Asymmetry in the μSR software landscape

This page compares Asymmetry with WiMDA, musrfit, and Mantid. The count-reduction guide below covers reviewed grouping, calibration-policy, timing, and loading conventions. Its agreement statements apply to the named versions, observations, and settings; they do not certify every analysis method or every supported file.

The intended reader is a μSR practitioner who already uses one of WiMDA, musrfit, or Mantid and wants to know where Asymmetry fits.

For end-to-end analyses analogous to the typical workflows in those tools, see Analysis workflows.

The reference programs provide different ways to organise an analysis. WiMDA offers grouped time and frequency workflows; musrfit uses authored .msr model and run descriptions; Mantid composes algorithms on workspaces. Asymmetry exposes a Python analysis core with a desktop front end. The comparison below concerns particular operations rather than a ranking of programs or a collective file-format guarantee.

Choose a comparison by topic

Start with the observation to be analysed. Loading a file, forming a count ratio, fitting that ratio, and reconstructing a field distribution are separate operations, even when one program presents them in the same window.

Methods and owning references

Topic

Comparison on this page

Asymmetry reference

Files and available observations

Files and available observations

Loading Data

Groups, t0, and good bins

Reviewed count-reduction conventions, Integer timing alignment

Detector Grouping and Layout

Alpha, beta, backgrounds, and deadtime

Stored and automatic calibration policies, Count-background estimates, Deadtime calibration methods

Alpha calibration, Beta calibration, Background correction

Binning and errors

Binning and timestamps, Asymmetry values, coordinates, and errors

Histogram binning

Fit objectives and constraints

Fit observations, objectives, and constraints

Fitting engine, Count-domain fitting (α calibration & single histograms)

Fourier phase, scale, and MaxEnt

Normalisation and transformed observations, Maximum-entropy reconstruction

Fourier analysis

Integral and RF scans

Count-integral condition scans

Integral scan mode (avoided-level-crossing field scans)

Program and documentation identities have different scopes. The executed cases name the actual release/build. A numbered Mantid algorithm revision (such as Fit v1) is separate from the documentation edition and installed release. The consulted musrfit manual identifies its edition as V1.10.0 and uses an unversioned address; it does not certify every installed build. WiMDA’s original manual is version 1.2 (February 2005); its author identifies it as older than the current program. Manual sections below describe that edition’s methods; only the named native cases establish release-specific observations. See the WiMDA author’s documentation notice.

Files and available observations

Asymmetry’s load(), CLI loading, and desktop file picker use the same registered loaders: ISIS muon NeXus V1/V2 (.nxs, .nexus), PSI BIN/MDU, and MusrRoot/LEM ROOT. HDF5 NeXus needs h5py; legacy HDF4 needs pyhdf; ROOT needs uproot. Generic column files and TRIUMF MUD have no built-in raw-run loader. Curve-array fitting is a separate entry point, not evidence of a raw-file importer. The loading reference describes supported layouts, period selection, metadata precedence, and unavailable-calibration limits.

For comparison, the WiMDA author’s format list names ISIS NeXus, PSI BIN/MDU, J-PARC NeXus, and TRIUMF TRI, and describes LEM access via conversion to BIN. That list is not a claim that TRI and MUD are interchangeable. The musrfit file-format manual distinguishes its PSI, ROOT, NeXus, MUD, and ASCII inputs. Mantid’s version 6.16.1 LoadMuonNexus v1 and LoadPSIMuonBin v1 describe particular acquisition loaders. These documented entry points are not an exhaustive cross-program format matrix or numerical loading comparison.

A suffix does not establish layout support, usable time calibration, detector geometry, or an independent-count model. The executed PSI/ROOT cases in PSI legacy ROOT calibration and PSI good-data intervals identify matched observations separately from loading coverage. The compared musrfit build has HDF4 support disabled; conversion would test a different input, rather than establish original-file support in that build. Unsupported or untested layouts here do not imply that another program cannot analyse them.

Evidence behind numerical comparisons

An analytic identity checks independently authored mathematics; a published numerical value checks a cited calculation or measurement. Neither executes another program. A recorded execution case preserves the native input, settings, version/build, numerical output, mapping, and justified tolerance. Replaying it is a regression against that recorded execution. A fresh comparison actually runs the configured external program and checks its identity and output. Documentation of a method alone establishes no numerical agreement.

The bounded synthetic controls cover complementary operations:

  • F/B reduction and count binning: musrfit 1.12.1, revision 708f777b, ROOT 6.40.04, uses owned tagged counts, t0 4/7, forward raw bins 20–59, backward raw bins 23–62, alpha one, zero background, and no deadtime. Packing-one ratios and positive-Poisson errors agree after fraction-to-percent conversion within relative \(5\times10^{-6}\), set by text precision. Summed-count packing three agrees for thirteen complete groups. Packing four exports nine groups and omits the tenth complete group retained by Asymmetry. The packed F/B path also has a first-point discrepancy; no whole-output packed F/B equivalence is claimed. These differences remain separate from the matched count-sum arithmetic.

  • Supplied-error constant fit: Mantid 6.16.1 and musrfit 1.12.1 (708f777b, ROOT 6.40.04) fit all three values 2, 4, 7 with errors 1, 2, 1. Independent normal equations give \(40/9\), standard error \(2/3\), and chi-square \(113/9\) with two degrees of freedom. Supplied absolute errors are not rescaled by residual variance. Mantid’s FlatBackground/Least squares/Levenberg–Marquardt result retains the stationary status Changes in function value are too small; its reduced chi-square is multiplied by two for this comparison. musrfit uses type 8, packing one, fit range −0.5 to 2.5, and MIGRAD/HESSE. Absolute tolerances are \(10^{-7}\) for Mantid and \(5\times10^{-5}\) for musrfit’s rounded text output. This scalar control does not certify nonlinear errors or coverage.

  • Manual-phase FFT: musrview 1.12.1 (708f777b, ROOT 6.40.04) applies 0° and 37° in a complete type-0 request on owned rounded cosine counts, 0.01 μs bins, t0 zero, explicit background 200000, packing one, 0.03–9.93 μs, no apodisation, and no requested padding. Observed real and imaginary ROOT-canvas coefficients follow \(z_{37}=z_0e^{-i37\pi/180}\) on native bins 0–493 within a per-bin float32 storage bound. The final native bin 494 remains unrotated at both settings and is retained as a discrepancy. This checks the same-bin phase relation; it does not establish full DFT scale, endpoint, or error agreement. The tested musrFT data-file ASCII and graphic exports remain unchanged when their command-line phase is varied.

These recorded controls use only owned synthetic input and safe numerical outputs. Ordinary tests need no reference installation, network, or private acquisition data. Fresh execution is optional and fails on a configured identity, command, parser, or numerical failure. The standalone Mantid pulse-source check has a separate checkout/module identity and scope, described in Fourier analysis; installed Mantid 6.16.1 does not identify that source module. None of these controls certifies general MaxEnt, wizard behaviour, physical calibration, or statistical coverage.

See the musrfit manual for packing, type-8 fitting, and the FOURIER block, and Mantid Fit v1 for supplied-error least squares.

Reviewed count-reduction conventions

Start with Count reduction and transformed signals for the standard F/B operation order. The closest reference depends on the operation, rather than on a single preferred program. The detailed contracts and versioned comparisons are linked below; different defaults must be matched before judging numerical agreement.

Conventions and meaningful differences

Operation

Asymmetry convention / closest comparison

Meaningful difference and detail

Membership and F/B pairs

Explicit selections, like Mantid detector IDs and musrfit histogram lists

Unique members and independent disjoint pairs are Asymmetry choices; reuse across projections remains valid. See Detector membership and exclusions.

Calibration policies

Stored versus explicit estimates, like WiMDA’s stored α and Estimate; integral α uses the Mantid AlphaCalc convention

Enabled automatic policies use current inputs; fixed and one-shot values stay fixed. See Stored and automatic calibration policies.

Alignment and NeXus width

Whole-bin alignment and declared-resolution fixed-bin counts, compared with musrfit

Selected exact origin and residuals are retained; Mantid also preserves coordinates and offers common-grid rebinning. See Integer timing alignment.

PSI intervals

Selected file-good intersection; musrfit is the primary PSI comparator

Explicit musrfit endpoints and omitted defaults differ; WiMDA endpoint membership is not established here. See PSI good-data intervals.

ROOT calibration and coverage

Legacy v4/v5 and modern layouts use common reduction; closest to musrfit

Stored bank selection, PPC uncertainty, and the directory-case limit have distinct scopes. See PSI legacy ROOT calibration.

Agreement means agreement for matched observations and uncertainty conventions, within the stated storage or export precision. Explained convention differences are identified separately. Unverified paths and unresolved uncertainty comparisons remain limits; acceptance of a selection alone does not establish correct covariance or physical calibration. Versioned WiMDA numerical checks cover the background and binning controls explicitly identified below. Other WiMDA descriptions compare workflows without claiming executable numerical agreement; the author’s homepage and manual provide general context.

Asymmetry values, coordinates, and errors

Asymmetry forms \(A=(F-\alpha B)/(\beta F+\alpha B)\) from corrected, packed counts, with alpha on the backward group. Its first-order error uses retained count variances, including existing shared-background/overlap terms when counts are merged. See Errors for the derivation, signed-count example, and Gaussian omission policy. Dataset values and errors are percent; count-to-ratio primitives are fractions. Neither scale imposes universal estimator bounds.

The WiMDA original manual, Sec. 5.2 defines the ordinary beta-one F/B balance. The versioned binning controls below compare values, without establishing identical positive-count or sparse-bin errors.

In Mantid’s AsymmetryCalc v1 documentation (reviewed 6.16 development), beta is one and alpha has no error. Its count-based rule is \(\sigma_A^2=(F+\alpha^2B)(1+A^2)/(F+\alpha B)^2\), treating numerator and denominator as independent. With 6000 forward and 4000 backward counts and unity alpha, both values are 20 %, but the documented rule gives 1.019803903 percentage points, compared with Asymmetry’s 0.979795897. This is documented arithmetic, not a native Mantid execution comparison.

musrfit 1.12.1 (708f777b, ROOT 6.40.04) type 2 fits the raw coordinate \(r=(F-B)/(F+B)\) with a transformed model and reciprocal alpha \(\alpha_m=1/\alpha\). Here raw means unbalanced; requested background subtraction is already applied. Its beta convention agrees after alpha conversion. For \(u=1+r,v=1-r\), converting to Asymmetry’s coordinate gives

\[A(r)=\frac{u-\alpha v}{\beta u+\alpha v},\qquad \sigma_A\simeq \left|\frac{2\alpha(1+\beta)}{(\beta u+\alpha v)^2}\right|\sigma_r.\]

For repeated counts \(F=60,B=40\), fixed \(\alpha=1.25,\beta=0.8\), zero backgrounds, corrections off, and packing one, native ASCII data/error columns give \(r=0.2,\sigma_r\simeq0.0979796\) with \(\alpha_m=0.8\). Converting both gives \(A=0.1020408163,\sigma_A=0.1147720100\) in fractions, agreeing with Asymmetry within the six-significant-digit export precision (relative tolerance \(5\times10^{-6}\)). The unity dense control also agrees. These checks use interior points on owned synthetic NeXus inputs, one native thread, and fixed balances/backgrounds. The fitted theory column is not a reduction oracle. See the musrfit type-2 manual and packing tutorial.

First-order local error agreement after a nonlinear coordinate change does not establish identical Gaussian objectives for arbitrary residuals or identical model recovery. musrfit’s empty-native-bin variance treatment also differs from Asymmetry’s Gaussian omission policy: musrfit supplies nonzero empty-native-bin variance, while Asymmetry omits exactly undefined or zero-first-order-variance histogram ratios. Asymmetry retains usable signed values and corrected zeros with positive variance. WiMDA’s sparse/signed fit-selection policy is not established by the value-only controls here. In Mantid, AsymmetryCalc uses the distinct error approximation above. These are deliberate method differences, not identical sparse-bin likelihoods. Asymmetry also checks final ratio usability after requested binning and period composition, reports omission reasons, and leaves raw/count-fitting access available. Native packing is not a statistical coverage or wizard-ranking test.

Asymmetry’s ordinary errors are conditional on the selected alpha/beta coefficients. A fitted calibration covariance is not propagated into a full reduced-bin/run covariance by that operation; see Calibration conditioning. This does not establish the covariance capabilities of other programs or of custom analyses built with them.

Binning and timestamps

Asymmetry sums counts and their retained variance before forming F/B asymmetry. Nonuniform target widths are evaluated at the actual left edge, rounded to the nearest positive whole number of acquisition bins (half-up), and stamped at the actual contributing-bin midpoint. The exact 10 μs Variable law, lifetime Constant error law, and pre-t0 clamp define the target, rather than its realised integer width. Base nonuniform bins retain inclusive selected support and a partial tail. Fixed packing and additional recorded-bin merges trim incomplete groups and retain all complete ones.

The closest width-selection comparison is WiMDA 1.326. With 0.1 μs native bins, initial width 0.12 μs, and Variable width at 10 μs of 0.25 μs, all eight Constant error values and eleven shared Variable values agree within 0.0005 percentage points. Actual midpoint times, terminal support, and the literal 10 μs law remain deliberate differences. Half-up is Asymmetry’s deterministic choice, not a verified WiMDA tie result. Positive-count uncertainty conventions also differ; this comparison does not certify matching corrected/sparse errors.

musrfit 1.12.1 (708f777b, ROOT 6.40.04) provides ordinary integer packing, with matching interior sums and mean native timestamps under matched support and integer timing. It does not establish the same nonuniform controls. Mantid histogram-centre conventions offer a coordinate comparison; histogram Rebunch retains a partial group, while Rebin can redistribute fractional bins. This Mantid method comparison does not establish matched Variable or Constant error controls. The recorded fixed-packing/history control below has a different scope. See Histogram binning for algorithm references, terminal qualifications, and uncertainty assumptions.

Asymmetry honours authored merges without a spectral peak veto. The controls above establish the particular compared bin membership, rather than a universal reference-program validation policy. Actual FFT requests still require a uniform grid. Asymmetry also checks that an authored count-fit factor and selection leave usable complete fit bins, before resolving unused corrections. The factor is never silently reduced to one. Raw inspection and request repair remain available; identical empty-bin diagnostics in other programs are not established.

Both fit wizards use counts-first automatic preparation for resolved histogram observations. In the analytic F = [90, 1], B = [10, 9] control, merging gives 65.454545…%, while averaging the separate ratios gives 0%. Curve-only, rotated, and unresolved historical/custom observations retain independent-curve means. The native count controls above support the input arithmetic and membership; they do not validate Asymmetry’s wizard ranking or entire search. The wizards continue fitting asymmetry, separately from TF alpha/beta count calibration.

Point-model fits use output timestamps and conditional point errors. Wide bins average the signal; their point-model approximation can attenuate oscillations or shift phase. Mantid’s separate histogram-evaluation option is not a general integration prescription for every count ratio or curve mean.

Detector membership and exclusions

Asymmetry groups unique acquired-channel ordinals and applies exclusions without rewriting raw histograms. Missing preset channels and empty effective groups make reduction unavailable. Typed overlapping ranges form a set union; structured duplicate membership is rejected. These ordinal IDs do not certify physical detector identity across reordered files.

Mantid’s MuonGroupingCounts documentation specifies detector-ID selection. The musrfit manual’s RUN grouping selects histogram numbers. Mantid 6.16.1 sums synthetic channels [10, 20] and [50, 60] to [60, 80], retaining supplied errors in quadrature within relative/absolute tolerance \(10^{-12}\). musrfit 1.12.1 (708f777b, ROOT 6.40.04), with corrections off, 0.1 μs bins, and packing one, exports 125000 counts/bin from constant channels 100000 and 25000; Poisson errors agree within relative \(5\times10^{-6}\), set by six-significant-digit text precision. Both reject the tested missing member and count a repeated member twice. Asymmetry’s unique membership rule deliberately rejects that repetition. The tests do not establish pair independence or equivalence of geometry, timing, or calibration methods.

Asymmetry requires disjoint effective acquired channels within each independent F/B pair, after exclusions. Native Mantid 6.16.1 MuonPairingAsymmetry rejects tested identical lists but accepts partial overlap; see its algorithm documentation. Native musrfit 1.12.1 accepts the tested identical and partially overlapping pairs. These observations describe request acceptance, not a covariance-aware error guarantee. Asymmetry retains EMU vector channel reuse across projections and overlapping auxiliary groups, and never infers identity from equal counts.

The WiMDA original manual, Secs. 5.1 and 5.10 describes numbered grouping tables and per-group timing. It also describes pregrouped acquisition channels: an acquired histogram may already combine physical detectors. Asymmetry’s exclusions select acquired histogram channels without rewriting the loaded data. Excluding such an aggregate removes the whole channel; hidden constituent observations cannot be recovered by regrouping. The manual alone does not establish WiMDA 1.326’s repeated-member, overlapping-pair, or unavailable-channel policies. See Detector Grouping and Layout and Detector exclusion.

Stored and automatic calibration policies

Asymmetry preserves fixed and one-shot calibrated coefficients during regrouping. Enabled automatic policies use current inputs: per_run_estimate α is the integral ratio of the current corrected grouped counts over the good window. It is distinct from repeating an optimising α fit. β has fixed and calibrated policies, without that automatic ratio policy. See Alpha calibration and Beta calibration for the contracts.

The WiMDA original manual, Secs. 5.2–5.3 describes stored grouping alpha and a separate estimate action. Mantid’s AlphaCalc v1 computes the integral efficiency ratio as a separate operation; it is the closest convention for Asymmetry’s automatic ratio, rather than a claim that Mantid automatically refits α after every grouping edit. musrfit specifies α/β through its RUN/model configuration. These policy distinctions do not establish numerical equivalence between calibration estimators. The cited Mantid documentation version is distinct from the 6.16.1 executable used in the numerical comparisons below.

WiMDA 1.326 quick Diamagnetic and General controls compare displayed alpha point estimates; they do not establish calibration uncertainty. Asymmetry’s quick methods return point estimates without an interval. Asymmetry’s Diamagnetic uses continuous minimisation and internal packing; its General uses a two-window decay balance. Native point agreement does not identify the internal algorithm of WiMDA’s quick estimates. Matched 16 ns synthetic beta = 1 controls over raw bins 4–498, background/deadtime off and bunch one agree within 0.001 absolute alpha: TF Diamagnetic 1.237 versus 1.236610, TF General 1.250 versus 1.249827, and relaxing General 1.250 versus 1.250073 (WiMDA then Asymmetry). These checks do not establish uncertainty coverage or equivalence on corrected or acquired observations. Asymmetry also checks rounding-level General contrast and a residual lifetime diagnostic; their purpose and limits are in Alpha calibration.

Asymmetry also checks that automatic α has finite positive F/B integrals and a finite positive ratio. An undefined estimate makes that reduction unavailable and editable; it does not replace the stored request with unity. Explicit fixed unity remains valid. This protects the ratio’s mathematical inputs, without claiming that all reference workflows use the same guard.

Shared TF count calibration

Asymmetry’s tf_calibration_fit core operation uses one exponential/common-phase F/B count model, fixing beta at one for alpha-only calibration or freeing it for joint calibration. Both backgrounds and other nuisances belong to one full covariance. Poisson deviance/dof counts unique free parameters once, and errors are conventional conditional HESSE standard errors, without residual scaling. The existing beta count_fit adapter uses this operation; the independent single_histogram ratio remains a separate protocol.

Applied File/manual deadtime corrections select the existing weighted count objective, retaining measured or supplied independent variances. Off/skipped deadtime uses raw Poisson counts. Backgrounds are fitted model terms; grouping fixed values, range estimates, and automatic-tail estimates do not subtract or constrain them. This shares the count model, not musrfit’s type-2 asymmetry objective.

Independent raw-Poisson calibration controls through the public operation support the conditional standard-error scale on the stated informative TF cases; their Monte Carlo precision, interval availability, selector outcomes, and misspecified/weak-contrast limits are given in Alpha calibration. The normal ±SE reference is approximately 68.27 %, without an exact guarantee. Optional MINOS profile intervals remain separate from HESSE errors. Neither native agreement nor these bounded controls establish physical deadtime coverage, selection-adjusted intervals, or a universal precision ranking. Quick Diamagnetic/General supply point estimates without intervals; Mantid 6.16.1 AlphaCalc returns scalar alpha without a calibration error.

Matched musrfit 1.12.1 type-0 chi-square controls use 16 ns native bins, two disjoint detectors, 100000 good frames, deadtime 0.02 μs, zero fixed backgrounds, and packing one/four. After matching its corrected-count variance convention, fixed/free-beta points agree within 0.0001 absolute and alpha errors within 0.1%. Asymmetry’s retained measured-count weights differ: one beta-one control gives alpha 1.249554 versus native 1.249574, and errors 0.001102 versus 0.001162. This is a weighting difference, not an uncertainty correction to copy. Mantid’s muon ApplyDeadTimeCorr preserves supplied errors. The recorded WiMDA FileBG control retains its sample-only export error; that is a separate operation. Those correction conventions do not establish equivalent native TF calibration objectives or complete deadtime uncertainty propagation. The comparisons are conditional weighted fits, not uncertainty coverage or a new detector-response likelihood.

The closest executed counterpart is musrfit 1.12.1 (708f777b, ROOT 6.40.04) global type-0 histogram fitting with MAX_LIKELIHOOD, described in the Single Histograms tutorial. Four owned controls (rounded means and one seeded Poisson realisation each, fixed/free beta) use identical 16 ns integer arrays, raw bins 4–498, fixed lifetime, zero backgrounds, phases 0/180 degrees, and matched free frequency/damping/amplitude parameters. Alpha/beta points agree within 0.0001 absolute after mapping native \(N_B/N_F\) to reciprocal Asymmetry alpha; native fractional amplitude maps to Asymmetry percent, MHz to MHz, and degrees to radians. This is not equivalence to musrfit type-2 asymmetry chi-square, a coverage study, or validation of acquired detector response.

The installed likelihood route reports a different error scale. With only the forward normalisation free, the exact Poisson MLE is 19999.9849 and its standard error is 9.24119 counts/bin. Asymmetry gives about 9.24122; native musrfit gives about 6.535, approximately \(1/\sqrt{2}\) as large. The independent control and full-TF information calculation support Asymmetry’s unit-deviance error criterion (see the Minuit2 manual). No compensating rescaling is applied. This observation is specific to that installed likelihood route; it makes no claim about other versions, type-2 chi-square, or all musrfit uncertainties.

WiMDA 1.326 quick Diamagnetic and Asymmetry’s quick method give about 1.237 on an owned beta-one TF control with generating alpha 1.25; the model fit gives 1.249995, changing mean reduced asymmetry by about −0.53 percentage points. These results describe different estimators, not a universal accuracy ranking. WiMDA’s separate paired/all-group model-fit errors remain unverified here; Mantid AlphaCalc is a ratio comparator, not a shared-TF likelihood comparison.

Asymmetry also checks raw-Poisson numerical inputs, disjoint detector banks, existing count-scale sanity, and free-beta precession/error diagnostics. Their scope and consequences, and the limits of processed/corrected/pulsed support, are in Alpha calibration and Beta calibration. Passing these checks does not certify that an exponential common-phase signal describes a calibration.

The separate select_tf_calibration script operation compares undamped, exponential, Gaussian, and stretched envelopes on identical count observations, using minimum summed objective without a complexity penalty or extra ambiguity threshold. All candidate states and conditional covariances remain available. This recommendation is an Asymmetry convenience. The existing single-model operation and beta adapter retain the exponential model.

musrfit’s MuSRFit interface lists Exponential Cosine, Gaussian Cosine, and Stretch Cosine. The musrfit THEORY documentation defines simpleGss with a Gaussian rate \(\sqrt{2}\) times Asymmetry’s \(\sigma\); generExpo uses \(\exp[-(\Lambda t)^p]\) at positive time. Fresh musrfit 1.12.1 (708f777b, ROOT 6.40.04) component controls for all four shapes agree within 0.0001 absolute alpha/beta with fixed/free beta, zero fixed backgrounds, phases 0/180 degrees, and the same 16 ns raw bins 4–498. Reciprocal alpha, fractional amplitude, Gaussian-rate, and degree/radian conventions are matched. This tests component points, not native automatic selection or an extension of the likelihood error-scale comparison above.

The Gaussian envelope convention determines the conversion to a field-distribution second moment; a function name alone does not. Asymmetry also checks captured Gaussian-component provenance at Brandt trend consumers and requires explicit conversion of values and errors. This is an Asymmetry request safeguard, not a claim that WiMDA, Mantid, or musrfit provides an equivalent provenance check. See Gaussian widths at second-moment consumers and Gaussian width convention; the component comparisons above retain their stated versions and conditions.

Mantid’s StretchExpMuon documentation defines the corresponding non-oscillatory stretched envelope and describes a usual exponent range up to two. Asymmetry’s positive stretching exponent has no upper physical cap; its numerical lower search default can be overridden with positive bounds. No fresh native Mantid stretched fit or WiMDA model-fit comparison is asserted. The outgoing Diamagnetic/reduced-signal comparison above concerns a different estimator and does not enter fit-cost ranking.

Integer timing alignment

Asymmetry retains integer shifts and resolves its exact reference and common observed support from the effective selected channels after exclusions. The maximum selected anchor defines the reference; acquired-channel ordinal breaks ties. Fractional residuals and their range are recorded in μs, without interpolating counts or assigning timing uncertainties.

Asymmetry uses declared resolution for fixed-bin NeXus count analysis, retaining its exact origins and bin-centre convention. Supplied auxiliary coordinates remain inspectable and do not override a usable resolution or infer an extra acquisition-start shift. musrfit 1.12.1 (708f777b, ROOT 6.40.04) exports identical arrays for regular, cumulative-float32, and shortened-terminal synthetic HDF5 coordinates with unchanged counts and resolution. Mantid 6.16.1’s legacy loader retains supplied coordinate variations in the compared HDF4 inputs while retaining their counts exactly. These cases do not certify physical t0 or terminal-bin exposure; Loading Data states the fixed-bin and axis-fallback conventions.

Asymmetry also checks that selected channels have compatible widths before integer alignment, protecting its whole-bin count transport. A failed request leaves raw histograms available. When no usable resolution exists, Asymmetry also checks the selected fallback axis’s shape, finiteness, monotonicity, and uniform spacing to obtain a physical width. Failure then concerns calibrated loading, because inventing a width would invent physical coordinates. Asymmetry also checks detector/period timing shapes and redundant integer/exact t0 agreement: unusable optional timing uses the remaining calibration, while an exact/integer disagreement is advisory and keeps integer timing for the affected detectors without discarding coherent exact origins elsewhere. Missing or unusable file t0 uses the existing search policy, and two-period selection retains the selected period’s own calibration status. These checks apply to the adopted interpretation, rather than requiring all auxiliary fields to conform. They do not imply that other programs omit such checks in every workflow.

The recorded musrfit 1.12.1 (708f777b, ROOT 6.40.04) type-0 integer grouping case uses 0.1 μs bins, forward 1 2, t0 2 4, data 1 9, packing one, and no corrections. Eight selected counts agree exactly with Asymmetry, with timestamps agreeing within \(10^{-12}\) μs. This is a count and coordinate check, not physical t0 certification. Mantid 6.16.1 keeps per-spectrum shifted X coordinates; direct grouping requires compatible grids, while explicit rebinning can redistribute counts onto common edges. See the Mantid muon reduction tutorial and musrfit manual. The tutorial version is separate from the recorded executable version. Detector Grouping and Layout gives the selected bins and output mapping.

File bounds and authored relative windows

Asymmetry keeps acquired detector bounds on raw bins and maps them through current selected-channel alignment. From file preserves the raw observations under a uniform common t0 change; Manual stores each supplied endpoint as a signed offset from effective t0. An omitted endpoint retains its mapped file limit. This separates file facts, timing, and an authored interval.

The closest comparator for preserved observations is musrfit’s explicit data channels, with timing specified separately in the RUN block. For tagged raw bins 11–14, musrfit 1.12.1 (708f777b, ROOT 6.40.04), packing one, and no corrections, unchanged data 11 15 preserves counts and Poisson errors when integer t0 moves from 10 to 15. Moving the explicitly supplied interval to data 16 20 instead matches Asymmetry Manual offsets 1 and 4. Counts and errors agree within the native ASCII precision (relative tolerance 5 × 10⁻⁶). An exact Asymmetry origin of 0.2074 μs differs from musrfit’s integer-centre origin 0.21 μs by a constant 0.0026 μs; changing t0 by five 0.02 μs bins preserves that difference. Fit-time limits and omitted data defaults are separate operations. Manually matched channels do not establish an automatic relative-window policy in musrfit’s GUI.

Mantid 6.16.1 CreateWorkspace and CropWorkspace provide the closest physical-coordinate comparison. With the same 32 raw observations, supplied Poisson variances, and 0.02 μs width, translated axes and explicitly matched crop limits retain identical counts and errors. Moving both raw crop endpoints by five bins selects 16–19 instead of 11–14. Centres agree within 10⁻¹⁵ μs. These constructed workspaces test the selected operation, not NeXus loader defaults or Mantid’s GUI policy.

The WiMDA original manual, Secs. 5.5–5.6 and 5.10 describes t0, a good-time offset, and per-group controls. It does not establish identical raw-bin membership under Asymmetry’s independent timing/window selectors and two relative endpoints. No WiMDA numerical agreement is claimed for this operation. The matched musrfit/Mantid synthetic operations above have no unexplained count disagreement; automatic-default equivalence remains unverified.

Asymmetry also checks that explicit manual endpoints are ordered and within common acquired support. This protects authored selection without inventing one-bin observations. The failed calculation remains editable and raw data remain available; it does not reject a facility file. This safeguard does not assert that all reference programs implement the same validation.

NeXus physical good-time selection

The muon NeXus definition, version 2 revision 8, p. 51 defines physical good times relative to raw_time. Asymmetry adopts inclusive raw-centre membership on its nominal fixed-bin grid and retains the resulting per-detector bounds through reduction and project reopening. Integer attributes retain precedence for their corresponding endpoints. The definition does not settle every writer’s integer base or sub-bin rounding convention.

In Mantid 6.16.1, a histogram workspace with the same 32 tagged bins, 0.02 μs width, and 0.2074 μs origin, cropped to raw limits 0.22–0.30 μs, selects the same four raw observations, centres, and Poisson errors. This comparison uses CreateWorkspace and CropWorkspace; it does not certify every NeXus loader default. Mantid retains supplied coordinate variation, whereas Asymmetry’s declared-resolution grid is uniform. A physical threshold near a stored edge can therefore select different observations. Matched selections must be established before comparing reduction values; legacy first-good metadata does not by itself establish equal defaults.

musrfit 1.12.1 (revision 708f777b, ROOT 6.40.04), with packing one and explicit data channels 11 15, selects tagged raw bins 11–14. Counts and Poisson errors agree within the ASCII export’s six-significant-digit precision (relative tolerance 5 × 10⁻⁶). With an explicit integer t0 of 10 channels, its timestamps agree with Asymmetry’s centre origin of 0.21 μs. The compared single-histogram path uses an integer channel anchor; its fractional t0 request does not reproduce Asymmetry’s exact 0.2074 μs calibration. This is a timing-convention difference, not a count-selection disagreement. Physical fit-time limits and omitted data defaults are separate controls; no native HDF4 agreement is claimed for the build with HDF4 support disabled.

WiMDA’s documented t0 and good-time controls do not establish identical sub-bin rounding or selected observations for these NeXus fields. The original manual, Secs. 5.5–5.6 describes those controls; no native WiMDA numerical comparison is claimed here.

Asymmetry also checks that the selected file-good intervals have nonempty common acquired support before reducing independent counts. An unavailable request retains the acquired histograms and editable grouping. Optional field shape or unit diagnostics do not impose whole-file rejection when another usable interpretation remains; see Loading Data.

PSI legacy ROOT calibration

Asymmetry adopts musrfit’s numeric TLemRunHeader v4/v5 calibration for legacy LEM ROOT: ns resolution and zero-based per-detector t0, followed by the shared count reduction. musrfit 1.12.1 (708f777b, ROOT 6.40.04) gives identical selected counts, times, and errors for equivalent synthetic BIN, legacy ROOT, and modern TFolder inputs. Asymmetry’s F/B asymmetry and Poisson errors agree within the six-significant-digit ASCII output precision (relative \(5\times10^{-6}\)); timing agrees within relative \(3\times10^{-8}\) from BIN float32 versus legacy decimal ns storage. Asymmetry reports percent asymmetry while musrfit reports fractions.

The compared selections are forward raw bins 20–59 and backward raw bins 23–62 with t0 4/7, packing one, alpha=beta=1, zero background, and no deadtime. Different endpoint requests need not select the same observations. Extra ROOT container bins remain in the loaded histograms. The generated TDirectory variant uses decayAnaModule, while this musrfit reader requests DecayAnaModule; Asymmetry reads that representation without claiming a native numerical comparison for it. WiMDA and Mantid coverage of this legacy class is unverified. The legacy v5 agreement above uses synthetic files; the modern acquired examples below extend a different part of ROOT coverage. Two acquired v4 examples in the musrfit 708f777b example set extend legacy coverage: all eight decay arrays retain their 66,601 stored bins, numeric 0.1953125 ns width, identities, and available calibration. Header slots 0–3 calibrate both bank 00 (NPP) and bank 20 (PPC); musrfit’s ROOT-NPP and ROOT-PPC select the respective stored bank. Distinct-anchor checks also confirm that the timing follows header slots rather than histogram ordinals. The musrfit LEM format description names these banks.

With channels 1/3, raw bins 3500–3539, packing one, alpha=beta=1, no added correction, and zero background, summed counts agree exactly. Times, F/B values, and counting errors agree within relative \(5\times10^{-6}\) ASCII precision, allowing absolute \(5\times10^{-7}\) for F/B values near zero. Both file t0 3419 and explicit t0 3429 are compared. The PPC check uses stored observations and the declared square-root-count error convention; it does not validate pileup physics, likelihood, or corrected-count covariance. The banks must not be combined as independent observations. The initial pairs in these two v4 files select bank-00 left/right; this is no general default geometry certification. Version 5 extra-bank calibration is outside this check.

Serialized v4 regression inputs remain optional external examples. Always-on normalisation tests cover sparse/reordered identities and unavailable anchors, but no independently authored serialized-v4 CI fixture is supplied. Loading checks over 37 external ROOT files establish compatibility and retained data, not numerical certification of every layout or default reduction. See Loading Data and the musrfit RUN settings.

PSI good-data intervals

Asymmetry adopts zero-based inclusive raw-bin bounds for file-good intervals and intersects them after whole-bin alignment over the selected detectors. BIN upper values at or beyond histogram length are clipped to the final stored index; interior values are retained. MDU M3/T4/T5 markers are converted at their hardware scale before clipping. Modern ROOT per-detector bounds use the same inclusive representation; legacy LEM v4/v5 without good fields defaults from each detector’s t0 through its final stored bin. Container-added bins remain stored, independently of their counts or the selected interval.

musrfit 1.12.1 (708f777b, ROOT 6.40.04) explicit data intervals at packing one have an exclusive upper bound. For matched tagged BIN and MDU M3/T4/T5 selections, summed counts agree exactly, raw-bin identities agree, and exported Poisson errors agree within relative \(5\times10^{-6}\) (six-significant-digit ASCII precision). Timestamp comparisons at that export precision use the same tolerance. The BIN and ROOT timing comparison above uses its separate storage-precision tolerance.

Omitted data does not mean the same default as Asymmetry’s file-good intersection. For the compared 128-bin BIN with t0 4, musrfit’s single-channel fallback selects raw 5..127 whether the supplied last-good field is 60, 127, or 128. The equivalent 129-bin legacy/modern folder ROOT storage also includes raw bin 128, a retained terminal zero. Asymmetry’s modern ROOT good fields can exclude that bin; legacy defaults can include it. Neither a terminal zero nor equal header integers establishes equivalent observations.

For the synthetic MDU marker cases, omitted-data single-channel fallbacks select raw 23..127 for M3 (t0 7, 0.625 ns bins) and 107..127 for T4/T5 (t0 5, approximately 0.09765625 ns bins). These differ from the converted file-good intervals 35..80 and 26..60 respectively.

Tested on musrfit 1.12.1 example ROOT files (revision 708f777b): modern LEM with a terminal good bound, and FLAME and HiFi with interior good bounds. Explicit requests matched to Asymmetry’s file-good observations give exact summed counts and raw-bin identities. Timestamps and positive-count F/B errors agree within relative \(5\times10^{-6}\); asymmetry exports also allow absolute \(5\times10^{-7}\) near zero because of ASCII precision. Omitted native data intervals select different observations in all three examples. The HiFi initial pair includes an empty channel; its error comparison uses an additional explicitly selected populated pair. Zero-count uncertainty conventions are outside that positive-count comparison. For one-sided bins, Asymmetry retains a unit-fraction uncertainty sentinel, while the native export supplies count-dependent finite errors. This is selected numerical coverage of three acquired examples, not agreement for every example file or every ROOT layout.

WiMDA’s original manual describes good-time controls, without establishing matching PSI endpoint membership in WiMDA 1.326. Mantid 6.16.1 LoadPSIMuonBin retains all 128 tagged BIN counts, exposes good-data limits in time units, and does not crop the raw workspace to those limits. This loading result does not establish Mantid’s default analysis selection. MDU and ROOT interval reductions in Mantid, WiMDA ROOT, and native musrfit numerical reduction of the generated lower-case-directory representation remain unverified here.

Asymmetry also checks that the selected channels have common acquired support and a nonempty intersection for a file-good reduction of independent counts. This prevents alignment padding from becoming observations. Raw histograms remain available when the initial pair is unavailable; the grouping request can be edited. These checks do not certify bin exposure from count values. See Loading Data, the musrfit manual, Mantid LoadPSIMuonBin, and the WiMDA manual.

NeXus deadtime units

Asymmetry converts explicitly declared NeXus-v2 microsecond and nanosecond deadtimes once to its canonical μs coefficients at loading. Thus 20 ns and 0.02 μs give identical corrected counts, including signed coefficients. Legacy v1/HDF4 uses fixed μs coefficients. Missing and other unit declarations retain the stored-value-as-μs behaviour. This choice does not establish WiMDA’s loader conversion or agreement on missing-unit defaults.

The compared musrfit 1.12.1 (708f777b, ROOT 6.40.04) NeXus-v2 path uses stored coefficient values as μs regardless of the tested ns declaration. Asymmetry intentionally differs on explicitly declared ns values. With 100000 counts, 0.1 μs bins, 100000 frames, and a 0.02 μs coefficient, the analytic inverse gives 125000 counts; an equivalent 20 ns declaration has the same meaning in Asymmetry. The musrfit manual describes its file deadtime correction option; the WiMDA manual provides the correction and calibration context.

Numerical WiMDA agreement and native Mantid 6.16.1 loader unit equivalence are not established. The coefficient interpretation does not change the correction formula or retained measured/supplied counting variance.

Optional file deadtime calibration

The WiMDA original manual, Sec. 5.9 describes file/manual correction and calibration-file loading. The absent-table fallback in current WiMDA is not established here. Asymmetry leaves the whole requested operation uncorrected with a warning when required File calibration is absent or incomplete, recording requested File versus actual skipped. It makes no automatic calibration-file search. A supplied complete zero table is an available identity.

musrfit 1.12.1 warns and leaves histograms without coefficients uncorrected; other histograms can still be corrected. Mantid 6.16.1 ApplyDeadTimeCorr accepts the tested partial table in its numerical API. This does not establish the same policy in its Muon GUI or every loader path. The musrfit missing, zero, and short-table count exports agree with these statements; Asymmetry deliberately skips the whole operation for a short table. Public context is provided by the WiMDA documentation, musrfit manual, and Mantid correction documentation.

Asymmetry also checks non-zero correction exposure and bin width and rejects saturated or nonfinite inverses over the required F/B observations. Standalone full-input preparation checks all supplied histograms. Those calculation failures remain errors; missing optional calibration does not prevent raw loading or other valid enabled operations.

Required F/B deadtime observations

Correction and selection are distinct operations. A caller that crops before correction supplies a different domain from a full-array correction followed by cropping. No universal reference-program domain is assumed here.

Asymmetry deliberately uses the exact required observations for F/B reduction, grouped raw-count fits, grouped Fourier inputs, and MaxEnt: the requested detectors and contributing raw bins, plus active range, tail-fit, and reference-background inputs. An unused saturated bin or detector cannot block the calculation, and a missing unused File coefficient cannot suppress it. A required missing coefficient skips the entire participating sample/reference operation with a warning; a required pole remains an error. Original detector ordinals and raw-bin coordinates are preserved. See Detector Grouping and Layout for the inverse and uncertainty convention. One joint count fit, selected Fourier average, or MaxEnt reconstruction is one cohort for File fallback. A final exclusion does not remove samples used by earlier normalisation, background estimation, conditioning, or phase fits. In particular, MaxEnt interior exclusions follow its weighted-baseline estimate; FFT preparation retains the input needed by its upstream conditioning. F/B-asymmetry FFT uses the F/B preparation directly; its group selector is inert. Standalone full-input preparation still checks every supplied observation. These preparation choices do not establish identical solver inputs or statistics across all three reference programs.

Public workflow context: WiMDA documentation, musrfit manual, and Mantid MuonPreProcess.

Signed deadtime coefficients

Asymmetry preserves finite signed calibration coefficients. A fitted negative coefficient reduces raw counts under the non-paralysable inverse; it is an empirical fit parameter, distinct from a detector’s physical recovery time. The published ISIS silver-calibration method and the numerical domain are explained in Detector Grouping and Layout.

A native Mantid 6.16.1 check of ApplyDeadTimeCorr v1 used raw counts [5, 20], 1 µs bins, 10 good frames and coefficients −0.1, 0 and +0.1 µs. Asymmetry agrees with Mantid and the analytic inverse to relative tolerance \(10^{-12}\). CalMuonDeadTime v1 recovered noiseless signed synthetic coefficients −0.02, 0 and +0.02 µs within \(2\times10^{-12}\) µs; its published examples also include negative fitted values. Asymmetry’s rational per-detector fit agrees on these noiseless inputs within \(10^{-9}\) µs. The inputs are \(N_0=10000\) counts/bin, lifetime 2.1969811 µs, one million frames, 0.016 µs bins with edges \(0.496+k\times0.016\) µs (\(k=0,\ldots,282\)), and fit range 0.5–5 µs. Asymmetry weights raw-count residuals by Poisson standard deviations, while Mantid fits the transformed regression; this is not a claim of noisy-data estimator equivalence. Calibration noise and model mismatch can produce negative fitted coefficients; the sign does not diagnose an individual calibration’s quality.

The WiMDA original manual, Sec. 5.9 and Appendix B describes correction and calibration workflows. It does not establish signed-coefficient acceptance or numerical agreement in WiMDA 1.326.

Deadtime calibration methods

A mean response and an estimator are separate choices. Asymmetry’s dedicated Cal/Estimate fits use the rational forward mean corresponding to the inverse above. They minimise raw-count weighted squared residuals separately for each detector, with a free non-negative amplitude, signed coefficient, fixed muon lifetime, and no count background or polarisation term. Weights use the observed count with a floor of one; supplied variance is not used by this calibration fit. The published transformed ISIS line is algebraically equivalent to the mean, but fitting that line is a different objective. Source histograms and windows also matter; Detector Grouping and Layout distinguishes the Grouping buttons, scripted API, and MaxEnt caller.

Cal retains per-detector coefficients. Estimate averages those fitted coefficients arithmetically and broadcasts the mean. The WiMDA original manual, Sec. 5.9 and Appendix B describes estimate and individual calibration workflows; their detailed objective and pooling rules are not established by that description. No numerical WiMDA calibration agreement is claimed here.

For an independent illustration using Asymmetry’s rational model, two detectors with the same 0.02 µs coefficient and equal rates give 0.02 µs both by averaging fits and by fitting their mean observations. With a 4:1 amplitude ratio, averaging the independent fits still gives 0.02 µs, while fitting the mean observations gives about 0.02636 µs. This noiseless illustration uses amplitudes 100000 and 25000 counts/bin at time zero, lifetime 2.1969811 µs, 100000 frames, 0.05 µs bins, and centres 0.525–6.475 µs. It demonstrates noncommuting operations for a nonlinear response; it is our own pooling experiment, not WiMDA output or a recommendation to change the estimator.

Mantid 6.16.1 CalMuonDeadTime v1 fits a transformed line separately for each spectrum, using goodfrm exposure. Its documented regression and Asymmetry’s rational fit agree on the signed noiseless inputs above. They do not minimise the same noisy-data objective. A bounded synthetic sensitivity comparison using matched observations, exposure, and selected bins shows method and window dependence under noise, a small constant background, and weak polarisation. Independent Poisson draws here are a chosen sensitivity model, not certification of physical deadtime statistics, estimator optimality, or confidence-interval coverage. A teaching silver-run comparison can show behaviour under matched selection; without known detector truth it cannot rank accuracy.

musrfit 1.12.1 (708f777b, ROOT 6.40.04) reports the compared deadtime-cor estimate option as unimplemented; its file correction is a separate operation. This limited observation does not imply that every musrfit calibration facility is unimplemented. The musrfit manual describes this option for pulsed/ISIS NeXus data.

The matched noiseless calibration is agreement under stated conditions; pooling, objectives, and exported count-error conventions are explained method differences. Physical continuous-source calibration and corrected ROOT bank statistics remain unresolved by these comparisons. Generic availability of Cal does not establish PSI applicability or a new exposure interpretation.

Asymmetry’s dedicated calibration errors are conditional Hessian errors from its chosen objective. Estimate combines independent detector errors in quadrature and divides by their number; it does not estimate detector scatter. These errors are distinct from the measured/supplied counting variance retained during correction below. Coefficient and exposure uncertainty are not propagated through that correction. A successful count-domain residual-DT₀ fit also need not supply a complete calibration: skipped File correction or unknown unused coefficients leaves it usable with a warning and no promotion. Explicit Off retains its supported zero-base behaviour; see Count-domain fitting (α calibration & single histograms).

Deadtime count errors

The deadtime inverse and its count-error convention are separate choices. Executed synthetic checks distinguish two reference programs:

  • Mantid 6.16.1, ApplyDeadTimeCorr, retains supplied workspace errors. With 1 µs bins, 100 frames, counts [100, 25], and a 0.2 µs coefficient, corrected values are [125, 26.315789473684212]; errors [10, 5] remain [10, 5]. Supplying [7, 3] instead retains [7, 3]. Signed coefficients, unequal detector rates, grouping, and bin merging preserve that behaviour.

  • musrfit 1.12.1, revision 708f777b with ROOT 6.40.04, exports single-histogram errors from corrected counts. With count-rate scaling disabled, 100000 counts/bin, 0.1 µs bins, 100000 frames, and a 0.02 µs coefficient give 125000 counts and error 353.553. The signed grid [−0.02, 0, 0.005, 0.02, 0.05] µs, single and unequal grouped detectors (100000 and 25000 counts/bin), and packing factors one and four agree with this convention to relative tolerance \(5\times10^{-6}\), set by the ASCII export precision.

Asymmetry retains measured counting variance, or supplied independent variance, through deadtime preparation, grouping, background subtraction, and binning. It agrees with Mantid’s supplied-error convention on the recorded synthetic inputs and differs from the tested musrfit exported-error convention. Calibration coefficients and frame exposure are fixed in this propagation; the error bars do not include their uncertainty or physical deadtime-induced correlations. These comparisons do not assert equality of low-count asymmetry formulas, background treatment, or fit objectives. The formulas and limitations are in Detector Grouping and Layout.

Reference timing and native coordinates

Asymmetry resolves sample and selected reference clocks independently, retaining whole-bin shifts and fractional file-origin residuals in F/B and grouped Fourier/MaxEnt preparation. Asymmetry also checks compatible selected native bin widths and required reference support before subtraction. It neither resamples counts nor restricts raw file loading.

Asymmetry offers Automatic and Manual reference scaling. Automatic uses compatible evidenced exposure ratios, or visibly assumes one when exposure is unknown. Whole-run PSI/ROOT deadtime normalisation one is not measured pulse exposure. Manual uses its authored positive coefficient regardless of metadata. WiMDA 1.326’s bounded FileBG check uses sample/reference 200/100 good frames, 200/30 counts per native 0.016 μs bin, one detector in group 1, bunch one, deadtime off, and Auto/Manual BG off. The exported 140 counts agree with the ratio-two arithmetic within 0.00051 counts (text precision). This does not establish a physical sample/holder calibration or an equivalent unknown-exposure label. Mantid 6.16.1 generic Scale expresses an explicit coefficient; no corresponding automatic Muon GUI or musrfit FileBG scale workflow is claimed. Fixed-scale independent variance excludes estimated-scale and shared-reference cross-result covariance.

WiMDA 1.326 FileBG uses a shared active channel shift. The synthetic NeXus check uses sample counts 200 per bin with t0 zero, reference counts 10 + rawbin with file t0 zero or two bins, 0.016 μs native width, equal 200-frame exposures, and scale one. WiMDA exports raw group 1 with one detector, bunch one, deadtime off, and Auto/Manual BG off. After restoring sample timing, both 1999-row exports agree exactly at count-export precision, beginning 190, 189, 188, 187 counts. Reference loading can change the active timing, so the matched comparison restores the sample first. Independent Asymmetry timing instead starts 188, 187, 186, 185 for the two-bin reference, checked against those tagged observations on bounded 32-bin sample support. The difference is explained by the alignment convention.

Mantid 6.16.1 generic Minus rejects incompatible X arrays; this is explicit workspace arithmetic, not an equivalent automatic Muon reference GUI workflow. musrfit’s run addition has separate t0 semantics and is distinct from FileBG subtraction. See Background correction for the subtraction contract and its independent-variance assumptions.

Reference subtraction: deadtime and counting errors

The following comparison reuses recorded native WiMDA 1.326 (64 bit, Wine 10.0) and Mantid 6.16.1 results on synthetic constant spectra. It is not a new native execution or a comparison on an acquired sample/holder pair. The sample has 200/150 counts per native bin in two detectors; the reference has 30/20. Each group contains one detector. Sample/reference exposures are 200/100 good frames, native width is 0.016 μs, raw t0 is zero, and the subtraction scale is two. Enabled Manual deadtime is 1 ns (0.001 μs) per detector. The table shows the first group with bunch one; WiMDA exports raw group 1 at nonnegative time, with Auto/Manual BG off and FileBG enabled. Its sample exposure is restored to 200 frames after reference loading, which can change the active frame count.

First-group count and standard error

Operation

Count

Standard error

Asymmetry / explicit Mantid, DT off

140

17.88854382

WiMDA FileBG, DT off

140

14.1421

Asymmetry / explicit Mantid, both sides DT-corrected

152.186836518

17.88854382

WiMDA FileBG, sample DT-corrected, raw reference

153.333

14.1421

For each acquisition, the independently evaluated inverse is \(c(n)=n/[1-n\tau/(F\Delta t)]\), with \(\tau\) and \(\Delta t\) in μs. Asymmetry gives \(c_S(200)-2c_R(30)=152.18683651804673\). The Mantid sequence explicitly applies ApplyDeadTimeCorr to each workspace, then Scale and Minus. A separately configured sample-only sequence gives \(c_S(200)-2(30)=153.33333333333334\), explaining the WiMDA FileBG result. These are authored workspace operations, not an equivalent automatic Muon GUI workflow.

Asymmetry and the configured Mantid arithmetic retain independent input variance \(200+2^2(30)=320\) with DT off or on. They do not infer Poisson variance from corrected means. WiMDA’s tested raw-group export retains the sample-only counting error \(\sqrt{200}\) after FileBG. These measurement- variance conventions exclude coefficient/exposure uncertainty and physical deadtime-induced correlations. The independent-variance rule also excludes estimated-scale and shared-reference cross-result covariance. Matching this example does not establish full uncertainty coverage or identical fit objectives.

The Asymmetry/Mantid independent-arithmetic checks use relative tolerance \(10^{-13}\). WiMDA’s text-export absolute tolerances are 0.00051 counts and 0.000051 in the standard error, set by print precision; all 1999 exported rows satisfy the constant-spectrum comparison. This bounds the claim to the recorded synthetic raw-group export, not noisy inputs, ROOT-bank statistics, or every GUI path. See Background correction for the count-reference contract and Reference timing and native coordinates for the separate channel-alignment convention.

Additive physical signals and amplitude units

The musrfit tutorial models sample and holder contributions with separate amplitudes, relaxation, and fields. These physical signals are distinct from the uncorrelated count background fitted in a single-histogram model. Asymmetry’s existing composite models likewise express additive physical signals; no external FileBG operation is implied. A constant holder component is appropriate only for a stationary, non-relaxing response in the chosen frame.

A reused musrfit 1.12.1 (708f777b, ROOT 6.40.04) check fits a generic fittype 8 ASCII curve \(a(t)=0.2\exp(-0.35t)+0.04\), with 81 equally spaced points from 0 to 8 μs, supplied error 0.001, packing one, and rate fixed at 0.35 μs⁻¹. MIGRAD/HESSE recovers both fraction-valued amplitudes within absolute tolerance \(5\times10^{-6}\) (printed parameter precision), starting at 0.15/0.02. Asymmetry’s Exponential + Constant forward evaluation uses amplitudes 20%/4%; dividing its percent output by 100 gives the same fractional curve (relative tolerance \(10^{-14}\), absolute \(10^{-16}\)). This is a model-composition and units illustration. It is not matched raw muon reduction, a new native fit, a holder-fraction calibration, or a comparison of fit covariance. See Background correction and Asymmetry units across the API for modelling advice and units.

Normalisation and transformed observations

Asymmetry’s internal fingerprint, peak and residual diagnostics preserve a uniform curve or prepare captured nonuniform histograms on fixed native bins (factor 1, merge 1) over exactly their retained native support. They do not change the user’s fit observations. A separate residual view evaluates the same fitted model/parameters, while original RMS/runs, scores and residual times/errors remain with the original fit. This is an explicit Asymmetry diagnostic policy, not an assertion that the reference programs implement an equivalent wizard.

Mantid 6.16.1 FFT v1 requires linear X coordinates and specifies output/shift conventions. A matched point-data 1.3 MHz control agrees with a direct DFT modulus, including interval scaling, to 2e-10 absolute/2e-11 relative. musrFT 1.12.1 (708f777b, ROOT 6.40.04) packing 1 and 3 controls locate that line within their actual frequency bins; native endpoint/packing and amplitude conventions are not equated. WiMDA 1.326’s fixed-bin, unfiltered F/B Fourier control locates the line within its actual 0.097656 MHz bin, matching the corresponding direct-DFT maximum. The manual control uses its observed 950-point display support and next-power-of-two padding; exact native endpoints, amplitude/phase conventions and errors are not equated. No equivalent wizard or new spectral uncertainty agreement follows from these controls.

A group count trace and an F/B ratio are different observations. Dividing out the muon lifetime creates a lifetime-corrected count signal, not a calibrated polarisation. A finite-window level then defines a fractional residual for Fourier analysis or a starting scale for a reconstruction. The Count reduction and transformed signals diagram shows these branches, and Fourier analysis specifies the actual weights and scale limits.

The native spectral cases name WiMDA 1.326 and musrfit 1.12.1 (revision 708f777b, ROOT 6.40.04). The method descriptions identify the older WiMDA manual separately from Mantid algorithm-v1 documentation at the 6.16 development snapshot (revision 1967ccb4). Formula and manual behaviour are distinct from executed numerical agreement. The bounded Mantid point-data FFT control above does not verify a complete muon-input Fourier workflow, PhaseQuad or MaxEnt; the separate standalone pulse-kernel comparison does not establish them either.

Operations and conventions

Observation / operation

Asymmetry

Reference convention and scope

Grouped Fourier counts

Lifetime correction, fractional level estimated with \(1/\sigma\), optional \(1/\sigma\) mean removal, and coherent-gain amplitude scale

WiMDA’s original manual Sec. 9 describes grouped Fourier averaging. Its exact level/scale convention is not established by that manual.

Theory-free single-histogram Fourier preparation

The grouped FFT uses the weighted level above

The musrFT 1.12.1 CLI --lifetimecorrection supplies a scale control; its recorded response does not identify the level estimator.

Lifetime removal and fitted constant

MaxEnt starts from an inverse-variance level and fits its own grouped phase/amplitude reconstruction model

In Mantid, RemoveExpDecay divides by a fitted constant after lifetime correction and subtracts one. This documented normalisation does not establish its fitting weights or equivalence of reconstruction algorithms.

Two measured transverse projections

Pointwise baseline/gain/sense rotation with retained shared counting noise

Mantid RRFMuon rotates two supplied components after explicit coordinate, phase, and unit conversion. PhaseQuad can prepare quadratures from a detector phase/amplitude table; it is a separate operation.

One scalar F/B ratio

Complex carrier plus FIR low-pass display; fits retain original laboratory observations

musrfit’s documented scalar RRF path uses its own packing and theory settings. Scalar-input processing does not establish vector covariance agreement.

The level estimators can differ even for noiseless oscillations when the window contains a nonzero oscillatory mean. Coherent-gain correction alone does not remove that difference or calibrate detector efficiency. Padding refines the sampled transform grid; it supplies no new observations. Neither peak height nor magnitude area is universally invariant to window, damping, phase, or apodisation.

The WiMDA homepage/manual and F. L. Pratt, Physica B 289–290, 710 (2000), describe grouped spectral analysis. The original manual’s all-poles discussion is distinguished from iterative reconstruction in Maximum-entropy reconstruction. The musrfit manual distinguishes single-histogram, asymmetry, and RRF analysis. In Mantid, see RemoveExpDecay v1, MuonMaxent v1, and RRFMuon v1.

Phase, clock, and geometry

An oscillatory model phase belongs to a chosen clock and sign convention. Asymmetry uses \(\cos(2\pi\nu t+\phi)\) with μs, MHz, and model radians; FFT manual tables and MaxEnt group controls use degrees. A positive manual FFT phase rotates the complex spectrum by \(e^{-i\phi}\). The first retained sample contributes a separate origin phase. Changing t0, source-response origin, correction recipe, or observation window can change a phase estimate; matching group IDs alone is insufficient to transfer it.

The manual FFT table and sampled Auto Phase estimates are distinct from the fitted MaxEnt group-phase store. MaxEnt’s Use fitted phases and Send phases to fit exchange group phases with the grouped time-domain fit using degree/radian conversion. They do not synchronise the FFT table. See FFT phase modes and Grouped time-domain fitting for the supported models and inputs.

The WiMDA author’s overview lists Fourier phase correction, including automatic options. It does not establish numerical phase transfer to Asymmetry’s FFT or MaxEnt phase stores. musrfit’s TFieldCos model uses \(\cos(2\pi\nu t+\pi\phi/180)\), with phase in degrees, according to the musrfit model manual. This names one model convention, not every component or full Fourier-display exchange. The recorded musrview manual-phase control separately checks the same-bin rotation on bins 0–493, with unrotated final bin 494; it does not validate phase transfer through the musrFT command-line exports.

In Mantid, CalMuonDetectorPhases v1 first determines a common frequency and then detector amplitudes and phases. PhaseQuad v1 uses a detector phase/amplitude table to form two quadratures. Its detector model is \(\cos(\omega t-\phi_i)\), with table phases in radians. At the same origin this corresponds to \(\phi=-\phi_i\) in Asymmetry’s plus-phase model, followed by conversion to degrees for its group controls. A shifted origin requires the additional frequency-dependent clock conversion. These are distinct from a sampled common-phase FFT fit per group. A group label or phase table does not establish orthogonal physical axes. Authored vector geometry and measured timing remain explicit assumptions; see Assumptions and limitations.

The two-projection rotation uses \(x=P_x-b_x\), \(y=(P_y-b_y)/g\), and \(\theta=2\pi\nu_{\rm RF}t+\varphi_{\rm RF}\): \(x'=x\cos\theta+s y\sin\theta\), \(y'=x\sin\theta-s y\cos\theta\). The sense \(s\) and second-axis sign must be mapped explicitly to Mantid’s real/imaginary convention, including nonzero phase. With Mantid inputs \(R=x,I=s y\) at the same \(\theta\), Asymmetry’s outputs are \(x'=R'\) and \(y'=-I'\). Mantid RRFMuon uses radians for its phase input; Asymmetry’s frame phase controls use degrees. The authoritative Rotating-frame projection describes the mapping, shared acquisition/reference/background counting covariance, conditional errors, and unequal-baseline pooled G+R safeguard. Lab G+R and rotating Red, Green, and G−R retain their supported behaviour.

A single real ratio cannot supply two measured projections pointwise. Its Rotating-reference-frame display display instead mixes with a complex carrier and filters. Same-bin quadrature covariance enters first-order magnitude errors; filtered bins remain correlated and low-signal magnitude remains biased. Correlated or anisotropic quadratures do not imply an exact Rician law. Corrected marginal bars establish neither full cross-bin covariance nor an independent joint likelihood; see Magnitude errors and Counting observations and conditional errors. The laboratory observations remain the scalar fit inputs; grouped count fits retain the separate What this mode fits.

Fit observations, objectives, and constraints

Asymmetry fits percent asymmetry with independent supplied-error Gaussian objectives through FitEngine.fit/fit_arrays; fit_observations also accepts eligible raw Poisson counts. Count-domain and grouped fits retain their own observation models. Fixed values, bounds, equality links, and single-run affine ties constrain parameters; affine ties are unavailable in global, count-domain, and grouped/series fits. Shared parameters across runs and across differently modelled series are described in Fitting engine, Grouped time-domain fitting, and Joint fits. A secondary parameter-model fit consumes fitted values/errors rather than the original counts; see Parameter trending.

The closest executed supplied-error objective comparison is the three-point constant fit. The closest raw-count TF comparison is musrfit global histogram fitting, which retains the installed likelihood error-scale discrepancy. Agreement for one objective does not establish agreement for another. Nonlinear parameter errors, model selection, and coverage retain their own assumptions.

For documented workflows, the WiMDA original manual, Secs. 7.1–7.9 describes F/B, single/all-group fitting, fixed parameters, parameter linking, and batch fits. The musrfit manual, FITPARAMETER, THEORY, and RUN blocks describes authored parameters/models and chi-square or maximum-likelihood fitting. Mantid’s Fit v1, version 6.16.1 documentation exposes cost function, minimiser, ties, constraints, and evaluation type as separate choices. These descriptions do not establish identical constraint languages, defaults, parameter errors, or automatic model selection. Match observation membership, units, bounds, objective, and fixed nuisances before comparing results.

Other analysis entry points

Asymmetry provides Composite models for model expressions, Fitting engine for minimisers and conditional parameter errors, Fit wizard for candidate-model comparison, and Parameter trending for secondary fits to run parameters. The model reference covers Fit functions; Frequency-domain conditioning and Spectral moments describe spectral estimators and derived quantities with their own assumptions. Integral scan mode (avoided-level-crossing field scans), Synthetic runs and degraded statistics, User functions, and Negative-muon capture-lifetime analysis (experimental) cover the separate scan, synthetic-data, plugin, and negative-muon workflows. Project files describes saved analysis state and compatibility. These are method pointers, without a claim that another program lacks an analogous or extensible workflow.

Automatic FFT phase estimation

Asymmetry’s Auto Phase estimates one physical phase per detector group using a nonnegative frequency distribution and the actual sampled FFT response. It preserves the rotated measured spectrum, includes windowing and known mean subtraction, and reports conditional uncertainty and processing sensitivity. Weak, incompatible, or unstable inputs are refused. It is distinct from fitting an entropy-plus-negativity objective or jointly subtracting a spectral baseline; see the published-method discussion and citations in the Fourier reference.

This is an FFT projection, not a maximum-entropy reconstruction or a full-detector timing calibration. Numerical agreement with another program’s automatic phase estimate depends on its objective, sampling conventions, band, and preprocessing. A comparison of mathematical formulations alone does not establish release-specific numerical agreement.

Maximum-entropy reconstruction

Asymmetry’s grouped-count MaxEnt prepares lifetime-corrected observations and an inverse-variance starting level, then fits a nonnegative spectrum through its response model with group amplitudes/phases and an entropy constraint. Its fitted phases belong to that reconstruction, separately from manual FFT rotation, sampled Auto Phase, and detector timing calibration. The MaxEnt reference describes preparation, errors, source assumptions, de-weighted windows, and reconstruction limitations.

Mantid’s MuonMaxent v1 documentation (6.16 development documentation snapshot) describes an iterative muon spectrum reconstruction with detector phases, deadtimes, and a forward response. This is a relevant workflow comparison, not an executed full reconstruction match. The Fourier reference gives the separately pinned standalone pulse-kernel control and its clock conversion; that control does not compare the fitted spectrum, objective, or covariance.

The WiMDA original manual, Sec. 9 explicitly distinguishes its maximum-entropy/all-poles transform from iterative entropy reconstruction. F. L. Pratt, Physica B 289–290, 710 (2000), also describes muon spectral analysis; these older method descriptions do not identify all reconstruction options in WiMDA 1.326. No matched native MaxEnt spectrum/error comparison is claimed for WiMDA or musrfit. A similar method name or a sharper peak does not establish algorithm equivalence, absolute polarisation, or additional measured information.

Selected automatic timing

Asymmetry retains the continuous maximum and pulsed leading half-height feature conventions. WiMDA’s original manual, Sec. 10, describes Search for T0, without specifying an identical estimator. musrfit 1.12.1 musrt0 -g writes per-detector prompt maxima. With two unique synthetic prompt features at raw bins 20 and 31, musrt0 and Asymmetry’s continuous search agree exactly (integer-bin tolerance zero). Grouped and detector maxima need not agree when channels have different delays. Pulsed maximum and leading half-height features also differ by construction; neither certifies physical time zero. No native WiMDA timing-estimator comparison is recorded here.

Mantid 6.16.1 MuonPreProcess applies supplied time-zero calibration and retains shifted spectrum coordinates. Explicit-table checks establish timing application, rather than automatic feature detection. Asymmetry uses whole-bin transport and carries file fractional residuals; comparisons require matched raw observations and origin conventions. See the MuonPreProcess manual, musrfit manual, and WiMDA overview.

Asymmetry also checks that automatic timing has an established strategy and that every required selected channel supplies its own usable estimate. This protects the requested detector alignment without making an unsuccessful optional search a file-loading failure. Unused failures do not block a valid selection. Missing-file fallback retains other usable file anchors and marks its own estimates explicitly; unresolved requests preserve raw inspection and editable settings. These are Asymmetry safeguards, not a claim about all three reference programs’ failure policies. Their full unknown-source and partial failure behaviour is outside the executed comparison.

Count-background estimates

Asymmetry fits one nonnegative exponential-plus-flat Poisson estimating objective on native bins. The exponential lifetime is fixed, and its shape is averaged across each bin. Profiling the total expected count and solving the bounded mixture score makes equivalent origins on identical observations return the same flat level and likelihood. Output bin merging does not refit this native-bin estimator. On deadtime-corrected or supplied-variance counts, the objective is an estimating convention, not a claim of raw Poisson statistics.

The WiMDA original manual, Sec. 5.8 describes automatic muon-decay-plus-background fitting, separately from manual/file/region options. Mantid’s Corrections tab documentation, version 6.11.0 describes FlatBackground + ExpDecayMuon with fixed lifetime. These method descriptions do not establish the current implementations’ exact weights, constraints, or sparse-count policies. A matched Mantid executable check on an independently generated dense exact exponential-plus-flat mean agrees with Asymmetry. Separate noisy-data checks with WiMDA and Mantid show different estimates and sensitivity to grouping, selected support, and supplied zero-bin errors. The WiMDA checks use paired corrected/uncorrected exports of seeded Poisson draws; Mantid checks use its public Fit algorithm with explicitly supplied count errors. These are selected calculations, not a statistical bias or coverage study or a full Mantid Muon GUI validation. The WiMDA homepage describes the program; the Mantid Corrections tab documentation is version 6.11.0 and is distinct from the compared runtime.

In musrfit 1.12.1 (708f777b, ROOT 6.40.04), asymmetry preparation uses an arithmetic region mean. A jointly fitted single-histogram background is a different operation. See the musrfit RUN settings. Asymmetry estimates its region mean on aligned grouped counts. Equal numeric bounds alone do not establish equal observations under another timing policy.

Asymmetry selects the late half of resolved good-data support on native bins: for inclusive [a,b] it starts at a + (b-a+1)//2. Output merging does not change these observations. Explicit shared/separate intervals are honoured exactly, including values equal to defaults. In the compared musrfit 1.12.1 region controls, selected intervals can shorten and depend on packing. Asymmetry confines its period convention to automatic defaults at native width; shorter-than-period defaults remain intact. Integer rounding and inclusive endpoints do not establish exact whole-cycle exposure. WiMDA’s manual describes region selection, without establishing equal membership in these controls.

Asymmetry propagates a shared estimated level and count-estimator covariance through merged counts. WiMDA’s compared background subtraction leaves exported count errors unchanged. The compared Mantid and musrfit marginal-error controls add fitted/estimated-level variance to individual points. Those controls do not establish shared or cross-bin covariance agreement. These are separate uncertainty conventions. Supplied variances in Asymmetry enter the fitted estimator’s linearised influence; they do not change its central estimating objective.

Asymmetry also checks sufficient fit support, expected-information rank, and finite positive uncertainty. These checks prevent a numerically unestimable level from being reported as usable, even when a solver might report success; they do not test residual physical polarisation or impose a chi-squared gate. Its linearised boundary errors do not promise calibrated coverage. An unusable automatic estimate, including a signed selected tail, leaves the whole requested result background-unsubtracted with a warning and recorded reason. A genuine zero-background optimum remains successful. Valid deadtime correction and supplied variance are retained; loading remains independent of estimability.

The Mantid Corrections tab’s documented Status rule skips a correction when fit chi-square exceeds 10 (version 6.11.0 documentation, separately from the 6.16.1 numerical controls). Asymmetry does not adopt that threshold: its information/finite-uncertainty checks protect its covariance calculation, and it coordinates the whole requested F/B/period result. This does not establish WiMDA’s current failure policy or identical native Mantid status handling. None of these checks establishes physical model adequacy.

See Background correction for the model, current window, and covariance equations.

Calibration persistence

Asymmetry’s project records preserve the effective TF model, count objective, units, correction assumptions, constraints, and full nuisance covariance of an applied core calibration. Saving and loading do not rerun a reference program, change the fit, or establish additional numerical agreement. The component, objective, and conditional-error comparisons above retain their stated versions and conditions. Model selection is an Asymmetry convenience; stored candidate covariances do not measure selection uncertainty or full reduced-data covariance.

Workflow calibration entry points

Explicit workflow/API/CLI TF requests share the documented core operation, model selection, and canonical alpha/beta record. --calibration-method also exposes Diamagnetic, General, and ratio as distinct operations. The no-option explicit CLI calibration uses TF calibration fit. Survey and automatic per-run calibration retain the separate integral-ratio policy, which is not WiMDA’s Diamagnetic or General estimate. These adapters add no reference-program numerical agreement or physical-accuracy claim to the versioned comparisons above. General count-domain fitting remains a separate capability.

Fresh explicit calibration routing

Asymmetry’s GUI, CLI alpha/--alpha-from, and omitted-option explicit API calibration use the shared TF count fit with four-envelope selection. Quick Diamagnetic and Decay balance remain distinct objectives, with stored general identity retained for Decay balance. Finite windows, weights, frequency, and model backgrounds can change the fitted detector balance; different estimators need not agree. The selected envelope and its conditional errors do not establish acquired physical alpha truth or pulse-model adequacy. Asymmetry also checks request freshness before displaying or applying GUI results, so changes to the source, pair, timing, or corrections cannot apply an obsolete result.

Asymmetry retains resolved F/B channel members and authored exclusions with a new calibration, so later regrouping cannot rewrite its source request. Legacy records keep uncaptured facts unknown. Mantid AlphaCalc returns a scalar ratio for the selected groups and window; it does not return a native alpha uncertainty or perform TF envelope fitting.

Asymmetry also checks that a displayed calibration coefficient is finite and positive before offering application. A numerical ratio of zero from an empty forward group cannot be used as a detector calibration; the workflow reports that diagnostic without changing the ratio formula or its error calculation.

Acquired periods and mapped sets

Asymmetry retains the original acquisitions behind a red/green mapping, including ignored periods. Independent supplied count variances add when periods are pooled; ordinary raw Poisson additions remain Poisson. A red-only mapping supports ordinary histogram preparation and period-1 selection. Stored data indices remain distinct from evidenced DAE sequence slots and dwell declarations. See Period mapping.

WiMDA 1.326’s recorded controls map acquired periods into red/green count sets. Mantid’s 6.16 development documentation describes separate period workspaces in MuonPreProcess v1. musrfit 1.12.1 selects period histogram families explicitly in its RUN/grouping manual. These comparisons establish selection behaviour, not identical uncertainties or general facility exposure interpretation. No native Mantid period result is claimed.

With unity calibration and corrections off, equal-total synthetic periods of 20% and 80% give Red/Green 20%/80%, G−R 60%, and G+R 50% in WiMDA 1.326 under Wine 10. musrfit 1.12.1 (708f777b, ROOT 6.40.04) selected/grouped controls give 20%/80%/50%. Asymmetry gives the same pooled G+R 50%, and its independent G−R is 60%. Unequal-total pooled results are not generally an arithmetic mean. The WiMDA value tolerance is 0.00051 percentage points, set by text precision. These controls do not establish matching error conventions.

Asymmetry corrects each acquired constituent before mapped pooling, with its own applicable detector coefficients and supported exposure. It then aligns whole bins using each resolved t0 on common observed support, retains supplied count variances, and applies group-background preparation once to the requested set. The unity-correction controls above do not establish WiMDA’s correction order for nonzero coefficients. Mantid’s documented MuonPreProcess prepares period workspaces; no native period-error comparison is asserted here.

For synthetic F/B counts (80000, 20000) and (10000, 90000), 0.1 μs native bins, 100000 frames per acquisition, and 0.05 μs coefficients, constituent correction then pooling gives −12.7378265%; raw-first correction gives −13.2890365%. musrfit 1.12.1 (708f777b, ROOT 6.40.04) exports −12.7378% for the equivalent histogram occupancies using a stored scalar 200000 frames and 0.1 μs coefficients (absolute tolerance 5×10⁻⁶ in fractional asymmetry). This checks correction order, not acquired per-period exposure interpretation or error equivalence.

Structured shifted-count controls also agree with musrfit grouping and ADDRUN on matched interior observations with explicit integer anchors, corrections off, and native bins at the same tolerance. ADDRUN’s edge and normaliser conventions are not adopted. Asymmetry records fractional residuals without interpolation and also checks common observed support and compatible native widths. One shared explicit α/β belongs to the comparison; there is no implicit separate calibration for each condition. General cross-period/reference likelihoods remain outside the conditional error contract; integrated scalars retain supported shared counting responses.

For G+R, group-background subtraction acts once on the pooled observation; Manual levels retain their requested-observation count units. Automatic reference scaling uses summed evidenced sample exposure, while Manual scale applies once. Asymmetry applies the requested group background to each independently reduced G−R set.

musrfit 1.12.1 provides a scalar RRF workflow. Mantid RRFMuon rotates a prepared transverse pair. These comparisons do not establish a native pooled baseline policy for unequal per-period baselines. Asymmetry also checks that Red and Green transverse baselines agree for pooled G+R rotation, removing the common pair once. Unequal baselines make that rotation unavailable; Lab G+R and rotating Red, Green, and G−R remain available. This is an Asymmetry safeguard. See Rotating-frame projection.

Integrated counting responses and fit support

Asymmetry’s count integral sums all selected native counts before classifying the ratio. Its explicit time-bin mean adopts the ordinary usable-ratio policy, with exclusive omission counts. Undefined ratios and defined zero-variance results have no Gaussian weight. Supported signed observations are retained. The first-order gradients and conditional calibration convention are described in Integral scan mode (avoided-level-crossing field scans) and Errors.

A reused reference contributes one set of counting observations. Its signed responses are combined before contraction, including within a scalar G−R. For independent equal sample periods, each four bins F=600/B=400, subtracting one reference of four bins F=60/B=40 at unity scale gives G−R=0 and standard error 0.024343224778007384. This independently derived linear-propagation control follows from cancellation of the identical reference responses; it is not a native-program comparison. Unequal responses do not cancel in general.

Asymmetry also checks whether the requested independent-error fit reuses known counting observations with nonzero cross-point covariance. That fit is unavailable, while inspection, export, and independent subsets remain available. Unknown identity retains an explicit distinct-observation assumption. This is a bounded safeguard, not a general joint likelihood or calibration covariance.

The documented Mantid AsymmetryCalc error approximation differs from Asymmetry’s derivative propagation. The checked musrfit 1.12.1 (708f777b, ROOT 6.40.04) positive-count packing comparison concerns 22 interior points per case within 5×10⁻⁷ export tolerance; first/endpoints remain qualified. These workflows do not establish a common reused-observation covariance contract. No fresh native WiMDA ALC/RF replay or full Mantid execution is claimed, and the analytic shared-reference control does not extend their agreement.

Shared counting noise in vector rotation

Asymmetry’s vector reduction retains acquisition/reference-bin identity, coefficients, and counting variance across projections. Its transverse rotation propagates this covariance through the actual gain, sense, and phase. Range and tail-background estimator influence survives rotation and display packing; cohort tail-fit uncertainty uses retained acquisition variance while preserving fitted levels. Errors remain conditional on calibration, baselines, gain, and correction values. Corrected marginals do not establish an independent joint likelihood or validate an overlapping-group MaxEnt objective.

In Mantid, RRFMuon v1 provides a comparable two-spectrum coordinate rotation after matching signs, phase, and units. The v1 documentation associated with the 6.16 development snapshot is a formula comparison, not native execution or shared-detector error agreement. The documented scalar RRF workflows use different inputs from this rotation; no equivalent covariance contract for these overlapping ratios is established. Independent analytic and seeded Poisson controls support Asymmetry’s propagation. Acquired ISIS and PSI file execution, including musrfit example ROOT files, checks the changed operation with authored groupings; it does not establish acquired vector geometry or reference-program numerical agreement. See Rotating-frame projection for the equations and limitations.

Scalar rotating-frame magnitude

Asymmetry’s scalar FIR display forms both quadratures from one asymmetry. It propagates their signed same-bin covariance through squared filter weights and includes the cross term in first-order Magnitude errors. Independent identity-filter and measured linear-response controls check this accounting, including phase invariance, heteroscedastic input errors, and finite masks. The errors assume independent supplied scalar input noise; unknown input temporal correlations, low-signal magnitude bias, and filtered-bin correlations remain. Fits continue to use laboratory observations.

Asymmetry retains a separate linear-error core comparison mode. That mode alone does not execute WiMDA. musrfit 1.12.1 scalar RRF preparation and packing differ from this FIR, while Mantid RRFMuon v1 rotates two supplied components. No new native-program magnitude error agreement is established. Execution on unchanged ISIS HDF4/HDF5, PSI BIN/MDU, and musrfit 1.12.1 example ROOT inputs checks the scalar operation with authored pair/frame settings, not acquired physical truth or reference agreement. See Rotating-reference-frame display for equations, sources, and limitations.

Run addition and subtraction

Run arithmetic: co-add and co-subtract compares the count operations and their limits. Asymmetry retains signed coefficients and independent squared-weight variance for ordinary and counterpart-period payloads. Raw independent unit sums retain the Poisson limit; weighted/subtracted or supplied-noise results use retained Gaussian variance and do not acquire a raw-Poisson origin from integer totals. Required original detectors are corrected with their own facts before signed weighting and pooling on whole-bin common observed good support. The raw pooled histograms remain inspection views. Group background and a selected reference act once on the pooled observation.

WiMDA’s original manual, Secs. 2.1 and 3.5, describes count co-addition and subtraction. musrfit 1.12.1 addRun/ADDRUN provides the executed count-addition comparison. Matched correction-off fit-time ADDRUN controls agree on unit-addition and interior asymmetry values from whole-bin-aligned counts within 5e-6 fractional asymmetry; this establishes neither uncertainty nor endpoint equivalence. The tested unequal-fact ADDRUN applies the primary RUN correction context to added data. Asymmetry deliberately retains each input’s correction facts and is invariant under primary-run reversal for shared authored intent. The correction order of WiMDA co-addition is not established by these controls. In Mantid, individual MuonPreProcess workspaces can be prepared before merging; generic merging does not determine correction order. Mantid 6.16.1 Scale/Plus/Minus provide the explicit independent weighted-arithmetic analogue: matched count/error controls reproduce 50/90, supplied 30/74, and signed 190/405. Separate MuonPreProcess deadtime preparation retains each input’s correction context before addition. These operations do not certify raw-Poisson eligibility or acquired-period identity. The owning page links the public manuals and algorithms and distinguishes each operation. Asymmetry retains the first ordered input’s displayed conditions and identity, with finite active constituent ranges; missing master conditions remain unknown. Mantid MergeRuns checks instrument names, spectrum counts, and units and offers configured sample-log merging. Asymmetry uses complete recognised orientation facts and evidenced implicit period permutations, retaining channel/data-order assumptions for absent or incomparable identity. It does not reject on instrument names alone or certify physical identity from matching shapes.

Asymmetry also checks flattened original observation reuse and invalid coefficient/payload requests to protect independent-noise calculations. Known nested repeated acquisitions or subtracted references are unavailable; a shared target reference applied once after sample pooling remains supported. This does not establish general covariance or reconstruct unknown processing history.

Pulse inputs and response origins

Ordinary time-asymmetry models in Asymmetry offer an explicit fixed double-delta response with equal incident strengths, survival weighting, and midpoint origin. The complete source expression is evaluated only at live ages and normalised; point evaluation does not claim integrated-bin or finite-width onset accuracy. The setting follows models through wizard candidates and saved/replayed fits. The WiMDA author’s overview lists data simulation; its current double-pulse population and fit conventions are not numerically established here. Asymmetry’s ordinary response does not supply that generator to the count-based pull diagnostic, which is unavailable for positive separation. Mantid DoublePulseFit’s fitting/error facilities and the musrfit authored-model check do not establish repeated-simulation pull coverage.

Mantid’s documented PulseOffset configures separation. Asymmetry likewise requires fixed supplied separation for its instantaneous count model; single-line post-arrival TF data cannot generally identify it separately from amplitude, phase, and count scale. Source uncertainty is not included in conditional fitted errors. Ordinary asymmetry fitting and TF balance calibration do not gain a finite-width response from this setting. See Count-domain fitting (α calibration & single histograms) for the supported count targets and Fourier analysis for the complex response and explicit origin conversion used in the standalone Mantid kernel comparison. No fresh native WiMDA or full Mantid pulse solve is claimed. An explicit musrfit 1.12.1 post-arrival two-component TF composition is a model control, not a verified generic pulse feature.

Asymmetry also checks active finite-width input domains and rejects missing or ambiguous widths rather than silently reporting identity compensation. These checks concern an authored analysis request; they do not restrict file loading or infer accelerator response from facility names or detector timing fields.

For Fourier compensation, Asymmetry retains the parabolic-plus-pion response shared with its MaxEnt forward model. WiMDA’s original manual, version 1.2, Sec. 9, describes Gaussian Fourier response compensation; these are different models (see the WiMDA original manual, Sec. 9 and Fourier analysis). The Mantid standalone MaxEnt kernel comparison establishes the shared response after an explicit origin conversion, not full native Fourier or MaxEnt reconstruction agreement. musrfit 1.12.1’s authored post-arrival TF composition does not test Fourier Power or its errors; corresponding automatic Fourier compensation is unverified here.

The Power multiplier follows squared-amplitude algebra: for amplitude gain g, amplitudes and their existing standard errors receive g, while Power and its standard errors receive g². Asymmetry also checks a sampled amplitude-gain limit, omitting the first positive frequency below the response floor and all following bins from numerical consumers. This grid-dependent safeguard is not a physical information-loss boundary or a recovery of response phase. Conditional errors do not include uncertainty in the fixed pulse response or full spectral covariance.

Fixed supplied TF count response

Asymmetry’s TF alpha/beta calibration uses equal incident instantaneous births, absolute muon survival, and complete age-local polarisation, with each flat background added once. Beta is fixed at one or fitted as a backward amplitude ratio. All four envelopes share the same source settings, observations, and objective; minimum summed F+B cost selects a usable candidate. Counts are predicted at packed-bin centres, without rate integration. Errors include all fitted nuisances and are conditional on fixed source and selection assumptions.

WiMDA’s native pulse-calibration equivalence is not established here. Mantid DoublePulseFit v1 provides supplied shifted functions and weights, a forward-model analogue, without establishing the same count calibration or covariance. musrfit 1.12.1 count models use an authored theory and reciprocal count balance; comparisons must match amplitude units, phase orientation, lifetime, and objective. Existing post-arrival shifted-TF polarisation agreement does not validate onset, bin integration, or calibration errors. No four-program estimator or uncertainty agreement is claimed. See Alpha calibration for the count equations.

Matched musrfit 1.12.1/ROOT 6.40.04 global-histogram likelihood controls on rounded double-pulse counts agree within 0.0001 absolute alpha/beta for beta fixed at one and free beta. The comparison uses 16 ns bins, raw bins 13–498, zero fixed backgrounds, fixed lifetime, a shared free phase (native degrees with a 180-degree backward offset), and reciprocal native count balance. The native post-arrival damped-cosine representation absorbs the supplied source into its amplitude, phase, and normalisation. This checks point estimates after both births; it does not test native onset, bin integration, or error equivalence. Independent count and full-nuisance information controls provide the separate scientific targets.

Count-integral condition scans

Asymmetry projects capture completed baseline, peak, and RF fit results and restore their numerical readouts and sampled overlays without optimisation. Settings-only histories require explicit fitting for new results. This is an Asymmetry history safeguard; the comparisons below do not establish identical project-restoration behaviour in WiMDA, Mantid, or musrfit.

Asymmetry runs baseline, peak, and RF fitting and prediction in cancellable background tasks and publishes only a complete applicable result. This does not imply identical Stop, threading, or saved-execution behaviour in other programs. A bounded Mantid 6.16.1 Fit v1 comparison on 61 owned scalar observations uses the same supplied Gaussian errors, a linear background, and a Gaussian peak. Mantid’s Height, PeakCentre, and Sigma match Asymmetry’s amplitude, centre, and Gaussian width; neither amplitude is an integrated area. Predictions agree within 0.005 of a supplied point standard error and fitted parameters within 0.02 of the native standard errors. This checks matched least-squares controls, not the specialised RF model or Asymmetry’s two-stage fitted-baseline propagation, and does not equate the programs’ cancellation granularity.

Asymmetry sums the selected corrected native F/B counts before forming \((F-\alpha B)/(\beta F+\alpha B)\) for ordinary and RF scans. G−R is \(A_G-A_R\) with shared chosen calibration; G+R pools prepared counts before the ratio. The explicit differential method is an equal-time-bin mean, distinct from a field derivative. Windows include whole bins by native centres; errors remain conditional on calibration. See Integral scan mode (avoided-level-crossing field scans).

Mantid’s PlotAsymmetryByLogValue v1 documentation, version 6.11.0 describes asymmetry versus a sample log, including period outputs. It is a related scan workflow, without a matched native count-integral execution here. That description alone does not establish identical dual-period observables, window membership, or errors. The scalar Fit control above compares fitted scan points, not the algorithm that constructs them.

The WiMDA original manual, Sec. 11.1 describes LCR scans made with histogram bunching, cursor selection, and batch fit tables. This is a useful time-window workflow comparison; it does not establish the dedicated ALC-column or Differential ALC algorithms in WiMDA 1.326. No native WiMDA ALC/RF numerical replay is claimed here. Asymmetry’s adjacent field-table derivative uses pair midpoints, positive spacing, an explicit optional max_gap, and fractional or percent units per selected G/K/run unit.

musrfit 1.12.1 (revision 708f777b, ROOT 6.40.04) packs counts before forming its fitting ratio, as described in its user manual. Executed positive-count and alternating-count controls with α=β=1, fixed zero background, and packing 20 agree on 22 interior values/errors per case within 5 × 10⁻⁷ (ASCII precision). The first packed value differs and the final nominal group is absent. Those unresolved boundaries are excluded from agreement; no whole-output or four-program numerical equivalence is claimed. Non-unit balance comparisons must also match musrfit’s reciprocal-alpha convention and fitting-coordinate transformation.

Asymmetry propagates retained shared counting responses into each scalar, with errors conditional on the chosen calibration and corrections. Its two-stage baseline/peak inference propagates the fitted baseline’s uncertainty with fixed model and regions; this does not establish an identical covariance contract in the other programs. Asymmetry also checks actual fit selections for unusable Gaussian observations and known nonzero source covariance. An independent selected window remains usable. It also restricts its scan fit adapters to original observations: adjacent derivatives share their middle observation, and marginal errors alone do not establish an independent likelihood. Derivative display/export remains available, with unavailable errors for known dependent input pairs. These are operation-specific Asymmetry support limits, not universal prohibitions on derivative or correlated-data analysis.

Effective reduction records

Asymmetry captures an immutable effective recipe with core reductions and supported transforms; see Count reduction and transformed signals. Work-directory, scan, project-result, and supported export records retain the captured account. It is distinct from raw replay, acquisition-content verification, and cache validity. Private records retain locators; publishable recipe fields omit nested locators and diagnostic prose without changing numerical identity.

musrfit 1.12.1 uses an explicit analysis configuration and writes fitted parameters to its .msr result while .mlog retains the preceding parameter input (manual). A bounded native 1.12.1/ROOT 6.40.04 request/result-file check confirms that behaviour; it does not establish acquisition hashing or general reduction equivalence. Mantid’s public workspace-history script generation describes applied operations separately from saved data and external replay inputs. A recorded Mantid 6.16.1 control saves and reloads an owned processed-NeXus workspace across processes, retains its Scale/Rebin history, and generates the applied operations as a script. This does not establish full Workbench .mantidproj restoration, matched Asymmetry project replay, or identical canonical analysis identity. WiMDA 1.326 saves and restores grouping settings and annotates reduced exports with grouping and corrections; its overview/manual supplies workflow context. These workflows do not establish identical canonical identity safeguards.

Asymmetry also checks canonical recipe identity independently of dictionary order, local paths, and display labels. The identity retains available source references; it does not verify file contents or recover unknown historical settings.

Asymmetry also checks actual sampled observations, retained errors, usable support, and interpretation when comparing saved single fits. Unknown or different input histories remain available without a shared ranking or current residuals. This safeguard is separate from model structure and recipe identity; it does not change the fitting objective, information-criterion definitions, or TF calibration selection.

Asymmetry also checks an explicit scientific-computation revision before automatic persistent reuse and current saved-result comparisons. This is an additional compatibility safeguard, separate from recipe, observation, and source identity. Historical results retain their old or unknown revision; ordinary documentation and package-version changes do not invalidate calculations. The operation-history and request/result-file behaviours described above do not establish an identical automatic computation-revision contract in WiMDA, Mantid, or musrfit. This check adds no muon-spin estimator and makes no new native numerical-equivalence claim.

Asymmetry also checks complete acquisition contents through an explicit cancellable core comparison API. Captured expected identities and newly observed candidate bytes remain separate; identical relocated bytes match independently of timestamp, and legacy expectations remain unknown. Capture binds the fact to a stable new read; a retained reference cache keeps the identity of its own counts. This is an additional Asymmetry provenance safeguard, not an established identical policy in WiMDA, Mantid, or musrfit. It changes no reduction estimator and makes no new native numerical-agreement claim. Project reopening runs this comparison before application, with explicit history, selected current-input replacement, and cancellation choices for known changed inputs. Saved historical records and calibration expectations remain intact; historical-only differences do not block valid active inputs. API/CLI actions use the same owner-aware core operation. These choices do not establish an identical replacement workflow in the other programs.

Conditional simulation and supplied-error refits

Asymmetry’s Conditional simulation: the pull distribution records attempted and usable seeds, explicit refit failures, and the actual ideal synthetic reduction separately from source context. Its descriptive mean/width do not certify acquired-data uncertainty or interval coverage. WiMDA’s documented generate test data and refit workflow is the closest operation; an identical automated seeded pull product is not established for the checked WiMDA 1.326 environment.

The supplied-error constant control retains all three observations, native stationary status, objective scaling, and recorded precision. Its absolute tolerances are \(10^{-7}\) for Mantid and \(5\times10^{-5}\) for musrfit. It concerns supplied Gaussian errors; the separate count-likelihood error-scale difference does not apply. No package-wide pull, coverage, or acquisition-chain agreement follows.

References

Software references:

For the underlying physics see S. J. Blundell, R. De Renzi, T. Lancaster, and F. L. Pratt, Muon Spectroscopy: An Introduction (Oxford University Press, Oxford, 2022).