Claude Fable 5 Simulates the Solar System from First Principles

1. What “first principles” actually means

There is an important distinction that gets glossed over in discussions of AI capability: the difference between retrieval and derivation. Retrieval is recalling a stored fact — “Mars orbits the Sun every 687 days.” Derivation is computing an answer from more fundamental relationships — taking the mass of the Sun, the mass of Mars, and Mars’s average distance from the Sun, applying Newton’s law of gravitation, and calculating that the orbital period must be approximately 687 days.

These are not the same cognitive act. Retrieval requires only memory; derivation requires reasoning. A student who has memorized “the orbital period of Mars” knows one fact. A student who can derive it from first principles knows something much more powerful: they understand the mechanism. They can compute the orbital period of any hypothetical planet, of a newly discovered exoplanet, of an asteroid on an unusual trajectory. The memorized fact fails the moment the situation differs from what was memorized. The derivation generalizes.

When Fable 5 built a solar system simulation “from first principles,” this is the distinction being invoked. It did not look up a table of planetary parameters and animate them. It reasoned from Newton’s law of universal gravitation — one of the most fundamental equations in physics — and derived the orbital behavior of the entire solar system from that single foundation. The fact that the simulation’s predictions matched observed astronomical reality is the proof that the derivation was correct.

This is the capacity that makes AI useful for scientific work, rather than just scientific recall. A system that can only retrieve stored answers is an encyclopedia. A system that can derive novel conclusions from fundamental principles is closer to a reasoning collaborator.

2. The physics of orbital motion

Newton’s law of universal gravitation states that any two objects with mass exert an attractive force on each other. The magnitude of that force is:

F = GMm / r²

where G is the gravitational constant (approximately 6.674 × 10⁻¹¹ N·m²/kg²), M and m are the masses of the two objects, and r is the distance between them. The force is directed along the line connecting the two objects, pulling them toward each other.

From this single equation, applied iteratively to every pair of bodies in the solar system, everything else follows. Kepler’s three laws — which Kepler derived empirically from Tycho Brahe’s observational data in the early 1600s, without understanding their cause — emerge naturally as consequences of Newton’s gravitation:

1. Orbits are ellipses with the central body at one focus. This is not assumed — it falls out of the mathematics when you solve the equations of motion for a two-body system under an inverse-square force.
2. Equal areas in equal times: a line from the orbiting body to the central body sweeps out equal areas in equal time intervals. This is a consequence of conservation of angular momentum.
3. The square of the orbital period is proportional to the cube of the semi-major axis (T² ∝ a³). This follows from combining the force law with the geometry of an ellipse.

In practice, simulating these orbits requires converting the continuous equations of motion into discrete steps. At each step, you compute the gravitational force on each body from every other body, use Newton’s second law (F = ma) to find the acceleration, and update the velocity and position accordingly. The simplest version of this — called Euler integration — takes the current velocity, multiplies it by a small time step, and adds it to the position. However, Euler integration accumulates errors rapidly; orbits that should be stable over centuries slowly spiral inward or outward in the simulation.

More accurate methods are essential for a simulation that is meant to predict real astronomical events. The Runge-Kutta 4th-order method (RK4) evaluates the derivative at four points within each time step and takes a weighted average, dramatically reducing the error per step. For solar system simulations, even better methods exist — symplectic integrators like the leapfrog algorithm exactly conserve energy over long timescales, preventing the artificial orbital decay that plagues simpler methods. Choosing the right integrator is not an academic detail; it is what makes the difference between a simulation that diverges after a decade of simulated time and one that accurately predicts positions centuries into the future.

3. Eclipse prediction as empirical validation

A solar eclipse is a precise astronomical event: the Moon must pass directly between the Earth and the Sun, its shadow falling on a specific region of Earth’s surface at a specific moment in time. The geometry required is extremely tight. The Moon’s orbit is tilted relative to Earth’s orbit around the Sun, which means most months the Moon passes above or below the Sun’s disk rather than directly in front of it. A total solar eclipse happens only when the Moon crosses the plane of Earth’s orbit (a node crossing) at the precise moment it is also near the Sun in the sky (near a new moon).

If a simulation predicts that a solar eclipse will occur at a specific date, time, and location on Earth, and that prediction matches what an astronomer actually observes, then the simulation’s model of the Moon’s orbit, the Earth’s orbit, the tilt of the Moon’s orbital plane, and the precession of that plane must all be correct. There are no loopholes: a simulation that gets any of these wrong will predict eclipses at the wrong times, in the wrong places, or miss them entirely.

This is why eclipse prediction is used as a validation test. It is a high-precision, empirically verifiable consequence of the orbital dynamics. Ancient astronomers, including the Babylonians, discovered the Saros cycle — an 18-year, 11-day period after which eclipse patterns repeat — by pure observation. They could predict eclipses without understanding why they recurred. Newton’s mechanics explained the why, and modern numerical simulations can predict the what with accuracy measured in seconds and kilometers. That Fable 5’s simulation passes this test is meaningful evidence that it derived the physics correctly rather than merely reproduced memorized orbital data.

4. Derivation vs. retrieval — why this matters for AI and science

The Fable 5 solar system demonstration matters beyond the visual spectacle because it addresses a foundational question about what AI systems can do in science: can they reason from principles, or only recall stored results?

A model that only retrieves answers is fundamentally limited by its training data. It cannot predict the orbit of a newly discovered asteroid for which no data exists in its training set. It cannot tell you how a planet’s orbit would change if the Sun were 10% more massive. It cannot compute the gravitational interaction of a proposed satellite configuration around a hypothetical moon. These are all questions that require derivation — applying known physical laws to novel configurations.

A model that can derive — as Fable 5 demonstrated — can address all of these questions, because the generative mechanism (Newton’s laws, numerical integration) is not dependent on having seen the specific case before. This is the same reason physicists value first-principles approaches: a first-principles calculation in a new domain carries more epistemic weight than a lookup in a table, because you understand exactly which assumptions went into the result and can reason about their validity.

For users of Claude, this suggests a useful heuristic: when you want to use Claude for scientific or mathematical problems, ask it to derive rather than recall. Ask it to show its work, to start from fundamental principles, to explain why each step follows from the last. This both improves the quality of the answer — derivation is more reliable than retrieval for novel questions — and makes it easier to verify, because you can check each step rather than having to trust the final answer blindly.

🧠 Check your understanding

5 questions · your answers are saved in this browser only

1. 1. What is the key distinction between 'retrieval' and 'derivation' in AI reasoning?

   Retrieval is faster; derivation is slower Retrieval recalls stored facts; derivation computes answers from fundamental principles Retrieval uses the internet; derivation uses training data Retrieval is for math; derivation is for language

   Retrieval reproduces memorized information; derivation applies fundamental laws or principles to compute answers that may never have been seen before. Derivation generalizes to novel situations; retrieval fails when the situation differs from what was memorized.

2. 2. What does Kepler's third law state?

   Planets orbit in circles with the Sun at the center All planets move at the same speed The square of the orbital period is proportional to the cube of the semi-major axis Gravitational force decreases linearly with distance

   Kepler's third law (T² ∝ a³) relates the orbital period of a planet to the size of its orbit. It falls out naturally from Newton's law of universal gravitation applied to elliptical orbits.

3. 3. Why does simple Euler integration fail for long-term solar system simulations?

   It cannot handle multiple bodies simultaneously It accumulates errors that cause orbits to drift over time It runs too slowly to complete in reasonable time It requires knowing the initial orbital periods in advance

   Euler integration approximates the continuous equations of motion with discrete steps. Each step introduces a small error, and these errors accumulate over time, causing simulated orbits to gradually spiral inward or outward rather than remaining stable.

4. 4. Why is eclipse prediction considered a rigorous test of a solar system simulation?

   Eclipses are visually dramatic and easy to observe Eclipse timing requires all orbital parameters — Moon orbit, Earth orbit, nodal tilt, precession — to all be correct simultaneously Eclipses occur frequently enough to provide many data points Eclipse geometry only depends on the Earth-Sun distance

   A solar eclipse requires the Moon to be precisely between Earth and the Sun, which depends on the Moon's orbital inclination, nodal precession, and synchronization with the new moon phase. A simulation must get all of these right simultaneously to predict eclipses correctly.

5. 5. Which integration method is best suited for long-term orbital simulations because it conserves energy exactly?

   Euler integration Trapezoidal integration Monte Carlo sampling Symplectic (leapfrog) integration

   Symplectic integrators like the leapfrog algorithm exactly conserve the total energy of the system over long timescales, preventing the artificial orbital decay or growth that affects non-symplectic methods.

Reset quiz

🛠️ Build it yourself

Follow these exact steps to reproduce it yourself · estimated time: ~30 minutes

Prerequisites

• Claude or another capable AI assistant
• A web browser or Python environment
• Basic familiarity with physics concepts (helpful but not required)

Step 1 — Ask Claude to derive orbital mechanics from scratch

Start with this prompt: “Please simulate a simple two-body gravitational system — a star and one planet — using only Newton’s law of gravitation. Derive the equations of motion from F = GMm/r² and Newton’s second law. Do not use any pre-baked orbital formulas. Implement this in Python (or JavaScript) and plot the resulting orbit.”

Step 2 — Verify the derivation step by step

Ask Claude to explain each step: “Walk me through why the acceleration of the planet at each time step is computed the way it is.” Check that each step follows logically from the previous one. Ask about any step that is not clear.

Step 3 — Extend to the Earth-Moon system

Once the single-planet simulation is working, say: “Now extend this to a three-body system: Sun, Earth, and Moon. Use realistic masses and distances (you can look these up in your training data). Show the Moon orbiting the Earth while both orbit the Sun.”

Step 4 — Ask for the next eclipse

Say: “Given the positions and velocities in this simulation, can you detect when a solar eclipse would occur — when the Moon passes between the Earth and the Sun? Find the next such alignment after January 1, 2026.” Compare the result with a real eclipse calendar (NASA publishes these online).

Step 5 — Reflect on the derivation

Ask Claude: “What would change in this simulation if the Sun were 20% more massive? Derive the new orbital period of Earth from first principles.” This tests whether the model is genuinely reasoning from the physics or just retrieving stored answers.