The Replicator

Principles — what the machine is, the small set of principles it runs on, and where each principle is worked out in full. Written to be read first, and to be explainable onward to a non-technical reader.

v1.1 · 2026-08-01 (v1.0 2026-07-28) · Changelog: instance content moved out to the hardware papers; the 2026-07-30 rulings (former §8c) integrated into the body sections they belong to; network map restructured to the four-layer set. · Role in the set: this paper carries no new results and no research-note detail. Every section states one principle, gives its status honestly, and links the member paper that owns it. If this paper and a member paper disagree, the member paper wins — then fix this one.

0 · The machine in one paragraph

The Replicator scans, builds, and un-builds objects using shaped fields instead of tools: a commanded boundary enclosing a build volume, nothing moving inside, every surface of the boundary a computed hologram. Sound is the muscle — the hands: standing acoustic holograms trap particles, carry them, and fuse them. Light — EM from microwave to optical — is the address, the measure, and the energy channel, and at the finest scale the fingertips, placing what the hands are too coarse to hold (optical gradient traps; R1-b). Objects are stored as holographic patterns — boundary recordings that reconstruct the object when replayed, the way a photo stores an image, not a list of pixel values — natively encoded, as a JPEG is, in frequency coefficients: the chord list, the resonances the object rings at with the boundary drive that addresses each one; like a progressive photo, the object sharpens as chords arrive. Scanning is call; building is call-and-response repeated until the workpiece answers with the target chord.

One realized instance (RH-1, a free-standing three-plate column) is specified in the hardware papers; a table-top single-plate device and an orbital shell are the same architecture at other points on the family.

The campfire version. Tap a wine glass and it sings — and the note tells you about the glass: its shape, its material, whether it is cracked. Every object has a voice like that. This machine is a room whose walls are speakers and antennas. To scan, it sings to the object and listens to how it sings back; the echo — the object's chord — is saved as a small file, the way sheet music is the song. To build, it runs the trick backwards: as sand on a vibrating plate jumps into patterns, the machine plays a chord whose sound forms an invisible mold in mid-air, powder falls into the mold's pockets, and light welds each grain where it sits. It keeps listening the whole time, and it is done when the object sings the same chord as the file — when it rings true. In four phrases: sound is the hands, light is the eyes, the pattern is the memory, and heat is the price — paid only at the weld.

1 · Three parts: storage, compute, chamber

The machine decomposes into exactly three subsystems, and — a ruling that decides the architecture — they are three separate cavities, not one:

One mathematics unifies the three: interfere a reference wave with an object wave, store the interference, re-illuminate to reconstruct. Memory, processor, and actuator are that one holographic operation wearing three hats, sized to three wavelengths. The compiler is what makes them one machine.

2 · Principle: everything happens from the walls

There is no tool inside the build volume — no nozzle, no gantry, nothing that moves. All the machinery is in the walls: dense arrays of small emitters and receivers. Walls are enough because waves superpose: choose each emitter's amplitude, phase, and frequency and you have chosen the summed field at every point of the volume at once — exactly so, by the Kirchhoff–Helmholtz theorem: the field on a closed surface fully determines the field everywhere inside it. Control the surface and you control the interior; there is nothing left inside to control.

Read backwards, the same statement is measurement: every wave scattering off the workpiece crosses the walls on its way out, so recording at the walls captures everything that can be known about the interior from outside. Command in, read out — one surface does both. And the walls buy parallelism no tool head can have: a mug is ~10²⁴ atoms; building it in minutes means ~10²² placements per second — a nozzle placing a billion atoms per second would need millions of years. Only a surface that patterns the entire volume in parallel reaches replicator rates (Transmutation §6.9).

3 · Principle: two carriers, one language

The chamber speaks two field types — phononic (sound: force-rich, 10⁵× shorter wavelength at a given frequency, needs a medium) and photonic (EM from microwave to optical: fast, penetrating, information-rich) — described by one knob list, treated by the compiler as one language:

Forces are derived quantities — quadratic, time-averaged combinations of the knobs: intensity, radiation pressure, the Gor'kov trap potential, acoustic streaming, angular-momentum torque, beat-frequency envelope forces, and heat. The honest asymmetries matter: air-borne sound gives up polarization to buy its short wavelength, and the EM channel earns its place by speed, bandwidth and sensing precision, not by mode count. Full treatment: The Two Carriers.

Control and muscle (R1, adopted 2026-07-30). Light addresses, measures, and calibrates; sound and induction exert force and heat; electrons only amplify at the boundary edge and carry DC power — never information. Not a preference: cross-carrier couplings in the volume are parametric (field compiler §17.6 bound i), so light cannot be muscle; its whole value is being the machine's independent information plane. Refined 2026-08-01 (R1-b): light's direct radiation force is negligible as bulk muscle but is the precision-placement channel at the smallest payloads — sound is the hands, light is the fingertips (The Photons §3b). (Full text and derivations for all rulings: REPLICATOR_CAP_STACK_RULINGS_2026-07-30.md, R1–R14 + amendments.)

The optical stem, and the topological alphabet (R2-b, R13, adopted 2026-07-30). The bore optical channel is one bidirectional holographic port: it speaks the caps' toroidal-pulse alphabet at optical frequency and reads back by state tomography — a density-matrix hologram in which even the loss of purity is information about the object. And on both carriers, field states can carry integer topological labels (skyrmion numbers) that survive the chamber's own reverberant chaos — the robust register of the machine's language. Full treatment: The Photons.

4 · Principle: every surface is a computed hologram

An instance's EM-plate spiral gratings, its graded phononic hex screens, and its q-BIC leakage maps are not three tricks — they are three outputs of one design procedure. Record the interference of the reference wave the feed actually launches, Ψ_ref, against the object wave wanted in the volume, Ψ_obj, and write it as surface impedance: Z(r) = X₀ + M·Re{Ψ*ref·Ψobj}. Running the reference wave across the modulated surface reconstructs the object wave. The plate pattern thus stops being a shape we choose and becomes a pattern we compute — the passive, printed half of the boundary joins the drive as compiler output; a printed hologram is fixed, so it steers by frequency, the driven half changes every microsecond, and together they are the boundary condition. MEASURED art on both carriers (leaky-wave holographic antennas; printed acoustic holograms). Full treatment: Mechanical Construction §2–§3.

Chirality (R4, adopted 2026-07-30). Driven structures are chiral and separately addressable; passive structures are achiral and neutral. Handedness mixed inseparably at fine scale cancels (the double gyroid is achiral); handedness in independently fed enantiomeric substructures — mirrored caps, contrawound winding pairs — is a knob. Diversity is only useful if addressable; unaddressable diversity averages to zero.

Worked example: one driver plate, dimensioned (added 2026-08-03, revised same day). The principle above — hologram, not shape choice — is not just a claim about the abstract surface impedance equation; it is buildable, and working the actual numbers surfaces real corrections, including one that questioned a whole capability. Each of the three plate assemblies is a dual-carrier coaxial horn: three PZT elements drive a common throat, a contrawound torus doubles as the horn's ring-radial flare, a graded ceramic gyroid carries the flare from throat to mouth while doubling as the EM dielectric, and one perforated plate — a Fibonacci-mesh micro-horn field (21×34 crossed spirals, the Vogel/sunflower lattice) plus 12 equiangular EM slots — is the shared radiating face. The bore recorded in the build spec (Ø130) turns out to be sized for a capability the machine doesn't need at that shared axis: neither the optical stem nor fine-feedstock transport needs more than a few mm: the real driver was bulk ingot-scale induction melting, which wants its own coil, not a shared one. Dropping that capability shrinks the bore to Ø12; the throat, no longer bore-constrained, is set instead by three PZT elements not overlapping at 120° (Ø36). Refitting the horn profile to the Webster area law with the new numbers gives a radius compression of 11.2:1 — this time overshooting the 6.7:1 a golden-ratio elongation would predict, having undershot it (1.84:1) before the bore fix. Landing on opposite sides of that number across two independent revisions is the point: the fit follows the physics each time, not an aesthetic prior (R3). Full derivation, the bore-sizing argument, and the ripple into the hardware spec: Mechanical Construction §3c.

5 · Principle: three modes, one field

The machine has three operational modes, all of them the same field used three ways. (Each mode has its own member paper, written easy-first-then-technical: Scan, Assemble, Dissolve — added 2026-08-01. The step lists below are the summaries; the mode papers own the detail.)

Assemble

1. Slice. The compiler turns the object file into a schedule of field states: for each build step, the points that should hold a grain and the boundary drive — per panel, per frequency, an amplitude and a phase — whose interference puts a low-pressure pocket at each of them.
2. Feed. Powder (grains tens of µm across), droplets, or a melt stream enters through the axis bore.
3. Trap. The boundary plays the drive; the crossing waves add into a three-dimensional pressure landscape. A grain in it feels a steady cycle-averaged force toward the nearest quiet pocket — the Gor'kov potential — and settles there, held from all sides like a marble in a bowl made of sound.
4. Move. The compiler ramps the phases smoothly; the pockets glide and the grains ride along, hundreds at once — this is where working from the walls pays.
5. Weld. When a grain reaches its final coordinate, energy — acoustic focus plus electromagnetic or optical heating — is concentrated on that one spot and the grain fuses to its neighbours. Heat is spent only here.
6. Check. 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 compound.

Every step composes MEASURED primitives — acoustic levitation, holographic traps, ultrasonic welding (composition status: §8b).

Scan

1. Ask. One gate sends a brief chord — many frequencies at once, at milliwatt level. Nothing in the volume moves.
2. Ring. The waves wash over the object; some frequencies land on its natural resonances and set it ringing — the wine-glass effect, at many pitches at once.
3. Record. The walls record what arrives back at every port: how loud, how delayed, in what phase. One gate's question, twelve ports' answers.
4. Repeat from each gate in turn, or in coded combinations. The collected answers are the object's full response book — the scattering matrix S(f).
5. Compress. The compiler keeps the few resonances carrying most of the signature, each with the port pattern that excites it — the chord list — and writes .pattern; roughly a megabyte describes a mug.
6. Identify materials. The electromagnetic sweep adds what sound cannot: dielectric response for material class, and optical spectroscopy down the bore for composition.

Dissolve

1. Play its echo backwards. Ordinarily an object scatters most of what hits it. Drive from all sides with the time-reversed copy of what it would itself radiate and the scattering runs in reverse: nothing reflects; all the incoming energy lands in the object (coherent perfect absorption — measured physics for beams and cavities).
2. Unbind. The absorbed energy is steered to the welds and bonds: joints soften where assemble fused them, and the object comes back apart into grains.
3. Reclaim. The pockets carry the freed grains back to the cartridge circle; electromagnetic energy returned to the boundary is rectified back into the drive rail rather than lost as heat.
4. Status, honestly. Each individual step is measured physics or ordinary engineering at powder scale; running the full chain — an arbitrary finished object back to sorted stock — is the research frontier OPEN. The deeper matter-loop sectors (transmutation) live entirely in the science paper under its own register.

The unity: assemble and dissolve are the same drive separated by one sign flip, and scan is the same physics at a millionth of the power. (The member papers call the pair ADD and REMOVE — "the two verbs" — a drive-level naming kept here only as a cross-reference.)

6 · Principle: call-and-response

Scan and build are one operation run in opposite directions. Scanning sends chords and records the ring; the collected book is the object's scattering matrix S(f) — its holographic pattern as seen from the gates — and .pattern its compressed, chord-encoded form. Building drives the field and keeps listening: the growing object shifts S(f), and the build is finished when it rings true — the chord list is at once the drive recipe and the acceptance criterion. Listening is milliwatt-class and always on, so every instant of fabrication is also a scan. Throughput is bandwidth: a resonance of quality Q at frequency f takes ~Q/f to ring up, so the machine parallelizes across the band — acoustic chords move matter slowly while the EM octaves listen ~10⁴× faster, verification effectively continuous. Hence: acoustic builds, EM verifies. Full treatment: Field Compiler §2–§4.5.

The Q budget (R6, adopted 2026-07-30). Calibration is a photograph of system state; its lifetime is measured in linewidths of drift. Damping buys calibration lifetime; continuous self-calibration buys back Q. Damp exactly down to the Q the calibration loop can chase, and not one linewidth further. Perfect manufacturing never removes calibration — the environment is state, not error.

7 · Principle: the carriers talk to each other

Sound is a density wave and density sets the refractive index, so the two carriers interact in flight — parametrically, with negligible energy exchange but large information value. Three interactions: sound steers light (the acoustic hologram as a reprogrammable volumetric optic — modulation-class in air, genuine beam shaping in the melt); light reads sound (schlieren/vibrometry — the machine photographs its own acoustic field, an independent witness for the forward model); light writes sound (photoacoustics — a virtual acoustic source at any optically visible point, including on the workpiece: position diversity with nothing moving). All three MEASURED art; machine-level use is design. Full treatment: The Two Carriers §6b — the canonical home of the crossings.

8 · Principle: scale-free by construction

The physics fixes ratios — aperture to volume, element pitch to wavelength, bore to aperture — but privileges no size. A phased boundary is a tiling: add area and channels and the volume grows with them, same compiler, same file format, same three modes. Desktop, workshop cell, industrial volume, orbital boundary — instances of one architecture. The far end of the ladder is bounded, not open: what the machine can never do (make matter from energy, beat Landauer, outrun Q/f) is computed against physical bounds in At the Limit.

Fields, not materials (R7, adopted 2026-07-30). The design variables are ρ(x), stiffness, loss, σ(x), ε(x), and piezo coupling e(x); conductors are skin-depth shells. What survives atom-placement manufacturing: diffraction, medium absorption, the Q budget, addressability rank (channels are information, not fabrication), Bode–Fano/Chu, racemic cancellation. Spec the invariants — angles, λ-ratios, winding numbers, count ratios, grading laws — and render per generation: gen-0 on mills and printers, gen-N as continuous fields. The milled plate is the optimum as pronounced by a Bridgeport; its job is to build the machine that pronounces it better.

8b · Register — what is claimed and what is not

The member papers tag every load-bearing claim MEASURED (published or bench-verified), DERIVED (arithmetic on measured quantities), or OPEN (registered bet with named falsification). This overview inherits those tags and adds none. Two standing hygiene notes. (i) The machine's primitives are measured art; their composition at machine scale is design until the bench says otherwise — build-rate, resolution, and trap-count numbers ship with their measurements, not their spec-table values. (ii) The toroidal-helical pulses' embedded skyrmion texture is a descriptor of the polarization map, not a stability claim: optical skyrmions carry no energy barrier and their invariants are construction-dependent (Chen, Forbes & Qiu, Light Sci. Appl. 2025). The pulses' measured robustness is space-time nonseparability, and pattern persistence in this machine comes from the commanded boundary, not from topology — which is the design's point, not a concession.

9 · The paper network

Maintenance rule: a new replicator paper joins this table and links back here in its banner, or it does not exist. Members own their detail; this paper owns only the principles and the map.