DFM-2015-R03 Transferring the unit of mass between weights ... · DFM‐2015‐R03 Page 2 of 16 O...

16
DFM‐2015‐R03 Page 1 of 16 Transferring the unit of mass between weights kept in air and in vacuum Lars Nielsen, DFM Abstract In the current watt balance and crystal density measurements used to assign a value to the Planck constant h in terms of the international prototype of the kilogram , the unit of mass has to be transferred from air to vacuum. In the future, when the kilogram has been redefined in terms of a fixed value of the Planck constant, the unit of mass has to be transferred from vacuum to air. This report describes how this mass transfer may be done in a way that enables evaluation of the standard uncertainty associated with the mass transfer. 1. Introduction The unit of mass, kilogram is currently defined as the mass of the international prototype of the kilogram immediately after cleaning and washing using a specified procedure involving solvents and steamሾ1ሿሾ2ሿሾ3ሿ. As the prototype is kept and used in humid air, a natural layer of adsorbed water on the surface of the artefact is contributing to the unit of mass. When measuring the Planck constant in the current watt balance experiments or the Avogadro constant in the current crystal density experiment, mass standards ሺweightsሻ that are traceable to the prototype but used in vacuum are required. Once a new definition of the kilogram in terms of a fixed value of the Planck constant is in place, the kilogram will be realized in vacuum and needs to be transferred to air before being disseminated. A gravimetric method is often used to measure the mass change due to water sorption/desorption associated with the transfer of weights between air and vacuum. This method uses at least two mass standards ሺknown as sorption artefactsሻ having the same nominal values of mass and volume, the same surface finish, but a large difference in their surface areas. Based on the assumption that the sorption per unit area is the same for the two sorption artefacts, the change in sorption can be calculated from a measured change in mass difference between the two sorption artefacts when transferred between air and vacuum. Experiments have shown that the sorption is not always reversible. The modelling described in this report accounts for that. 2. Modelling Assume that sorption artefact S is cycled repeatedly between the two media air ሺAሻ and vacuum ሺBሻ. The cycling is described by a media sequence ܯൌ ሺA, B, A, B, … ሻ, so that ܯA, ܯB, ܯA, etc. When the sorption artefact is transferred from medium ܯ to medium ܯାଵ the mass of the artefact is assumed to change according to the model: ,ାଵ , ܣ ݏ , ሺ1ሻ where , is the mass of the artefact S in medium ܯ , ,ାଵ is the mass of the artefact S in medium ܯାଵ , ܣis the geometrical surface area of the artefact S,

Transcript of DFM-2015-R03 Transferring the unit of mass between weights ... · DFM‐2015‐R03 Page 2 of 16 O...

Page 1: DFM-2015-R03 Transferring the unit of mass between weights ... · DFM‐2015‐R03 Page 2 of 16 O Ü is the change in sorption per unit surface area of the artefact when transferred

DFM‐2015‐R03 Page1of16

TransferringtheunitofmassbetweenweightskeptinairandinvacuumLarsNielsen,DFM

Abstract

InthecurrentwattbalanceandcrystaldensitymeasurementsusedtoassignavaluetothePlanckconstanthintermsoftheinternationalprototypeofthekilogram ,theunitofmasshastobetransferredfromairtovacuum.Inthefuture,whenthekilogramhasbeenredefinedintermsofafixedvalueofthePlanckconstant,theunitofmasshastobetransferredfromvacuumtoair.Thisreportdescribeshowthismasstransfermaybedoneinawaythatenablesevaluationofthestandarduncertaintyassociatedwiththemasstransfer.

1. IntroductionTheunitofmass,kilogramiscurrentlydefinedasthemassoftheinternationalprototypeofthekilogram immediatelyaftercleaningandwashingusingaspecifiedprocedureinvolvingsolventsandsteam 1 2 3 .Astheprototype iskeptandusedinhumidair,anaturallayerofadsorbedwateronthesurfaceoftheartefactiscontributingtotheunitofmass.WhenmeasuringthePlanckconstant inthecurrentwattbalanceexperimentsortheAvogadroconstant inthecurrentcrystaldensityexperiment,massstandards weights thataretraceabletotheprototype butusedinvacuumarerequired.OnceanewdefinitionofthekilogramintermsofafixedvalueofthePlanckconstantisinplace,thekilogramwillberealizedinvacuumandneedstobetransferredtoairbeforebeingdisseminated.

Agravimetricmethodisoftenusedtomeasurethemasschangeduetowatersorption/desorptionassociatedwiththetransferofweightsbetweenairandvacuum.Thismethodusesatleasttwomassstandards knownassorptionartefacts havingthesamenominalvaluesofmassandvolume,thesamesurfacefinish,butalargedifferenceintheirsurfaceareas.Basedontheassumptionthatthesorptionperunitareaisthesameforthetwosorptionartefacts,thechangeinsorptioncanbecalculatedfromameasuredchangeinmassdifferencebetweenthetwosorptionartefactswhentransferredbetweenairandvacuum.

Experimentshaveshownthatthesorptionisnotalwaysreversible.Themodellingdescribedinthisreportaccountsforthat.

2. ModellingAssumethatsorptionartefactSiscycledrepeatedlybetweenthetwomediaair A andvacuum B .Thecyclingisdescribedbyamediasequence A, B, A, B, … ,sothat A, B, A,etc.Whenthesorptionartefactistransferredfrommedium tomedium themassoftheartefactisassumedtochangeaccordingtothemodel:

, , , 1

where

, isthemassoftheartefactSinmedium ,

, isthemassoftheartefactSinmedium ,

isthegeometricalsurfaceareaoftheartefactS,

Page 2: DFM-2015-R03 Transferring the unit of mass between weights ... · DFM‐2015‐R03 Page 2 of 16 O Ü is the change in sorption per unit surface area of the artefact when transferred

DFM‐2015‐R03 Page2of16

isthechangeinsorptionperunitsurfaceareaoftheartefactwhentransferredfrommedium tomedium ,

isafactorforthesorptionartefactSconvertingitsgeometricsurfaceareatoaneffectivesurfaceareaforthesorption.

Notethatthequantity maybenegativeorpositivedependingonthemediabetweenwhichtheartefactistransferred.

Ifasetof sorptionartefactsS , S , . . . , S arecycledbetweenthetwomedia,equation 1 isreplacedby

, ,

, ,

⋮, ,

2

Thesuccessfuluseofsorptionartefactsrelyontheassumptionsthatthechangeinsorption isthesameforallsorptionartefactsinvolvedinthesametransferbetweenmediaAandB.Inordertotestthatassumption,atleastthreesorptionartefactsneedtobeused.Uptonowithasbeenusualpracticetoassumethatthefactors , … , areidentical andequalto1exactly .Thisimpliesthatthe

sorption isdefinedintermsofgeometricarearatherthanintermsofeffectivearea,andthattheratioofeffectiveareatogeometricalareaisthesameforallsorptionartefactsinaset.Thefactors

, … , canbeinterpretedastheratiosofeffectiveareatogeometricareasofthesorption

artefacts.Ifthesurfaceroughnessesoftheartefactshavebeenmeasured,someinformationabouttheseratiosisavailable.Inthatcase,bestestimates largerthan1 andassociatedstandarduncertaintiesmaybeassignedtothefactors , … , .Ifnoinformationoftheeffectiveareasofthe

artefactsisavailable,variationinthesurfacefinishofthesorptionartefactsinasetmaybetakenintoaccountbyassigningapriorvalue1toallfactors , … , butwithnon‐zerostandarduncertainties.

AssumethatoneweightM ispermanentlystoredinair mediumA ,andanotherweightM ispermanentlystoredinvacuum mediumB ;bothweightshavethesamenominalmassvaluesasthesorptionartefactsS , S , . . . , S .Thegoalistomeasurethemass oftheweightM storedinvacuumintermsofthemass oftheweightM storedinair,orviceversa.ThevolumesoftheweightsM andM aredenoted respectively ,andthevolumesofthesorptionartefactsS , S , . . . , S aredenoted , , . . . , .

Atstage inthemediasequencethesorptionartefactsS , S , . . . , S arecomparedwitheachotherandwitheithertheweightstoredinairortheweightstoredinvacuum.Thiscomparisonismodelledby

∆ 3

where

,, , , , … , , isavectorofthemassvaluesoftheweightsandthesorption

artefactsatstage inthemediasequence, , , , … , isavectorofthevolumevaluesofthemassandsorptionartefacts

involvedinthecomparisons, isthedesignmatrixaccordingtowhichthemassandsorptionartefactshavebeencomparedat

stage inthemediasequence, isthevectorofairdensitiesmeasuredforeachmasscomparisonatstage inthemedia

sequence,

Page 3: DFM-2015-R03 Transferring the unit of mass between weights ... · DFM‐2015‐R03 Page 2 of 16 O Ü is the change in sorption per unit surface area of the artefact when transferred

DFM‐2015‐R03 Page3of16

∆ isthevectorofindicationdifferencesmeasuredforeachmasscomparisonperformedatstage ofthemediasequence,

isthescalefactorofthemasscomparatoratstage ofthemediasequence.

Whenweighingisdoneinair A ,theweightM storedinvacuumisnotinvolvedinthemasscomparison,soif A,thentheelementsofthesecondcolumnof areallzero.Similarly,whenweighingaredoneinvacuum B ,theweightM isnotinvolvedinthecomparisons,soif B,thentheelementsofthefirstcolumnof areallzero.Themasses and areassumedtoconstant;thisassumptionisplausibleonlyiftheweightM iskeptpermanentlyinair,andtheweightM iskeptinpermanentvacuum.Whenweighingisdoneinvacuum B ,theairdensityisnegligible,soif B,then .

Theequations 2 and 3 formthenecessarybasisfortransferringtheunitofmassbetweenairandvacuum.TheseequationsareeasilysolvedbythegeneralmethodofleastsquaresdevelopedatDFM 4 5 ,knownasDFM‐LSQ.Toillustratethis,considerthatthemassoftheweightM keptinairisknown,andthatwewanttomeasurethemassoftheweightM keptinvacuumusingtwosorptionartefactsS andS .Iftheshortestpossiblemediasequence A, B isapplied,threemassdifferencesmaybemeasuredinair,

Δ ,

Δ ,

Δ ,

1 1 01 0 10 1 1

,

,

, , 0

, 0 ,

0 , ,

,

,

, 4

andthreemassdifferencesmaybemeasuredinvacuum:

Δ ,

Δ ,

Δ ,

1 1 01 0 10 1 1

,

,

. 5

Themassesofthesorptionartefactsinairandinvacuumarerelatedtoasingle negative sorptioncoefficient throughtheequations

,

,

,

,6

Inthismeasurementthereare 6quantities,forwhichnopriorinformationisavailable,

,,,

,,

,,

,, , 7

and 16quantities

, , Δ , , . . , Δ , , , ,,,

,, , , , 8

forwhichwehavebestestimates andanassociatedcovariancematrix .Amongthequantities andthereare 8constraintsontheform

, , 9

whicharederivedfromequations 4 , 5 ,and 6 bysubtractiontherighthandsidesfromthelefthandsidesoftheequations.Followingref. 5 ,bestestimates and ofthequantities and ,aswellastheassociatedcovariancematrix,isfoundbyminimisingthechi‐squarefunction

10

subjecttotheconstraint , .Theconsistencyofmodelanddatashouldbecheckedbycomparingtheobservedchi‐squarevalue

Page 4: DFM-2015-R03 Transferring the unit of mass between weights ... · DFM‐2015‐R03 Page 2 of 16 O Ü is the change in sorption per unit surface area of the artefact when transferred

DFM‐2015‐R03 Page4of16

11

withitsexpectationvalue,whichisequaltothedegreesoffreedom oftheleastsquaresadjustment:

, 12

where isthenumberofconstraintsand numberofquantities,forwhichnopriorinformationisavailable.Assumingthat istheoutcomeofachisquaredistribution ,consistencybetweendataandmodelmaybequestionedif

Pr , 13

where istheprobabilitythat bychance theoutcomeofachi‐squaredistribution islargerthanthevalueactuallyobserved,and isachosenlevelofsignificance.Itisrecommendedtousealevelofsignificanceintherange0.1 0.2.

Thelargerthenumberofdegreesoffreedomare,thestrongeristheconsistencytestexpressedbyequation 13 .Inthecasedescribedabove,wehave 8 6 2degreesoffreedomfortheleastsquaresadjustment;thesetwodegreesoffreedomwouldreducetozero,ifthetwo redundant masscomparisonofthesorptionartefactsinequations 4 5 wereomitted.Evenifthetworedundantmasscomparisonswereincluded,therewouldbenoredundancyinthedeterminationofthesorptioncoefficient.Inordertoobtainsuchredundancy,athirdsorptionartefactwouldhavetobeincluded.Assumingthatthemassvalues and areconstant,furtherredundancywouldbeobtainedbyexpandingthemediacyclefromjusttwostagestothreeormorestages.

Inadditiontothegeneraltestforconsistencydescribedabove,itisrecommendedtotestforconsistencybetweeneachmeasuredvalue inthearray withthecorrespondingadjustedvalue inthearray .Thisisdonebycalculatingthenormaliseddeviations

, 1, … , . 14

Measuredvalues forwhich| | 2arepotentialoutliers,althoughthereisaprobabilityofabout0.05thatthiscouldhappenbychance,assumingthat followsanormaldistributionN 0,1 .

3. AnalysisofsorptiondataUnfortunatelytherearenodataavailableforamasstransferbetweenmassandvacuum,whichhasbeencarriedoutusingtherecommendedprocedure.However,inasorptioncomparisoncarriedoutinthetaskgroupCCMWGMTG1,datawerereportedthathavebeenusedtodemonstratehowthemethodworksandhowaccurateitis.

IntheTG1sorptioncomparisonthreesorptionartefactS , S ,andS werecirculatedamonganumberofparticipants.ThesorptionartefactS wasanintegralweightofstainlesssteelidentifiedas‘71DD’,S wasastackoftwostainlesssteeldiscsidentifiedas‘Stack1’,andS wasastackoffourstainlesssteeldiscsidentifiedas‘Stack2’.Themeasuredvolumesandareasarelistedintable1.

Table1.Measuredvolumesandsurfaceareasofsorptionartefacts

S1 S2 S3V /cm3 126.7730 126.7737 126.7700

u (V )/cm3 0.0011 0.0014 0.0010

A /cm2 140.0 188.3 285.4

u (A )/cm22.3 2.6 3.3

K48(m1 kg)/mg 0.160

u (m )/mg 0.008

V /cm346.4879

u (V )/cm30.0005

Page 5: DFM-2015-R03 Transferring the unit of mass between weights ... · DFM‐2015‐R03 Page 2 of 16 O Ü is the change in sorption per unit surface area of the artefact when transferred

DFM‐2015‐R03 Page5of16

Table2.MasscorrectionandvolumeofDanish nationalprototypeK48Eachparticipatinglaboratorydidmeasurementsinair mediumA andvacuum mediumB usingamediacycle A, B, … , withatleastonemeasurementinvacuumandtwomeasurementsinair.Thelaboratoriescomparedthethreesorptionartefactswitheachotherinairandinvacuum,andwithanunspecifiedreferencestandardkeptinair.Thesorptionartefactswerenotcomparedtoaweightkeptincontinuousvacuum.Ateachstageinthemediasequence,thelaboratoriesreportedthemeasuredmassvalueof andthethreemassdifferencesmeasuredbycomparingS , S ,andS .Theaverageairdensitiesassociatedwiththemasscomparisonsinairwerealsoreported.

Inordertoassesstheuncertaintyatwhichthemassofanationalprototypekeptinaircouldbetransferredfromairtovacuum,theindicationdifferences,whichwouldhavebeenobservedbycomparingtheintegralsorptionartefactS withtheDanishprototypeK48inair,werecalculatedusingtheairdensitiesandmassvaluesofS reportedbythelaboratories,thevolumeofS listedintable1andthemass andvolume oftheprototypelistedintable2.Thecalculationwasdoneusingequation 3 with 1 exactly .Similarly,allreportedmassdifferencesmeasuredinairwereconvertedintoindicationdifferences.

Measurementdatawerereconstructedthiswayforfiveparticipatinglaboratoriesthatprovidedthenecessaryinformation:BIPM,LNE,PTB,METAS,andINRIM.ThereconstructedmeasurementdatawereanalysedusingDFM‐LSQ.Acommonstandarduncertainty Δ 0.001mgwereassignedtothereconstructedindicationdifferencesΔ ,andforthefactors , , usedtodescribevariations

inthesorptionefficienciesofthesurfacesofthesorptionartefacts,apriorvalueequalto1withstandarduncertainty 0.1wasassignedtoeachsorptionartefact.

Theresultsoftheanalysisofthereconstructedmeasurementdataareshowninfigure1inthecaseofINRIM.Thefigureindicatesthatthesorptionisnotreversible;whatisremovedfromthesurfaceinthefirsttransitionfromairtovacuumislessthanthatremovedinthesubsequenttransitionsfromairtovacuum.Furthermore,moreisaddedtothesurfacebythetransferfromvacuumtoairthanthatremovedbytheprecedingtransitionfromairtovacuum.Asaresult,themassinairofthesorptionartefactsincreasesfromonestageinairtothenext,andthesameistrueforthemassinvacuum.Thisiswhyitisrecommendedtocomparethesorptionartefactswithaweightpermanentlykeptinvacuumaswellaswithaweightpermanentlykeptinair.Bydoingsoitcanbetestedifthemassdifferencebetweentheweightkeptinairandtheweightkeptinvacuumisconstantwithinthemeasurementuncertainty.

InthecaseofINRIM,thedegreesoffreedomoftheleastsquaresanalysisis 21,andtheobservedchi‐squarevalueis 2.5.Inthiscase,thereisnoindicationofinconsistenciesbetweendataandmodel.Thisisconfirmedbythenormaliseddeviationsshowninfigure2,whichallfallswellwithintherange| | 2.Iftheprioruncertaintiesofthefactors , , arereducedbyafactorof10 from

0.1to 0.01 ,theobservedchi‐squarevalueincreasesfrom 2.5to 11.1,whichalsonotlargerthanexpected.However,thenormaliseddeviationsassociatedthesurfaceareas

, , andthefactors , , failtomeetthecriteria| | 2byasignificantmargin

2.4 | | 3.1 .Thisindicatesthatthereisasignificantdifferenceinthesorptionefficienciesofthesurfacesofthethreesorptionartefacts.Hadonlytwosorptionartefactsbeenused,suchadifferencecouldnothavebeendetected.

Thestandarduncertaintiesofthemassvaluesfoundforthesorptionartefactsinairisabout0.010mg,whereasthestandarduncertaintiesofthemassvaluesfoundinvacuumareabout0.008mg.This

Page 6: DFM-2015-R03 Transferring the unit of mass between weights ... · DFM‐2015‐R03 Page 2 of 16 O Ü is the change in sorption per unit surface area of the artefact when transferred

DFM‐2015‐R03 Page6of16

meansthattheuncertaintyassociatedwiththemasstransferfromairtovacuum orfromvacuumtoair isabout0.006mg.

FormtheanalysisofthedataprovidedbyBIPM,LNE,PTB,METAS,andINRIMintheCCMWGMTG1sorptioncomparison,fivesetsoffactors , , werefoundforthecirculatedsorptionartefacts

S , S ,andS .Thesefivesetsareshowninfigure3.Itisnotedthatthefivesetsofvaluesareconsistenttakingintoaccounttheassociatedstandarduncertainties,andthatsorptionartefactS seemstohaveasorptionefficiencythatisabout10%higherthanS andS .

Figure1.Thesorptioncoefficients andmasses , , , , , ofthesorptionartefactsS , S ,andS measuredbyINRIMattheninestagesofthemediasequence A, B, A, B, A, B, A, B, A ,wheremediumAisair oddvaluesofstageindex andBisvacuum evenvaluesofstageindex .Theerrorbarsindicatestandarduncertainties.

Figure2.Thenormaliseddeviationsassociatedwiththe48measuredquantitiesinvolvedintheleastsquaresadjustment.

1.50

1.52

1.54

1.56

1.58

1.60

1 2 3 4 5 6 7 8 9(m

i1 kg)/m

gStage in media sequence, i

Mass of S1

‐0.0003

‐0.0002

‐0.0001

0.0000

0.0001

0.0002

0.0003

1 2 3 4 5 6 7 8

s i/(mg/cm

2)

Stage in media sequence, i

Sorption coefficient

1.64

1.66

1.68

1.70

1.72

1.74

1 2 3 4 5 6 7 8 9

(mi1 kg)/m

g

Stage in media sequence, i

Mass of S2

1.40

1.42

1.44

1.46

1.48

1.50

1 2 3 4 5 6 7 8 9

(mi1 kg)/m

g

Stage in media sequence, i

Mass of S3

‐2.00

‐1.00

0.00

1.00

2.00

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47

d

Measured quantity no.

Normalised deviations

Page 7: DFM-2015-R03 Transferring the unit of mass between weights ... · DFM‐2015‐R03 Page 2 of 16 O Ü is the change in sorption per unit surface area of the artefact when transferred

DFM‐2015‐R03 Page7of16

Figure3.Thefactors , , measuredbyfivelaboratoriesforthesamesetofsorptionartefactsS , S ,andS .Theerrorbarsindicatestandarduncertainties.

4. SimulationstudyInordertovalidatethemasstransferfromairtovacuumusingthemodeldescribedinsection2,asimulationstudywascarriedout.ThestudyinvolvedoneweightM permanentlystoredinairmediumA ,oneweightM permanentlystoredinvacuum mediumB ,and 3threesorptionartefactsS , S ,andS cycledbetweenairandvacuumusingthemediasequence A, B, A, B, A, B, A, B, A .Ateachstageinthemediasequence,allsixpossiblemassdifferencesbetweenthethreesorptionartefactsandtheweightkeptpermanentlyinthemediawereassumedtobemeasured.

4.1. SimulationofdataInthefirststep,truevalueswereassignedtothefollowingquantities:

Themass oftheweightkeptpermanentlyinvacuum. Themass andvolume oftheweightkeptpermanentlyinair. Thevolumes , , ,geometricalareas , , ,conversionfactors , , andinitial

masses,,

,,

,ofthesorptionartefactsS , S ,andS .

Theairdensities , , , , duringmasscomparisonsin assumedtobeconstantduringmasscomparisonsatagivenstageofthemediasequence .

Thebalancescalefactors . Theeightsorptioncoefficients , … , .

Usingequation 2 ,truemassvaluesofthethreesorptionartefactsatstage2–9inthemediasequencewerecalculated,andusingequation 3 atotalof9 6 54truedifferences∆ , , … , ∆ , werecalculated.

Inasecondsteparealisticstandarddeviation wasassignedtoeachquantity exceptforthefactors , , ,andameasuredvalue wasdrawfromanormaldistributionN , ,where

0.80

0.85

0.90

0.95

1.00

1.05

1.10

1.15

1.20

BIPM LNE METAS CEM INRIM

Factors p of sorption artefacts

S1

S2

S3

Page 8: DFM-2015-R03 Transferring the unit of mass between weights ... · DFM‐2015‐R03 Page 2 of 16 O Ü is the change in sorption per unit surface area of the artefact when transferred

DFM‐2015‐R03 Page8of16

isthetruevalueofthequantity assignedorcalculatedinthefirststep.Forthefactors , ,

measuredvaluesequalto1wereassigned.

Atotaloftwentysimulationssimulationwascarriedoutinfourcases:

Case1:Reversiblesorption/desorptionofthesameamountofwater 0.0001mg/cm2 ateachmediatransfer.

Case2:Reversiblesorption/desorptionasinCase1,butacontaminationof0.010mgwasaddedtoallthreesorptionartefactsinthetransferfrommediastage4 vacuum tostage5 air andnotremovedagain.

Case3:Irreversiblesorption/desorptionsimilartothatobservedbyINRIMinthesorptioncomparisondescribedinsection3,seefigure1.

Case4:Irreversiblesorption/desorptionasinCase3,butacontaminationof0.010mgwasaddedtoallthreesorptionartefactsinthetransferfrommediastage4 vacuum tostage5 air asinCase2.

Ineachofthefourcases,thesameseedoftherandomnumbergeneratorwasusedforthesimulations.Thismeansthesequenceofrandomcomponentsisthesameineachofthefourcases;onlythetruevaluesofthequantitiesaredifferent.

Forthemassoftheweightkeptinvacuum,atruevalue 1kgwasassigned,andforthemasskeptinair,thebestestimatesgivenintable2wereusedastruevaluesofthemass andvolume oftheweightkeptinair,whereasthestandarduncertaintiesintable2wereusedasstandarddeviationsforthesimulation.Forthesorptionartefacts,thebestestimatesgivenintable1wereusedastruevaluesofthevolumes , , andthegeometricalareas , , ,whereasthestandarduncertainties

intable1wereusedasstandarddeviationsforthesimulation;twentysetsofconversionfactors, , randomlyselectedfromanormaldistributionN 1, with 0.1wereusedastrue

values,onesetforeachofthetwentysimulations.Asinitialmassvalues,,

,,

,ofthe

sorptionartefacts,thebestestimatesfoundbyINRIMinthesorptioncomparisondescribedinsection3 seefigure1 wereusedastruevalues.ThefiveairdensitiesreportedbyINRIMinthesorptioncomparisonwereusedastruevaluesoftheairdensities , , , , ;thestandarddeviationusedforthesimulationofairdensitieswas 0.00005kg/m .Forsimplicity,thevalues 1 exactly wereselectedforthebalancescalefactors.Forthesimulationoftheindicationdifferences∆ ,thestandarddeviation 0.001mgwasused.

4.2. AnalysisofsimulateddataToeachsimulatedvalue ofaquantity exceptthefactors , , ,astandarduncertainty

wasassigned.Forthefactors , , ,thebestestimate 1andfivedifferentbut

commonstandarduncertainties wereassigned:

1,

, ∈ 0.2, 0.1, 0.05, 0.02, 0.01 .

Eachsimulatedsetofdatawasanalysedfivetimes,onetimeforeachstandarduncertainty assignedtothefactors , , .Onlyoneofthestandarduncertainties isequaltothe

standarddeviation 0.1usedtosimulatevaluesofthefactors , , .Analyseswithlargerand

smallervaluesof wereperformedinordertoassesstheimportanceofselectingarealisticvalueof .Asummaryoftheresultsoftheanalysesperformedareshowninfigure4‐7.

Page 9: DFM-2015-R03 Transferring the unit of mass between weights ... · DFM‐2015‐R03 Page 2 of 16 O Ü is the change in sorption per unit surface area of the artefact when transferred

DFM‐2015‐R03 Page9of16

Incaseofcompletelyreversiblesorption/desorption Case1 ,theadjustedvalueofthemass oftheweightkeptinairisfairlyindependentofthevalueoftheuncertainty assignedtothefactors , , intheleastsquaresanalysis,seefigure4.Onlyinthecase 0.2,wherethe

uncertaintyisoverestimatedbyafactoroftwo,thereareafewcases simulationno.14and19 ,wheretheadjustedvaluesofthemass aresignificantlydifferentcomparedtothecases,where

isequaltoorsmallerthanthestandarddeviation 0.1usedforthesimulationofthefactors , , .Thestandarduncertainty ,thevalueofwhichisindicatedwitherrorbarsinfigure

4,decreasesas decreases,butasthereareothersignificantuncertaintycontributionsto ,e.g.fromthestandarduncertainty 0.0080mgassignedtothemass ofthereferenceweightkeptinair,thedecreasein ispredominantwhen isreducedfrom 0.2,where0.8mg 1.4mg,to 0.1,where0.0096mg 0.0103mg.Inthecase

0.2,thevariationintheuncertainty oftheadjustedvalueofthemass isverylarge,especiallyforthoseadjustedvaluesof thatdiffersmostlyfromthetruevalue.Whathappensinthesecasesisthattheadjustedvaluesof , , areratherextremeinsuchawaythatthe

differencesoftheeffectivesorptionareas , , becomesrathersmall,orevengetthe

wrongsign.Itshouldbenotedthatforallfivestandarduncertainties ,theadjustedvalueofthemass isconsistentwiththetruevaluewithin 2 .Thesameisnottrueforthefactors , , ;forthesefactorsthetruevaluesareonlyexpectedtoberecovered,ifthevalueof is

equaltoorlargerthanthestandarddeviation 0.1usedforthesimulationofthetruevaluesof , , .Thegraphsintherightcolumnoffigure4confirmthat.Thegraphsinthemiddleoffigure

4showthatthetestofconsistenceofdatawithmodelisalmostindependentofthevalueoftheuncertainty assignedtothefactors , , .Atfirstthismightbeseenasaparadox,sincefor

smallvaluesof theadjustedvaluesof , , differsignificantlyfromthetruevalues.The

explanationis,however,thatduetothesignificantuncertaintyintheairbuoyancycorrection,theinformationaboutdifferencesintheeffectivesorptionareas , , cannotbeextracted

fromthemeasuredmassdifferenceswhensorption/desorptionisreversible.

Page 10: DFM-2015-R03 Transferring the unit of mass between weights ... · DFM‐2015‐R03 Page 2 of 16 O Ü is the change in sorption per unit surface area of the artefact when transferred

DFM‐2015‐R03 Page10of16

Case1:Reversiblesorption/desorption

Figure4.ResultsoftheleastsquaresanalysisoftwentysimulationsinCase1withstandarduncertaintya 0.2,b 0.1,c 0.05,d 0.02ande 0.01assignedtothefactors , , .Thegraphtotheleftshowstheadjustedvaluesfoundforthemass ofweightkeptinvacuumcomparedtoitstruevalue.Thegraphinthemiddleshowsobservedchi‐squarevalue comparedtotheexpectationvalue fullline andthecriticalvaluecorrespondingtosignificancelevel 10% brokenline .Thegraphtotherightshowstheadjustedvalues opendots ofthefactors , , comparedtotheirtruevalues closeddots .Errorbarsindicatestandarduncertainties.

a

b

c

d

e

Page 11: DFM-2015-R03 Transferring the unit of mass between weights ... · DFM‐2015‐R03 Page 2 of 16 O Ü is the change in sorption per unit surface area of the artefact when transferred

DFM‐2015‐R03 Page11of16

Case2:Reversiblesorption/desorption+0.01mgcontamination

Figure5.ResultsoftheleastsquaresanalysisoftwentysimulationsinCase2withstandarduncertaintya 0.2,b 0.1,c 0.05,d 0.02ande 0.01assignedtothefactors , , .

a

b

c

d

e

Page 12: DFM-2015-R03 Transferring the unit of mass between weights ... · DFM‐2015‐R03 Page 2 of 16 O Ü is the change in sorption per unit surface area of the artefact when transferred

DFM‐2015‐R03 Page12of16

Case3:Irreversiblesorption/desorption

Figure6.ResultsoftheleastsquaresanalysisoftwentysimulationsinCase3withstandarduncertaintya 0.2,b 0.1,c 0.05,d 0.02ande 0.01assignedtothefactors , , .

a

b

c

d

e

Page 13: DFM-2015-R03 Transferring the unit of mass between weights ... · DFM‐2015‐R03 Page 2 of 16 O Ü is the change in sorption per unit surface area of the artefact when transferred

DFM‐2015‐R03 Page13of16

Case4:Irreversiblesorption/desorption+0.01mgcontamination

Figure7.ResultsoftheleastsquaresanalysisoftwentysimulationsinCase4withstandarduncertaintya 0.2,b 0.1,c 0.05,d 0.02ande 0.01assignedtothefactors , , .

a

b

c

d

e

Page 14: DFM-2015-R03 Transferring the unit of mass between weights ... · DFM‐2015‐R03 Page 2 of 16 O Ü is the change in sorption per unit surface area of the artefact when transferred

DFM‐2015‐R03 Page14of16

Incase2,wherethereversiblesorption/desorptionisdisruptedbyadding0.010mgcontaminationtoallthreesorptionartefactsbetweenstage4and5inthemediacycle,itisonlypossibletoobtainconsistencybetweendataandmodeliftheadjustedvaluesofthefactors , , aresothatthe

differencesineffectivesorptionareas , , areclosetozero.Thisisachievedif

≅ 1.2, ≅ 1and ≅ 0.6asinthegraphtotherightoffigure5a .Thatis,toachieve

consistencyalargevalueof isrequired,suchas 0.2.However,asthedifferencesintheeffectivesorptionareas , , areclosetozero,theuncertainty becomes

extremelylargeasseeninthegraphintheleftinfigure5a .For 0.05,theuncertainty reachesausefullevel,buttheadjustedvaluesof aresignificantlybiasedcomparedtothetruevalue,seefigure5c .Forsmallervaluesof ,thisbiasdisappearsasseeninfigure5d ‐e .However,theresultswouldnot andshouldnot betrustedbytheexperimenter,astheconsistencetestshowsasignificantinconsistencybetweendataandmodel,whichisduetothefactthatacontaminationof0.010mgwasaddedbetweenstage4and5inthemediacyclewithoutaccountingforitinthemodelusedtoanalysethedata.

Contaminationofthesorptionartefactsisaproblemonlyifitisnotuniformlydistributedoverthesurfacesofthesorptionartefacts.Infact,uniformlydistributedcontaminationallowstheratiosoftheeffectivesorptionareastobemeasuredfairlyaccurate.ThisisillustratedinCase3,whereuniformirreversiblesorption/desorptionhasbeenassumed.Asseeninthegraphsintheleftcolumnoffigure6,theadjustedvaluesofthemass oftheweightkeptinvacuumarevirtuallyindependentofthevalueassignedto .Theassociatedstandarduncertainties alsohaveonlylittledependenceofthevalueassignedto intheleastsquaresanalysis;ontheaverageitdecreasesfrom

0.0096mgfor 0.2to 0.0087mgfor 0.01.Incontrasttocase1,thetestofconsistencydependsonthevalueassignedto ,asseeninthegraphsinthemiddlecolumnoffigure6.However,evenfor 0.01theobservedchi‐squarevaluefallsbelowthecriticalvalueforatestofconsistenceata10%levelofsignificanceinelevenoutoftwentysimulations.Asshowninthegraphsintherightcolumnoffigure6,theadjustedvaluesofthefactors , , areconsistent

withthetruevaluesaslongastheuncertainty isequaltoorlargerthanthestandarddeviation0.1usedforthesimulation.For 0.1,thestandarduncertaintiesassociatedwiththefactors

, , areabout30%smallerinCase3thaninCase1.Thereasonisthatthegaininmassofthe

sorptionartefactsfromonestageingivenmediatothenextstageinthesamemediaismeasuredbycomparingthesorptionartefactswiththeweightstoredinthatmedia,whichisassumedtohavethesamemassatallstagesinthemediacycle.Thelargerthesemassgainsare,themoreinformationtherewouldberegardingthevaluesofthefactors , , .

Incase4,theuniformirreversiblesorption/desorptionwasdisruptedbyaddinga0.010mgcontaminationtoallsorptionartefactsbetweenstage4andstage5inthemediacycle.Asseeninthegraphsintheleftcolumnoffigure7,thisleadtoasignificantbiasintheadjustedvaluesofthemass oftheweightkeptinvacuum,whichislargerthelargerthevalueassignedto is.For 0.2,theaveragebiasis0.0300mg,andfor 0.01,theaveragebiasis0.0061mg.Forcomparison,theaveragebiasesfoundinCase3were0.0034mgfor 0.2and0.0028mgfor 0.01.Asseeninthegraphsintherightcolumnoffigure7,theadjustedvaluesofthefactors , ,

seektoreducethedifferencesintheeffectivesorptionareas , , ,justasincase2.It

islesspronouncedinCase4,however,duetotheadditionalamountofinformationabouttheeffectivesorptionareas.Forthemajorityofthesimulations,thechi‐squaretestindicatestheinconsistency

Page 15: DFM-2015-R03 Transferring the unit of mass between weights ... · DFM‐2015‐R03 Page 2 of 16 O Ü is the change in sorption per unit surface area of the artefact when transferred

DFM‐2015‐R03 Page15of16

betweendataandmodelcreatedbyaddingthe0.010mgcontaminationtothedatawithoutaccountingforitinthemodel;seethegraphsintherightcolumnoffigure7.

5. DiscussionWhentransferringtheunitofmassfromairtovacuum,atraditionalassumptionhasbeenthatonlywaterisadsorbedanddesorbed,andthatthattheadsorption/desorptionisreversible.ThesorptioncomparisoncarriedoutinCMMWGMTG1shows,thatadsorption/desorptionisirreversibleingeneral,sothatthesorptionartefactsgainmassesastheyarerepeatedlycycledbetweenairandvacuum.Theirreversibilityseemstobemostpronouncedifthesorptionartefactsarecleanedjustbeforethemediacyclingexperiment.Aftersomecyclesbetweenairandvacuum,thesorption/desorptionseemstoapproachreversibility,andthemassesofthesorptionartefactstendtostabilizeinairaswellasinvacuum.

Thesimulationexperimentshowsthatirreversibilityofthesorption/desorptionisinfactusefulasitenablesameasurementoftherelativeabsorptionefficiencies , … , ofthesurfacesofasetof

sorptionartefactsS , S , . . . , S .Thequantities , … , maythereforebedeterminedbyperforming

anair‐vacuumcyclingexperimentonfreshlycleanedsorptionartefacts.Oncethesequantitieshavebeenmeasured,theirvaluesandassociatedcovariancematrixcouldbeusedasinputtotheanalysisofsubsequentair‐vacuumcyclingexperimentswithoutcleaningofthesorptionartefacts,wheretheirreversibilityofthesorption/desorptionprocessmightbelesspronounced.

Incaseofreversiblesorption/desorption,atleastthreesorptionartefactsareneededinordertotestthehypothesisthatthemassofthesorptionisproportionaltothegeometricalareaofthesorptionartefacts.Ifasorptionartefactismadeupof discs withspacers havingthesamenominalmass,theuniformityofthesorptionefficienciesamongthediscscouldbetestedinaseparateair‐vacuumcyclingexperimentinwhichthemassdifferencesamongthediscs withspacers aremeasuredinairandinvacuum.

Thesimulationswereperformedonanair‐vacuumcyclingexperiment,inwhichtheweightstoredinairwasaplatinum‐iridiumstandard,whereastheweightstoredinvacuumandthesorptionartefactsweremadeofstainlesssteel.ThiswasdoneinordertoevaluatethestandarduncertaintywithwhichthemassofaweightkeptinvacuumcouldbemeasuredintermsofanationalprototypekeptinairusingthesorptionstandardscirculatedintheCCMWGMTG1sorptioncomparison.Thisuncertaintyturnedouttobeontheorderof0.010mgandwasactuallydominatedbythestandarduncertainty0.008mgassignedtotheprototypeitself.Thevalidityofthisuncertaintycanonlybeprovenbyperformingaratherlargenumberofair‐vacuumcyclingexperiments similartotheperformedsimulation thatleadstoconsistentmassvaluesoftheweightstoredinvacuum,takingintoaccountthemeasurementuncertainty.Insuchanexperiment,theweightkeptinairshouldhavethesamenominalvolumeasthesorptionartefactsinordertoreducetheuncertaintyduetobuoyancyeffectswhenmeasuringinair.Forthesamereasonshouldthevolumes orvolumedifferences ofthesorptionartefactsandtheweightkeptinairbemeasuredwiththesmallestpossiblestandarduncertainty.

6. ConclusionAprocedurefortransferringtheunitofmassbetweenaweightkeptinairandaweightkeptinvacuumhasbeenpresented.Theprocedureinvolvestheuseofsorptionartefactsthatarerepeatedlycomparedtotheweightkeptinairandtotheweightkeptinvacuuminanair‐vacuumcycle.Amodelofthemeasurementbasedonhomogeneousabsorption/desorptionhasbeendescribed.Themodel

Page 16: DFM-2015-R03 Transferring the unit of mass between weights ... · DFM‐2015‐R03 Page 2 of 16 O Ü is the change in sorption per unit surface area of the artefact when transferred

DFM‐2015‐R03 Page16of16

hasbeentestedonadaptedmeasurementdataprovidedbyparticipantsinasorptioncomparisonandonsimulateddata.Thesensitivitytomodellingerrorshasbeentestedbysimulationaswell.

Ithasbeenshownthatamassvalueinairhavingastandarduncertaintyof0.008mgcanbetransferredtoamassvalueinvacuumwithastandarduncertaintyof0.010mg,whichmeansthatanuncertaintycontributionof0.006mgfromtheair‐vacuumtransfercanbeachieved,atleastinprinciple.Themodelisbasedonthecrucialassumptionthatsorption/desorptionishomogenousoverthesurfacesofthesorptionartefacts.Ifthisassumptionfailsinreality,themassvaluecalculatedfortheweightkeptinvacuummightbesignificantlybiased.Althoughtheproposedmethodforanalysingthedatafromanair‐vacuumcyclingexperimentincludesatoolfortestingtheconsistencyamongdataandmodel,thereisaratherlargeprobability,thataninvalidassumptionmightnotbedetectedinasingleair‐vacuumcyclingexperiment.

7. References1 BIPM2006TheInternationalSystemofUnits SI 8thedn,Sévres,BIPM2 DavisR2003TheSIunitofmassMetrologia,40 2003 299–3053 GirardG1990ThewashingandcleaningofkilogramprototypesattheBIPM,Sévres,BIPM4 NielsenL1998LeastsquaresestimationusingLagrangemultipliersMetrologia35,115‐18

NielsenL2000Metrologia47,183 erratum 5 NielsenL2002EvaluationofmeasurementsbythemethodofleastsquaresAlgorithmsfor

ApproximationIVedJLevesleyetal UniversityofHuddersfield pp170‐86RecommendationG1 2010 oftheCCM,Considerationofanewdefinitionofthekilogram.