Kerr  Effects  in  the  Infra-red  Spectrum.  51 
general  this  angle  was  made  45°.  When  ready  to  take  a 
set  of  readings,  the  small  compensating-coil  was  adjusted  in 
position  till  the  galvanometer-zero  was  unchanged  by  the 
action  of  the  magnet. 
Because  of  the  greater  freedom  from  magnetic  as  well  as 
mechanical  vibrations,  the  observations  were  all  made  at 
night.  With  the  spectrometer  set  at  the  desired  wave-length, 
the  total  energy  I  was  first  measured  with  reduced  sensibility. 
Then,  with  the  slit  still  open,  the  sensibility  was  increased 
step  by  step,  the  bolometer  at  the  same  time  being  balanced 
to  keep  the  reading  on  the  scale.  When  the  magnet  was 
excited  the  galvanometer  showed  a  deflexion,  indicating 
either  an  increase  or  decrease  of  intensity  dl.  The  magnet- 
current  was  then  reversed,  the  new  deflexion  observed,  and 
the  operation  repeated  so  as  to  give  usually  five  readings  in 
either  direction.  The  galvanometer  connexions  were  then 
reversed,  and  the  process  repeated;  for  in  this  way,  it  will  be 
observed,  any  small  outstanding  effect  due  to  the  direct  action  of 
the  magnet  on  the  galvanometer  would  be  eliminated,  while 
at  the  same  time  a  second  set  of  readings  was  obtained  which 
.served  to  form  a  very  good  check  on  the  first.  An  occasional 
set  of  readings  with  the  carbon-bisulphide  cell  removed 
served  to  show  if  any  other  cause  was  also  acting  to  produce 
rotation. 
The  mean  difference  of  the  galvanometer  readings  when 
the  magnet-field  was  direct,  and  when  it  was  reversed,  gave 
2  dl,  and  the  first  measurement  of  the  total  intensity  gave  I, 
and  their  ratio,  on  the  basis  of  the  same  sensibility,  deter- 
mined the  angle  of  rotation.  The  method  of  computing  the 
mean  d\  was  mainly  graphical  and  need  not  be  described  in 
detail,  but  it  will  serve  to  remark  that  it  was  designed  to 
render  as  ineffective  as  possible  errors  due  to  bolometer 
drift  and  variations  of  the  source,  as  well  as  to  eliminate 
completely  any  chance  of  personal  bias  as  to  any  one  or 
more  observations. 
Sources  of  Error. 
It  is  to  be  particularly  noted  at  the  outset,  that  onlv  those 
errors  are  of  first  importance  which  are  selective  in  action, 
and  hence  affect  the  determinations  for  different  wave-lengths 
unequally.  For,  while  it  is  of  course  desirable  to  know  the 
absolute  magnitudes  o£  the  rotations,  for  convenience  in 
checking  the  results  by  oilier  methods,  it'  the  error  can  be 
shown  to  be  a  constant  one  which  applies  equally  to  all 
wave-lengths  it  can  have  no  effect  on  the  correctness  of  the 
results   aimed    at,   t.  c.    the   form  of   the   rotatory  dispersion 
E  2 
