The Tindalosian Observation Problem

A speculative physical model of angular time, observation, and the Hounds of Tindalos

There is an old, uncharitable way of reading “The Hounds of Tindalos”: a man drugs himself, peers somewhere he shouldn’t, attracts a predator, and gets eaten for his trouble. It reads the story as a cautionary tale about curiosity. I have never found it satisfying, because it leaves every interesting question on the table. Why plaster? Why corners? Why does being seen matter to a thing that can move through time?

The rules of this exercise

Frank Belknap Long was writing cosmic horror in 1929. He was not anticipating quantum information theory, and nothing here claims that quantum mechanics predicts monsters, that consciousness collapses wave functions, or that information is some supernatural fluid you can bottle.

What I want to do instead is stricter than hand-waving. I am going to start from a handful of things that real physics and mathematics genuinely allow. Measurement creates correlations. Information gets copied into an environment. Boundary geometry can localize a field. Some quantum processes have no definite causal order. To that pile of respectable facts I will add exactly one fictional axiom, stated plainly, and then hold myself to a rule: everything I attribute to the Hounds has to follow from the established mathematics or live inside that single axiom. Nothing sneaks in through the back.

The goal is a universe where Hounds could exist without the equations having to lie on their behalf.

Long’s original problem

In the story, Halpin Chalmers takes a drug that rewrites how he experiences time. He claims he can see the whole of history at once, and he draws a line between what he calls curved time and angular time. The things he meets out there have no bodies in their own condition. They travel through angles that ought not to connect, they turn toward him the moment they notice him, and afterward they come through into his rooms by way of corners. His response is to plaster over every join and edge until the room is as close to a sphere as he can make it. The 1929 text.djvu/91) is careful about the order of events: Chalmers looks, the Hounds catch his “scent,” and the connection outlasts the vision that started it.

Read as pursuit, the story is simple. He looked into their world, drew the attention of something hungry, and it followed him back.

That reading always fought with Long’s own insistence that the Hounds are not evil in any way a human would recognize, and it makes the geometry feel like set dressing. Plaster stops something that walks through time? Observation is what does the damage? A predator that could presumably leave chooses to kill instead?

The model I am going to build treats the visible Hound differently. In its native state it is an angular thing with no particular location. Observation is what forces it into a fixed relationship with our spacetime, and that bound relationship is a kind of wound. The violence is the thing trying to end the process that is pinning it here. On this reading, every encounter we ever get to witness is a trapped projection, never the creature as it actually is.

Part I: Observation does not mean consciousness

The word observer has done more damage to public understanding of quantum mechanics than almost any other, because it smuggles in the idea that the universe is waiting on a conscious mind to look. The formalism asks for much less. It asks for a physical interaction that produces a correlation, and it does not care whether anything is awake to appreciate it.

A photon lands on a detector. A molecule bounces off a mote of dust. A photographic emulsion changes where the light struck it. Each of those is an observation in the only sense the mathematics needs.

Write for the physical degrees of freedom belonging to a Hound, and for those of some observing apparatus. Before they meet, an idealized joint state can be written as a clean product:

Here is a density operator, which is just a compact way of bookkeeping the probabilities and coherences a quantum system has available. The tensor product is the mathematical way of saying that knowing about tells you nothing about .

Let them interact, and in general that clean split stops being possible:

The two systems are now correlated. Sometimes that correlation rises to entanglement, but entanglement by itself is not a homing beacon and cannot carry a message faster than light, so we should not let the word do any heavy lifting it has not earned.

You can put a number on the total correlation using the quantum mutual information:

where

is the von Neumann entropy. Strip away the notation and is answering one question: how much do these two systems now tell you about each other?

That gives Long’s “scent” an honest translation. The original encounter lays down a physical correlation between Chalmers’s world and the angular thing on the other side of the drug.

Correlation on its own cannot chase anyone or pull a monster through a wall. For that we need physical systems that carry the record, and an interaction through which a Hound field can actually couple to them. Both come later. First the record has to spread.

Part II: How an encounter spreads into the world

Chalmers does not keep the encounter sealed inside his skull. He remembers it. He writes it down. He talks about it, remodels his rooms, and frightens the people around him into acting differently. Every one of those is a physical record, sitting in ordinary matter.

Let the relevant Hound alternatives be , each with probability . Some fragment of the environment, a page of Chalmers’s notes, a camera sensor, the inside of another person’s head, picks up a conditional state . How much a later inspector could actually learn about from that fragment is capped by the Holevo quantity:

Behind the intimidating symbols is a plain question. If you handed someone fragment , how reliably could they tell you which Hound-related thing happened?

As more and more environmental degrees of freedom line up with the event, decoherence sets in. If the full description includes an environment we have no hope of monitoring, the state actually available to Chalmers is

That partial trace is us throwing away the environmental detail we cannot track. Interference between the alternatives becomes, for all practical purposes, unrecoverable. Nothing was erased by human ignorance. The information that would let you tell the alternatives apart simply scattered into places no one can reach.

There is a whole subject, quantum Darwinism, devoted to how selected information ends up redundantly stamped across many environmental fragments. Call the smallest slice of the environment that still carries at least of the available classical information. The redundancy is then

If one percent of the environment already tells you a Hound was seen, then something like a hundred separate fragments are each carrying that fact. That redundancy is exactly why a roomful of people can agree an event happened without any of them interrogating the original system.

The scent as a difference between histories

Now “scent” can be pinned down properly. Take two possible environmental states at the same moment . One,

comes from a history where the Hound was observed. The other,

comes from an otherwise identical history where it was not. The distance between them is measured by

Those double bars are a trace norm, but the operational meaning matters far more than the construction. Trace distance tells you how well the best conceivable measurement could tell the two states apart. Zero means no experiment on could ever separate the histories. Anything above zero means some physically recoverable difference is still out there.

So here is a Tindalosian diagnostic worth naming:

with standing in for scent strength. The trace distance is real physics. The claim that a Hound can feel it is the fiction, and I want to keep that line bright.

This also heads off a mistake people make with the story. The scent is not necessarily sitting on Chalmers. Right after the encounter, sure, the sharpest differences are in his nervous system and his behavior. Give it time, though, and the difference gets copied into Frank, into written records, into the reshaped architecture, into scattered photons and stirred air and a mountain of microscopic states. A small fragment can lose almost all of its telltale power,

even while the environment taken as a whole stays distinguishable:

The scent can survive as a global fact long after it has become too thin for any one local fragment to reveal.

There is a subtlety I do not want to paper over. If you treat the post-encounter universe as genuinely closed and let both alternatives run forward under the same unitary evolution, their global trace distance never changes at all. In practice no observer ever controls the whole environment, so the locally recoverable evidence can get fantastically hard to dig out. “Still encoded somewhere in the global state” is a very different thing from “usable by a frightened investigator.” My Hound is being defined, on purpose, as sensitive to precisely that distributed causal displacement that a human never could recover.

The phrase cross-temporal correlation can be demystified too. Ordinary physics already uses connected two-time correlation functions,

to ask whether a state now is still statistically tied to a state then. Non-Markovian process tensors push the same idea out to whole sequences of pokes and measurements, describing the cases where a system’s later behavior cannot be predicted unless you remember what was done to it earlier.

So the conservative, physics-respecting version of the claim is this:

The Hound does not smell backward through time. It detects present physical systems whose states are still correlated with a past encounter.

And the stronger version, the one I am charging entirely to fiction:

The Hound perceives the complete multi-time correlation as a single angular object, rather than reconstructing it one moment at a time.

For Tindalos, that has a cruel implication. Wiping Chalmers’s memory does not necessarily wipe the encounter. Frank still remembers. The notes are still on the desk. The plastered room is itself a piece of evidence. Every independent record is one more physical carrier of the original correlation, and the Hound does not care which one it reads.

But none of those records exerts some abstract “informational force” on their own. To make anything happen we have to write down an ordinary interaction between their physical degrees of freedom and a Hound field :

The source is built from the actual matter carrying the records, not from information floating free of any substrate, and sets how strongly the two couple. The Holevo information and the redundancy tell you how far the source has spread. It is that physically drives the field. Keeping those two ideas separate is the only thing standing between this model and turning information theory into a magic wand.

Part III: Why a corner can hold a Hound

General relativity does not care about corners. It assigns no special moral or metaphysical status to a sharp angle. A room corner is a boundary in matter, not a tear in the manifold, and anyone telling you otherwise has left physics behind.

Spectral geometry is more helpful, and it hands us a genuine mechanism by which boundaries and corners can trap localized field modes.

Picture a two-dimensional wedge with opening angle , which you can think of as a slice across a three-dimensional edge. Let the Hound field satisfy a Robin boundary condition on each wall,

where measures how fast the field changes heading straight out of the wall. Robin conditions blend a field’s value with its slope, and they turn up all over wave mechanics, diffusion, acoustics, and field theories that live inside boundaries. There is nothing exotic about them.

For a wedge with , the function

is an exact eigenfunction of the Robin Laplacian. It piles up at the vertex and dies off exponentially as you move away. Differentiate it and you get

so the corner eigenvalue is

A lone flat wall starts its spectrum at . The corner sits below that edge continuum by

and is the binding advantage the corner throws in for free.

The physical reading is about as direct as physics ever gets. Open the wedge toward and the two walls flatten into one surface, falls to zero, and the special corner mode dissolves back into the continuum. Close the wedge to an acute angle and the binding tightens. The localization length along the bisector is

which says the sharper the corner, the harder it grips the mode. A hairline crack can be worse than an honest right angle, and that is going to matter later.

Corner-localized Robin eigenfunctions, how they shift with opening angle, the tunnelling between separated corners: all of that is settled spectral theory. The fiction only walks in when we decide those modes are alive.

Part IV: From a bound mode to a manifestation

Give some ordinary field dynamics:

Here is the propagation speed inside the projected Hound field and fixes its natural frequency scale. I am not assuming either equals the speed of light or the mass of any known particle. They are parameters of the proposed sector, and I would rather leave them honestly undetermined than pretend to know them.

Expand the field in spatial modes and keep the corner mode:

Project onto and what falls out is a driven, damped oscillator:

where

is the slice of the record source that actually overlaps the corner, and

That single equation is a two-factor theory of manifestation, and I like it because it refuses to let either half of the story win alone. The geometry supplies a mode. The record source drives it. A cathedral of a thousand corners produces no Hound if there is no Tindalosian correlation to feed it, and an archive drowning in records produces no local body if there is no geometry to hold the mode. The Hound shows up only where epistemology and architecture happen to cross.

The same equation sorts behavior into three regimes:

If , the critical opening angle is

Below the field does not merely fit into the corner. The un-manifested state goes unstable, and a defect that sharp will nucleate a manifestation the instant it is coupled to even a feeble source. Sharp geometry does not invite the Hound. It stops giving the Hound any choice.

Part V: Edges, corners, cracks, and plaster

The pecking order is easiest to see in rectangular geometry. In a half-space the mode carries eigenvalue . Fold two walls into a right-angle edge and you get

Bring in the third wall at a room corner and

Every extra intersecting boundary deepens the binding, which gives a natural ordering:

A Hound should not bloom from just any corner it fancies. It should settle first at the defect with the lowest compatible eigenvalue and the strongest overlap with , the way water finds the lowest drain.

Chalmers’s plaster erases the exact polygonal vertex, and that is worth something, but “rounded” and “safe” are not the same word. Smooth curved boundaries can hold Robin bound states too. The parameter that decides it is roughly , with the radius of curvature. When , the boundary bends slowly compared to the field’s own localization scale, and the sharp-corner enhancement is strongly suppressed. When , the field can still read the rounding as a tight confining defect and settle in anyway.

So a real defense needs a large enough radius, the right boundary material to set , and someone who keeps checking for cracks, because an acute enough crack will bind the field harder than the corner it replaced. Chalmers plastering in a panic, missing a hairline split behind the wainscoting, is not a plot hole. It is the model working exactly as written.

Part VI: More than one corner

Two similar corners a distance apart carry nearly identical modes and . Where those modes overlap, they combine into symmetric and antisymmetric versions:

The splitting between their eigenvalues is exponentially small,

with a prefactor that depends on the whole geometry.

Until some environmental difference picks a winner, the lowest Hound mode can be spread across both corners at once. This is honest mode tunnelling, not a big animal standing in two places like a bad special effect. Any small asymmetry, a draught, a warm wall, a floor that shivers when a truck passes, a slightly stronger record-source overlap on one side, tips the balance and drops the manifestation into a single corner.

It also predicts that jittery, teleporting movement between equivalent corners without the Hound ever bothering to cross the room in between. It was never really in the room. It was in the modes.

Part VII: What happens when the records are destroyed

Say you strip out every reachable source at time , so . The field does not vanish on the spot. In the underdamped regime it rings down:

with

and a characteristic decay time

An amnestic operation buys you nothing unless it removes every relevant physical record, and even a perfect one leaves the existing bound mode to ring itself out. For the length of that ring-down the Hound is still here, still localized, getting less stable by the second, and by every reasonable reading of its behavior, desperate to leave.

One aside on the famous blue vapor. It should not be described literally as entropy leaking out. Entropy is a property of a state, not a gas you can watch pour across a floor. In this model the vapor is ordinary matter or radiation, thrown off as energy drains out of the decaying Hound mode and back into fields we can actually measure.

Part VIII: Memory is not yet time travel

An open system carrying environmental memory can obey a non-Markovian equation of motion:

The kernel says how strongly the past still leans on the present. A kernel that decays slowly gives the Hound a long physical memory of what it has interacted with, and that alone can account for persistence, for delayed reactions, for a scent that outlives the moment that made it.

What it cannot do is let the future reach back and edit the past. Ordinary non-Markovian dynamics stays stubbornly causal, and I am not going to pretend otherwise for the sake of a scarier story.

Long’s stronger idea treats the observation and the observer’s eventual death as two edges of a single relation that the Hound experiences whole. Getting that requires the one genuinely Tindalosian thing I promised to declare out loud:

The angular sector is described by a causally nonseparable process whose projection into curved spacetime becomes ordered only when it couples to ordinary records.

Indefinite causal order is not my invention. It is a real research topic. Process-matrix theory allows correlations that refuse to be sorted into any fixed order like or , while still keeping every local quantum operation perfectly valid. It does not hand you arbitrary time travel for free. My fictional addition is narrow: Hound processes live in such a sector as their natural home, and they only pick up a definite order when they couple to our environment.

To Chalmers, the whole thing runs in sequence:

To the un-projected Hound those are not four experiences in a row. They are the ordered shadow of one relation. Observation is what forces the relation into a sequence, and forces the Hound, along with it, to take on here, now, before, after, and at the very end, I.

Part IX: Why it attacks

Killing Chalmers cannot unmake the observation. Frank still holds records, so do the room, the notes, and the wider environment. His death only sets his own source term to zero:

That closes off one proliferating branch and leaves every other record in the network untouched.

Nothing in the model lets the Hound reverse a finished measurement by murdering the man who made it. Its useful targets are the systems still actively producing records and still driving the localized mode. Anyone who remembers, talks, investigates, or reconstructs the encounter has made themselves part of the apparatus holding the Hound inside curved time. That is a grim thing to realize about yourself halfway through an investigation.

From where we stand it looks like predation, because it keeps converging on witnesses. From the Hound’s side it is closer to an animal chewing its own leg out of a trap, violence aimed at the machinery of its confinement.

It also explains the nastiest property of the whole business: rebuilding an erased encounter can bring it back. An investigator who reasons the missing cause out of blank pages, suspiciously rounded architecture, and a terror nobody can source is manufacturing fresh, physically distinguishable records. climbs back off the floor. If a workable corner mode is still there, the oscillator gets driven all over again, and the act of understanding becomes the act of summoning.

The Hound follows the counterfactual trail of everything that would never have happened if Chalmers had kept his eyes shut.

What the model predicts about Tindalos

Laid out, the deductions form a fairly tight set:

1. Visible Hounds are projected bodies. A corner localizes a much more general angular excitation.
2. A sharper defect yields a tighter, lower-energy manifestation. Cracks can beat corners.
3. Three-wall vertices bind harder than edges. Manifestation should choose the highest-codimension defect on offer.
4. Geometry alone is not enough. A physical record source has to overlap the localized mode.
5. Records alone are not enough. Without compatible geometry the field never forms a local body.
6. Rounding is a quantity, not a guarantee. Safety rides on radius, material coupling, and the field’s localization length.
7. Equivalent corners can share one mode. Tiny environmental nudges decide where the Hound actually appears.
8. Erasure has a decay time. Pull the last source and relaxation begins; the thing does not blink out.
9. Killing a witness cannot erase distributed records. It only stops that witness from making new ones.
10. The violence is escape behavior. Definite sequence, place, and identity are what injure the projected organism.

Which leaves the last deduction, the one about Tindalos itself.

If Hounds are defect-bound excitations, they may not be apex predators back home at all. They may be as unremarkable there as surface ripples, phonons, or eddies are here. We only ever meet them as monsters because the very coupling that makes one visible is the thing that wounds it. Our looking drags an angular process into causal order, our redundant records hold it in place, and an architectural defect gives the pinned thing a body to scream with.

We call what follows a hunt because we only ever get the ordered projection: first we look, then it appears, then someone dies. For the Hound, that same sequence is what it feels like to become finite.

Where established physics ends

Established mathematics or physics, all of it load-bearing here:

• Density operators, mutual information, and the Holevo quantity.
• Decoherence and redundant environmental records.
• Robin boundary conditions.
• Corner-localized Robin eigenmodes and their angle-dependent spectra.
• Tunnelling and exponentially small splitting between separated corner modes.
• Damped mode dynamics and non-Markovian memory kernels.
• Operational frameworks for indefinite causal order.

Fictional assumptions, declared so nobody has to guess:

• A Hound field coupled to physical record carriers.
• The values and meaning of , , , , and .
• Identifying corner-bound modes with manifestation.
• A native angular sector built from causally nonseparable processes.
• The claim that projection into ordinary causal order hurts, and that it manufactures a self.

The mathematics cannot prove the fictional premises, and it was never asked to. What it can do is guarantee that, once those premises are on the table, their consequences follow instead of being invented as convenient. That is the whole of the Tindalosian Observation Problem. The puzzle was never why looking reveals the Hound. It is why being looked at forces the Hound to become something that can be revealed.

References

• Frank Belknap Long, “The Hounds of Tindalos,” Weird Tales 13, no. 3 (March 1929), original text at Wikisource.djvu/91).
• M. Khalile, T. Ourmières-Bonafos, and K. Pankrashkin, “Effective operators for Robin eigenvalues in domains with corners,” arXiv:1809.04998.
• B. Helffer and K. Pankrashkin, “Tunneling between corners for Robin Laplacians,” arXiv:1404.4765.
• P. Exner and A. Minakov, “Curvature-induced bound states in Robin waveguides and their asymptotical properties,” preprint.
• W. H. Zurek, “Decoherence, einselection, and the quantum origins of the classical,” doi:10.1103/RevModPhys.75.715.
• M. Zwolak, H. T. Quan, and W. H. Zurek, “Quantum Darwinism in non-ideal environments,” arXiv:0911.4307.
• H.-P. Breuer, E.-M. Laine, and J. Piilo, “Measure for the Non-Markovianity of Quantum Processes,” doi:10.1103/PhysRevLett.103.210401.
• F. A. Pollock et al., “Non-Markovian quantum processes: Complete framework and efficient characterization,” doi:10.1103/PhysRevA.97.012127.
• O. Oreshkov, F. Costa, and Č. Brukner, “Quantum correlations with no causal order,” doi:10.1038/ncomms2076.