110 G. H. KNIBbS. 
+206, °103, 069 and ‘052. It is difficult to rely upon measure- 
ments of F to less than say ‘05 mm., corresponding to an error of 
from 1 in 600 to 1 in 4000, according to the length of the contact 
piece. Again if the director be set to 0-2 mm. as mentioned in 
§ 6, the prism is set to + ‘030, so that from this cause there is an 
uncertainty of about the same order as that arising from defective 
measurement of the test-piece. And further the distance of the 
scale from the mirror, viz. Z, cannot easily be fixed to within 
0-25 mm., nor can the right angle condition be very perfectly 
satisfied, so that here again an absolute error of the order of about 
one four-thousandth also exists. 
It is evident from what has been educed that this Extensometer 
will give extremely accurate results so far as relative extensions 
or compressions are concerned, but the relation of these to the 
absolute length of the test-piece is not of the same order of accuracy. 
By lengthening the handle-bars, the setting of the prisms could be 
made more exact, and by making the grooves less deep and finish- 
ing them more carefully the value of # could be ascertained with 
greater accuracy. In this way, and by fixing the relations of 
parts of the apparatus which require to be relatively adjusted, 
allowing only for small adjusting mevements, the absolute results 
can be made more nearly comparable in precision to the differential 
ones, in which a very high order of accuracy has been already 
reached, thanks to the very ingenious contrivance of Professor 
Martens, and the excellent workmanship of Herr Béhme. 
One point remains to be noticed, viz., that the scale need not 
necessarily be parallel in two planes to the test-piece ; hence after 
defining the plane in which it is important it should be carefully 
adjusted, [ have said that it should be made merely ‘“ approxi- 
mately parallel to the test-piece,” see (c) this section. Even were 
the scale at right angles thereto, it is evident that the mirror 
rotation would be similarly registered. "When however the vari- 
ation caused by this rotation in the length of Z, and the errors 
due to the shifts of the test-piece, are considered, it will be see™ 
that the position indicated is the best. 
