Thursday, 23 April 2026

How Disciplines Misunderstand Their Own Success — 1 The Demand for Convergence

After ten years of increasingly meticulous experimentation, physicists have not converged on the value of the gravitational constant. [See Nature article here.]

They have, however, converged on something else: the limits of the assumption that such convergence must occur.

The constant in question—G, the parameter that quantifies the strength of gravitational attraction—has been measured for over two centuries. Yet it remains the least precisely known of the fundamental constants. A recent decade-long replication effort, involving the relocation of experimental apparatus across continents, has produced a value that disagrees with both earlier measurements and the current internationally recommended figure.

This is not an isolated anomaly. Measurements of G have never settled cleanly. What is new is the degree of precision with which this failure now reproduces itself.

The expectation, however, has not shifted. Independent measurements are still assumed to converge on a single value. Disagreement is still treated as provisional—evidence of hidden error, underestimated uncertainty, or incomplete control.

But this expectation is not an empirical result. It is a condition placed on what counts as a successful result.


The quiet contract of measurement

At the centre of the problem is a largely unexamined demand:

Independent measurements must converge.

This appears methodological. In practice, it is ontological. It encodes a set of commitments:

  • that the quantity being measured is invariant
  • that variation is accidental
  • that error is eliminable in principle

Within this frame, divergence can only ever be temporary. If results do not align, the work is not yet finished.

And yet, in the case of G, the work has been extended, refined, and repeated across decades—without producing convergence.

This does not look like noise waiting to be eliminated. It looks like variation persisting under increasing control.


Error, or structure?

The standard response is to treat discrepancies as clues: signs that something in the experimental setup has not been properly accounted for.

And indeed, each new generation of experiments identifies new influences:

  • material imperfections
  • environmental couplings
  • previously unnoticed forces

These are not failures of method. They are successes of refinement.

But notice the pattern:

every reduction in one source of variation reveals another.

If variation were merely accidental, we would expect it to diminish. Instead, it reorganises itself at higher levels of precision.

At some point, the question shifts:

Are these deviations from a true value,
or are they structured differences between experimental configurations?


The impossibility of isolation

Gravity presents a specific difficulty. It cannot be screened, cancelled, or confined. Every mass contributes. Every configuration matters. There is no neutral background against which the interaction can be cleanly extracted.

In practice, this means that each experiment does not measure “gravity itself.” It stabilises a relation within a field that cannot be fully bounded.

The torsion balances, the atom interferometers, the free-fall systems—all are designed to isolate a tiny signal. But what they actualise is always a particular configuration of relations, under highly specific constraints.

The expectation, however, is that these configurations differ only superficially—that beneath them lies a single invariant parameter.

It is this expectation that the data fails to support.


Refinement without release

Modern experiments are acutely aware of potential bias. Measurements are blinded. Independent offsets are introduced and later removed. Procedures are rigorously controlled.

This is reflexivity at a high level of sophistication.

But it operates within a fixed frame. It asks:

Where might the procedure be distorting the result?

It does not ask:

What must be true for there to be a single result at all?

So refinement proceeds without releasing the underlying assumption. The demand for convergence remains untouched.


What actually stabilises

A detail often treated as incidental deserves emphasis: most practical applications do not require G in isolation. They rely on combined quantities—products of G with mass—that can be determined with far greater precision.

What is stable, in other words, is not G as an independent value, but relations in which G participates.

This suggests a different interpretation. G functions less as a directly measurable property and more as a parameter within a network of constraints—a value that stabilises certain models under certain conditions, rather than one that can be cleanly extracted from them.


The persistence of invariance

Why, then, does the expectation of a single value persist?

Because it is not derived from the experiments. It is inherited from a broader picture in which:

  • explanation requires invariance
  • measurement approximates pre-existing values
  • reality is decomposable into independent parameters

Within this picture, convergence is not just expected. It is required for the intelligibility of the result.

Which is why its absence is so difficult to register.


Reframing the problem

If we suspend, even provisionally, the demand for convergence, the situation becomes legible in a different way.

The question is no longer:

What is the true value of G?

But:

  • Under what conditions do measurements of gravitational coupling stabilise?
  • How do different configurations systematically vary?
  • What constraints govern these variations?

From this perspective, the spread of values is not a failure. It is data about the structure of the measurement itself.

The task shifts from eliminating variation to mapping it.


A constant, reconsidered

This does not require abandoning G. It requires reconsidering what kind of entity it is.

Not:

a number waiting to be discovered

But:

a parameter that emerges within, and is inseparable from, the conditions of its measurement.

The decade-long effort to refine its value has not failed. It has revealed, with increasing clarity, that the object of inquiry does not behave like the invariant it was assumed to be.

The question now is not why the measurements do not converge.

It is why convergence is still being demanded.

No comments:

Post a Comment