52  Mr.  L.  R.  Ingersoll  on  the  Faraday  and 
curve,  and  hence  need  be  only  approximately  determined. 
This  point  will  be  mentioned  again  in  connexion  with  certain 
o£  the  errors  to  be  considered. 
Errors  may  otherwise  be  conveniently  grouped  into  two 
general  classes  :  those  incidental  to  the  measurement  o£  the 
ratio  dl/l9  and  those  occurring  in  the  interpretation  o£  this 
result  as  rotation  for  radiations  of  a  definite  wave-length. 
Those  o£  the  first  class  will  in  the  main  be  somewhat  com- 
pensating, and  their  effect  may  be  best  observed  by  plotting 
from  the  above  table  dl/L  with  wave-length,  and  noting  how 
closely  the  points  follow  a  regular  curve.  Those  of  the 
second  class  may  be  systematic,  and  their  effect  is  more  diffi- 
cult to  estimate  directly. 
As  sources  of  errors  of  the  first  kind  may  be  mentioned : — 
Variations  in  brilliancy  of  the  Nernst  glower. 
Disturbing  effect  of  magnet  on  galvanometer. 
Bolometer  unsteadiness  and  drift. 
Changes  of  temperature  due  to  heating  of  magnet. 
Inaccurate    determination   of  the    sensibility  reduction 
ratios. 
Considering  them  brieflv  in  the  order  given, — the  variations 
of  the  glower,  as  explained  above,  were  small,  and  their  effect 
was  eliminated  by  taking  the  average  of  several  deflexions  in 
each  direction,  in  rapid  succession.  The  graphical  method 
of  computing  this  average  was  especially  useful  in  enabling 
one  to  detect  and  allow  for  such  accidental  variations. 
It  was  impossible  to  eliminate  entirely  the  effect  of  the 
magnet  on  the  galvanometer  even  with  the  most  careful 
shielding  and  with  the  aid  of  the  small  compensating  solenoid, 
because  of  the  unequal  heating  effects  in  the  different  coils. 
However,  by  frequently  adjusting  this  solenoid,  by  altering 
its  distance  from  the  galvanometer,  the  effect  might  be  made 
small  ;  and  by  reversing  the  galvanometer  connexions  and 
averaging  the  two  sets  of  readings,  it  was  practically 
eliminated. 
The  matter  of  bolometer  drift  was  rather  troublesome,  and 
would  seem  to  be  due  partly  to  the  high  sensibility  which 
was  demanded  of  the  instrument,  and  partly  to  the  fact  that 
the  method  of  observation  necessitated  frequent  changes  of 
balance.  However,  the  effect  of  drift  was  far  less  disturbing 
in  this  than  in  most  spectrobolometric  problems,  and,  like 
errors  due  to  the  variations  of  the  glower,  it  was  largelv 
eliminated  from  a  series  of  readings  by  taking  the  graphical 
mean. 
Heating  of  the  magnet  doubtless  gave  rise  to  slight  changes 
