Skip to contents

Changelog

newmark 1.6.0

Breaking changes

  • The JB07 kill-switch is removed (owner ruling 2026-08-07): family aggregates (ensemble, rigid, flexible) now honor exactly the declared weights. JB07 = 1 includes Jibson (2007) in its families; JB07 = 0 excludes it, like every other model. Projects that declare JB07 = 0 (all current production data.R) are unaffected.

newmark 1.5.0

Breaking changes

Bug fixes

  • The strong-demand end of the Dn(ky) inversion (target below the entire drawn curve) now inverts each model’s own closed equation instead of extrapolating log-log linearly through the last two grid points: an exact quadratic in ln ky for BT07/BM17/BM19, and the exact r = ky/PGA.s shape for AM88/JB07/SR08 with the rigid-block domain ending at r = 0.9999 (SR08 recovers its amplitude-dependent epsilon first; JB07 remains an exact unbounded power law). Linear extrapolation over SR08’s structural plateau had produced e17-class roots from high sampled demands, dominating published means.

New features

  • sampleSaCorr(), fitDnCurve(), and the fitDnScenario() row contract accept zTrunc: symmetric renormalized truncation of each Gaussian-copula marginal at ±zTrunc, mirroring the hazard calculation’s truncation_level, with recentering to the truncated mixture mean. Analytic mixture input only; quantile input is bounded by its own endpoints and rejects the argument.

newmark 1.4.1

Breaking changes

  • The analytic identifiers mu_Ln and sigma_Ln are renamed muLn and sigmaLn everywhere (function returns, mixture component columns, parameters, and documentation). The underscore spelling violated the project naming contract; no alias is kept.

  • The Dn-to-kmax inversion adopts a declared uniform boundary policy: displacements below the one-micron physical floor are censored to zero, a target above a realisation’s physical ceiling saturates at kmax = 0 (null demand, realisation kept in the pool) instead of log-linear tail extrapolation through underflow, and a crossing into the floored tail resolves to the first floored ky. Flat all-zero curves no longer abort the inversion; the exact flat-AM88 producer-limit inverse is preserved. Log-scale consumers must filter kmax > 0.

  • Scenario runners surface worker errors directly: a failed mclapply worker now stops with the original diagnostic instead of a downstream rbindlist type error.

newmark 1.4.0

Breaking changes

  • fitSaLognormal() replaces OpenQuake quantile families with a period-wise analytic lognormal representation. It preserves every reported mean and fits sigma_Ln(Tn) independently at each period with a mean-anchored Huber loss.

  • fitSaFMixture() now consumes and returns analytic components named mu_Ln and sigma_Ln. It no longer materializes report quantiles; the returned finite mixture is the canonical input for downstream composition and sampling.

  • sampleSaCorr(), getDnKy(), and the existing fitDn*() scenario APIs now consume those analytic mixtures directly. Correlated sampling inverts each mixture CDF at the Gaussian-copula probabilities and preserves its exact mean without reconstructing report fractiles or a quantile spline.

  • Quantile-table inputs remain a distinct contract for spectra whose distribution is already represented by quantiles, including the current MCE envelope workflow. They are not an alternate OpenQuake calibration path: OpenQuake fractiles are consumed by fitSaLognormal() and discarded.

newmark 1.3.9

Numerical changes

  • Dn realization tables now retain the sampled PGA as PGA.s. When an AM88 realization is flat because its sampled PGA lies below the stored ky grid, kmax recovers the unique root on the physical branch 0 < ky / PGA.s < 0.9999. Other curves retain the existing tabular inversion, and flat curves from other models still fail explicitly.

newmark 1.3.8

Behaviour changes

  • The public kmax APIs now expose one estimator only: epistemically weighted aggregation of the per-model root distributions. The transitional aggregationMethod selector and the historical root-averaging implementation were removed. The legacy implementation remains reproducible from commit 8538e78.

  • fitSaF() now declares the scalar default integrationMethod = "monte-carlo" and validates the accepted choices separately. Omitted calls retain the same Monte Carlo behaviour without depending on the order of a default vector.

newmark 1.3.7

Behaviour changes

  • The three public kmax APIs now declare the scalar default aggregationMethod = "model-mixture". Omitting the argument therefore uses the approved epistemic mixture. The historical "root-mean" estimator remains available only by explicit selection for legacy reproduction. Accepted values are validated separately from the default and no longer depend on the order of a default vector.

newmark 1.3.6

New features

  • invertDnDraws(), fitKmaxScenario() and fitKmaxScenarios() add aggregationMethod. The compatible default, "root-mean", preserves the historical estimator that averages model roots within each shared realisation. "model-mixture" instead treats the model root distributions as epistemic branches and applies the same weighted-draw aggregation used for displacement curves.

Performance

  • The kmax scenario path now inverts each (IDn, s) displacement curve once and reuses that ephemeral root population for individual models and all aggregate families. It no longer repeats the numerical inversion for each family.

Numerical changes

  • Quantile-mode fitSaF(integrationMethod = "quadrature") now integrates the two Gaussian-copula variables after their exact transformation to uniform probability space, using tensor Gauss–Legendre quadrature. This preserves the reconstructed marginals, copula, mean recentring, site-response model and tail policy while avoiding the oscillatory Gauss–Hermite convergence observed for steep upper quantile segments. The Monte Carlo path and fitSaFMixture() Gauss–Hermite integration are unchanged.

newmark 1.3.5

Behaviour changes

  • The standard kmax inversion family now includes the median (p = 0.50) together with 0.05, 0.10, 0.16, 0.84, 0.90, 0.95 and mean. Both invertDnDraws() and the faster scenario path use the same family. The inversion still operates on every simulated Dn(ky) curve before summarising; no quantile is copied from a pre-summarised displacement table.

newmark 1.3.4

New features

  • fitSaF() adds an opt-in integrationMethod = "quadrature" backend for quantile-defined probabilistic spectra. It integrates the PGA–Sa Gaussian copula by tensor Gauss–Hermite quadrature and the conditional site-response residual analytically, so repeated calls are independent of the random number state. Before mean re-centering, the quantile interpolant passes through the supplied knots and retains their Hyman interpolation; unobserved tails use explicit non-decreasing linear continuation instead of the folded cubic extrapolation possible in buildQSpline(). Marginals are then re-centered to the published means. integrationMethod = "monte-carlo" remains the default and is unchanged draw-for-draw.

newmark 1.3.3

New features

  • getDesignParameters() returns the ASCE 7-22 Sec. 21.4 design acceleration parameters SDS, SD1, SMS and SM1 from the ordinates of a site-specific spectrum. designUHS() already performed this derivation internally without exposing it, so consumers that needed the parameters had to recover them from the shape of its output — which stopped being possible once producers began publishing ordinate-level spectra with no plateau to read them from. The derivation now has one entry point and designUHS() calls it.

Behaviour changes

  • designUHS() output is unchanged, verified identical across five input spectra, both levels and four vs30 settings including the error paths. Its cleaning and Sec. 21.4 error messages now name getDesignParameters() instead, since that is where the check lives.

newmark 1.3.2

New features

  • solveRTGM() implements the risk-targeted ground motion of ASCE 7-22 Sec. 21.2.1: it solves for the spectral ordinate whose lognormal collapse fragility, convolved with the site hazard curve, reaches the target collapse probability. Defaults are the ASCE 7 target set (10 percent conditional collapse at the ordinate, logarithmic standard deviation 0.60, 1 percent in 50 years); every tunable is an argument so that other target sets, such as the AASHTO guide specification, can be selected. Returns the ordinate, the uniform-hazard ordinate and their ratio, the risk coefficient.

    The procedure follows the USGS Risk-Targeted Ground Motion Calculator (Makdisi 2024, doi:10.5066/P9A2F2B8), which ASCE 7-22 Commentary C21.2.1 traces to Luco et al. (2007). It reproduces that implementation to 1.8e-15 relative across all 340 published NEHRP-2020 fixtures; 60 of them ship as the regression oracle in tests/testthat/.

    The hazard curve is resampled and log-log extrapolated to [0.10, 10] times the 2475-year uniform-hazard ordinate before integrating. Restricting the integral to the tabulated curve domain instead is what makes low-hazard, long-period curves fail to bracket.

    This is the probabilistic branch only. The Sec. 21.2.2 deterministic motions, the Sec. 21.2.3 80-percent floor and the Sec. 21.2 maximum-direction scaling remain out of scope, as they already are for designUHS(). The two functions are complementary, not chained: designUHS() consumes a uniform-hazard spectrum and performs the Sec. 21.4 parameterisation, while solveRTGM() consumes a hazard curve and returns an ordinate that Sec. 21.3 scales by two thirds directly.

newmark 1.3.1

Fixes

  • sampleSaCorr() and its internal full-correlation path now select every marginal from the requested Tn[j], independent of input row order. The previous data.table expression resolved both uses of Tn as the table column, so column j used the period stored in input row j; ordinary period-grouped tables could consequently sample every channel from the PGA marginal. The Gaussian copula and random-number stream are unchanged. PGA-only displacement models are unchanged, while flexible models that use Sa(1.3Ts) or Sa(1.5Ts) change when the affected input ordering was used.

newmark 1.3.0

CRAN release: 2026-07-16

Model corrections (audit-driven; change numerical results)

  • designUHS() now derives the spectral parameters per ASCE 7-22 Sec. 21.4 (site-specific input): at MCER level, SMS = 0.9 * max SaM(T) over T in [0.2, 5] s and SM1 = max(0.9 * max[T * SaM(T)], SaM(1 s)) over [1, 2] s when vs30 > 442 m/s, else [1, 5] s (new optional vs30 argument; NULL uses the conservative wide window). The previous seeding (Ss = Sa(0.2 s), S1 = Sa(1.0 s), Fa = Fv = 1) matched no ASCE 7-22 pathway for site-specific spectra. ASCE-table products shift on the order of 10 percent. The Sec. 21.2.3 floor and the Sec. 21.2.2 deterministic cap remain out of scope (documented).

Fixes

  • designUHS(): the output grid beyond TL was degenerate whenever TL >= 5 (including the default 8 s), so the long-period branch (ASCE 7-22 Eq. 11.4-5) never appeared in the returned table; the grid now samples one octave beyond TL. Documentation now states precisely which parts of ASCE 7-22 the function implements (the Sec. 21.4 derivation and Sec. 11.4.5.2 two-period shape) and which it does not (the Sec. 21.2 floor and deterministic cap).

Breaking changes

  • The dead Rrup argument has been REMOVED from rhoBJ(), fitSaF(), fitDnCurve(), and fitDnModel(), and from the fitDnScenario() records contract (a legacy Rrup column in records is tolerated and ignored). The Baker & Jayaram (2008) inter-period correlation model has no distance dependence, so the argument never influenced any result: outputs are identical draw-for-draw under the same seed. Callers passing Rrup by position or by name must drop it.

newmark 1.2.1

Model corrections (audit-driven; change numerical results)

  • Dn_BT07(): restored the published + 1.50 * Ts term of Bray & Travasarou (2007) Eq. 5, previously missing (the function received Ts but never used it), and the -0.22 intercept for nearly rigid sliding masses (Ts < 0.05 s). The implementation now reproduces the worked example published by Bray (2007). Median displacements increase by exp(1.5 * Ts) (about x2.1 at Ts = 0.5 s).

  • geSiteTable() / getSiteProperties(): the Shibata-Soelarno (1975) row of the Ishihara (1996) shear-modulus table now uses its published LINEAR void-ratio form F(e) = Ce - e/(1 + e); the quadratic form of the other models had been applied to it, under-predicting that model by a factor of about 65 and dragging the Sands-group mean modulus down (about -20 percent in Go for typical sand profiles).

  • fitModel.Ts(): rebuilt on the classical Rayleigh quotient with the static-deflection trial shape. The previous trial shape violated the base boundary condition under package usage and overestimated the fundamental period of normal (stiffness-increasing) profiles by tens of percent (e.g. +44 percent on a 200/400 m/s two-layer column, +27 percent on the vignette profile). The rebuilt estimator is within -0.4 to -2.8 percent of the exact layered solution across test profiles and is invariant to the layer ordering of the input. All Ts-dependent products (site properties, flexible-block displacements) shift on regeneration.

  • Vs30toSID(): the class-B lower bound is now 910 m/s (3000 ft/s, ASCE 7-22 Table 20.2-1), matching the packaged SiteClass table; it was 900. The class-E row in SiteClass now prints the correct inequality (<150 m/s).

Behaviour changes

  • Dn_BM19() (and fitDnCurve(), fitDnModel()): near-fault pulse behaviour is now selected with the explicit pulse argument (default FALSE = ordinary motion), matching the papers’ definition of pulse as a declared property of the record and the authors’ spreadsheet. The automatic PGV > 115 cm/s switch is retired; PGV remains a required predictor inside the pulse equations, whose published 150 cm/s sub-regime is unchanged.

newmark 1.2.0

New functions

  • F_ST20() implements the final 2020 NGA-East ergodic site-amplification generation (Stewart et al. 2020 linear model and F760 impedance/gradient weighting; Hashash et al. 2020 nonlinear model with period-dependent reference velocity and PGA-dependent nonlinear sigma). Coefficients are regenerated from the USGS nshmp-lib machine-readable table (data-raw/SiteAmp2020.R); the implementation is validated at machine precision against the OpenQuake nga_east.py reference functions.

  • fitSaFMixture() computes site-amplified spectra from an analytic scenario mixture (one lognormal component per rupture-variant x GMPE branch, as produced by the DSHA scenario engine). The composition of the rock mixture with the lognormal site dispersion is exact: Gauss-Hermite quadrature over the PGA channel plus monotone mixture-CDF inversion. No random numbers are drawn - outputs are byte-stable; Sa and SaF are quantiles of the same conditional demand distribution and AF = SaF / Sa is the amplification of the p-quantile.

Behaviour changes

  • fitSaF() gains a working models menu: "ST20" (new default) and "ST17". Pass models = "ST17" to reproduce the 2017-generation amplification.

  • F_ST17() sigma corrected: the nonlinear component now applies the PGA-dependent multiplier of its source model (sigma_Fnl = sigma_f2 * ln((PGA_3000 + f3)/f3), PEER 2017/05 Eq. 2.5 as implemented by USGS nshmp-lib and OpenQuake). Earlier versions used bare sigma_f2, overstating dispersion at low PGA and understating it at high PGA. muLnF is unchanged; sdLnF now varies with PGA.

Documentation

  • Corrected provenance: the F_ST17() family implements the PEER Report 2017/05 NGA-East recommendations (not Seyhan & Stewart 2014 as previously documented). The inter-period correlation model is Baker & Jayaram (2008); it has no distance dependence and the Rrup arguments are retained for backward compatibility only.

newmark 1.1.1

New functions

  • fitDnScenario() fits the Newmark displacement curve and Monte Carlo draws for a single scenario row (one (TR, IDg) combination). Returns a list with curve and draws data.tables matching the contract of fitDnCurve().

  • fitDnScenarios() applies fitDnScenario() over a records data.table of scenario rows and row-binds the results. Accepts useParallel and workers arguments for fork-based parallel execution on multi-core systems.

  • fitKmaxScenario() fits the seismic coefficient kmax for a single scenario row given pre-computed displacement draws and model weights. Returns a data.table with Da, p, kmax, PGA, Kh, TR, IDg, and IDn.

  • fitKmaxScenarios() applies fitKmaxScenario() over a records data.table and row-binds the results. Supports the same useParallel / workers interface as fitDnScenarios().

Performance

  • fitDnScenarios() now runs Dn scenario rows in parallel when useParallel = TRUE and workers > 1L. On systems where fork-based parallelism is available (Linux, macOS), wall time scales nearly linearly with workers for large scenario grids. Sequential execution (the default) is unchanged.

Internal tooling

  • inst/cran-release.R simplified to a two-phase source() from the garage canonical scripts (cran-check.R then cran-submit.R). The prior inline copy of those steps is removed; the garage copy is the single source of truth.

newmark 1.1.0

CRAN release: 2026-05-19

Bug fixes

  • inst/CITATION: replace utils::packageVersion("newmark") with meta$Version so the citation file evaluates correctly when the package is not yet installed (CRAN pre-test failure).
  • inst/WORDLIST: add 135 domain-specific terms (author surnames, technical vocabulary, package-internal identifiers) so spelling::spell_check_package() exits clean. Add Language: en-US to DESCRIPTION.

Documentation

  • Vignettes rebuilt for fidelity to the implemented code and the underlying paper (Verri Kozlowski 2026, working). Hallucinated cross-references to deleted articles, oqt-pipeline terminology leaking into newmark prose, and inconsistent function-name references were removed.

  • New vignette dynamic-site-response documents the dsra-side methodology merged into newmark 1.1.0: Ishihara (1996) small-strain shear-modulus model, Gazetas & Dakoulas (1985) inhomogeneous truncated shear-beam fundamental period, synthetic-profile generation via getSiteProperties() / geSiteTable(), and the ASCE 7-22 site-class utilities Vs30toSID() / SIDtoVs30().

  • pipeline vignette rewritten as a function-level overview of the four library workflows (dynamic site response, hazard import, displacement curves, seismic coefficient), with a single top-of-file note that orchestration on a hazard grid (looping scenarios, persisting tables) is the consumer’s responsibility.

  • ensemble-formulation vignette now states the inverse-problem formulation explicitly, documents the BM19 near-fault PGV > 115 cm s⁻¹ correction and the NFC D100/D50 selector, and lists the operational defaults (Monte Carlo size, k_y grid, reported quantiles) that the package and downstream consumers share.

  • newmark-quickstart vignette: stale links to deleted articles removed; Next steps rewritten as in-package vignette() and ?fnname references.

Improvements (no behavioural change)

  • interpolateSaTable() and buildQSpline() now emit a descriptive warning() when called on a UHS table with duplicate (Tn, p) rows, pointing the user to the standard disaggregation key (siteID, ID, Vs30, TR). R’s cryptic regularize.values: collapsing to unique 'x' values warning is suppressed in favour of the new message — it is replaced, not stacked on top. Numerical behaviour is unchanged: duplicate rows are still silently collapsed via stats::approx / stats::splinefun ties = "mean". SoT-validated bit-identical on clean fixtures (5/5 direct + 36/36 pipeline) and byte-equal output on duplicated input.

Internal renames (no public API change)

  • Private helpers .summarizeDn(), .summarizeDnDraws(), .aggregateWeighted() parameter p.vals renamed to pVals. Producer .sampleDnScenario() returns the list element under the new key pVals (was p.vals). Consumers fitDnCurve() and fitDnModel() updated to read .sample$pVals. All references are intra-package; no exported symbol changes signature. SoT-validated (Stage B helpers direct identity 3/3; Stage C swap-in-place via assignInNamespace over the 6 affected functions running the full fitDnCurve grid 216/216, both bit-identical).

Breaking changes

  • getDnKy() parameters renamed for STYLE.md compliance: ky.n → kyN, p.ref → pRef. Output is bit-identical for any call that uses the defaults or passes the args by position. Callers that passed by name must update the keyword. No live caller in the ecosystem (oqt, research scripts, MCP server, skill templates) was found to pass these by name; the rename is a clean keyword change. SoT-validated (Stage A direct identity 27/27 across 3 fixtures × 3 Ts × 3 kyN values; Stage C swap-in-place via assignInNamespace 216/216 across the full fitDnCurve pipeline).

  • fitDnCurve() no longer accepts NS.draws. The parameter was a partial-retention knob (filter draws to the first NS.draws realisations) that no caller used; every observed call site set NS.draws = NS. The default NS.draws = 0L produced draws = NULL, which silently broke invertDnDraws() for callers who followed the README pattern. draws is now always returned as a populated data.table (or an empty data.table with the four canonical columns when no model produces output, replacing the prior NULL corner-case return).

    Migration: drop NS.draws = ... from any fitDnCurve() call. Output for NS.draws = NS callers (the universal pattern) is bit-identical (verified across vignette + 2 real UHSTable fixtures × 3 seeds × 2 NS × 3 Ts × 2 NFC × 2 weight sets).

newmark 1.0.0

  • First CRAN release.