André Nachbin

Tradutor simultâneo

Equações Diferenciais em Modelagem Matemática Computacional, André Nachbin (IMPA) and Esteban Tabak (Courant), 21º Colóquio Brasileiro de Matemática, IMPA, July 1997, ISBN 85-244-0127-3. Chapter 1, § 1.1.1, before any mathematics:

Two words in that do the work. Simultâneo: the translator never finishes one side before starting the other, which is precisely how a dictionary written for one direction ends up being used in both without anyone deciding to. And e vice-versa, the harder half — carrying a result back. That is where this corpus has lost things: a theorem proved about a model, quoted about the thing. Volume XIII, Chapter 9 is the theory of that sentence, and it is open.

A dictionary is a model plus its error term

The full water-wave problem has dispersion $\omega^{2} = gk\tanh(kh)$. The long-wave model everyone actually computes with replaces it by $\omega = \sqrt{gh}\,k\,[\,1 - (kh)^{2}/6 + 19(kh)^{4}/360 - \cdots]$, and that $(kh)^{2}/6$ is where KdV's dispersive derivative comes from. ch-nachbin-verify.py computes both: at $kh = 0.05$ the model is exact to one part in $10^{10}$; at $kh = 1.5$ it is 15% wrong. Between them the error rises like $(kh)^{6}$ — doubling $kh$ from 0.2 to 0.4 multiplies it by 63.

That is the entire difference between a dictionary and an analogy. A dictionary is exact in its limit, wrong outside it, and tells you the exponent of its wrongness. You can push it until it breaks and know when it broke. An analogy has no such term; it cannot be pushed, only agreed with.

Time reversal, walking droplets, and a wave at a fork

Time reversal in random media. Send a wave through a disordered medium, record it at a small aperture, reverse the recording in time and send it back: it refocuses on its source — and refocuses more sharply when the medium is rough than when it is smooth. Disorder improves resolution, because multiple scattering makes a small aperture behave like a large one. Fouque & Nachbin (2003), then with Garnier and Muñoz-Grajales in Phys. Rev. Lett. (2004). This corpus spent yesterday asking whether a system's path can be undone; here is a field where that question is asked with instruments.

Pilot-wave hydrodynamics. A droplet bouncing on a vibrating bath walks, guided by the wave field it has itself created — so its future depends on the field, and the field is a record of where it has been. Milewski, Galeano-Rios, Nachbin & Bush, J. Fluid Mech. 778 (2015); Nachbin, Milewski & Bush on tunnelling, Phys. Rev. Fluids 2 (2017); and Nachbin alone on Walking droplets correlated at a distance, Chaos 28 (2018). That is path dependence with a mechanism, in a bath you can photograph. The holonomy test this series wrote yesterday is hunting for the same property in storms, where nobody can see the wave field.

Solitary waves at branching points. Nachbin & Simões, Solitary waves in forked channel regions, J. Fluid Mech. 777 (2015), and the earlier paper on channels with abrupt turns (2012). A wave arriving at a fork in its domain — the topology of the container deciding what the dynamics can do. A series whose central claim is that the shape of the medium governs the transition has, in that pair of papers, a worked instance of exactly that, with experiments.

At $\gamma = 0$ the flow is symplectic and the pulse comes home to machine precision. Turn damping on and it is not symplectic any more — a damped mechanical system is the standard example of contact Hamiltonian dynamics, the extra Reeb direction carrying the dissipated action — and the pulse does not come home at all. The saturation near 1.0 is just the field having decayed to nothing.

Why the momentum, and why damping ends it

This is the layer the series usually skips — the result gets stated and the reason gets left as an exercise nobody sets. It is worth four paragraphs.

A trajectory is not a sequence of positions. Photograph a pendulum at one instant and you cannot tell whether it is swinging left or right. The photograph is $u$, and the ambiguity is not a limitation of the camera — the information is genuinely absent from the picture. What answers “which way is this going” is the velocity, and it is a separate coordinate, not a derived one. The state is the pair. That is why reversal touches only $v$: position is what the system is, velocity is where it is headed, and only the second one has a direction to undo.

And the equation cooperates because of a parity. The wave equation contains $u_{tt}$ and nothing else in time — an even number of time derivatives. Substitute $t \to -t$ and $u_{tt}$ is unchanged, so the equation is literally the same equation run backwards. What changes is the solution: $u$ stays, $u_t$ flips sign. The reversal is not a trick performed on the system; it is a symmetry the system already has, and $(u,v) \to (u,-v)$ is simply how you enter that symmetry. Which is also why the failed attempt failed. Recording $u$ at an aperture and replaying it is photographing every frame of a film and then asking the projector to run it backwards by showing the same photographs in the same order. Every frame is correct. The direction was never in them.

Damping breaks the parity, and that is the whole of it. A term $\gamma u_t$ has an odd number of time derivatives. Under $t \to -t$ it changes sign, so the time-reversed equation is not the original equation — it is one in which damping has become pumping. To run a damped system backwards you would need a different law, one that spontaneously amplifies exactly what the original dissipated, matched mode for mode. Nothing supplies it. This is why you can watch a film of a real pendulum and know at once which way the film is running, and cannot for an ideal one: the ideal pendulum's law is blind to the direction of time and the real one's is not.

Which is why the two numbers in Part III are worth having. $3.6\times10^{-15}$ and $0.96$ are not a demonstration that damping damps. They are the receipt for a trade this series made in Volume I and has been spending ever since: every fold that cannot be undone, every threshold that is crossed once, every claim that history matters and the path is not recoverable, was bought at the counter where reversibility was handed over. The escape from Liouville is not a technical convenience. It is the moment the framework chose to be about systems that remember, and remembering is what it costs to be unable to go back.

Yes: it is reverb, and the word is not a metaphor

Clap once in a cathedral and once in a cupboard. Same clap. What comes back is entirely different, and what differs is not the clap — it is the room. That tail is the impulse response, and it is the cleanest everyday instance of the thing said above: a feature of the space, not of the thing.

And the parameter is the same parameter. Reverberation time is set by absorption; a damped mode has poles a distance $\gamma$ off the real axis; the resonance half-width is $\gamma$; the decay rate is $\gamma$; and the reversal defect measured in Part III is $\gamma$. One number wearing four names, depending on whether you are in the time domain, the frequency domain, an architecture textbook, or a phase space:

Read that table sideways and it says something worth keeping. A resonance sharp enough to be measured is already a resonance that has spent its reversibility. A perfectly reversible cavity has poles exactly on the axis — infinitely sharp lines, infinite $Q$, a tail that never ends and therefore never tells you anything. The moment a line has width, something has been lost, and the width is how much. Linewidth and irreversibility are the same fact read in two domains.

Noise rather than pulses is the part this series should notice. You cannot clap at a hurricane. You cannot impulse a cell, or a market, or the arithmetic contact form. What every one of those has instead is ambient noise, continuously, for free — and the result above says the reverberant field of a noise source is enough to reconstruct where the source was. The requirement is not a controlled input. It is enough receivers.

The last turn of it is the one that matters for this series. Reverb is a room remembering a sound after the sound has stopped. That memory is only possible because the room absorbs — a perfectly reflective room does not reverberate, it simply never goes quiet. So the tail exists because energy is being lost, and the tail is informative about the room for exactly the same reason. Memory and dissipation are not two things. Which is the same sentence as before, arriving from the other side: an attractor is a memory of having forgotten, and a reverb tail is a room telling you what it took.

You cannot clap at a hurricane. It turns out you do not have to.

The objection that has stood against every attempt in this series to test a real system is that you cannot excite it. There is no impulse you can deliver to a storm, a cell, a market, or an arithmetic contact form, so the impulse response — the thing that carries the medium's whole structure — appears to be out of reach. That objection is twenty-five years out of date, and the papers that retire it are sitting in the reference list above.

Correlate the ambient noise recorded at two passive sensors, over long enough, and the Green's function between them emerges from the correlation. No source, no control, no experiment performed on the system at all — only listening, twice, and multiplying. Weaver & Lobkis got it from thermal fluctuations and titled the paper Ultrasonics without a source. Campillo & Paul got it from the diffuse seismic coda, in Science, which is to say they recovered the Earth's impulse response between two stations from the Earth's own background rumble.

Two more from the same list are worth naming because each answers something this corpus has been circling. Draeger & Fink refocused elastic waves in a chaotic silicon cavity using one channel — a single transducer, where a clean cavity would need an array, because chaos plus reverberation does the aperture's work. And Weaver & Lobkis, Temperature dependence of diffuse field phase, is the medium-drift confound measured rather than argued: exactly how much a degree of warming decorrelates the field between recording and replay. This series raised that same confound about SST and latitude two days ago without knowing acoustics had put a number on it in 2000.

They had the problem and the solution. The translation is the missing part.

This corpus has been in archaeoacoustics for a while without calling it that. ch-seismic § 8 argues Borobudur's seventy-two latticed stupas and Bagan's bell-shaped zedi are cavity resonators on a self-similar base, and it does so under a withdrawal notice: an earlier draft claimed the stupas form a frequency ladder covering the seismic spectrum, no measurement was found to support it, and the claim was pulled and restated as hypothesis. There is also a Ħal Saflieni page — a tone generator for Newark Wellness Soundworks, not a piece of mathematics. What follows checks the physics that page rests on, and finds one thing wrong with it.

Now put those numbers through the identity this chapter spent its last three sections building. A reverberation time is a decay rate, a decay rate is a resonance half-width, and a half-width is an inverse Q. Nothing is being compared; it is one quantity written four ways. From γ = 3 ln 10 / T and Δf = γ/π:

The brass pots in the walls

They had the problem and they had the solution. Vitruvius describes bronze vessels set into Greek theatre seating; medieval builders put earthenware jars into church walls and vaults, necks open to the room. A recent review of roughly fifty European churches puts their resonances at 80–450 Hz, mean 226 Hz, and measures up to 25 dB of amplification inside the cavity, about 6 dB at 10 cm from the neck, and nothing beyond that. The effect on a room's reverberation time comes out inconsistent — increases in some studies, decreases in others, one confined-chamber test going 0.58 s to 0.71 s on opening the pots — and the reviewers say plainly that the differences can be the same size as the fluctuation in the measurement.

Read that as a failure and you have the wrong problem in mind. Here is the translation.

So the pots are not a failed broadband treatment. They are a working narrowband one, being judged by a broadband instrument. The builders were not trying to change T. They were trying to kill a tone — the one note in the nave that rang when the choir hit it, the mode whose Q was too high for the room to be sung in. They put a matched oscillator in the wall at that frequency and drained it. What they engineered was not γ. It was γ(f).

The count is group theory, not dynamics

Gallot, Catheline and Roux find coherent backscattering enhancement not only at the source but at other points in the cavity: one in a one-dimensional cavity, in a disk, and in a symmetric chaotic plate; three in a two-dimensional rectangle; seven in a three-dimensional parallelepiped. One, three, seven. A box with a mirror plane in each of d axes has reflection group ℤ₂^d, of order 2^d; one element is the identity, which is the source itself. 2^d − 1 reproduces all three counts.

The disk and the chaotic plate give the same answer as the one-dimensional cavity, because each has one reflection axis. Chaotic ray dynamics does not destroy the enhancement; only breaking the symmetry does. The refocus lands on the orbit of the source under the cavity's symmetry group, and the group does not care how the trajectories behave in between.

Which puts a question to the Oracle Chamber that is answerable with a tape measure. The niche in that room is where the sound is presumed to have been made. If the chamber has a mirror plane and the niche sits on the image of somewhere else, then the niche is where the sound arrives. A voice at the source and a listener at the image both sit on the enhancement. This series has no measurements of that room and is not going to pretend otherwise — but the prediction is cheap, falsifiable, and nobody in the archaeology literature found here has run it.

Place in the Series

The register is standard-setting, not evidential. Nachbin does not work on dm³, has no connection to it, and nothing here is evidence for anything in this corpus. What his work supplies is the benchmark the series keeps failing to meet: a translation between two descriptions with the error term written down, pushed until it breaks, and the breaking point reported. Over the past week this series has been wrong by a factor of five on a timing prediction, wrong about which SST column was a control, and wrong in calling an asymptote a constant — every one of them a translation asserted without its error term.

He belongs beside Tao for a precise reason. Tao's Collatz paper is a gap closed by changing the measure a statement was made in; Nachbin's working method is the discipline that makes such a change legitimate rather than convenient — you may swap descriptions whenever you like, provided you can say what the swap costs. Together they are the two halves of Chapter 9: what a translation transports, and how far it can be trusted before it stops.

One attribution note, because this gallery exists for them. André Nachbin is not to be confused with Leopoldo Nachbin (1922–1993), the Brazilian mathematician of Nachbin's theorem and holomorphy in infinite dimensions. No relation between them is established in any source read here, and none should be assumed in this corpus.

References