A Little Help for My Friends

A plain-English explainer of what this research program is actually about,
written for non-physicists. Share with friends, family, anyone curious who
doesn’t want to wade through the technical paper.

The question

Picture two people watching the same thing through different dirty windows.
Each window has its own pattern of smudges, random, unrelated to the
other’s. Each person describes what they see. How much do their descriptions
disagree?

That’s basically what this paper is about, translated from quantum gravity.

The “thing” is a quantum state. The “dirty windows” are what physicists call
non-isometric codes, the imperfect encoding from the higher-dimensional
bulk (where quantum gravity lives) down to the lower-dimensional boundary
(where we measure). The “smudges” are random because the universe doesn’t
hand you a nicely-labeled encoding; you get some specific random one. Each
observer is tracing their own picture back from the boundary to the bulk,
and because of the smudges, their pictures won’t match exactly. The question
is: by how much do they disagree?

People had been measuring this disagreement via a specific quantity (the
entropy of each person’s reconstruction, minus the other’s). The natural
guess was: it should fall off at some universal rate as you let the
observers see more clearly. One fixed answer.

The surprise

It’s not one answer. It depends on what you’re looking at.

Here’s the analogy I’d use: imagine our two people are listening to the
same song, each through a scratchy phone line with its own random pattern
of pops and static. They write down what they hear.

• If the song is one long note held at middle C, a little static really
  matters. One listener hears “C, maybe a bit flat.” The other hears “C,
  with a weird warble.” Their transcriptions differ noticeably, and they
  keep differing even if you upgrade their phone lines a lot.

• If the song is a dense symphony with a thousand instruments playing at
  once, the static still scrambles things, but the scrambled version still
  sounds like “dense symphony.” Both listeners write down essentially the
  same description. Their transcriptions converge fast as you upgrade the
  phones.

That’s exactly the pattern we found. Call the “phone line quality” $d_B$, how much information the observers can actually receive. Then:

• Simple (unstructured) quantum state (the “one held note”):
  disagreement shrinks as $1/\sqrt{d_B}$. Double the resolution, cut the
  disagreement by ~30%. Slow improvement.
• Complex (maximally entangled) quantum state (the “symphony”):
  disagreement shrinks as $1/d_B^{3/2}$. Double the resolution, cut the
  disagreement by ~65%. Much faster improvement.

The gap between those two exponents is exactly one, one full power
of $d_B$ separating how fast the two regimes converge. That’s not an
artifact. It’s a real structural statement about how observer disagreement
works.

Why it’s nice

Two things make this result worth writing up rather than just noting:

First, we derived the numbers, not just fit them. The prefactors
$0.608$ and $0.798$ aren’t adjustable parameters. They came out of the
math. When we compared against computer simulations, the agreement was
within the statistical noise at every point we tested, and at points we
didn’t use to calibrate.

Second, there’s one clean fact underneath both cases. If you average
over all the possible “smudge patterns”, all the random dirty windows
the universe might hand you, what’s left is just the diagonal of the
original quantum state (in a particular basis). All the off-diagonal
crosstalk washes out. That one clean fact (“averaging leaves the
diagonal”) is what does the work for both the simple case and the complex
case. The two look like different problems but they’re the same problem
applied to different inputs.

So what

The bigger picture: observer complementarity in quantum gravity has been
studied at the level of inner-product errors, how wrong can one observer
be about a given state, and there the answer is universal. At the level
of entropies across two observers, the answer is not universal. It
depends on the complexity of what’s being observed, and in a specific,
derivable, integer-separated way.

Which is to say: it’s not that the universe has “one answer” for how much
two observers can disagree. It has a hierarchy of answers, indexed by how
structured the thing they’re looking at is. And we can actually compute
the hierarchy.