
What this covers
This Veritasium video by Derek Muller works as a history of nuclear test-ban diplomacy that happens to contain an algorithm. It begins with the postwar moment when the U.S. was open to international control, including the Baruch Plan proposal to put all radioactive materials under an international body. It then moves through the 1954 Shrimp test at Bikini Atoll, which yielded two and a half times its expected energy and sent fallout over nearby atolls and a Japanese fishing boat. That test fed the public outcry that produced the peace symbol. The 1958 Geneva conference and its pause in testing follow, with Richard Garwin and John Tukey among the American scientists arguing over seismometers with the Soviet delegation.
The technical thread is verification. Underground blasts keep most of their radiation contained, and the Soviets treated onsite inspection as espionage. Distant seismometer readings are also ambiguous, because a deeper explosion looks smaller for a given yield. Telling a bomb from an earthquake therefore meant heavy spectral analysis. A brute-force discrete Fourier transform needs N² multiplications, which for a million samples would take years on 1960s machines. The video follows Tukey's insight at a 1963 science advisory meeting attended by Kennedy, and Garwin's approach to James Cooley at IBM, who was not told the real motive. It also covers the 1965 Cooley–Tukey paper and the discovery that Carl Friedrich Gauss had derived the method in 1805 while studying asteroid orbits, then lost it in non-standard Latin notation in his posthumous collected works.
The range then widens to the Fast Fourier Transform) in streaming, radar, sonar, 5G and WiFi. The most contested claim is counterfactual: the 1963 Partial Test Ban Treaty banned only testing that could be verified and pushed the arms race underground, where warheads grew to roughly 70,000 by the mid-1980s. Garwin says an earlier FFT "might have turned the tide." The video closes with a sponsor segment on career choice that frames a career as 80,000 hours.
The Fast Fourier Transform, discovered by Gauss in 1805 but rediscovered by Cooley and Tukey in 1965, is the most important algorithm of our time, enabling signal processing across radar, communications, and compression; its delayed adoption prevented earlier detection of covert nuclear tests and may have changed the course of the Cold War nuclear arms race.
- FFT reduces discrete Fourier transform calculations from O(N²) to O(N log N), making large-scale signal analysis computationally feasible
- Scientists needed FFT-level detection capability in early 1960s to verify underground nuclear test compliance and secure a comprehensive test ban before arsenals proliferated
- Gauss derived the FFT algorithm in 1805 but published it posthumously in untranslatable Latin notation, delaying its impact by 160 years
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
The Fast Fourier Transform was actually discovered earlier than Cooley and Tukey—mathematician Carl Friedrich Gauss derived it in 1805 while studying the orbits of newly discovered asteroids Pallas, Ceres, and Juno.
“it was discovered earlier but then forgotten. All the way back in 1805, Mathematician Carl Friedrich Gauss was studying the newly discovered asteroids of Pallas, Ceres and Juno. To determine the orbit of Juno, Gauss devised a novel approach to harmonic analysis and he jotted it down in his notes”
At the end of World War II, Canada and the United Kingdom requested a meeting with the U.S. to discuss international control of nuclear weapons, and the U.S. was open to these discussions rather than assuming unilateral dominance was inevitable.
“At the end of the war, Canada and the U.K. requested a meeting to discuss what would be done about nuclear weapons. And contrary to what I expected, the U.S. was actually open to these discussions.”
In 1954, the U.S. tested a thermonuclear device called Shrimp at Bikini Atoll that used lithium deuteride as fuel; the explosion was expected to release energy equivalent to six million tons of TNT but actually released two and a half times that amount due to unanticipated reactions with lithium-7.
“In 1954 at Bikini Atoll in the South Pacific, the U.S. tested a new thermonuclear design that used lithium deuteride as a fuel and they expected the device code named Shrimp to release the energy equivalent of six million tons of TNT but they were wrong. It released two and a half times that due to unanticipated reactions with lithium-7.”
The Shrimp test and its fallout triggered public outcry against nuclear testing, leading to the modern peace sign being designed in the 1950s by combining semaphore letters N and D for nuclear disarmament.
“These events triggered public outcry against nuclear testing. The modern peace sign was designed in the 1950s by combining the semaphore letters for N and D standing for nuclear disarmament.”
The breakthrough in fast DFT computation came in 1963 at a meeting of the President's Science Advisory Committee where President John F. Kennedy was present along with Richard Garwin and John Tukey, and Tukey discovered how to compute discrete Fourier transforms with far fewer calculations while doodling during a boring meeting discussion.
“And the breakthrough came in 1963 at a meeting of the President's Science Advisory Committee. President John F. Kennedy was there, as were Garwin and Tukey, the physicist and mathematician from the Geneva meeting. Although they were discussing issues of national importance like the Apollo program and nuclear fallout shelters, the meeting was apparently pretty boring. Garwin observed Tukey doodling throughout but what he was actually doing was working on discreet Fourier transforms.”
Richard Garwin approached IBM researcher James Cooley to program the FFT algorithm on a computer, but did not tell him the motivation was detecting covert Soviet nuclear tests; instead, Garwin claimed the purpose was to work out the spins in a crystal of helium-3.
“Garwin approached an IBM researcher, James Cooley to program a computer to run the FFT algorithm but he didn't tell him the reason was to detect underground Soviet nuclear tests. He said it was to work out the spins in a crystal of helium-3.”
The U.S. offered under the Baruch Plan to decommission all its nuclear weapons if other nations pledged never to develop them, proposing that an international body control all radioactive materials globally from mining through peaceful nuclear power generation.
“They realized that they wouldn't be the only nation with nukes for long. So they offered to decommission all their nuclear weapons if other nations would pledge never to make them. This was known as the Baruch plan and it proposed that an international body would control all radioactive materials on earth, from mining to refining to using these materials for peaceful purposes as a nuclear power generation.”
Richard Garwin, when asked about the counterfactual, said an earlier FFT discovery 'would've helped and it might have turned the tide,' validating the historical contingency argument.
“To check how realistic this is, I asked Richard Garwin himself... 'Well, it's a good story.' It is a good story. 'It would've helped and it might have turned the tide.'”
Gauss's FFT algorithm was lost to history because it only appeared after his death in volume three of his collected works, and was written in non-standard notation in 19th-century Latin, making it inaccessible to later researchers.
“The reason his breakthrough was not widely adopted was because it only appeared after his death in volume three of his collected works and it was written with non-standard notation in a 19th century version of Latin.”
The Shrimp test's radioactive fallout affected neighboring atolls with residents only evacuated three days after the test, causing acute radiation sickness, and also affected a Japanese fishing boat crew 23 strong, one of whom died from radiation complications six months later.
“more radioactive material was created and it rained down over a much larger area than planned. Residents of neighboring atolls were only evacuated three days later suffering from radiation sickness. And further east, the 23 crew of a Japanese fishing boat were caught in a flurry of radioactive white ash. And by that evening, they were suffering from acute radiation syndrome. One of the crew died from ensuing complications six months later.”
The 1963 Partial Test Ban Treaty banned atmospheric, underwater, and space-based nuclear testing—places where compliance could be verified—but did not ban underground testing because it was almost impossible to verify compliance.
“And this is ultimately why when a test ban treaty was signed in 1963, it was only a partial ban. It banned testing underwater, in the atmosphere and in space, places where compliance could be verified, but it didn't ban testing underground for the simple reason that it was almost impossible to verify compliance.”
At the 1958 Geneva meeting, American physicist Richard Garwin and mathematician John Tukey debated with the Soviet delegation over which nation's seismometers were superior; Garwin simulated both countries' devices on a computer at CERN and the next day everyone agreed there wasn't much difference.
“In the American delegation at the Geneva meeting, there was a physicist, Richard Garwin and a mathematician John Tukey. They got into a debate with the Soviet delegation over which nation's seismometers were superior. So Garwin simulated the responses of both countries' devices on a computer at CERN. The next day, everyone agreed there wasn't much difference.”
Cooley and Tukey published the Fast Fourier Transform algorithm in a seminal 1965 paper, and its use immediately took off across science and engineering.
“Cooley and Tukey published the algorithm in a seminal 1965 paper and its use immediately took off”
The Fast Fourier Transform is used ubiquitously in modern signal processing including video streaming, radar, sonar, 5G, and WiFi applications.
“you use it all the time including right now to watch this video and it's used in radar and sonar, 5G and WiFi. Basically, anytime a signal is processed, there's a good chance that a Fast Fourier Transform is involved.”
The U.S. conducted nuclear weapons testing in Nevada, and the radioactive fallout blew across the country, resulting in Americans outside of Japan being the most adversely affected by American nuclear weapons.
“The U.S. also created a nuclear testing site in Nevada, the radioactive products of which blew across the country so that outside of Japan, the people most adversely affected by American nuclear weapons were Americans themselves.”
In an average career, someone works 40 hours per week, 50 weeks per year, for 40 years, totaling 80,000 hours of work, meaning that a career might be the biggest opportunity to make a difference in the world.
“You know in an average career, you work 40 hours a week, 50 weeks a year for 40 years and that works out to 80,000 hours. It means that your career might be your biggest opportunity to make a difference in the world.”
The partial test ban treaty sent the nuclear arms race underground rather than deescalating it; nations that could only test underground began testing extensively to avoid falling behind
“And the partial test ban treaty, far from deescalating the nuclear arms race just sent it underground. The thinking was if you were only allowed to test underground, then you better be testing extensively so as not to fall behind all the other nuclear states”
The core problem with negotiating a comprehensive nuclear test ban was verifying that the other side would comply, with the U.S. worried the Soviets would test covertly and leapfrog them technologically, and vice versa.
“But there was a big problem with negotiating a comprehensive test ban which is how do you know that the other side will hold up their end of the bargain? The U.S. worried that the Soviets would continue testing covertly and leapfrog them technologically and the Soviets similarly distrusted the U.S..”
The atomic bomb dropped on Hiroshima released approximately one thousand times more energy than the largest conventional explosive of that era.
“The bomb dropped on Hiroshima released a thousand times more energy than the biggest conventional explosive”
Underground nuclear tests are much harder to detect from afar because their radiation is mostly contained, and the Soviets refused to allow onsite compliance visits which they regarded as espionage.
“But underground tests are much harder to detect from afar because their radiation is mostly contained and the Soviets refused to allow onsite compliance visits which they regarded as espionage.”
The extensive underground testing during the Cold War arms race facilitated the construction of an enormous number of nuclear warheads; at the peak in the mid-1980s, 70,000 nuclear warheads were in existence globally.
“This testing facilitated the construction of an absurd number of nukes. At the peak in the mid 1980s, 70,000 nuclear warheads were in existence.”
Seismometer signal depends not only on the yield of a nuclear device but also on how deep it was buried—deeper explosions appear smaller for a given yield—creating ambiguity that scientists needed to resolve to determine whether a signal was from a bomb and how big and how deep.
“In addition, part of the purpose of detecting your adversaries tests is to spy on them, to know how big of an explosion they can make. But the seismometer signal depends not only on the yield of the device but also on how deep it was buried. For a given yield, deeper explosions appear smaller.”
The brute-force discrete Fourier transform requires N² complex multiplications to transform N data points, meaning a million-sample signal would require one trillion calculations, taking over three years on 1960s computers.
“To complete one discrete Fourier transform requires multiplying N data points by N different frequency waves. So N squared complex multiplications. Now this is doable for eight samples but if you had say a million samples, that would require a million squared or one trillion calculations. At the speed of 1960s computers, that would take over three years to complete, all for a single transform.”
World leaders called for a comprehensive nuclear test ban treaty, which was taken seriously by nuclear powers who entered into negotiations at the Conference on the Discontinuance of Nuclear Weapon Tests held in Geneva in 1958, and stopped all testing during negotiations so that 1959 was the only year in a long stretch when no nuclear weapons were detonated.
“A number of world leaders called for a comprehensive test ban, no more testing of nuclear weapons by any state and this call was actually taken seriously by the world's nuclear powers. They entered into negotiations at meetings with very literal names like the Conference on the Discontinuance of Nuclear Weapon Tests held in Geneva in 1958. And to show just how serious they were, they even stopped all testing during the negotiations which is why 1959 is the only year in a long stretch when no nuclear weapons were detonated.”
The Soviet Union rejected the Baruch Plan, viewing it as a mechanism for the U.S. to maintain nuclear dominance rather than a genuine disarmament offer, which led directly to the initiation of the global nuclear arms race.
“But the Soviets rejected the proposal. They saw it as just another ploy to maintain American nuclear dominance and so the global nuclear arms race began.”
Image compression using FFT works by taking the pixel brightness values of each row of an image, performing FFT to determine which frequencies are present in the brightness values, and then performing another FFT for the columns of pixels (the order doesn't matter) to get a 2D FFT of the original image.
“Here's how the Fast Fourier Transform allows you to compress an image. You take the pixel brightness values for each row of an image and perform a Fast Fourier transform. So essentially, you're figuring out what frequencies are present in the brightness values of the rows of an image. Here, brightness represents the magnitude of each frequency component and the color represents the phase, how shifted that frequency is to the left or the right. And then you perform another FFT for the columns of pixels. It doesn't matter if you do the rows or columns first, which you need is a two dimensional FFT of the original image.”
Detecting atmospheric nuclear tests was straightforward because radioactive isotopes disperse in the atmosphere and can be detected thousands of kilometers away, and underwater tests produce distinctive sounds that travel efficiently around a kilometer below the ocean surface and can be picked up by hydrophones.
“Now detecting atmospheric tests was fairly straightforward. The radioactive isotopes they produced disperse in the atmosphere and can be detected thousands of kilometers away. Underwater tests produce distinctive sounds that travel efficiently around a kilometer below the surface of the ocean and these can be picked up by hydrophones.”
Approximately 300 earthquakes with magnitude three or greater occur each day worldwide, making the task of distinguishing nuclear test signals from earthquake signals at that magnitude scale extremely difficult.
“Every day around the world, there are close to 300 earthquakes with a magnitude of three or greater.”
A Fourier transform is a mathematical method of decomposing a signal into pure sine waves, each with its own amplitude and frequency, that combine additively to reconstruct the original signal, and the sum of sine waves can look arbitrarily complicated while bearing no visual resemblance to its component parts.
“A Fourier transform is a way of decomposing a signal into pure sine waves, each with its own amplitude and frequency that add to make it up. This seems a bit like magic since the sum of a set of sine waves can look arbitrarily complicated and nothing like its component parts.”
In a Discrete Fourier Transform, the frequency spectrum is also discrete and finite, with frequencies falling into a finite number of bins whose number and size are determined by the number of samples in the signal and how closely spaced the samples are.
“And one of the distinguishing features of a Discreet Fourier Transform is that the frequency spectrum is also discreet and finite. You can think of the frequencies as falling into a finite number of bins. And what determines the number and size of these bins is the number of samples in the signal and how closely spaced they are.”
The more spaced out the samples are in a signal, the lower the maximum frequency that can be measured, because the samples aren't close enough together to capture high-frequency oscillations.
“For example, the more spaced out the samples are, the lower the maximum frequency you can measure. Because the samples aren't close enough together to capture high frequency oscillations”
The Fast Fourier Transform algorithm works by exploiting the periodic and overlapping nature of sinusoids: due to redundancy in the frequency wave computations, a calculation expecting to require 64 multiplications (8×8) can be reduced to only 24 multiplications.
“Consider the example from before with eight samples. Each of those data points must be multiplied by the eight different frequency waves. Here I'm only showing cosines for simplicity. So you would expect this to require eight times eight or 64 complex multiplication. But due to the periodic nature of sinusoids, these waves of different frequencies overlap in a predictable way... So instead of performing 64 calculations, you need only 24.”
To determine how much of a particular sine wave is present in a signal, multiply the signal by that sine wave at each point and sum the area under the curve; if the frequencies don't match, the positive and negative areas cancel to zero, indicating that frequency is not in the signal.
“If you wanna know how much of a particular sine wave is in a signal, just multiply the signal by the sine wave at each point and then add up the area under the curve... if you multiply this signal by a sine wave of an arbitrary frequency, the two waves are uncorrelated. You're just as likely to find places where they have the same sign, both positive or both negative as where they have opposite signs and therefore when you multiply them together, the area above the x-axis is equal to the area below the x-axis. So these areas add up to zero which means that frequency sine wave is not a part of your signal”
If Gauss's FFT algorithm had become widely adopted after his discovery in 1805, the FFT may have transformed our world even more dramatically than it has, as most people don't think about the cumulative impact of their life's work.
“If only it had become more widely adopted after Gauss discovered it, the FFT may have even more dramatically transformed our world. Now I don't think Gauss could ever have imagined how important the FFT would be, just as most people don't think about the cumulative impact of their life's work.”
The real obstacle to detecting underground explosions wasn't seismometer accuracy but the vast amounts of computation required to Fourier transform seismometer signals.
“The real obstacle to detecting underground explosions wasn't the accuracy of the seismometers, it was the vast amounts of computation required to Fourier transform the seismometer signals.”
If scientists had been confident in their ability to remotely detect underground nuclear tests in the early 1960s using FFT-level signal processing, then a comprehensive test ban could have been achieved, potentially stopping the nuclear arms race before it proliferated to 70,000 warheads.
“If only scientists had been confident in their ability to remotely detect underground tests in the early 1960s, then a comprehensive test ban could have been reached, stopping the nuclear arms race before it really got going.”
Total spending on nuclear weapons in the 20th century is estimated at approximately $10 trillion each for the United States and Soviet Union when adjusted for inflation.
“Total expenditure on nukes in the 20th century is estimated at around $10 trillion each for the U.S. and the Soviet Union adjusting for inflation.”
Using Euler's formula, you can calculate sine and cosine amplitudes with a single complex exponential term: the real part is the cosine amplitude and the imaginary part is the sine amplitude
“You can calculate these sine and cosine amplitudes separately or you can use Euler's formula so you only need to multiply your signal by one exponential term. Then the real part of the sum is the cosine amplitude and the imaginary part is the sine amplitude”
With modern seismometers, computing, and the Fast Fourier Transform, detection capability enables identification of magnitude 3 seismic events (approximately 1 kiloton explosions) anywhere on earth
“With our modern network of seismometers and computing and the Fast Fourier Transform, today, we have the ability to detect magnitude three events which correspond to a one kilo ton or so explosion, basically anywhere on earth”
By the time the FFT was published in 1965, the UK, France, and China had joined the Soviet Union and U.S. as nuclear powers
“By that time, the U.K., France and China had joined the Soviet Union and the U.S. as nuclear powers”
For almost all real-world images, most values in the FFT transform are close to zero, especially toward the edges where high frequencies are located, allowing 99% of the transformed image information to be discarded while still preserving a good representation
“For almost all real world images, you find that a lot of the values in the transform are close to zero. I've taken the log of the transform values so you can see them but if I don't, then it's clear most of the values, especially toward the edges are very small and these correspond to high frequencies. And what this means is that you can throw out a lot of the information in the transformed image say 99% of it. But when you invert that result, you still get a fairly good representation of the original image”
Mathematician Gilbert Strang called the FFT 'the most important numerical algorithm of our lifetime'
“mathematician Gilbert Strang called the FFT the most important numerical algorithm of our lifetime”