Kerr  Effects  in  the  Jnfra-red  Spectrum.  53 
of  field-strength,  although  determinations  with  a  small  test- 
coil  showed  the  effect  to  be  small.  But  any  possibility  of 
this  error  being  selective  was  removed  by  so  choosing  the 
succession  of  wave-lengths  to  be  tested,  that  the  rotation  for 
the  same  or  neighbouring  parts  of  the  spectrum  might  be 
measured  at  different  times  under  very  different  heating  con- 
ditions. The  temperature  of  the  carbon  bisulphide  was  kept 
constant  to  within  one  or  two  degrees  by  a  water-jacket 
around  the  containing  tube,  hence  the  error  from  this  source 
would  be  less  than  one  per  cent. 
Sensibility  ratios  were  determined  by  measurements  of  the 
same  intensity  of  radiation  with  the  two  sensibilities,  the  ratio 
of  which  was  desired.  Although  the  discrepancy  between 
successive  determinations  made  in  this  manner  was  some- 
times as  much  as  three  or  four  per  cent.,  so  many  tests  were 
made  that  the  probable  error  of  the  accepted  values  was  very 
much  smaller. 
Judging  from  the  agreement  in  different  results,  for  the 
same  or  neighbouring  wave-lengths,  the  cumulative  effect  of 
the  errors  of  this  class  is  to  produce  an  uncertainty  of  probably 
less  than  two  or  three  per  cent,  over  the  greater  range  of 
spectrum  studied,  although  at  the  ends  it  is  considerably 
increased.  To  be  certain  that  this  rotation  was  not  in  part 
due  to  some  other  cause  than  the  Faraday  effect  in  the 
carbon -bisulphide,  the  cell  containing  the  latter  was  removed 
and  a  number  of  tests  made,  which,  as  may  be  seen  from  the 
table  of  results,  showed  that  such  spurious  rotation  was  too 
small  to  be  definitely  detected.  The  rotation  due  to  the  thin 
glass  ends  of  the  cell  was  entirely  negligible,  being  only  about 
one-fifth  of  one  per  cent,  of  the  other. 
As  causes  of  the  second  sort  of  errors  may  be  mentioned: — 
Inability  of  the  pile  of  plates  to  polarize  completely. 
The  use  of  a  non-parallel  beam  of  light. 
Impurity  of  spectrum  due  to  stray  light  and  to  the  use 
of  a  broad  bolometer  strip  and  slit. 
Regarding  the  first  source  of  error,  it  is  evident  that  the 
light  would  be  incident  on  the  polarizer-plates  over  quite  a 
range  of  angle,  because  of  the  appreciable  size,  as  well  as 
nearness  of  the  glower,  which  was  only  about  10  ems.  distant: 
hence  complete  polarization  could  not  be  expected  in  the 
reflected  light.  Then,  since  the  formula  used  to  reduce  the 
quantity  d\j\  to  a  rotation  through  a  em-tain  angle  was 
deduced  on  the  supposition  of  completely  polarized  light,  the 
results  obtained  with  it  must  be  subject  to  a  correction  which 
will  now  be  calculated. 
