C. The Conception of a Limiting Film Thickness .sx The formation of an impervious and relatively insoluble film on the surface of a metal should ultimately lead to a decrease in the velocity of corrosion , since further corrosion could only take place as a result of diffusion through the film .sx The solubility of the film will presumably vary for different metals ; in certain cases it is conceivable that the solubility will decrease as corrosion proceeds , since the rain will probably remove the most soluble products and the less soluble compounds will accumulate on the metal .sx If the velocity of film formation exceeds the rate of solution in rain water , a stage will ultimately be reached when the film has thickened to such an extent that corrosion is practically at a standstill ; any further attack will then only occur by the solution of the outer layers of corrosion product , resulting in a temporary thinning of the film , which will render possible further diffusion of the air to the metal .sx Thus an equilibrium thickness of film will be produced in which the rate of solution of the outer layers in rain water is just balanced by the rebuilding of the film resulting from diffusion .sx In the case of a non-pervious and insoluble film , the velocity of both processes may approach zero ; this is probably the cause of the extremely long life of copper roofing .sx The existence of a limiting film thickness has already been demonstrated by Vernon in the case of indoor exposure for lead and aluminium and , in the case of aluminium exposed outdoors , Wilson has shown that the radial thickness of corrosion was the same after 23 years' exposure as after 8 years' , although during this period there was a further loss of metal .sx It is not suggested , however , that a limiting film thickness will be reached in the case of all metals , and alloys ; whether this is so or not will depend upon the properties of the corrosion product .sx If this hypothesis is correct even in a limited number of cases , it follows that the corrodibility of different metals can only be truly compared after the limiting film thickness has been attained in all the possible cases .sx ( This really leads back to the view already expressed that the corrosion time curves should be determined .sx ) The results given in Tables XXVIII .sx to XXXII .sx show that the weights of corrosion product adhering to the plate specimens at the various stations differed considerably for the same material .sx It is therefore probable that the limiting film thicknesses had not been reached after a year's exposure in the least corrosive atmospheres and doubtful whether this was the case in the most corrosive ones .sx It may also be remarked that the ratio of the maximum and minimum values of the weight of corrosion product adhering to the plates of the same material but at different stations was 7.3 for 80/20 nickel-chromium , 7.8 for nickel and 4.4 for 60/40 brass , as against 2.2 for arsenical copper , 2.5 for H.C. copper and 2.3 for lead .sx This fact points to the conclusion that the limiting film thickness will be reached more rapidly in the case of the latter materials than in that of the former , and agrees with the evidence as to the relative solubilities of the respective corrosion products .sx In cases where the metal becomes covered with a uniform film of corrosion product , e.g. , copper , arsenical copper , tin bronze , etc. , the .sx subsequent course of corrosion is probably largely determined by the perviousness and solubility of the film .sx This in turn depends upon the composition of the metal itself .sx The chief effect of a small amount of alloying element , such as 0.3 per cent .sx of arsenic or 1 per cent .sx of cadmium , in the case of copper , on the atmospheric corrosion of the basis metal would probably result from its influence on the composition and properties of the corrosion film .sx ( In the initial stages , the effect of the added element on the potential of the metal surface might come into play , but this would probably be negligible as soon as the surface was covered with corrosion product .sx ) It is also clear that in some cases it may take a considerable time for the effect of the small amount of alloyed element to become perceptible ; there may be a gradual accumulation of the added element in the corrosion film as a result either of differences in the solubility of the constituent compounds of the corrosion product or of a reaction between the uncorroded metal and the corrosion product , as in the case of brass .sx The effect of a small difference in composition may thus be cumulative ; in support of this view , some evidence has been obtained that arsenic accumulates in the corrosion product formed on arsenical copper .sx The corrosion product scraped off from the arsenical copper plates that had been exposed at Birmingham for a year , was found to contain 0.81 per cent .sx of arsenic and 45.2 per cent .sx of copper , i.e. , 1.78 parts of arsenic per 99.6 parts of copper .sx As the original arsenic content was 0.43 per cent .sx , the ratio of arsenic to copper is four times greater in the corrosion product than in the metal .sx It appears , therefore , that the arsenic does accumulate in the corrosion film , and it is possible that the superior resistance of arsenical copper to atmospheric corrosion , as compared with H.C. copper and judging by the results of loss in weight tests , may be due to this fact .sx This may also be the reason why the difference between arsenical copper and H.C. copper was much more pronounced in Vernon's experiments , at the end of 4 years , than in the present tests , which refer to one year's exposure .sx The arguments advanced in the preceding paragraphs are admittedly more or less hypothetical , but even if they are only approximately correct they serve to show that it is unwise to compare the relative resistances of different metals to corrosion , merely on the result of tests extending over a period of one year .sx The development of a protective film will , ceteris paribus , occur much more rapidly on metals yielding a relatively insoluble corrosion product , such as copper and bronze , than on those yielding a relatively soluble one , such as nickel and zinc .sx Consequently , the relative corrodibility of the different metals will apparently vary with the duration of exposure , and a true comparison could only be made after the corrosion time curve had been established in each case ; as already stated , this would require several years' exposure .sx For this reason , in discussing the results of the present field tests , it has been thought advisable to refrain from a too detailed comparison of the results for different metals and only the salient points have been mentioned .sx Comparisons of the behaviour of the materials , so far as was warranted by the experimental results , have already been made at the end of the sections describing the tests by each experimental method , so that it is hardly necessary to reiterate them here .sx There is , however , one pronounced case of a peculiar effect of local conditions .sx Whether judged by the weight-increment test , the resistance test , or the loss in weight test , the nickel specimens were inordinately corroded , relatively to the other materials , in the Wakefield atmosphere .sx As shown in the analyses given in Table XII .sx , the nickel corrosion product at this station consists largely of sulphate .sx The sulphur content of the atmospheric pollution at Wakefield is known to be high .sx If there is any connection between the two facts , it points to the conclusion that the presence of sulphur compounds in the atmosphere has a particularly harmful effect on the resistance of nickel to corrosion .sx D. Comparison of the Results Obtained by the Different Methods .sx The effects of corrosion are so diverse that no single experimental method can be expected to define them completely ; a much clearer insight into the effects of corrosion can probably be obtained by contrasting the results of several types of test .sx With this object in view , a brief reference will be made to the average results for the five stations shown in Table XXXVI .sx The following seem to be the chief conclusions to be drawn from the inter-comparison of the experimental results :sx 1 .sx The rate of corrosion is , as a rule , appreciably greater out-doors than in a Stevenson screen .sx The ratio of the values obtained under the two conditions is much higher for the copper-rich alloys than for the copper-nickel series , the actual figures based on the mean results for the five stations being 5.6 for arsenical copper , 5.8 for H.C. copper , and 6.3 for cadmium-copper , as against 2.8 for 80/20 copper-nickel and 80/20 nickel-chromium , and 2.4 for 70/30 nickel-copper and for nickel itself .sx ( See the last column of Table XXXVI .sx ) This difference in ratio is concomitant with the fact that the nickel alloys are relatively much more resistant to corrosion under conditions of complete exposure than when exposed in a Stevenson screen .sx The probable explanation is that the relative deliquescence of the corrosion products is not such a potent factor under the former conditions , where the rain tends to remove the corrosion product periodically and , moreover , falls indiscriminately on all the specimens .sx 2 .sx The close agreement between the electrical resistance tests and the determinations of the decrease in the breaking load on nine of the exposed materials , shows that , in these cases , there is very little pitting or inter-crystalline corrosion .sx The two brasses , which give anomalous results , suffer " dezincification , " their behaviour will be discussed in one of the following sections .sx The absence of pitting is confirmed in the case of copper by the fact that the increase in resistance of H.C. copper wires on exposure has been found to be inversely proportional to the diameter .sx 3 .sx It will be seen from Table XXXVI .sx that there is an appreciable difference in the corresponding values for the average thicknesses of the corroded layer deduced from the resistance tests and the loss in weight tests , the former being much the greater .sx Moreover , as shown in the last column of this Table , the ratio of the two results varies from one material to another , and the order of corrodibility differs somewhat in the results of the two tests .sx Taking an average value , the results for the loss in weight .sx tests are approximately half the corresponding results for the increase in resistance tests .sx The explanation of this is probably to be found in one or more of the following ways :sx The leeward side of a solid plate is much more sheltered from the wind than that of a stretched coil of thin wire .sx That the amount of air circulating over a specimen may have an appreciable effect on the corrosion has been demonstrated by the laboratory experiments of Vernon , by the positional effect in a Stevenson screen and by the results of some exposure tests on specimens exposed on the roof of the Royal School of Mines , in which it was found that the specimens nearest to the edge of the roof were much more corroded than the others .sx Greatest corrosion is , of course , associated with greatest circulation of air .sx The loss in weight test takes no account of any slight penetration of the metal that may occur .sx This point will be clear from Fig. 16 .sx It is probable that some slight penetration of the corrosion product takes place along the crystal boundaries , if the corrosion film is at all adherent .sx This penetration might affect the resistance appreciably but the loss in weight test would take no account of it , since the scraping of the specimen is discontinued as soon as there is a risk of abrading metal .sx As already mentioned , Vernon has shown that the loss in weight is consistently less than the total corrosion ( as measured by a gravimetric method ) ; in his experiments the ratio of the two was 0.89 for copper , 0.67 for 70/30 brass , 0.57 for 60/40 brass and 0.91 for zinc .sx It has also been found that when resistance and weight-increment determinations are made simultaneously on the same specimens exposed in a Stevenson screen , the observed resistance change is generally some 50 per cent .sx greater than that calculated from the weight of metal corroded .sx