of the middle band of the triplet in the blue, which does not appear shifted, but of this I am not sure for the photographs shew a trace of a washed-out band about midway between the two extreme bands of the triplet in addition to the stronger band which is more refrangible. With glycerol the continuous diffuse absorption also creeps down the spectrum as with alcohol.

In order to observe the effect of a crystallizable solvent other than water, some didymium acetate was prepared and dissolved in glacial acetic acid, and for comparison with it an aqueous solution of didymium nitrate was made of equal concentration. Plate 20 shews the photographs of their spectra. Comparing the absorptions directly by eye, the band in the red appeared stronger in the acetate and sensibly shifted to the less refrangible side, the feeble band in the orange also was shifted in the same direction, the strong group in the yellow considerably extended towards the red but its more refrangible edge not apparently shifted, doubtless because the widening of the bands compensated the shift which was visible in all the other bands of the acetate though they otherwise had the same general appearance as those of the nitrate. The shift and change of character produced by acetic acid was less than was produced by alcohol.

Didymium tartrate is very insoluble in water, but the compound produced by potassium hydrogen tartrate acting on didymium hydroxide dissolves in a solution of ammonia. The spectrum given by this solution is contrasted with that of an aqueous solution (not exactly of the same concentration), of didymium chloride in plate 23. With the exception of the group in the yellow, the less refrangible of the groups in the green, and the narrow band in the indigo, the bands seem all a good deal washed out. All the bands are shifted towards the red, and the apparent shift increases as the bands become more refrangible, but probably this appearance is the effect of the greater dispersion of the more refrangible rays.

I had no crystals of didymium salts sufficiently large to enable me to see how the diminished freedom of the molecules in the solid would modify the spectrum, but had a rod of fused borax coloured with didymium. This was made by mixing weighed quantities of didymium oxide and dried borax, fusing the mixture, and sucking the fused mass into a hot platinum tube. After cooling the rough ends were cut off and polished, and I was thus able to compare the spectrum given by a thickness of 25 mm. of this glass with that of an equivalent solution of didymium chloride. Photographs of these spectra are shewn in plate 21. They are somewhat marred by dust on the slit of the spectroscope, but this does not prevent a fair comparison. It will be seen that the modifications produced by the glass are on the whole similar in character to those produced by some of the liquid solvents. The strong group in the yellow is much expanded and the components of the group unequally shifted towards the red, the less refrangible of the groups in the green is shifted and its appearance modified for the same reason. The more refrangible bands are much washed out and their shifts appear very unequal. Nevertheless they appear to be still essentially the same bands modified as to their rates of vibration by the diminished freedom of the molecules producing them.

VOL. XVIII.

40

On a review of the whole series of observations I conclude that the characteristic absorptions of didymium compounds, namely those which are common to dilute aqueous solutions, and are only modified by concentration, by heat, and by variations of the solvent, are due to molecules which are identical in all cases, though their vibrations are modified by their relations to other molecules surrounding them. The like conclusion holds for erbium compounds. It appears to me quite incredible that the atoms of didymium should retain in chemical combination so much individuality and freedom as to take up their own peculiar vibrations unaffected by the rest of the matter combined with them, as must be the case if we supposed the combined didymium in the molecules to give the common spectrum of all the salts in dilute solution. When I speak of atoms of didymium in the salts, I mean of course masses equal to the atoms of didymium metal, but having different energy, which means different internal motions, probably different structure, and different capabilities of vibration. No chemical compounds shew the absorptions which their separate elements exhibit. Sodium vapour, though monatomic, has a very strong absorbent power which is quite lost when it has parted with energy in combining with chlorine. Nevertheless the molecule of a chloride breaks up, in general, into masses equal to those of the atoms of its elements more easily than in any other way, and there is pretty good evidence that in encountering a molecule of water this also is sometimes broken up, and ultimately, if not immediately, new molecules of hydroxide and acid are formed, as well as, by a similar process, new molecules of the salt. In the interval between the rupture of a molecule and the recombination of its parts with each other, or with parts of other molecules, the parts have a certain freedom, and capability of vibrating, which they do not possess in combination. Now if we suppose the number of such parts as have the capability of taking up vibrations of frequency corresponding to the characteristic absorptions of didymium to be directly proportional to the concentration of the didymium salt and to the time of their freedom, the observed facts will be all in agreement with the hypothesis. Increased concentration, and increased temperature, will mean more frequent encounters amongst the molecules, and more frequent ruptures, but at the same time more frequent encounters of the parts and consequent shortening of their times of freedom. These effects will exactly compensate each other and leave the average number of absorbent parts of molecules constant under changes either of concentration or of temperature. The continuous absorption of the more rapid vibrations increasing with concentration and rise of temperature points to an action depending only on the number of encounters of the molecules of the salt with one another. It is not every encounter which is attended with disruption, and the continuous absorption may be due to molecules in encounter without rupture, but at all events it seems due to the condition of the molecules during encounter, but not to occur at the encounters of a molecule of salt with the very much less massive molecules of water. Encounters of a molecule of salt with a molecule of acid will in all probability cause effects very similar to those of encounters between two molecules of salt, and this supposition is quite in agreement with the observed facts.

The time of complete freedom of a vibrating part of a molecule must be very

short, but probably shorter when the complementary part is more massive, as in the case of a nitrate, than it is in the case of a chloride. But between complete freedom and complete incorporation in a chemical compound there is a considerable gradation, and the capacity of the part to vibrate at particular rates will have a corresponding gradation, and the part may moreover be frequently under the influence of molecules, or parts of molecules, with which it does not combine. This influence will probably be greater as the molecule exerting the influence is greater whether more massive, or, as in the case of such colloids as alcohol, more voluminous. These considerations reconcile all the facts as to the spectra I have observed with the hypothesis I have made.

There are, however, other facts to be reconciled with that hypothesis. I mean the facts of ionization, of osmotic pressure and the correlative facts of the rise of boiling point, and fall of crystallizing point, of solutions. In regard to all these effects the freedom of the parts is the primary postulate, far more definitely so than in the case of vibrations such as my observations relate to. The laws I have tried to investigate appear to hold good up to the point of saturation of the solutions, which is not the case with the laws of osmotic pressure and of change of boiling and freezing points, which have been established for dilute solutions. Further, ionization implies a certain distribution of energy in the field, the ions are charged with electricity. That is not necessary for the absorption of light, which will depend, primarily at least, on the form of the internal energy of the vibrating mass, that is on its structure. That a redistribution of energy occurs at every rupture of a molecule seems certain, solution is attended with thermal effects and so is dilution, and it is only when equilibrium is reached, and as much change takes place in one direction as in the opposite, that the manifestation of such redistribution ceases. How much of the intrinsic energy of the molecules takes the form of heat and how much is retained in the field at the rupture of the molecules we do not know. It is however quite conceivable that the circumstances under which the rupture takes place may determine whether any, or how much, energy is retained by the field, that is whether any, or how many, of the ruptured parts become ions.

The plates, which are all reproductions of photographs, will be found at the end of the volume.

XV.

The Echelon Spectroscope. By Professor A. A. MICHELSON, Sc.D.

[Received 19 October 1899.]

THE important discovery of Zeeman of the influence of a magnetic field upon the radiations of an approximately homogeneous source shows more clearly than any other fact the great advantage of the highest attainable dispersion and resolving power in the spectroscopes employed in such observations.

If we consider that in the great majority of cases the separation of the component lines produced by the magnetic field is of the order of a twentieth to a fiftieth of the distance between the sodium lines, it will be readily admitted that if the structure of the components themselves is more or less complex, such structure would not be revealed by the most powerful spectroscopes of the ordinary type.

In the case of the grating spectroscope, besides the difficulty of obtaining sufficient resolving power, the intensity is so feeble that only the brighter spectral lines can be observed, and even these must be augmented by using powerful discharges-which usually have the effect of masking the structure to be investigated.

Some years ago I published a paper describing a method of analysis of approximately homogeneous radiations which depends upon the observation of the clearness of interference fringes produced by these radiations. A curve was drawn showing the change in clearness with increase in the difference of path of the two interfering pencils of light,—and it was shown that there is a fixed relation between such a "visibility curve" and the distribution of light in the corresponding spectrum-at least in the case of symmetrical lines*.

It is precisely in the examination of such minute variations as are observed in the Zeeman effect, that the advantages of this method appear,―for the observations are entirely free from instrumental errors; there is practically no limit to the resolving power; and there is plenty of light.

There is however the rather serious inconvenience that the examination of a single line requires a considerable time, often several minutes, and during this time the character of the radiations themselves may be changing.

Besides this, nothing can be determined regarding the nature of these radiations until

* In the case of asymmetrical lines another relation is necessary, and such is furnished by what may be called the "phase curve."

is complete, and analyzed either by calculation or by an equivalent

Notwithstanding these difficulties, it was possible to obtain a number of rather interesting results, such as the doubling or the tripling of the central line of Zeeman's triplet, and the resolution of the lateral lines into multiple lines; also the resolution of the majority of the spectral lines examined, into more or less complex groups; the observation of the effects of temperature and pressure on the width of the lines, etc.

so serious

It is none the less evident that the inconveniences of this process are that a return to the spectroscopic methods would be desirable if it were possible (1) to increase the resolving power of our gratings; (2) to concentrate all the light in one spectrum.

It is well known that the resolving power of a grating is measured by the product nm of the number of lines by the order of the spectrum. Attention has hitherto been confined almost exclusively to the first of these factors, and in the large six-inch grating of Prof. Rowland there are about one hundred thousand lines. It is possible that the limit in this direction has already been reached; for it appears that gratings ruled on the same engine, with but half as many lines, have almost the same resolving power as the larger ones. This must be due to the errors in spacing of the lines; and if this error could be overcome the resolving power could be augmented indefinitely.

In the hope of accomplishing something in this direction, together with Mr S. W. Stratton, I constructed a ruling engine in which I make use of the principle of the interferometer in order to correct the screw by means of light-waves from a homogeneous source. This instrument (only a small model of a larger one now under construction) has already furnished rather good gratings of two inches ruled surface, and it seems not unreasonable to hope for a twelve-inch grating with almost theoretically accurate rulings.

As regards the second factor-the order of the spectrum observed, but little use is made of orders higher than the fourth, chiefly on account of the faintness of the light. It is true that occasionally a grating is ruled which gives exceptionally bright spectra of the second or third order, and such gratings are as valuable as they are rare, for it appears that this quality of throwing an excess of light in a particular spectrum is due to the character of the ruling diamond which cannot be determined except by the unsatisfactory process of trial and error.

If it were desirable to proceed otherwise to attempt to produce rulings which

FIG. 1.

should throw the greater part of the incident light in a given spectrum, we should try to give the rulings the form shown in section in Fig. 1.