�So����6�all��|w���e�ha�v�e�to�do�is�plug��X����=���x�(�P�c��)�in�to�this�p�Golynomial�iden�tit�y��*�,��tand�simplify�using����6�the�UUrelations�for��x����1���Ȳand��y���^���[ٱ2��l�0����L�:���A������������&I(3�x�������2����0����S�+�8�4�A�)(�x����1�����4�y�������[ٱ2����0����L�)���������͸������������������(3�x�������3����0����S��8�5�Ax����0�����27�B��q�)(�y�������[ٱ2����0����L�)���������=��4�A������3���S�+�8�27�B���q�����2����:������r��6लAll���of�the�v��q�ariables�in�this�equation�are�in���tegers.�U0No�w���it�is�clear�that��y���^���[ٱ2��l�0����ܲdivides�the����6�left�UUhand�side;�therefore�it�m���ust�divide�the�righ�t.���������G��6ल5.2.��Doud's�@�Algorithm.��The��hnaiv���e�approac�h�to�nding�the�torsion�on�elliptic����6�curv���es��o�v�er��Q�,��wthen,�w���ould�b�Ge�to�c�hec�k�eac�h�of�the�nitely�man�y�pairs�of�co�Gordinates����6�(�x;���y�[ٲ)�Vsuggested�b���y�the�Nagell-Lutz�theorem,�.�kno�wing�b�y�Mazur's�Theorem�(4.1)����6�that�]3a�torsion�p�Goin���t�cannot�ha�v�e�order�greater�than�12.��aIn�fact,�_*this�approac�h�w�as����6�widely��implemen���ted�un�til�fairly�recen�tly��*�.��J.W.�Cremona's�1997�b�Go�ok��[�5��]�of�elliptic����6�curv���e���algorithms�ga�v�e�it�as�the�standard�recip�Ge�for�the�computation�of�torsion.����6�Ho���w�ev�er,�g:Darrin�0sDoud�[�6��]�p�Goin���ts�out�t�w�o�disadv��q�an�tages�to�this�metho�Gd:�(rst,�g:it����6�requires��5factorization�of�the�discriminan���t,��mwhic�h��5ma�y�b�Ge�large;�*�second,��mfor�eac�h����6�square��divisor�of�the�discriminan���t,�&^a�cubic�equation�m�ust�b�Ge�solv�ed,�&^and�there�ma�y����6�b�Ge��"man���y�suc�h�divisors.�3a(Remem�b�Ger�that�the�n�um�b�Ger�of�divisors�gro�ws�exp�Gonen�tially����6�in�UUthe�n���um�b�Ger�UUof�prime�factors�of�the�discriminan���t.)����B�In���its�place,�ODoud�suggests�using�the�W��*�eierstrass��}��function�asso�Gciated�to�a����6�complex�_lattice���Ҹ2��C�.���Giv���en�_the�lattice,�?"w�e�asso�Gciate�to�it�the�doubly�p�erio�dic����6�function���ҍ��9�}�(�z�p��)��=�����<$��D1���K�w�fe	���	(֍�z����r�2��������+�����������u

cmex10�X�����8��2��;�6�=0�����%�ğ��^�������<$��=���1��.5k�w�fe$��	(֍(�z��w��8��	z�)���r�2������U�(������<$��a۲1��l�w�fe
됟	(֍���	z��r�2��������֟��^����u�8�:��r�6लIt��is�a�standard�result�(see�[�17��
],���c���hapter�VI,��x�3-5)�that�for�the�curv�e�asso�Gciated�to����6�the�lattice,�*�there�is�an�isomorphism�of�groups���㮲:��C�=���!��E����(�C�)�giv���en�b�y���(�z�p��)��=����6�[�}�(�z�p��)�;���}���^��0���9�(�z��)�;��1].��(Here�8�the�asso�Gciation�of�curv���e�and�lattice�means�that�the�curv�e��E��z�has����6�W��*�eierstrass��co�Gecien���ts�that�come�from�the�Lauren�t�series�expansion�of�the�function����6व}�,�")whic���h�]dep�Gends�on�.�\uThe�precise�algebraic�w�a�y�that�the�co�Gecien�ts�deriv�e�from����6�the�C0Lauren���t�expansion�is�not�essen�tial�to�these�remarks.)�k�It�is�also�kno�wn�that�the������à�7
��������6ऱ14����SETH���KLEINERMAN���V������6लrev���erse��Qasso�Gciation�can�b�e�made:�ٿgiv���en�an�elliptic�curv�e��E����(�C�),��Pthere�is�a�lattice�����6�suc���h���that�the�Lauren�t�expansion�of��}�(�z�p��)�pro�Gduces�the�W��*�eierstrass�co�ecien���ts�for����6वE����.����B�This�H�suggests�a�natural�w���a�y�H�of�c���hec�king�H�for�rational�torsion�in��E����,�~�namely��*�,�consider����6�the�Btorsion�p�Goin���ts�in��C�=��and�c�hec�k�to�see�whether�their�images�are�rational�under����6�the��3map���.�UaSince�the��n�-torsion�p�Goin���ts�of��C�=��are�just�������N�1���f��&�fe�~�����n�����
���=�,��*there�isn't�m�uc�h�to����6�compute.�M�The�IJessen���tial�steps�are�therefore�pro�Gducing�the�lattice�from�the�curv�e����6�in�h�the�rst�place,��hwhic���h�is�quic�k�relativ�e�to�the�rest�of�the�algorithm,��hand�then����6�follo���wing���the��n�-torsion�from�the�lattice�in�to�the�curv�e,���c�hec�king�to�see�whether����6�they're���in���tegral�p�Goin�ts.�FEDoud�uses�reduction�(to�determine�a�m�ultiple�of�the�torsion����6�subgroup),��in���conjunction�with�Mazur's�Theorem,�to�determine�whic���h�v��q�alues�of��n����6लneed�UUto�b�Ge�observ���ed.����B�So��	Doud's�metho�Gd�is�rather�quic���k.�RYThe�only�p�ossible�implemen���tational�pitfall�to����6�the�R�algorithm�is�its�analyticit���y;�S�one�needs�to�tak�e�care�that�one�uses�enough�terms����6�of��Nthe��}��function.���Doud�pro���v�ed��Nthe�required�minimal�precision�for�the�algorithm����6�in�Īhis�pap�Ger,��so�the�theory�is�ne.���Ho���w�ev�er,�if�Īthe�error�b�Gounds�are�not�carefully����6�implemen���ted,���one���migh�t�falsely�truncate�the�series�for��}�(�z�p��)�to�Go�early�and�pro�duce����6�a���p�Goin���t�on�the�curv�e�that's�\close"�to�an�in�tegral�p�Goin�t�but�not�the�one�it's�actually����6�con���v�erging�ȟto.�˥In�that�case�the�algorithm�w���ould�return�less�torsion�than�actually����6�presen���t���in�the�elliptic�curv�e.�0+In�fact,���this�mistak�e�w�as�made�in�the�original�MA�GMA����6�implemen���tation��2of�Doud's�metho�Gd.�;gIn�2001,���Da�vid�Y��*�eung�observ�ed�that�MA�GMA's����6�implemen���tation��of�the�algorithm�w�as�returning��Z�=�8�Z��for�the�torsion�subgroup�of����6�a�curv���e�he�tested,�/�while�P��*�ARI��pro�Gduced��Z�=�8�Z��d���Z�=�2�Z�.�~After�a�bit�of�confusion,����6�the���precision�of�the�analytic�metho�Gd�w���as�increased,���Doud's�algorithm�pro�duced����6�the�|�righ���t�answ�er,���and�there�ha�v�e�b�Geen�no�kno�wn�further�problems.��The�algorithm����6�represen���ts�UUa�signican�t�sp�Geed�increase�o�v�er�the�Nagell-Lutz�approac�h.��&%
���t�-6.����K�Ba��ck�gr�ound��on�Modular�Abelian�V���Uarieties����6��Denition��V6.1.��Z�An�C��algebr��}'aic���gr�oup�劵V�|ֲis�an�algebraic�v��q�ariet���y�that�is�also�a�group,����6�in���that�there�are�t���w�o���regular�maps,���an�in���v�erse���map��i�L��:��V��ڸ!��V�ތ�and���a�comp�Gosition����6�map��~�m��\�:��V��߸����V��@�!��V�8�,��Isatisfying�the�usual�group�axioms�on�the�p�Goin���ts�of��V��.�CAn����6��ab��}'elian���variety�7ҵA�UU�is�a�pro��8jectiv���e,�irreducible�algebraic�group.������6��R��}'emark��T�6.2�.��?�In�B�order�to�sho���w�that�the�algebraic�v��q�ariet�y�with�the�ab�Go�v�e�prop�Gerties����6�is�*ab�Gelian,�Q�one�needs�to�use�the�fact�that��A��is�pro��8jectiv���e.��EThe�pro�of�is�found�in����6�Shafarevic���h�UU[�15��
],�4.3.����6�6.1.��Mo�Q�dular���Groups�&�Mo�dularit��9y��
�.��Go�Go�d�Mreferences�for�the�follo���wing�deni-����6�tions�UUinclude�[�13��
],�[�14��],�and�[�16��].����B�Denote���the�op�Gen�upp�er�half�of�the�complex�plane�b���y��+�%n�

eufm10�h�����C�.�3RIf���w�e�also�include�the����6�p�Goin���ts��in��Q��and�a�p�oin���t��f1g��at�innit�y��*�,��w�e'll�call�it��h���^������.�V�(There's�a�sligh�tly�p�Geculiar����6�top�Gology�traround�the�added�p�oin���ts�(kno�wn�as�\cusps"),�|:as�there�w�ould�ha�v�e�to�b�Ge:����6�see�B�[�16��
],�F[�x�1.3,�for�a�reference�on�this,�the�P���oincar���Ge�metric.)�k�The�group��SL���؟��2���K�(�R�)�acts����6�on���h���^����&�b���y�linear�fractional�transformations,��0as�follo�ws:��&giv�en�an�elemen�t��g��j�=���%����b���������	e��a���՝b���\q���	�Xc�����d�����[email protected]���b�������6लwith�UU�ad�8���bc���=�1,�and��z�7��2��h���^������,�let������P�g�[�z�7��=�����<$���K�az��w�+�8�b���K�w�fe�֟	(֍��cz��w�+�8�d����� T:������W��7
�������@v��ON���THE�TORSION�POINTS�OF�ELLIPTIC�CUR��#�VES�&�MODULAR�ABELIAN�V�ARIETIES���15����V������6लIt�UUis�not�to�Go�hard�to�c���hec�k�UUthat��h���^�����9�is�stable�under�the�action�of��SL���#����2����(�R�),�since��39����w��Im����h(�g�[�z�p��)��=���Im�����K���^�������<$����(�az��w�+�8�b�)(�c����ɲ���z���O��+��d�)����w�feEJ[�	(֍�3��j�cz��w�+�8�d�j���r�2�������a����^����k?�=�����<$���K(�ad�8���bc�)��Im���
��(�z��)���K�w�feA��	(֍�Y��j�cz��w�+�8�d�j���r�2������F�h�;�����6लwhic���h�s�is�p�Gositiv�e�if�the�original�imaginary�part�w�as�p�Gositiv�e.�&�Notice�that�the�elemen�t������6ट��b�����ꪍ��= ���1���Pc�0����.���C`�0���J#���1�����V���b����]��is��zthe�only�non-iden���tit�y��zelemen�t�that�acts�trivially�on�all�of��h���^������.�S)So�in�what��.��6�follo���ws,���w�e���tak�e�care�to�mo�Gd�out�all�the�groups�b�y���1.��Th�us��PSL���R8���2��Ϋ�(�R�)�is�a�group����6�that�UUacts�faithfully�on��h���^������.����B�No���w�ڧconsider�the�discrete�subgroup���=��PSL���c����2����(�Z�)�ڧof�matrices�with�co�Gecien�ts�in����6��Z�,��&again���mo�Gdulo���1.�*#It�is�called��the��pmo��}'dular�gr�oup�,��&and���inherits�an�action�on��h���^������.����6�F��*�or�Fan���y�in�teger��N��,�I*w�e�can�pro�Gduce�three�subgroups�of�the�mo�dular�group��whic���h����6�also�UUact�on��h���^������:���5�����������Y������0��|s�(�N��)�������xz=�������^�����
G���^�����d������a���!a�b������8c��� �d�����&!/���^����0D��2����PSL���c����2����(�Z�)��UUs.t.�����T���^������d���1ȵa���-��b�������sc���-z�d�����2�j���^����<����������^������d���
#�����#�������
#��0���#�������#����^�����1*��mo�Gd��G���N�����^�������������������Y�Ҳ����1��|s�(�N��)�������xz=�������^�����
G���^�����d������a���!a�b������8c��� �d�����&!/���^����0D��2����PSL���c����2����(�Z�)��UUs.t.�����T���^������d���1ȵa���-��b�������sc���-z�d�����2�j���^����<����������^������d���
#��1���#��������
#��0���#�1�����#����^�����1*��mo�Gd��G���N�����^�������������������^1E�(�N��)�������xz=�������^�����
G���^�����d������a���!a�b������8c��� �d�����&!/���^����0D��2����PSL���c����2����(�Z�)��UUs.t.�����T���^������d���1ȵa���-��b�������sc���-z�d�����2�j���^����<����������^������d���
#��1���#�0������
#�0���#�1�����#����^�����1*��mo�Gd��G���N�����^��������3:��6ल(Here�גthe���s�can�b�Ge�an���ything.)�G�No�w�גif�w���e�quotien�t��h���^����pv�b�y�the�action�of�these�groups,����6�w���e�UUget�Riemann�surfaces�([�16��
],��x�1.5).��
���6��Denition�a16.3.���F��*�or��Wan���y�p�Gositiv�e�in�teger��N��,��the�Riemann�surfaces��X����0��|s�(�N��),���X����1���(�N��),����6�and�UU�X���(�N��)�are�the�spaces�����0��|s�(�N��)�n�h���^������,�����1��|s�(�N��)�n�h���^������,�(�N��)�n�h���^�����9�resp�Gectiv���ely��*�.����6��R��}'emark�~�6.4�.����The�?�quotien���t�is�traditionally�written�on�the�left�b�Gecause�the�group����6�action��is�written�on�the�left.�W�Also,�Mb�Gecause�w���e�ha�v�e�the�inclusion�of�groups�(�N��)�������6ल����1��|s�(�N��)���������0���(�N��),�1the��Riemann�surfaces�ha���v�e��corresp�Gonding�relationships,�in�rev���erse����6�order:�q�X���(�N��)��*is�the�\biggest,"�and�the�other�t���w�o��*are�progressiv���ely�quotien�ts�of�it.����6�Finally��*�,��these��are�algebraic�curv���es�o�v�er��Q��(this�is�not�ob�vious,��and�is�pro�v�ed�in�[�16��
],����6सx�6.7).�J�F��*�or���example,��m[�20��
]�notes�that��X����1��|s�(13)�is�the�desingularization�of�the�pro��8jectiv���e����6�closure�UUof�the�ane�curv���e��y��[ٟ�^��2���d�=���x���^��6���S�+�8�2�x���^��5���+��x���^��4���+�2�x���^��3���+�6�x���^��2���+�4�x��+�1.����B�Giv���en��ka�compact�Riemann�surface��X��M�of�gen�us��g�[ٲ,��1it�has�a�Jacobian��J��9�(�X���)�asso-����6�ciated�M
to�it,���whic���h�is�a�complex�torus��C���^��g����=��for�a�certain�lattice�;���w�e�can�also����6�think���ab�Gout�it�as�the�space�of�divisors�of�degree�zero�on��X����up�to�linear�equiv��q�alence.����6�(Recall�wthat�a��divisor����on��X�?��is�a�formal�sum�of�p�Goin���ts��p��in��X��with�in���teger�co�Gecien�ts,��36���=��D�5�=����ܶ����X����7����p�2�X���ŵn����p���R�p;���n���2��Z�;��t���6लits�UU�de��}'gr�e�e���is�the�sum�of�those�co�Gecien���ts,�������Ģdeg���ő(�D�G�)��=����ܶ����X����7���p�2�X���ŵn����p���R�;����6लand�Վan���y�meromorphic�function��f�[�:�Gt�X�V�!��C�Վ�has�a�naturally�asso�Gciated�divisor,����6�namely��
qˍ���=(�f���)��=����ܶ����X����7���p�2�X���d�(�ord���x���p��8�(�f��))�p:���A��6लThen�͵J��9�(�X���)�is�the�space�of�divisors�mo�Gdulo�the�divisors�of�meromorphic�functions.)����6�It�#is�not�terribly�ob���vious�that�the�space�of�divisor�classes�of�degree�0�has�the�struc-����6�ture���of�a�complex�torus;�Lthe�result�is�due�to�Ab�Gel�and�Jacobi.�_MF��*�or�details�of�the������C��7
��������6ऱ16����SETH���KLEINERMAN���V������6लisomorphism��at�a�more�or�less�elemen���tary�lev�el,��Qincluding�what�the�lattice��is����6�explicitly��*�,�UUsee�the�author's�junior�pap�Ger�[�9��].����B�W��*�e�+.ha���v�e�dened�three�compact�Riemann�surfaces�ab�Go�v�e,�3�and�brie
y�touc�hed�on����6�the�UUnotion�of�the�Jacobian.�q�The�next�denition�should�therefore�b�Ge�in���tuitiv�e.��33��6��Denition�0�6.5.�/��Let���J����0��|s�(�N��)�and��J����1���(�N��)�b�Ge�the�Jacobians�of�the�Riemann�surfaces����6वX����0��|s�(�N��)�UUand��X����1���(�N��),�resp�Gectiv���ely��*�.����B�And���no���w�w�e�can�see,���at�least�et�ymologically��*�,���where�the�follo�wing�denition�comes����6�from:����6��Denition�
46.6.����An�d�ab�Gelian�v��q�ariet���y��A��is��mo��}'dular�x�if�it�is�the�quotien�t�of��J����1��|s�(�N��)����6�for���some�p�Gositiv���e�in�teger��N��,���that�is,�if�there�is�a�surjectiv���e�map��J����1��|s�(�N��)�7[���A�.�;�The����6�smallest�UUin���teger��N�lp�for�whic�h�this�holds�is�referred�to�as�the��level�]޲of�the�v��q�ariet�y��*�.����6��R��}'emark�|�6.7�.�2.�Ev���ery�S�elliptic�curv�e�is�mo�Gdular.�
l�This�result�w�as�conjectured�b�y����6�T��*�aniy���ama�<nand�Shim�ura�and�has�b�Geen�established�as�an�extension�of�the�celebrated����6�w���ork�UUof�Wiles�and�T��*�a�ylor-Wiles�(see�[�2��]).����6��R��}'emark� `�6.8�.�NݲThe���formal�groups�mac���hinery�that�w�e�established�earlier�for�elliptic����6�curv���es���w�orks�on�mo�Gdular�ab�elian�v��q�arieties�as�w���ell,��although�there�is�no�general����6�form���ula�>�to�w�ork�with.�jNBut�one�can�establish�that�the�reduction�map�acting�on�the����6�torsion���p�Goin���ts��A�(�Q�)����tor��
��,��UX�!�-q�A�(�F����p���R�)����tor��O3�is�injectiv�e�as�w�ell,��using�similar�metho�Gds.�*W��*�e����6�do�UUnot�giv���e�the�details�here.����6��R��}'emark��(�6.9�.���Since�w:�J����1��|s�(�N��)�has�quite�large�dimension,���w���e�will�w�ork�en�tirely�with����6वJ����0��|s�(�N��)���in�the�follo���wing�mac�hinery�and,��(later,�in���the�computations.��(This�is�purely����6�a��pragmatic�consideration,�9*since�all�ab�Gelian�v��q�arieties�that�are�quotien���ts�of��J����0��|s�(�N��)����6�are�ealso�quotien���ts�of��J����1��|s�(�N��),�Jiand�so�are�mo�Gdular.���W��*�e�should�b�e�a���w�are,�Jiho�w�ev�er,����6�that��>there�exist�ab�Gelian�v��q�arieties�that�are�quotien���ts�of�the�latter�but�not�the�former;����6�these�UUwill�remain�un���treated�in�what�follo�ws.����6��R��}'emark�Nd�6.10�.�l߲Although�
�w���e�are�only�going�to�consider�mo�Gdular�ab�elian�v��q�arieties,����6�there�Дare�man���y�ab�Gelian�v��q�arieties�that�are�not�mo�dular.��It�is�conjectured�(in�[�12��
])����6�that�jan�ab�Gelian�v��q�ariet���y�is�mo�dular�if�and�only�if�it�is�of�\�GL������2���X�-t���yp�e".��Brie
y��*�,�<�an����6�ab�Gelian�'v��q�ariet���y��A=�Q��is�said�to�b�e�of��GL���@���2���s�-t���yp�e�if�there�exists��K�~4����Q��l�
���End�������$qy��msbm7�Q���޲(�A�)�'for����6�whic���h��C[�K�0k�:�yO�Q�]�=��dim��#�(�A�).���Here,����End���ŭ���Q���q�(�A�)�is�the�ring�of�endomorphisms�of��A��o���v�er����6��Q�.���This��fc���haracterization�is�a�generalization�of�the�T��*�aniy�ama-Shim�ura�conjecture,����6�since���an�elliptic�curv���e�is�a�dimension�1�ab�Gelian�v��q�ariet�y��*�,��Xand�taking��K�V8�=���Q��in�the����6�ab�Go���v�e�UUdenition�w���e�see�that�it�is�automatically�of��GL���n:���2��ꭲ-t�yp�Ge.����6�6.2.��Hec��9k�e��TOp�Q�erators�and�Mo�dular�Sym��9b�ols.��>F��*�rom�&`here�on,�/�the�theory�will����6�explicitly�Jiin���v�olv�e�the�mo�Gdular�subgroup�����0��|s�(�N��)�and�its�asso�ciated��X����0��|s�(�N��)�and��J����0���(�N��),����6�b�Gecause�UUof�a�preceding�remark�(6.9).����6��Denition��6.11.�A��The��5space�of��mo��}'dular�Ѭsymb�ols�TV�is��5a�shorthand�for�writing�elemen���ts����6�of�۴the�homology�group.��W��*�e�write��f��	z;������g��
2��H����1��|s�(�X����0���(�N��)�;��Q�)�۴for�the�homology�class����6�represen���ted�UUb�y�in�tegration�from���^ϲto�����,�b�Goth�elemen�ts�of�the�set�of�cusps��P���^��1��|s�(�Q�).����6��R��}'emark�ۇ�6.12�.�8��F��*�or�w�an���y�pair�of�elemen�ts���	z;����N4�2���P���^��1��|s�(�Q�),��the�elemen�t��f��	z;������g��is�in�fact�an����6�elemen���t�ubof�the�rational�homology��*�,�}fas�men�tioned�ab�Go�v�e;��ithis�is�a�theorem�of�Manin����6�and�UUDrinfeld.�����v��7
�������@v��ON���THE�TORSION�POINTS�OF�ELLIPTIC�CUR��#�VES�&�MODULAR�ABELIAN�V�ARIETIES���17����V������BलSince�Helemen���ts�of�����0��|s�(�N��)�act�on��h���^������,��they�should�also�act�on�the�mo�Gdular�sym�b�Gols����6�b���y�]~�g�[ٲ(�f��	z;������g�)�Բ=��f�g��	z;���g����g�.��BHo���w�ev�er,�_�for�]~�g�0��2�Բ�����0��|s�(�N��),�this�do�Ges�nothing�to�the�homol-����6�ogy��class,�Gb�Gecause�it�amoun���ts�only�to�a�c�hange�of�v��q�ariables�in�the�in�tegration.�Y�This����6�suggests��that�w���e�should�formalize�the�b�Geha�vior�of�mo�Gdular�sym�b�Gols,�!�using�what�w�e����6�already�UUkno���w�of�the�structure�of��H����1��|s�(�X����0���(�N��)�;����Q�).��]���6��Prop�Q�osition�
6.13.���The��.mo�Gdular�sym���b�ols��f��	z;������g��2��H����1��|s�(�X����0���(�N��)�;��Q�)��.inherit�the�fol-����6�lo���wing�UUprop�Gerties�from�prop�erties�of�in���tegration:��kЍ����G�(1)����Y�ĸf��	z;�����g���=�0�:�������Gॲ(2)����Y�ĸf��	z;������g���=��f��;����	z�g�:�������Gॲ(3)����Y�ĸf��	z;������g�8�+��f��;���
��8�g���=��f��	z;�
��8�g�:�������Gॲ(4)����Y��As�UUobserv���ed�ab�Go�v�e,�for�an�y��g�"�2�������0��|s�(�N��),��f�g�[ٲ(��	z�)�;���g��(����)�g��=��f��	z;�����g�.����6��Denition��|6.14.��7�Giv���en��a�prime��p��not�dividing�the�lev�el��N��,�Pdene�the��He��}'cke����6�op��}'er�ator�h�T����p���acting�UUon��H����1��|s�(�X����1���(�N��)�;����Q�)�UUb���y��/ύ�vrP�T����p���R�(�f��	z;������g�)��=������^������d���
#��p���+��0������
'�0���+�1�����+����^����'2��(�f��;������g�)�8�+�����PǴp��1���獍�Eş���X������k�+B�=0�����o���^������d���ˏ�1���(ː�k������ˏ�0���)
;�p�����.P���^����7X�(�f��;������g�)�:��^���6लThese�8�op�Gerators�comm���ute.��The�ring�of�Hec�k�e�op�Gerators�is�denoted�b�y��T���=��Z�[�T����1��|s�;���T����2���;��:�:�:��
UO�],����6�the�UUset�of��Z�-linear�com���binations�of�the�individual�op�Gerators.����6��R��}'emark��)�6.15�.�,ϲThe��individual�op�Gerations������b�����Ǎ��	��p����;�0���_I���	�s0����;1�����mV���b�����U�(�f��	z;������g�)�and������b�����>����	��1����\�k���؍��	��0����{�p����������b�������(�f��;������g�)�are�not��ݍ�6�w���ell-dened,��in�}that�they�dep�Gend�on�a�c�hoice�of�coset�represen�tativ�e��f��	z;������g�.��%Ho�w-����6�ev���er,��}the��sum�in�the�denition�of�the�Hec�k�e�op�Gerator�do�es�not.�,�Also�observ���e�that�the����6�\action"���of�a��GL��������2��|�(�Z�)�matrix�on�a�mo�Gdular�sym���b�ol,���and�therefore�on�its�constituen���t����6�elemen���ts���in��P���^��1��|s�(�Q�),��jhas�not�y�et�b�Geen�men�tioned.�E�By�a�sligh�t�abuse�of�condence�but����6�not��wof�denition,���w���e�can�extend�our�concept�of�matrix�action�on��P���^��1��|s�(�Q�)�to�include����6�all���elemen���ts�of��GL�������2���3�(�Z�),��since�the�imaginary�part�of�an�elemen�t�of��P���^��1��|s�(�Q�)�is�zero�and����6�the�UUset�is�therefore�preserv���ed�b�y�linear�fractional�transformations.��<0���k�'7.��x�E�Comput��UTa�tions��with�Modular�Abelian�V���Uarieties����BलHere��Jw���e�will�brie
y�men�tion�an�algorithm�(mostly�giv�en�in�[�5��],���c�h.�JnI�GI)��,for�nding����6�a�M�m���ultiple�of�#�A�(�Q�)����tor��
^�for�mo�Gdular�ab�elian�v��q�arieties�where�w���e�explicitly�kno�w�the����6�smallest�c�lev���el��N�z��of�a�surjectiv�e�map��J����1��|s�(�N��)����!��A�.���This�c�is�essen�tially�b�y�w�a�y�of����6�giving�UUbac���kground�for�the�computations�w�e�p�Gerform�in��x�8.��Hۍ�6�7.1.��Finding���the�Quotien��9ts�of��J����0��|s�(�N��)�.��The���rst�thing�to�do�is�compute�the����6�homology���group��H����1��|s�(�X����0���(�N��)�;����Q�),�ըas���represen���ted�b�y�the�group�of�mo�Gdular�sym�b�Gols.����6�This�J�group�is�isomorphic�to�a�certain�nite�group��H����(�N��)�with�relations�that�are����6�relativ���ely�easy�to�w�ork�with�(this�is�a�result�due�to�Manin,�Ecited�in�[�5��],�theorem����6�2.1.4),��so���the�computations�are�easier�than�one�w���ould�exp�Gect�giv�en�the�denition�of����6�the��[mo�Gdular�sym���b�ols�space.���The�group��H����(�N��)�decomp�oses�in���to�simple��T�-mo�dules����6वM����1���'�:���:�:��|jM����2�r��8�under���the�action�of�the�Hec���k�e���algebra,��and�there�is�a�w���a�y���in�whic���h�these����6�mo�Gdules��pair�up�canonically�to�asso�ciate�with�eac���h�of�the��r�$�ab�elian�v��q�arieties�of�lev���el����6वN��.����6�7.2.��The��T��
�orsion�Multiple.��It���is�a�deep�fact�that�the�c���haracteristic�p�Golynomial����6�of�Rcthe�Hec���k�e�Rcop�Gerator��T����p���acting�on�the��M����i�����asso�ciated�to�a�v��q�ariet���y��*�,���ev�aluated�Rcat����6वp����+�1,�Ggiv���es��the�n�um�b�Ger�of�p�oin���ts�in�the�reduction�o�v�er��F����p���R�.�VAThis�is�analogous�to�the����6�situation���with�elliptic�curv���es,�'although�without�a�W��*�eierstrass�equation�the�Hec�k�e�����#���7
��������6ऱ18����SETH���KLEINERMAN���V������6लop�Gerators��Lare�the�only�w���a�y��Lto�nd�the�size�of�the�reduced�v��q�ariet���y��*�.�]�F�or��Lprimes�of����6�go�Go�d���reduction,���the�torsion�subgroup�is�injectiv���ely�reduced�and�its�order�is�a�divisor����6�of���UUc���harp�Goly���(�x(�T����p���R�)(�p�8�+�1).����B�So�konce�w���e�ha�v�e�the�v��q�arieties�asso�Gciated�to�the�c�hosen�lev�el�represen�ted�as��T�-����6�mo�Gdules�`�M����i��TL�,�^w���e�compute�the�c�harp�Goly�of��T����p�����acting�on��M����i��TL�,�^and�ev��q�aluate�it�at��p����+�1,����6�and��+then�tak���e�the�GCD��of�the�resulting�torsion�m�ultiple�for�v��q�arious�primes��p��not����6�dividing�UUthe�lev���el.��:B|����z�8.���B�Examples����BलUsing�}�t���w�o�computer�w�orkstations,���MECCAH�}�(the�Mathematics�Extreme�Com-����6�putation�{�Cluster�at�Harv��q�ard)�and�Neron�(named�for�Andr�����Ge�N���Geron),���William�Stein����6�computed�ythe�c���haracteristic�p�Golynomials�of�the�rst�eigh�t�prime�Hec�k�e�op�Gerators����6�acting��on�all�simple�quotien���ts�of��J����0��|s�(�N��)�up�to��N��,�=�o5000,��Pwith�less�comprehensiv�e����6�calculations�Wup�to��N���=���7244,�Wzas�w���ell�as�the�dimensions�of�these�v��q�arieties.�v�W��*�e�used����6�this�t�data�to�compute�a�m���ultiple�of�the�torsion�group�for�eac�h�one,�|�as�ab�Go�v�e,�|�using����6�primes�UUup�to�19.����B�This�B�rst�table�sho���ws�the�p�Gossible�torsion�m�ultiples,�~ sorted�b�y�dimension,�~ for����6�lo���w�UUdimensions.�q�F��*�ull�ranges�are�giv�en�b�y�a�dash.���G^���Bट���ff]�ۤfd����ͤ���ff��͟�fdDimension���ff����8�T��*�orsion�UUMultiples�F�ound�\������ff����ff]�ۡ����ͤ���ff�P^��fd1���ff����W31-10,�UU12,�16,�20���^����s������ff����ff]�ۡ����ͤ���ff�P^��fd�2���ff���S�D1-17,�UU19,�20,�22-24,�27,�28,�30-32,�36,�37,�44,�52,�56,�64������ff����ff]�۟}r����ͤ���ff�P^��fd3���ff���G*�1-20,�UU22-24,�28,�31,�32,�36-38,�49,�58,�64,�72,�80,�83���^��y��*��,�92,�160��^����ff����ff]�ۡ����ͤ���ff�P^��fd4���ff���BNB1-17,�UU19-29,�31,�32,�34-36,�38,�40,�44,�48,�49,�55,�56,�64,�74,�76,�
����ff�������͟���ff�7���ff�����96,�UU152,�176,�208,�256,�328���^��y��[ٟ���ff����ff]�۟}r����ͤ���ff�P^��fd�5���ff���A�{1-14,�UU16,�18-25,�27-29,�31,�32,�34-36,�38,�41���^��y��*��,�44,�48,�50,�54,�56,�	b%����ff�������͟���ff�7���ff���=#58,�UU63,�64,�68,�72,�86���^��y��*��,�92,�104,�110,�128,�212���^��y���,�224,�272,�288,�320��͟���ff����ff]�۟}r����ͤ���ff�P^��fd6���ff���E�	1-18,�UU20,�22,�24,�25,�27,�28,�31,�32,�34,�36,�37,�40,�41���^��y��*��,�48,�49,�
E�����ff�������͟���ff�7���ff���@��52,�UU53���^��y��*��,�56,�64,�70,�74,�76,�80,�96,�104,�120,�124,�128,�192,�212���^��y���,�[email protected]����ff�������͟���ff�7���ff�����N248,�UU288,�296,�432�k������ff����ff]�ێ���B^¡��B�Tw���o�uEthings�jump�out�at�us.�іFirst,��@the�starred�n�um�b�Ger�in�the�ab�o���v�e�uEc�hart�is����6�the�oonly�case�where�the�torsion�m���ultiple�is�immediately�seen�to�b�Ge�to�o�large,�N5b���y����6�Theorem���4.1.�KJThe�v��q�ariet���y�is�the�ten�th�in�lev�el�2310.�KJSince�2310��=�2�Q���3����5����7����11,���it���is����6�lik���ely�hHthat�w�e�merely�didn't�use�enough�primes�(the�computations�could�only�tak�e����6�adv��q�an���tage���of�the�primes�13,���17,�and���19).���A���quic�k�follo�w-up�searc�h�in�the�mo�Gdular����6�forms��hdatabase�sho���ws�that�our�surmise�w�as�correct:���using�primes�through�53,��-the����6�GCD�Sof�hthe�torsion�m���ultiples�is�reduced�quite�drastically��*�,�1to�4,�whic���h�is�w�ell�within����6�the�UUb�Gound�imp�osed�b���y�Theorem�4.1.����B�Next,�(the���daggers�in�the�table�represen���t�torsion�m�ultiples�divisible�b�y�a�prime����6�greater�\Qthan�40.���Since�there�isn't�ev���en�an�y�11-torsion�in�dimension�1,�^it's�in�terest-����6�ing��to�in���v�estigate��some�examples�of�torsion�p�Goin���ts�of�large�prime�order�in�higher����6�dimensions.���W��*�e'll�~mak���e�a�separate�table�of�these��y�-ed�v��q�arieties;��`where�p�Gossible�w�e����6�use��the�mo�Gdular�forms�database�to�see�whether�the�torsion�m���ultiple�gets�smaller����6�using�UUprimes�up�to�53.������7Р�7
�������@v��ON���THE�TORSION�POINTS�OF�ELLIPTIC�CUR��#�VES�&�MODULAR�ABELIAN�V�ARIETIES���19����V��������Bट�]m�ff/㲤fd����ͤ���ff��@��fd�N���ff���1��Num���b�Ger�lΟ���ff���f�&Dimension��͟���ff�����IMultiple�UUof��͟���ff����#�Multiple�UUof�torsion��͟���ff�������ͤ���ff��͟�fd(lev���el)���ff���*8�within�UUlev���el��͟���ff���������ff�����Ktorsion�
�ϟ���ff����`(using�UUMFD)�	A����ff����ff/㲡����ͤ���ff�i?��fd5010���ff���>�16�����ff���{J�3�P^����ff������83��{����ff������{�+,ӟ���ff����ff/㲡����ͤ���ff�
�?��fd498���ff���A7������ff���{J�4�P^����ff����x�328�wz����ff�����82�(�ҟ���ff����ff/㲡����ͤ���ff�
�?��fd830���ff���>�11�����ff���{J�5�P^����ff������41��{����ff�����41�(�ҟ���ff����ff/㲡����ͤ���ff�i?��fd1038���ff���>�10�����ff���{J�5�P^����ff������86��{����ff�����86�(�ҟ���ff����ff/㲡����ͤ���ff�
�?��fd642���ff���A8������ff���{J�5�P^����ff����x�212�wz����ff������106�&,ҟ���ff����ff/㲤}r����ͤ���ff�
[email protected]��fd83���ff���A2������ff���{J�6�P^����ff������41��{����ff�����V41���^��z��*�(����ff����ff/㲡����ͤ���ff�
�?��fd�326���ff���A5������ff���{J�6�P^����ff������41��{����ff�����V41���^��z��*�(����ff����ff/㲤fd����ͤ���ff�i?��fd�2158���ff���>�10�����ff���{J�6�P^����ff������41��{����ff�����41�(�ҟ���ff����ff/㲡����ͤ���ff�i?��fd1070���ff���>�12�����ff���{J�6�P^����ff������53��{����ff�����53�(�ҟ���ff����ff/㲟}r����ͤ���ff�
�?��fd422���ff���A6������ff���{J�6�P^����ff����x�212�wz����ff�����V53���^��z��*�(����ff����ff/㲎���I����BलLet��ous�consider�the�second�en���try�in�the�table:�?�the�7th�mo�Gdular�ab�elian�v��q�ariet���y����6�of��Olev���el�498;�
�it�has�dimension�4.�ܴUsing�primes�up�to�19,��w�e�computed�328�as�the����6�torsion���m���ultiple�and�40709�(a�prime)�as�the�m�ultiple�of�the�eld�discriminan�t.�:;Since����6�498�mis�only�divisible�b���y�t�w�o�of�the�primes�on�our�list�(2�and�3),�5one�ma�y�b�Ge�inclined����6�to��\trust,"�more�or�less,���the�computation�of�the�torsion�m���ultiple.�M9In�particular,�328����6�has��a�large�prime�factor,�Dnamely�41,�and�w���e'd�lik�e�to�kno�w�whether�w�e�can�observ�e����6�41-torsion�UUin�the�v��q�ariet���y�explicitly��*�.����B�There�UUis�a�series�of�maps�� ������p�X����0��|s�(�N��)���,��UX�!��J����0���(�N��)����A����6लwhere��the�rst�map�b�Get���w�een��p�oin�ts��and�divisors�is��P�K_�7!��в[�P����]1�].�y{The�image�of����6�the�z�cusps�in��X����0��|s�(�N��)�under�this�map�can�b�Ge�used�to�generate�a�subgroup�of�the����6�Jacobian��called�the�cuspidal�subgroup,��C���.��It�is�a�fact�that��C��(�Q�)�����J����0��|s�(�N��)(�Q�)����tor��
�u�.����6�Then��under�the�quotien���t�map�in�to��A�(�Q�)�it�will�end�up�as�a�subgroup�of�the�torsion����6�on���the�v��q�ariet���y��*�.�,�Th�us���this�is�a�w���a�y���of�explicitly�constructing�a�lo���w�er���b�Gound�on�the����6�torsion,�UUb���y�actually�nding�a�subgroup�of�it.����B�So� tthe�natural�thing�to�do�is�to�try�that�with�this�v��q�ariet���y��*�.�`'W�e� tconstruct�it�using����6�MA���GMA,�UUas�follo�ws:��RX��6��>�?�S�:=�CuspidalSubspace(ModularSymbols(498,2,+1));����6�>�?�D�:=�TraceSortDecomposition(NewformDecomposition(S));����6�>�?�A�:=�D[15];����6लThen�#�the�command��RationalCuspidalSubgroup(A)�#��giv���es�the�subgroup�generated����6�b���y���the�image�of�the�cuspidal�subgroup�under�quotien�t.�i[In�this�case,��}it�returns����6��Abelian�?�Group�of�order�1�.����B�(Note���that�in�the�v��q�arieties�mark���ed�with�a��z��in�the�table�ab�Go�v�e,��this�metho�Gd����6�pro�Gduces���a�non���trivial�subgroup�and�therefore�conrms�the�high�prime�torsion�ex-����6�plicitly��*�.)����B�P���erhaps�'�unfortunately��*�,�0�in�the�example�w�e�w�ere�considering,�0�the�subgroup�of�the����6�torsion���group�that�this�metho�Gd�pro�duces�is�trivial.�But�there�do�es�seem�to�b�e�41-����6�torsion���in�the�v��q�ariet���y��*�.�6�If�w�e�use�primes�up�to�53�(computed�in�the�mo�Gdular�forms����6�database),��the��torsion�m���ultiple�is�rened�to�82.�FWAll�this�is�v�ery�suggestiv�e,��but�w�e'd����6�lik���e�UUto�kno�w�whether�41�isn't�just�a�red�herring�that�k�eeps�app�Gearing.�����Iz��7
��������6ऱ20����SETH���KLEINERMAN���V������BलT��*�o�e�this�end,�jw���e�can�turn�to�the�Birc�h�and�Swinnerton-Dy�er�conjecture�to�mak�e����6�predictions.�q�The�UUconjecture�supp�Goses�an�equalit���y��'ݍ�����<$���-m�L�(�A;����1)���-m�w�fe��	(֍��V
����A������Įײ=�����Y����
/ٟ���Q�������p�j�N��&aM�c����p���2��8�#����hV1
wncyr10�X���
�n�(�A�)���K�Y��fea\~�	(֍#�A�(�Q�)����tor���U��8�#�A���r�_��㏲(�Q�)����tor������f���;���O��6लwhere���the�left�hand�side�is�an�analytic�in���v��q�arian�t���asso�Gciated�to�the�v�ariet���y�and�the����6�righ���t�hwhand�side�con�tains�certain�algebraic�in�v��q�arian�ts.��-The�ones�w�e�are�particularly����6�concerned�A�with�are�the�torsion�group�sizes�in�the�denominator�on�the�righ���t,�x�of�course.����6�In�)�the�example�ab�Go���v�e�)�with�lev���el�498,�^�man�y�of�these�quan�tities�can�b�Ge�computed����6�explicitly��*�.�0�Using��bthe�MA���GMA��0commands��LRatio��and��TamagawaNumber��(the�former����6�predicts��!the�left�hand�side�of�the�equation�up�to�a�constan���t�conjecturally�equal�to����6�1),�UUw���e�nd�that�the�BSD�conjecture�predicts�that��Ӎ�����56��=�����<$���K(2296�8���164����1)#���X���
�n�(�A�)���K�w�fec�ğ	(֍�8##�A�(�Q�)����tor���U��8�#�A���r�_��㏲(�Q�)����tor������h�B�:����6लAll�the�remaining�unkno���wn�quan�tities�are�in�tegers,� Jand�the�n�umerator�on�the�righ�t����6�hand�q�side�is�divisible�b���y�41���^��2��|s�,�ywhile�the�left�hand�side�has�no�divisors�of�41.�ǫHence����6�if�k�BSD�k�is�correct,��jw���e�should�exp�Gect�the�v��q�ariet�y�and�its�dual�to�ha�v�e�41-torsion.����6�(W��*�e���exp�Gect�b�oth�of�them�to�ha���v�e���it�b�ecause�neither�of�them�can�ha���v�e���t�w�o�factors����6�of�]�41,���since�the�torsion�m���ultiple�is�an�upp�Ger�b�ound�on�all�curv���es�in�the�isogen�y����6�class.)�[�Often��w���e�can�use�the�BSD��conjecture�to�mak�e�statemen�ts�ab�Gout�the�torsion����6�subgroup.��R���6�8.1.��Bounds��using�Go�Q�o�d��Reduction.��If��a�A��is�of�dimension�2�and�has�go�Go�d��areduc-����6�tion�g�at�the�prime�3,�lthen�#�A�(�Q�)����tor��
�޸��i�#�A�(�F����3��|s�).��ZIt�is�kno���wn�that�the�c�haracteristic����6�p�Golynomial�2�of�the�Hec���k�e�2�op�erator��T����3���;�factors�as�(�x�́���a����3��|s�)(�x����a���^�����l��3���b��).�
 F��*�urther,�j%a�2�re-��
�ɍ�6�sult�H�due�to�Deligne�states�that��j�a����p���R�j�;����j�a���^������p���b��j�\���2�p����G��33�k����1��33��i�W	���P����2�����T�,���where�Hʵk��a�is�the�dimension.�L%In��
�ˍ�6�this�zBcase,��|�a����3�����and�its�conjugate�are�b�Gounded�in�absolute�v��q�alue�b���y�2�����P�p���UW���P�fe�E���3����
UX.��W��*�riting��6��6�these�3�as�elemen���ts�of�the�ring�of�in�tegers�in�the�real�quadratic�eld��Q�[�������p���UW�����fe���wv��D������],�:fw�e�ha�v�e���@��6सj�a�=��+��b�������p���UW�����fe���wv��D������j�;����j�a����b�������p���UW�����fe���wv��D�����j�����2�����P�p���UW���P�fe�E���3����
UX.���W��*�e'd�\�lik���e�to�b�Gound�the�p�ossible�v��q�alue�of�the�c���harac-����6�teristic�yOp�Golynomial�ev��q�aluated�at�4=3+1,��Mwhic���h�w�ould�tell�us�the�n�um�b�Ger�of�p�oin���ts����6�o���v�er�UU�F����3��|s�.����B�Plugging�?in��x��G�=�4�?and�m���ultiplying�out,�4�w�e�need�to�nd�the�maximal�v��q�alue�of����6�16�L+�(�a���^��2��Ȉ���b���^��2��|s�D�G�)����8�a�,�yYwhic���h�r%is�at�its�largest�when��b���=�0�r%and,�if�the�absolute�v��q�alue����6�of����a��is�b�Gounded,���when��a�S�<��0���is�minimized�within�the�range.�n�W��*�e�do�some�ddling����6�with�UUDeligne�b�Gound�and�the�triangle�inequalit���y�to�nd�this�range:������������������j�2�a�j���������~��j�a�8�+��b������p���UW����fe����q��D������j��+��j�a����b������p���UW����fe����q��D�����j��������������������j�2�a�j���������~���4����=V�p���UW��=V�fe�ª��3�����������������������j�a�j���������~���2����=V�p���UW��=V�fe�ª��3��������g⍑6�Since�u��a��can�only�tak���e�on�discrete�in�teger�or�p�Gossibly�half-in�teger�v��q�alues,�}��a����=���3�u�is����6�the��least�p�Gossible;�uand�th���us�the�n�um�b�Ger�of�torsion�p�oin���ts�on�a�dimension�2�mo�dular����6�ab�Gelian�UUv��q�ariet���y�with�go�o�d�reduction�at��p���=�3�UUis�at�most�(4�8���(��3))���^��2��C��=��49.����B�T��*�o���conrm,���w���e�searc�h�our�data�for�v��q�arieties�with�dimension�2�and�a�computed����6�torsion�UUm���ultiple�greater�than�49:������\y��7
�������@v��ON���THE�TORSION�POINTS�OF�ELLIPTIC�CUR��#�VES�&�MODULAR�ABELIAN�V�ARIETIES���21����V���������Bट�LΉff>�{�fd����ͤ���ff��͟�fd�N�UU(lev���el)���ff���5=Num���b�Ger�UUwithin��͟���ff����x�Dimension��͟���ff�����Multiple�UUof�eld��͟���ff���	�Multiple�UUof��͟���ff�������͟���ff�/�ׄff���L��lev���el�IC����ff����L����ff����`discriminan���t��Ο���ff���� torsion�
�ϟ���ff����ff>�{�����ͤ���ff�S��fd390���ff���S×8�#�'����ff������2�P^����ff���׀32�"�џ���ff�����56��{����ff����ff>�{�����ͤ���ff�
���fd2574���ff���QC�26�!&����ff������2�P^����ff����164� Lџ���ff�����64��{����ff����ff>�{�����ͤ���ff�
���fd3120���ff���QC�29�!&����ff������2�P^����ff���׀32�"�џ���ff�����56��{����ff����ff>�{�����ͤ���ff�
���fd4368���ff���QC�32�!&����ff������2�P^����ff���׀17�"�џ���ff�����52��{����ff����ff>�{�����ͤ���ff�
���fd5070���ff���QC�29�!&����ff������2�P^����ff���׀32�"�џ���ff�����56��{����ff����ff>�{�����ͤ���ff�
���fd5850���ff���QC�64�!&����ff������2�P^����ff���׀32�"�џ���ff�����56��{����ff����ff>�{����0����B�Happily�E�enough,�H�all�the�lev���els�here�are�divisible�b�y�3,�H�so�w�e�don't�ha�v�e�to�w�orry����6�ab�Gout���the�torsion�m���ultiple�violating�our�b�ound.��In�fact,��Yfor�a�curv���e�with�go�o�d����6�reduction�35at�3,�:the�highest�torsion�m���ultiple�in�the�data�w�e�ha�v�e�computed�so�far�is����6�22.�B2The���v��q�ariet���y�in�whic�h�this�putativ�e�torsion�o�Gccurs�is�a�quotien�t�of��J����0��|s�(322).�B2As����6�in�]zthe�previous�section,�_�for�this�v��q�ariet���y��RationalCuspidalSubgroup��is�trivial,�but����6�the���BSD��uconjecture�suggests�that�there's�at�least�11-torsion�in�the�v��q�ariet���y�and�its����6�dual.����B�This�PDtable�also�serv���es�to�illustrate�t�w�o�con�tradictory�eects,���since�the�torsion����6�m���ultiple��is�high�and�man�y�small�primes�divide�the�lev�el.��First,���since�man�y�small����6�primes���divide�the�lev���el,��the�torsion�m�ultiple�ma�y�b�Ge�to�o�high,��since�w���e�ha�v�e�few�er����6�n���um�b�Gers���in�the�GCD�ܼcalculation.�fSecond,���if�the�torsion�is�high,�it�ough���t�to�ha�v�e����6�man���y���small�primes�dividing�the�lev�el,�Гso�as�to�ha�v�e�bad�reduction�at�the�small����6�primes�Sand�therefore�the�capacit���y�for�higher�torsion.���It�is�not�easy�to�tell�whic�h����6�eect�UUis�dominan���t�in�a�particular�example.�����6�8.2.��Conjectures.��Sifting��Fthrough�the�data,�$�w���e�nd�strong�evidence�for�the�fol-����6�lo���wing�UUconjectures.���k���6��Conjecture��T8.1.����(1)�������Giv���en�t۵p����1��|s�;���p����2���;�p����3���N�distinct�t�primes,���if�the�n���umerator�of�����������p���i��*���1����);�fe���������12�����a��,����Y��whic���h��w�e'll�call��`����i��TL�,��Qis�a�prime�for�some��i�,�then�there�is�an��`����i��TL�-torsion�p�Goin���t����Y��in�UUsome�mo�Gdular�ab�elian�v��q�ariet���y�whose�lev�el�is��p����1���S��8�p����2�����p����3��|s�.�������G�(2)����Y��An�UUanalogous�statemen���t�holds�for�an�y�o�Gdd�n�um�b�Ger�of�distinct�primes��p����i��TL�.��xፍ���G�(3)����Y��The��3ab�Go���v�e�statemen�ts�hold�for�an�y�prime��`����i����dividing�the�n�umerator�of����������f�p���i��*���1���f�);�fe���������12������W�.����6��Conjecture���8.2.��=�F��*�or�5`an���y�prime��p��>��2,�;�there�5`is�a��p�-torsion�p�Goin�t�in�some�mo�Gdular����6�ab�Gelian�UUv��q�ariet���y�whose�lev�el�is��p���^��3��|s�.����B�F��*�or��6an�example�of�the�rst�statemen���t,�orecall�the�ab�Gelian�v��q�ariet�y�with�lev�el�498��=����6�2�xt���3����83���ab�Go���v�e;��cthe�BSD���computation�suggested�strongly�that�it�had�a�41-torsion����6�p�Goin���t.��;Similar�&computations�can�b�e�made�with�other�v��q�arieties�in�that�table:��jthe����6�one���with�lev���el�642�(>=�2�_����3����107���app�Geared�to�ha�v�e�a�53-torsion�p�Goin�t.� �In�fact,��2up�Gon����6�closer��insp�Gection,��@all�the�v��q�arieties�with�high�torsion�that�w���e�found�seem�to�satisfy����6�this�UUconjecture.����B�This���rst�observ��q�ation�is�b�Gound�to�b�e�related�to�the�follo���wing�theorem,��awhic�h�is����6�pro���v�ed�UUin�[�10��
]:��8L��6��Theorem��K8.3��(Mazur)�.��;�L��}'et���p��>��3��b��}'e�a�prime,��Gand�let��n��b�e�the�numer�ator�of�����������p��1����);�fe[Ɵ����1p12�����J�.����6�Then�ΐthe�torsion�sub��}'gr�oup�ΐof�the�Mor��}'del���l-Weil�gr�oup�of��J����0��|s�(�p�)��is�a�cyclic�sub�gr�oup����6�of���or��}'der��n�.����BलAlthough���this�theorem�addresses�the�case�of�prime�lev���el,��!it�is�not�immediately����6�clear�+!wh���y�Conjecture�8.1�seems�to�hold�when�the�lev�el�is�the�pro�Gduct�of�an�o�dd�����n���7
��������6ऱ22����SETH���KLEINERMAN���V������6लn���um�b�Ger�d�of�distinct�primes.��jIt�is�also�w���orth�noting�that�lev�els�that�are�the�pro�Gduct����6�of�UUan�ev���en�n�um�b�Ger�of�distinct�primes�do�not�exhibit�this�b�eha���vior.����B�The�i�second�observ��q�ation�is�a�bit�more�puzzling�still,���and�w���e�do�not�suggest�a����6�reason�UUfor�its�truth�here.����B�It��5is�en���tirely�p�Gossible�that�b�oth�observ��q�ations�ha���v�e��5b�een�made�b�efore�or�already����6�pro���v�ed;�UUho�w�ev�er,�w�e�do�not�at�this�p�Goin�t�kno�w�of�a�source�for�them.��fd������9.���a��A��Uckno��wledgments����BलI��Bw���ould��Qlik�e�to�thank�m�y�adviser,��Benjamin�P�eirce�Assistan�t�Professor�William����6�A.�UUStein,�for�his�en���th�usiasm�UUand�explanations.��������7
�������@v��ON���THE�TORSION�POINTS�OF�ELLIPTIC�CUR��#�VES�&�MODULAR�ABELIAN�V�ARIETIES���23����V������͸��References������; ��[1]���HY�M.F.���A�Îtiy�ah�and�I.G.�Macdonald.��.#�f�cmti8�Intr���o�duction��"to�Commutative�A���lgebr���a�.�Addison-W��J�esley�,��
��HY�1969.������; �[2]���HY�C.��Breuil,���B.�Conrad,�F.�Diamond,�and�R.�T��J�a�Îylor.��On��the�Mo���dularity�of�El�p[liptic�Curves�over����HY��p�p�msbm8�Q�:�i�Wild�M�3-adic�Exer���cises�.�+Journal�of�the�American�Mathematical�So�<rciet�Îy��J�,�%14,�no.�+4�(2001),����HY�843-939.������; �[3]���HY�J.W.S.�/�Cassels�and�E.V.�Flynn.��Pr���ole�gomena�xqto�a�Midd�p[lebr���ow�A���rithmetic�of�Curves�of�Genus����HY�2�.��XCam�Îbridge�Univ�ersit�y�Press,�1996.������; �[4]���HY�G.��XCornell�and�J.H.�Silv�Îerman,�eds.��A���rithmetic�~Ge���ometry�.�Springer-V��J�erlag,�1986.������; �[5]���HY�J.E.�7Cremona.��A���lgorithms�nfor�Mo���dular�El�p[liptic�Curves�,�V�second�7edition.�Cam�Îbridge�Univ�ersit�y����HY�Press,��X1997.������; �[6]���HY�D.���Doud.��A�Pr���o�c�e�dur�e�"to�Calculate�T��Wworsion�of�El�p[liptic�Curves�Over��Q�.���Man�Îuscripta�Mathe-����HY�matica,��X95�(1998),�463-469.������; �[7]���HY�E.V.���Flynn,��IF.�Lepr�Î���nev�ost,�E.F.���Sc�haefer,�W.A.���Stein,�M.�Stoll,�and�J.L.�W��J�etherell.��Empiric���al����HY�Evidenc���e�թfor�the�Bir�ch�and�Swinnerton-Dyer�Conje�ctur�es�for�Mo�dular�Jac�obians�of�Genus�2����HY�Curves�.��XMathematics�of�Computation,�236�(2001),�1675-1697.������; �[8]���HY�N.���Katz.��Galois��{Pr���op�erties�of�T��Wworsion�Gr�oups�on�A���b�elian�V��Wwarieties�.���In�Îv�en�tiones�Mathemati-����HY�cae,��X62�(1981),�481-502.������; �[9]���HY�S.J.��Kleinerman.��The�Jac���obian,�I!the�A���b�el-Jac�obi�Map,�I!and�A���b�el's�The�or�em�.��Submitted�as����HY�junior��Xpap�<rer,�Harv��ard,�2003.������6�[10]���HY�B.��	Mazur.��Mo���dular���Curves�and�the�Eisenstein�Ide�al�.��	Publications�math�Î���nematiques�de��
�鍑HY�l'I.H.������P�E.S.,��X47,�no.�2�(1977),�33-186.������6�[11]���HY�L.���Merel.��Bornes���p���our�la�T��Wworsion�des�Courb�es�El�p[liptiques�sur�les�Corps�des�Nombr�es�.���In�Îv�en-����HY�tiones��XMathematicae,�124�(1996),�437-449.������6�[12]���HY�K.A.��&Rib�<ret.��A���b���elian�PV��Wwarieties�over��Q��and�Mo�dular�F��Wworms�.��&Article�found�in��Mo�dular�PCurves����HY�and�~A���b���elian�V��Wwarieties�.��XBirkha�<r����user,�2004.������6�[13]���HY�K.A.�4�Rib�<ret�and�W.A.�Stein.��Mo���dular�kF��Wworms,��DHe�cke�Op�er�ators,��Dand�Mo�dular�A���b�elian�V��Wwari-����HY�eties�.��XPreprin�Ît.������6�[14]���HY�J.-P��J�.��XSerre.��A�~c���ourse�in�A���rithmetic�.��XSpringer-V�erlag,�1973.������6�[15]���HY�I.R.��XShafarevic�Îh.��Basic�~A���lgebr���aic�Ge�ometry�1�.��XSpringer-V��J�erlag,�1994.������6�[16]���HY�G.��Shim�Îura.��Intr���o�duction��fto�the�A���rithmetic�The���ory�of�A�utomorphic�F��Wwunctions�.��Princeton����HY�Univ�Îersit�y��XPress,�1971.������6�[17]���HY�J.H.��XSilv�Îerman.��The�~A���rithmetic�of�El�p[liptic�Curves�.��XSpringer-V��J�erlag,�1986.������6�[18]����HY��ff���cDo.��X�A���dvanc�e�d�~T��Wwopics�in�the�A���rithmetic�of�El�p[liptic�Curves�.�Springer-V��J�erlag,�1994.������6�[19]���HY�J.H.��XSilv�Îerman�and�J.�T��J�ate.��R���ational�~Points�on�El�p[liptic�Curves�.��XSpringer-V�erlag,�1992.������6�[20]���HY�W.A.���Stein.��The���A���rithmetic�of���2cmmi8�J��q��Aa�cmr6�1��*��(�p�).���Notes�for�a�talk�deliv�Îered�at�Brandeis�in�Octob�<rer�2003.������6�[21]���HY�W.A.�yvStein.��Explicit��Appr���o�aches�to�Mo�dular�A���b�elian�V��Wwarieties�.�yvPh.D.�thesis,��}Univ�Îersit�y�yvof����HY�California��Xat�Berk�Îeley��J�,�2000.������W���;��7
���.#�f�cmti8�+�%n�

eufm10�*��<x

cmtt10�(t}\�cmti7�&�-�
cmcsc10�%�':

cmti10�$qy��msbm7�#���

msbm10�!����

msam10��-�

cmcsc10��"V

cmbx10�p�p�msbm8��2cmmi8��Aa�cmr6�|{Ycmr8��hV1
wncyr10�
!",�

cmsy10�O!�cmsy7���0ncmsy5�
�b>

cmmi10�	0e�rcmmi7�O
�\cmmi5�K�`y

cmr10�ٓ�Rcmr7���Zcmr5���u

cmex10��M�������