FAUST: Protostellar Science, Formal Methods, DSP

• FAUST is a polysemous term referring to an ALMA protostellar survey, a MATLAB toolbox for uncountable-state Markov process abstraction, and a real-time audio DSP language.
• In astronomy, FAUST uses high-resolution ALMA observations to resolve chemical and kinematic structures in young protostars on scales around 50 au.
• In formal methods and DSP, FAUST² and Faust leverage abstraction algorithms and compiler techniques to optimize stochastic model verification and fixed-point signal processing.

Searching arXiv for FAUST-related papers to ground the article in current research. {"query":"FAUST arXiv ALMA Large Program FAUST² Faust compiler", "max_results": 10} In the literature summarized here, FAUST denotes several distinct technical enterprises rather than a single framework. The dominant usage is the ALMA Large Program “Fifty AU STudy of the chemistry in the disk/envelope system of Solar-like protostars”, which was designed to resolve the chemistry, kinematics, and physical structure of young protostellar systems on scales of roughly $50$ au. Separate usages denote FAUST$^2$, a MATLAB toolbox for formal abstractions of uncountable-state discrete-time Markov processes, and Faust, a purely functional, block-diagram-oriented language for real-time audio and DSP whose recent compiler work includes automatic fixed-point format determination and CLAP plugin generation (Codella et al., 2021, Soudjani et al., 2014, Herrou et al., 2024, Franchino et al., 11 Jun 2026).

1. FAUST as an ALMA Large Program in protostellar science

The ALMA FAUST program was introduced to reveal and quantify the variety of chemical composition of the envelope/disk system at scales of $\sim 50$ au in a sample of Class 0 and I protostars representative of the chemical diversity observed at larger scales. Its stated objectives were to disentangle the components of the $50$–$2000$ au envelope/disk system, characterize organic complexity in each component, probe the ionization structure, and measure molecular deuteration. The original program description specifies a sample of 13 Class 0/I protostars at distances $\le 250$ pc and luminosities $\le 25\,L_\odot$, observed with three frequency setups—$214$–$219$ GHz and $229$–$^2$0 GHz, $^2$1–$^2$2 GHz and $^2$3–$^2$4 GHz, and $^2$5–$^2$6 GHz and $^2$7–$^2$8 GHz—at $^2$9 km s$\sim 50$0 spectral resolution and $\sim 50$1–$\sim 50$2 mas angular resolution, using both the 12 m array and the 7 m ACA. One spectral window per setup was dedicated to continuum measurements from $\sim 50$3 mm to $\sim 50$4 mm (Codella et al., 2021).

The molecular strategy was explicitly partitioned into tracer groups. Envelope probes included c-C$\sim 50$5H$\sim 50$6 and CS; centrifugal-barrier probes included CH$\sim 50$7OH, SO, and SiO; disk probes included H$\sim 50$8CO, C$\sim 50$9O, and HC$50$0N. Organic-complexity diagnostics targeted CH$50$1OH, NH$50$2CHO, CH$50$3CHO, CH$50$4OCH$50$5, and HCOOCH$50$6. Ionization diagnostics used H$50$7CO$50$8, DCO$50$9, and N$2000$0H$2000$1, while deuteration diagnostics included c-C$2000$2HD, N$2000$3D$2000$4, HDCO, D$2000$5CO, and CH$2000$6DOH (Codella et al., 2021).

Later program-context descriptions retain the same $2000$7 au emphasis while broadening the characterization of the survey. One such summary defines FAUST as a program designed to resolve and characterize young Class 0/I disks and their infall envelopes on $2000$8 au scales, combining multi-frequency continuum and spectral-line mapping to study both chemistry and dust growth, and states that the sample includes $2000$9 Sun-like protostars (Yang et al., 21 Jun 2026). This suggests that FAUST functions both as a tightly specified observing program and as an evolving publication series centered on uniform, high-resolution, multi-scale studies of young Solar-like systems.

2. Chemical regimes, molecular complexity, and deuteration in the FAUST series

A central result of the astronomy FAUST papers is that chemically rich inner regions are not restricted to the earliest Class 0 phase. FAUST I reported a hot corino toward the Class I protostar L1551 IRS5, with methanol non-LTE analysis giving a gas temperature of $\le 250$0 K, density $\le 250$1 cm$\le 250$2, and an emitting radius of $\le 250$3 au. The methyl formate and ethanol relative abundances were found to be compatible with those measured in Class 0 hot corinos, and the study concluded that little chemical evolution from Class 0 to I hot corinos occurs (Bianchi et al., 2020).

FAUST also identifies sources in which hot-corino chemistry coexists with carbon-chain chemistry. In CB68, CH$\le 250$4OH, HCOOCH$\le 250$5, and CH$\le 250$6OCH$\le 250$7 were detected toward the protostar, with a CH$\le 250$8OH rotation temperature of $\le 250$9 K and an emitting region of $\le 25\,L_\odot$0 au, while c-C$\le 25\,L_\odot$1H$\le 25\,L_\odot$2 and CCH were detected on a $\le 25\,L_\odot$3 au scale. The source was therefore described as having a hybrid chemistry (Imai et al., 2022). In VLA 1623–2417 B, compact CH$\le 25\,L_\odot$4OH emission and methyl formate were detected; LVG analysis yielded a size of $\le 25\,L_\odot$5–$\le 25\,L_\odot$6 arcsec ($\le 25\,L_\odot$7–$\le 25\,L_\odot$8 au), $\le 25\,L_\odot$9–$214$0 cm$214$1, kinetic temperature $214$2 K, and volume density $214$3 cm$214$4, while the line profiles were interpreted in terms of a chemically enriched ring of radius $214$5 au close to the centrifugal barrier (2206.13339).

Several FAUST studies connect chemical enrichment to shocks and filamentary accretion. In [BHB2007] 11, more than 45 CH$214$6OCHO lines, 8 CH$214$7OCH$214$8 transitions, one H$214$9CCO transition, and four trans-HCOOH transitions were detected, and the compact iCOM emission was found to encompass both protostars while tending to align with the southern filament. The study concluded that the detected methanol and the other iCOMs are generated by shocked gas from incoming filaments streaming toward the two sources (Vastel et al., 2024). In Elias 29, SO rotational temperatures reached $219$0 K at the interaction point of the outflow and the southern ridge and $219$1 K within the southeastern outflow, while $219$2 remained $219$3–$219$4 K in the quiescent southern ridge; the warm condition was attributed to the nearby B-type star HD147889, whereas local hot spots were associated with outflow or jet interactions (Oya et al., 18 Jan 2025).

Deuteration is a second major axis of the FAUST chemical program. FAUST X measured H$219$5CO, HDCO, and D$219$6CO toward [BHB2007] 11, deriving $219$7–$219$8 cm$219$9, $229$0–$229$1 cm$229$2, and $229$3–$229$4 cm$229$5, corresponding to an average D/H ratio of $229$6–$229$7; the same work tentatively interpreted a second large-scale H$229$8CO feature as an asymmetric molecular outflow launched by a wide-angle disk wind (Evans et al., 2023). FAUST XVII reported the first detection of deuterated formaldehyde in a planet-forming disk, IRS 63, with $229$9–$^2$00 and $^2$01; notably, D$^2$02CO was strongly asymmetric and peaked where the streamer strikes the disk, and it was also detected in two outflow spots, leading to an interpretation in which HDCO is dominated by gas-phase formation while D$^2$03CO is mainly grain-mantle material released by shocks (Podio et al., 2024). FAUST-XXII extended this approach to VLA 1623–2417, finding $^2$04 K toward the hot corino, $^2$05–$^2$06 K in the outflow cavities, and $^2$07 K in the streamers, with $^2$08 in the hot corino, $^2$09–$^2$10 in the outflow cavities, and $^2$11 in a streamer. The similar deuteration values across components were interpreted as evidence that prestellar deuteration is inherited mostly unaltered into the protostellar phase (Mercimek et al., 25 Feb 2025).

Taken together, these results show that FAUST chemistry is not limited to cataloging line detections. It uses spatially resolved abundances, isotopologue ratios, and excitation diagnostics to distinguish hot corinos, warm carbon-chain chemistry, shock-enhanced chemistry, grain-surface inheritance, and gas-phase reprocessing on scales from tens to thousands of au.

3. Disk-envelope coupling, streamers, outflows, and multiplicity

FAUST astronomy also emphasizes kinematic continuity and discontinuity across scales. In VLA 1623AB, the envelope rotation axis traced by H$^2$12CO$^2$13 was measured as $^2$14, whereas the circum-binary disk minor axis of VLA 1623A was $^2$15, implying a misalignment of about $^2$16. The outflow velocity gradient seen in CCH is perpendicular to the outflow axis and agrees with the envelope rotation axis rather than the disk minor axis; under the assumption of outflow rotation with constant specific angular momentum $^2$17 au km s$^2$18, the launching radius was estimated as $^2$19–$^2$20 au. The same study detected, for the first time, a velocity gradient associated with rotation toward the VLA 1623B disk, with a sense opposite to that of the large-scale envelope, outflow, and circum-binary disk (Ohashi et al., 2022).

A subsequent FAUST study of the same protocluster detected an accelerating SO streamer plausibly feeding VLA 1623B. The streamer extends over $^2$21–$^2$22 au, has $^2$23 K, $^2$24 cm$^2$25, $^2$26, a total mass of $^2$27, and an accretion rate of $^2$28–$^2$29. That rate is close to the inferred mass accretion rate of VLA 1623B itself, which the study used to argue that asymmetric streamer infall is an important contributor to protostellar disk growth. The same work reported the first SiO detection toward VLA 1623–2417, identifying a compact $^2$30–$^2$31 au jet around source B and shock-heated high-velocity SO with $^2$32 K and $^2$33 K in the red and blue components (Codella et al., 2024).

FAUST has also been applied to binaries with circumbinary structure and entangled outflow morphology. In L1551 IRS 5, continuum and C$^2$34O analysis yielded a circumbinary disk mass of $^2$35, northern and southern circumstellar disk masses of $^2$36 and $^2$37, a centrifugal barrier radius $^2$38 au, and a specific angular momentum $^2$39 au km s$^2$40. The line-of-sight velocity field required an infalling-rotating envelope rather than a pure Keplerian curve, and an analytic jet-driven outflow model fit the broad X-shaped C$^2$41O cavities with density power-law index $^2$42 and envelope rotation velocity $^2$43 km s$^2$44 (Durán et al., 12 Jun 2025).

A related L1551 IRS 5 study resolved two continuum over-densities at the edge of the circumbinary cavity, with the northern feature being about 20% brighter than the southern one. Using C$^2$45O kinematics and previous astrometry, it derived $^2$46, semimajor axis $^2$47 au, eccentricity $^2$48, inclination $^2$49, longitude of ascending node $^2$50, and argument of periastron $^2$51. Three-dimensional gas-dust SPH simulations with Phantom, followed by MCFOST post-processing, reproduced the brightness contrast and the S-shaped velocity twist, supporting a binary–disc interaction origin for the dust concentration (Cuello et al., 11 Dec 2025).

Another FAUST development is methodological rather than purely descriptive. FAUST XXIX proposed a luminosity estimator for embedded protobinaries based on the sublimation radius of OCS, aided by quantum-mechanical calculations of the OCS binding-energy distribution. Applied to NGC 1333 IRAS 4A, the method yielded luminosities of $^2$52 for A1 and $^2$53 for A2 (Saury et al., 4 Dec 2025). In GSS 30, FAUST multi-wavelength continuum analysis showed that the spectral index in IRS3 increases radially from $^2$54 at the center to $^2$55 at the disk edge while dropping to $^2$56–$^2$57 along the outflow direction; SED fitting gave $^2$58 and a dust mass of $^2$59–$^2$60. The same paper reported millimetre variability in IRS2, with the $^2$61 mm flux decreasing from $^2$62 mJy to $^2$63 mJy over $^2$64 s, consistent with a magnetic flare (Yang et al., 21 Jun 2026).

Paper	System	Key result
(Ohashi et al., 2022)	VLA 1623AB	$^2$65 disk-envelope misalignment
(Codella et al., 2024)	VLA 1623–2417	$^2$66 au streamer and compact SiO jet
(Durán et al., 12 Jun 2025)	L1551 IRS 5	$^2$67 au, $^2$68 au km s$^2$69
(Cuello et al., 11 Dec 2025)	L1551 IRS 5	$^2$70 dust over-density contrast from binary–disc interaction
(Saury et al., 4 Dec 2025)	NGC 1333 IRAS 4A	OCS-based luminosities for A1 and A2
(Yang et al., 21 Jun 2026)	GSS 30	Radial $^2$71 structure and rapid mm variability

These studies show that FAUST treats chemistry and dynamics as a coupled problem. Misalignments, counter-rotation, streamer impact zones, cavity walls, and circumbinary pressure maxima are not secondary complications; they are the structures within which the observed chemistry is organized.

4. FAUST$^2$72: formal abstractions of uncountable-state stochastic processes

Outside astronomy, FAUST$^2$73 is a formal-methods software tool for generating abstractions of possibly non-deterministic discrete-time Markov processes defined over uncountable state spaces. A dtMP model $^2$74 is specified in MATLAB and abstracted as a finite-state Markov chain or Markov decision process, with the abstraction procedure formally related to the concrete model through a user-defined maximum threshold on the approximation error. The toolbox can export abstract models to PRISM or MRMC, or can compute PCTL properties internally and refine the results back on the concrete dtMP using quantified error bounds (Soudjani et al., 2014).

Its architecture is divided into four components: a Model Definition Module, an Abstraction Engine, an Error Analysis Module, and an Output & Verification Interface. State and input spaces are given as box-shaped bounds in $^2$75 and $^2$76, and the stochastic kernel can be supplied by a linear-Gaussian template, a nonlinear-Gaussian template, or a user-defined density function. The GUI exposes “Formula-free,” “PCTL Safety,” and “PCTL Reach-Avoid” modes, as well as gridding assumptions and error controls (Soudjani et al., 2014).

The abstraction itself is partition based. FAUST$^2$77 chooses measurable cells $^2$78, assigns representative points $^2$79, and defines transition probabilities by integrating the concrete kernel over target cells: $^2$80 For controlled systems, the same construction yields an MDP kernel $^2$81. The implementation relies on MATLAB’s vectorized integral routine, or quad in older releases, and can parallelize these computations if the Parallel Computing Toolbox is available (Soudjani et al., 2014).

A distinguishing feature of FAUST$^2$82 is explicit error quantification. Under a global Lipschitz assumption on the transition density, the horizon-$^2$83 abstraction error satisfies

$^2$84

where $^2$85 is the partition diameter and $^2$86 is the user-supplied tolerance. For safety and reach-avoid properties, FAUST$^2$87 also supports a local Lipschitz, formula-dependent bound that drives adaptive sequential refinement of cells with large local errors (Soudjani et al., 2014).

The verification interface supports two workflows. One exports the abstract MC or MDP and labeling map to PRISM or MRMC. The other evaluates bounded-until PCTL formulas internally by dynamic programming on the abstract graph and then returns a guaranteed interval enclosure for the concrete probability. In the controlled case, the toolbox additionally solves a max or min over inputs to synthesize an $^2$88-optimal policy $^2$89 (Soudjani et al., 2014).

The reported case studies illustrate the intended operating regime. A 2D room-temperature control example with safe set $^2$90, horizon $^2$91, and $^2$92 produced about $^2$93 states in under $^2$94 s, with maximum one-step error about $^2$95 and total safety error about $^2$96. A 3D two-room extension using adaptive gridding produced about $^2$97 cells in under $^2$98 s (Soudjani et al., 2014). In this context, FAUST$^2$99 occupies a specific position between stochastic control, probabilistic verification, and numerical abstraction of continuous-state systems.

5. Faust in real-time audio DSP, fixed-point inference, and CLAP compilation

In audio and DSP research, Faust is a purely functional, block-diagram-oriented language for real-time audio and DSP. Faust programs are parameterized over the abstract type FAUSTFLOAT, which typically maps to IEEE-754 32-bit float in generated C++ or LLVM back ends. Recent work has focused on extending the compiler toward hardware-efficient fixed-point realizations and modern plugin targets (Herrou et al., 2024).

The fixed-point format work augments the compiler’s symbolic propagation pass with interval analysis and precision inference. For each signal $\sim 50$00, the compiler computes an interval $\sim 50$01 and chooses

$\sim 50$02

to prevent overflow. The least-significant-bit position $\sim 50$03 is then chosen using a pseudo-injectivity criterion intended to preserve distinctness through quantized operators. In the generated code, signal types become Ap_fixed<w_s,m_s> with $\sim 50$04, and casts are inserted at core operations (Herrou et al., 2024).

The numerical motivation is practical. The paper reports that on a mid-range FPGA a single-precision floating-point adder uses about 313 LUTs and 11.4 ns of combinational delay, whereas a 24-bit two’s-complement integer adder uses 24 LUTs and 1.7 ns. Preliminary audio tests yielded SNR values of 32 dB for a sine generator with step $\sim 50$05, 25 dB for a sine generator with step $\sim 50$06, and 33 dB for Karplus–Strong (Herrou et al., 2024). The work therefore treats fixed-point inference not as a manual back-end tweak but as a compiler-level semantic analysis problem.

A separate compiler-development line is faust2clap, described as the first officially maintained compilation pathway from Faust DSP specifications to the CLAP format. The framework has two modes. A static mode performs ahead-of-time compilation to native binaries through faust2clap.py, clap-arch.cpp, and a CMake build. A dynamic mode uses the Faust interpreter backend and a filesystem watcher to hot-reload DSP code without restarting the host application (Franchino et al., 11 Jun 2026).

The major algorithmic issue in faust2clap is parameter identity under structural mutations of the DSP graph. The framework therefore uses an address-based identity matching algorithm to preserve parameter values across reloads and a stable slot allocation scheme to preserve host automation bindings. The implementation comprises approximately 2,400 lines of C++ architecture and plugin code plus approximately 200 lines of Python, and has been integrated into the main Faust distribution (Franchino et al., 11 Jun 2026).

The reported performance figures place the dynamic mode well within interactive use. Interpreter overhead per 256-sample block at 48 kHz ranges from 0.008 ms for a resonant low-pass filter to 0.27 ms for dm.zita reverb, all below the 5.33 ms real-time deadline. Reload latency scales with DSP size but remains $\sim 50$07 ms, while compilation during hot reload typically takes 5–60 ms (Franchino et al., 11 Jun 2026). In this literature, Faust is thus simultaneously a language, a compiler infrastructure, and a target-adaptation framework for embedded, FPGA, and plugin workflows.

6. Disambiguation, common patterns, and scholarly significance

The term FAUST is therefore genuinely polysemous across current technical literature. In astronomy it refers to an ALMA large program on the chemistry and dynamics of Solar-like protostars (Codella et al., 2021). In formal methods it denotes a MATLAB toolbox for abstraction and verification of uncountable-state stochastic processes (Soudjani et al., 2014). In audio DSP, stylistically often written Faust, it denotes a language and compiler ecosystem for real-time signal processing, with current research on automatic fixed-point formats and CLAP plugin generation (Herrou et al., 2024, Franchino et al., 11 Jun 2026).

The domains are unrelated in subject matter, but a common structural pattern is visible. ALMA FAUST translates multi-line observations into a coherent view from envelope to disk and outflow scales; FAUST$\sim 50$08 translates continuous-state stochastic dynamics into finite probabilistic models with quantified error; Faust compiler research translates high-level DSP graphs into fixed-point circuits or CLAP plugins. This suggests that the shared name recurs in contexts concerned with representation change across scales or abstractions, even though the scientific objects—protostars, dtMPs, and audio signal graphs—are entirely different.

For scholarly use, explicit disambiguation is essential. Citations such as FAUST I, FAUST XVII, or FAUST XXXI belong to the ALMA protostellar survey; FAUST$\sim 50$09 belongs to stochastic-process abstraction and probabilistic model checking; and Faust in compiler papers denotes the audio DSP language. Treated separately, each of these literatures is technically coherent. Treated together under a single bare acronym, they require context, arXiv identifier, or subtitle to avoid category errors.