Vannevar Bush

Vannevar Bush

Engineer and science administrator · 1890–1974 · United States

Bush's work joins three distinct problems: calculating a changing physical system, finding useful connections among documents, and organizing research institutions. They deserve separate mechanisms and separate evidence. A mechanical differential analyzer is not a digital computer, and a proposed personal microfilm library is not the World Wide Web.

Historical bearings

Vannevar Bush was born on 11 March 1890 and died on 28 June 1974. After study at Tufts and doctoral work associated with MIT and Harvard, he joined MIT's faculty in 1919. He led a collaborative programme that produced the differential analyzer described in 1931. Harold Hazen, Samuel Caldwell, and other engineers and students contributed to this work. The larger Rockefeller machine followed, with a demonstration in 1941. Calling all these machines one man's invention obscures the engineering effort. National Academy memoir

Bush became director of the Office of Scientific Research and Development when it was established in 1941. OSRD coordinated important wartime research through contracts. This does not mean that Bush personally directed every American scientist or every aspect of the Manhattan Project. Administrative authority, military command, and laboratory scientific leadership are different roles. Library of Congress collection overview

Worked mechanism: turn a differential equation into connections

This is an original mathematical model of analog integration, not a reconstruction of a particular Bush machine setup. A differential analyzer represents variables by continuous physical quantities such as shaft rotation. An integrator produces a quantity proportional to an accumulated input. Scaling, initial conditions, and connections determine the equation being represented. Mechanical integration overview

Consider the oscillator

y'' = -4y
initial conditions: y(0)=1, y'(0)=0.

Introduce velocity v=y'. The system becomes v'=-4y and y'=v. A conceptual arrangement has two integrators:

y --multiply by -4--> v' --integrate, initial 0--> v
v ---------------------> integrate, initial 1--> y

The output y feeds back into the first input. The coefficient 4 is represented by an appropriate scale ratio, and the negative sign requires sign reversal. The equation has solution y(t)=cos(2t) and v(t)=-2sin(2t). At t=π/4, y is zero and v is -2; at t=π/2, y is -1 and v is zero. Those values provide independent checkpoints for a plotted curve.

An invariant gives a stronger check. Let E=v²+4y². Differentiate:

E' = 2v v' + 8y y'
   = 2v(-4y) + 8yv
   = 0.

The initial value is E=4. A large drift away from 4 signals error in the physical implementation, setup, or readout. The invariant checks the modeled dynamics; it does not guarantee that a device is frictionless or calibrated.

Now approximate the same equations digitally with simultaneous forward Euler steps of size h:

y_next = y + h v
v_next = v - 4h y

Substitution gives E_next=(1+4h²)E. At h=0.1, each step multiplies energy by 1.04. After ten steps, E≈4×1.04^10≈5.921. The continuous system conserves energy, but this numerical method systematically increases it. This comparison is educational: neither “analog” nor “digital” alone determines accuracy. The physical apparatus and the numerical algorithm each introduce their own errors.

Memex: links are a design choice

In As We May Think, published in July 1945, Bush described a proposed personal microfilm system with retrieval controls, annotations, and associative trails connecting records. The proposal emphasized organizing and reusing intellectual connections. It did not specify network addresses, HTTP, distributed servers, or the Web's document formats. Original essay

Here is an original representation exercise. Suppose a researcher has records A, B, and C. A subject index can map “oscillators” to {A,C}. A trail can instead preserve an ordered argument: A's experimental observation, then B's criticism, then C's revised model. A trail is more than a set of search results because ordering and annotations can record the researcher's reasoning.

A modern system might store a trail as ordered pairs of stable record identifiers and comments. That is a design analogy, not a claim about Bush's microfilm encoding. A robust implementation must decide what happens when a record changes, when two trails share a record, and when a recipient lacks a referenced document. Those problems show how far a suggestive concept is from a complete interoperable system.

Influence can sometimes be documented directly. In a 1962 letter, Douglas Engelbart recalled reading Bush's essay in 1945 and described its importance to his programme for improving intellectual effectiveness. This testimony supports a particular connection. It does not establish every arrow in a retrospective family tree of the Web. Engelbart's letter

Institutions are designed too

Bush's 1945 report Science, the Endless Frontier argued for federal support of basic research and proposed a new funding institution. Read it as a policy argument: it makes claims about public benefit, researcher autonomy, and how responsibility should be allocated. Original report

NSF was established in 1950 after legislative revisions. Its origin cannot be reduced to one report instantly becoming law. Nor does advocating basic research prove that all innovation follows a one-way sequence from discovery to product. These are historical and causal claims that require separate evidence. NSF institutional history

Exercises

1. For y''=-9y, with y(0)=2,v(0)=0, specify the two-integrator arrangement, the exact solution, and a conserved quantity. Compute the conserved quantity's initial value.
2. Starting from y=1,v=0, take two Euler steps with h=0.1 for y''=-4y. Calculate the energy after each step and compare it with the exact invariant. Explain why reducing h improves a fixed-duration simulation but does not make forward Euler exactly conservative.
3. Design a three-record trail that includes a claim, contrary evidence, and a conclusion. Specify how your system would identify records and preserve annotations when the underlying records change. Explain which parts are your implementation choices and which general idea is supported by Bush's essay.

Compare your reasoning with the worked solutions

1. Feed -9y into the velocity integrator, initialized at zero; feed v into the position integrator, initialized at 2. The solution is y=2cos(3t), v=-6sin(3t). The quantity v²+9y² has derivative 2v(-9y)+18yv=0, and its initial value is 36. The factor 9 and initial position 2 must be encoded separately; changing one is not equivalent to changing the other.
2. The first step gives y=1,v=-0.4, hence E=4.16. The second uses both old values: y=0.96,v=-0.8, hence E=0.64+4×0.9216=4.3264. Each equals the previous energy times 1.04. Over duration T, the factor is (1+4h²)^(T/h), which approaches 1 as h tends to zero. For every positive h and nonzero state, however, each step increases energy.
3. One valid design stores (record ID, version ID, annotation) at each ordered step. The first annotation states the claim, the second records a challenge, and the third explains the revised conclusion. Immutable versions preserve what the trail's author saw; stable record IDs can separately lead to newer versions. Bush's essay supports associative trails and annotation, not this particular versioning schema. A user interface should show when a cited version has been superseded rather than silently rewriting the evidence.