Why Exploding Stars Demand Good Explanations

What supernovae teach us about the creativity of explanatory knowledge and the limits of LLMs.

Jul 11, 2025

Supernova 1987A (SN 1987A) exploded in February 1987

(OK, so in previous posts I have been banging on about what David Deutsch calls "good explanations" without perhaps being clear about why they are so fundamental to the problems with the current state of generative AI (LLM) development

So here is my attempt at describing an example of how good explanations come about by human creativity, ones that could not have been arrived at merely by an incremental remixing of existing knowledge - which is what LLMs do.)

When a massive star dies, it creates one of the universe's most violent and spectacular events. The story of supernova core collapse reveals what physicist and philosopher David Deutsch calls a "hard-to-vary" good explanation. Let me take you on a journey from stellar death to the deepest principles of how we understand reality itself.

Imagine a massive star, at least eight times heavier than our Sun, spending millions of years in a delicate dance between gravity trying to crush it inward and nuclear fusion pushing outward. For most of its life, fusion wins. The star burns hydrogen into helium, then helium into carbon, carbon into oxygen, and so on, creating concentric layers like a cosmic onion.

This stellar alchemy has a fatal endpoint: iron. When the star's core begins fusing silicon into iron, everything changes. Because iron fusion absorbs energy rather than releasing it. At this fatal point in it life, the star loses its eternal battle against gravity.

What happens next is almost incomprehensibly violent. Without nuclear fusion pushing outward, gravity crushes the iron core from Earth-sized down to less than 20km (~12 miles) across in less than a second. The core becomes so dense that protons and electrons get smashed together to form neutrons, creating what astrophysicists call a neutron star. A teaspoon full would weigh as much as Mount Everest.

This collapse releases an enormous shockwave that bounces off the ultra-dense neutron core and races outward through the star's layers at thousands of kilometres per second. The shockwave heats everything to billions of degrees and blasts the star's outer layers into space. The explosion becomes so bright it can outshine an entire galaxy of a hundred billion stars for weeks, scattering heavy elements throughout the universe. These ejected materials that will eventually form new stars, planets and life itself (including you!).

During the core collapse, the extreme conditions create an enormous flood of neutrinos. These are “ghostly” particles that barely interact with matter. At one time they were believed to be massless but developments in asteroseismology have shown that this is not the case, more later. These neutrinos carry away most (~99%!) of the supernova's energy, streaming out in all directions at nearly the speed of light.

Along their way, something remarkable happens during their journey. The neutrinos produced in the supernova core start out predominantly as one type known as electron neutrinos. But neutrinos come in three "flavours": electron, muon, and tau neutrinos. As these particles travel the vast distances from distant supernovas to Earth something called neutrino oscillation occurs.

The three neutrino types do in fact have slightly different masses, and quantum mechanics causes them to continuously transform from one flavour to another as they journey through space. By the time they reach our detectors on Earth, only about one-third remain as electron neutrinos which is the the type our detectors are best at catching. The other two-thirds have transformed into muon and tau neutrinos, which largely slip through our instruments undetected. See ‘Background: The solar neutrino problem’, below.

This discovery solved a major puzzle in physics which was ultimately explained by the insight that neutrinos must have mass, confirming the findings from asteroseismology mentioned earlier. This was a breakthrough that fundamentally changed our understanding of particle physics and earned multiple Nobel Prizes.

OK, so that ‘s a nice story… but why is it a good explanation? And why does it fit the bill of being hard-to-vary? How do we know it is even true? (And what has it got to do with LLMs?) No human has ever directly observed a supernova core collapse. No one has watched neutrons form in real-time or seen neutrinos change flavours. Yet we're confident these theories describe reality with extraordinary accuracy and fit the criteria what David Deutsch calls a "hard-to-vary" good explanation.

Deutsch argues that good explanations are those that are difficult to modify while maintaining their explanatory power. Bad explanations, by contrast, can be easily tweaked to accommodate almost any observation. The supernova story is a very good example of this principle in action.

Consider how tightly interconnected every aspect of the explanation is. The critical mass at which core collapse begins (the Chandrasekhar limit) comes from extremely precise calculations in quantum mechanics and stellar physics. The millisecond collapse time follows directly from gravitational equations. The formation of neutrons requires specific nuclear physics. The shockwave rebound depends on the equation of state of ultra-dense matter.

You cannot arbitrarily change any piece of this explanation. You can't say "maybe cores collapse more slowly in different galaxies" or "perhaps the mass limit is different now than in the past" without breaking the entire coherent framework. Each component supports and constrains every other component, creating a mutually reinforcing coherent web of explanatory knowledge that reaches from quantum mechanics to galactic astronomy.

To me, the neutrino oscillation discovery makes this extraordinary. When detectors found only one-third of the expected neutrinos, scientists didn't just shrug and say "close enough." The discrepancy demanded an explanation and the asteroseismologists piped up. Their studies of the physical conditions inside stars constrain what processes are possible and which ones aren’t, these implications required neutrinos to have mass and flavour mixing Otherwise their observations made no sense. This conceptual and creative leap was entirely new physics that's now essential to our understanding of the universe. (I will include a link below to the work done at The University of Birmingham that helped resolve this stellar sized conundrum).

This “story” makes this a truly profound example of scientific explanation in that it describes processes that are fundamentally invisible and impossible to experience directly. We cannot visit a collapsing stellar core. We cannot feel neutrinos passing through our bodies (though trillions do every second). We cannot observe quantum oscillations across interstellar distances.

Yet through the power of good explanations (theories that connect diverse phenomena through hard-to-vary relationships) we can understand these hidden realities with confidence. The supernova theory successfully predicts everything from stellar lifetimes to the abundance of elements in the universe to the behaviour of subatomic particles. Not bad!

This is what Deutsch means when he emphasises that science is about explaining the unseen. Good explanations reveal the underlying structure of reality that gives rise to what we observe. They tell us not just what happens, but why it must happen that way.

The next time you look up at the night sky, remember that you're seeing the afterglow of countless stellar deaths, their heavy elements now part of the Earth beneath your feet and the life around you. And streaming through your body at this very moment are neutrinos from supernova explosions, carrying with them the signature of some of the most extreme physics in the universe, invisible messengers that confirm our best explanations of how reality works at its deepest levels.

So at the danger of labouring the point, the creativity behind these explanations is staggering. Every component of the supernova theory represents an incredible leap of human imagination that went far beyond existing knowledge.

Each component required extraordinary leaps of creativity that couldn’t have been arrived at incrementally:

Stellar Evolution Theory required abandoning the then physics establishment view of unchanging stars and proposing that they undergo nuclear fusion, a process not even discovered when the theory began developing.

The Chandrasekhar Limit emerged from a 19-year-old's (!!!) calculations on a ship voyage to England. Subrahmanyan Chandrasekhar's conclusion that stars beyond a certain mass must collapse was initially rejected by established astronomers because it seemed impossible.

Neutron Stars demanded the extraordinary idea that matter could be compressed to nuclear density, that an entire star could become a city-sized ball of neutrons. This wasn't extrapolation from existing physics; it was a completely new conception of matter itself.

The Neutrino represents perhaps the most insane theoretical leap. In 1930, Wolfgang Pauli off the top of his head invented an invisible, nearly massless particle as what he called a "desperate remedy" to save energy conservation in his calculations, otherwise he found that energy was not being conserved. It was initially a “placeholder” particle forced on him to make his calculation balance. He was so uncertain about proposing something undetectable that he famously said he had "done a terrible thing." Nothing in existing physics suggested such particles should exist.

Neutrino Oscillation required overturning the Standard Model of particle physics when Ray Davis kept detecting fewer solar neutrinos than predicted. Rather than assume experimental error, as mentioned above scientists proposed that neutrinos could spontaneously change identity, something that required them to have mass, contradicting decades of established theory.

None of these insights could have been reached incrementally by simply reinterpreting existing knowledge or looking for patterns in known data. Each required genuine creative leaps into unexplored conceptual territory. You cannot induce the existence of neutrinos from 19th-century physics, nor derive stellar collapse from classical astronomy.

This illustrates something profound about how knowledge grows. David Deutsch emphasises that scientific progress comes from creating new explanations that reveal previously inconceivable aspects of reality. The supernova story shows this process in action: each breakthrough opened entirely new territories of understanding that were literally unimaginable before someone made the creative leap to propose them.

The theory we now take for granted required multiple generations of scientists to propose ideas that seemed impossible, pursue concepts that contradicted established wisdom and, importantly, persist with explanations that initially had little empirical support. What we now recognise as a beautiful, coherent explanation of stellar death and cosmic element creation emerged from acts of pure theoretical creativity, the human mind reaching beyond the known to grasp the structure of the unseen universe. Try that LLMs!

There is also something profound about good explanations: hard-to-vary explanations tend to be universal. That is they are applicable everywhere across space and time. The supernova story isn’t just applicable to one supernova and maybe different for another, it is true about how reality works everywhere and everywhen.

The Chandrasekhar limit has been operating literally as an iron law (bad pun intended) throughout cosmic history. Every massive star that ever lived and died, from the very first generation of stars 13 billion years ago to explosions happening right now in distant galaxies, has been governed by exactly the same physics Chandrasekhar derived on his sea voyage from India to Cambridge in 1930. I often wonder what he would have come up with if his voyage had lasted more than 18 days…

Those ancient supernovae that scattered the heavy elements eventually becoming part of Earth, our bodies and even the neurons in Chandrasekhar's brain as he performed his calculations all obeyed the same 1.4 solar mass limit. The neutrinos streaming through the universe from stellar deaths billions and billions (sorry, couldn’t resist) of years before our solar system existed were oscillating between flavours according to the same quantum mechanical principles we discovered only in the late 20th century.

The same physics that governs a supernova in the Andromeda Galaxy today governed the stellar explosions that created the gold in your jewellery billions of years ago. The universe has been running this supernova "program" for cosmic ages, following rules that took human minds millennia to even glimpse.

What makes human creativity so precious is precisely this capacity to grasp universal truths that go far beyond our immediate experience. Through acts of pure theoretical imagination, we can understand processes that occurred before Earth existed and will continue long after it's gone. We can comprehend the invisible dance of neutrinos, the crushing violence of stellar collapse, and the nuclear alchemy that forged the elements of our existence. And, crucially, we can’t get to these types of explanations just by incrementally remixing and finding patterns in existing knowledge, although we do have a lot to work with.

Good explanations are hard to vary because they discover objective features of reality itself. The universe doesn't care whether we understand it, but the fact that we can understand it, that human minds can reach across billions of years and light-years to grasp the fundamental workings of stellar death, represents something extraordinary about the nature of our ability to create explanatory knowledge and consciousness in the cosmos.

</Carl Sagan Mode>

I hope you enjoyed reading this article as much as I enjoyed writing it.

Thank you,

Bern.

[Note: You may have noticed that this article mentions both "stars at least eight times heavier than our Sun" and "the 1.4 solar mass Chandrasekhar limit." These refer to different masses at different stages: massive stars (8+ solar masses) lose most of their material during their lifetime through stellar winds and the supernova explosion itself. What remains to collapse is the iron core, which triggers collapse when it exceeds the Chandrasekhar limit of about 1.4 solar masses.]

Background: The solar neutrino problem.

Helioseismology (now known by the broader umbrella term asteroseismology) has also had implications not only for the understanding of Sun-like stars and their evolution in the Universe, but also for the field of particle physics. Theoretical studies indicate that the nuclear fusion occurring in the Sun’s core must generate enormous quantities of neutrinos (Davis et al., 1968). Neutrinos interact extremely weakly with matter, having no electric charge and possibly no mass which makes them vary hard to detect (D¨appen et al., 1986). Travelling at nearly the speed of light they are produced in vast quantities by the fusion reactions in the Sun. Ironically, before the development of helioseismology, neutrino astronomy was the only means of obtaining information form the Sun’s core.

There are ground based experiments created to detect neutrinos which have large collecting volumes in order to detect any low probability neutrino interactions (Abdurashitov et al., 1995). So far such experiments have been designed to detect only the electron neutrino but not the muon and tau neutrinos that are also predicted by theoretical physics.

Solar models can provide predictions of the flux of neutrinos produced in the nuclear reactions at the Sun’s core. Nuclear reactions are dependant on a very high power of the internal temperature which can be provided by solar models that have been refined by the advances from helioseismology.

Unfortunately the predicted and measured fluxes differ by a factor of two or three, which introduces what is known colloquially as the ‘solar neutrino problem’ (Elsworth et al., 1990). There are two possible ways to resolve the problem. The first of these is that the models of the solar interior are wrong and our estimate of the core temperature T is also wrong. The second solution may be that our understanding of the behaviour of neutrinos is incomplete.

The observed frequencies from Helioseismology agree extremely well with those predicted by solar models which leads to some confidence in current solar theories. If this is the case then we should be confident that we have a reasonable understanding of the conditions in solar interior and hence the expected level of neutrinos predicted by these models, which in turn suggests that a modification to the understanding of the behaviour neutrinos is required.

One proposed idea is that neutrinos are able to change type during their journey from the Sun to the ground based detectors here on Earth. That is, electron neutrinos may not always stay as electron neutrinos but may flip between all three states; the muon and tau as well as the electron states. These changes of type or state are referred to as neutrino oscillations. This would explain why experiments designed to detect only electron neutrinos may have missed the other two thirds of the neutron flux. In order for neutrino oscillations to occur, particle physicists require that neutrinos must have a small amount of mass.

Future Pathways

Discussion about this post

Future Pathways

Why Exploding Stars Demand Good Explanations

What supernovae teach us about the creativity of explanatory knowledge and the limits of LLMs.

Further Reading

Discussion about this post