and accurately the same in diameter-about 2.8 cm. Each cylinder has its ends ground as flat and parallel as possible and at right angles with its axis. The members of each pair of cylinders have very closely the same weight, and the combined weight of the bismuths differs from the combined weight of the zincs by a fraction of a milligram only. The weight of each pair is 304.541 + grms. Of course, the zinc cylinders are considerably higher than the bismuths on account of their lower specific gravity.

It is essential that the axes of the cylinders be equi-distant from, and parallel with, the axis of the member b. This adjustment is first approximately made by means of a jig in which all the parts are placed. At the left of the figure one of the cylinders is shown in section, which also shows the rod c ground wedge-shaped for a considerable part of its length within the cylinder. Into the wedgeshaped spaces thus left long thin wedges of soft wood are driven, the upper or lower one more than the other, until the axis of the cylinder is parallel with the axis, of b. The wedges also serve to fasten the cylinder tightly onto the rod. After the final adjustments the protruding parts of the wedges are broken off. The weight of the wedges is small, and being virtually the same for each cylinder, is negligible.

The most important adjustment of all is the radii of gyration, because an error here of only 0.03 mm. in both radii would double or obliterate the whole difference found in the behavior of the zinc and bismuth, though this difference, as will be seen, is very considerable. The aim is to make the double radius as accurately as possible

the same always when the zincs and bismuths are exchanged. This adjustment is made by gently tapping with a very small hammer one or both projecting ends of the rod c, or driving either cylinder, as indicated, further onto its rod with a block of soft wood until the desired over-all dimension is obtained, indicating a standard (though much smaller) double radius. In making this adjustment for the following experiments the most patient care was exercised, and a micrometer caliper of high precision was used.

The driving clock had been running about two months since winding when the experiments hereafter detailed were made, yet the amplitude of the new pendulums was much greater than necessary to unlock the escapement. One tooth of the "scape-wheel" was marked, and this came round to a certain definite position every five minutes by the face hands of the clock, indicating 40 gyrations of the pendulum. Thus the clock was used as an accurate and dependable counter of pendulum gyrations. A good watch was kept always in the same position beside the clock to measure elapsed time! and always wound at the beginning of an experiment. Room temperature remained nearly constant throughout the experiments. A glass hood always covered the clock to keep out air drafts. Runs of 22 to 24 hours were usually made, to average up irregularities in watch and clock rates. Total elapsed time in seconds was divided by the counted number of gyrations to get the periods in seconds of a single gyration.

Many preliminary experiments were made to detect lack of homogeneity in any cylinder, if such existed. To this end each cylinder was separately turned about-its inner element made the outer-or its upper end the lower. In no case was there any observable change of period. The matter of unequal air resistance of the two pairs of cylinders was gone into quite extensively. Of several attempts to equalize this, the following was found the most trustworthy. A pair of short hollow cylinders of thin firm paper, closed at one end, and of such diameter as to slip snugly on the upper ends of the bismuth cylinders were prepared. These cylinders or caps were pushed just far enough down on the bismuth cylinders. to make them equal the zinc cylinders in height, plus thickness of

PROC. AMER. PHIL. SOC. VOL. LX, E, DEC. 20, 1921.

the paper cylinder ends. The performance of the bismuth cylinders thus equipped was carefully compared with that of the zinc cylinders wearing the same paper caps pushed down to contact. Of course, the paper caps, because of their weight, slightly increased the periods of both zinc and bismuth cylinders. But, within the limits of experimental error, it increased them equally, thus showing no effect from the equalization of air resistances. Hence the paper caps, apparently serving no useful purpose, were not used in the following experiments.

Each

Table III. embodies the results of a series of experiments with the zinc and bismuth cylinders on the torsion pendulum. period is computed from a run of 22 to 24 hours. All the observed periods are given-not those selected from a larger number.

All bracketed periods were obtained with one setting of the cylinders. After one period or set of periods was observed with either pair of cylinders, these were removed from the pendulum frame and the other pair substituted. This involved each time a new and very careful adjustment of the radii of gyration as before explained.

It is seen that the bismuth pendulum had the shorter period by one part in 1333. This is a very gratifying confirmation in kind of the earlier gravity pendulum findings.

It is also extremely interesting to note that the weight-mass difference between bismuth and zinc appears to be many times. greater (350001333) when the mass is measured by the accelerating force of a steel spring than when it is measured by the accelerating force of gravity as in the former pendulum experiments. This is very suggestive and will be discussed in a future paper. Of course, these relative values are only rough first approximations; but their difference is very much too large to be attributed to experimental

errors.

More torsion-pendulum experiments are in progress, and all, thus far, confirm the above findings. A pair of pure silver cylinders are being prepared to compare both with the zinc and with the bismuth

ones.

Although the three distinctly different lines of experiment detailed herein loyally support each other, I hesitate to draw general conclusions from their (to me) astonishing results until after much further experimentation for additional data, and much more time for reflection.

I am greatly indebted to Charles F. Brush, Jr., for efficient aid in preparing the apparatus described, in reducing observation data and in making computations.

CLEVELAND, April, 1921.

ROSE ATOLL, AMERICAN SAMOA.

BY ALFRED GOLDSBOROUGH MAYOR.

(Read April 22, 1921.)

His Excellency the late Commander Warren Jay Terhune, U. S. N., then Governor of Samoa, was so kind as to invite me to accompany him on the U. S. S. Fortune to visit the little known Rose Atoll in S. Lat. 14° 32', W. Long. 168° 12', and we spent twenty-four hours upon this island from June 5 to 6, 1920. There has been no scientific account of the island since 1839.

The island is an atoll, the lagoon being encircled by a narrow ring of limestone composed chiefly of lithothamnium, which is everywhere nearly awash at low tide, excepting on the northeast side, where there is a narrow entrance about six to nine feet in depth, out of which a current constantly flows. The ring of limestone which surrounds the lagoon is quite uniformly about 500 yards in width, while the central lagoon is about two miles wide and appears to have a maximum depth of not more than eight fathoms. There are only two small islets upon the atoll rim, Sand Islet and Rose Islet. The only map of the atoll is U. S. Hydrographic Chart of the Samoan Islands No. 90, based on the survey of the U. S. Exploring Expedition in 1839. This shows Rose Islet as occupying the entire width of the atoll rim, whereas at present it is confined to the inner half of the width of the reef rim. Moreover, this chart shows trees covering the entire area of the islet, whereas at present only the southern half of the islet bears trees. The chart states that Rose Islet is 33 feet high, but at present the land of the islet is 11 feet above high tide, and the tallest trees, as measured by means of a sextant, are about 80 feet high, and thus the total height of the landfall as seen from the ocean is about 90 feet.

Rose Islet is at present about 240 yards S.S.W.-N.N.E., and about 200 yards wide. The southern and southeastern half of the ilet is densely covered with a forest composed exclusively of