App execution and arbitrary-precision coverage¶
This inventory separates concurrent execution from arbitrary-precision support.
All numerical apps offer native precision choices, subject to the build; run
bethe-info for configured formats and providers. mpreal means explicit
bits-N/digits-N precision, not a promise that a chosen mesh or iteration
budget reaches that accuracy. See precision controls
and Async ownership.
The table records implemented execution paths and numerical limits. An app can
remain serial when its work is too cheap to benefit from scheduling; independent
parameter sweeps can still use the generic library grid. All 45 numerical
frontends below expose explicit MPFR when that scalar family is configured.
Names omit the bethe- executable prefix. This inventory describes the current
source tree. Rebuild an older
installation to obtain these capabilities. Linked PRs preserve the original
precision/provider configurations and independent numerical or serial/concurrent
validation; their measurements are historical records, not new integration runs.
| Apps | Existing execution path | MPFR frontend support | Scope and numerical limits |
|---|---|---|---|
xxx-pbc, xxx-obc |
Continuous real-root scans; owning Async scan API | Yes, MPFR scalars | Owning options survive through final consumers; real-root families only |
xxx-structure-factor |
Async preparation, fused root/form-factor runners, ordered reduction | Yes, MPFR scalars; CPU LU available | Two-spinon family; failed references and factors remain explicit |
lieb-liniger-pbc, lieb-liniger-obc |
Shared continuous real-root scanner; owning Async scans (#17) | Yes, MPFR scalars; CPU LU available | Finite label windows; independent requests can use different precisions |
richardson, central-spin |
Individual solver calls; owning grid-sweep example and full-state composition checks (#17) | Yes, MPFR scalars | Continuation within each solve remains sequential |
xxx-structure-factor-thermo |
Bounded ordered two-spinon grid; shared Async sector, node and replicate work | Yes, scalar MPFR | Two spinons by default; selected 2/4/6 partial sums with separate quadrature, sampling, contour, arithmetic and cutoff diagnostics |
xxx-moments-thermo |
Shared Async sector queries; bounded node and replicate work | Yes, scalar MPFR | Direct fixed-momentum/global M0/M1; physical channel/exchange conversion and optional six-spinon covariance; finite domains require refinement |
lieb-liniger-dispersion |
Prepared solver and bounded grids; shared Async backgrounds | Yes, MPFR scalars; CPU LU available (#9) | Arithmetic precision does not replace mesh, cutoff or inversion convergence |
lieb-liniger-thermal |
Bounded independent CLI temperature solves; owning mixed-precision library grids | MPFR scalars, grand-canonical and fixed-density equilibrium | Mesh, rapidity cutoff and density inversion budgets remain independent of arithmetic precision |
hubbard-dispersion |
Shared Async preparation and one branch-spanning query window (#14) | Yes, MPFR scalars; doped backgrounds use CPU/provider LU | Explicit quadrature, stable gamma-ratio phase and shared immutable kernel |
hubbard-continuum |
Bounded independent extrema queries (#12) | Yes, no MPLAPACK required | Explicit constituent and extrema precision; independent work budgets |
xxz-pbc, xxz-obc |
Continuous real-root scans; owning parameter sweeps (#15) | Yes, MPFR scalars; CPU LU available | Ground/excitation domains are unchanged by precision or concurrency |
q-boson-pbc |
Continuous real-root scans; owning mixed-precision queries (#25) | Yes, MPFR scalars; CPU LU available | Ground and canonical states retain full zero-update/failure diagnostics |
su3-pbc |
Individual nested solves; owning mixed-precision sweeps (#13) | Yes, MPFR scalars; CPU LU available | Balanced singlets retain their existing physical scope |
tb-pbc |
Single-request CLI stays serial; owning mixed-precision Async size sweeps compare complete states (#37) | MPFR scalars, including finite string deviations and owning complex rapidities | Even-ring singlet ground-state branch; the residual tolerance is not an energy-error bound |
hubbard-pbc, hubbard-obc |
Individual finite-size solves; owning parameter sweeps (#16) | Yes, MPFR scalars; CPU LU available | Supported sectors, both interaction signs and symmetry mappings |
gaudin-yang-pbc |
Individual nested solves; owning mixed-precision sector sweeps (#18) | Yes, MPFR scalars; CPU LU available | Restricted repulsive shells; failed continuation can stop below requested coupling |
sun-fermions-pbc |
Single-request CLI stays serial; owning mixed-precision population/coupling grids preserve complete states (#38) | Yes, MPFR scalars; CPU LU available | Existing odd-population interacting branch, free limits and partial accepted stages; independent two-body and boson-reduction checks |
tj-pbc |
Single-request CLI stays serial; owning one/two-worker mixed-precision grids preserve complete states | Native and explicit MPFR; CPU LU available | Existing doped parity restrictions remain; higher precision does not resolve the known large dilute-sector seed/Newton limitation |
ladder-pbc |
Owning prepared branch catalogue with dependent Async field/rung queries; complete mixed-precision results checked against direct scans | Native and explicit MPFR; CPU LU available | Fixed-Sz/singlet scans retain their direct paths, physical branch scope and whole-scan budgets; one precision per preparation |
bose-fermi-pbc |
Individual nested solves; owning mixed-precision requests (#19) | Yes, MPFR scalars; CPU LU available | Restricted mixed shells; zero-update diagnostics remain explicit |
su3-dispersion |
Serial analytic CLI; owning mixed-precision Async queries (#20) | Yes, MPFR scalars | Individual formulas remain too cheap to justify per-point scheduling |
haldane-shastry-pbc, sutherland-pbc |
Serial analytic CLIs; owning mixed-precision window/parameter scans (#21) | Yes, MPFR scalars | Gap/order labels remain exact integers; cheap formulas stay serial |
tb-dispersion |
Serial analytic CLI; owning mixed-precision Async queries (#22) | Yes, MPFR scalars | Analytic spinon kinematics; no operator-dependent spectral weights |
sine-gordon-dispersion |
Serial analytic CLI; owning mixed-precision Async queries (#23) | Yes, MPFR scalars | Particle kinematics are distinct from scattering and finite-volume numerical convergence |
xyz-dispersion |
Serial CLI; owning bound branches and shared Async preparation / parameter curves (#24) | Explicit MPFR scalars | Shared Uni20 complex theta jets; owning mixed-precision bound readers; no MPLAPACK required |
xxz-structure-factor-thermo |
Bounded ordered Async momentum/frequency grid | Explicit scalar MPFR; no MPLAPACK required | Massive transverse two-spinon contribution; both sheets, endpoint diagnostics and independent kernel/arithmetic controls |
xxz-dispersion |
Immutable band with cheap queries; bulk preparation remains serial | Explicit scalar MPFR | Scaled theta constants, finite quadrature and analytic limits; no MPLAPACK required |
xxz-qg-obc |
Continuous regular-root scans; owning regular/free-sector queries (#26) | Yes, MPFR scalars | Critical regular-root/free-endpoint scope; cheap free blocks stay serial |
sine-gordon-bethe-yang |
Bounded independent level scans (#27) | Explicit MPFR scalars | No MPLAPACK; Fourier, iteration and omitted-wrapping limits remain separate |
biquadratic-obc |
Continuous real-root and owned bound-cluster scans; independent AF sector minima (#35) | Yes, all existing modes with MPFR scalars; CPU LU available | Restricted candidate families, ranking and explicit failure diagnostics; selected states and Q-system/singlet paths stay direct |
lee-yang-vacuum, lee-yang-excited |
Owning mixed-precision parameter sweeps; optional vacuum/excited overlap | Yes, MPFR scalars | Nonlinear, mesh, cutoff and work convergence are separate from arithmetic bits |
sine-gordon-vacuum |
Serial single-scale CLI; owning mixed-precision Async parameter sweeps | Explicit MPFR/MPC; no MPLAPACK required | Independent kernel, nonlinear, mesh, cutoff and contour controls; more bits can require larger work budgets |
sine-gordon-excited |
Owning mixed-precision parameter sweeps; optional independent vacuum/excited overlap | Explicit MPFR/MPC; no MPLAPACK required | Selected p>=1, I=1/2 or 3/2 family; independent Fourier, hole, nonlinear and refinement budgets; coupled iterations remain sequential |
kondo-response |
Bounded field scans and owning Async queries (#29); single-point/default one-worker requests stay direct | Yes, MPFR scalars; no MPLAPACK required | Precision-aware log-Gamma amplitude, Euler-series tails and endpoint-aware quadrature; independent work budgets and explicit failure diagnostics |
potts-pbc |
Bounded charged-momentum scans and owning mixed-precision levels after one vacuum solve | Yes, MPFR scalars; no MPLAPACK required | Explicit constants, twists, auxiliary roots and complete failure diagnostics; vacuum and charged one-hole family only |
asep-pbc |
Serial CLI; owning mixed-precision shared TASEP seeds and dependent Async rate grids | Explicit MPFR/MPC; CPU/provider real LU | Existing continuation and work budgets; tiny-bias drift remains explicit |
tasep-pbc |
Serial CLI; owning mixed-precision size/filling requests | Explicit MPFR/MPC; CPU/provider real LU | Existing gap branch and work budgets; no full-spectrum claim |
xyz-pbc |
Serial CLI; owning mixed-precision size/eta/t sweeps | Explicit MPFR; CPU/provider LU | Shared complex theta jets, precision-aware roots, Newton controls and failure results |
bethe-info reports capabilities and has no numerical parallelism to add.
The inventory covers 45 numerical frontends; it does not extend any model's
physical state family. Generic scalar functions and numerical primitives belong
in Uni20; model equations and physical normalization remain in Bethe.
Validation records¶
Historical precision and execution checks retain the feature-checkpoint comparisons, provider configurations and test counts. See testing and CI for the current validation workflow and benchmarks for measured workloads.