APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

PROJECT WHIRLWIND

SUMMARY REPORT NO. 29 FIRST QUARTER 1952

Submitted to the

OFFICE OF NAVAL RESEARCH Under Contract N5ori60 Project NR 048-097

and the

UNITED STATES AIR FORCE Under Contract AF19(l22)-458

DIGITAL COMPUTER LABORATORY MASSACHUSETTS INSTITUTE OF TECHNOLOGY Cambridge 39, Massachusetts APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

TABLE OF CONTENTS

FOREWORD QUARTERLY REVIEW (AND ABSTRACT) SYSTEM ENGINEERING . 1 Storage Reliability and Checking . 2 Input-Output 3 Operational Control Center CIRCUITS AND COMPONENTS . 1 Vacuum Tubes . 2 Component Replacements in WWI . 3 Ferromagnetic and Ferroelectric Cores .4 Transistors , ELECTROSTATIC STORAGE . 1 Tube Program INPUT-OUTPUT . 1 Magnetic Tape . 2 Magnetic Drums MATHEMATICS, CODING, AND APPLICATIONS . 1 Problems Being Solved . 2 Subroutines Completed

ACADEMIC PROGRAM IN AUTOMATIC COMPUTATION AND NUMERICAL ANALYSIS . 1 Automatic Computation and Numerical Analysis . 2 Seminars on Computing Machine Methods APPENDIX . 1 Reports and Publications . 2 Professional Society Papers . 3 Visitors

e APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

1. QUARTERLY REVIEW tion of the 16x16 ceramic array has been (AND ABSTRACT) operated, but much work remains to be done. FOREWORD A multi-position ferroelectric switch has been developed which can perform many of the During the firstquarter of 1952 the Whirl­ switching tasks required in an information- wind I computer used one bank of storage handling system. A magnetic-materials group tubes with a density of 32x32 (1024) spots. has been formed for the purpose of develop­ Project Whirlwind The original bank of 16 x 16 tubes was remov­ ing ferromagnetic and ferroelectric materi­ ed from operation. The recently installed als for use in computer circuits. Project Whirlwind at the Massachusetts Institute of Technology Digital parity check has increased the reliability of A system has been set up to keep a com­ Computer Laboratory is sponsored by the Office of Naval Research under the storage section and has facilitated the plete record of each transistor being used in ContractN5ori60and the United States AirForce under Contract AF19( 122)- location of faults. About 85% of the scheduled development work at the Laboratory in order 458. The objectives of the Project are the development of an electronic applications time was useful during this peri­ digital computer of large capacity and very high speed, and its application to study the effects of various parameters to problems in mathematics, science, engineering, simulation, andcontrol. od. affecting transi stnrs as computer components. At the present time Project resources are about equally divided between As integration and testing of its various A two-transistor flip-flop has operated e {])°P ration Qf the computer and improvement of its reliability ;(2) applica­ components approach completion, the com - reliably at prf's up to 1 megacycle, and a tions of the computer to engineering and scientific problems; (3) storage puter becomes available for increased use by promising transistor gate circuit has been research and development; and (4) design of additional terminal facilities. the applications group. Section 6 of this re­ designed. port describes a number of engineering and The 256-spot storage tubes-removed from scientific problems to which the computer is The Whirlwind Computers the system, all of which were operating sat­ being applied. This work is providing valu­ isfactorily, had been in service an average of The Whirlwind computer is of the high-speed electronic digital type, in able experience in the development of tech­ 1635 hours. Four of the tubes had been used which quantities are represented as discrete numbers, and complex prob­ niques for the use of large-scale digital com­ for more than 3000 hours. The 17 tubes lems are solved by the repeated use of fundamental arithmetic and logical puting machines and the training of people to now operating at a density of 1024 spots have (i.e., control or selection) operations. Computations are executed by frac­ operate them. In addition, useful solutions' been in service an average of 1000 hours. tional-microsecond pulses in electronic circuits, of which the principal are being obtained. ones are (1) the flip-flop, a circuit containing two vacuum tubes so connected Two factors that have caused some of the that one tube or the other is conducting, but not both; (2) the gate or coin­ Failures of vacuum tubes were at a new tubes to have narrow margins have been cidence circuit; (3) the electrostatic storage tube, which uses an electron low during this quarter. Research on the re­ eliminated, so that the yield of good tubes is beam for storing digits as positive or negative charges on a storage surface. liability of tubes has been at a low ebb during expected to improve over the 65% during this Whirlwind 1 (WWI) may be regarded as a prototype from which other this period because of the heavy demands for quarter. Research toward the improvement computers will be evolved. It is being used both for a study of circuit tech­ acceptance tests of production tubes required of electrostatic storage tubes continues. niques and for the study of digital computer applications and problems. for new input-output equipment. Other com­ Whirlwind I uses numbers of 16 binary digits (equivalent to about 5 The interim magnetic-tape input-output decimal digits). This length was selected tolimit the machine to a practical ponents, particularly the new glass-envelope system is in operation in the computer. Cir­ size, but it permits the computation of many simulation problems. Calcu­ crystals, also showed increased reliability. cuits for the final magnetic-tape system have lations requiring greater number length are handled by the use of multiple- The 16x16 metallic magnetic-memory all been designed, and some units have been length numbers . Rapid-access electrostatic storage initially had a capacity array has been running with read-write times constructed. of 4096 binary digits, sufficient for some actual problemsand for prelimin­ reduced to 8-16 microseconds. An 8x8 por­ ary investigations inmost fields of interest. This capacity is being gradual­ ly increased toward the design figure of 32,768 digits. Present speed of the computer is 20,000 single-address operations per second, equivalent to about 6000 multiplications per second. This speed is higher than general scientific computationdernands at the present state of the art, but is needed for control and simulation studies.

Reports Quarterly reports are issued to maintain a supply of up-to-date infor­ mation of the status of the Project. Detailed information on technical aspects of the Whirlwind program maybe found in theR-,E-,and M-series reports and memorandums that are issued to cover the work as it pro­ gresses. Of these, the R-series are the most formal, the M-series the least. A list of the publications issued during the period covered by this Summary, together with instructions for obtaining copies of them,appears in the Appendix.

— APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

u 2. SYSTEM ENGINEERING

2. SYSTEM ENGINEERING total storage available. Therefore Bank A was removed from operation. This reduced the required storage-tube maintenance time and made more computer time available for The computer has been operating during experimental studies of storage-tube opera­ this quarter with one bank of storage tubes tion. Tests made in an effort to determine using a density of 32x32 (1024) registers. optimum rewrite times resulted in a better The computer has shown a general reliability adjustment of the parameters involved in the of 85% useful application time during the rewrite operation. There is, however, still period. more experimental work which should be done to complete the study of this problem. 2. 1 STORAGE RELIABILITY AND Under certain operating conditions, evi­ CHECKING dence of a small deflection shift in the storage tubes was found. Special programs devised Electrostatic storage has been function­ for investigating this fault resulted in the ing with the parity-check system outlined in detection of a plate-current shift in the 715C Summary Report 26. The parity system has amplifier tubes in the electrostatic-storage the advantage that it stops the computer im­ deflection output panels. Individual tests on mediately when an error occurs in information these tubes revealed that current shift existed read from storage . Without the parity check, only in the older tubes. (See Section 4.12 for a program may continue to operate using false further details.) This deflection trouble was information and thus produce erroneous re­ cured by replacing the bad tubes. However, sults, or it may try to perform an unallowed a further deflection shift that was then un­ operation as a result of a previous error in covered was found to be caused by an ion stored information. space charge in the body of the storage tube. The parity check has been very valuable The holding-gun operating voltages were re­ for two major reasons: adjusted, and this second deflection shift re­ (1) The check has uncovered various duced to a low value. problems in electrostatic storage and has Recently constructed storage tubes have been a great aid in locating these faults. The large operating margins, which provide the fact that the computer is stopped immediate­ required high reliability of operation with ly by a parity-check alarm prevents subse­ substantially less maintenance. Daily auto­ quent operations from masking the failure matic marginal checking has been extended symptoms. to the storage section and has been quite valu­ Fig. 2-1. Servicing of electrostatic-storage section. (2) In the event of a failure of one of the able in keeping it in good operating condition. storage tubes, the parity digit column may Fig. 2-1 shows this section (right center) be used as a replacement for the failing digit being serviced; a small section of the arith­ 2. 3 OPERATIONAL CONTROL CENTER in its flexibility: new methods of operation column by rerouting a few cables; operation metic element appears on the left. can be installed simply by adding new test of the computer (but then without the parity The operational-control-center equipment equipment or by changing interconnections of check) is thus made possible until the failing of the Whirlwind computer is located in a the presently used test equipment. storage tube can be serviced. series of 6-footx 19 -inch racks; it consists During the experimental stages of Whirl­ The use of the parity check with the in­ 2.2 INPUT-OUTPUT largely of various types of standard test wind operation, this control center was en­ creased variety of application programs has equipment. This control center (Fig. 2-2) tirely adequate and fulfilled all the desired shown up several weaknesses in the storage Tests of an interim system of magnetic- is used during the operation of the computer requirements. The increaseduse of the com­ section during this quarter. These difficulties tape terminal equipment in conjunction with by the applications groups; it is also used as puter by the applications groups, however, have been investigated and many of the weak­ the computer have shown that adequate relia­ a center for trouble-shooting. The equip­ has indicated that the addition of a control nesses eliminated by the use of improved bility can be obtained. While a final system ment in the control center is designed so that desk would be desirable. A newly designed storage-tube lineup procedures. New test is being designed, the interim system will special test signals can be generated for in­ control desk will include all of the frequently programs have been written to investigate remain connected to the computer. Applica­ sertion in the computer; synchronizing sig­ used push-buttons and switches for control­ more fully methods for obtaining better focus tion groups are encouraged to use this equip­ nals for scopes are also available. The main ling the computer and will contain the paper of the writing and reading beam and for ob­ ment. advantage of this type of control center lies tape and printer terminal equipment. taining complete erasure of a stored spot, The plug-in unit display decoders have enlargedby spot-interaction effects. A much been operating very reliably during the past better understanding of storage-tube operation two months. These 2048-increment d-c undera variety of different modes has yield­ coupled decoders, which replaced the old ed muchfaster and more efficient lineup pro­ 256-increment breadboard decoders, have cedures . been quite satisfactory in all respects. The The availability of 1024 registers of stor­ resolution of displayed patterns is now ex­ age in Bank B resulted in a substantial reduc­ cellent even on the 16-inch scope, and the tion in the use of Bank A, whose 256 registers compact plug-in equipment has not caused represented only a small proportion of the any maintenance problems.

- APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

2. SYSTEM ENGINEERING 9

3. CIRCUITS AND COMPONENTS with t"'3 figure in mind, it is possible to interpret the data of Figure 3-1 and deter­ mine relative rates of failure for compari- 3. 1 VACUUM TUBES son with previous experience. With about 2100 7AD7 tubes in service, 3. 11 Vacuum-Tube Life 21 tubes failed. This number is the lowest of any quarter since the computer has been During the past quarter, the WWI com- in full operation. This rate of 1 percent per puter was operated for about 1000 hours. thousand hours should be compared to the

Reason for Failure; Number failed Total in Hours at Change in Type Service Failure Characteristics Mechanical Burn-out Gassy 0-1000 1 1000-2000 1 1 3000-4000 1 1 4000-5000 3 2 7AD7 2100 5000-6000 3" 6000-7000 2 8000-9000 4 2 9000-10000 2 7AK7 1700 7753 1 1000-2000 * 1 2000-3000 2 2C51 20 5000-6000 2 8000-9000 2 3D21A 3 915 1 3E29 160 5256 1 4000-5000 1 5U4G 24 6000-7000 1 1 7000-8000 1 300-400 1 500-600 1 6AS7 80 600-700 2 Fig. 2-2. Operational control center. 800-900 1 6L6G 80 6312 1 1000-2000 3 20 2000-3000 1 1 715 5000-6000 1 8000-9000 5 1 2000-3000 1 3000-4000 2 4000-5000 1 6SN7 400 5000-6000 1 6000-7000 2 8000-9000 2 1000-2000 1 2000-3000 1 ELC16J 12 3000-4000 1 5000-6000 1 6000-7000 1 VR150 47 6926 1 Fig. 3-1. Tube failures in WWI January 1 - March 31, 1952 APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

10 3. CIRCUITS AND COMPONENTS 3. CIRCUITS AND COMPONENTS 11

value of 3.4 percent per thousand hours com­ 3. 12 Vacuum-Tube Research Raytheon Manufacturing Co. as an aid to electric storage and switching. puted to August of 1951. their cathode-materials research program. Brief mention should be given to two other There may have been some actual im­ During the past quarter, life-test activi­ Information concerning test methods for in­ lines of endeavor. A master's thesis re­ provement in the life of 7AD7 tubes, but not ties have been at a rather low ebb. The only terface impedance has been supplied to a search is getting under way in the field of by a factor of 3. The probable reason for active life test at the moment involves some number of other organizations, including low-power pulse transformers; the aim is a the change is the modification of marginal- 6AS7G tests, which are described below. General Electric, Raytheon, Sylvania, and re-evaluation of pulse-transformer design in checking procedure made during the past However, it is expected that investigations of Superior Tube Co. the light of new materials and techniques. quarter. Flip-flops are no longer checked certain new and modified tubes will soon re­ The WWI pulse transformer was easily re­ individually, only as part of the computing quire renewed activity in this area. produced in a smaller and cheaper package as system. If this change is the cause of the During December 1951 it was found that 3.2 COMPONENT REPLACEMENTS a first step in the experimental work; a few reduced failure rate this quarter, the rate 6AS7G tubes were causing a great deal of IN WWI of these samples will go on life test in the can be expected to be somewhat higher later, trouble in some of the storage-tube control five-digit multiplier. Two theses are com­ even though it may not reach the old high circuits. The cathode coating was peeling Figure 3-2 lists the replacements of mencing which have to do with the applica­ level again. in tubes of one lot, causing grid-cathode components other than tubes during the first tion of magnetic cores to computer circuits shorts. These tubes were used in a type of quarter of 1952. The very low number of and systems other than the internal memory, The single 7AK7 failure is also the lowest such as flip-flops, counters, adders,etc. ; number for any quarter since full operation cathode-follower circuit, with a high-imped­ crystal failures confirms the prediction that the newer glass-envelope crystals would re­ this work will be stressed rather heavily was attained. However, in this case no change ance grid return, so that the grid-cathode from now on. has been made in the checking procedure. shorts caused spurious pulses across the sult in increased reliability. Statistical fluctuations may account for this cathode load, considerably disrupting opera­ tion of the storage tubes. As a result, sam­ Total in No. of Hours of low number. Component The 9 failures of 6SN7-GT tubes cor­ ples of all future lots of 6AS7G tubes will be Type Service Failures Operation Comments respond to a failure rate of 2. 25 percent per life-tested before use, to prevent a recur­ Mica thousand hours. This rate should be com­ rence of this rather serious difficulty. 0.001 3300 1 1210 Mechanical pared to the rate of 0. 9 percent computed Techniques for the measurement of cath­ mfd to the middle of 1951. The increase is pos­ ode interface impedance have been further refined. This work has been written up as Inverted sibly due in part to statistical fluctuations Can and also to more severe requirements of R-207, The Use of a Transconductance Bridge Capacitors (oil filled) 8 1 1280 Insulation breakdown circuits. in the Measurement of Cathode Interface Im­ 4 Other tube types are not used in suf­ pedance. mfd ficient numbers to make meaningful com­ A by-product of this work has been the parisons. discovery of a rapid change in the cathode Bath Tub Some difficulties were encountered dur­ interface impedance when the cathode current 1.0 75 1 7334 Oil leakage ing the past quarter with the 715B and 715C of an involved tube is switched on. It has mfd tubes used in the deflection amplifier for been common knowledge for several years D-357 4 drift Crystals 7500 5 4800-5012 electrostatic storage (ESDO). The trouble that the cathode interface impedance would (1N34A) 1 open found was a short-time instability, with a slowly decrease when current was drawn from time-constant near 1 second, when the se­ the cathode. This is the effect which causes Fig. 3-2. Failures of components in WWI lected point was changed from one location tubes operated Class A to have low values of January 1 - March 31, 1952 to another in a different part of the storage interface impedance. However, there is surface. This instability, which was trouble­ another important effect, which has not been some only with high-density storage of 1024 previously reported. In some tubes the in­ 3. 3 FERROMAGNETIC AND 3.31 16x16 Metallic Array points, was traced to small changes in the terface resistance will increase rapidly, by FERROELECTRIC CORES characteristics of the 715 tubes when the perhaps as much as a factor of 3, within the This memory array has continued to duty factor or plate current was changed. first minute after cathode current is switched Summary Reports 24 through 27 contained operate with fair reliability and margins. Only tubes with ages over 8000 hours dis­ on. discussions of a scheme for storing digital Improvements in some of the surrounding played this difficulty. Changes in plate cur­ Checking the interface resistance with information in a three-dimensional array of test equipment have made for a reduction in rent as high as 15 ma in 150 were found to pulses at a low duty factor will give the low magnetic cores and the research and develop­ minimum read-write time to 8 microseconds depend upon the previous operating condi­ initial value of interface resistance, while ment work which has demonstrated the feasi­ using a 3:1 selecting-current ratio and 16 tions of the tube in the seconds preceding the checking under small-signal Class A con­ bility of that scheme. Two 256-core memory microseconds using a 2:1 ratio. Further test. No interface impedance was found. ditions will give the larger value. The tran­ arrays were reported on; one had seen full work on the surrounding equipment continues . Marginal-checking procedures are available sition between the two conditions has not yet operation, the other partial operation. Some Whenever possible, the memory is run for to detect these changes, so that bad tubes been investigated at all completely; however, work on the use of ferroelectric slabs in the periods of several hours holding some arbi­ maybe changed before serious operating dif­ both the initial low resistance and the final dual of the ferromagnetic-core scheme was trary pattern of information; results continue ficulties are encountered in the future. high resistance can be observed using the also discussed. to be encouraging. Work on the punched-card record system transconductance-bridge technique mentioned The following paragraphs cover activity mentioned in SR26 and SR27 is progressing above if the cathode current is keyed. during the last quarter, including (1) experi­ rather well. The activation of this system Cooperation with other laboratories has mental operation of the metallic array, (2) 3. 32 16x16 Ceramic Array and Switch has been somewhat slower than expected, but continued. Additional tests have been made further partial results of work on the ceramic data will probably be available on major types as a part of the ASTM cathode committee array and matrix switch, (3) the expanding Portions of this array up to 8 x 8 in size by July 1. About 10,000 cards will be in­ program( see Summary Report 27. ) More program in new materials, testing, and meas­ have been in operation with fair reliability, volved. ASTM diodes have been analyzed for the urement, and (4) the research work on ferro- poor margins, and read-write times of about

1

i APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

12 CIRCUITS AND COMPONENTS 3. CIRCUITS AND COMPONENTS 13

3 microseconds. The array and the two 8- parameters involved in the switch, and its position magnetic-core matrix switches can coupling to the memory, are far from opti­ be seen in Fig. 3-3. A close-up of the memo­ mum, and present work, is directed toward ry cores is shown in Fig. 3-4. improving them. It is an encouraging sign Testing of single cores over a major for the memory that operation was at all portion of the memory array indicated a sig­ possible with such a large variation in co­ nificant, but probably tolerable, variation ordinate-current shapes and amplitudes. in core responses. Testing of a 2x2 square was then tried over a portion of the array and, finally, testing of a 4x4 square was 3. 33 New Materials, Testing, tried over the entire memory. With current and Measurement readjustments at each new position, it was possible to get the large majority of the cores A magnetic -materials group has been to handle information successfully. On the formed for the purpose of developing and basis of these tests, a favorable 8x8 region providing ferromagnetic and ferroelectric was chosen and, as indicated above, it was materials for use in computer circuits. One just barely possible to operate. function of the group is the improvement and Tests indicate that the major immediate evaluation of materials for magnetic-core obstruction to better operation is the unsym- circuits now being developed. Another func­ metrical and unequal current waveshapes tion is the study of ferromagnetic and ferro­ being put on the coordinate lines of the memo- electric materials to uncover characteristics ory by the magnetic -core matrix switch. The which suggest application in computer cir-

Fig. 3-4. Close-up of magnetic-memory array.

cuits. These functions call for close coopera­ for Insulation Research at MIT, under the tion with those engaged in circuit development direction of Professor A.R. von Hippel, is and also with outside groups engaged in the preparing to conduct more fundamental in­ experimental or theoretical investigation of vestigations for the improvement of ferrites. these materials. The improvement of ferrite The Laboratory will be able to synthesize cores for the ceramic memory array has re­ ferrites and to analyze and measure their ceived most attention during the lastquarter. composition and properties. The ceramics The research staff of the General Ceram­ group of the Laboratory for Insulation Re­ ics and Steatite Corporation, using empirical search recently produced its first samples techniques, has developed an improved square- of ferrite, a material very similar to the loop material for the memory. A hysteresis MF-666 made by General Ceramics. loop of the old MF-666, used in the first The evaluation of ferrite cores for the I6x 16 memory array (see Summary Report memory is done using a tester which operates 28), and a loop of the new material, MF-1118, a single core in much the same manner as it are shown in Fig. 3-5. Since the develop­ would be operated in a memory array. All ment of MF-1118, General Ceramics has the important parameters such as current been striving to produce a square-loop ma­ amplitude and rise time are variable, so that terial with lower coercive force. the optimum operating conditions for a given Fig. 3-3. Magnetic-core memory array and matrix switch. In order to supplement the empirical core can be determined. One tester is now work at General Ceramics, the Laboratory being used for the study of core operation, APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

1-1 3. CIRCUITS AND COMPONENTS 3. CIRCUITS AND COMPONENTS 15

3. 34 Ferroelectric Storage and Switching 3.41 The Transistor as a Component aid of the circuit of Fig. 3-7. Thin sheets of barium titanate ceramic At the end of 1951, preliminary defini­ are being pulse-tested in order to find ma­ tions of the transistor parameters were 9K "ELECTRA" CARBON FILM RESISTOR terials suitable for storage and switching adopted. These definitions and the methods applications. Materials have been found for measuring the parameters are contained which are on the borderline of suitability for in Engineering Note E-441.

memory-array operation and which promise The transistor equivalent circuit which SQUARE 50MFD fast operation. was adopted in E-441 was suggested by R. WAVE WV Adler.* It consists of two diodes (emitter GENERATOR A multi-position ferroelectric switch ELECTROLrTlC (essentially the dual of the magnetic-core and collector), a resistance (base), and a matrix switch described in Summary Report current generator (/ele) placed across the 27) has been developed (Fig. 3-6) which can collector diode. accomplish many of the switching tasks in an A transistor history log containing a Fig. 3-7. Circuit for study of transit time, information-handling system; in particular, history card for each transistor and a card dispersion, and hole storage effects. it can select among the rows and columns of for each manufacturer was devised. Each a ferroelectric memory array. The logical history card contains the collector charac­ A square pulse of current is forced into circuitry of the ferroelectric switch can be teristic family, a typical negative-resistance the germanium from the emitter contact, and painted directly onto the two sides of a thin curve, the point measurements, and the col­ the current wave at the collector is observed ferroelectric sheet. When compared with lector-current rise and fall times. The across a 220-ohm resistance. Input- and existing methods of switching, this new meth­ history of each transistor is recorded on the output-current waveforms are superimposed Fig. 3-5. Hysteresis loops of magnetic back of the card. On the manufacturer's card on the same photograph, so that the time memory materials. od shows great promise wherever size, weight, and cost are important factors. the point measurements on all of a certain relationships are apparent. The resultant Upper: MF-666, used in first 16x16 transistor type are summarized. This log photo appears as in Fig. 3-8. array. has been used since the middle of January. Lower: MF-1118, a newly developed Other groups (R. Rediker at MIT and Air INPUT CURRENT WAVE material. Force Cambridge Research Center) have 3.4 TRANSISTORS adopted the system so that a larger sample of the data can be obtained. while different parameters such as current The work in the first quarter of 1952 was amplitude, ambient temperature, pulse-repe­ divided into two phases: After the measurements had been made tition frequency, etc , are varied. A sec­ 1. A study of the transistor as a circuit on all of the transistors available at the time FOR SLOWEST TRANSISTORS .I'M) component. (February 5, 1952), the results were sum­ ond tester is being used at the research marized inE-447. At that time, five General laboratory of General Ceramics for evalua - 2. A study of computer circuits which r- INPUT WAVE employ the transistor as a circuit component. Electric GllA's were compared with five tion of new materials. Raytheon CK7l6's and one Bell 1698. To determine how the back resistance of the collector diode varied as a function of

collector voltage, plots of collector back re­ FOFt F.ST TB.NStSIO.S HT3«) sistance versus collector voltage were made for 30 transistors. These plots are contained Fig. 3-8. Input and output waveforms. in E-451. To determine what effect self-heating The first step(l) is due to the dropacross would have on the back resistance of the col­ the mutual resistance. The second step (2) lector diode, an experiment was performed results when the holes arrive at the collector. in which a steady back voltage was applied to The time between 1 and 2 is the transit time the collector, and the collector current was of the holes. Dispersion and variation of plotted as a function of time measured from hole path lengths cause step 2 to be smeared. the turn-on time. Data on ten transistors The fall-off time is due to hole storage ef­ was taken. From this data a plot of back re­ fects (i. e. , the time required for the holes sistance as a function of time was made (see to be cleared out of the germanium). For E-448). Changes in resistance during warm- the slow transistor such as the 1768, the up varied from 4% to 50% of the final value. output waveform appears on Fig. 3-8. The A device for making all the measure­ results of this experiment will be reported ments outlined in E-441 has been built. in E-455. Studies will be made of transit A study of transit time, dispersion, and time as a function of several circuit varia­ hole storage effects has been made with the tions including collector voltage, pulse ampli-

* Adler, Richard B. , Large Signal Equivalent Circuit for Transistor Static Characteristics. MIT, RLE Transistor Group T-2, August 30, 1951. (Revised October 2, 1951) Fig. 3-6. Ferroelectric multi-position switch. APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

16 3. CIRCUITS AND COMPONENTS 17 tude, and collector switching. sistor parameter values, except during the switching time. 4. ELECTROSTATIC STORAGE volts is felt to be necessary to insure a high Several diode gate circuits which can degree of reliability in storage. 3 42 Transistor Circuitry be operated by the voltages obtained from During this period, a total of 24 tubes the collector of a flip-flop stage have been 4. 1 TUBE PROGRAM were replaced in the computer. Of these, A study of memory circuits has been designed. At least one of these seems satis­ 15 had low margins, and 3 had lost emission made. The designs of these circuits have factory. A transistor gate using a regenera­ During this quarter, the first bank of 17 from one or both guns. The remaining 6 had been based on studies of the negative-re­ tive amplifier that has been designed shows storage tubes (Bank A) operating at 256 spots non-uniform writing characteristics caused sistance characteristic. Calculations of the promise of being more useful than the diode per tube was removed from service. When by buckling of the mica target, low secondary voltage-current characteristics of several gate. Bank B became available for reliable opera­ emission, or surface blemishes. The aver­ of the basic circuits which present this nega - Triggering of transistor circuits with tion at 1024 spots per tube, the bank of 16x16 age life of the tubes removed from Bank B of tive resistance are contained inMemorandum voltage sources was the subject of a short tubes, although operating properly, was used Whirlwind was 625 hours, ranging between M-1406, and an analysis treating the base- study. The results are reported in M-1430. so infrequently that further aging was unwar­ 10 and 1700 hours. Many of the failures lead­ stabilized transistoras a four-terminal non­ A master's thesis proposal (M-1401) to ranted. ing to this high replacement were revealed linear network is contained inM-1400. Sever­ build a 1-megacycle regenerative pulse am­ The yield of good 1024-spot tubes re­ by the parity check, which was put into ser­ al single-transistor flip-flops have been de­ plifier has been presented. As part of this mains lower than is desirable. However, the vice in Bank B during this period. This fea­ signed and built which would operate at prf's work, a regenerative pulse amplifier em­ two main factors causing this shrinkage have ture gives an odd-even check of every readout greater than 1 megacycle on alternating po­ ploying a 1734 transistor has been designed. been isolated and are under control, and a from storage. It has proved to be invaluable larity pulses and with rise times in the order This circuit will serve either as a positive significant improvement in acceptable tubes in locating the source of trouble'irnmediately of 0.05 to 0.1 microsecond, depending upon the or as a negative pulse amplifier, depending was evident during the last month of this instead of at some later time when the pro­ transistor. Unfortunately, the problem of upon the bias level at the emitter. It oper­ quarter. gram has stopped. The parity check has also reliably triggering the flip-flop with unidirec - tional pulses was not solved, so it was ates on a 0. 1-microsecond trigger pulse 2 helped to uncover a small deflection shift in abandoned in favor of a two-transistor flip- volts in amplitude and produces a 13-volt the storage tubes. This shift appears to be a flop. 0. 35-microsecond output pulse with a rise 4. 11 Tube Experience in Whirlwind I function of holding-gun cur rent and duty cycle. time of 0. 05 to 0. 1 microsecond. This cir­ The seriousness of this shift in giving faulty A two-transistor flip-flop that has been cuit operates satisfactorily at prf s of 1 mega­ When the bank of 256-spot tubes was re­ readouts has been alleviated by operating designed requires, with its complement cir­ cycle. moved from the computer, their average age with the holding-gun current set below 1 ma, cuit, 6 diodes in addition to the two transis­ The transistor counter proposed in M-13 53 was 1635 hours, with values ranging between while research efforts are directed toward tors.' It has operated at prf's as high as 1.5 is nearing completion. A two-stage counter 300 and 4000 hours. Four of the tubes had finding the best way to eliminate it. megacycle, but it operates reliably up to about has been built and operated unreliably up to been used for more than 3000 hours; only one 1 megacycle. Above this frequency the output about 500 kc. This counter employs a two- less than 750 hours. The 300-series tubes waveform becomes distorted because of the transistor flip-flop and diode gates. removedfrom Bank. Aare being kept on hand. 4. 12 Tube Production slow fall-off time. Rise times are in the Plans have been made to build an accumu­ In the event of an unexpected failure of the order of 0.1 microsecond. Fall-off times lator using Whirlwind block diagrams but present operating bank, these tubes could be A total of 43 new and 10 reprocessed are in the order of 0. 3 microsecond. Base substituting transistor circuitry. Two-tran­ returned to service in a few days. tubes were made during the first quarter of stabilization and load switching are both sistor flip-flops, regenerative pulse ampli­ The 17 tubes in Bank B(the parity-check 1952. Two tubes with Philips type "L" cath­ used, so that the voltage levels at the col­ fiers, and diode gates are to be used. digit described on page 6 of Summary Report odes are included in this group, as well as lectors are almost independent of the tran­ 26 has now been added) have now operated an 13 other research tubes intended for studies average of 1000 hours, ranging between 120 of the stannic oxide coating or certain as­ and 2080 hours. This average age of 1000 pects of 32x32 storage. A weekly summary hours represents a rather small increase of tube production is shown in Fig. 4-1. over the figure of 735 hours at the end of the Of the 38 tubes made for 32x32 opera­ last quarter. It is indicative of a large num­ tion in the computer, 14 were good, 14 were ber of replacements during this period. As marginal, and 10 were rejected. This gives described further under Section 4.12, a com­ an acceptance figure of 35% good tubes, which bination of two troubles caused many of the is considerably lower than the previous rate tubes to operate with rather narrow margins, for 16x16 tubes. However, of the bad and so that as soon as tubes were available with marginal tubes more than half were rejected wider margins they were immediately put because of a low secondary-emission ratio into service. (Such "margins" refer to the below 100 volts. In order to eliminate a con­ excursions which are possible in the ampli- tamination of the guns, a vacuum bakeout tube of the video gates applied to the tube procedure was added to the processing sched­ during writing.) For example, in an average ule. This bakeout reduced the secondary good tube, the amplitude of the gate applied emission of the surface to the extent that to the grid of the high-velocity gun for writing positive spots were not maintained under positive spots may be varied from about 20 holding-gun bombardment at the normal value to 60 volts without interfering with operation of 100 volts acceleration. The tubes were of the tube. For a marginal tube, this range, classedas marginal because a higher holding- or the range of the write-minus high-velocity gun voltage had to be used, which brought gun and/or signal-plate gates, would be re­ about a reduction in the useful operating mar­ duced to possibly 10 or 15 volts. A range in gins. This situation is more critical in the these three gate amplitudes of at least +20 400-series tubes than in the 300-series, APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

18 4. ELECTROSTATIC STORAGE 4. ELECTROSTATIC STORAGE 19

NUMBER OF TUBES

POOR SECONDARY EMIS5I0N indefinitely. An increase in the required off, or a special tube with an ion-collector RT-294 — RT-293-1 ~ RT-299-1 BELOW 100 VOLTS holding-gun voltage of about 10 volts, as the element of some type will be used. I- auxiliary collector is raised from 100 to 400 Preliminary results from the thesis stu­ r."-»»- ' «r-W '1«-Wi volts, cannot be accounted for simply on the dy of the current-density distribution in the basis of the number of ions reaching the sur­ fringe region of the high-velocity beam are face. *I \ 1 GUNS DID NOT ACTIVATE PROPERLY now available . The results indicate that high- 2 HIGH-VELOCITY GUN MISALIGNED Brief experiments have been made con­ energy secondary electrons, scattered from cerning a slight deflection shift as a function the auxiliary-collector screen, are responsi­ SURFACE BLEMISHE5 of holding-gun duty cycle. The results indi­ ble for the current in the region beyond rough­ cate that this shift might be related to the ly 100 mils radius from the center of the

1 POOR SECONDARY EMISSION BELOW 100 VOLTS ion density in the tube and the rate at which beam. The rate of spot growth beyond this 2 COLLECTOR-TO-AUXILIARY-COLLECTOR SHORT it is dispersed after the holding beam is turn­ point is primarily a function of the spacing ed off. With the auxiliary collector at its between auxiliary collector and surface; it is i POOR SECONDARY EMISSION BELOW IOC VOLTS normal potential of +400 volts, the entire greater for wide spacings. Furtherwork will m 2 COLLECTOfl-TO-THIRD -ANOOE SHORT body of the tube becomes a "trap" for positive include an approximate theoretical analysis CBmwz ^D ions. Further tests are planned. Either the of this problem and a consideration of the role 1 WEAK HIGH-VELOCITY GUN ST-503 ST-504 2 POOR SECONDARY EMISSION BELOW 100 VCLTS third anode of a standard tube will be gated of the collector screen in contributing to the negatively during the time the holding gun is fringe region by a similar mechanism.

1 POOR SECONDARY EMISSION BELOW 100 VOLTS K 2 AIR INCLUSIONS 1 GASSY 2 SURFACE BLEMISH AND POOR SECONDARY EMISSION BELOW 100 VOLTS 3 MICA BUCKLED

I POOR SFCONDARY EMISSION BELOW 100 VOLTS 2. NO CONNECTION TO THIRD ANODE 3 HIGH-VELOCITY GUN MISALIGNED

BUCKLED MICA

TT^; REPROCESSED AS A GUN RESEARCH TUBE

LEAD BROKEN DURING BASING

E RESEARCH TUBE } MARGINAL TUBE

Fig. 4-1. Storage tube production record. because until recently the 400-series tubes 4. 13 Research have had both thinner mica targets and smal­ ler collector-to-surface spacing. With these Research activities were reduced during dimensions it is possible for the mica target this quarter because of the higher demand to buckle out toward the collector screen for production testing. Not only were more under the electrostatic force of attraction be­ tubes tested, but the scope of the testing was tween the collector and a negative storage increased in anattempt to more nearlydupli- surface. We have gone back to wider col­ cate computer operation in the test procedure. lector spacings until a new target design is As an outgrowth of the tests on tubes hav­ made which will hold the target more rigidly. ing low secondary emission below 100 volts, Both of the tubes with Philips type "L" it was found that the potential of the auxiliary cathodes processed this period were good; they are presently on life test. collector has a definite effect upon the holding- gun voltage required to hold a positive spot APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

20 21

5. INPUT-OUTPUT A breadboard of this printing system was built up, and in preliminary tests the type­ writer printed characters that had been re­ 6. MATHEMATICS, CODING, with boundary conditions H(x0, t) * H(-x0,t) = corded on tape at a rate of over 100 charac­ AND APPLICATIONS Ho, H(x,0) - 0 (mentioned in Summary Re­ 5 1 MAGNETIC TAPE ters per second. Since the typewriters oper­ ports 24, 26, and 28) is being programmed. ate at a speed of 8 characters per second, The method, which has been described by The interim magnetic-tape system is this represents a saving of more than 90 per­ The group working on scientific and Hartree'-, consists of replacing the t-deriva- still in operation in the computer, and studies cent of the computer print-out time. engineering applications of the Whirlwind tive by a finite difference. The differential and improvements continue to be made on it. computer have three principal goals: (1) De­ equation then becomes Present plans are to begin using the interim velopment of a simple, effective organization making use of standard procedures and coded system regularly on applications problems 5.2 MAGNETIC DRUMS 82K(x,t) B(x.t) - B(x,t ) within the next month. The magnetic tape subroutines to facilitate preparation and —i i LLL . 0 execution of coded programs. (2) Training will be used to store the conversion program The construction of two magnetic-drum 9x' At in order to reduce the time required with systems for the Whirlwind Computer is pro­ and assisting interested and qualified people paper tape. The circuits for units to be used ceeding satisfactorily at Engineering Re­ (those who have problems to solve on the The partial differential equation is now re­ in the final magnetic-tape system, as de- search Associates, Inc., of St. Paul, Min­ computer and those MIT students who want duced to a system of ordinary differential scribed in Summary Report 28, have all been nesota. These systems, one of which is to to learn about computing machines) to set up equations which can then be solved by the designed, and some units have been con­ be used for auxiliary storage of programs their own coded programs. (3) Solution of various numerical techniques available for structed. and numbers and the other for buffer storage certain problems, each of which is typical ordinary differential equations (e.g., the In an effort to save some of the time that of input and output data, have been described of some class of problems and is therefore Runge-Kutta method). the computer spends in producing a lot of in some detail in Summary Report 28. Al­ valuable not only in its own right but also in Results obtained using this method will typewritten data, a small system is being de­ though delivery of the firstof these drums is broadening the experience of the program­ be used as a check on the accuracy of re­ signed that will permit a typewriter to be not expected until the last quarter of 1952, mers in the group. sults obtained using the 6-point difference operated by data on a magnetic tape indepen­ detailed planning for their installation is al­ Perhaps 40% of the time of the staff of equation described in Summary Reports 24 , dent of the computer. With this system the ready under way. Since a large amount of ten in the group is devoted to the first goal, 26, and 28. computer will record its output data on mag­ additional terminal equipment will be inte­ development of procedures, of which much netic tape, and then the tape can be trans­ grated with the computer at approximately has been said in recent Summary Reports. Problem #24. Matrices ferred to the printing system while the com­ the same time, careful studies are being Another 20% is devoted to the third goal, puter is doing something else. The time made of all engineering and shop schedules solution of problems, of which the magnetic This quarter has been used to retrace saving is not as much as might be hoped for, for the laboratory for the next twelve to eight­ flux density studies (#8) are typical. The some of the steps made during previous because blank, spaces must be run between een months to insure that adequate manpower remaining 40% is devoted to aiding others to quarters using (15, 0, 0) arithmetic in the characters on the tape in order to provide will be available. The planning for installa­ use the computer . Problems are being solved (24, 6, 0) number scheme. General packaged stopping and starting spaces needed during tion and the preparation for systems testing by MIT students in connection with MIT sub­ solutions have been prepared for arbitrary the typing operation. The tape has to come will be intensified in the near future so that jects 6. 535, 6. 537, and 6. 539. Thesis prob­ matrices using "direct method"solutions in­ to a stop after a character is read and it has lems are being done by several students. cluding Jordan and Gauss Elimination. The after delivery of the drum equipment it can Stiefel method of "conjugate gradients" is to accelerate to full speed before the first be quickly tied in with the computer. Appointees of the MIT Committee on Machine line of data of the next'character is reached. Methods of Computation are using the com­ also being programmed. Iterative packaged puter. In addition, students or staff repre­ schemes for symmetric positive definite senting 12 of the 19 different MIT professional matrices have been programmed and are now departments have used the machine during the under test. The Gauss (Southwell) Relaxation past quarter, as have several outside users. scheme and the Seidel iterative scheme are under test.

6. 1 PROBLEMS BEING SOLVED Problem #28. Ambipolar Diffusion In the following paragraphs some of the In connection with the study started some various problems now being actively solved time ago by Professor Allis of the MIT Phys - by the Whirlwind I computer are described ics Department (see Section 6. 10 of Summary and the status of the work is reported. Report 26), a floating-point (24, 6, 0) pro­ gram has been written during this past quar­ Problem #8. Magnetic Flux Density terly period and is in the process of being Study operated and tested. Briefly, the problem has to do with an electrical discharge in which A second method for the numerical solu­ electrons and positive ions are generated at tion of the non-linear partial differential a rate(r^) proportional to the electron con­ equation centration (77). In the steady state, both par­ ticles must flow to the walls of the discharge tubeatthe same rate (.F), whichis determined 6B by the concentration gradients of the elec- B = f(H) trons(v« )and positive ions(v )and by the 6x n

1 Hartree, D.R., Calculating Instruments and Machines, University of Illinois Press, 1949. APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

II 6. MATHEMATICS, CODING, AND APPLICATIONS 6. MATHEMATICS, CODING, AND APPLICATIONS

space charge field (Es). The field (Es) is, in verting decimal instructions typed on Flexo- turn, determined by the space charge writer equipment to the milling-machine code, The arrowheads indicate the direction of al­ basis. For the network of Fig. 6-1, for are being revised to simplify their use. The lowed passage of information. example, under the equi-probability assump­ wing template problem will be taken under As a first approximation to a theory of tion, the machine produced the table shown consideration again when this work is com­ the behavior of the network, Dr. Luce has below. e v pleted. propounded a first simple theory that passage It is apparent that the statistics of the of information is an equi-probable process, problem are consistent over each thousand with a member of the network having an equal tries. Similar results have been obtained for When the equation of continuity V- /""" - T\ ?7_is Problem #41. Binary Matrix Product chance of yielding his information to each of a total of fifteen networks. Each solution (for added, the result is four first order equations Statistics the allowed other members of the network. one particular network) took about twenty for the four variables Tj^TjFE. The boundary Sucha supposition was used in the Whirlwind minutes. conditions are that/"'and Es should vanish at This problem originated with R. D. Luce, the center of the discharge tube and 7? and 77 _ program, and the machine was instructed to If the results prove to give satisfactory + of the Group Networks Laboratory of the Re­ calculate the number of steps of passage of agreement with the basic theory, a test of a at the walls. By the introduction of dimen- search Laboratory of Electronics, MIT. The sionless variables, the number of arbitrary information before every member of the net­ more elaborate theory using conditional pro­ originally proposed problem has now been work got the information on this probabilistic babilities will be attempted. parameters is reduced and the task of finding solved, and plans are under way for an ex­ compatible values of electron and ion con­ tension of the problem and solution under a centration necessitates solving two second set of more complicated basic assumptions. Number of time Number in each order, second degree differential equations. This problem has provided the first Number in each Number in each steps before all category after category after category .after A three-point Newton-Cotes integration successful performance of a "Monte-Carlo" possess information 1000 tries formula is used. The one starting point is type of solution on Whirlwind. The machine 2000 tries 3000 tries found by a Taylor's series expansion. The is instructed to act successively in different 1 0 equations are integrated until either one or manners dependent on a sequence of random 0 0 2 0 0 0 the other vanishes {or until the independent numbers fed into it from an outside tape. The 3 6 variable is greater than its upper bound, in random numbers used were a translation into 16 24 which case the increment is changed positive­ machine binary form of the first 10,000 deci­ 4 205 417 628 ly). From the knowledge of which equation mal digits (ranging from 0 through 9, with 5 318 628 935 vanishes first, the sign of the increment of 8's and9's omitted)in the Kendall-Babbington- 6 238 473 705 the second equation is known. Then the sign of Smith table of random decimal digits. In 7 106 215 333 the "previous" increment is compared with their binary form, these become equivalent 8 69 139 206 the sign of the present one. If they differ, to a sequence of random binary digits stored 9 30 58 86 the increment is bisected; if they are the fifteen to a register. A special read-in pro­ 10 22 41 63 same, the increment is left unchanged. gram is used to feed in one random binary 10 6 13 20 Three parameters have been operated digit (0 or 1) to the Whirlwind accumulator. by the computer, and satisfactory data was Dependent on the value of this digit (or suc­ obtained. However, the remaining param­ cessive digits in pairs), the program is in­ eter, yet to be operated, will be subject to structed to perform certain operations. more difficult criteria which may necessitate The basic problem is as follows: cer­ changes in the method. tain experiments in group communications have been performed in the Group Networks Problem #42. Hyperbolic Partial Problem #30. Digitally-Controlled Laboratory. These consist of a step-by-step Problem #43. Generation of Random Differential Equations: Numerical Numbers Milling Machine Program passage of information among five test per­ Integration sonnel, occupying positions in a "network. " The servomechanically controlled milling Certain of the positions are allowed to pass A set of hyperbolic non-linear partial One attempt to devise a repeatable ran­ machine built by the MIT Servomechanisms information only to certain other ones. A differential equations describing the propaga­ dom-number generation device has been to Laboratory (see Section 6.24 of Summary Re­ typical network is pictured below in Fig. 6-1. tion of spherical waves has been under study. translate already-tested sets of random num­ port 28) is currently undergoing tests using The equations have been put into finite dif­ bers into proper form for machine input. For punched paper instruction tapes prepared both ference form along the characteristic direc­ this purpose, the 100,000 decimal digits of by hand methods and by the Whirlwind com­ tions of the problem, and a numerical process the Kendall-Babbington-Smith random-num­ puter. The tapes which Whirlwind can pre­ has been coded for solution on the Whirlwind ber tables have been translated into a seven- pare can be grouped into two classes, one 5 computer. Runs have been made for three hole Whirlwind tape through the use of the consisting of tapes whose preparation en­ different mesh widths in the characteristic punched-card-to-tape translation equipment tails a large amount of preliminary compu­ net. of the MIT Electrical Engineering Depart­ tation, the other of tapes which require little Results have been satisfactory using ment. A translation program was then writ­ or no advance computations. Work on writing 4 -* >. ) numbers of single-register length, but for ten to translate the digits, encoded in the a program for a typical problem of the first further study and investigation of regions in card-to-tape converter code, into 5-5-6 bi­ class was begun, and temporarily discon­ which the onset of a spherical shock wave is nary input form. tinued, at a time when Whirlwind had insuf­ \ / expected, double-precision, floating-point In addition, a direct read-in program ficient internal storage. All of the programs 3-* **2 methods must be called into use. Initial test has been written for use with'the translated written for the routine preparation of instruc­ runs are currently being made on this phase. tape, so that whenever control is transferred tion tapes for cutting straight lines and cir­ The general behavior of the solution in to the program, a random binary 0 or 1 will cular arcs, as well as the program for con­ Fig. 6-1. Typical network the case of a shock formation is to be con­ be brought into the Whirlwind accumulator. sidered in particular, and it is hoped that a The program refers to a previously good method of solving such regions numeri­ stored set of 50 registers of random binary cally may be worked out. digits for one digit. When this set of 50 is APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

24 6. MATHEMATICS. CODING, AND APPLICATIONS 6. MATHEMATICS, CODING, AND APPLICATIONS 25

exhausted, the program calls in another Problem #47. Partial Differential order integro-differential equations has been equationfor self-consistent field calculations group of 50 from a set of such registers which Equation of an Internal Combustion derived to describe the above situation, and of various atoms. Since there are six eigen­ are on the translated 5-5-6 tape. Because of Engine a method nf solution making use of finite values and six associated wave functions the speed of input of the photoelectric reader, difference methods has been set up for three which must be evaluated for an atom such as this process has proved to be practicable. This problem is anattempt to use Whirl­ different cases. A program is now being Cu , this is a lengthy computational task. wind's rapid calculating abilities for design written for computing a solution to the dif­ The prime reason for programming this prob­ purposes. The geometry of the inlet and ference equations in response to a sharp- lem on WWI is to utilize the high computing Problem #45. Crystal Structure exhaust valves and passages of internal com­ edged gust load for one of the three cases. speed and to formulate a program which will bustion engines affects the efficiency of such The numerical data obtained from this pro­ permit the determination of a single eigen­ This project, which originated with the engines measurably. Mr. Donald Tsai, of gram will be used to compute aircraft re­ value as an almost automatic procedure. MIT Electrical Engineering Department In­ Sloan Laboratories and the Mechanical En­ sponse to a graded gust load. The basic form of the second order dif­ sulation Research Group and the Chemistry gineering Department, MIT, is attempting ferential equation under investigation is: Department has been set up to study the to solve digitally the partial differential equa­ application of digital computing equipment to tions of such a system, so as to see what Problem #50. Lattice Analogy Applied various problems arising in the study of differences from the exact solution arise to Shear Walls Mi crystal structures by means of X-rays. 2E - 2V - p = o (i) under certain approximations, and also to dr l ^} The initial phase of the problem con­ determine quantitatively the behavior under G.D. Galletly of MIT has obtained useful sidered was the computation of a finite Fourier certain parameter changes. results on the problem of lattice analogy ap­ series of the form, The first goal set in this problem was plied to shear walls. The form of lattice analo­ subject to the following boundary conditions, to calculate a set of tables for certain pres­ gy that he utilizes is the one devised by Pro­ sure ratios and valve geometries. The pro­ fessor J.R. Benjamin of Stanford University. P(r -0 ) = 0 (2) The problem concerns itself with the determi­ f>(x,y,z) • L Fukecos27^hx+ky+lz-akl) gram for this computation is operating satis­ (u,k,l)e/" factorily. However, no block of time long nation of the load deflection characteristics of (1) enough to print out the complete table has as a shear wall under lateral forces such as P(r 0 (3) yet been made available. might be present in an earthquake or high where h, k, I, are integers and/a 3-dimen- Following the generation of the above winds. The type of wall being studied is con­ In this 2quation V « V(r) represents the po­ structed of reinforced concrete. tential of the central field, E represents the sional lattice domain, and where Fufci and program.it was decided to program the solu­ fll nkl are given. In crystallographical ap­ tion of the partial differential equation itself. The problem is to find the stresses in the energy of the electron in this field, and plications, the F's are taken to be the "crys­ The method employed was an iterative inte­ lattice bars so that they will satisfy the equa­ tal structure factors" obtained from X-ray gration along the characteristics of the hyper­ tion of statics and the equation of elastic ill + 1) diffraction data, and p(x,y,z) is then the bolic equation, a method which would par­ compatibility. Serious difficulty is encounter­ 2 ed when one considers the wall under cracked electron density at the point (x, y, z) of the ticularly give results similar to some already r crystal cell. A special case of equation 1, obtained by other methods. Such a program conditions. With a desk computer, to solve is the angular momentum term. the centrosymmetric case, has now been written using the (24, 6, 0) num­ the problem of a cracked wall using a 5x3 The presence of the singularity at the ber system, and is awaiting test on the ma­ lattice requires approximately 15 hours for origin increases the complexity of the prob­ chine. The solution will be obtained over each parameter. However, after a program lem. Therefore, the accuracy of the desired |0(x, y) = L L FhkQ cos 277-(hx+-ky) (2) several cycles in order to determine an for the solution of this problem has been solution is strongly influenced by such fac­ approximate "steady-state" solution. This written for Whirlwind I, each of the param­ tors as: problem should yield satisfactory results in eters requires only a few seconds for a com­ 1. The truncation error incurred in the was coded for WWI and has been producing this manner, since valve changes make the plete solution exclusive of print-out time. finite difference approximation useful results since the middle of the quarter. behavior of one cycle relatively independent We obtain the stresses and deflections 2. The choice of the actual interval of The values of f>(see equation 2) are com­ of the preceding one. for the cracked case by first considering integration used during the computation puted for (x, y) in a rectangular lattice. the uncracked case. When the stress of a 3. The number of decimal digits carried Normally, the crystallographer will plot the particular bar exceeds the tensile strength during the computation, i.e., the register values of p thus obtained and manually obtain Problem #48. Gust Loads on Rigid of the concrete, we merely eliminate that length used a contour map of the projection of the crys­ Airplanes in Two Degrees of Freedom specific bar and proceed to consider the next 4. The round-off error incurred during tal. Following Kendrew and Bennet (Acta one in our wall. the computation, etc. Crystallographer, I, 1952), a method of hav­ The Aero-Elastic and Structures Re­ To date, 81 parameters have been run The significance of these factors has been ing the machine plot the values of P on the search Laboratory of MIT under contract with the 5x3 lattice analogy program. A lattice was coded and successfully run. ascertained by previous detailed studies on with the Bureau of Aeronautics is now con­ second and more complicated program con­ punched-card equipment such as the 602A, The next step in the procedure is to ob­ sidering the effectof sharp-edged and graded sidering a 2x2 lattice, with slightly dif­ the 604, and the Card-Programmed Calcu­ tain an approximation to the position of the gust loads on rigid and elastic airplanes with ferent geometric configurations, has been lator (CPC). As a matter of fact, a solution atoms of the crystal by computing the cen- pitching taken into consideration. The analy­ successfully tested and is being run with obtained on the CPC using an 8-digit floating- troids of the peaks in the contour map de­ sis presumes two degrees of freedom - verti­ several different sets of parameters. A decimal board provided the necessary check scribed above. Methods of mechanizing this cal displacement and pitching rotation. The third program is envisaged in which the dis­ solution for ascertaining the accuracy of the process are now under consideration, and it forward velocity of the airplane is assumed tribution will be represented by a 5x4 lattice. WWI program. is expected that the mechanization of this to remain constant and the control surfaces The given differential equation is re­ phase of the problem will be completed in the are considered fixed. Problem #53. Solution of Boundary next quarter. placed by a finite difference approximation A system of two simultaneous second Value Problems on the Whirlwind which has a truncation error: Computer T/r)fe,J&.—a-(^P ) (4) llr,-240 ( . 6 ' This problem is the Schrodinger wave dr

.' APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

26 6. MATHEMATICS, CODING, AND APPLICATIONS 6. MATHEMATICS, CODING, AND APPLICATIONS 17

where h represents the interval of integra­ ing plants, water storage facilities, thermal index. the characteristic polynomial in the (24,6,0) tion. Since this is a fairly accurate approxi­ generating plants, and a distribution network The optimizing computations and eleva­ system also works satisfactorily, except in mation, the precision of the resulting solution which must supply a given load. The heart of tion corrections are being made at a finite the case of degenerate eigenvalues (double is quite good. the problem lies in the fact that water used at number of equally spaced ordinates of the roots of the polynomial). Round-off and loss The actual evaluation of the desired any hydro-gene rating plant, whether it be elevation curves over a time interval of of significant figures causes a definite lack eigenvalues consists of a trial-and-error natural flow or storage, yields an amount of fifty-two weeks. This spacing of ordinates of accuracy with certain polynomials. An method of solution; i.e., a trial eigenvalue energy dependent upon the head at that par­ can be made as small as desired, but a point attempt will be made to use the zeros of is selected, the solution is performed, and ticular time, which in turn is dependent upon of diminishing returns will be met. The derivatives of the polynomials to increase the resulting terminal conditions are ob­ the past operation of that plant. Similarly, minimum spacing desirable appears to be accuracy. served. The eigenvalue is then modified in present operation will determine future ener­ weekly. The third part of the over-all solution accordance with the observed discrepancy gy yield, since the consumption of water has also been written and is under test. This between the terminal conditions and the given affects the headand thereby affects the amount program, given a symmetric matrix and an boundary conditions (eqs. 2 and 3). The so­ of energy per unit discharge through that Problem #56. Determining Pupil eigenvalue of known multiplicity, will deter­ lution is then repeated using the new eigen­ plant. Data and Two Chromatic Aberrations mine the eigenvectors corresponding to the in Optical Lens Systems eigenvalue. value, etc. It is the mechanization of this Among the factors being considered are trial-and-error process which is the chief It is hoped that after some further ex­ the relations at each hydro plant between net Dr. F. Wachendorf of the Retina Founda­ perimenting and analysis of the round-off purpose of this investigation using the Whirl­ head, discharge, and generation; between wind computer. tion is attempting to use the computer for error involved, it may be possible to com­ stored water and storage elevation; and be­ the computation and evaluation of optical lens bine these three programs into a packaged The preliminary work performed on tween tail water elevation and discharge. systems. Programs are now being written program that will, given a matrix, deliver Whirlwind to date consists of four solutions. Also involved are the thermal cost charac­ for the evaluation of the third order mono­ eigenvalues and corresponding eigenvectors. The first two solutions were obtained for teristics, in dollars per megawatt-hour of chromatic lens aberrations {astigmatism, Although this may prove difficult in the de­ 160 points of the independent variable r using replacement energy, the time of water flow spherical aberration, coma, field curvature generate case, it may prove completely satis­ a constant interval length of 0.005. These between dams, and the characteristics of the and distortion) in optical lens systems. The ' factory in the case of simple eigenvalues. solutions are incomplete, since they cover distribution network. third order chromatic aberrations and the One result that is already apparent is that onlya portion of the desired range of integra­ The originator, R. Cypser of the MIT fifth order lens aberrations will betaken into for extremely ill-conditioned matrices such tion. However, they did serve to ascertain Electrical Engineering Department, has se- consideration in future programs. as those derived in the oxygen molecule case, the accuracy of the Whirlwind program com­ lecteda specific combination of hydroelectric a (39, 6, 0) number system may prove more pared to the results obtained on the Card- plants and storage facilities. These, in con­ useful than the present (24, 6, 0) scheme, be­ Programmed Calculator. junction with a thermal power source, form Problem #58. Determination of Energy cause of the need to avoid round-off losses The presence of the singularity at the a basic system having all the ingredients Levels of Oxygen Molecule in accuracy. origin and the associated large excursion of necessary to study the long-term coordination the digits in the coefficients of the given of water utilization and reduction of fuel cost. This problem, although it deals with a equation necessitated the use of the (24, 6, 0) The hydroelectric characteristics have been gas, is of fundamental interest in solid-state Problem #60. Determination of interpretive subroutine on Whirlwind. The taken from the system involving the Hungry physics, since the physics involved will be Energy Levels of the Deuteron as solution obtained on Whirlwind agreed within Horse generating plant and storage reservoir applicable later on to solids. The mathemat­ a Function of the Form of the Proton 7 significant decimal digits with that obtained on the South Fork of the Flathead River, and ics involved has been the development of a (or Neutron) Potential Function on the 8 -significant-decimal-digit Card-Pro­ the Kerr Dam and its associated storage set of "standard programs, " such that given grammed Calculator. These results are con­ reservoir at Flathead Lake, both located in asymmetric matrix, the program would print To determine theoretically the energy sistent with the 7-decimal-digit accuracy in­ Western Montana. The energy requirements out the eigenvalues and corresponding eigen­ levels of the deuteron (a proton and a neutron herent in the (24, 6, 0) code. of the basic systemare patterned after those vectors of the matrix. bound together by nuclear forces), one must The third and fourth solutions were ob­ of the Montana Power Company. The hydro The method originally decided upon, the solve Schrodinger1 s equation in three dimen­ tained using a varying interval length so deficiency is provided in the basic system method of successive traces, is an exact sions. For this particular problem, it is as­ chosen as to span the desired range of inte­ from a single plant representing the thermal solution scheme which under perfect condi­ sumed that the nuclear potential V is a func­ gration. This was done because of the limited source of energy. This plant is characterized tions should yield the characteristic poly­ tion only of the distance, r, from the proton amount of storage available on Whirlwind by an operating-cost characteristic in dollars nomial of such a matrix. Such a method had (or neutron); i.e. , the potential has no angu­ (— 160 double registers) after the necessary per hour versus megawatts generated. been used previously by Mr. Alvin Meckler lar dependence. of the Physics Department, the originator, requirements of the (24, 6,0)code and associ­ For the predicted flow and load patterns The nuclear potential function assumed ated interpretive routines are fulfilled. Ad­ and it was hoped corresponding check point is of the form: in the interval being optimized, the operator results could be made available. mittedly, a great loss of precision is thus estimates desirable storage elevation curves incurred; however, the use of a varying inter- as a function of time for each dam. These A program to determine the characteris­ :-Cxi valwill permit a consideration of the desired estimates are then subjected to a computation­ tic polynomial was first tested satisfactorily AV(x) automatic program previously described. al program which recommends small modifi­ for a number of fourth and fifth order matri­ J This program is currently being prepared. ces. At present, the characteristic poly­ cations to the elevation curves . These recom­ nomial of eight9x9and eight 12x 12matrices mendations are based ona numerical approach have also been calculated on the machine. Problem #54. Optimizing the Use to the corresponding problem in the calculus where A and C are parameters, x = r/rQ, The question of the effect of round-off error and r0 is some value determiried by scatter­ of Water Storage in a Combined of variations. The mathematics evaluates by on the characteristic polynomial, using (24, Hydro-Thermal Electric System ing experiments or binding energy. With the gradient method the influence on the per­ 6, 0) computations, has not yet been deter­ this potential function, and assuming spheri­ formance index of changes in elevation at mined completely. cal symmetry, Schrodinger's equation re­ The problem is concerned with the opti­ each dam throughout the time interval. The duces finally to: mization of system performance over a period results of the computations specify the direc­ A second program to find the roots of of time. The system whose operation is being tion and the relative amount of corrections optimized consists of hydroelectric generat- which will "best" reduce the performance APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

28 6. MATHEMATICS, CODING, AND APPLICATIONS 6. MATHEMATICS, CODING, AND APPLICATIONS 29

(1) i4 =[E-AV(x)l y OD H (rk) 6.2 SUBROUTINES COMPLETED L J dx ( m-o H' Hrkv ) m ' Library Subroutine Tape where E is proportional to the binding energy Number Number Title of the deuteron. The boundary conditions on this second order linear equation require The calculation of these Fourier coefficients OT2.2t T-664-3 Print C(AC) as Decimal Fraction, Sign and Magnitude, that y2, the probability density, be zero at (at a given r) was done by hand, and pro­ Point, Single Column Layout x = 0 and at x-co. The problem is to deter­ gramming is under way, but not yet completed. mine the wave functions and the eigenvalues, Since the equations are complex, we PA 8. It T-724-1 Operations on Real (15,0,0) Fixed-Point Single Register E, which satisfies these boundary conditions, shall be dealing with two interdependent sets Numbers (Interpretive) for various combinations of the parameters of linear algebraic equations in the numerical AandC. By comparing these binding energies approximation. Each set will represent 24 TF 1. It T-705 Form Sinewy from y Stored in AC, and Leave Result with those obtained by experiment, and using equations, and the Gauss-Seidel method will values of r0 based on scattering experiments, be used to find the solution. in AC (15,0,0) it is hoped that a better idea may be obtained of the form of the nuclear forces which bind TF 7. It T-750-1 Form Sine——x from x Stored in AC, and/or Form the deuteron. II Cosine——y from y Stored in AC, Leave Result in Once this nuclear binding potential is AC (15,0,0) determined, corrections will be added to it Problem #63. MIT Seismic Project to take into account the effects of relative ED 1. It T-874-2 Programmed Whirlwind I Operation (Interpretive)(15, 0,0) spin orientations and, in the case of the pro­ For the past two years a seismic research ton, the coulomb potential. projectunder the direction of Professor G. P. ED 1. 2t T-828-1 Octal Print of PC and Address Section of Instruction Wadsworth has been under way to determine the statistical behavior of seismic records on sj> or cp (-) only Problem #62. Reflection of Scalar and its relationship to wave propagation, geo­ Waves from a Cylinder with Inhomo- logical structure, and instrumentational char­ ED 1 3t T-727-6 Print Function Letters for Error Diagnosis geneous Acoustical Properties acteristics. Of particular interest is the detection of weak seismic wave reflections, OC1.3t T-694 1st Quadrant Axis Display Monochromatic sound waves are reflect­ often missed by present interpretation tech­ ed from a cylinder whose acoustical proper­ niques. A statistical method utilizing auto- OC 1.4t T-752 2nd Quadrant Axis Display ties are a function of the distance around and cross-correlation concepts was success­ the cylinder. The problem is to find the ful in this respect on several records. These OC1.5t T-753 3rd Quadrant Axis Display angular distribution of pressure on the cylin­ results greatly encouraged the desire to der surface, and the phase and amplitude of further test the values and limitations of the OC1.6t T-754 4th Quadrant Axis Display the reflected wave at any point outside the method in a large number of cases. In Feb- cylinder. ruaryl950 the project was greatly expanded, OT 1.3t T-542-3 Print C(AC) as Octal Number, Sign Digit and Complement, The equation is an inhomogeneous Fred- and integrated with the Geophysics Depart­ Point, Single Column Layout. holm integral equation of the second kind: ment under Professor Hurley. For an under­ OT 2.51t T-883 Print C(AC) as Decimal Integer, Magnitude Only, Initial taking of such magnitude, hand computation irkcOS Zero Suppression, Print Final Zero, No Layout ^(r.0) = e W U/(r,tfo)G(r^; r,^)d0 was not considered feasible. At this time -'o programming of the seismic project was PA 3. 5t T-721-1 Operations on Real (30, 0,0) Fixed Point Double Register where (fris the desired function, and G is the begun on the Whirlwind Computer; it has not Numbers (General Routine with Sign Agreement) Green's function, been completed yet. PA 3. lOt T-798 Operations on Real (30,0,0) Fixed Point Double Register Numbers (Short, Fast Routine without Sign Agreement, giving 28 Binary Digit Accuracy in mr)

VM 11. It T-802 Generate Scalar Matrix, Symmetric Form, From Scalar Stored in Accumulator, Matrix Order as Program Parameter (15, 0, 0)

AD 0. It T-856-3 Differentiate n-th Degree Polynomial to order K(24, 6, 0)

ED2.lt T-727-10 Programmed WW Operation, Print Function Letters for Error Diagnosis

ED 3. It T-780-5 Programmed WW Operation, Print C(AC) as 5-digit Signed Decimal Fraction

—• — APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

30 6. MATHEMATICS, CODING, AND APPLICATIONS 6. MATHEMATICS, CODING, AND APPLICATIONS 3

Library Subroutine Tape Library Subroutine Tape Number Number Title Number Number Title

ED 4. It T-828-3 Programmed WW Operation, Octal Print of C(PC) OT 103. It T-860-2 30, 0, 0 MRA Print and/or Punch, Decimal Fraction, and Address Section of Instruction on ££ and cp(-) Sign, Number of Digits Arbitrary, No Carriage only Return, No Sign Agreement (Interpreted)

ED 4.2t T-827-4 Programmed WW Operation, Decimal Print of C(PC) OT 105. lOt T-903-1 45, 0, 0 MRA Print and/or Punch, Decimal Fraction, and Address Section of Instruction on sp and cp(-) Sign, Number of Digits Arbitrary, No Carriage Only Return, Sign Agreement Program Included (Interpreted) ED5.lt T-1004 Programmed WW Operation, Print Octal C(PC) and PA 5. lOt T-902 Operations on Real 45, 0, 0 Numbers (Basic Instruction Decimal C(AC) on s£ and c_p_(-) Only Code Without Sign Agreement Giving 42 digit accuracy in mr) NR2.lt T-554-1 Square Root of C(AC), Result in AC PA 6. lOt T-929 Operations on Real 60, 0, 0 Numbers NR202.lt T-880-1 24, 6, 0 MRA Square Root Subroutine PM 102. 1 24, 6, 0 Interpretive Subroutine, Post Mortem OC2.5t T-891-1 ES Storage Octal Integer Display, Sign, Layout, Magnitude OC 3. It T-1021-1 .' , ' MRA Deflection Display Subroutine 24.6,y 0

OT 0. It T-979-1 Page Layout Subroutine for AC Print Out

OT 0. lOt T-981-1 Page Layout Subroutine for MRA Print Out

OT 1.4t T-923 Print, C(AC) as Octal Number, Sign and Magnitude, Point, Single Column Layout

OT 1 5t T-764-5 Print C(v3) through C(v4) as Octal Number, Sign Digit and Complement, Point, Single Column Layout

OT 1.6t T-927-3 Print C(v3) through C(v4) as Octal Number, Sign and Magnitude, Point, Single Column Layout

OT 1.7t T-773-3 Print C(v3) through C(v4) as Octal Number, Sign Digit and Complement, Point, Page Layout

OT 2. 3t T-855-3 Print C(v3) through C(v4) as Decimal Fraction, Sign and Magnitude, Point, Single Column

OT 2.52t T-937 Print C(AC) as Decimal Integer, Sign, Zero Suppression, Final Zero

OT 2.53t T-928 Print C(ES) as Decimal Integer, Sign, Initial Zero Suppression, Final Zero, Horizontal or Vertical Layout, Preset Parameters

OT 3. 2t T-786-2 Print Out Flexowriter Characters with Arbitrary Insertion (to be used with OT 3. 2at)

OT 3.2at T-787-1 Read In Flexowriter Characters, Stored Two to a Register (to be used with OT 3. 2t)

OT 101. It T-879-3 15, 15, 0 MRA Decimal Conversion and Output Print Subroutine (Column Layout)

OT 102. It T-829-6 24, 6, 0 MRA Output Decimal Conversion and Print Subroutine (Column Layout)

— —

r APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

12 ACADEMIC PROGRAM 33

7. ACADEMIC PROGRAM IN student, as well as prepared on punched AUTOMATIC COMPUTATION paper tape and executed on Whirlwind I Date Title Speaker AND NUMERICAL ANALYSIS computer. January 9 Some Remarks on the Stability of A special intensive two-week summer Difference Equations Professor F. B. Hildebrand program is to be offered July 21 through 7. 1 AUTOMATIC COMPUTATION August 2. This program, which will cover January 16 Discussion of Papers by F. John AND NUMERICAL ANALYSIS much of the material contained in the two and G. Blanch on the Integration subjects listed above, will carry no academic of Parabolic Partial Differential Mr. D.G. Aronson The MIT catalogue now lists Automatic credit and will be intended primarily for Equations by Difference Methods Mr. J. D. Porter Computation and Numerical Analysis as one people in industry who are interested in of four graduate options in Electrical Engi­ learning enough about the use of digital com­ February 7 Review of the Theory of Charac­ neering. The program includes subjects in puters to realize their potentialities and their teristics Professor C. C. Lin numerical analysis, analog computers, and limitations. This subject, which will consist other related subjects as listed in Summary of5-l/2hours of lectures and demonstrations February 14 The Use of Characteristics in the Report 23 and in the MIT Catalogue. There each day throughout the two-week period, is Numerical Solution of the Equations are three subjects directly concerned with described in the catalogue as follows. Representing Spherical Wave the application of digital computers. The Propagation Miss P. A. Fox first-semester subject, Introduction to Digi­ 6.539. DIGITAL COMPUTERS AND tal Computer Coding and Logic, has been THEIR APPLICATIONS. Survey of vari­ February 21 On the Truncation Error in the inc reased from two class hours to threeclass ous possible applications of digital com­ Numerical Solution of Hyperbolic hours a week. The second-semester subject, puter. Brief description of the logical Differential Equations Mr. D.G. Aronson offered for the first time in the spring of 1951 structure of real, as well as idealized, as A Special Problem in Electrical Engineer­ February 28 The Mathematics Panel at Oak ing, has again been offered during the spring digital computers. Detailed discussion of the techniques used in preparing coded Ridge and the Computing Machinery of 1952 as a regular subject, carrying the Connected with It. Dr. Clay Perry number 6. 537 and the name Digital Computers programs needed to instruct a general- purpose computer to solve a particular and Their Applications. Catalogue descrip­ A Philosophy of Programming tions of these subjects are as follows. problem. Emphasis on the use of sub­ March 6 routines. Brief treatment of such prob­ for High-Speed Machines Dr. J. W. Carr III lems as the handling of various forms of 6.535T. INTRODUCTION TO DIGITAL- terminal equipment, location of mistakes March 13 Construction and Use of Optimum COMPUTER CODING AND LOGIC (A). in programs, detection of computer mal­ Interval Mathematical Tables Dr. H.R.J. Grosch Interpretation of arithmetical and logical functions. Outline of the special mathe­ sequences into digital computer instruc­ matical methods needed for digital com­ March 20 Analysis of Round-off Error in tions. Basic principles of digital-com­ putation. Use of examples selected from the Tabulation of Some Fundamental puter use with examples selected from typical engineering scientific and business Functions Dr. A.J. Perlis common engineering mathematics. Logi­ problems and from real-time control ap­ cal analysis of a problem, flow diagrams, plications. Use of the Whirlwind I com­ The proposed program for April is: scale factor control, use of sub-programs puter at MIT to provide group demonstra­ and iterative sequences, and introduction tions and to permit individual practical April 3 The Method of Kernal Functions to real-time applications of digital com­ experience. No previous digital computer for Solving Boundary Value puters. Visits to Whirlwind Computer to experience necessary. Problems in the Theory of Linear witness the execution of programs dis­ Elliptical Partial Differential cussed in class. Equations Dr. Stefan Bergman 7. 2 SEMINARS ON COMPUTING 6. 537. DIGITAL COMPUTER APPLICA­ MACHINE METHODS April 10 Numerical Solution of Boundary TIONS PRACTICE (A). Advanced treat­ Value Problems by Kernal ment of the preparation of coded programs The Seminars on Computing Machine Function Methods Dr. Franz D. Alt for automatic, electronic digital comput­ Methods are now being arranged jointly by ers. Techniques for handling various representatives of the MIT Committee on April 17 Sampling Methods in Analog forms of storage and terminal equipment, Machine Methods of Computation and the MIT Computation Dr. F.M. Young for detecting errors and mistakes in pro­ Digital Computer Laboratory. Various speak­ grams and for controlling scale factors. ers from other MIT activities and elsewhere, April 24 Report on the paper, "On the Emphasis on the use of subroutines. Labo­ as well as members of the two sponsoring numerical solution of the ratory work demonstrating performance groups , participate in these weekly seminars, Dirichlet problem for Laplace's by Whirlwind I computer of programs pre­ whichare held in a lecture room at the Insti­ difference equation, " by J.B. pared by the class. At least one problem tute. The program during the past quarter Diaz and R.C. Roberts, Quart, of his own choice programmed by each was as follows. of Appl. Math., January, 1952 Mr. N.J. Hicks APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

M 8. APPENDIX 3^

8. APPENDIX No. of No. Title Pages Date Author 8.1 REPORTS AND PUBLICATIONS E-453 Error Diagnosis Subroutines for Use with Standard Single Length Project Whirlwind technical reports and memorandums are routinely distributed to only a Program 6 3-17-52 D. Combelic restricted group who are known to have a particular interest in the Project. Other people who need information on specific phases of the work may obtain copies of individual reports by E-454 A Non-Destructive Read System making requests to John C. Proctor, Digital Computer Laboratory, 211 Massachusetts Avenue, for Magnetic Cores 3 3-24-52 D.A. Buck Cambridge 39, Massachusetts. The following reports and memorandums were among those issued during the first quarter M-1371 Magnetic Core Activity 5 1-15-52 W.N. Papian of 1952. M-1381 Magnetic-Core Memory Matrix No. of Analysis (Effect of Driver No. Title Pages Date Author Impedance) 4 i-24-52 D.A. Buck SR-27 Summary Report Number 27, M-1383 Block Diagram of the Buffer Third Quarter, 1951 29 Drum System 5 1-25-52 E.S. Rich SR-28 Summary Report Number 28, M-1400 Stabilized Transistor as a Four Fourth Quarter, 1951 30 Terminal Non-Linear Network 12 3-6-52 J.F. Jacobs R-90-1 The Binary System of Numbers 9 Revised M . F. Mann 2-29-52 8.2 PROFESSIONAL SOCIETY PAPERS Title R-204 A Head for Static Reading of Author Magnetic Recording. M.S. Thesis 72 1-15-52 D.H.A. Hageman At the Winter General Meeting of the "Digital Computers - AIEE held in New York on January 24, Jay Present and Future R-206 Charles Babbage - Scientist W. Forrester presented a paper on "Three- Trends" Jay W. Forrester and Philosopher 28 1-31-52 R. R. Rathbone Dimensional Magnetic Storage." Mr. For­ rester spoke on "Digital Computers" to the "The Whirlwind I R-207 The Use of a Transconductance Boston Section of the AIEE on February 12. Computer" R.R. Everett Bridge in the Measurement of Five members of the Project Staff de­ Cathode Interface Impedance 20 3-13-52 H. B. Frost livered papers at the IRE National Convention "Evaluation of the in New York during the first week of March: Engineering Aspects E-421-1 Transistor Bibliography 8 2-19-52 N. T. Jones of Whirlwind I" N.H. Taylor Title Author E-440 WWI Operating Speed 8 12-21-51 R. P. Mayer E-441 Standardized Transistor "Digital Computers 8.3 VISITORS J. F. Jacobs in Control Systems" Jay W. Forrester Parameter Measurements 8 1-3-52 N. T. Jones During the past quarter the Laboratory has had among its visitors the following: E-442 Analysis of Vacuum Tube "The Measurement of L. Sutro Cathode Interface Mr. C .P. Marsden of the National Bureau Failure Data 6 1-8-52 H. B. Frost Impedance" H. B. Frost of Standards, who studied the vacuum-tube M. Mackey construction facilities. "Magnetic Matrix Mr. R. A. Aldridge, Mr. R. Berlin, E-443-1 Interface Formation in 6AG7 Tubes 5 1-25-52 H. B. Frost Switches K.H. Olsen and Mr. CD. Cockburn of General Electric Co. , who discussed air traffic control. E-444 Changes in Design of the re and rd "The Digital Computer Mr.S.E. Knudsen and Mr. E.K. Kelley Orders 3 1-31-52 E.S. Rich as a Control Element" C.R. Wieser of General Motors. Mr. L. Petter of General Electric, who E-448 Variation of Transistor Collector J. F. Jacobs "Analysis of Control discussed air traffic control. Resistance Due to Self Heating 2 2-5-52 N.T. Jones Systems Involving Mr. R. C. Davis of the Laboratory for Digital Computers" W.K. Linvill Electronics, who discussed general aspects E-450 Whirlwind I Terminal Equipment of magnetic drums. (Talk given at Digital Computer Laboratory Seminar January 22, The "Review of Electronic Digital Com­ L. B. Melson and R. H. Lee of the U.S. 1952) puters" (an account of the Joint AIEE-IRE Navy. 2-21-52 E.S. Rich Computer Conference held in Philadelphia Professor Hideo Yamashita of the Elec­ trical Engineering Department of Tokyo Uni­ E-451 Variation of Transistor Collector December 10-12, 1951), published by the AIEE N. T. Jones in February 1952, contains three papers by versity. Resistance with Collector Voltage 3-3-52 R.J. Callahan members of the Project Staff: Mr. R. Wood, Mr. I. Levy, and Mr P. APPROVED FOR PUBLIC RELEASE. CASE 06-1104.

)6 8. APPENDIX

Mannix of Raytheon Manufacturing Co. , who Mr. H. F. Shannon of United States Steel discussed means of measuring interface re­ Company. sistance. Mr. C.L. Schepens and Mr. F. Wachen- Mr. J. H. Burnett of Electrons Inc., dorf, of the Retina Foundation at the Massa­ Newark, N.J. , who discussed thyratrons in chusetts Eye and Ear Infirmary, who were d-c power supplies. interested in the applications of the computer. Mr. CE. Kerr, Mr. S. W. Coppock, Mr. Mr. D. J. Rose of Bell Telephone Labora­ K. V. Diprose, and Mr. R.H. Tizard, from tories. England, who discussed matters of general Mr. G. L. Hollander and Mr. R. L. Stern­ interest in digital computers. berg, of the Laboratory for Electronics, Mr. R. Fallows and Mr. B.H. Alexander who were interested in using the computer for of Sylvania, who discussed transistor ap­ applications purposes. plications . Mr. D. Holmes of the Shell Development Capt. LB. Smith, Lt CE. Rodgers, Company. and R. D. Crow, of the United States Air Mr. John E. Garret of Olin Industries, Force. who discussed the possibility of submitting Mr. B. W. Pollard of Ferranti Ltd. , problems for the computer. England. Mr. Clay Perry of Oak Ridge National Mr. N.B. Saunders and Mr. F. A. Browne, Laboratories. Jr. of Transducer Corporation, who discus­ Mr. W. R. Ashby of Barnwood House, sed magnetic-core applications. Gloucester, England, who discussed some Mr. I.L. Auerbach, Mr. J. Pairinen, general aspects of the computer. and Mr. T. C. Chem of Burroughs Adding Dr. Warren McCulloch of the University Machine Co. , who discussed magnetic-core of Chicago. memories. Mr. R. E. Wilson, Mr. F. G. Miller, Mr. R.S. Williams, Mr. M. Rosenberg Mr. S.W. Spaulding, and Mr. I. H. Sublette, Mr. L. Person, and Mr. J. Rajchman, of of R. C.A. , who were interested in obtaining R.C. A. , who discussed magnetic-core stor­ information on the Whirlwind design. age and switching. Mr. E. G. Shower of Radio Receptor Co. , Mr. H.J. Crawley of National Research who was interested in our transistor facili­ Development Corporation, who discussed ties. materials for magnetic storage. Mr. H.C. Rambug of Bonneville Power Mr. W.J. Yost of Magnolia Petroleum Administration. Company, who was interested in the applica­ Mr. L. D. Wann of the Continental Oil tions aspects of the computer. Company, who was interested in the coding Mr. J.M. Crawford and Mr. D.E. Mitchell aspects of the computer. of the Continental Oil Company. Mr J. D. Noe of Stanford Research Mr. L. Laskaris and Mr. L. Whorton Corporation, who discussed the Whirlwind of the Atlantic Refining Company, who dis­ approach to the computer problem. cussed general aspects of the computer.