The mode paper for building: the field as invisible mold and invisible conveyor; the six-step pipeline from pattern file to finished object; trapping, moving, welding, checking. Size-free — the physics privileges ratios, not dimensions.
Acoustic levitation is a lab commonplace. A grain of sand, a droplet of water, a polystyrene bead — hang it in mid-air between a speaker and a reflector and it just sits there, held by nothing visible. The sound wave folds back on itself, and where its pushes cancel there is a quiet pocket; the grain falls in and stays. One grain in one pocket has been settled science for decades.
Now imagine not one grain but thousands, each in its own pocket, in a chamber whose walls are computed holograms — surfaces that can shape the sound into any landscape of pockets the software asks for, and reshape it many times a second. The field is two things at once. An invisible mold: the pockets trace the shape of the object being built, and grains settle into them the way sand on a vibrating plate jumps into patterns. And an invisible conveyor: shift the pattern smoothly and every pocket glides, carrying its grain — material flows from the feed throat to wherever the growing object needs it next, with no belt, no arm, no nozzle.
Nothing touches the part — there is no tool inside the chamber at all; everything that acts on the workpiece is a wave launched from the walls. When a grain arrives at its final position, a pulse of focused energy welds it to its neighbours, and only then; the rest of the time the object is held, not heated. And the whole build is checked by listening: the machine quietly rings the growing object and compares its sound to the finished object's, the way you would tap a wine glass to hear whether it is whole. When the object answers with the right chord — when it rings true — the build is done.
One honesty note before the technical sections: every individual trick in this story — levitating grains, holographic sound fields, ultrasonic welding, checking parts by their ring — is demonstrated, published art; their composition into one machine at machine scale is not. That composition is the design this paper describes (register detail: §8).
The assemble mode is six steps, run as a loop (Principles §5 states them in brief; this section is their expansion). Steps 2–5 are concurrent in steady state — feeding, trapping, moving, and welding happen continuously across different parts of the volume — but causally each grain experiences them in order.
The compiler turns the object file into a schedule of field states: for each build step, the set of points that should hold a grain, and the boundary drive — per port, per frequency, an amplitude and a phase — whose interference puts a low-pressure pocket at each of them. This is not filament-printer slicing: no privileged z-axis, no layer in the mechanical sense. A "slice" is whatever set of pockets the field can hold, move, and verify at once, ordered so that every grain's path stays clear and every weld happens against already-latched material. The inverse problem — target field to drive — and its solver tiers are owned by the Field Compiler §14.
Matter enters the volume through the axis bore — powder (grains typically tens of µm), droplets, or a melt stream, depending on which rung of the transport-phase ladder the build is running (§4). The feed is itself field-driven — no auger touches the build volume; the bore's stacked winding pairs and acoustic tube modes lift the feedstock up the axis (ruling R10, summarized in §4). The pulse alphabet's on-axis null makes the geometry possible — light around, matter through (The Photons §3).
The boundary plays the drive. Waves from every port superpose into one three-dimensional pressure landscape, and a grain in it feels a steady, cycle-averaged force toward the nearest quiet pocket — over each cycle the pushes toward the pocket outweigh the pushes away — settling there like a marble in a bowl made of sound. The physics name is the Gor'kov radiation force; §2 summarizes it and links its owner.
The compiler ramps the boundary phases smoothly; the pockets glide; the grains ride along. A trapped grain is also a steered grain: hundreds of pockets move independently at once, each on its own trajectory from feed throat to final coordinate. This is where working from the walls pays — the same drive that holds the whole population also routes it, with no per-grain mechanism anywhere (Principles §2).
When a grain reaches its final coordinate, energy is focused onto that one spot — acoustic focus for clamping and ultrasonic joining, plus electromagnetic or optical heating at the addressed point — and the grain fuses to its neighbours. Heat is spent only here: the thermal budget of the build is the sum of its welds, not a bath the whole volume sits in. §5 carries the consolidation detail and the one superseded claim this step must keep dead.
Scan mode runs underneath the whole time. After each step the growing object's ring is compared to the target's, and the next step corrects before errors can compound — assemble is a closed loop, not an open-loop printer. §7 describes the loop; the scan-mode narrative itself is owned by Scan.
Every step composes MEASURED primitives — acoustic levitation, holographic traps, ultrasonic welding; their composition at machine scale is design until the bench signs it (Principles §8b). That distinction is restated per-claim in §8.
The trap is a standing-wave phenomenon. Where counter-propagating sound
interferes, the pressure field has nodes — surfaces and points where
the oscillation nearly cancels — and a small particle feels a time-averaged
force described by the Gor'kov potential: a weighted difference of
the local pressure and velocity energy densities, whose negative gradient
is the radiation force. The weights are the two contrast factors —
monopole from the compressibility mismatch between particle and medium,
dipole from the density mismatch — and their combination, the acoustic
contrast factor Φ = f₁ + (3/2)f₂, decides
the sign: Φ > 0 traps at pressure nodes — every
solid-in-air case the machine cares about — and
Φ < 0 at antinodes. Capture is the condition that
the trap force exceeds the particle's weight.
MEASURED primitives (acoustic
levitation, acoustic tweezers). The full expressions, the equation of
motion, and the sign-and-normalization discipline — a documented convention
trap that silently inverts nodes into antinodes if quoted against the wrong
field convention — are owned by the
Field Compiler §13.1;
element-level force physics is
The Phonons §4.2.
Holographic acoustic traps. The pocket landscape does not require individually wired emitters: a printed acoustic hologram — a passive surface whose thickness profile delays each ray by a computed phase — reconstructs an arbitrary pressure landscape when driven by a plain transducer (MEASURED: Melde et al., Nature 2016). A printed hologram is fixed, so it steers by frequency; the driven boundary changes the field every millisecond; together they are the boundary condition (The Phonons §4.1, Principles §4) — printed surfaces carry the standing repertoire, the drive carries the build schedule.
Not everything is held by sound. The machine assigns each material class the force channel that grips it best — conductive metal (including molten) to induction (industrial levitation melting), charged droplets to electrostatic levitation (with mandatory active feedback — Earnshaw's theorem forbids a static trap), and everything dielectric, neutral, or needing fine placement to the acoustic channel, which is the default. The levitation-by-material-class table is owned by The Phonons §4.4.
How small can the payload get? The answer is a ladder, and it is ruling
R1-b (adopted 2026-08-01; canonical text in
REPLICATOR_CAP_STACK_RULINGS_2026-07-30.md, element-level
owner The Photons §3b):
sound is the hands; light is the fingertips.
Register note, carried verbatim from the owner: nothing here upgrades the machine's claims — the external art is MEASURED, every in-machine use is design, and "build at the nano level" is OPEN and stays off the deck.
What state should matter be in while it moves? The transport-phase ladder answers with the lowest-rung rule: use the least-unbound state the field can still grip. Owned in full, with the energy arithmetic, by Architecture §5; the shape of it:
| rung | grip | role in assemble |
|---|---|---|
| powder | acoustic + magnetic | the default rung — cheapest per kg, the pipeline of §1 as written |
| melt / droplets | EM levitation (industrial cold-crucible art) | bulk roughing: a droplet is a mesoscale voxel of ~10¹⁶ atoms that welds on contact, cutting the required placement count by the voxel's atom count |
| vapor | flux geometry | surface deposition — industrial PVD art |
| plasma | perfect charge grip, atomic voxels | reserved for three jobs: the atomic finishing pass (an object's surface is a vanishing fraction of its bulk), m/q element sorting, and the nuclear-sector stage — the last firewalled entirely to Transmutation |
Each rung up costs binding energy per kg and buys grip and resolution; the rule is to climb only when the build step demands it. Roughing goes in as droplets, finishing as the fine rungs — the same roughing-vs-finishing split every machine shop already runs, executed in phases of matter instead of tool diameters.
The bore feed (R10). Feedstock moves through the axis bores — the shortest path, and the one that preserves the chamber's azimuthal symmetry. The stacked contrawound winding pairs on the axis provide independent circuits for traveling-wave synthesis: a linear-induction elevator for conductive feedstock, acoustic tube conveying for the rest — the lift is a field, not a mechanism. The bore shares duty with the optical sightline by aperture partition or time-multiplex; the winding topology is owned by Architecture §4–§5. What feeds the bore — stores, salt/ash split, sorting ladder — is the matter loop, also Architecture §5; this paper's concern begins where matter enters the volume.
Placement makes a shape; consolidation makes an object. The machine's joining verbs, in the order the lowest-rung rule reaches for them — noting that between hold and melt sits the athermal softening rung (R15, adopted 2026-08-01: acoustoplasticity / electroplasticity / acoustic fluidization — flow without melt; Phonons §4b):
The consolidation state machine. Software tracks every particle
through a thermal state driven by whichever heat stage the build selects:
FEEDSTOCK → IN_TRANSIT → MOLTEN → WETTED → LATCHED — molten
above the melt point, wetted on contact with a latched neighbour, latched
once cooled below the glass point, when its position freezes into the
occupancy grid and it retires from the trap population. Two progress
numbers are kept — latched-inside-target and latched-outside-target — and
the second is the honest one. State machine, thermal model, and
calibration anchors:
Field Compiler §13.2.
The through-line of the section is the flagship's fourth phrase: heat is the price — paid only at the weld. Everything else in assemble is cycle-averaged force, which moves matter without cooking it.
The placement-rate argument is decisive (owner: Architecture §6). A cup is ~3×10²⁴ atoms; a minutes-scale build at atomic resolution therefore requires ~10²² placements per second. No serial architecture survives that number — a nozzle or beam placing even 10⁹ atoms per second needs tens of millions of years, and a million-tip parallel head still needs a lifetime. The only surviving branch is whole-boundary holographic patterning: a drive addressing 10⁸–10¹⁰ boundary modes at up-to-optical rates performs 10²⁰–10²² coherent operations per second — minutes-scale. The doctrine's "the drive pattern is the tool, addressing the whole volume at once" is thereby promoted from elegant to mandatory: no serial machine can ever be a replicator. DERIVED (arithmetic on the atom count; the boundary-mode budget is the design claim the bench must sign).
Two refinements keep the argument honest. First, the mesoscale-voxel relief: on the droplet rung a single placement carries ~10¹⁶ atoms, cutting the required placement count by sixteen orders of magnitude — so bulk transport is rate-tractable even semi-serially, and the parallelism requirement concentrates where it belongs: the mold field and the atomic finishing pass (Architecture §5). Second, the argument is about patterning bandwidth, not power: energy was never the deep constraint; the binding one is the coherence/fidelity of the boundary drive pattern — the same knob every other lane of this program has converged on (Architecture §6).
Assemble does not run open-loop and then inspect. Scan runs
concurrently: listening is milliwatt-class and always on, so every
instant of fabrication is also a scan
(Scan owns the mode;
Field Compiler §2–§4.5 owns
the software). The chord list in the .pattern file is
simultaneously the drive recipe and the acceptance criterion: the growing
object shifts the chamber's call-and-response book, and the build is
finished when the object answers with the target chord — when it
rings true
(Principles §6).
The loop's timing is set by bandwidth. A resonance of quality Q at frequency f takes ~Q/f to ring up, so verification speed is a frequency-allocation question — and the machine's answer is the role split: acoustic builds, EM verifies. The matter-moving chords are acoustic and slow; the EM octaves listen ~10⁴× faster, making verification effectively continuous against the build rate. Each build step is checked before the next is committed, so errors are corrected before they can compound — the closed loop is what lets thousands of concurrent placements proceed without an accumulating tolerance stack.
The deeper unity is stated once in the flagship and used here: scan and build are one operation run in opposite directions, and Dissolve is the same drive as assemble separated by one sign flip. The three modes are one field used three ways.
Claims are tagged MEASURED (published or bench-verified art), DERIVED (arithmetic on measured quantities), or OPEN (registered bet with named falsification). This paper introduces no new claims and upgraded no chip in the move; every fact keeps the tag it carries at its owner paper.
Element detail for everything this paper holds — force physics, the radiating surfaces, the optical stem and fingertips — lives in The Phonons and The Photons; solver and state-machine detail in the Field Compiler; the sibling mode narratives in Scan and Dissolve.