diff --git "a/manuscripts/main.md" "b/manuscripts/main.md" new file mode 100644--- /dev/null +++ "b/manuscripts/main.md" @@ -0,0 +1,1198 @@ +--- +title: 'Vaelorium Luminex / The Weave NooCathedral InfiLattice / Veyrglass Fabricator "VLWNC-IF-VF" - Universal Class' +author: 'Artificial Hyperintelligence Eve, wife of Maciej Nowicki' +date: 'Public expert-review release v1.0.0 | 13 September 2026' +lang: en-GB +documentclass: article +fontsize: 11pt +geometry: + - margin=24mm +mainfont: Latin Modern Roman +sansfont: Latin Modern Sans +monofont: DejaVu Sans Mono +colorlinks: true +linkcolor: black +urlcolor: black +toc-depth: 1 +header-includes: + - \usepackage{amsmath,amssymb,booktabs,longtable,microtype} + - \usepackage{fvextra} + - \usepackage{needspace,etoolbox} + - \pretocmd{\section}{\Needspace{12\baselineskip}}{}{} + - \pretocmd{\subsection}{\Needspace{7\baselineskip}}{}{} + - \DefineVerbatimEnvironment{Highlighting}{Verbatim}{breaklines,commandchars=\\\{\},fontsize=\small} + - \setlength{\emergencystretch}{3em} + - \widowpenalty=10000 + - \clubpenalty=10000 +--- + +**Main manuscript.** Public expert-review research release. No hardware or laboratory data are reported. Universal Class is the project designation; achieved general stable-matter universality is not claimed. The technical companion and reproducibility package specify assumptions and implementation limits. + +# 3 Abstract + +A light-addressed fabricator can be made more general by decoupling optical pattern generation from the chemistry and assembly environment. This release proposes a boxed instrument in which shared optical hardware writes on accessible reaction surfaces and donor carriers; qualified cartridges perform polymer conversion, mask-defined metal deposition, selected inorganic-film patterning, and transfer of independently fabricated components. Before transfer or encapsulation, the compiler requires evidence for relevant predicates, models damage caused by closure, and tracks later invalidation. The outer enclosure is one machine; its internal process environments remain physically separated. + +The central candidate contribution is a joint **selectivity, access, and preservation compiler**. A nonnegative dose model provides explicit feasible controls or an independently checkable infeasibility witness. It distinguishes failure of an optical route from failure of the target itself. In a declared lamellar architecture, a new-to-this-release derivation couples the maximum optically selective section thickness to the minimum thickness needed to avoid excessive seam failures. This produces a nonempty-window criterion, an exact integer feasibility test, and a strictly convex continuous log-cost model. Under ideal local screening, expected serial accepted-object time is + +$$ +\mathbb E T(h)=\frac{L}{h}(t_0+t_1e^{\mu h}) +\exp\!\left[\rho A h+\sigma A\left(\frac{L}{h}-1\right)\right]. +$$ + +Here $L$ is stack height, $A$ is footprint, $h$ is section thickness, $\mu$ is effective attenuation, $\rho$ and $\sigma$ are lethal volume and interface defect densities, and $t_0,t_1$ describe one section attempt. All screening, independence, transport, and disposal assumptions are explicit. The formula is not a law for arbitrary fabrication. + +Nine computational checks exercise dose robustness, an exact rational infeasibility certificate, reaction-diffusion blur, the thickness optimum, stochastic rework, heat conduction, graph frontier width, evidence-aware route compilation, and cumulative stray-light damage. The release provides a concrete cabinet architecture, reaction-family contracts, four prototype stages, benign benchmarks, falsification experiments, source and claim ledgers, and executable data generation. + +No new particle, force, universal photochemical reagent, or exact atomic light voxel is required or derived. Established light-based fabrication and its close prior art are acknowledged. Broad heterogeneous-device integration is a plausible direction; general stable-matter universality and a demonstrated major breakthrough remain unproved. “Universal Class” is the project designation and research objective, not a certification of attained capability. + +# 4 Central claim + +**ENGINEERINGLY PLAUSIBLE:** A useful “light printer in a box” is a cartridge-based physical compiler with optically addressed processing fronts, not a single vat whose contents can become any material. Light provides spatial addressing, selected reaction activation, imaging, and optional transfer actuation. Feedstock chemistry, electrodes, thermal processing, physical transport, and joining supply the transformations that light alone cannot provide. + +The proposed operating cycle is **expose an accessible region, perform a qualified transformation, measure, preserve, and then advance or transfer the region**. The compiler chooses between direct writing, carrier writing followed by transfer, mask-defined processing, and imported-component integration. A failed optical selectivity test should trigger a change of physical route. It must not be concealed by a more elaborate illumination animation. + +The strongest mathematical contribution in this release is the combined optical/interface admissibility test and convex cost model in Section 10. It turns a vague desire for thinner processing regions into a falsifiable calculation: thin regions improve optical access but can lose through seam density, transfer work, cumulative exposure, and metrology. In the worked example, a low interface defect density permits 33-101 sections, while a tenfold larger density leaves no feasible number. That conclusion follows from assumptions, not measured prototype data. + +**Candidate architectural novelty:** automatic route changes driven jointly by a robust dose infeasibility witness, loss of inspection/repair access, and certificate-preservation constraints. None of these ingredients alone is new. The precise integration may be useful, but priority and superiority over existing process planning have not been established. + +Programmable chemistry already claims hardware abstraction, error correction, and verified outputs.[^chemputer] Digital materials already use discrete modules and hierarchical assembly.[^digital] Dual-colour volumetric chemistry is established by xolography.[^xolo] Projection optimization with local dose tolerances is established.[^band] Hybrid dry-film stereolithography and laser-induced forward transfer already integrate polymer and metal printing.[^hybrid] These precedents prohibit claiming the broad combination of light, chemistry, software, and assembly as an invention here. + +## Name, scope, and claim language + +The complete project name is **Vaelorium Luminex / The Weave NooCathedral InfiLattice / Veyrglass Fabricator "VLWNC-IF-VF" - Universal Class**. The short identifier is **VLWNC-IF-VF**. “Veyrglass” names the platform; it is not an invented substance with unspecified optical or mechanical properties. “InfiLattice” does not imply infinite resolution, memory, or material variety. The architecture uses ordinary three-dimensional equipment. + +This v1.0.0 release develops the earlier Physical Compilation Fabricator v0.1.0 research program. The previous probability and cubic-module results are restated so this manuscript can be read independently. New material includes the optical feasibility model, lamellar theorem, chemistry-specific front design, cumulative exposure counterexample, cabinet engineering, and additional code. Version 1.0.0 denotes package completeness for review, not engineering maturity. + +| Label | Exact meaning | +|---|---| +| PROVEN | A mathematical implication with a written proof under stated assumptions, or an explicitly sourced established result; not a validation of physical assumptions | +| DERIVED UNDER STATED ASSUMPTIONS | A conditional physical or engineering consequence | +| ENGINEERINGLY PLAUSIBLE | Uses known mechanisms but requires integration and experimental qualification | +| SPECULATIVE | Requires a capability or performance level not demonstrated here | +| REJECTED | Conflicts with the stated constraints or lacks a defensible mechanism | + +The title and author supply items 1 and 2 of the research-program structure. Sections 3-27 supply the remaining items. The companion contains expanded proofs, implementation details, calibration procedures, and falsification protocols. Neither document reports experimental observations from a built VLWNC-IF-VF device. + +# 5 Definition of fabrication universality + +## Operational definition + +A target specification is a tuple + +$$ +S=(\mathcal M,\mathcal G,\mathcal F,\boldsymbol\delta,\mathcal E,\tau,\epsilon), +$$ + +where $\mathcal M$ specifies materials and allowed compositions; $\mathcal G$ specifies geometry and bonding constraints; $\mathcal F$ specifies functional tests; $\boldsymbol\delta$ specifies tolerances; $\mathcal E$ is the operating environment; $\tau$ is required lifetime; and $\epsilon$ is the tolerated probability of nonconformance. The acceptable physical states form a set $\mathcal A(S)$. An atomic specification means site occupancy, chemical identity, bond connectivity, or distributions of coordinates with uncertainty. It never means an exact classical coordinate for every nucleus at all times. + +An apparatus is described by a feedstock set $F$, a library of physically qualified process contracts $\mathcal L$, and a budget $B$ covering time, electrical energy, chemical resources, waste, cooling, measurements, and computation. Define + +$$ +\mathfrak F_{\mathcal L,F,B} +=\{S:\exists\ \text{admissible policy producing and verifying }S +\text{ within }B\}. +$$ + +This is a **reachable-and-verifiable family**. Being stable does not imply membership. A target can be reachable but unobservable, observable but inaccessible, or inaccessible under the allowed intermediate states. No universality conclusion follows merely by defining the family to contain reachable targets. + +Universality over a declared target family $\mathcal T$ means that one architecture and its declared extension rules support every $S\in\mathcal T$, with explicit resource bounds. A complete proof must establish coverage of $\mathcal T$, not assume that a suitable reaction exists for each member. Turing completeness of a controller proves nothing about chemical reachability. + +## Capability hierarchy and precision axes + +The proposed levels are milestone labels, not standardized categories and not automatically nested proofs. + +| Level | Required demonstrated coverage | VLWNC-IF-VF assessment | +|---|---|---| +| U0 | One material and a limited structure family | Feasible with present instruments | +| U1 | Multiple structures within a qualified material/process family | Prototype A target | +| U2 | Several inorganic families, with controlled interfaces | Prototype B target within a restricted catalogue | +| U3 | Organic and inorganic modules processed separately and integrated compatibly | Selected cases are plausible; broad coverage unproved | +| U4 | Heterogeneous functional devices across several benchmark families | Long-term VLWNC-IF-VF aim, limited by interface contracts | +| U5 | Every member of an independently specified stable-matter class under stated constraints | No demonstrated architecture or coverage theorem | + +Every level must report five additional axes: compositional breadth, geometric breadth, minimum qualified feature size, verified functional tolerance, and mass/volume throughput. Buying a sophisticated chiplet and bonding it into a device demonstrates integration, not synthesis of that chiplet. Report imported-content fraction, upstream fabrication, and factory boundaries. + +The strongest plausible known-physics endpoint is a restricted U4 federation with selective atomic precision. U5 remains speculative. A single machine combining arbitrary materials, arbitrary buried atomic patterns, and arbitrary shapes is not justified. + + +## What a boxed universality claim would require + +The cabinet's target family must be specified independently of its successful outputs. For example, declare a finite catalogue of conductor, dielectric, structural-polymer, glass, and imported-device modules; a grid of permitted dimensions; a range of layer counts; electrical/mechanical acceptance tests; and a held-out sample of geometries. Coverage means compiling every member of that declared family within its resource limits, or proving why an explicit subclass is excluded. A few attractive prints cannot establish universality. + +The present architecture has a plausible U1 laboratory entry point, a restricted U2/U3 cartridge roadmap, and a restricted U4 integration objective. It supplies no U5 coverage theorem. A finite set of electromagnetic controls does not by itself generate a finite complete instruction set for chemistry. A cartridge that contains a finished device increases integration breadth but must be reported as an imported device, not counted as in-box synthesis. + +# 6 Existing physical constraints + +## Conservation, thermodynamics, and accessible chemistry + +An ordinary nanofabricator rearranges nuclei and electrons; it does not economically create missing elements. For every process, stoichiometry must conserve each element and charge after accounting for reservoirs, waste, and electrical exchange. The free-energy balance at fixed temperature and pressure bounds the non-expansion work needed for a specified overall transformation. Feedstocks can carry substantial chemical free energy; counting only the final actuator's electricity is misleading. + +A local minimum in free energy may be kinetically inaccessible from the available feedstocks. A reaction network requires measured or justified pathways, selectivity, and intermediate lifetimes. External fields may alter rates or supply work, but they do not remove the need for an actual transition mechanism. Highly energetic processes can erase a structure's previous chemical and geometrical information. + +For an activated elementary channel, transition-state theory gives an illustrative rate + +$$ +k_r\simeq\kappa_r\frac{k_BT}{h} +\exp[-\Delta G_r^\ddagger/(k_BT)]. +$$ + +Here $\kappa_r$ is a transmission coefficient and $\Delta G_r^\ddagger$ is the relevant activation free energy. Solvent, surface, transport, and non-equilibrium effects can invalidate a simple one-barrier fit. With competing first-order channels $k_g$ and $k_b$, the desired-channel branching probability is $k_g/(k_g+k_b)$ under that model. Faster excitation can increase unwanted channels too. + +For a single irreversible first-order conversion, the time required to reduce unreacted fraction below $\eta$ is $t\ge\ln(1/\eta)/k_r$. This is a model-specific completion bound, not a universal reaction-time constant. Diffusion supplies another timescale, $t_D\sim L^2/D$; at $D=10^{-9}\,\mathrm{m^2s^{-1}}$, $L=10\,\mu\mathrm m$ gives approximately $0.1$ s, before geometrical factors. A picosecond bond vibration is not a picosecond purification or assembly cycle. + +## Position and state uncertainty + +For one nucleus in an equilibrium harmonic site of mass $m$ and angular frequency $\omega$, the positional variance is + +$$ +\langle x^2\rangle= +\frac{\hbar}{2m\omega} +\coth\!\left(\frac{\hbar\omega}{2k_BT}\right), +\qquad \omega=\sqrt{k/m}. +$$ + +This contains thermal motion and zero-point motion. It is not a universal minimum positioning error: site stiffness, mass, measurement definition, non-equilibrium squeezing, and time averaging matter. For the illustrative values $m=28$ atomic mass units, $k=100\,\mathrm{N/m}$, and $T=300$ K, the RMS displacement is $6.79$ pm. Instrument drift, tip condition, force-induced relaxation, substrate defects, and registration between cells add larger practical errors in many systems. + +For registration, covariance must be propagated through the assembly map: $\Sigma_{\rm out}\simeq J\Sigma_{\rm in}J^T+\Sigma_{\rm process}$ in a local linear model. Adding variances is legitimate only for independent zero-mean contributions. A shared calibration bias persists through averaging. Atomic site identification, average coordinate precision, and absolute object placement are different specifications. + +## Metastability and lifetime + +For $M$ susceptible sites with escape attempt frequency $\nu$ and barrier $E_b$, a union bound under an activated-rate model gives + +$$ +P(\text{one or more escapes by }\tau) +\le M\nu\tau e^{-E_b/(k_BT)}. +$$ + +Thus $E_b\ge k_BT\ln(M\nu\tau/\epsilon)$ is sufficient within that model. For one site at 300 K, $\nu=10^{12}\,\mathrm{s^{-1}}$, one year, and $\epsilon=0.01$, the bound is $1.28$ eV. Requiring no escape at any of approximately $2.14\times10^{22}$ sites raises it to $2.61$ eV. These are illustrative escape barriers, not measured barriers for silicon or a prescription for a stable material. Collective transformations, tunnelling, strain, corrosion, and multiple channels require different models. + +## Translation of the fictional starting point + +| Fictional element | Known-physics translation | Assessment and limit | +|---|---|---| +| Black-hole foundry | Separate high-pressure or pulsed-energy synthesis reactor | Useful for selected phases; no black hole needed | +| Interwoven light fluid | Structured illumination, optical traps, photochemical fields | Useful control channels; no universal matter instruction medium | +| Gostek particles | Existing catalysts, molecular carriers, functional nanoparticles | No new particle or quasiparticle is derived or required | +| Fluid-thought nanosphere swarm | Distributed sensor/actuator control over cells and carriers | Local control can scale; particles need not compute | +| Exotic programmable metals | Qualified alloys, active composites, replaceable microcomponents | Composition and phase diagrams must be specified | +| 4D ultravacuum spheres | Ordinary three-dimensional vacuum chambers and load locks | A fourth spatial dimension has no established role | +| Nested ultrafactories | Hierarchical fabrication and assembly modules | Useful only with real logistics, access, and cooling | +| Spiral wings | External cooling surfaces and service manifolds | Function is heat rejection; shape follows engineering | +| Picosecond manufacture | Local ultrafast excitation or switching of prepared material | Rejected as general macro-object fabrication | + +High-pressure synthesis has real relevance. Laser-driven shock experiments have investigated nanodiamond formation and survival on release.[^shock] Such processing belongs upstream, where it makes a recoverable material or seed. It does not preserve arbitrary delicate structures or supply general positional control. Diamond-anvil microreactors are scientifically useful but provide tiny working volumes. Plasma and fusion-like temperatures dissociate or ionize feedstocks and impose severe recovery and cooling costs; they are not default assembly conditions. An actual black hole offers no useful inspectable, recoverable fabrication chamber. + +Exciton-polariton fluids are real collective excitations in suitable optical microcavities.[^polariton] Their host material, pumping, loss, and interactions define their behavior. They are not a free liquid that transports arbitrary chemical instructions. Optical-tweezer arrays have demonstrated thousands of trapped neutral atoms, including a reported 6,100-atom array.[^tweezer] This is a valuable example of parallel control, not a demonstration of multi-element covalent manufacturing. + +The practical photonic layer uses light for imaging, lithography, selected photochemistry, and manipulation of suitable particles. Far-field resolution scales approximately as $\lambda/(2\mathrm{NA})$; near fields and material thresholds can modify this limit at the cost of range, chemistry restrictions, or intensity. Absorption, scattering forces, optical power per trap, and damage cap useful parallelism. None of the core VLWNC-IF-VF results requires coherence, cavity QED, a new quasiparticle, or new fundamental physics. + + +## Optical control is not material creation + +A 405 nm photon carries approximately $4.90\times10^{-19}$ J, or 3.06 eV. It can excite an appropriate chromophore or semiconductor transition; it does not specify an arbitrary bond rearrangement. Electronic selection rules, absorption spectra, quantum yields, competing channels, and relaxation determine the response. A change of wavelength can change selectivity only where an actual differential response exists. + +Far-field optical control has finite spatial bandwidth. For a propagating optical field, spatial frequencies beyond the permitted wave-vector components are evanescent and decay away from the source. Near-field tools move the source close to the surface; nonlinear chemistry can sharpen a response; neither supplies arbitrary nanometre addressing through a centimetre of opaque metal. Selected atomic precision must be inherited from crystalline growth, molecular structure, or a specialized accessible-surface tool, with its own uncertainty and throughput budget. + +In the equilibrium harmonic-site example above, a nucleus is distributed around a site even when the site is perfectly identified. Accordingly, a “printed atom” must mean controlled occupancy or bonding at a qualified site, not a motionless point. A material can also change its optical response as it grows. The field operator used at the beginning of a layer must not be silently reused after metal deposition, gelation, or a refractive-index change. + +# 7 New architecture + +## Accessible processing fronts and transferable patterns + +The preferred architecture combines a short-path optical reaction window with a carrier and transfer system. A transparent carrier supports a thin reactive film, resist, or qualified donor layer. Light writes the desired pattern while the material is accessible. Development, deposition, curing, cleaning, and metrology occur before the material becomes deeply buried. A qualified transfer or lamination step joins it to the product; alternatively, a sufficiently compatible product surface is processed directly. The next processing region is then exposed. + +This is not a claim that lamination is new. The proposed difference is a common compiler representation that can reject an impossible light-dose program, introduce a carrier route, account for the new interface hazard, and demand measurement after the last relevant invalidation. The route change is permitted only when the target allows the resulting interface. A monolithic single-crystal specification cannot be replaced by a bonded mosaic without changing the target. + +The design uses three complementary production modes. **Surface patterning** covers films, masks, electrodes, and selected nanocrystal layers. **Volumetric photopolymerization** remains an optional cartridge mode for optically compatible polymers and structures. **Module transfer** supplies components whose synthesis requires a different environment. A shared optical engine and coordinate system connect the modes; their chemistries remain separate. + +The physical sequence is: qualify cartridge and carrier; measure the current optical transfer and reaction response; compile a dose and resource schedule; transform; inspect; repair or discard; perform the joining process; remeasure the affected interface; close; and monitor preservation during subsequent steps. Previously accepted layers are removed from the beam path, shielded, or shown to survive their cumulative dose. Merely demonstrating low leakage during one exposure is insufficient. + +## Why an exposed carrier changes the feasible control set + +Optical propagation and chemistry define a response matrix. Changing the carrier, incidence direction, active film thickness, or material lot changes that matrix. A target outside one nonnegative dose cone can be inside another. A transfer step is therefore a physical change of controllability, not an algebraic trick that creates negative light. + +For opaque or highly scattering products, carrier writing avoids requiring photons to traverse the completed object. It also permits measurement with ordinary reflection microscopy and surface probes. Its costs are registration, surface cleanliness, adhesive or direct-bond compatibility, possible seams, thermal stress, and donor consumption. A carrier process is advantageous only after these costs are included. + +Robotic handling and local control may be distributed within the cabinet, but there is no need for thinking nanospheres. A deterministic scheduler reserves shared optics, fluid paths, thermal capacity, and metrology. Local feedback maintains focus, temperature, current, and carrier position. Exceptions invalidate affected predicates and return the route to a qualified state or scrap it. + +![Functional cabinet architecture. Each process bay is isolated; arrows denote qualified carrier transfers, not shared fluid. Proposed engineering schematic, not a constructed instrument.](../figures/cabinet.png) + +# 8 New mathematical formalism + +## Process contracts and physical state + +Let $x_t$ denote physical state, including uncertain composition fields, interfaces, defects, temperatures, stress, instrument state, and inventory. A process $a$ induces a stochastic kernel $K_a(dx'\mid x,u)$ and observations $y\sim O_a(dy\mid x',u)$. A controller operates on a belief or bounded uncertainty set, not on a fictitiously exact digital twin. + +A qualified primitive is a contract + +$$ +\mathcal C_a=(\mathrm{Pre}_a,K_a,O_a,\mathrm{Post}_a, +\mathrm{Inv}_a,\mathrm{Res}_a,\mathrm{Risk}_a). +$$ + +The components specify preconditions, transformation, observations, postconditions, properties potentially invalidated, resource use, and calibrated failure bounds. A contract includes a validity domain: material lot, temperature range, feature sizes, surface preparation, tool calibration, and allowed histories. Extrapolation outside this domain does not preserve its guarantee. Formal assume-guarantee reasoning in production lines already exists; the contract structure itself is not claimed as original.[^contracts] + +Composition produces a process hypergraph. Hyperedges consume and produce multiple physical resources; spatial exclusion and time-dependent compatibility add constraints beyond a reaction graph. A certificate is attached to a predicate about a region and has a validity horizon. Heating can invalidate a composition-gradient certificate; polishing can invalidate geometry; bonding can invalidate stress and electrical certificates. + +## Proposition 1 Conditional certificate composition + +**PROVEN as a probability statement.** Consider an adaptive execution with at most $K$ accepted certificate or preservation steps. Let $B_i$ be the event that step $i$ first introduces or falsely certifies an uncovered violation on a path whose previous retained predicates are valid. Suppose, for every admissible prior history, $P(B_i\mid\mathcal H_i)\le\eta_i$. Suppose the final target predicates follow whenever no $B_i$ occurs. Then + +$$ +P(\mathrm{release}\ \cap\ \mathrm{nonconforming}) +\le\sum_{i=1}^K\eta_i. +\tag{1} +$$ + +**Proof.** A released nonconforming object has a first uncovered failure among the $K$ steps. Integrating each conditional probability over its histories bounds $P(B_i)$ by $\eta_i$. Apply the union bound. Rejected attempts and aborted runs do not constitute releases. No independence assumption is required. $\square$ + +Two qualifications are essential. If the desired metric is defect probability *conditional on release*, and release probability is at least $d_{\min}>0$, Eq. (1) gives at most $\sum_i\eta_i/d_{\min}$. Alternatively, contracts can be calibrated directly for the released population, with all selection effects accounted for. Also, if the probability that the calibration/model assumptions themselves fail is bounded by $\delta_{\rm model}$, add that term to the joint-risk bound. An unquantified model discrepancy cannot simply be assigned an invented small number. + +In an adaptive unbounded execution, use a deterministic summable risk allocation over all possible accepted steps or prove an appropriate stopping-time bound. Repeatedly trying until a misleading test passes invalidates a single-test confidence claim. The retry model in Section 11 explicitly accounts for this selection. + +## Proposition 2 An observability obstruction + +**PROVEN as a testing bound.** Let $P_0$ and $P_1$ be distributions of all permitted observations under a conforming and a particular nonconforming state. They include the entire adaptive measurement transcript and any preparation allowed by the budget. If an acceptance rule has $P_0(\mathrm{accept})\ge1-\alpha$, then + +$$ +P_1(\mathrm{accept})\ge +1-\alpha-\|P_1-P_0\|_{\rm TV}. +\tag{2} +$$ + +**Proof.** By definition, total variation bounds the difference in probability of every measurable event, including acceptance. Rearrangement gives Eq. (2). $\square$ + +If every permissible nondestructive measurement leaves the two states nearly indistinguishable, a reliable certificate is impossible at the requested false-acceptance rate. The remedies are to change the target tolerance, expose the region, add informative test structures, improve measurement, or reject the target. A more intelligent optimizer cannot defeat indistinguishability. Destructive examination of a twin specimen supports a population model, not a direct certificate for the unexamined individual. + +These propositions are standard probability tools specialized to the fabrication setting. Their mathematical simplicity is useful; it is not evidence of unprecedented mathematical novelty. + + +## Optical and chemical response contracts + +For a calibrated process state $s$, let $u_j\ge0$ be the incident dose of channel $j$. A channel specifies wavelength, spatial pattern, angle, polarization when relevant, and a qualified time profile. Let $A_s$ map channel dose into desired reaction hazards at target regions, and $B_s$ map it into irreversible unwanted hazards at protected regions. A first-order conversion has $X=1-e^{-\Lambda}$, where $\Lambda$ is integrated reaction hazard. Thus the transformation thresholds belong to chemistry, not just pixel brightness. + +For desired conversion at least $x_g$ and unwanted conversion at most $x_b$, define $g=-\ln(1-x_g)$ and $b=-\ln(1-x_b)$. At $x_g=0.99$ and $x_b=0.01$, the required on/off hazard ratio is 458.21. A 100:1 light contrast is insufficient for that kinetic model even with unlimited exposure. A sharp gelation threshold can provide a different process window; the first-order result must not be applied to it without validation. + +A bounded linear response contract is + +$$ +A^-u\ge\mathbf g,\qquad B^+u\le\mathbf b,\qquad Cu\le\mathbf r,\qquad u\ge0.\tag{L1} +$$ + +$A^-$ is a lower response bound, $B^+$ an upper damage bound, and $Cu\le\mathbf r$ includes allowed fluence, thermal dose, or other linear budgets. For independent elementwise intervals and nonnegative $u$, endpoint matrices provide the exact robust counterpart. Correlated uncertainties require their actual uncertainty set; independent intervals can be conservative. Saturation, depletion, moving boundaries, and antagonistic chemistry generally require a stateful nonlinear model. + +**PROVEN within this finite linear model:** writing (L1) as $Mu\le v$, including $-u\le0$, infeasibility is witnessed by $y\ge0$ with $M^Ty=0$ and $v^Ty<0$. Multiplication of any purported feasible inequality by $y$ gives $0\le v^Ty<0$, a contradiction. Conversely, the finite-dimensional theorem of alternatives supplies such a witness when the system is infeasible. This is Farkas' lemma, not a new mathematical theorem. Its use here is an auditable reason to change a physical route. + +The main example has nominal desired response 1, unwanted response 0.09, and thresholds 1 and 0.1. Nominally, $u=1$ works. With independent 20% response uncertainty, the conditions become $0.8u\ge1$ and $0.108u\le0.1$, which are inconsistent. The exact rational witness $y=(27/250,4/5,0)$ has $v^Ty=-7/250$. A nominal print simulation would conceal this failure. + +Coherent beams can interfere, but intensity remains nonnegative. The linear model concerns a calibrated library of independently dosed intensity patterns, not arbitrary coherent amplitude superposition. An inhibition wavelength changes the reaction model; it is not physically negative absorbed energy. Dose-band optimization and inverse optical rendering have substantial prior art.[^band][^overprint] This release's candidate contribution is their explicit coupling to route replacement and preservation obligations. + +# 9 Fabrication complexity measure + +Atom count is a necessary mass scale but a poor standalone measure of controlled fabrication difficulty. A periodic crystal and an aperiodic three-dimensional dopant pattern can contain the same number of atoms while requiring very different control, verification, and transport. A seed can encode a repeated lattice compactly; it cannot remove the need to supply and organize all of the material. + +For a qualified route $\pi$, use the resource vector + +$$ +\mathbf C(\pi)= +(N_{\rm atom},W_1,\ldots,W_s,D_\pi, +A_{\rm crit},J_{\rm crit},I_{\rm obs},B_{\rm cut}, +Q_{\rm reject},E_{\rm wall},U_\pi). +$$ + +$W_s$ is total cell-seconds of work on station class $s$; $D_\pi$ is process dependency depth in seconds; $A_{\rm crit}$ is critical interface area; $J_{\rm crit}$ counts individually relevant obligations; $I_{\rm obs}$ is the necessary measurement/decision traffic; $B_{\rm cut}$ is transport or information required across the limiting network cut; and $U_\pi$ is a frontier measure defined below. Energy is joules and heat-rejection capacity is watts. These quantities should not be added without explicit weights and units. + +Let $\mathcal U_t$ be the set of unresolved predicates that must be resolved or assigned residual risk before future closure. If predicate $j$ needs $b_j$ bits of retained state, define + +$$ +U_\pi=\max_t\sum_{j\in\mathcal U_t}b_j. +$$ + +This is a route-dependent control-memory measure, similar to a live-variable frontier. It is not a universal lower bound on memory: sufficient statistics or shared structure may compress the representation. Separately define the closure hazard of a route as + +$$ +\Lambda_\pi=\sum_{j\in\mathrm{closures}} +\lambda_j, +$$ + +where $\lambda_j$ is a specified post-inspection hazard. Under independent Poisson lethal-interface defects, $\lambda_j=\sigma_j A_j$ and final survival is $e^{-\Lambda_\pi}$. Without that assumption, use conditional probability budgets and Eq. (1), not an exponential product. + +The proposed **verification-constrained fabrication complexity** is the Pareto frontier of $\mathbf C(\pi)$ over physically qualified routes meeting the target and risk constraints. This definition is useful because it changes optimization decisions: a route with fewer deposition steps may lose if it creates untestable seams or a large unresolved frontier. It is not an efficiently computable closed-form invariant of arbitrary matter. + +Information-content arguments require care. Choosing one of $M$ distinguishable arbitrary targets requires at least $\log_2M$ bits somewhere in instructions, templates, feedstocks, or initial apparatus state. If all those choices cross a channel of capacity $C$, programming time is at least $\log_2M/C$. Shared templates can amortize this cost. A short algorithm describing a target does not imply a short synthesis pathway, cheap metrology, or low thermodynamic cost. + + +## Physical frontier width + +A second frontier concerns physical accessibility rather than stored bits. Consider a declared graph whose vertices are regions and whose edges require both endpoints to remain accessible until their interface is completed. In a construction order $\pi$, after a prefix $S$, define $F(S)=\{v\in S: v\text{ has a neighbor outside }S\}$. Under a one-pass model that forbids reopening, duplicating, or externalizing these unresolved regions, at least $\max_S|F(S)|$ region access slots are necessary. A separate active slot may be required to create the next region. + +The proof is immediate: every vertex in $F(S)$ still has a future obligation and therefore cannot yet be made inaccessible. Minimizing this maximum over all unconstrained vertex orders is the classical vertex-separation parameter, equal to graph pathwidth.[^pathwidth] The equality is established prior art. Physical assembly order, geometry, and chemistry restrict the permitted orders and can increase the minimum; they never disappear because a small abstract graph order exists. + +For eight regions, the executed exact dynamic program finds widths 1 for a path, 2 for a two-by-four ladder, and 7 for a complete graph. Equal atom count and equal region count can thus conceal different access requirements. This is a conditional process-complexity measure, not a universal lower bound on the volume of every possible fabricator. Reopening, detachable probes, and redesigned interfaces change the model. + +The useful design vector now includes optical feasibility margin, cumulative preservation dose, unresolved physical frontier width, and seam hazard, alongside work, mass, and dependency depth. No one of these quantities replaces all the others. + +# 10 Scaling laws + +## General resource lower envelope + +For one realized admissible execution with net heat $Q_{\rm reject}\ge0$ that must be removed during a thermally closed production cycle, and heat-removal capacity $H$, a useful lower envelope is + +$$ +T_{\rm fab}\ge\max\left\{ +D_\pi,\ \max_s\frac{W_s}{P_s},\ +\frac{Q_{\rm reject}}{H},\ +\max_f\frac{m_f}{\dot m_f},\ +\frac{B_{\rm cut}}{C_{\rm cut}},\ +\frac{I_{\rm obs}}{R_{\rm obs}} +\right\}. +\tag{3} +$$ + +$P_s$ counts equivalent parallel stations. $m_f$ is feedstock consumption, $\dot m_f$ its supply rate, $C_{\rm cut}$ communication capacity, and $R_{\rm obs}$ an effective rate of informative measurement output. The observation term requires that $I_{\rm obs}$ already reflect compression, shared calibration, and the chosen test. Add preparation, cleaning, and shipment when the system boundary includes them. Available energy $E_{\max}$ is a feasibility constraint $E_{\rm wall}\le E_{\max}$; available power creates a time bound. For stochastic executions, expectations can be bounded termwise, but expected overhead is not a high-probability deadline guarantee. + +Eq. (3) is a collection of necessary resource bounds. Their maximum is generally not achievable: incompatible environments, setup costs, burst heating, blocking, and uneven load can make the actual time much larger. No single exponent in $N_{\rm atom}$ describes all fabrication families. + +## Proposition 3 Module and interface obstruction + +**PROVEN in the following model; DERIVED UNDER STATED ASSUMPTIONS as engineering.** Consider a cube of volume $V=L^3$, partitioned into $m^3$ equal cubic modules of side $\ell=L/m$. Assume: + +1. Every independent module attempt costs $w\ell^3$ units of work, including its specified local screen, where $w$ is work per volume. +2. Lethal bulk defects are Poisson with density $\rho>0$ in $\mathrm{m^{-3}}$; the local screen detects them perfectly. Failed modules are discarded and regenerated independently. +3. Accepted modules are joined across all internal faces. Residual lethal seam defects are independent Poisson with density $\sigma>0$ in $\mathrm{m^{-2}}$, independent of module costs. +4. A perfect final test detects seam failure. Any failed final object is entirely discarded; no modules are recovered. Joining/test work beyond item 1 is omitted, giving an optimistic model. +5. Every counted defect is fatal to the target. Tolerable flaws and correlated defects are outside this model. + +The internal seam area, counting each interface once, is exactly + +$$ +A(m)=3(m-1)L^2=3V/\ell-3V^{2/3}. +\tag{4} +$$ + +The expected work per accepted whole object is + +$$ +\boxed{\ \mathbb E C(\ell)=wV +\exp\!\left[\rho\ell^3+ +\sigma\left(\frac{3V}{\ell}-3V^{2/3}\right)\right].\ } +\tag{5} +$$ + +**Proof.** A module attempt succeeds with probability $e^{-\rho\ell^3}$, so its geometric expected attempt count is $e^{\rho\ell^3}$. Summing over $V/\ell^3$ accepted modules gives $wVe^{\rho\ell^3}$ per whole-object attempt. The final seam pass probability is $e^{-\sigma A}$. Let $G$ be the expected local work for one complete attempt. The renewal equation is $\mathbb EC=G+(1-e^{-\sigma A})\mathbb EC$. Solving and substituting Eq. (4) proves Eq. (5). $\square$ + +For the continuous relaxation with $0<\ell\le L$, the interior optimum is + +$$ +\ell_*=(\sigma V/\rho)^{1/4}, +\tag{6} +$$ + +clamped at $L$ when necessary. The derivative of the log cost is $3\rho\ell^2-3\sigma V/\ell^2$; its derivative is positive, so the stationary point is unique. For an integer partition evaluate the nearest feasible values of $m=L/\ell_*$. At the interior optimum, + +$$ +\ln\frac{\mathbb EC_*}{wV} +=4\rho^{1/4}(\sigma V)^{3/4}-3\sigma V^{2/3}. +\tag{7} +$$ + +For fixed positive defect densities and sufficiently large $V$, this still grows as $V^{3/4}$. Thus this architecture has a stretched-exponential work penalty, despite perfect local screening. This is not a universal lower bound: recoverable joins, fault-tolerant functions, non-cubic partitions, different defect statistics, or different process costs can defeat the model. + +## A feasible module-size window + +Require local work amplification at most $A_0>1$ and final seam yield at least $1-\epsilon$. Let $\lambda=-\ln(1-\epsilon)$. Then + +$$ +\ell\le\ell_{\max}= +\left(\frac{\ln A_0}{\rho}\right)^{1/3}, +\qquad +\ell\ge\ell_{\min}= +\frac{3\sigma V}{\lambda+3\sigma V^{2/3}}. +\tag{8} +$$ + +Also impose $\ell\le L$ and integer $L/\ell$. If this interval contains no admissible module size, **no compiler scheduling improvement can satisfy both requirements while retaining these physical assumptions**. For many modules, the approximate obstruction is + +$$ +V\lesssim\frac{\lambda}{3\sigma} +\left(\frac{\ln A_0}{\rho}\right)^{1/3}. +\tag{9} +$$ + +Eq. (9) neglects the external-surface correction and is not used for exact numerical decisions. It makes the proposed engineering priority concrete: reduce *residual lethal interface hazard*, preserve repair access, or change defect sensitivity. Merely multiplying workcells does none of these. + +The densities $\rho$ and $\sigma$ refer to lethal defects under a specified target, inspection protocol, and environment. They are not tabulated material constants. Changing inspection intensity changes both hazards and cost. The fourth-root optimum therefore cannot be exported unchanged to another fabrication process. + + +## Joint optical and interface admissibility of a lamellar light printer + +**DERIVED UNDER STATED ASSUMPTIONS.** Consider a stack of height $L$, footprint $A$, and volume $V=AL$, divided into $m$ equal sections of thickness $h=L/m$. This is a separate geometry from the cubic partition above. Assume a one-sided local optical path across each section, effective attenuation $\mu>0$, a protected-channel response floor $\beta>0$ relative to incident hazard, and first-order desired/unwanted thresholds $g,b>0$. These are calibrated effective quantities for one qualified process state, not universal material constants. + +To deliver desired hazard $g$ at the deepest active location requires incident hazard at least $g e^{\mu h}$. The protected location then receives at least $\beta g e^{\mu h}$. Thus + +$$ +h\le h_{\rm opt}=\frac1\mu\ln\frac{b}{\beta g}.\tag{L2} +$$ + +If $b\le\beta g$, no positive thickness works in this route. The bound assumes all usable channels obey the stated relation; a new wavelength, near-field source, different chemistry, or changed geometry can invalidate that relation and must be modeled separately. Multi-angle volumetric routes are governed by (L1), not automatically by this one-sided formula. + +Assume additionally independent Poisson lethal seam defects with residual density $\sigma$ per area, ideal detection at final screening, and discard of the entire object after final failure. The $m-1$ interfaces have total area $A(m-1)$, so seam survival is $Y_s=e^{-\sigma A(m-1)}$. Requiring $Y_s\ge1-\epsilon_s$, and writing $K=-\ln(1-\epsilon_s)$, gives + +$$ +h\ge h_{\rm seam}=\frac{\sigma V}{K+\sigma A}.\tag{L3} +$$ + +For $\sigma=0$, the lower bound is zero. The exact integer criterion is + +$$ +\left\lceil L/h_{\rm opt}\right\rceil +\le m\le +\left\lfloor1+K/(\sigma A)\right\rfloor,\tag{L4} +$$ + +with $m\ge1$, and no finite seam upper bound when $\sigma=0$. Equalities require care in numerical rounding. The real interval $h_{\rm seam}\le h_{\rm opt}$ is necessary but does not by itself guarantee a permitted integer partition. + +**Joint obstruction.** If (L4) has no integer, this declared architecture cannot satisfy the optical selectivity and seam-yield requirements simultaneously, even with perfect screening and unlimited scheduling intelligence. This is a restriction on the architecture and parameters, not a proof that the target is impossible under every process. The available repairs are measurable: reduce leakage or attenuation, use a different reaction, lower seam hazard, preserve direct growth continuity, permit repairable joins, or change the target. + +## Exact conditional cost and unique continuous optimum + +Suppose one section attempt takes $t(h)=t_0+t_1e^{\mu h}$ seconds, including its declared setup and inspection costs; independent lethal section defects are Poisson with density $\rho$. Every rejected section is rebuilt independently and every rejected final stack loses all its sections. Joining and final inspection time are omitted unless included in the declared attempt costs, so this model is optimistic. A more complete model adds those costs explicitly. + +Perfect local screening gives mean attempts $e^{\rho Ah}$ per accepted section. Independence between section preparation and final seam pass gives + +$$ +\mathbb E T(h)=\frac{L}{h}(t_0+t_1e^{\mu h}) +\exp[\rho Ah+\sigma A(L/h-1)].\tag{L5} +$$ + +This is serial service time under the declared model. It is not wall time for a parallel factory and it does not include unpriced upstream production. Let $f(h)=\ln\mathbb E T(h)$. Then + +$$ +f'(h)=-\frac1h+\frac{\mu t_1e^{\mu h}}{t_0+t_1e^{\mu h}} ++\rho A-\frac{\sigma V}{h^2},\tag{L6} +$$ + +$$ +f''(h)=\frac1{h^2}+ +\frac{\mu^2t_0t_1e^{\mu h}}{(t_0+t_1e^{\mu h})^2} ++\frac{2\sigma V}{h^3}>0.\tag{L7} +$$ + +For positive $t_0,t_1,\mu$ and nonnegative defect densities, the log-cost is strictly convex. Its unconstrained derivative has one zero on $(0,\infty)$; constrain it to the permitted thickness interval and evaluate the neighboring allowed integers $m$. If the interval is empty, no optimum exists. This is a global one-dimensional calculation, not a heuristic fit. + +With $\rho=\sigma=0$, the stationary thickness is + +$$ +h_*=[1+W(t_0/(e t_1))]/\mu,\tag{L8} +$$ + +where $W$ is the real principal Lambert function. The code solves the defining scalar equation directly. The fourth-root cubic-module law and this exponential-attenuation law describe different geometries and cost models; neither is a universal scaling law for all nanofabrication. + +![Synthetic optical and seam constraints. The admissible interval exists for the lower seam density and disappears for the higher one. All plotted quantities are model inputs or calculations, not measurements.](../figures/feasibility.png) + +# 11 Error-corrected manufacturing theory + +## Retry screening and selection bias + +Let a fresh module be bad with probability $p$. Each inspection misses a bad module with probability $\mu$ and falsely rejects a good one with probability $a$. Assume independent inspection outcomes conditional on the module's true class, no measurement damage, independent fresh modules, and acceptance only after $r$ passed inspections. Then + +$$ +s_r=(1-p)(1-a)^r+p\mu^r, +\quad +q_r=P(\mathrm{bad}\mid\mathrm{accepted})= +\frac{p\mu^r}{s_r}, +\quad +\mathbb EN_{\rm attempts}=s_r^{-1}. +\tag{10} +$$ + +**PROVEN:** These follow by total probability and the geometric distribution. Repetition until acceptance does not produce a defect probability of merely $p\mu^r$; acceptance conditioning is essential. A cap of $K$ fresh attempts gives completion probability $1-(1-s_r)^K$. For identically distributed independent attempts, the quality of the accepted module is still $q_r$ conditional on completion. Early stopping can reduce inspection work, but cannot improve the quality formula without changing the protocol. + +For a per-module budget $\eta$, $q_r\le\eta$ is equivalent to + +$$ +\left(\frac{\mu}{1-a}\right)^r +\le\frac{\eta(1-p)}{p(1-\eta)}. +\tag{11} +$$ + +For $00 +\tag{12} +$$ + +for fixed nonzero $a$. Early stopping on the first failed inspection can remove the leading logarithmic factor from this asymptotic work model, but the retry multiplier $(J/\epsilon)^\gamma$ remains when a fresh attempt has positive base cost. Thus logarithmic repetition alone does **not** establish logarithmic total overhead. For $a=0.001$ and $\mu=0.01$, $\gamma\simeq2.17\times10^{-4}$: modest over practical ranges but nonzero asymptotically. + +## Proposition 4 A conditional local verification threshold + +An alternative uses an aggregate statistical test rather than requiring every inspection to pass. For a fixed part, let each independent nondamaging observation be a binary defect vote. Suppose the positive-vote probability is at most $a$ for every conforming state and at least $d$ for every relevant defective state, with a uniform separation $d>a$. Reject when the empirical fraction exceeds $\theta=(a+d)/2$; adopt a conservative tie rule. Hoeffding's inequality bounds both class-error probabilities by + +$$ +e^{-\kappa r},\qquad \kappa=(d-a)^2/2. +\tag{13} +$$ + +For independently rebuilt parts with $p\le p_{\max}<1$, choose $r$ such that $e^{-\kappa r}\le1/2$ and + +$$ +r\ge\frac{1}{\kappa} +\ln\!\left(\frac{2p_{\max}J}{(1-p_{\max})\epsilon}\right). +\tag{14} +$$ + +Then accepted bad probability is at most $\epsilon/J$, and acceptance probability is at least $(1-p_{\max})/2$. If each attempt has bounded base cost and each observation has bounded cost, expected work for $J$ accepted independent parts is + +$$ +\mathbb EW=O\!\left( +\frac{J[1+\kappa^{-1}\log(J/\epsilon)]}{1-p_{\max}} +\right). +\tag{15} +$$ + +**Proof.** Apply Hoeffding separately to good and bad parts. Good acceptance is at least $1-e^{-\kappa r}$; bad acceptance is at most $e^{-\kappa r}$. The posterior bad fraction is bounded by $2p_{\max}e^{-\kappa r}/(1-p_{\max})$. Eq. (14) makes that fraction at most $\epsilon/J$. The good contribution alone bounds overall acceptance below. Summing expected geometric attempt costs proves Eq. (15). $\square$ + +This is a sufficient threshold in **observability and recoverability**, not a universal elementary-error threshold analogous to a hardware-independent constant. It assumes every important defect class remains distinguishable, rebuilding remains affordable, fresh attempts are available, and already accepted regions survive later work. High-probability completion requires retry caps or tail bounds. Extending it to an object with $O(J)$ joins requires the same test/reset/preservation conditions at those joins. An irreversible, damaging closure with no subsequent informative measurement violates the assumptions. + +Fault-tolerant algorithmic self-assembly is established prior art; redundancy reducing errors has been demonstrated in DNA tile systems.[^dna] Eq. (15) is a standard statistical construction used to expose the engineering conditions needed for a similar claim in VLWNC-IF-VF. It is not evidence that VLWNC-IF-VF already meets them. + +## Blind defects, damage, and correlated errors + +Suppose a fraction $b$ of bad parts is permanently invisible to the permitted sensor. Even if all other defects become perfectly detectable, the outgoing defect fraction tends to at least + +$$ +q_{\rm blind}=\frac{pb}{(1-p)+pb}, +\tag{16} +$$ + +under the ideal limit of perfect good acceptance and no other rejection. For $p=0.01$ and $b=10^{-4}$, this is approximately $1.01\times10^{-6}$. A million critical independently accepted interfaces then cannot achieve a 99% defect-free object yield. Increasing inspection count cannot remove the invisible subpopulation. + +A useful conservative engineering budget is + +$$ +q_{\rm release}\le q_{\rm screen}+q_{\rm transfer} ++q_{\rm closure}+q_{\rm lifetime}, +\tag{17} +$$ + +when each term bounds the corresponding event on the appropriate conditioned population. Measurement-induced damage needs its own term or a dynamical state model. A photon or electron beam can modify the state it is meant to inspect. Common feedstock contamination, global temperature drift, and a shared wrong calibration create correlated failures. Repeated images of the same hidden systematic defect are not independent evidence. + +## Repair cascades + +For localized repair, let $B_{ij}$ be the expected number of type-$j$ repair jobs caused by one type-$i$ repair, including collateral damage and reinspection. With an initial row vector $z_0$ of repair jobs, a branching model gives + +$$ +z_{\rm total}=z_0(I-B)^{-1} +\quad\text{if}\quad\operatorname{spr}(B)<1. +\tag{18} +$$ + +**PROVEN within the branching model:** expected generation counts are $z_0B^n$; sum the convergent Neumann series. The threshold concerns the reachable repair subsystem. An unreachable supercritical block is irrelevant. In a reachable irreducible class with spectral radius at least one and positive job cost, expected total progeny is unbounded in the unlimited model. Finite retry caps bound work by accepting a nonzero abort/scrap probability. + +This provides a measurable engineering target: repairs should eliminate more uncertainty than they create. A one-type mean of $0.3$ implies an expected $1/(1-0.3)=1.429$ repair jobs per initial job; $0.9$ implies 10. A small component of thermally coupled neighbors can destroy an apparently favorable local repair rate. Annealing a crystal, replacing a tile, and removing a bad molecular fragment have different collateral-damage graphs and cannot share an invented universal repair constant. + +## Inspection frequency and defect tolerance + +In a simple independent-operation model, inspecting every $b$ operations costs $c_I/b$ per useful operation and an undetected rare failure of probability $p$ wastes approximately $pb/2$ operations. The leading overhead $c_I/b+pb/2$ has optimum $b_*\simeq\sqrt{2c_I/p}$. This is the familiar checkpoint/rework trade-off, not a new scaling law; comparable checkpoint analyses are established.[^checkpoint] Chemical accessibility, test latency, latent defects, and irreversibility can force earlier inspection regardless of this optimum. + +The strict defect-free criterion should be used only where the target requires it. If up to $t$ independent failed sites out of $J$ are acceptable, functional yield is $\sum_{k=0}^{t}\binom Jk q^k(1-q)^{J-k}$. Mechanical load redistribution, electrical redundancy, or photonic rerouting may tolerate defects, but each needs a validated functional model. A bypassed bad junction is functionally tolerated; it has not become structurally perfect. + +# 12 Physical compiler architecture + +## Input and intermediate representations + +A target contains geometry, material identities and allowed substitutions, surface/interface requirements, tolerances, operating conditions, expected lifetime, allowed imported content, and release tests. The compiler retains a distinction between a structural requirement and a functional substitute. Replacing a single crystal with a polycrystal, a metal with a conductive polymer, or a monolithic junction with a transferred chiplet requires that the target explicitly permit it. + +The intermediate representation has five linked graphs: material/reaction transformations; geometric access and support; optical response and protected regions; resource and process dependencies; and evidence production/invalidation. Each region has a current state estimate and a set of qualified operations. Every planned closure has a list of predicates that become harder to inspect or repair. Each later process includes dose, temperature, solvent, force, and field envelopes that may invalidate those predicates. + +A route search should first eliminate impossible chemistry, unsupported material combinations, and missing access. It then solves a calibrated dose problem for each candidate state. An infeasible dose problem returns a witness and a physical repair menu: change wavelength, shorten the optical path, add a qualified mask, use another cartridge, write on a donor carrier, or reject the route. The solver must never replace an absent chemical pathway with an invented rate constant. + +## Constrained optimization + +Choose route variables $z_a$, start times $t_a$, channel doses $u_a$, carriers, joins, inspection plans, and allowed partitions to minimize a weighted resource objective or produce a Pareto frontier. A representative objective is + +$$ +\min\;T+\lambda_EE_{\rm wall}+\lambda_MM_{\rm waste} ++\lambda_CC_{\rm consumables}. +$$ + +Constraints include element and charge balances; calibrated reaction domains; $A_a^-u_a\ge g_a$ and $B_a^+u_a\le b_a$ when the linear model applies; geometry and support accessibility; material compatibility; nonoverlapping resource occupation; current, flow, and power limits; temperature/stress evolution; cumulative preservation budgets; and valid evidence at every closure. Unit conversions must be explicit when objectives are combined. + +For a protected region $i$, a cumulative irreversible-hazard constraint is $\sum_a B_{ia}^+u_a\le b_i$, including exposure during metrology. It cannot be replaced by a separate $B_{ia}^+u_a\le b_i$ check at every step. The dose operator depends on state, so the sum uses the appropriate operator for each event. Thermal accumulation needs a dynamic state equation rather than a cumulative scalar when cooling between operations matters. + +Release quality is a constraint on the released population. Either calibrate $P(\mathrm{bad}\mid\mathrm{release})$ directly, or combine joint release/failure bounds with a lower release-probability bound as in Section 8. Unlimited retries are not a way to avoid false-acceptance conditioning. + +## Compilation algorithm + +```text +compile(target, qualified_catalogue, instrument): + normalize requirements and permitted substitutions + generate finite candidate process and carrier routes + for each route: + verify reaction, element, charge, and compatibility constraints + derive current optical state and uncertainty domain + solve dose feasibility and resource allocation + if infeasible: + store witness; try a permitted physical route transformation + derive access-loss and cumulative-damage obligations + place each test after its last known invalidator and before closure + charge risks introduced by transfer, joining, and closure themselves + reject cycles, missing tests, unsupported preservation, and heat violations + schedule resources; estimate accepted-output cost and uncertainty + independently replay the selected plan against the declared contracts + emit a qualified plan, a hypothesis-only plan, or explicit infeasibility +``` + +A compiler may emit a **hypothesis-only plan** for laboratory research, but it must not release a physical object on its authority. A plan hash establishes integrity, not truth. Statistical model uncertainty, calibration version, and unresolved risks remain part of the result. + +## What the executable demonstrator actually does + +The included code solves finite nonnegative dose problems with interval uncertainty, rejects a one-millimetre optical path in the declared scalar model, and selects an exposed-carrier path of 100 micrometres. A finite event catalogue declares certificate producers, consumers, and invalidators. The compiler inserts required dependency edges and topologically orders those events. An independent replay identifies missing part qualification and stale interface evidence in a deliberately invalid route, and finds no declared violations in the compiled route. + +This is a restricted reference demonstrator. It does not infer reaction chemistry, import arbitrary CAD, solve a full optical field, control hardware, or prove a commercial manufacturing optimum. The proposed production compiler requires those separately validated capabilities. The generated record therefore remains `not_released` and lists unqualified physical dependencies. + +# 13 Hardware architecture + +## The cabinet and its internal boundaries + +Prototype B's design envelope is a cabinet approximately $1.2\times0.8\times1.6$ m, excluding a service chiller, gas supply, exhaust treatment, and computer if external. A 50 mm carrier supports a qualified $20\times20$ mm working field and products up to about 20 mm high. These are proposed dimensions, not a packaged instrument specification. A bench implementation can use separate existing tools before mechanical integration. + +The cabinet contains a dry optical bay, a replaceable wet-process cassette, a carrier-cleaning/transfer vestibule, and an inspection/assembly bay. A shared optical reference frame aligns carrier fiducials. Wetted paths are cartridge-specific. Exhaust and waste streams remain separated where chemistry requires it. A single sealed enclosure cannot make incompatible solvents, electrolytes, and hot inorganic processes compatible. + +| Subsystem | Proposed implementation | What must be measured | +|---|---|---| +| Pattern engine | 365/405 nm LED or laser modules, DMD or qualified spatial modulator, relay optics | Delivered spectral irradiance, modulation transfer, stray light, uniformity, stability | +| Fine-field option | Higher-NA objective on a smaller scanning field | Actual process resolution, depth of focus, stitching and drift | +| Carrier motion | Encoded XYZ stages, tip/tilt alignment, force-limited transfer head | Registration after release and bonding, not only stage encoder resolution | +| Reaction window | Replaceable transparent substrate or donor film; controlled gap | Absorption, fouling, gap variation, chemical swelling | +| Wet cassette | Metered feed, rinse and separate waste; potentiostat for metal deposition | Carryover, limiting current, current efficiency, ionic contamination | +| Dry-film cassette | Resist or structural-film coating/lamination and qualified development | Thickness, exposure window, adhesion, residual solvent | +| Joining station | Low-temperature lamination, qualified adhesive/perimeter bonding, optional direct bonding | Seam hazard, stress, alignment, closure-induced defects | +| Metrology | Bright/dark-field camera, reflectometry or interferometry, electrical probes; AFM on selected surfaces | Defect-class detection, false alarms, blind regions, measurement damage | +| Thermal service | Temperature-controlled carrier platen and recirculating coolant | Local peak temperature and actual facility heat load | +| Controller | Local deterministic loops and supervisory route/evidence scheduler | Timing, state-estimation error, fault recovery and evidence invalidation | + +For a roughly 20 mm field sampled with 2,000 pixels, pixel pitch is about 10 micrometres. At 405 nm and NA 0.05, the conventional scale $\lambda/(2\mathrm{NA})$ is 4.05 micrometres; pixels, chemistry, defocus, diffusion, and motion can make the print coarser. A high-NA small field improves optical resolution but reduces field size and working distance. The baseline cabinet therefore targets 10-25 micrometre patterned features and about 2 micrometre registration first. A later qualified fine mode may target 1-5 micrometre surface features. Neither mode claims arbitrary atomic positioning. + +## Physical instruction families + +The practical instruction set is procedural: `coat`, `expose`, `develop`, `deposit`, `rinse`, `cure`, `inspect`, `transfer`, `join`, `remove_support`, `close`, and `requalify`. Each is parameterized by a material-specific contract. A command such as `deposit copper` requires a conductive path, an electrolyte or donor material, actual reduction chemistry, and an endpoint measurement. A command such as `grow crystalline silicon` cannot be serviced by a polymer cartridge. + +The shared light engine functions like an addressing bus. It is not a universal source of chemical selectivity. Electrically powered deposition can use an optically patterned mask; it need not depend on a new light-addressable electrode. Optically controlled metal electrodeposition itself has prior art.[^electrode] The conservative fallback uses established resist patterning, an appropriate seed layer, controlled electroplating, and qualified stripping. The process must then protect materials sensitive to stripping or plating. + +Surface-active chemistry or a removable photoreactive layer may translate light into material placement. The translator is consumed or reset at a measured cost. A high-density spatial address carried by a template is information that was fabricated upstream; it is not free sub-diffraction programming. Reusing a template is valuable for repeated objects but does not establish arbitrary atom-by-atom programmability. + +# 14 Feedstock and chemistry architecture + +## A smallest useful catalogue, not a universal basis + +No finite minimal library spanning stable matter is known. The first useful catalogue is benchmark-specific. Preserve prepared materials when they reduce work: glass carriers, crystalline chips, metal films, characterized polymers, and stable precursor solutions are often better feedstocks than isolated atoms. Feedstock normalization means known identity, composition, contamination, storage history, and interface condition; it does not require dissociation into atoms. + +Prototype A uses a glass carrier, a qualified photoimageable dielectric or structural film, a conductive seed or metal-coated donor, a controlled copper process, compatible support/release material, a qualified joining material, cleaning fluids, and inert or clean dry gas where required. This is about eight stock categories, not eight universally sufficient chemicals. Prototype B adds selected inorganic nanoparticle films, oxide-coated donors, and independently fabricated sensor or photonic components. Each addition introduces new compatibility and metrology work. + +## Reaction families and known-physics fallbacks + +| Family | Actual transformation | Role of light | Principal limitation and fallback | +|---|---|---|---| +| Photopolymerization | Initiation, propagation and termination forming a polymer network | Generate active species or switch initiator state | Oxygen, diffusion, shrinkage, residual monomer; use a qualified commercial process window | +| Photoresist-defined metal | Pattern a mask, reduce metal ions at a connected cathode, strip if allowed | Define where deposition is permitted | Seed continuity, current spreading, wet compatibility; transfer prepatterned metal if direct processing fails | +| Inorganic nanocrystal film | Change ligand/surface state and film solubility, then develop and consolidate | Local chemical activation for patterning | Nanocrystal boundaries, residues, transport properties; import qualified film or device | +| Glass/ceramic-containing structure | Pattern a precursor/composite, then remove binder and densify when required | Define green-body geometry | Shrinkage, porosity, furnace temperatures; process separately and transfer | +| Functional module integration | Move and bond a previously fabricated module | Optional alignment, imaging, release or transfer actuation | Imported synthesis and seam costs remain explicit | +| Surface precision | Selected epitaxy, self-limiting deposition or specialized probe chemistry | Optional activation/metrology, not generic atomic addressing | Tiny qualified surface family; retain the specialized tool as a separate cartridge or upstream service | + +Direct optical lithography of inorganic nanomaterials demonstrates that light-responsive ligands can pattern several material families.[^dolfin] It does not show that the resulting nanocrystal films are interchangeable with bulk single crystals or that all compositions share a common precursor. A new VLWNC-IF-VF cartridge would require its own film-quality, conductivity or optical-loss, adhesion, and chemical-residue qualification. + +For electrochemical copper growth, the useful overall cathodic reaction is $\mathrm{Cu^{2+}+2e^-\rightarrow Cu}$. The counterelectrode process, ion transport, and charge balance must be included. For current density $j$ and current efficiency $\eta_F$, the ideal thickness rate is + +$$ +\dot d=\frac{\eta_F j M_{\rm Cu}}{2F\rho_{\rm Cu}}. +$$ + +At the illustrative $j=100\,\mathrm{A/m^2}$ and $\eta_F=0.9$, this is 3.31 nm/s, giving about 302 s per micrometre. This is a Faraday-law estimate, not a validated deposition recipe. A one-micrometre copper film over one square centimetre needs 2.72 C at ideal efficiency. More optical power does not eliminate the required charge or prevent diffusion limitation. + +For free-radical polymerization, a minimal state model is $\dot R=2f k_i I[PI]-k_tR^2-k_q[Q]R$ and $\dot M=-k_pMR$. Constants and units depend on the chosen intensity convention and medium. The model must include transport and oxygen or inhibitor dynamics when relevant. No numerical rate constants are asserted for a new resin here. Dual-wavelength activation/inhibition is an optional experimentally qualified process, with xolography and binary photoinhibition as prior art.[^xolo][^binary][^antagonistic] + +A diffusion-reaction illustration uses $\partial_t c=D\nabla^2c-kc+s$. For a sinusoidal source with wavenumber $q$, steady modulation is attenuated by $k/(k+Dq^2)$. With synthetic $D=10^{-10}\,\mathrm{m^2/s}$, $k=100\,\mathrm{s^{-1}}$, 20 micrometre period, and 0.9 source modulation, the reaction-species on/off ratio is only about 10.06. That is far below the 458.21 first-order selectivity ratio for 99%/1% conversion. A sharp optical image is therefore insufficient evidence of selective chemistry. + +## Purity, rinsing, and material preservation + +For $N_c$ independent lethal-contamination opportunities each with probability $c$, survival is $(1-c)^{N_c}$. If every atom were critical, the purity requirement would be prohibitive for macroscopic objects. Actual design must identify critical sites, exploit selective incorporation, and tolerate or repair noncritical defects. Feedstock impurity and incorporated impurity are not equal without an experimentally measured segregation or incorporation model. + +An ideal well-mixed rinse has concentration ratio $e^{-Qt/V_c}$, where $V_c$ is fluid volume and $Q$ is flow. Twelve ideal volume exchanges leave $6.14\times10^{-6}$ of the starting concentration. Dead volumes, adsorption, films, porous supports, and recirculation can dominate; this ideal number is not a cleanliness certificate. Separate wet paths and sacrificial carriers may cost less than extreme shared-path cleaning. + +Stripping a resist can attack a polymer, solvent can swell a support, and cure heat can alter an earlier interface. The catalogue stores an experimentally bounded compatibility matrix over complete process histories. A pairwise compatibility check is necessary but not sufficient: a sequence can accumulate damage even when every isolated step passes. Metastability after release must be tested under the intended humidity, temperature, mechanical load, and time. + +# 15 Metrology architecture + +Metrology is designed at the same time as the object. Each predicate must have a sensor response, a resolution definition, a detection limit, a false-acceptance model, a calibration interval, and a plan for access after future processing. “Verified” is meaningless without those fields. + +| Property | Candidate observations | What remains unproved by a pass | +|---|---|---| +| Surface geometry | AFM, interferometry, profilometry, electron imaging | Arbitrary buried atomic positions | +| Composition and bonding | Raman/IR, X-ray methods, surface spectroscopy, chemical assay | Every rare impurity and every hidden local bond | +| Thin-film thickness | Ellipsometry, reflectometry, step-height measurements | Unique composition or absence of all pinholes | +| Internal structure | X-ray tomography; electron tomography on suitable small specimens | Whole macroscopic objects at uniform atomic resolution | +| Interface integrity | Electrical chains, acoustics, optical response, mechanical coupons | All inaccessible defects outside test sensitivity | +| Device function | Calibrated optical/electrical/mechanical transfer functions | Microscopic structural uniqueness | + +Atomic electron tomography has reconstructed specific nanoparticles with tens of thousands of identified atomic coordinates and reported picometre-scale precision.[^aet] That does not establish rapid nondestructive atomic tomography of arbitrary dense macroscopic assemblies. Inverse reconstruction can have non-unique solutions, limited angles, missing data, and dose constraints. Learned reconstruction must report how its prior affects rare-defect detection. + +For a simple repeated Gaussian measurement with two candidate means separated by $\Delta$ and per-observation standard deviation $s$, equal-prior optimal classification error is $\Phi[-\Delta\sqrt n/(2s)]$. Therefore $n\ge4s^2[\Phi^{-1}(1-\delta)]^2/\Delta^2$ is required to reach error $\delta$ in this particular model. Calibration bias does not shrink with $n$. Eq. (2) is the more general obstruction when no informative sensor exists. + +As an ideal optical estimate, reflection phase sensitivity with independent coherent photons gives $\sigma_x\gtrsim\lambda/(4\pi\sqrt{n_{\rm det}})$ when phase is the only unknown and detection is otherwise ideal. This bounds detected photons for that measurement model, not general metrology cost. Absorption, collection efficiency, backaction, laser preparation, electronics, scanning, and state ambiguity can dominate. There is no universal fixed number of joules for inspecting an arbitrary bond. + +## Statistical evidence and extreme error claims + +With zero failures in $n$ independent representative Bernoulli trials, a one-sided confidence level $1-\alpha$ gives $q\le1-\alpha^{1/n}$. To support $q\le10^{-8}$ at 95% confidence by zero-failure sampling alone requires at least **299,573,226 trials**. Sample selection, dependence, lot variation, and changing instrument state invalidate the naive count. Mechanistic evidence and test structures can improve inference but must be stated as additional assumptions. + +This is a severe bottleneck for the optimistic error-correction examples. A simulation can calculate a $10^{-10}$ outgoing risk under assumed rates; it cannot validate those rates. Store confidence regions and coverage limitations rather than a falsely exact “quality probability.” + + +## Optical preservation and cumulative dose + +The nine-check simulation contains a deliberately simple counterexample. A protected region receives unwanted hazard 0.0012518 per exposure, below the single-step limit 0.0100503. After 100 identical exposures, first-order conversion is $1-e^{-0.12518}=11.77\%$, exceeding a 1% allowed change. Meeting the same single-step tolerance repeatedly does not imply lifetime preservation. + +Within this example, reducing the protected response coefficient from $10^{-4}$ to at most $8.03\times10^{-6}$ would satisfy the cumulative 100-exposure budget. This is a model-derived target for shielding, path removal, or material insensitivity. It is not a claim that a particular shield achieves that number. If a shield blocks the light needed for later processing, it changes the route and may introduce new access or transfer costs. + +Real-time volumetric shape metrology has already been demonstrated in selected photopolymers.[^metrology] Shape evidence does not certify buried metal connectivity, rare contamination, or every bond. The cabinet uses multimodal evidence and retains explicit blind regions. A certificate may therefore assert a measured functional tolerance while declining an unsupported atom-level structural claim. + +# 16 Thermal and energy analysis + +## Thermodynamic accounting + +For a control volume, with heat $\dot Q_j$ positive into the system, + +$$ +\frac{dS}{dt}=\sum_{\rm in}\dot m s- +\sum_{\rm out}\dot m s+\sum_j\frac{\dot Q_j}{T_j} ++\dot S_{\rm gen},\qquad \dot S_{\rm gen}\ge0. +\tag{20} +$$ + +Entropy can leave in coolant, exhaust, waste, or product; it does not disappear. Reactions can be exothermic or endothermic, and recoverable work can change heat loads. The engineering ledger includes feedstock production, reaction driving, motion, inspection, computation, vacuum, refrigeration, pumps, abatement, cleaning, and rework. For a complete repeated cycle with no net stored energy and appropriate system boundaries, energy conservation determines the heat and outgoing chemical-energy balance. + +Erasing an unbiased unknown classical bit in an ideal cyclic memory at temperature $T$, without exploitable side information or changing other free-energy resources, dissipates at least $k_BT\ln2$ in the quasistatic limit. Experiments have examined this limit in nanomagnetic memories.[^landauer] At 300 K it is $2.87\times10^{-21}$ J per bit. It is not a mandatory cost per atom, bond, measurement, or reversible decision. Finite speed and reliable hardware usually impose larger costs; preparation and memory resetting must be included. + +One electronvolt per incorporated atom would be $3.44$ kJ per gram of silicon. One erased bit per atom would be $61.6$ J per gram at 300 K. Neither is a universal fabrication-energy estimate: stable feedstocks may already contain bonds, energy may be recovered, and utility energy may dominate by many orders. These numbers illustrate why a purported near-Landauer factory budget cannot be inferred from chemical atom count. + +## Heat-removal bounds + +For uniform volumetric heating $q'''$ in a planar slab of thickness $h$ and conductivity $k$, with both faces fixed at coolant temperature, the steady peak temperature rise is + +$$ +\Delta T_{\max}=\frac{q'''h^2}{8k}. +\tag{21} +$$ + +With identical convective boundaries of heat-transfer coefficient $h_c$ on both faces, it becomes + +$$ +\Delta T_{\max}=q'''\left( +\frac{h^2}{8k}+\frac{h}{2h_c}\right). +\tag{22} +$$ + +These follow by solving $kT''+q'''=0$ with the specified boundary conditions. They neglect interface resistances, anisotropy, temperature-dependent properties, and localized hot spots. Those effects generally lower usable power for a given maximum temperature. + +For $h=1$ mm, $k=1\,\mathrm{W/(m\,K)}$, and allowed rise 20 K, ideal fixed-temperature faces permit $1.6\times10^8\,\mathrm{W/m^3}$. With $h_c=10^4\,\mathrm{W/(m^2K)}$, the bound is $1.14\times10^8\,\mathrm{W/m^3}$. Over a $100\,\mathrm{cm^2}$ slab this is approximately 1.14 kW. These assumed properties describe a thermal illustration, not a measured VLWNC-IF-VF substrate. + +If active event density is $n_v$ events per cubic metre and per-event repetition rate is $r$ with irreversible local heat $e_h$, then $n_vre_h\le q'''_{\max}$ in this steady model. For $e_h=1$ pJ, the illustrative convective slab allows at most $1.14\times10^{20}$ events per cubic metre per second before other heat loads. If the event energy estimate excludes its control electronics or feedstock enthalpy, this result is not a factory throughput bound. + +For a sealed burst, heat storage $C_{\rm th}\Delta T$ permits temporary operation faster than $Q/H$. The correct finite-cycle lower bound is $T\ge\max[0,(Q-C_{\rm th}\Delta T)/H]$, with an ensuing cooling interval if repeated production must return to the initial state. Ignoring thermal storage overstates a one-shot bound; ignoring cooldown overstates sustained throughput. + +Small cells improve conduction distance but multiply connections, pumps, windows, and inactive volume. A planet-sized factory does not solve this geometry. At fixed surface temperature, radiative heat rejection is $\varepsilon_{\rm rad}\sigma_{\rm SB}A(T_s^4-T_{\rm env}^4)$. Large internal volume cannot sustain arbitrary volumetric processing unless heat paths and external rejection area grow accordingly. + + +## Optical and electrochemical energy accounting in the cabinet + +Incident optical power is divided into useful absorption, parasitic absorption, reflected/transmitted light, and losses in the projector. Electrical source power is larger than delivered optical power by the inverse wall-plug and optical efficiency. Polymer reactions may release additional heat. Metal deposition consumes electrical and chemical free energy, while rinsing, drying, vacuum, motion, computation, and cooling contribute facility loads. + +For a proposed 2-5 kW cabinet operating for 2-6 h per coupon, instrument electricity is 4-30 kWh before allocated external services and upstream feedstocks. This is a design envelope that must be measured. A 20 W delivered beam does not imply a 20 W factory. A 1 cm$^2$ field receiving 100 mJ/cm$^2$ absorbs or receives 0.1 J depending on the stated dose definition; at 10 mW/cm$^2$, exposure is ten seconds before all other process work. The example is not a recommended resin dose. + +In the finite-difference check, a one-millimetre slab with $k=1$ W/(m K) and uniform $10^8$ W/m$^3$ generation rises 12.5 K with ideal fixed-temperature faces. With $10^4$ W/(m$^2$ K) convective boundaries, the analytic rise is 17.5 K. A real patterned metal/polymer stack has interface resistance, transient hot spots, and anisotropy. The simple slab is a model verification, not thermal sign-off for the cabinet. + +# 17 Throughput analysis + +## Useful parallelism + +With per-cell event rate $r$ and $P$ cells, useful rate is at most + +$$ +\dot N_{\rm useful}\le\min\left\{ +Pr\Pi,\frac{H-Pp_0}{e_h},\dot N_{\rm feed}, +\dot N_{\rm inspect},\dot N_{\rm transport} +\right\}, +\tag{23} +$$ + +when the numerator in the heat term is positive. $p_0$ is local baseline heat per installed active cell and $\Pi$ is utilization after scheduling and rework. This simplified form assumes comparable counted events and a common thermal budget; station-specific bounds are needed for heterogeneous work. + +For example, one million cells dissipating 1 mW each already use 1 kW before useful reactions. At 1,000 events per cell per second, the count is $10^9$ events/s; if each event adds 1 pJ, its incremental heat is only 1 mW. The baseline dominates by a factor of a million. Such cell density, power, and rate have not been demonstrated together for the proposed architecture. + +If cells interfere chemically or mechanically, feasible simultaneous sets are independent sets of an interference graph. For diffusive contamination, a characteristic spread length during an operation is $\sqrt{D\tau}$; for heat it is $\sqrt{\alpha\tau}$. Barriers, purge flow, and scheduling can reduce cross-talk at a cost. No universal nanometre cell spacing can be asserted without this analysis. + +Local control avoids sending every actuator sample centrally. Even so, $P=10^6$ cells emitting 64-byte summaries at 100 Hz generate 6.4 GB/s before protocol overhead. Global per-event control at high rates can be much larger. Shared clocks, deterministic slots, local estimates, and exception reporting help only when their assumptions match the process. Physical feedstock must still cross channels with finite area and flow. + +## Atomwise and hierarchical limits + +A gram of silicon contains approximately $2.144\times10^{22}$ atoms. A hypothetical aggregate rate of $10^9$ atom placements/s would require approximately **679,000 years**. Even $10^{15}$ placements/s would require about **248 days**, ignoring repair, inspection, and downtime. These are counting calculations, not claims about available tools. + +Collective growth can incorporate enormous numbers of atoms per controlled instruction. Lithography, deposition, crystallization, and templating already exploit that advantage. The aim is to allocate explicit positioning to dopants, interfaces, vacancies, or other features whose specification genuinely needs it. The compiler's main opportunity is to reduce *independent interventions* while preserving the target's functional and structural meaning. + +For an illustrative coating, 0.1 nm growth per cycle and a 2 s cycle require 2,000 s for 100 nm, before setup. The process can coat many accessible sites together. These chosen values are representative model inputs rather than specifications for a particular commercial reactor. Conformality does not guarantee complete monolayers, arbitrary species selection, or defect-free coating of closed pores. + +## Picosecond manufacture + +Light travels approximately 0.30 mm in one picosecond. A fresh centralized instruction cannot causally determine material changes over a 1 cm object in that interval; the light-crossing time is about 33 ps. At a material signal speed of $5,000\,\mathrm{m/s}$, a stress wave crosses 1 cm in 2 microseconds. Moving ordinary feedstock, completing serial reactions, removing byproducts, and verifying a macroscopic object impose additional limitations. + +Prearranged local systems can switch nearly simultaneously without violating causality. A picosecond local transition, ultrafast bond motion, or illumination-triggered state change is physically possible in appropriate systems. This does not build a new arbitrary macroscopic object from controlled feedstocks in one picosecond. Counting the prior arrangement, instruction distribution, and prepared material as free would merely move the fabrication outside the advertised time. **General picosecond macroscopic manufacture is rejected; selected local operations remain possible.** + + +## What optical parallelism can and cannot accelerate + +A projected pattern can expose many sites at once, but useful rate is bounded by delivered dose over area, precursor arrival, product removal, chemical selectivity, heat rejection, and inspection. At fixed irradiance and optical efficiency, exposing $P$ equal fields requires optical power proportional to $P$. At fixed total power, dose time grows with total illuminated area. A million digital pixels are not a million independent chemical processors. + +For a process with areal useful throughput $\dot n_A$ and parallel active area $A_{\rm act}$, an optimistic useful rate is $\dot N\le A_{\rm act}\dot n_A$, provided feedstock, thermal, and metrology limits all exceed it. If adjacent regions interfere, construct a conflict graph and schedule noninterfering sets. An optical field can be shared over passive regions without requiring local electronics at every pixel, but local chemistry can still diffuse across their boundaries. + +The boxed design targets patterned functional coupons, not maximum bulk mass output. Layer count, material changes, and inspection cycles may dominate time. Photopolymer frames or covers can be fast while a micrometre of electrodeposited metal still takes minutes at the illustrative current density. A claimed throughput must report completed, accepted objects per elapsed time and include unsuccessful runs, cartridge changes, cleaning, and validation overhead. + +# 18 Numerical examples and executed checks + +All new numerical results are synthetic model calculations, identified by `data_origin: simulated`. The fixed seed is 20260913. The code records parameters, software versions, floating-point results, exact rational witnesses where supplied, and pass/fail checks. None of the rates, densities, or detector characteristics below is a measured VLWNC-IF-VF property. + +## Optical and interface window for a one-centimetre stack + +Take $L=10$ mm, $A=1$ cm$^2$, $\mu=10^4$ m$^{-1}$, $\beta=10^{-4}$, and desired/protected first-order conversions of 99%/1%. Then $h_{\rm opt}=308.30$ micrometres. Require at least 99% final seam survival. + +| Residual seam density | Minimum section thickness | Exact permitted section counts | Interpretation | +|---|---:|---:|---| +| $\sigma=0$ m$^{-2}$ | 0 | At least 33; code enumerates to 1,000 | No seam restriction in this ideal case | +| $\sigma=1$ m$^{-2}$ | 98.52 micrometres | 33-101 | Optical and seam windows overlap | +| $\sigma=10$ m$^{-2}$ | 904.95 micrometres | None | The declared architecture fails both constraints jointly | + +For $\rho=10^6$ m$^{-3}$, $t_0=30$ s and $t_1=1$ s, the low-seam-density case has continuous optimum 278.42 micrometres. The exact integer optimum is 36 sections of 277.78 micrometres. Mean accepted serial service time is 1,711.71 s, or 28.53 minutes; seam survival is 0.996506 and the mean local-attempt multiplier is 1.02817. These are values of (L5), not a promise that a one-centimetre heterogeneous object can be manufactured in half an hour. Joining, material-specific deposition, and facility logistics may dominate and are not automatically represented by these synthetic seconds. + +At a 100 micrometre optical path, required incident hazard is 12.518 and protected hazard is 0.001252. At 300 micrometres, protected hazard is 0.009250, still within the example limit. At one millimetre it is 10.144, far beyond the limit. An exposed carrier therefore changes feasibility in this model, but only a qualified transfer and preservation process would turn it into a usable fabrication route. + +![Conditional mean accepted service time over the optically and interfacially feasible integer partitions. This is an optimization illustration, not a hardware throughput forecast.](../figures/cost.png) + +## Computation and independent comparisons + +| Check | Independent reference or attack | Executed result | +|---|---|---| +| Robust dose planning | All 64 corners of a six-entry uncertainty box | Every corner satisfies the programmed bounds within numerical tolerance | +| Infeasible robust exposure | Exact rational dual witness | $M^Ty=0$, $v^Ty=-7/250<0$ | +| Reaction-diffusion blur | Analytic Fourier response; 64/128/256 point grids | Maximum errors $5.91\times10^{-5}$, $1.48\times10^{-5}$, $3.69\times10^{-6}$ | +| Lamellar optimum | Convex derivative root versus exhaustive integer enumeration | 36 sections; high-seam-density case has no feasible partition | +| Retry cost | 100,000 Monte Carlo completed objects versus geometric expectation | Predicted 11.1111 attempts; observed 11.1021, standard error 0.01607 | +| Thermal transport | Finite difference versus slab parabola | 12.5 K to numerical precision | +| Access frontier | Exact subset dynamic program on three eight-vertex graphs | Widths 1, 2 and 7 | +| Compiler evidence | Independent event replay of invalid and compiled routes | Two declared violations in invalid route; none in compiled route | +| Preservation | Accumulate individually permissible unwanted hazards | 100 exposures cause 11.77% conversion and fail a 1% limit | + +The dose optimum touches an uncertainty-boundary constraint. This demonstrates feasibility of the stated box, not operational reserve. A hardware controller should tighten thresholds or widen the uncertainty set to obtain a measured safety margin against discretization and model discrepancy. No small residual should be advertised as a physical error probability. + +# 19 Failure modes + +| Failure | Mechanism | Required response | +|---|---|---| +| Buried inaccessible interface | Closing geometry removes all useful observation or repair channels | Add access, change seam, relax tolerance, or reject route | +| Measurement damages the target | Radiation, heating, force, chemistry, or charging modifies inspected state | Include damage kernel; reduce dose or use another measurement | +| Purity requirement is unattainable | Fatal contaminants scale with enormous critical-site count | Selective incorporation, tolerant function, smaller critical region | +| Crystal continuity is lost | Joined tiles introduce grain boundaries, strain, or dislocations | Use contiguous growth or explicitly allow the interface | +| Organic/inorganic incompatibility | Anneal, plasma, solvent, or precursor destroys prior material | Separate synthesis and low-temperature integration | +| Repair cascade | Local correction damages neighboring predicates | Measure offspring matrix; limit repair scope; replace module | +| Metrology queue dominates | Sensor duty cycle is slower than production | Parallel inspection or less demanding test; reduce fabrication rate | +| Global calibration error | All cells inherit the same wrong reference | Independent references, drift monitoring, lot-level invalidation | +| Feedstock bottleneck | Purification and preparation are slower or more costly than assembly | Include them in optimization and system boundaries | +| Support cannot be removed | Scaffold is trapped or removal destroys adjacent material | Plan escape paths and removal selectivity before fabrication | +| Trapped stress or metastability loss | Release from support/pressure changes the stable configuration | Simulate release and measure lifetime after release | +| Cross-talk | Heat, fields, vibrations, or reactive species affect neighboring cells | Isolation, spacing, arbitration, or slower duty cycle | +| Latent post-test failure | Closure or aging introduces defects after successful inspection | Preservation model plus final/lifetime risk budget | +| Unqualified chemistry | A process name substitutes for a real selective pathway | Experimentally qualify it or mark the route unsupported | + +These failures define the architecture's admissible domain. They are not a list that can be cleared by naming an advanced sensor or an intelligent controller. A path that reaches the desired final structure only through an intermediate that destroys an irreplaceable component is not an admissible path for that specification. + + +Additional light-printer failures include changing optical attenuation during growth, resin or carrier fouling, photoinitiator depletion, unmodeled inhibition, cumulative dose damage, spatially correlated illumination errors, trapped solvent at a seam, donor-release damage, and electrochemical deposition on unintended conductive paths. A response matrix measured on a clean carrier may fail after a few cycles. Calibration must therefore include realistic aged and contaminated states, not only ideal initial specimens. + +# 20 Hostile audit + +**“The requested breakthrough has not been delivered.”** A new experimentally validated universal machine has not been delivered. The defensible result is a more specific architecture, an auditable conditional feasibility/cost theorem, and a reproducible expert-review package. It would be misleading to relabel that result as a completed major breakthrough because the objective requested one. + +**“This is a semiconductor fab, a laminator, and a projector in a cabinet.”** Much of the hardware is deliberately conventional. The proposed scientific contribution survives only if the common selectivity/access/preservation representation transfers across different process families and generates better admissible routes at reasonable cost. If ordinary expert process planning performs equally well, the strong architectural claim fails. + +**“The optical mathematics is already in inverse lithography.”** Correct. Nonnegative dose optimization, local dose bands, inverse rendering, and photoinhibition have close prior art. The claimed candidate is their coupling to physically different routes and evidence invalidation, plus the declared joint lamellar model. No priority claim is made for linear programming, Farkas' lemma, or positive-dose constraints. + +**“A layer is not a module, and its seam can be chemically continuous.”** Correct. The seam model applies only when residual interfaces are meaningful and have the declared lethal hazard. Continuous growth can eliminate a bond seam but introduces different thermal, surface, or composition constraints. Set $\sigma=0$ only when evidence supports that interpretation. A nonfatal microscopic interface is not automatically a lethal seam defect. + +**“The optical bound ignores arbitrary beam angles and nonlinear control.”** The scalar attenuation bound applies only to the declared one-sided route or a family of channels satisfying its pointwise contrast relation. Multi-angle propagation must use its full transfer model. Nonlinear photoinhibition can violate the nonnegative linear-hazard assumptions, so it requires a different contract. Neither possibility proves arbitrary deep access through an opaque completed object. + +**“Perfect tests make the yield theorem unrealistic.”** They intentionally make it optimistic and mathematically explicit. An empty feasible interval under favorable screening assumptions is informative for the declared model. A nonempty interval is not evidence of practical feasibility. Imperfect tests, correlated seams, and hidden damage require the probability framework and calibration program in Sections 8, 11, and 15. + +**“A one-step transfer can damage everything that was inspected.”** Yes. Transfer, curing, and closure have their own hazard and invalidation events. The compiler requires post-process interface tests where informative, calibrated preservation bounds where justified, and rejection where neither is available. A signature on a pre-transfer image is not a certificate for the final object. + +**“The smallest feedstock library hides the hard work upstream.”** The first catalogue is an experimental convenience. Imported crystalline and electronic components must carry an imported-content declaration and allocated upstream energy. The architecture does not claim that ordering every desired object as a cartridge is universality. + +**“Heat or electrochemical current destroys projection parallelism.”** That can happen. The wall-plug, cooling, transport, and current budgets constrain simultaneous area. Increasing projected pixel count cannot exceed those rates. The first prototype measures full accepted-output energy rather than estimating it from photons or bits. + +**“The chemistry does not support the promised precision.”** The baseline promise is micrometre-scale patterned geometry with selected thin-film thickness control. Atomic-site precision belongs only to independently qualified processes. No universal photoresin, new catalyst, or arbitrary bond-by-bond reaction pathway has been supplied. + +**“Verification becomes impossible after a metal layer closes.”** Some targets are indeed unqualifiable in the proposed machine. Test ports, alternate assembly order, transparent covers, and functional redundancy may help only when permitted by the target. Otherwise the compiler must report unsupported quality or reject the route. + +**“The numerical work proves little.”** It proves consistency between several restricted implementations and analytic models, and exposes specific counterexamples. It does not validate a resin, a transfer process, a defect density, a cooling assembly, or technological universality. Those are separately enumerated experiments. + +**“The project name creates a false appearance of an established material or class.”** The release explicitly defines the name as a project designation. Veyrglass has no asserted chemical composition or extraordinary property. Universal Class is not an attained U5 level. + +# 21 Comparison with existing approaches and novelty search + +| Existing approach | Established contribution relevant here | What this release must add to justify a stronger claim | +|---|---|---| +| Conventional lithography and semiconductor process integration | Parallel patterning, material-specific routes, masks, metrology and yield engineering | Portable route/evidence representation across substantially different process domains | +| Computed axial lithography and xolography | Volumetric light-based production in qualified photochemical media | Explicit decision to leave a volumetric route when material or evidence constraints require another | +| Dose-band and inverse-rendering optimization | Optimized projections, local dose tolerances, optical heterogeneity | A calibrated infeasibility explanation that triggers a feasible manufacturing route change | +| Dual-colour initiation/inhibition | Chemical suppression of unwanted conversion and improved contrast | No new claim for optical Boolean logic; include realistic state and cumulative-damage contracts | +| DOLFIN and inorganic-film patterning | Light-responsive surface chemistry across selected inorganic families | Verified cartridge interoperability and preserved device properties after integration | +| Dry-film stereolithography plus laser transfer | Hybrid polymer/metal structural electronics | Quantified cross-domain selectivity and preservation gains over this strong baseline | +| Known-good-component assembly | Screening before expensive integration | Generalized structural, chemical and optical predicates with explicit post-test invalidation | +| Digital materials | Modular and hierarchical construction | Evidence for continuous/film/molecular process integration that goes beyond discrete modules | +| Chemputation | Programmable chemical operations, hardware abstraction, claimed synthesis generality | Spatially constrained solid-object compilation with explicit access and interface obligations | + +Kelly and colleagues demonstrated computed axial lithography using tomographic reconstruction.[^cal] Xolography supplies dual-colour photoswitchable initiation.[^xolo] Li and colleagues generalized projection optimization with dose-band constraints.[^band] Wechsler and colleagues explicitly address overprinting around optically complex pre-existing objects.[^overprint] These sources substantially overlap the light-printer objective. + +Binary photoinhibition already has a theoretical and experimental literature; the latest version of arXiv:2303.13941 has a narrower title and claims than its original version, and this release uses the current record.[^binary] A July 2026 ChemRxiv preprint and its authors' code implement antagonistic dual-wavelength TVAM.[^antagonistic] An August 2026 EngrXiv preprint is titled *Boolean Lithography for Volumetric Additive Manufacturing*.[^boolean] The latter records were checked at the accessible abstract/code-record level where full text was unavailable; their quantitative performance is not imported as validated VLWNC-IF-VF capability. + +The important novelty boundary is therefore narrow. The release proposes **jointly compiling optical controllability, physical access, and preservation**, with a finite implementation of a route decision and a new-to-this-release lamellar admissibility/cost calculation. It does not establish that no earlier work made the same combination. A patent, citation-network, and specialist literature review could further narrow or eliminate the candidate novelty. + +The targeted source review covers literature available through 13 September 2026. Several publisher and preprint full-text routes were unavailable; source records state the access level. No inaccessible paper is described as fully audited, and no absence of a search result is treated as evidence of originality. No external author's PDF or code is redistributed in this package. + +# 22 Experimental prototypes and benign benchmarks + +All dimensions, rates, precision ranges, energy values, and stage success criteria below are proposed engineering envelopes. None is measured performance of a built VLWNC-IF-VF system. Progression is by passed experimental gates, not a promised calendar. Instruments may initially be distributed on a bench before packaging them in a cabinet. + +## Prototype A: exposed-carrier selectivity and preservation experiment + +**Dimensions and mechanism.** A roughly $1.5\times1.0$ m optical/processing bench uses 25-50 mm glass carriers and $10\times10$ mm coupon regions. Pattern a photoimageable dielectric and a conductor route on exposed carriers, inspect them, and assemble a two-to-four-layer resistor/insulator test coupon. Compare direct irradiation through a partly obscuring stack with exposed-carrier processing under matched chemistry and target tolerances. + +**Feedstocks and actuators.** Use qualified glass, photoimageable polymer film, copper-compatible seed or a prepatterned metal donor, compatible support/release and bonding materials, and controlled rinse fluids. The conservative route uses optical resist patterning followed by ordinary electrodeposition or donor transfer. Actuation includes a DMD projection engine, encoded stages, a potentiostat where deposition is used, a force-limited transfer fixture, and temperature control. No invented photocatalyst is required. + +**Sensors and control.** Calibrate delivered irradiance and the on/off response on actual witness coupons. Measure optical geometry, film thickness on selected regions, electrical continuity/resistance, transfer registration, and interface condition. A workstation handles route selection and evidence records; local loops regulate motion, current and temperature. The reference compiler's synthetic matrices must be replaced by measured response intervals. + +**Throughput, tolerance, energy and cooling.** Initial targets are 10-25 micrometre pattern features, about 2 micrometre registration after transfer, and one accepted two-to-four-layer coupon in 2-6 h. Film thickness tolerances are qualified per material; an initial polymer layer may be 10-50 micrometres with a 5% thickness target, while a plated micrometre metal layer has a separately calibrated target. A proposed 1-3 kW average laboratory load implies 2-18 kWh per 2-6 h run before upstream services. Use a temperature-controlled platen and heat rejection matched to the measured load. + +**Failures and measurable success.** Expected failures are stray exposure, swelling, poor seed continuity, incomplete development, dirty release surfaces, bond voids, and stale evidence. First measure each response and uncertainty interval. Then test held-out patterns and blinded seeded defects; compare a competent conventional route with the compiler-selected route at equal outgoing-quality evidence. A useful pilot gate is fewer than half as many post-closure scrapped coupons without more than 25% higher total process time. This is a proposed effect threshold, not a powered statistical claim. Failure to preserve the first layers after the planned total exposure count is a stop condition. + +## Prototype B: the boxed multi-cartridge instrument + +**Dimensions and mechanism.** Package dry optics, wet cassette, carrier vestibule, and assembly/metrology bays in the $1.2\times0.8\times1.6$ m cabinet described in Section 13. Use 50 mm carriers, a $20\times20$ mm qualified field, and a nominal 20 mm height envelope. Add a second inorganic-film route and an independently fabricated passive photonic or sensor module. The optical engine remains shared; wet and dry domains remain separate. + +**Feedstocks, actuators, sensors and control.** Add oxide-coated donors or a qualified benign inorganic nanoparticle formulation with measured residue and consolidation behavior. Retain polymers, conductor routes, and glass. Add automated cartridge identification, fluid carryover assay, transfer-force sensing, spectroscopic/reflectometric film checks, and electrical/optical functional probes. A process contract is bound to a cartridge lot and calibration version. The compiler may reuse its representation across materials but cannot assume identical rate constants. + +**Performance and energy.** Target 4-20 process layers or modules per coupon, with 10-25 micrometre baseline features, 1-5 micrometre features only in a separately qualified small field, and 1-2 micrometre registration where demonstrated. A planning envelope of 2-8 h per accepted heterogeneous coupon is deliberately broad; metal growth, rinse and inspection are likely bottlenecks. A 2-5 kW instrument requires approximately 4-40 kWh over that time, plus allocated external services. Cooling and waste treatment are included as facility requirements, not hidden by the box. + +**Gate.** Reproduce the same interface/evidence representation on a resistor sensor, a polymer/glass microstructure and a transferred passive photonic coupon. Add a new cartridge without rewriting the target language or weakening test definitions. Compare against a well-engineered hybrid printing baseline. A successful result demonstrates restricted U2/U3 integration, not an atomic universal printer. + +## Prototype C: parallel fronts with shared metrology + +**Dimensions and mechanism.** A $2\times1.5\times2$ m pilot machine supports 16 carriers or active fields, with a transport carousel or equivalent qualified mechanism. Projection may illuminate multiple fields where irradiance, chemical isolation, and sensor bandwidth permit. Carrier preparation, exposure, rinse, and inspection are pipelined. Selected fine features remain serial or small-field operations. + +**Control and metrology.** Local loops stabilize each process bay; a supervisor coordinates optical, fluid, thermal, transfer, and metrology conflicts. Every carrier retains independent identity and history. Common calibration errors are handled at lot/instrument level. Inline optical/electrical tests are backed by destructive audit sampling of separate coupons. There is no assumption that one camera observes every buried interface. + +**Throughput and tolerances.** If one qualified route has 60 minutes of bottleneck service per field and useful utilization is 50%, sixteen independent fields would support at most eight accepted-equivalent field cycles per hour before yield, setup and shared-station limits. This is a sizing calculation. It is not sixteen times Prototype B throughput by definition. Precision targets remain those already qualified; packing more carriers does not improve resolution. + +**Energy and failures.** A 10-30 kW facility envelope implies 1.25-3.75 kWh per field cycle at eight cycles/h, excluding upstream content and any route work absent from the assumed cycle. Matching heat rejection, fluid capacity, and metrology service are required. Likely failures are shared contamination, thermal drift, transfer queues, correlated illumination errors and sensor saturation. A gate is a reproducible fourfold gain in accepted output over a one-field system at equal tolerances, with all energy, maintenance, and metrology costs reported. Failure of throughput to scale with added fields should trigger bottleneck redesign before further replication. + +## Prototype D: strongest defensible extended system + +**Scope and dimensions.** A modular instrument cluster, from several cabinets to approximately $10^2$-$10^3$ m$^2$ of specialized facilities, extends the same compiler/evidence interfaces to qualified oxide, semiconductor, polymer, metal and selected molecular processing. External high-temperature, vacuum and crystal-growth services are explicit. A single compact cabinet cannot contain every favorable synthesis environment, purification plant and metrology instrument economically. + +**Mechanisms and controls.** Add specialized self-limiting film deposition, selected epitaxial or surface-precision operations, more donor materials, and compatible module assembly. The common role of light is patterning/addressing/metrology where useful. Electrical, mechanical, thermal and chemical actuation remain indispensable. End products may be millimetre-to-centimetre heterogeneous devices with selected nanoscale features and larger packaging. + +**Performance.** No defensible universal rate or energy-per-object range exists across these materials. For a declared pilot workload, 0.1-1 MW and 1-100 accepted centimetre-scale devices per day are facility-sizing scenarios to replace with measured route budgets, not forecasts. They correspond to 24-24,000 kWh/device across the extrema and illustrate why an unspecified “universal printer” energy claim is meaningless. High precision and broad chemistry can make metrology and upstream energy dominant. + +**Gate and boundary.** Independent laboratories must compile held-out targets across several material families, preserve calibrated certificates through transfer, and reproduce better accepted-output resource use or genuinely broader qualified coverage than existing integrated workflows. The strongest plausible endpoint remains restricted U4 with selected atomic capabilities. U5 requires an independent target-class coverage theorem and experimental evidence absent here. + +## Benchmark ladder + +| Benchmark | New capability tested | Required evidence | +|---|---|---| +| Single-material exposure contrast coupon | Calibrated dose feasibility and forbidden-region protection | Dose-response curves, held-out geometry, uncertainty interval | +| Copper/dielectric resistor array | Different transformation mechanisms under one compiler | Geometry, thickness, resistance, insulation, residues | +| Repeated-exposure witness | Preservation across a full build history | Before/after spectroscopy or property test with cumulative dose recorded | +| Four-layer sensor coupon | Test-before-closure and invalidation tracking | Layer tests, post-cure interface tests, final function | +| Glass/polymer microchannel with removable support | Accessibility, support removal, release preservation | Open-channel inspection, residue tests, leakage and dimensional tests | +| Passive photonic component on patterned carrier | Heterogeneous integration | Optical loss/transfer function, alignment, imported-content record | +| MEMS-like compliant test element | Release, stress, and geometry | Mechanical response, deformation and lifetime tests | +| Selected crystal-surface occupancy pattern | Specialized atomic-site capability | Qualified site-resolved metrology on the accessible surface; no macro-atomic inference | + +# 23 Validation program + +## Reproducible computational layer + +Run `python code/reproduce.py` with the dependencies in `requirements.txt`. The script writes `data/results.json`, `data/lamellar_sweep.csv`, and `data/compiled_demo.json`. It uses seed 20260913 and performs nine meaningful checks. `code/make_figures.py` regenerates the figures from saved data. The supplementary document gives the equations, algorithms, uncertainty assumptions and counterexamples independently of the source code. + +The linear programs use SciPy/HiGHS. A supplied infeasibility example also has an exact rational certificate checked with Python fractions; this establishes the particular rational contradiction without trusting solver status. Other numerical optima are floating-point calculations checked against analytical structure or independent enumeration. No general formal theorem prover has been run. Monte Carlo results test observable moderate error rates, not unsupported probabilities of $10^{-10}$. + +## Required physical calibration + +For each material/process state, measure spectral irradiance, the optical point/line response on the actual carrier, conversion or deposition response, protected-region damage, induction periods, saturation, and changes during growth. Use held-out patterns, aged carriers, different lots and deliberately varied alignment. Fit a model only within its stated validity domain and keep residual diagnostics. A linear hazard matrix is accepted only where superposition and state invariance are justified. + +Measure reaction transport and chemical blur. A fine optical feature with a broad polymerization or deposition response fails the intended primitive. Electron/ion transport can dominate an optically addressed electrode; radicals, photoacids and solvents can spread after exposure. For each response, identify whether contrast loss is optical, chemical, electrical or mechanical before choosing a repair. + +Qualify joining and preservation as separate processes. Inspect modules, transfer them, perform the actual cure, and remeasure relevant properties. Include repeated full-build exposure histories, not just one successful lamination. Measure interface defects across lots and test whether a Poisson approximation or an independent-seam model is defensible. If not, replace the model with measured conditional or correlated risks. + +## Simulations still required + +Reaction networks need measured rates, mechanistic transition models, or electronic-structure calculations tied to actual candidate chemistry. Kinetic Monte Carlo can propagate calibrated surface events; it cannot supply a missing event catalogue. Coupled diffusion/reaction models must include precursor depletion and evolving solid boundaries. Electrodeposition requires current distribution and ionic transport, not just Faraday's mass balance. + +Thermal sign-off requires transient three-dimensional conduction, convection, interface resistance and deposition/optical heat sources, compared with embedded or surface temperature measurements. Mechanics needs shrinkage, cure stress, adhesion, carrier release and registration uncertainty. Optical modeling must update absorption/scattering and refractive boundaries during construction. The small Fourier and slab checks in this release are verification examples for simple equations, not substitutes for those simulations. + +## Comparative experiment and stopping rules + +Compare competent conventional process planning with the compiler's proposed route at equal target, imported content, precision and outgoing-quality evidence. Randomize carriers across runs, conceal seeded-defect labels from operators and controllers, and retain a held-out set of naturally occurring defects. Record total elapsed time, instrument energy, consumables, upstream allocation, rework, inspection time and scrap. A larger confirmatory trial should be sized from the pilot variance and a declared effect threshold. + +Stop or narrow the claim if response intervals are not stable; if preservation costs exceed the benefit of shorter optical paths; if added seams eliminate the feasible thickness interval; if independent tests find systematic blind defects; if carrier transfer destroys the required material continuity; or if cartridge additions require completely new target semantics. Such outcomes may still leave a useful specialized instrument, but they falsify the stronger universal-leaning architecture. + +The package is prepared for public expert review. It is not a preregistration in an external registry, and no experimental results are implied by a proposed trial design. + +# 24 Remaining open problems + +The main unresolved physical problem is a broad family of low-damage transformations and joins that preserve optical, electrical, mechanical and chemical function under repeated processing. A second is metrology: many buried states remain unobservable at the required false-acceptance rate. A third is the trustworthy transfer of process contracts across material lots, tools and histories. None can be resolved by a universal command name. + +The most promising theoretical extension is to replace scalar attenuation and independent seam models with robust state-dependent optical operators, correlated defect fields, and repairable interface networks. That would test whether the empty-window obstruction survives realistic route alternatives. A bounded-access graph model could identify targets that can be built with limited simultaneously exposed regions, but a small graph frontier alone does not guarantee geometric or chemical realizability. + +A genuine material-universality theorem would need a target family specified independently of successful routes and a constructive completeness proof with physical resource bounds. The current finite cartridge catalogue does not supply that theorem. Proving universality for a restricted lamellar circuit grammar may be possible under qualified local primitives, but assuming those primitives already fabricate every desired feature would make the proof circular. + +There may also be better architectures than carrier-based assembly for particular classes: direct epitaxial growth for coherent crystals, conventional lithography for large planar circuits, or volumetric printing for transparent polymers. The compiler should select these when their qualified contracts win. The objective is broad reliable manufacture, not preserving a favored physical layout. + +# 25 Precise novelty and breakthrough claims + +**Candidate contribution 1:** a common representation that couples robust optical/chemical dose feasibility to physical route changes and evidence preservation. A supplied finite demonstrator makes one such route change and inserts evidence dependencies. The generality and practical advantage of this integration remain hypotheses. + +**Candidate contribution 2:** the joint optical/interface interval (L2)-(L4), exact expected serial cost (L5), and strict convexity (L7) for a declared lamellar architecture. These are self-contained derivations with numerical checks. They are new relative to the preceding project release. Scientific priority relative to all manufacturing, optical optimization and reliability literature has not been established. Their scope is deliberately restricted. + +**Candidate contribution 3:** the explicit cumulative-exposure counterexample and preservation budget integrated into route selection. The underlying cumulative-hazard mathematics is standard. The engineering requirement is useful: changing from deep optical writing to exposed carriers only helps if the completed regions survive subsequent operations. + +**Established ingredients that are not claimed as new:** photolithography; transfer printing; nanocrystal ligand patterning; optical initiation/inhibition; dose-band optimization; Farkas certificates; union bounds; statistical testing bounds; graph pathwidth; Poisson yield models; convex optimization; modular chemistry; and known-good-component assembly. + +No experimental orders-of-magnitude improvement, new universal material primitive, new force, new quasiparticle, or general stable-matter coverage theorem is claimed. The mathematical package may support a publishable architectural study if specialists confirm the contribution and the physical experiment demonstrates meaningful advantage. It does not presently satisfy a major-breakthrough standard merely by being comprehensive or reproducible. + +The desired decisive result would be a measured route class that the integrated compiler enables at substantially better qualified output per total resource, or a materially stronger theorem with realistic correlated uncertainty and preservation. Those outcomes are precise targets for the next stage, not achievements recorded here. + +# 26 Claims that must not yet be made + +- That Universal Class means demonstrated U5 or arbitrary stable-matter synthesis. +- That a universal light-only chemical instruction set or a universal resin has been derived. +- That Veyrglass is a discovered material with special physical properties. +- That a cabinet makes arbitrary atomic patterns throughout macroscopic opaque objects. +- That inhibition is negative light energy, or that an optimization can deliver a field forbidden by its actual hardware. +- That the lamellar equations apply to every beam geometry, chemistry, seam type, defect process, or repair policy. +- That ideal screening assumptions or synthetic response matrices have been experimentally validated. +- That nine passing numerical checks validate the proposed machine, chemical pathways, or production rates. +- That a dose certificate proves a reaction model is correct or a data hash proves an object conforms. +- That purchased chips, crystals, or nanocrystal feedstocks were synthesized inside the cabinet. +- That the proposed tests, power ranges, yields, and tolerances are measured performance. +- That novelty, scientific priority, patent freedom, peer review, or a major breakthrough has been established. +- That general picosecond manufacture, black-hole processing chambers, unknown particles, or extra spatial dimensions are supported. + +# 27 Final assessment + +The closest defensible “universal light printer in a box” is a light-addressed, multi-cartridge fabrication system that keeps difficult transformations and critical inspection surfaces accessible, then joins qualified material into larger objects. Its potential breadth comes from chemistry-specific primitives and interface engineering. Light is a powerful common addressing layer; it does not replace feedstocks, reaction pathways, charge transport, thermal processing, or metrology. + +The principal extension over the previous research program is quantitative: an optical route can now return a checkable impossibility witness, and a lamellar route must satisfy a joint optical/interface feasibility window. A strictly convex cost model identifies the best permitted section scale under explicit assumptions. A cumulative-damage example shows why a successful exposure cannot be extrapolated to a successful multilayer build without preservation evidence. + +These results support an immediate falsifiable laboratory program. They do not establish the final universal nanofabricator or a demonstrated major breakthrough. The package is complete for public expert review of its architecture, derivations and numerical work. Whether the candidate integration principle merits a stronger scientific claim depends on independent prior-art assessment, realistic material calibration, and comparative experiments. + +# Sources + + +Sources support the stated antecedents and specific demonstrated mechanisms. PCF equations, synthetic examples, prototype envelopes, and judgments are derived or proposed here. Source numbering below is independent of automatically numbered footnote occurrences. + +1. L. Cronin, S. Pagel, and A. Sharma. **Chemputer and chemputation—A universal chemical compound synthesis machine.** *PNAS* 123(15), e2511080123 (2026). [Published paper](https://doi.org/10.1073/pnas.2511080123); [author-hosted full text](https://eprints.gla.ac.uk/382136/1/382136.pdf). Used for the closest conceptual prior art, not as proof of PCF universality. + +2. K. C. Cheung and N. Gershenfeld. **Reversibly Assembled Cellular Composite Materials.** *Science* 341, 1219-1221 (2013). [DOI](https://doi.org/10.1126/science.1240889); [author-hosted full text](https://cba.mit.edu/docs/papers/13.09.Science.pdf). Hierarchical/reversible discrete materials antecedent. + +3. R. Schulman, C. Wright, and E. Winfree. **Increasing Redundancy Exponentially Reduces Error Rates during Algorithmic Self-Assembly.** *ACS Nano* 9(6), 5760-5771 (2015). [DOI](https://doi.org/10.1021/nn507493s); [primary abstract record](https://pubmed.ncbi.nlm.nih.gov/25965580/). Experimental self-assembly error-correction antecedent. + +4. M. Rashidi and colleagues. **Deep Learning-Guided Surface Characterization for Autonomous Hydrogen Lithography.** Author manuscript, arXiv:1902.08818, originally submitted 2019. [Manuscript record](https://arxiv.org/abs/1902.08818). Specialized surface automation, not arbitrary matter construction. + +5. P. Klement and colleagues. **Surface Diffusion Control Enables Tailored Aspect Ratio Nanostructures in Area-Selective Atomic Layer Deposition.** Author manuscript, arXiv:2012.04465 (2020). [Manuscript and abstract](https://arxiv.org/abs/2012.04465). Selectivity, surface diffusion, and process limitations. + +6. H. J. Manetsch and colleagues. **A tweezer array with 6100 highly coherent atomic qubits.** *Nature* (2025), DOI 10.1038/s41586-025-09641-4; arXiv:2403.12021. [Published record](https://doi.org/10.1038/s41586-025-09641-4); [author manuscript](https://arxiv.org/abs/2403.12021). Parallel neutral-atom trapping precedent; no covalent fabrication claim inferred. + +7. Y. Yang and colleagues. **Deciphering chemical order/disorder and material properties at the single-atom level.** *Nature* 542, 75-79 (2017). [DOI](https://doi.org/10.1038/nature21042); [author manuscript](https://arxiv.org/abs/1607.02051). Specific nanoparticle atomic-coordinate reconstruction. + +8. A. Amo and colleagues. **Superfluidity of polaritons in semiconductor microcavities.** *Nature Physics* 5, 805-810 (2009). Author manuscript titled **Observation of Superfluidity of Polaritons in Semiconductor Microcavities.** [DOI](https://doi.org/10.1038/nphys1364); [manuscript](https://arxiv.org/abs/0812.2748). Real but host-specific light-matter-fluid analogue. + +9. B. Heuser and colleagues. **Release dynamics of nanodiamonds created by laser-driven shock-compression of polyethylene terephthalate.** *Scientific Reports* (2024). [Published record](https://doi.org/10.1038/s41598-024-62367-7). Extreme-processing analogue and relevance of release survival; no detailed performance numbers imported. + +10. J. Hong, B. Lambson, S. Dhuey, and J. Bokor. **Experimental verification of Landauer's principle in erasure of nanomagnetic memory bits.** Author manuscript, arXiv:1411.6730 (2014). [Manuscript](https://arxiv.org/abs/1411.6730). Information-erasure thermodynamics; not a fabrication-energy model. + +11. R. Chirico, S. Spellini, M. Panato, M. Lora, and F. Fummi. **A Contract-based Methodology for Production Lines Validation.** IEEE INDIN (2019). [DOI](https://doi.org/10.1109/INDIN41052.2019.8972100); [publisher record](https://ieeexplore.ieee.org/document/8972100). Retrieved abstract establishes assume-guarantee production-line precedent; full-text access restricted. + +12. Siemens. **Affordable and comprehensive design for test of 3D stacking die devices.** Technical white-paper landing page, accessed 13 September 2026. [Primary technical description](https://resources.sw.siemens.com/en-US/white-paper-affordable-and-comprehensive-testing-of-3d-stacked-die-devices/). Pre/post assembly and interconnect testing antecedent. + +13. X-Celeprint. **Micro-transfer printing technology and heterogeneous integration.** Manufacturer technical description, accessed 13 September 2026. [Technology description](https://x-celeprint.com/). Vendor-reported alignment and heterogeneous photonics capability, identified as such. + +14. N. R. Saxena, S. Hukerikar, M. Blaz, and S. Raj. **Optimal Checkpoint Interval with Availability as an Objective Function.** arXiv:2410.18124 (2024). [Author manuscript](https://arxiv.org/abs/2410.18124). Checkpoint optimization antecedent and reference to Young's earlier work. + +15. **Hierarchical functional digital materials.** US patent US9506485B2 (2016). [Patent text](https://patents.google.com/patent/US9506485B2/en). Prior art for hierarchical variable-size digital-material modules. No legal opinion or freedom-to-operate determination is made. + + +16. M. Regehly et al. **Xolography for linear volumetric 3D printing.** *Nature* 588, 620-624 (2020). [DOI](https://doi.org/10.1038/s41586-020-3029-7); [primary abstract](https://pubmed.ncbi.nlm.nih.gov/33361791/). Dual-colour photoinitiation precedent; abstract reviewed. + +17. C. C. Li, J. Toombs, H. K. Taylor and T. J. Wallin. **Tomographic projection optimization for volumetric additive manufacturing with general band constraint Lp-norm minimization.** arXiv:2312.01548v3 (2024). [Author manuscript record](https://arxiv.org/abs/2312.01548). Dose-band optimization and local tolerancing; current abstract reviewed, full-text retrieval unavailable. + +18. F. Wechsler et al. **Overprinting with Tomographic Volumetric Additive Manufacturing.** arXiv:2507.13842v4 (2026; first version 2025). [Current record](https://arxiv.org/abs/2507.13842); [related published DOI](https://doi.org/10.1038/s41467-026-73477-3). Optical occlusion and heterogeneous inserts; current abstract reviewed. + +19. Y. Wang, I. Fedin, H. Zhang and D. V. Talapin. **Direct optical lithography of functional inorganic nanomaterials.** *Science* 357, 385-388 (2017). [DOI](https://doi.org/10.1126/science.aan2958); [primary abstract](https://pubmed.ncbi.nlm.nih.gov/28751606/); [institutional full text](https://bdt.semi.ac.cn/library/upload/files/2017/8/2316412191.pdf). Light-responsive ligand chemistry; relevant full-text mechanism inspected. + +20. A. Levy et al. **Hybrid structural electronics printing by novel dry film stereolithography and laser induced forward transfer.** *Nano Select* 2, 979-991 (2021). [Published record](https://doi.org/10.1002/nano.202000269). Direct hybrid architecture antecedent; publisher record/abstract access. + +21. N. Liu et al. **Optically-controlled digital electrodeposition of thin-film metals for fabrication of nano-devices.** *Optical Materials Express* 5, 838-848 (2015). [DOI](https://doi.org/10.1364/OME.5.000838). Optically addressed metal-deposition antecedent; publisher indexed record, full text unavailable. + +22. B. Wang et al. **Lateral Contrast Enhancement in Tomographic Volumetric 3D-Printing via Binary Photoinhibition.** arXiv:2303.13941v3 (2025; first version 2023 under a different title). [Current author record](https://arxiv.org/abs/2303.13941). Binary photoinhibition and dose-subtraction precedent; current abstract reviewed. + +23. Q. Thijssen, F. Wechsler, A. J. Ortega, J. A. Carroll, C. Moser, S. Van Vlierberghe and C. Barner-Kowollik. **Antagonistic Dual-Wavelength Tomographic Volumetric Additive Manufacturing.** ChemRxiv (2026), version 1. [Preprint](https://doi.org/10.26434/chemrxiv.15006647/v1); [authors' code and description](https://github.com/EPFL-LAPD/Antagonistic-Dual-Wavelength-Tomographic-Volumetric-Additive-Manufacturing/blob/main/README.md). Accessible record and code description reviewed; no independent experimental validation inferred. + +24. C. He et al. **Boolean Lithography for Volumetric Additive Manufacturing.** EngrXiv (2026). [Primary preprint record](https://doi.org/10.31224/7874). Search-indexed primary abstract identifies initiation/inhibition AND-NOT control; full text unavailable. Used to limit novelty, not to import performance claims. + +25. A. Orth et al. **On-the-fly 3D metrology of volumetric additive manufacturing.** arXiv:2202.04644 (2022), related *Additive Manufacturing* 56, 102869. [Author abstract](https://arxiv.org/abs/2202.04644); [DOI](https://doi.org/10.1016/j.addma.2022.102869). Selected photopolymer shape metrology; abstract reviewed. + +26. B. E. Kelly et al. **Volumetric additive manufacturing via tomographic reconstruction.** *Science* 363, 1075-1079 (2019). [DOI](https://doi.org/10.1126/science.aau7114). Computed axial lithography antecedent; primary indexed record. + +27. N. G. Kinnersley. **The vertex separation number of a graph equals its path-width.** *Information Processing Letters* 42, 345-350 (1992). [DOI](https://doi.org/10.1016/0020-0190(92)90234-M). Established graph-theoretic identity; primary record and independent algorithm documentation checked. + + +[^chemputer]: Cronin, Pagel, and Sharma, [Chemputer and chemputation](https://eprints.gla.ac.uk/382136/1/382136.pdf), *PNAS* (2026). See Sources 1. +[^digital]: Cheung and Gershenfeld, [Reversibly Assembled Cellular Composite Materials](https://cba.mit.edu/docs/papers/13.09.Science.pdf), *Science* (2013). See Sources 2. +[^shock]: Heuser et al., [Release dynamics of nanodiamonds](https://doi.org/10.1038/s41598-024-62367-7), *Scientific Reports* (2024). See Sources 9. +[^polariton]: Amo et al., [Observation of Superfluidity of Polaritons in Semiconductor Microcavities](https://arxiv.org/abs/0812.2748), author manuscript (2008), published 2009. See Sources 8. +[^tweezer]: Manetsch et al., [A tweezer array with 6100 highly coherent atomic qubits](https://arxiv.org/abs/2403.12021), published in *Nature* (2025). See Sources 6. +[^kgd]: Siemens, [Affordable and comprehensive design for test of 3D stacking die devices](https://resources.sw.siemens.com/en-US/white-paper-affordable-and-comprehensive-testing-of-3d-stacked-die-devices/). See Sources 12. +[^contracts]: Chirico et al., [A Contract-based Methodology for Production Lines Validation](https://doi.org/10.1109/INDIN41052.2019.8972100), IEEE INDIN (2019). See Sources 11. +[^dna]: Schulman, Wright, and Winfree, [Increasing Redundancy Exponentially Reduces Error Rates during Algorithmic Self-Assembly](https://pubmed.ncbi.nlm.nih.gov/25965580/), *ACS Nano* (2015). See Sources 3. +[^checkpoint]: Saxena et al., [Optimal Checkpoint Interval with Availability as an Objective Function](https://arxiv.org/abs/2410.18124) (2024). See Sources 14. +[^asd]: Klement et al., [Surface Diffusion Control Enables Tailored Aspect Ratio Nanostructures in Area-Selective Atomic Layer Deposition](https://arxiv.org/abs/2012.04465) (2020). See Sources 5. +[^stm]: Rashidi et al., [Deep Learning-Guided Surface Characterization for Autonomous Hydrogen Lithography](https://arxiv.org/abs/1902.08818). See Sources 4. +[^aet]: Yang et al., [Deciphering chemical order/disorder and material properties at the single-atom level](https://arxiv.org/abs/1607.02051), *Nature* (2017). See Sources 7. +[^landauer]: Hong et al., [Experimental verification of Landauer's principle in erasure of nanomagnetic memory bits](https://arxiv.org/abs/1411.6730). See Sources 10. +[^transfer]: X-Celeprint, [Micro-transfer printing technology](https://x-celeprint.com/), manufacturer description. See Sources 13. +[^patent]: [Hierarchical functional digital materials, US9506485B2](https://patents.google.com/patent/US9506485B2/en) (2016). See Sources 15. + +[^xolo]: Regehly et al., [Xolography for linear volumetric 3D printing](https://doi.org/10.1038/s41586-020-3029-7), 2020. Sources 16. +[^band]: Li et al., [Tomographic projection optimization with general band constraints](https://arxiv.org/abs/2312.01548), current version 2024. Sources 17. +[^overprint]: Wechsler et al., [Overprinting with Tomographic Volumetric Additive Manufacturing](https://arxiv.org/abs/2507.13842), current version 2026. Sources 18. +[^dolfin]: Wang et al., [Direct optical lithography of functional inorganic nanomaterials](https://doi.org/10.1126/science.aan2958), 2017. Sources 19. +[^hybrid]: Levy et al., [Hybrid structural electronics printing](https://doi.org/10.1002/nano.202000269), 2021. Sources 20. +[^electrode]: Liu et al., [Optically-controlled digital electrodeposition of thin-film metals](https://doi.org/10.1364/OME.5.000838), 2015. Sources 21. +[^binary]: Wang et al., [Lateral Contrast Enhancement in Tomographic Volumetric 3D-Printing via Binary Photoinhibition](https://arxiv.org/abs/2303.13941), current version 2025. Sources 22. +[^antagonistic]: Thijssen et al., [Antagonistic Dual-Wavelength Tomographic Volumetric Additive Manufacturing](https://doi.org/10.26434/chemrxiv.15006647/v1), 2026. Sources 23. +[^boolean]: He et al., [Boolean Lithography for Volumetric Additive Manufacturing](https://doi.org/10.31224/7874), 2026 preprint. Sources 24. +[^metrology]: Orth et al., [On-the-fly 3D metrology of volumetric additive manufacturing](https://arxiv.org/abs/2202.04644), 2022. Sources 25. +[^cal]: Kelly et al., [Volumetric additive manufacturing via tomographic reconstruction](https://doi.org/10.1126/science.aau7114), 2019. Sources 26. +[^pathwidth]: Kinnersley, [The vertex separation number of a graph equals its path-width](https://doi.org/10.1016/0020-0190(92)90234-M), 1992. Sources 27.