# How did Dennis Ritchie Produce his PhD Thesis? A Typographical Mystery

> An annotated edition of this paper on Fermat's Library, where scientists and readers explain its ideas, equations and history in the margins. Annotated edition: https://fermatslibrary.com/s/how-did-dennis-ritchie-produce-his-phd-thesis-a-typographical-mystery

- **Authors:** David Brailsford et al.
- **Published:** 2022
- **Topic:** Computer Science
- **Original:** https://s3-us-west-1.amazonaws.com/fermatslibrary/uploads%2F1749532291493-m90e8vl1g4n-e4d2f04ee6ae1aeab4e217fac55898d7%2Fritchie.pdf
- **Pages:** 10
- **Annotations:** 18 comments in 17 threads
- **Featured:** Fermat's Library Journal Club
- **Annotated edition:** https://fermatslibrary.com/s/how-did-dennis-ritchie-produce-his-phd-thesis-a-typographical-mystery

## Annotations

### Annotation 1 (page 1)

Beside the passage: “Dissertation, an article about Dennis’slong-lost PhD thesis, ProgramStructureand Computational Complexity. Brock’sarticle [1] makes for fascinating reading. Much of it is focused on the thesis’scontributions to recursive function theory and early theoretical computer science. To over-simplify,the thesis showed that a …”

**Annotation** by João Batalha (2025-06-10):

> Here is the article:
> https://computerhistory.org/blog/discovering-dennis-ritchies-lost-dissertation/

### Annotation 2 (page 1)

Beside the passage: “ABSTRACT Dennis Ritchie, the creator of the C programming language and, with Ken Thompson, the co-creator of the Unix operating system, completed his Harvard PhD thesis on recursive function theory in early 1968. But for unknown reasons, he neverofﬁcially receivedhis degree, and the thesis itself disappeared for …”

**Annotation** by João Batalha (2025-06-10):

> Dennis Ritchie was born in 1941 in Bronxville, New York, and grew up in Summit, New Jersey. His father worked at Bell Labs, which exposed him early to science and engineering. At Harvard, he started in physics but became interested in computing after attending a lecture on the UNIVAC I. He switched to applied mathematics for graduate school, where he began exploring theoretical computer science.
>
> ![](https://i.imgur.com/2tjBGMl.jpeg)
> *Dennis Ritchie at Harvard*

### Annotation 3 (page 1)

Beside the passage: “Dissertation, an article about Dennis’slong-lost PhD thesis, ProgramStructureand Computational Complexity. Brock’sarticle [1] makes for fascinating reading. Much of it is focused on the thesis’scontributions to recursive function theory and early theoretical computer science. To over-simplify,the thesis showed that a …”

**Annotation** by João Batalha (2025-06-10):

> Think of a loop program as working with a handful of “registers,” each holding a non-negative integer.
> You are allowed only three basic instructions:
>
> ```
> X ← 0          # reset X to 0
> X ← X + 1      # add 1 to X
> X ← Y          # copy the value of Y into X
> ```
>
> And one control construct:
>
> ```
> LOOP X TIMES
>     … instructions …
> END
> ```
>
> The body runs exactly the number of times currently stored in `X`. Because loops can be placed inside other loops, the total number of iterations can grow explosively, letting these tiny building blocks mimic any algorithm.
>
> #### Example: Multiplying 3 × 4
> Registers: `A = 3, B = 4, R = 0`
>
> ```
> LOOP B TIMES          # outer loop runs 4 times
>     LOOP A TIMES      # inner loop adds 3 to R each pass
>         R ← R + 1
>     END
> END
> ```
>
> After the nested loops finish, `R` holds 12. Even with nothing but resets, increments, copies, and fixed-count loops, you can write addition, subtraction by counting down, conditional tricks (store a flag, skip work when it is 0), and eventually simulate a full Turing machine.

### Annotation 4 (page 1)

Beside the passage: “Dissertation, an article about Dennis’slong-lost PhD thesis, ProgramStructureand Computational Complexity. Brock’sarticle [1] makes for fascinating reading. Much of it is focused on the thesis’scontributions to recursive function theory and early theoretical computer science. To over-simplify,the thesis showed that a …”

**Annotation** by João Batalha (2025-06-10):

> Harvard in the late 1960s followed a tight three-step routine. First a typed draft, usually from an IBM Selectric, went to the dissertation committee for edits. After fixes, the candidate held a small departmental defense, then a Public Oral Examination advertised two weeks in advance in the Harvard Crimson. Passing the orals unlocked the paperwork stage. The student had to produce a flawless typescript on acid-free paper, margins 1 ½ in. left and 1 in. elsewhere, each page initialed by the advisor. Three hard-bound copies were made at a local bindery: one for University Archives, one for Widener Library, one for the department. A bindery receipt plus the signed degree form had to reach the registrar by the semester deadline; otherwise the degree slipped to the next term.

### Annotation 5 (page 1)

Beside the passage: “Dissertation, an article about Dennis’slong-lost PhD thesis, ProgramStructureand Computational Complexity. Brock’sarticle [1] makes for fascinating reading. Much of it is focused on the thesis’scontributions to recursive function theory and early theoretical computer science. To over-simplify,the thesis showed that a …”

**Annotation** by João Batalha (2025-06-10):

> Ritchie entered Harvard in 1959 and did finish an undergraduate degree in physics in 1963. Later he joked that his “undergraduate experience convinced me that I was not smart enough to be a physicist, and that computers were quite neat”

### Annotation 6 (page 2)

Beside the passage: “throughout his lifetime. This includes citations for the Turing Award and the Japan Prize, both of which mention his Harvard doctorate. Dennis was so guileless his entire life that this must have weighed on him, but it is unlikely that we will everknowfor sure. This part of the story,still a work in progress, is told …”

**Annotation** by João Batalha (2025-06-10):

> In the late 1960s, when Ritchie wrote his thesis, there were no digital typesetting systems like LaTeX or Word. Mathematical documents were usually typed on electric typewriters—often by hand or by hired typists. Symbols like Greek letters or integrals had to be inserted manually, sometimes using special typeballs on IBM Selectric machines, or drawn in by hand afterward. Subscripts and superscripts were produced by physically adjusting the paper in the typewriter - rolling it up or down by tiny increments.
>
> The first widely used computerized formatting tool, RUNOFF, had just been written at MIT in 1964, but it couldn’t handle math. Troff and eqn, which could handle mathematical typesetting, wouldn’t arrive until the 1970s. LaTeX came much later, in the 1980s. So at the time of Ritchie’s thesis in 1968, producing a clean, complex math thesis with precise formatting required a level of mechanical precision, patience, and ingenuity that most people today associate with programming - not typing.

### Annotation 7 (page 2)

Beside the passage: “throughout his lifetime. This includes citations for the Turing Award and the Japan Prize, both of which mention his Harvard doctorate. Dennis was so guileless his entire life that this must have weighed on him, but it is unlikely that we will everknowfor sure. This part of the story,still a work in progress, is told …”

**Annotation** by João Batalha (2025-06-10):

> ## troff and eqn
>
> **troff** (short for *typesetter roff*) was developed at Bell Labs in the early 1970s by Joe Ossanna and later improved by Brian Kernighan. It was written in C and designed to format text for high-quality typeset output, especially on Unix systems. The name “troff” comes from a chain of earlier tools: RUNOFF (from MIT) → roff → nroff (“new roff”) → troff (“typesetter roff”).
>
> Here’s a basic `troff` example:
>
> ```
> .TL
> Unix Programmer's Manual
> .AU
> Dennis Ritchie
> .PP
> This is a paragraph in a troff document.
> ```
>
> **eqn** was introduced in 1974 by Brian Kernighan and Lorinda Cherry as a `troff` preprocessor for typesetting mathematics. It let users write math expressions in plain text that would be rendered cleanly in print.
>
> Example:
>
> ```
> .EQ
> e = m c sup 2
> .EN
> ```
>
> This would render: $e = mc^2$

### Annotation 8 (page 2)

Beside the passage: “2. Background: 1960s Typing Technology The 1960s was the decade of the electric typewriter,but not yet the electronic typesetter.IBM, the leader in the ﬁeld, wasfocused on the burgeoning business community rather than scientiﬁc research; their primary goal was ofﬁce integration and efﬁciencyrather than the elaborate …”

**Annotation** by João Batalha (2025-06-10):

> An electric typewriter mechanically prints characters directly onto paper using fixed spacing and limited fonts, making precision layout and special symbols difficult. An electronic typesetter, however, digitally arranges text with precise control over fonts, sizes, spacing, and alignment, ideal for mathematical and technical documents - which can later be printed.

### Annotation 9 (page 2)

Beside the passage: “2.1. IBM Electronic Typewriters In 1959 IBM introduced the Model C basket-and-typebar typewriter,then in 1967 the Model D of the same design. Other manufacturers including Royal, Olympia, Smith Corona and Olivetti offered typewriters with similar design and features. Forthe most part these machines used a ﬁxed width …”

**Annotation** by João Batalha (2025-06-10):

> IBM’s Model C and Model D typewriters were popular office machines known for their durability and standard fixed-width fonts (usually 10 or 12 characters per inch). Unlike later IBM Selectrics - which had interchangeable typeballs - these earlier models used traditional mechanical typebars arranged in a basket. Precise typographical layout with these machines was challenging, as spacing and character sets were rigidly limited.
>
> ![](https://i.imgur.com/QrwTk7b.png)
> *IBM Model C*  ![](https://i.imgur.com/W5tjAo0.png)
> *IBM Model D*

### Annotation 10 (page 2)

Beside the passage: “2.1. IBM Electronic Typewriters In 1959 IBM introduced the Model C basket-and-typebar typewriter,then in 1967 the Model D of the same design. Other manufacturers including Royal, Olympia, Smith Corona and Olivetti offered typewriters with similar design and features. Forthe most part these machines used a ﬁxed width …”

**Annotation** by João Batalha (2025-06-10):

> IBM’s Selectric typewriter (1961) was revolutionary due to its spinning typeball - nicknamed the “golf-ball” - which replaced the conventional basket-and-typebar system. With the Selectric, the paper stayed stationary while the typeball moved across it, dramatically improving typing speed and print clarity. Importantly for mathematical and scientific typing, users could quickly swap typeballs to insert special characters like Greek letters or mathematical symbols, rather than manually inserting separate type sticks. This greatly simplified producing technical documents before digital typesetting became common.
>
> ![](https://i.imgur.com/wt6lrYg.gif) *IBM Selectric*

### Annotation 11 (page 3)

Beside the passage: “from a typographic perspective isvastly superior to most other math dissertations of the period.”

**Annotation** by João Batalha (2025-06-10):

> Here you can see a page from Stephen Hawking’s PhD dissertation Properties of Expanding Universes (1966), where he filled out the equations by hand.
>
> ![](https://i.imgur.com/Qm8jort.png)

### Annotation 12 (page 3)

Beside the passage: “MikeFischer echoes this: “With my own thesis, I beganby typing a rough draft myself that I then edited using scissors and tape and pencil and white-out. Once a page became too patched up, I made a Xerox copy to create a clean newversion that I could then edit and chop further as needed. Once my draft was ﬁnalized, I …”

**Annotation** by João Batalha (2025-06-10):

> RUNOFF, created by Jerry Saltzer at MIT in 1964, was one of the earliest text formatting programs. It allowed users to produce formatted documents from plain-text inputs with simple markup commands (for instance, line breaks, paragraphs, or page numbers). However, RUNOFF was relatively advanced for the era and designed specifically for MIT’s CTSS operating system, limiting its portability.
>
> When Brian Kernighan needed a formatting tool for his own Princeton PhD thesis in 1968, he developed a simplified version called Roff. Written in Fortran for the IBM 7094, Roff removed many specialized features to fit within the constraints of his environment.
>
> Despite its limitations, Roff’s simpler design and greater portability made it easier to adapt to new computing environments. This eventually helped it become a foundational text-processing tool at Bell Labs, evolving into nroff and later troff, which significantly influenced document formatting in Unix systems.

### Annotation 13 (page 4)

Beside the passage: “3.4. Roman Numeral Lists Roman numeral lists were a frequent component of the thesis; Dennis used such lists twenty three times. With the ﬁrst ﬁvelists, all typed characters stayed in their respective 1/12" by 1/12" cells. Starting with the Roman numeral list on page 39, the spacing changed to create a newtypographic …”

**Annotation** by João Batalha (2025-06-10):

> Here you can see the difference in alignment of Roman numeral lists:
> ![](https://i.imgur.com/ivIHrDB.png)

### Annotation 14 (page 5)

Beside the passage: “4.1. Font Issues In a previous restoration project [4] we discovered the importance of ﬁnding that a set of fonts created for the Linotron 202 typesetter had been migrated into the modern era in Adobe Type 1 PostScript format. All the mechanical limitations of the 202 could be sidestepped when its fonts could be used …”

**Annotation** by João Batalha (2025-06-10):

> The IBM 2741 was a computer terminal introduced in 1965 that used the Selectric typewriter mechanism as its printing engine. In this context, a terminal was not a computer itself, but a device used to send and receive text from a mainframe - a large, centralized computer located elsewhere. The 2741 had to be connected to a mainframe to function, typically over phone lines or dedicated data links.
>
> It offered much higher print quality than line printers or teletypes of the time and was used in research labs, universities, and places like Bell Labs and MIT’s Project MAC. Because of its precision and support for interchangeable typeballs, it was one of the few devices in the 1960s capable of producing the kind of clean, symbol-rich output seen in Ritchie’s thesis.
>
> ![](https://i.imgur.com/Fqeq8Td.png)
> *IBM 2741 Terminal*

### Annotation 15 (page 6)

Beside the passage: “Our font consultant, Chuck Bigelow, determined that the 2741 character set had indeed been converted to a ﬁxed-pitch Bitstream Inc. font called ‘Pica 10’ with the same metrics as the original golf-ball element. Unfortunately this font only includes the standard character set. Despite intensive efforts we have been …”

**Annotation** by João Batalha (2025-06-10):

> Y&Y’s MTMS script refers to a digital font collection developed by the company Y&Y, which specialized in high-quality TeX and PostScript font support in the 1990s and early 2000s. MTMS stands for Mathematical Typesetting Math Script, a family of script-style fonts designed for use in mathematical and scientific documents.
>
> These fonts mimicked handwritten or calligraphic math script (used for things like ℒ, 𝓕, etc.) and were compatible with typesetting systems like TeX and troff.

### Annotation 16 (page 6)

Beside the passage: “Our font consultant, Chuck Bigelow, determined that the 2741 character set had indeed been converted to a ﬁxed-pitch Bitstream Inc. font called ‘Pica 10’ with the same metrics as the original golf-ball element. Unfortunately this font only includes the standard character set. Despite intensive efforts we have been …”

**Annotation** by João Batalha (2025-06-10):

> The monus symbol (∸) is used to represent truncated subtraction. It behaves like regular subtraction, except the result is never allowed to go below zero:
>
> $$
> a ∸ b = \max(a - b,\ 0)
> $$
>
> For example:
> $$5 ∸ 3 = 2$$
> $$2 ∸ 5 = 0$$
>
> This operation is common in theoretical computer science, especially when working with natural numbers or defining functions in models where negative values aren’t permitted

### Annotation 17 (page 7)

Beside the passage: “Figure8: Braces produced with eqn”

**Annotation** by Steve Powell (2025-06-11):

> I notice that the colons in the upper braced section are misaligned with the subscripts.

**Reply** by Steve Powell (2025-06-11):

> I notice that the colons in the upper braced section are misaligned with the subscripts.

## Full text

Extracted from the PDF. [Annotation N] marks where each annotation above sits.

#### Page 1

-- --

Howdid Dennis Ritchie Produce his PhD Thesis?

ATypographical Mystery

David F.Brailsford School of Computer Science

University of Nottingham Nottingham NG8 1BB, UK

dfb@cs.nott.ac.uk

Brian W.Kernighan Department of Computer Science

Princeton University Princeton, NJ 08540, USA

bwk@cs.princeton.edu

William A. Ritchie

Thinkfun Inc. 1725 Jamieson Ave Alexandria, VA22314

bill.ritchie@thinkfun.com

ABSTRACT Dennis Ritchie, the creator of the C programming language and, with Ken Thompson, the co-creator of the Unix operat- [Annotation 2] ing system, completed his Harvard PhD thesis on recursive function theory in early 1968. But for unknown reasons, he neverofﬁcially receivedhis degree, and the thesis itself disappeared for nearly 50 years. This strange set of circumstances raises at least three broad questions:

•What was the technical contribution of the thesis? •Why wasn’tthe degree granted? •How was the thesis prepared? This paper investigates the third question: howwas a long and typographically complicated mathematical thesis produced at a very early stage in the history of computerized document preparation?

CCS CONCEPTS I.7 [Document and Text Processing]: I.7.1: Document and Te xtEditing, I.7.2: Document Preparation.

KEYWORDS mathematical typesetting, electric typewriter,digital restoration, archiving, troff, Postscript fonts, IBM 2741

ACMReference Format

David F.Brailsford, Brian W.Kernighan and William A. Ritchie. 2022. Howdid Dennis Ritchie Produce his PhD Thesis? A Typographical Mystery.InProceedings of The 22nd ACM Symposium on Document Engineering (DocEng2022). ACM, NewYork, NY, USA, 10 pages. https://doi.org/10.1145/3558100.3563839

1. Introduction In June 2020, David Brock, a historian of technology and director of the Computer History Museum’sSoftware History Center,published Discovering Dennis Ritchie’sLost Permission to makedigital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for proﬁt or commercial advantage and that copies bear this notice and the full citation on the ﬁrst page. Copyrights for components of this work owned by others than ACM must be honored. Forall other uses, contact the authors. DocEng2022, Sept 2022, Virtual Event, CA USA. ©Copyright held by the authors 978-1-4503-1789-4/13/09. https://doi.org/10.1145/3558100.3563839

Dissertation, an article about Dennis’slong-lost PhD thesis, [Annotation 1] ProgramStructureand Computational Complexity. Brock’sarticle [1] makes for fascinating reading. Much of it is focused on the thesis’scontributions to recursive function theory and early theoretical computer science. To over-simplify,the thesis showed that a class of programs expressed as assignments, increments, and nested loops was capable of performing arbitrary computations. Quoting Brock, “In loop programs, one can set a variable to zero, add 1 to a variable, [Annotation 3] or move the value of one variable to another.That’sit. The only control available in loop programs is ... a simple loop, in which an instruction sequence is repeated a certain number of times. Importantly,loops can be "nested," that is, loops within loops.”Inmore modern terms, these loop programs are a Turing-complete computational model, equivalent to Turing machines and Church’slambda calculus. The ﬁrst section of Brock’sarticle, "Everything but Bound Copy,"explores an intriguing open question. Although the thesis was essentially ﬁnished, lacking only a handful of trivial typographical corrections and presumably a pro forma ﬁnal public oral exam, the thesis was neversubmitted to Harvard (or so it is believed), it deﬁnitely was never [Annotation 4] accepted by Harvard, and thus Dennis neveractually receivedhis PhD. Whywasn’tthe thesis accepted by Harvard? Whydidn’t Dennis everget his PhD? Indeed, whydid he neverexplicitly acknowledge the unusual situation? And howdid the thesis simply disappear for nearly 50 years, coming to light only after Dennis’ssister Lynn tracked down a copyafter his death in October 2011? One oft-told story was that Harvard wanted a fee for processing the thesis and Dennis thought that he shouldn’thav e to pay it. If thesis rejection was as simple as a library fee dispute, however, weshould expect that Dennis would have recounted the story,embraced it in his typically self-deprecating way,and turned it into a life lesson similar to the way he described how he wasn’tsmart enough to become a [Annotation 5] physicist so he turned to computing. Instead, the thesis disappeared for 50 years and was never mentioned by Dennis. More signiﬁcant is how he allowed the uncorrected story that he had a doctorate to spread

1

#### Page 2

-- --

throughout his lifetime. This includes citations for the Turing Award and the Japan Prize, both of which mention his Harvard doctorate. Dennis was so guileless his entire life that this must have weighed on him, but it is unlikely that we will everknowfor sure. This part of the story,still a work in progress, is told by Bill Ritchie, Dennis’syoungest brother, at dmrthesis.net, a web site devoted to the thesis [2]. This paper explores a different set of questions that are particularly appropriate for a conference on document engineering. Howdid Dennis manage to produce multiple versions of a long and complicated document with exceptionally high quality and accuracy, atatime when computer preparation of mathematical documents was not even in its infancy? What hardware and software might he have used, and howdid it relate to the state of the art at the time? [Annotation 6] We also describe our attempt to recreate the thesis using standard Unix document preparation tools like troff and eqn that became widely available only a fewyears later [3]. A [Annotation 7] machine-readable version of the thesis would enable a number of studies that are too hard with only imperfect scans of typewriter-likeprintouts, and thus might shed some light on the early history of computational complexity. The results of our experiments and supporting documents for this paper are available at www.cs.princeton.edu/˜bwk/dmr.

2. Background: 1960s Typing Technology The 1960s was the decade of the electric typewriter,but not yet the electronic typesetter.IBM, the leader in the ﬁeld, [Annotation 8] wasfocused on the burgeoning business community rather than scientiﬁc research; their primary goal was ofﬁce integration and efﬁciencyrather than the elaborate scientiﬁc notation that was required for academic research papers.

2.1. IBM Electronic Typewriters In 1959 IBM introduced the Model C basket-and-typebar typewriter,then in 1967 the Model D of the same design. Other manufacturers including Royal, Olympia, Smith [Annotation 9] Corona and Olivetti offered typewriters with similar design and features. Forthe most part these machines used a ﬁxed width system, either 10 or 12 characters per inch, with 6 lines per inch vertical spacing. In 1961 IBM introduced the revolutionary newSelectric typewriter with its spinning typeball or ‘golf-ball’ design, [Annotation 10] which kept the paper stationary while the typeball movedits wayacross the platen. Inserting a special purpose character such as a mathematical symbol or Greek letter was much easier with a Selectric: the typist could simply swap in a new typeball rather than having to insert a supplementary type stick into the basket, which was the method with traditional machines. While not fully modern, the Selectric was more suitable for electronic control than were traditional typewriters. In 1964 IBM introduced a standalone word processing machine, the

MT/ST (magnetic tape, Selectric typewriter), which was marketed primarily to large businesses along with dictation equipment with the intention to better integrate secretaries and their bosses in the executive suite. The Selectric also formed the guts of the IBM 1050 printer terminal, which was introduced in 1963 for use with the System/360 and other mainframe computers, and the more streamlined 2741 terminal which launched in 1965. Compared with other printing devices of the time, Selectric-based terminals were faster,had better print quality,were quieter and had more easily interchangeable special characters and type fonts. But IBM always sawthese machines as proprietary and nevermade anyattempt to use ASCII standards or otherwise embrace the coming electronic revolution.

2.2. Early Formatting Programs In 1964 Jerry Saltzer,while working for Project MACat MIT,wrote the text formatting program RUNOFF to help type his thesis proposal and subsequently his thesis. (Conveniently he had an IBM 1050 terminal in his home that was connected through CTSS to the 7094 mainframe at Project MAC.) As much of a historical breakthrough as this was, RUNOFF did not have any facility to help change golf-balls in the middle of printing a manuscript; likewise it had no commands for superscripts or subscripts as these were not supported by the Selectric. According to Saltzer,“If you look skeptically at its list of features you will discoverthat it includes just enough to allow me to prepare my own PhD thesis, nothing more. For example, my thesis had no equations, so RUNOFF had no facility for them. Development of RUNOFF features ended in 1966 when I turned in my thesis.” So what were the conditions that Dennis faced as he was completing work on his doctoral thesis in late 1967 and early 1968? MikeFischer (a fellowHarvard grad student and brother of Dennis’sﬁrst thesis advisor,Patrick Fischer) says “Computerized typesetting was in its infancy in the 1960s. Of course people had the idea of having the computer print their paper instead of a typist, but printers were limited in both their quality and their range of allowed fonts. Forthe time that Dennis was at Harvard, it was electric typewriters.” It may be hard for readers today to appreciate just howlaborintensive itwas to prepare documents before the creation of word processing programs, when there were only mechanical typewriters—better than clay tablets or quill pens, to be sure, but anychange of more than a fewwords in a document could require a complete retype. Thus most documents went through only one or tworevisions, with handwritten changes on a manuscript that had to be laboriously retyped to makeaclean copy. As Jerry Saltzer said “The standard procedure for preparing aPhD thesis in the 1960s was to assemble a rough draft either by typing or in longhand and then hire a professional thesis typist.”

2

#### Page 3

-- --

MikeFischer echoes this: “With my own thesis, I beganby typing a rough draft myself that I then edited using scissors and tape and pencil and white-out. Once a page became too patched up, I made a Xerox copy to create a clean newversion that I could then edit and chop further as needed. Once my draft was ﬁnalized, I paid $500 to a professional typist ($4,300 today) to retype it into a ﬁnal dissertation that I could bind and submit to Widener Library.Ispeciﬁcally recall that my typist used typebar special characters, which would have meant that she used an IBM Model C or D typebar typewriter.” Brian Kernighan created a stripped-down version of [Annotation 12] RUNOFF that he called Roff to print his Princeton PhD in 1968. Roffwas written in Fortran and ran on an IBM 7094. Input was on punch cards, so making a revision required replacing some cards with newones, then submitting 3 boxes of cards (about 20 pounds or 10 kg) to an operator. Kernighan’sthesis also avoided mathematical expressions as much as possible because his output device, an IBM 1403 line printer,couldn’thandle them. Subscripts and superscripts were eliminated, and special characters were inserted by hand after the fact. Roff’sone novelty was that it did automatic capitalization of the ﬁrst letter of each sentence, since punch cards really only supported upper case. In short, typing a mathematical paper in the 1960s was hard work: time consuming, detailed, and laborious. So in practical terms what were Dennis’soptions for typing his thesis? As indicated, manydoctoral students used a department secretary or professional typist to ﬁnalize their theses. This is certainly a possibility,though in this event Dennis still must have exercised close oversight overall character placements. He could have typed it himself, or perhaps he had a Bell Labs technical typist help him with ﬁnal drafts. In late 1967 when he was ﬁnalizing his dissertation he was working at the Labs at Murray Hill NJ and living at his family home in Summit. However, neither family members nor Bell Labs colleagues everheard him discuss anything likethis. Dennis worked on Project MACduring graduate school and had a CTSS account which he kept active while moving to the Labs and starting work on Multics. Bill Ritchie recalls that at some point shortly after starting work, Bell Labs installed a toll-free phone line and a 2741 terminal in his home basement ofﬁce, though he is not certain exactly when. If there was a way to use these to help type his thesis, one would think that Dennis would have ﬁgured it out, especially with RUNOFF as a model. But there is no evidence that he did this, or eventhat he could have done so giventhe technical limitations described above. Dennis nevertalked about his thesis, to colleagues or to family, so no one knows howthe document was actually produced. For the purposes of this paper it really doesn’tmatter.What is important is the actual work product, which

from a typographic perspective isvastly superior to most [Annotation 11] other math dissertations of the period.

3. The Thesis Document The ﬁnal draft of “Program Structure and Computational Complexity” is 180 double-spaced pages and includes nearly 40 different math symbols and Greek letters scattered across manypages. Nearly ev ery page after the introduction contains multiple sub- and superscript expressions, often three layers deep; manypages contain dozens of sub- and superscripted characters. Figure 1 shows half of one comparatively easy page.

Figure1: Half of page 46

3.1. The Basic Grid Clearly the document was typed on a ﬁxed width typewriter, 12 characters per horizontal inch and double spaced at 6 lines per vertical inch. Most typewriters at the time had a manual platen adjustment that allowed 1/12" line height half-steps, which in principle allowed a typist to organize the character placement of anypaper onto a 1/12" by 1/12" grid. We applied a 12x12 grid overevery tenth page to evaluate character placement. The document is a scan of a copyso not everything lines up perfectly but the precision of positioning is evident in the excerpt in Figure 2 and other ﬁgures; a full set of sample pages is available at dmrthesis.net. The equations line up perfectly on the overlaid 1/12" by 1/12" lattice. Every character is placed exactly inside its appropriate cell, with no errors or out of place symbols. For the most part this precision is evident throughout the entire document. Evaluating the thesis after it surfaced, MikeFischer explained “The fact that it was typed on a 1/12" x 1/12" grid is no surprise to me. 12 characters/inch was the standard "elite" typewriter.6lines/inch was also standard, but the

3

#### Page 4

-- --

Figure2: Character positioning

IBM typewriters had a platen that would allowhalf-spacing, i.e., 12 half-spaces per inch. So there was no need to do manual continuous positioning in order to type a page where all letters ended up on the grid. Rather, in order to type a superscript "2", one would roll the paper down one click, type the "2", and roll it back up one click to continue.” In our research we found only a fewmath doctoral theses from the 1960s where sub/superscripts consistently followa 1/12" by 1/12" grid, and none of these were nearly as complex as Dennis’sthesis. One possible explanation is that it wasjust too hard to keep track while typing the individual characters of a complicated expression onto the page one character at a time.

3.2. Vertical Positioning But there are fascinating departures from the grid as well. We start with vertical control, the up and down spacings that create the sub and superscripts. There are manyexamples through the thesis where, evenwith multiple layers and extensive undulations, the characters lie directly in the grid. There are several dozen cases sprinkled throughout the document where characters seem to be wedged into place outside the grid system. In some instances this could be a relaxation, where the typist took a moment offfrom the rigors of the grid and opened the escape mechanism to allowasingle character to be inserted free form. Other times the placement seems intentional, as if an individual character belonged in a half-step between lines. The argument for intentional placement comes into more clear focus when we examine some of the more complex mathematical expressions contained within the document. Forarelatively small but not insigniﬁcant number of instances, a superscript is itself an exponent expression containing its own sub or superscripts. In these cases Dennis appears to have doubled the precision of the vertical grid, centering the exponent on a 1/24" grid so that its own sub/superscripts can center onto the main 1/12" grid.

3.3. Horizontal Positioning and the + Symbol Dennis was also able to split the horizontal grid of 12 characters per inch. We hav e identiﬁed twospeciﬁc situations where he manipulated the document copyonto a 1/24" center in order to reposition one or more characters. Typewriters of the day were not set up to handle this kind of adjustment; repositioning the paper or adjusting the typing mechanism by this small precise amount would be difﬁcult, exacting work. And yet in each of these cases it appears that he made adjustments to widen or justify horizontal spacing purely for aesthetic reasons. When it appears in a subscript or superscript position, the "+" symbol is treated likeother character,with no blank space either side. But when a " + " appears in the baseline of an exponent, Dennis almost always inserted one or more blank characters before and after the symbol to improve visual spacing. His primary spacing rule consisted of positioning the "+" symbol between twoadjacent blank cells, then striking the "+" directly between these cells on a 1/24" center.The "+" symbol thus becomes a 1/12" wide character surrounded by two1/24" blank half-spaces: a single character centered within twocells, as shown in Figure 3, marked with blue arrows.

Figure3: Plus sign example Dennis’ treatment of spacing around the "+" sign stands out as among the least disciplined parts of the document. He kept to his standard “2 cell spacing” about 80% of the time, butthere are plenty of instances when other spacings are used as well. We hav e no idea how it would even be possible to make these horizontal adjustments with the typing machines of the time. Whydid Dennis not simply put a full character space left and right and avoid splitting the grid? He did this a few times; nothing stood in the way of this decision. The "+" sign leaves us with a basket of unresolved issues.

3.4. Roman Numeral Lists Roman numeral lists were a frequent component of the thesis; Dennis used such lists twenty three times. With the ﬁrst ﬁvelists, all typed characters stayed in their respective 1/12" [Annotation 13] by 1/12" cells. Starting with the Roman numeral list on page 39, the spacing changed to create a newtypographic effect. Dennis accomplished this by shifting right the characters of Roman number (i) by 1/12", then shifting the characters on line (ii) right by 1/24", centering both numerals on numeral (iii) below.

4

#### Page 5

-- --

Then every evennumber in each list after this follows this newtypographyrule, adjusting to a 1/24" center alignment; see Figure 4, where the center line is marked.

Figure4: Centered roman numeral list It is as if for this Roman numeral list Dennis found a new skill which he then employed on the remaining 13 lists. We have noidea how he did this. Moreover, the visual distinctions between the center-justiﬁed and non-justiﬁed lists are so subtle as to seem almost gratuitous. Thus we also have no idea why he did this. And yet this treatment of the Roman numeral lists is what Dennis elected, including a collection of 1/24" half steps, seemingly for the pure fun of it.

3.5. Version Control Since Dennis’sdeath in 2011, twoversions of the thesis have surfaced. In January 1968 he had sent copies of a late draft, nearly complete but containing approximately 60 minor typos with pencil corrections noted, to his student colleague Albert Meyer and to his initial thesis advisor Patrick Fischer. These are identical except for a few of his pencil comments in margins. Sometime around 2015, David Brock discovered a thesis manuscript in Dennis’spersonal papers, which had been donated to CHM by the Ritchie family.This version was nearly identical to the January draft except that all 60 typos had been corrected. This undated draft was likely made in late January or February 1968 and was extremely close to ﬁnished; only 6 minor corrections were still needed. Forthe most part the revisions from January to February version were simple typos, replacing a fewcharacters on a single line of copy, as in Figure 5, which shows a handwritten note from page 8 of Albert Meyer’scopy, and the correction on the ﬁnal version. It appears that Dennis was able to replace incorrect copy with newtext in the surrounding original copy, but note the extra space in the corrected version. Other than corrections, the January and February versions are identical in every way: February is not a newversion, but a copy of the January draft except for corrections. To test whether this could really be true, we chose six sample pages at random, scanned January and February versions of each page, then overlaid them to see where differences might emerge. The pages are available at dmrthesis.net. So this leaves us with our ﬁnal mystery: Howcould Dennis have done this? The natural explanation is that he used white-out and a Xerox copymachine. Except that there have

Figure5: Handwritten correction noted and made

been no reports from anyone from this era that white-out / replace / Xerox editing treatment was possible at the precision and reproducibility seen here.

4. Recreating the Thesis Document The previous sections have highlighted anynumber of mysteries about howthe thesis was produced. When we began this work, the thesis was available only as a bitmap PDF from a scanned original. It seemed that the thesis would be more accessible for further study if it could be recreated as a searchable text document. Thus we have recreated the thesis by converting it to troff format, with extensive use of eqn for the mathematical parts. In a sense, this is anachronistic, since nroff,the typewriter-only precursor of troff,dates from about 1972, and eqn wasnot available until 1974, six years after Dennis’sthesis was completed early in 1968. But it seems certain that Dennis would have used nroff, eqn and related tools if theyhad been available to him.

4.1. Font Issues In a previous restoration project [4] we discovered the importance of ﬁnding that a set of fonts created for the Linotron 202 typesetter had been migrated into the modern era in Adobe Type 1 PostScript format. All the mechanical limitations of the 202 could be sidestepped when its fonts could be used on anyPostScript-capable typesetting machine. Forour present work, we have assumed that Dennis did use an IBM 2741 terminal to print his thesis; this seems the most [Annotation 14] likely possibility.Ifonly someone had made the 2741 character set available as a font on a typesetter or laser printer— rather than as a set of metal characters on a golf-ball for a near-obsolete device—then we could makeDennis’sthesis machine-readable and amenable to further research.

5

#### Page 6

-- --

Our font consultant, Chuck Bigelow, determined that the 2741 character set had indeed been converted to a ﬁxed-pitch Bitstream Inc. font called ‘Pica 10’ with the same metrics as the original golf-ball element. Unfortunately this font only includes the standard character set. Despite intensive efforts we have been unable to trace anymodern form of the ﬁxedpitch mathematical symbol and script characters that Dennis so clearly had available on golf-ball elements in early 1968. Fortunately the Adobe Symbol font (for Greek characters and mathematical symbols) and as used in troff and eqn, worked well when supplemented by YandY’sMTMS script collection. Occasionally using the eqn ‘deﬁne’ capability, [Annotation 15] we were able to confect missing symbols such as ‘monus’, which appears at the right-hand end of the ﬁrst lines of Fig- [Annotation 16] ures 6 and 9.

Figure6: Constructing a weird symbol

4.2. Retyping the Thesis Once the Pica 10 font had been tested we realized that a ‘thesis-rebuild’ would be possible, though it would need a major effort to bring it about. We hav e set up a repository web site [5] devoted to creating a canonical bitmap-PDF ﬁle of what the original 180-page thesis was intended to contain, together with source ﬁles for a step-by-step rebuild of the entire thesis, using tbl, eqn and troff,leading to a ﬁnal highquality PDF with full text and graphics. The starting point for the recreation was a set of scannedpage bitmaps, principally created from two of the original 1968 hard copies. More recently,dating from the release of PDF in 1989, it has been possible to enclose these scans in a PDF wrapper. Formaterial of this sort the Adobe Acrobat PDF viewer provides an ingenious device for making the acquired pages seem to be, at least partly,searchable. An OCR engine analyses each page and calculates point-size string widths and bounding box coordinates for all the words it identiﬁes. It then re-typesets the page, often using a generic font such as Helvetica, in what is called TextMode 3. This mode is a sort of electronic ‘invisible ink’ which is fully compliant with all of PostScript’scapabilities for sizing and placing of text, except that it omits the ﬁnal commands that render material and make it visible. If this correctly-sized invisible layer is laid down in exact registration with the underlying bitmap original, a search request by the user, to highlight the word ‘scanner’ (say), delivers the bounding box of that string in the invisible layer, little realizing that the exact registration properties result in the highlight illuminating the bitmap version of the word in the overlaid bitmap layer.

Regrettably this impressive OCR capability is only of limited use to us in the present project. Less than 5% of the pages in the original thesis are in what might be called “plain English.”For this small subset, the good performance of the Pica 10 font under OCR led to a set of pages that simply needed a fewprooﬁng corrections and the insertion of some troff markup. The rest of the pages were all to some degree highly mathematical and the OCR engine, lacking any AI expertise in Dennis-authored material, could not cope with dense strings of what it saw as anomalously long non-existent words. Accordingly,one of us (DFB) spent manyhours with successive pages of the bitmap original displayed in a left-hand workstation window, while he worked out (and typed into a text-editor window) line-by-line, eqn coding to mimic what was on display.BWK did a great deal of proofreading, and occasionally contributed advice on how to use eqn to create exotica likemulti-line braces that appear on manypages, such as those in Figure 7. Figure 8 shows the same display created with eqn.

Figure7: Hand-drawn braces

5. Two Challenges We now takeadetailed look at two of the problems encountered in bringing Dennis’smaterial into the modern era.

5.1. Center Justiﬁcation Revisited We hav e already pointed out that although font characters in Pica 10 were of ﬁxed width, Dennis seemingly had the ability to center-justify a column of Roman numerals around the mid-point of the longest of the strings. At the very minimum this would seem to require a horizontal half-space capability.Figure 9 shows what we are up against. The longest string is (viii),which has an evennumber of characters and so the center of the string lies between the (vi and ii) sub-strings.

6

#### Page 7

-- --

Figure8: Braces produced with eqn [Annotation 17]

Nowconsider the shortest string, which is (i).The midpoint of this string aligns with the center point of the letter i. Visual inspection shows that, somehow, Dennis has managed to get the alignment very close to perfect. The task is to achieve the same effect using the troff typesetting programs. There is no macro in the standard ms macro package that can achieve this effect and eqn has no facility either.This may well be because anyimplementation of the effect requires look-ahead to ﬁnd the longest string in the grouping followed by calculation of its mid-point. Giventhat all the characters in the Pica 10 font are of ﬁxed width it is possible to pad out the strings with thin spaces to makeeverything line up for center alignment. But ideally it would be better to have a more general facility that could cope with text strings in variable-width fonts. Fortunately Michael Lesk’s tbl program [6] for typesetting tables provides the necessary facility.Figure 9 may be thought of as a two-column table. The left-hand column contains the Roman numeral strings that are to be centered. The next column is a left-justiﬁed set of text strings which deﬁne the assumptions inherent in the overall Lemma. When coping with the left column the tbl preprocessor uses the fact that low-levelcommands in troff itself enable text string lengths to be computed accurately, in machine units provided the individual character widths are known. So tbl’s task is to emit code that causes troff to measure the width of all strings and also to emit low-levelinterleavedspacing code for troff to obey, that will cause the entire left-hand column to be center-justiﬁed.

Figure9: Original centered roman numerals

Figure 10 shows that center-justiﬁcation of the entire lefthand column has indeed taken place.

Figure10: Figure 9 with Bitstream Pica 10 and tbl

5.2. Line Drawings Our second example where we have allowed ourselves the luxury of ‘anachronistic’ improvements is in the diagram of Figure 12.5 in the thesis. Figure 12.5, shown here as Figure 12, is a complexbox-and-line diagram showing the way that different varieties of loop programs are inter-related. In the early 1970s devices such as the 2741 or a Teletype Model 37 could be drivenfrom a computer using nroff. Dennis would have been well aware that such diagrams could only be drawn with hyphens or underscores for horizontal lines, together with sequences of bar | symbols for vertical lines. The results from these techniques were never elegant. For all these reasons we can conjecture that he thought it better to place his character symbols on the page by careful measurement and dead-reckoning, leaving the

7

#### Page 8

-- --

boxes and lines (sometimes dotted or dashed) to be drawn in by hand. Figure 12 shows the hand-boxed original page where the various box and line elements have clearly been hand-drawn, around carefully placed characters in a script font. Manyof the script characters are not exactly elegant. Clearly Dennis had problems ﬁnding golf-balls for all the characters he needed. We used pic [7] to create a version of the ﬁgure, Figure 13. It can place elements not just at absolute positions but also at positions relative to box and line boundaries, and other places within a diagram. So as not to stray too far from the constraints on the original material we did use absolute coordinates to place the original lines, boxes, etc.

6. Conclusion It seems that the more we try to ﬁgure out howDennis produced his thesis, the less certain we are in anyconclusions we might draw. For example, the typed material is remarkably precisely laid out on the page, matching a grid almost perfectly.How did he (or some typist) manage to sustain such precision for 180 pages, with endless sequences of subscripts on superscripts? Howdid he manage the fractional spacing, especially horizontally,where so far as we know, the devices of the day did not provide a mechanism. Howdid he deal with the multitude of characters—Greek, Fraktur,mathematical symbols, script letters—without losing track of their positions on the page. And as a weird aside, whyare there twoversions of the digit "4", one with a closed top and one with an open top, shown in Figure 11?

Figure11: Tw o versions of the digit 4

Howdid he drawall the brackets around the vertical constructs of loop programs? The braces are not perfect— clearly theyhav e been done by hand—but theyare very precise. Howdid he manage the sole large diagram, Figure 12.5, with its intricate patterns of dashed and dotted lines? Howdid he manage hyphenation? Howdid he manage to keep pages all approximately the same height? And so on: more questions than answers. We hav e greatly enjoyed working on this project and we feel that there is a lot of mileage left in it yet. It is in manyways a blessing that Dennis’sthesis did go missing for so long. Anyattempt at early restoration, in the period from 1968 to 1985, would have had to use the newly developed tools such as troff, eqn, tbl and pic to confect a version of the document that was suitable for an electromechanical device such as a daisywheel printer or evena

2741. It wasnot until the mid-1980s that economical laser printers at decent resolution became widely available. Most important was the development of PostScript by Adobe in 1984, followed by its more declarative form, PDF,in 1989. PDF has nowbecome the de facto standard for interchanging visually complexmaterial such as Dennis’sthesis. And along with PostScript and PDF came font technologies such as Type1, TrueType and Open Type, which made possible our starting point of the Pica 10 font as a faithful representation of what was originally available on engravedgolfball elements. Of course Dennis worked at Bell Labs from 1967 through to his death in 2011 and he was certainly aware of (and a user of) all the Unix-based text processing tools that we have mentioned in this paper.Sadly,whateverhappened in 1968 seems to have killed offany chance of him using these tools to have another go at rebuilding and submitting his thesis. Despite the important result from his research that loop programs provide another Turing-complete computability model, it was still the case that working with Ken Thompson on the Unix operating system and the C compiler was much more in line with what he wanted to do in his professional career.Although Dennis did not get his Harvard doctorate, we should be thankful that this did not stop him from a lifetime of amazing contributions to practical computing.

Acknowledgements Katie LaSeur of TheCreativeFold.com was enormously helpful with preparation of ﬁgures for this paper and for dmrthesis.net. Weare also grateful to Steve Bagley, Chuck Bigelow, David Brock, Stu Feldman, MikeFischer,Harry Lewis, Doug McIlroy, Sean Riley, John Ritchie, Jerry Saltzer and Tom Van Vleck for generously sharing their expertise, experience, and memories.

References [1] David C. Brock, Discovering Dennis Ritchie’sLost Dissertation,

https://computerhistory.org/blog/discovering-dennis-ritchies-lostdissertation, June 2020. [2] William A. Ritchie, Dennis Ritchie’s"missing" PhD thesis.

https://dmrthesis.net [3] Brian W. Kernighan and Lorinda L. Cherry,“ASystem for Typeset-

ting Mathematics,”Communications of the ACM, vol. 18, no. 3, p. 151-157, 1975. [4] Steven R. Bagley, David F.Brailsford, and Brian W.Kernighan,

“Revisiting a Summer Vacation: Digital Restoration and Typesetter Forensics,” in Proceedings of the ACM Symposium on Document Engineering (DocEng13), p. 3-12, ACM Press, 10-13 September 2013. DOI: 10.1145/2494266.2494275 [5] Repository of material for this paper: www.cs.prince-

ton.edu/˜bwk/dmr [6] M. E. Lesk, Tbl—A Program to Format Tables, 1976. Bell Labs

memorandum [7] B. W. Kernighan, “PIC—A Language for Typesetting Graphics,”

Software--Practice and Experience, vol. 12, no. 1, p. 1-21, January, 1982.

8

#### Page 9

-- --

Figure12: Original Figure 12.5

9

#### Page 10

-- --

Figure13: Pic version of Figure 12.5

10

---

Source: [Fermat's Library](https://fermatslibrary.com/) — https://fermatslibrary.com/s/how-did-dennis-ritchie-produce-his-phd-thesis-a-typographical-mystery. More annotated papers: https://fermatslibrary.com/llms.txt
