A  NOTE  ON  THE  HYPERGEOMETRIC  AND  BESSEL'S  EQUATIONS  ( II  )  .sx

By  J.L.  BURCHNALL  ( Durham  ) and  T.W.  CHAUNDY  ( Oxford  )  .sx

Received  26  May  1931   .sx

1 .sx   Introduction   .sx

IN  part  I  of  this  Note  we  called  attention  to  certain  elementary  propositions  in  the  theory  of  the  generalized  hypergeometric  equation    , ( 1  ) which  enable  us  to  transform  equations  of  this  type  , by  substitutions  of  the  form    , ( 2  ) into  equations  of  similar  or  simpler  character  in  z.   .sx

Our  arguments  were  based  on  the  main  theorem  I  ( 2  ) , namely :sx   

If  z  is  a  solution  of  the  equation    , then    is  a  solution  of  the  equation   .sx   

In  part  I  we  applied  this  theorem  to  a  special  class  of  equations  ( 1  ) , namely  , those  of  the  generalized  Bessel's  type    , ( 3  ) employing  it  mainly  to  throw  light  on  the  classification  of  such  equations  by  the  types  of  functions  entering  into  their  general  solution .sx   

In  this  part  II  we  undertake  a  similar  inquiry  for  the  more  general  equations  ( 1  ) , in  which  , however  , some  of  the  simplicity  of  the  earlier  inquiry  will  be  lacking  for  the  following  reason .sx   The  substitution  ( 2  ) derives  the  complete  solution  of  the  y-equation  from  the  z-equation  is  less  than  that  of  the  y-equation  , or  else  a  solution  of  the  z-equation  is  annihilated  by  the  operator   .sx   

With  equations  of  the  generalised  Bessel's  type  ( 3  ) neither  of  these  possibilities  occur .sx   On  the  one  hand  the  order  of  the  y-equation  is  the  difference  of  the  orders  of  the  o-operators  on  the  two  sides  of  the  equation  , which  is  invariant  for  substitutions  of  the  type  ( 2) .sx   The  z-equation  cannot  therefore  be  of  lower  order  than  the  y-equation  , and  in  cases  of  interest  is  of  the  same  order  , when  consequently  it  is  also  of  the  generalized  Bessel's  form  ( 3) .sx   Therefore  , on  the  other  hand  , no  solution  of  the  z-equation  can  be  annihilated  by  the  operator    , for  this  operator  annihilates  only  finite  sums  of  the  form    , whereas  the  generalized  Bessel's  equations  have  no  finite  solutions  of  this  form .sx   

Thus  , as  we  have  found  in  part  I  , the  complete  solution  of  the  z-equation  leads  at  once  to  the  complete  solution  of  the  y-equation .sx   With  equations  of  the  more  general  type  ( 1  ) this  need  not  always  be  the  case  , and  we  must  then  fill  in  the  'missing'  solution  ( or  solutions  ) from  other  sources .sx   

2 .sx   Amplification  of  Theorem  I  ( 2  )  .sx

We  begin  by  extending  the  theorem  I  ( 2  ) into  a  form  which  takes  account  of  these  missing  solutions .sx   

We  evidently  lack  a  solution  y  , if  and  only  if  some  z  is  a  finite  sum  of  powers   .sx   

For  this  the  necessary  and  sufficient  condition  is  that    have  a  factor    , where  s  has  one  of  the  above  values .sx   If  there  are  several  such  factors  , suppose    is  that  with  greatest  s  , and  consider  the  equation   .sx   ( 4  )  .sx

The  substitution    now  gives  for  z  the  equation    or  sufficiently  , if    ,  .sx   

In  this  case  , then  , as  compensation  for  missing  a  solution  , we  have  been  able  to  lower  the  order  of  the  equation .sx   

To  replace  the  missing  solution  we  observe  that  the  full  equation  for  z1  is  actually    , or  sufficiently  ,   , which  has  the  first  integral    , ( 5  ) where  C  is  some  constant .sx   The  substitution  connection  y  , z  is  now    , ( 6  ) and  z1  is  annihilated  by  the  operator  on  the  right  , only  if  it  be  a  sum  of  powers   .sx   

If  such  as  z1  be  a  solution  of  ( 5  ) , then  , whether  C  be  zero  or  not  , the  last  of  the  above  powers  must  be  annihilated  by   .sx   Thus  &formula  must  have  had  a  factor    with    , which  is  contrary  to  our  hypothesis  about  q.  Consequently  the  complete  solution  of  ( 5  ) with    gives  , through  ( 6  ) , the  complete  solution  of  ( 4) .sx   Replacing    by    we  may  enunciate  this  result  as  follows :sx   

( 1  ) If  q  , r  , s  be  integers  such  q    r  and    have  no  factor    where  q    s    r  , then  the  complete  solution  of  the  equation    is  given  by    , where  z  is  the  complete  solution  of  the  equation    and  C  is  a  constant  other  than  zero .sx   

If    ,   themselves  have  each  a  factor  of  the  type    , where  q  , r  are  subject  to  the  conditions  stated  in  ( 1  ) for  q  , r  , then  we  can  once  again  apply  the  procedure  of  ( 1  ) to  diminish  the  order  of  the  equation  , adding  a  further  term  independent  of  z  on  the  right .sx   The  term    already  on  the  right  is  replaced  by  some  X  defined  by  an  equation   .sx   

In  general  it  is  enough  to  write   .sx   

But  , if  the  operator  on  the  left  contain  a  factor    , we  have  to  take  X  of  the  form   .sx   

We  can  clearly  proceed  in  this  way  so  long  as    ,   possess  corresponding  factors  of  the  requisite  forms .sx   

We  can  replace  ( 1  ) by  the  more  general  theorem :sx   

( 2  ) If  q  , r  , s  be  integers  such  that    have  no  factor    where    , and  if  p  be  any  positive  integer  , then  the  complete  solution  of  the  equation    is  given  by    , where  z  is  the  complete  solution  of  the  equation    and  C  is  a  constant  other  than  zero .sx   

We  suppress  the  proof  , since  no  new  principle  arises .sx   

There  is  a  theorem  , complementary  to  ( 1) :sx   

( 3  ) If  r  is  a  positive  integer  and  z  any  solution  of  the  equation    , then    is  a  solution  of  the  equation   .sx   

In  this  case  the  order  of  the  z-equation  exceeds  by  unity  that  of  the  y-equation  , and  one  solution  z  must  be  annihilated  by  the  operator   .sx   We  see  from  the  form  of  the  z-equation  that  it  has  a  finite  solution  led  by    and  terminated  by   .sx   This  is  evidently  the  solution  annihilated .sx   

In  the  same  way  that  ( 2  ) is  the  generalization  of  ( 1  ) , so  we  have  the  generalization  of  ( 3) :sx   

( 4  ) If  r  , p  are  positive  integers  and  z  is  any  solution  of  the  equation    , then    is  a  solution  of  the  equation   .sx   

3 .sx   Equations  soluble  by  elementary  functions   .sx

In  analogy  with  I  ( II  ) we  now  consider  the  equation    , where  the  order  of  the  -operators  on  each  side  is  equal  to  the  index  of  xm  and  the  differences    ,   are  all  integers  incongruent  to  modulus  m.  We  may  write  this  equation  more  conveniently  as    , ( 6  ) where  Nr  , are  integers  of  either  sign  or  zero  , and  a  , b  are  any  real  numbers .sx   By  I  ( 4  ) , this  is  satisfied  by  some    , where  z  is  a  solution  of    , ( 8  ) and  N  , N  are  the  algebraically  least  integers  of  the  sets  ( Nr  ) , ( Nr) .sx   By  I  ( 6  ) this  equation  is  satisfied  by  the  solution  of    , and  therefore  ( 8  ) has  the  m  independent  solutions   .sx   ( 9  )  .sx

Unless    be  a  negative  integer  or  zero  , the  expressions  ( 9  ) for  z  are  not  expressible  as  finite  sums  of  powers  of  x  and  therefore  are  not  annihilated  by  the  operator   .sx   We  can  thus  assert  that   .sx

( 5  ) In  general  , the  equation  ( 7  ) is  soluble  by  algebraic  functions .sx   

In  the  exception  case  , when    is  a  negative  integer  or  zero  , the  substitution    in  ( 7  ) and  a  cyclic  rearrangement  of  factors  on  the  right  make  it  enough  to  consider  , in  place  of  ( 7  ) , the  equation   .sx   ( 10  )  .sx

The  exceptional  case  is  now  that  in  which   .sx   If  we  write  the  right-hand  side  of  ( 10  ) in  the  form    and  apply  ( 1  ) , we  see  that  solutions  are  'missing'  only  when   .sx   

Let  us  divide  the  integers    into  two  groups  ( s  ) , ( t  ) , according  as   .sx   

By  ( 1  ) and  the  subsequent  discussion  it  appears  that  the  complete  solution  of  ( 10  ) is  given  by    , where  z  is  the  complete  solution  of   .sx   ( 11  )  .sx

No  logarithms  appear  , since  the  factors    that  have  been  removed  from  the  right  are  mutually  incongruent  to  modulus  m.   .sx

Operation  on  ( 11  ) with  the  operator    gives  the  equation    , that  is  to  say  ,  .sx   ( 12  )  .sx

No  terms  have  been  lost  on  the  right  , since   .sx   The  complete  solution  of  ( 12  ) thus  includes  the  complete  solution  of  ( 11) .sx   In  ( 12  ) the  'complementary  function  , is  a  sum  of  powers  of  x  ; a  'particular  integral'  evidently  introduces  , besides  powers  of  x  , the  logarithms    , where  m  =  1 .sx   Terms  involving  log  x  could  also  conceivably  appear  in  the  particular  integral .sx   In  this  case  z  must  include  a  term    which  gives  rise  to  a  term    on  the  right  of  ( 12) .sx   But  there  are  no  such  terms  , since  t  , m-s  are  never  congruent  to  modulus  m.  Thus  no  logarithms  appear  except  those  of    , and  the  complete  solution  of  ( 12  ) involves  at  most  these  and  a  finite  set  of  power  of  x.  The  same  is  true  of  ( 7  ) in  the  exceptional  case  , and  we  may  thus  sum  up  and  say   .sx

( 6  ) The  differential  equation    , where  Nr  , Nr  are  integers  of  either  sign  or  zero  , and  a  , b  are  any  real  numbers  , is  , in  general  , soluble  in  algebraic  functions  , and  in  any  case  is  soluble  in  algebraic  functions  together  with  the  logarithms    , where  m  =  1 .sx   


4 .sx   Legendre's  equation   .sx

In  illustration  we  may  apply  the  preceding  theory  to  a  discussion  of  Legendre's  equation    ( 13  ) in  the  case  in  which  n  is  a  positive  integer .sx   If  in  ( 2  ) we  write    ,   ,   ,   ,   ,   , we  obtain  the  solutions    ,   , recognizable  as  constant  multiples  of    ,  .sx   

Since  n  , -(n+1  ) are  of  opposite  parity  , Theorem  ( 6  ) is  applicable  to  Legendre's  equation  , which  , as  is  otherwise  known  , is  soluble  in  terms  of  algebraic  functions  together  with  the  logarithms   .sx   Various  formulae  expressing    more  or  less  explicitly  in  algebraic  and  logarithmic  parts  may  be  obtained  by  transforming  Legendre's  equation  into  the  symbolic  form    , ( 14  ) and  suitably  manipulating  the  fractional  operator .sx   

If  we  apply  ( 4  ) to  ( 13  ) , we  obtain  the  known  formula    , determining  the  numerical  factor  from  a  comparison  of  leading  coefficients .sx   

We  may  obtain  certain  less  usual  formlae  for  Pn  , Qn  , if  we  separate  the  cases  of  n  even  , n  odd .sx   Write  first  n=2p  and  in  ( 1  ) set    ,   ,   ,   ,  .sx   

We  then  have  the  complete  solution  of  Legendre's  equation  ,   , where  z  is  the  complete  solution  of  the  equation    , ( 15  ) and  so   .sx   ( 16  )  .sx

From  the  coefficient  of  C  we  obtain  , on  comparison  of  leading  coefficients  , the  formulae    , which  may  also  be  written  as    ,  .sx   

We  may  express  these  in  the  form  of  contour-integrals  as    ,   , where  in  both  integrals    is  a  contour  in  the  z-plane  which  encloses  z=x  but  z=0 .sx   These  contour-integrals  are  evidently  special  cases  of  those  which  represent  the  hypergeometric  function .sx   

Again  from  the  coefficient  of  C  in  ( 16  ) we  similarly  obtain    ,  .sx   

If  n  still  equals  2p  and  we  apply  ( 3  ) to  ( 13  ) , we  see  that  , if  y  is  any  solution  of  Legendre's  equation  of  order  2p  , then    ( 17  ) is  a  solution  of   .sx   ( 18  )  .sx

Since  ( 18  ) is  satisfied  by    and  P2p  is  annihilated  by  the  operator  on  the  right  of  ( 17  ) , we  have  , on  comparison  of  leading  coefficients  ,   , which  may  be  written  in  the  equivalent  forms    ,  .sx   

Corresponding  results  for    are    ,  .sx     ,   ,   ,  .sx   

In  similar  fashion  we  can  obtain  formulae  for  the  associated  Legendre  functions  by  considering  the  differential  equation    satisfied  by   .sx   


SOME  SELF-RECIPROCAL  FUNCTIONS   .sx

By  G.  N.  WATSON  ( Birmingham  )  .sx

Received  6  July  1931   .sx

1 .sx   THIS  NOTE  contains  a  few  minor  developments  of  a  recent  paper  entitled  'Self-reciprocal  functions'  by  Hardy  and  Titchmarsh .sx   The  problems  which  I  discuss  have  arisen  out  of  a  question  which  was  put  to  me  by  Mr.  E.  G.  Phillips  ; he  had  encountered  for  formula    , ( 1  ) where  p  is  the  greatest  integer  such  that    and  the  dash  indicates  that  the  first  term  of  the  sum  is  to  be  halved  ; and  his  question  was  whether  an  analogous  formula  existed  for  the  series  obtained  from  ( 1  ) by  changing    into   .sx   

In  this  paper  I  first  consider  the  more  general  series    ( 2  ) and  I  transform  it  into  an  infinite  series  which  is  a  natural  modification  of  the  sum  on  the  right  in  ( 1  ) ; it  is  then  a  trivial  matter  to  obtain  the  transformation  in  the  special  case    by  a  limiting  process .sx   





6   .sx