558 A N A L Y T I C A L C H E M I S T R Y

Ackci and Frediani, ISD. Esc,. ( ' H E M . , ANAL. ED., 17, 793 (1945). Aepli and McCarter,Ibid., 17,316 (1945). Almy, Griffin, and Wilcox, I / , i d . , 12, 392 (1940). Am. SOC. Testing Materials, Conmiittee D-2, "Standards uf

Petroleum Products and Lubricants," Philadelphia, 1948. Boeke. J., Phi l l ips Tech. Rea., 9, S o . 1, 13 (1947). Fischer, Karl, Angew. Chcm., 48, 304 (1935). Gester, C . G., Cheni. Eng. Progress, 43, 117 (1947). Graefe, E., J . SOC. Chem. Ind., 25, 1035 (1906). Gremeko, B., S m o s t i T t khn ik i , 6, 43 (1938). Griswold and Kasch, I i id . E i i g . C'hem., 34, 804 (1942). Groschuff. E., 2. Elektrochon.. 17, 348 (1911). Hachmuth, K. H., WerterrL(;ria. 8, 55 (1931). Johansson, A , , Svensk Pappo 'a t idn , , 50, 11B, 124 (1947). Larsen, R. G., ISD. E x . <'HEX., .%SAL. ED., 10, 195 (193s~. Levin, Uhrig, and Roberts, Ihid., 17, 212 (1945). McKinney and Hall, Ib id . , 15, 460 (1943). hlitchell and Smith, "Aqunnieti~y," p. 59, Xew York I t i r r t -

Rising and Hicks, J . A m . Cliein. S u e . , 48, 1929 (1926). Roberts and Fraser, J . Suc. Cheni. Ind., 29, 197 (1910). Smith and Bryant, J . A m . Chrm. Soc., 57,841 (1935). Smith, Bryant, and Mitchell, Ib id . , 61, 2407 (1939). Swann, 14, H., IXD. ESG. CHEM., ANAL. ED., 18, 799 (19lti I .

Tarasenkov and Poloshintzeva, Ber., 65B, 186 (1932). Taubmann, A, Z. anal. Chem., 74, 161 (1928). Toennies and Elliott, J . Ani. Chem. SOC., 59, 902 (1937). Weaver, E. R., Ib id . , 36, 2462 (1914). Wernimont and Hopkinson, ISD. ENG. CHEM., ASAL. ED., 15,

Zerban, F. W., Ibid. , 18, 138 ( 1 <Mi),

'

science Publishers, 1948.

272 (1943).

used for several years in the authors' laboratory with very satisfactory results in the following special applications:

Determination of water in stocks of high vapor pressure such as propane and the butanes.

Determination of water in colored stocks such as lubricating oils, transformer oils, etc., in which the color bodies are not solubla in the glycol.

Studies of the water solubility-temperature relationship for hydrocarbons over the temperature range 60" to 180" F. There seems no reason to doubt that the upper temperature may not be further increased.

SUMMARY

-4 modified Karl Fischer method for determining the water content of hydrocarbons and petroleum fractions involves ex- traction of the water from the hydrocarbon by dry ethylene glycol and subsequent titration of the glycol with Fischer reagent. With one extraction, over 90% of the watc,i present in the hydro- carbon is absorbed by the glycol.

Increased accuracy with stocks of low water content mav be obtained by concentration of the water from a large volume of hydrocarbon in a small volume of extract. The difficulty of titra- tion in a two-phase liquid is eliminated. The method is applic- able to high vapor pressure stocks such as liquefied petroleum gases. Colored stocks, such as lubricating oils, transformer oils, etc., in which the color bodies are not soluble in glycol, may be analyzed for water without difficulty. This method may be used to determine the solubility of water in hydrocarbons and petro- leum frartions at temperatures up to about 350" F.

Rapid Routine Calculation with Punched

RECEIVED July 15, 1940.

o f Multicomponent Mixtures Card Machines

ASCHEH OI'LEH Creut It estern Division, T h e Dow Chemirul Company , I'ittsbtrrg, G t l i f .

To facilitate rapid routine determination of ten-component samples, niulti- component anal) sis was performed on an infrared spectrometer. The formidahle task of calculating hundreds of samples was handled by punched card machines. Csing standard accounting office equipment, solutions were obtained (after contentional iniersion of the matrix of absorption coefficients) in a mean time of 3 minutes per sample. Limits to the accuracy of the method are discussed,

11E laborious calculation of tlic composition of multi- T component mixtures may be handled hg automatic coni- puting equipment v ith speed and economy. Such automatic equipment is frequently available in accounting installations in the form of I.B.M. punched card machines. In the method de- scribed below, advantage is taken of thc sprrd of these machines in performing routine calculations.

A multicomponent mixture may tie completely analyzed by measuring sufficient independent, additive properties, if these properties of the individual components are knovn. A set of simultaneous equations may be foimed from the latter and may be solved for the concentrations of the components. The caw where these properties are infrared absorbances has been treated by Brattain, Rassmussen, and Cravath ( 2 ) , Fry, Nusbaum, and Randall (6) , and others. There is some loss in accuracy by assuming linear relationship between concrntration and ab- sorbence when practical slit widths are employed. Khen greater accuracy is not required, rapid routine multicomponent analysis may be set up as follows:

1. Obtain the coefficients of the linear equations, UUZI +

crl,.ra+. . . . a l , , ~ , L = k l , vilierc citch coefficient ai , is t,he slopc, o f the absorbence us. concentration curve at wave length Xi. Ki is the total measured absorbance a t Xi .

2. Invert the matrix of cocficients, using either a desk cal- culator ( S , 9 ) or an analog computer (1).

3. Solve each set of inverse equations by an arithmetical substitution into the inverse matrix.

In conjunction with a process development study, a large number of multicomponent samples were submitted for infrared analysis. Preliminary survey showed that only ten compounds \\.ere likely to appear in t,he reaction products. Although few samples contained all ten, evrry sample was treated as a ten- component mixture, in order t o eliminate the handling of each sample as an individual problrni. The over-all procedure u.sccl for each sample was as follo~vs:

One milliliter of sample was diluted to yolume with carbon disulfide in a 10-ml. volumetric flask.

Using the turret pin mechanism of a Beckman 1R-2 spectrometer and a predetermined slit, program, absorbences wert? measured at ten selected wave lengths. Samples were measured in a 0.1-mm. cell.

After correcting absorhrnres for false energy (Z), thc usual

1.

2.

3.

V O L U M E 2 2 , NO. 4, A P R I L 1 9 5 0

wll-plus-solvent blank was subtracted from each reading. The \ample absorbences were recorded on specially devised sheets.

Data for 25 samples were arcumulated into a “set” arid wr!t to the tabulating department.

The calculation and the preparation of the final report w r e handled by punched card machines as described below. The final report (plus check shccts) was retuined to the research lnboratoi y.

USE OF PUNCEIEL) c m n MACHIYES

4.

a.

Gcmeral principles of punched cwd computation h a w bccn tiiwusscd by Eckert ( 4 , 5 ) and King (8). The folloning Tritcr- national Business Machines Cor poi at ion equipment was svailablr~ foi the computations:

1. 2. Reproducing punch No. 519.

Keypunch, verifier, and csrri-counting sorter.

559

Accounting machine KO. 405 (standard equipment, p,lus two digit selectors, card cycle total transfer and group indication c,liniination).

These machines, which represent a basic accounting group, \yere used for the calculation. I t was recognized that the usc of a calculating punch and a collator would expedite the operation, but these machines were not available.

Detailed description of the machine mt.thods and wiring diagrams are available on request from the author. “Pointer 461” ( 7 ) may also be consulted.

The sums of small groups of products can best be obtained on an addition-type machine by a process known as “tligiting.” Although this process seems long and involved, the rapid opera- tion of punched card machines more than compensates for the apparent clumsiness of the mcthod. This process is carried out in three steps, each perfoi nied automatically.

1. Each multiplic~aiitf in the group is entered into three ctounters controlled by thr multiplier digits as follows: The iilultiplier digits are factored successively into sums of f 1, +3, and + 5 . The presenfie or absence of each factor impulses three counters labeled “1, “3,” and “5” to add, subtract, or skip the corresponding mu1 tiplicand.

3.

4 general description of the method is given below.

Example.

‘ I l ” “3”

347 X 6 + 289 X 7 = 4105 Counter

“5”

- 289 280 289 7 = -1 + 3 + t 5

347 347 6 = 1 + 5

2. The three counters ale totaled. (‘1’’ i ‘3’ (‘5”

58 589 636

3. The corresponding totals are added, respectively, 1, 3, and 5 times, then added together.

Euample. 58 X 1 = 58 289 X 3 = 867 636 X 5 = 3180

Total 4105

4. When the multiplier consists of more than one digit, this process must be performed for each digit. The corresponding sums of products are multiplied by 10, 100, 1000, etc., and then added together.

In practice, the operation is performed by the tabulating de- partment using a simple routine. Absorbence data are punched into cards, whichare merged with a permanent set of cards contain- ing the inverse matrix coefficients. The merged cards are passed

t h r o u g h the repro-

INFRARED DATA SHEET

J b 5 6 ? 8 g 3 4 5 6 7 8 9

Data by Key Punched by .d- Verified by ///1~-

Figure 1. Form Used to Transmit Absorbence Data to Tabulating Department

Fourth line in upper left-hand box would be keypunched 211030690. 2110 in sample number, 3 is a code for wave length, and 0690 represents

an absorbance of 0.690

- ---OOND--1 C O M P O U N D 2 x T 3 8 - - 2 9 0 1 C O M P O U N D 3 3 C O M P O U N D 4 a 0 3 8 2 9 0 1 C O M P O U N D 5 C O M P O U M D 6 1 0 0 1 a o 38 2 9 0 1 C O M P O U N D 7 1 C O M P O U N D 8 2 0 3 8 2 9 0 1 C O M P O U N D 9 C O U P O U N D 1 0 1 a o 3 8 __

2 9 0 2 C O M P O U N D 1 2 4 1 C O M P O U N D 2 __ 5

2 9 0 2 C O M P O U N D 5 4 0 C O M P O U N D 6 3 7 7 2 0 3 8 7 F v m r u - T T - 6 3 a ----fb- 2 9 0 8 C O M P O U N D 9 1 9 C O M P O U N D 1 0 5 2 0 3 8

2 9 0 3 C O M P O U N D 3 - 1 7 1 C O M P O U N D 4 2 0 2 0 3 8

2 9 0 3 C O M P O U N D 7- 7 c o u P o u N D 8 4 7 2 0 3 8 2 9 0 3 C O M P O U N D 9 1 5 c o M P o u l l o 1 0 * 6 2 0 38

d ~ U U N U J A * I cuw?-van-u * 4 4

__

& Y U J C U n N l J I 6 J > c u r r & U i V T ~ 6 Y 6 U J O

- 9 0 3 C U M P O U N D 5 > > CU- b 4 u 4 1 u 38

--

2 9 0 4 C O M P O U N D 1 1 1 1 C O M P O U N D 2 3 2 0 3 8 2 9 0 4 C O M P O U N D 3 1 2 4 C O M P O U N D 4 1 6 2 0 3 8 2 9 0 4 C O Y P O U N D 5 1 6 C O M P O U N D 6 5 7 3 2 0 3 8 a 9 0 4 C O M P O U N D 7 2 C O M P O U N D 8 2 4 2 0 3 8 2 9 0 4 C O M P O U N D 9 8 C O M P O U N D 1 0 2 a 0 3 8

Figure 2. Form Printed by I.B.R.I. Machine as Final Calculations Are Being Made Sample number is at extreme left, followed by two double columns listing compounds and their corresponding volume per cent. Final t w o columns are checks which indicate correct machine operation. “Sample 2901” is a check substitution of part

of original data into inverse matrix

(Blank column represents decimal point, since machines do not print deeimal points.)

- ducer, yielding a new set containing both the absorbence and the matrix element o n t h e s a m e c a r d which is processed as above. T h e f i n a l step, multiplication by 1, 3, and 5, is combined w i t h t h e p r e p a r a t i o n of the f i n a l r e p o r t a s follows : Previously prepared “compound name cards” as well as cards to control the proper repetitive transfer of totals are m e r g e d w i t h cards summarizing the rc- sults of step 2 above.

560 A N A L Y T I C A L C H E M I S T R Y

Samples of the submitted data sheet and the final report ale shonn in Figures 1 and 2.

SOURCES AND COWTROL OF ERROR

In a complex multicomponent analysis and calculation, sources of error may be both numerous and cumulative. Therefore, it is essential to recognize every source of error. In the analysis dis- cussed above, errors may be classified as those of measurement, calculation, and theory. Sources of measurement errors in- clude dilution and spectrometry. When instrument tempera- ture was maintained constant to 0.5” C., no significant wave- length shifts were observed. The instrument maker’s cIaim of less than 0.5% transmittance error was well substantiated.

In the computation, human error may be involved in the key- punching of the data on cards. Use of the verifier should greatly reducc the probability of undetected error. Once the data are piope111 punched machine errors may be minimized by in- corpoi ating numei ous automatic safeguards. The pessimistic assumption is made that the machines will make all possible e n ors The operation incorporates check methods which are printed on the final report. Thus, the machines will signal an e11 0 1 0 1 fail to print the proper check values in the event of a mis- placwi or missing card, incorrect operation, machine part failure, etc. .1 final proof consists of the substitution of a column of the calihration matrix for the absorbence data. The results appear

0. There is a small systematic error of about 0 .17 , which is the cumulative effect of dropping rather than rounding off the last significant figure in successive steps.

By far the largest error is that introduced by the neglect of cuivature in the absorbence-concentration curve. Although it is possible to perform second-order corrections ( 2 ) using the I B.M. machines, the increased machine time outweighs the inci eased accuracy for most process study programs. The effect of curva- ture was minimized by measuring all samples in dilute solution. The nominal concentration thus averaged 1 to 3 7 for any one component, When the compounds that evidenced the gi eatest

curvature viere high in concentration, accuracy decreased as evidenced by a departure from 10091, total concentration.

CONCLUSIO\S

Use of automatic computing equipment reduced calculation time to 3 minutes from a minimum of 15 minutes by desk cal- culator or 11 minutes by analog computer. The 3 minutes represents combined operator and machine time per sample w\.hen calculated in groups of 25 samples. The machines require operator attention less than half of the time. The cost of materials consists primarily of 10 cents‘ worth of cards per sample computed.

In a 6-month period, 750 ten-component analyses and compu- tations were performed without an undue strain on the spectroa- copy group. The rapid accumulation of an array of analytical figures greatly facilitated the n-ork of the process study group.

ACKNOWLEDG\IER[T

The wi te r is indebted t o David J. Pye for his interest and criticism of this work. Thanks are due to R. E. Clement of the International Business Machines Corporation for his cooperation. The permission of The Dow Chemical Company to publish this material is gratefully acknowledged.

LITERATURE CITED

(1) Beiry, C. E., Wilcox, D. E., Rock, S. II., and Washburn, H. IT - , J . davlied Phus.. 17. 262 (1946).

Brattain: R. R.,kassmussen. R. ’S,, and Cravath, d. 51.. Ihid. ,

Crout, P., Trans. Am. Inst. Elec. Eiigr.?., 60, 1235 (1941). Eckert, IT. J., J . Chem. Education, 24, 54 (1947). Eckert, W. J., “Punched Card Methods in Scientific Computa-

Fry, D. L., Xusbaum, R. E., and Randall, H. hI., J . d p p l i r d

International Business Machines Corp., Sew I’ork, “Poiriter

King, G. TT., J . Chsm. Ediacution, 24, GI (1947). TTaugh, F. V., and Dwyer, P. S.. Ann. M n t h . Stat., 16, 259 (19453.

~ E I V E D September 1, 1919.

14, 418 (1943).

tion.” New York, Columbia University Press, 1946.

Phys. , 17, 150 (1946).

461.”

Rapid Photometric Determination of Iron in High Temperature Alloys

MICHAEL STEVENS PEP1 Fairchild Engine and Airplane Corporation, Farmingdale, Long Island, S. Y .

A rapid method for the photometric determination of iron in high temperature alloys is based on the reaction of the ferrous ion with 1,lO-phenanthroline. The sample is dissolved in aqua regia and a few drops of hydrofluoric acid, and diluted to 500 ml. A 10-ml. aliquot sample is reduced with hydroxylamine hydro- chloride and the orange colored complex is formed by the addition of 1,lO- phenanthroline. The reproducibility is good and the average error is *0.02% of the amount present.

HE chief sources for the work reported in this paper are T Fortune and hiellon’s complete report on the spectrophoto- metric method for the determination of iron with 1,lO-phenan- throline (2), and the miter’s work on the determination of iron in aluminum alloys ( 4 ) . EXPERIMENTAL WORK

velopment of a rapid and accurate method for the determination of iron in high temperature alloys based upon the orange colored complex formed by ferrous iron and 1,lO-phenanthroline.

A number of papers ( 1 , S , 5 , 7 ) have reported on the use of 1 , l O - phenanthroline for the determination of iron in a variety of des . A summary of the work done with 1,lO-phenanthroline up

The preliminary work covering the follotving problems is fully described by Fortune and Mellon ( 2 ) and the writer ( 4 ) .

io 1944 was made by Smith and Richter (6). A suitable reductant for iron. A 17‘ solution of hydroxylamine The purpose of the work described in this paper was the de- hydrochloride was reported as the most effective reductant.