物理 · Physics

General Relativity

Day 16 · 2026 · Phase D Relativity
Nobody is pulling the apple down. It has been travelling in a straight line all along — it is just that time runs a little slower down here.
Newton was the first person to be unhappy with his own law of gravitation: how could one body reach across empty space, through no medium at all, and instantly push on another? He called the idea absurd, and insisted he was only describing, not explaining. Two centuries later Einstein's answer was to cancel the question: there is no such force as gravity. There is only spacetime, bent by matter, and objects travelling dutifully straight through it.

The Equivalence Principle

Einstein · 1907
Intuition A windowless elevator. You have weight underfoot and a tossed ball accelerates to the floor — can you tell whether you are parked on the ground or strapped to a rocket accelerating at 9.8 m/s²? You cannot. Cut the cable, and the ball simply hangs in front of you; now you cannot tell falling from drifting in deep space either. In 1907 Einstein called this the happiest thought of my life.
At rest on the ground ball accelerates to the floor ≡ no measurable difference Rocket accelerating in deep space a = g ball accelerates to the floor Elevator with a cut cable free fall ball just hangs in mid-air ≡ no measurable difference Capsule drifting in deep space ball just hangs in mid-air
Left and right are indistinguishable from inside the box. Since no measurement separates them, simply declare them the same thing — that step is the foundation of the whole theory.
Mechanism Why the coincidence? Because of an equation that had been sitting unclaimed since Newton:
minertial = mgravitational
minertial is the m in F=ma — how hard it is to push something. mgravitational is the m in the law of gravitation: a gravitational charge, the analogue of electric charge in an electric field. Conceptually the two have nothing to do with each other, yet they are equal everywhere: they cancel, the m drops out of the acceleration, and a feather falls beside an iron ball in vacuum. The coincidence has only been pinned down harder — from Eötvös's torsion balance to the MICROSCOPE satellite, whose final results came out in 2022, the ratio of two materials' falling accelerations differs from 1 by less than 10−15. Einstein stopped calling it a coincidence and made it an axiom.
The counterintuitive part The person in free fall is the one feeling no force. Standing on the ground, you are being shoved upward by the floor — right now you are accelerating at 9.8 m/s². This is not wordplay: the accelerometer in your phone, sitting still on a table, reads 9.8 and not 0 (the digit zero); throw it out the window and for those few hundred milliseconds it honestly reads 0. What matters more is that the equivalence principle holds only locally: gravity can be transformed away, but only one small patch at a time, and what refuses to go is curvature.
Cross-disciplinary read · promoting "no detectable difference" to an axiom Turning a string of failed attempts to tell things apart into a defining principle is one of science's most productive moves:
  • Thermodynamics: nobody could build a perpetual motion machine — that purely empirical failure was promoted to the second law, which then yields the Carnot limit and entropy.
  • Cryptography: modern security is defined as indistinguishability — given ciphertexts of two plaintexts, an adversary must not guess which is which appreciably better than a coin flip. Security is not "well hidden", it is "the two worlds look alike".
In one line: free fall is not being dragged along — it is the only state of motion with no force acting.
Think: If free fall cancels gravity, can one coordinate system cancel the whole Earth's gravity at once?
No — only within a small neighbourhood of each point. Objects falling freely on opposite sides of the Earth head in opposite directions, and no single acceleration matches both. What survives is the tidal effect: no change of coordinates removes it, so it is real.

Gravity Is Geometry

Geodesics · Field equations · 1915
Intuition Two people stand on the equator a hundred kilometres apart and both set off due north. Neither ever turns, yet they drift together and collide at the pole. If they insisted that "straight lines do not meet", they would have to invent a force that pulls them together and — awkwardly — is independent of their weight. On a curved surface the truth is plainer: no force, the surface is curved. The action-at-a-distance tug of "Motion, Force and Determinism" becomes a single sentence here: objects travel straight all along; it is the meaning of "straight" that spacetime rewrote.
Mechanism A "straight line" in curved spacetime is a geodesic — the locally straightest path, one that never turns of its own accord. A freely moving body does exactly one thing: follow a geodesic. How spacetime bends is fixed by the field equations Einstein wrote down in 1915:
Gμν = 8πGc4 Tμν
μ and ν (read "mu" and "nu") are indices, each running over 4 values — one time direction plus three space directions — so this is not one equation but a whole set. Gμν is the Einstein tensor, pure geometry; Tμν is the stress-energy tensor, holding energy density, momentum and pressure. Watch out: the two Gs look alike and are unrelated — the one with indices is a geometric tensor, the one in 8πG is Newton's gravitational constant. The coefficient 8πG/c4 ≈ 2×10−43 (SI units) is absurdly small: that is "gravity is weak" in quantitative form. Wheeler compressed the whole equation into one line — spacetime tells matter how to move; matter tells spacetime how to curve.
Mechanism · how to measure curvature Curvature has an operational definition: release two freely falling bodies and watch their relative acceleration. Neither feels a force, yet the distance between them changes — that is geometry talking. The effect has a more familiar name, tidal force: it is the difference, not the pull itself, that is curvature.
Two initially parallel lines neither turned, yet they meet north pole: collision start: both due north equator no force — the surface is curved A freely falling cloud no forces, yet the shape changes stretched squeezed squeezed dashed: the original shape mass lies this way
The left panel is what causes curvature; the right panel is how you read it. Tidal force is not a side effect of gravity — it is the one part of gravity no change of coordinates can erase.
The counterintuitive part The commonest question about "curved spacetime" is curved into what? The answer is: into nothing. Curvature is intrinsic. An ant that can never leave a sphere can measure the angles of a triangle, or the ratio of a circle's circumference to its radius, and work out that the surface is curved — without ever knowing the sphere sits in three-dimensional space. Spacetime is the same: whether some higher-dimensional "outside" exists is neither needed nor of interest to physics.
Cross-disciplinary read · go round a loop and come back different: holonomy Curvature has an equivalent reading: carry an arrow around a closed loop and it comes back rotated. Go from the north pole down a meridian to the equator, a quarter-turn along the equator, then back up to the pole — you never turned the arrow, and it has swung by 90°. This is holonomy:
  • The Foucault pendulum: the swing plane rotates 360°×sin(latitude) per day — about 260° in Beijing, and not at all on the equator. The Earth is not "dragging" the pendulum; this is the holonomy of parallel-transporting the swing direction once around a circle of latitude.
  • Mechanics and robotics: a cat dropped upside down has exactly zero angular momentum throughout, yet turns a net 180° — by tucking and extending so its shape traces a closed loop, with the net rotation being that loop's holonomy.
  • Quantum mechanics: drag a quantum system's parameters slowly around a loop back to the start and the wavefunction picks up a phase fixed by the loop's geometry alone — the Berry phase (1984), now basic vocabulary for topological phases of matter.
In one line: gravity is not a force — the definition of "straight" changed.
Think: If gravity is pure geometry, why does the equation still need a measured constant G?
Geometry itself does not know "how much matter bends how much". G (together with c) sets the exchange rate converting kilograms of matter into units of curvature. An exchange rate is not a geometric fact — it has to be measured. And its being as small as 2×10−43 is exactly why we can stand on a planet without being shredded by the folds in spacetime.

Why the Apple Falls

Gravitational time dilation · Maximal proper time
Intuition Put two facts together. The first comes from "Spacetime": a freely moving body follows the worldline of maximal proper time — the path on which its own watch ticks the most. The second is new: clocks lower down run slower. So "falling" gets a new explanation: spending time higher up buys extra proper time, but getting up there requires speed, and speed slows the clock again. The optimal compromise between those two ledgers is exactly a parabola.
Mechanism In a weak gravitational field, two clocks separated by a height h tick at a ratio:
Δττ ≈ ghc2
τ (read "tau") is proper time, the reading on the watch you carry; Δτ/τ is how much faster one clock runs than the other; g is the local gravitational acceleration, h the height difference, c the speed of light. Put in numbers for the Earth's surface: each metre of altitude makes a clock gain 1.09×10−16 seconds per second, so your head ages roughly 3 nanoseconds per year more than your feet. Laughably small, and measured directly: in 2010 NIST resolved a 33 cm height difference with two optical clocks, and in 2022 JILA resolved the gravitational redshift across millimetres inside a single cloud of cold atoms.
height h ct (time × c) low · clocks slow high · clocks fast → thrown back down stay on the ground clocks slowest here the real trajectory longest proper time too high, too fast speed slows the clock: a net loss all three paths join the same two events
Height runs horizontally, time vertically — this is a spacetime diagram, not a trajectory. Over a one-second flight the ball banks about 4×10−17 s of proper time more than staying put. For those few tens of attoseconds it climbed 1.25 metres.
The counterintuitive part Everyday gravity comes almost entirely from the curvature of time, not of space. The reason is the conversion: every second you advance 300 million metres along the time axis, while moving perhaps one metre through space. That tiny bend in the time direction (of order 10−16 per metre), amplified by a 300-million-metre "displacement in time", becomes a visible fall. The spatial share only becomes comparable near c — starlight grazing the Sun is deflected by 1.75 arcseconds, whereas counting only the curvature of time gives half that, and the missing half is exactly what space contributes.
Cross-disciplinary read · the metric decides what "steepest" means Once you admit that "how distance is measured" must be stated first, several fields' core problems change shape:
  • Optimisation and AI: a neural network's parameter space is not flat — equal-sized parameter changes affect behaviour wildly differently. Natural gradient descent (Amari, 1998) treats the Fisher information matrix as the metric, which rewrites the "direction of steepest descent"; K-FAC and Adam's per-coordinate scaling are cheap approximations of the same idea.
  • Optics: Fermat's principle says light takes the path of stationary optical length. A weak gravitational field can be written equivalently as a medium of refractive index n ≈ 1 + 2GM/rc2 — starlight bending past the Sun and light bending in graded-index glass are the same mathematics.
In one line: the apple falls because it is buying its own watch the largest possible reading.
Think: If time really runs slow down low, why does nobody in a basement feel their day is shorter?
Because everything slows together — heartbeat, metabolism, atomic clocks, thought — so no local experiment can detect it. The difference exists only when you bring two clocks together and compare. GPS cannot dodge it, because it is permanently comparing two clocks twenty thousand kilometres apart in altitude.

The Verdicts

Mercury · 1919 · GPS
Intuition Elegance counts for nothing. General relativity opened by staking several bets it could lose — each one a definite number, with no dial to turn afterwards.
Mechanism · four bills 1. Mercury's perihelion. The long axis of Mercury's orbit swings 5600 arcseconds per century; the Newtonian account explains 5557, and the leftover 43 hung there for half a century — people even went looking for a planet called Vulcan to take the blame. Einstein's calculation gave exactly 43, with no adjustable parameter anywhere in the equations.
2. Light deflection. Starlight grazing the Sun's edge should bend by 1.75 arcseconds; treating light as a Newtonian particle with mass gives precisely half that, 0.87. Eddington's 1919 eclipse expedition picked the former.
3. Gravitational redshift. In 1959 Pound and Rebka set a gamma-ray source and absorber at the top and bottom of a 22.5-metre tower at Harvard and used the Mössbauer effect to resolve a shift of 2.5×10−15 — light emitted below is reddened as it climbs.
4. GPS. Navigation satellites orbit twenty thousand kilometres up at about 3.9 km/s. Altitude makes the onboard clock gain roughly 45.7 microseconds a day, speed makes it lose about 7.2, netting +38 microseconds per day; uncorrected, position errors accumulate at around 10 km a day.
The counterintuitive part The theory's status has changed: from the thing being tested to the ruler doing the testing. Gravitational lensing is standard astronomical equipment, with galaxy clusters used as natural telescopes to magnify earlier galaxies behind them. And an honest engineering detail: the frequency of a GPS satellite clock is detuned at the factory so it runs correctly once in orbit — the relativistic correction is not a software patch, it is welded into the hardware.
Cross-disciplinary read · what it takes to convict a theory
  • Philosophy of science: Popper used that 1919 eclipse as his model case for falsifiability — Einstein had committed in advance to 1.75 arcseconds, and any other value would have sunk the theory. He set it against theories that can explain any outcome, arguing that their explanatory reach is precisely their weakness.
  • Statistics and machine learning: the same criterion goes by preregistration and held-out test sets — state in advance what you predict and on which unseen data it will be checked. Tuning after the fact to fit known data is the same disease as inventing Vulcan to absorb those 43 arcseconds.
In one line: only a theory a single observation could have killed deserves to survive.
Think: What if Eddington's 1919 measurement had come out at Newton's 0.87 arcseconds?
General relativity would have had to concede — there is no dial that reaches 0.87. But history has another side: Eddington's error bars were large, and people have questioned ever since whether he discarded data with a thumb on the scale. What nailed the result was decades of repetition: very-long-baseline interferometry has pinned the deflection to within a few parts in ten thousand of 1.75 arcseconds.

Going deeper

The bowling ball on a rubber sheet — what exactly is wrong with that picture?
Three things. First, it explains gravity using gravity: the small ball rolls into the dip only because there is a "down" outside the frame pulling it; take the sheet into space and nothing happens. Second, it draws only the curvature of space, while everyday gravity comes almost entirely from the curvature of time — and the picture has no time axis at all. Third, it invites the question "curved into what?", when curvature is intrinsic. A better substitute is a spacetime diagram of worldlines.
Where exactly do general relativity and quantum mechanics collide?
Look at the two sides of the field equations. The right-hand side holds matter, and matter is quantum — it can be in superposition; the left-hand side is a definite classical geometry. Put a particle in a superposition of "on the left" and "on the right": which way does spacetime bend? Replacing the right side with an expectation value (semiclassical gravity) is known to yield conclusions in conflict with quantum mechanics. What makes it worse is that the two only meet head-on at the Planck scale (around 1019 GeV, more than ten orders of magnitude above existing colliders). That is territory for "The Open Problems of Physics".
Why are they called "the field equation" when there are 10 of them?
μ and ν each take 4 values, giving 16 components; both tensors are symmetric (swapping indices changes nothing), leaving 10 independent ones. The Bianchi identities then supply 4 constraints — those 4 equations govern the self-consistency of the initial data rather than evolution. Genuinely independent dynamical degrees of freedom: 2, matching the two polarisations of a ripple travelling through vacuum. That counting exercise pays off in "Black Holes and Gravitational Waves".

Further reading