Derivation
of
a
Distribution
Function
of
Relaxation
Times
for
the
(fractal)
Finite
Length
Warburg.
Bernard
A.
Boukamp
UniversityofTwente,Fac.ofScienceandTechnology&MESA+InstituteforNanotechnology,P.O.Box217,7500AE,Enschede,TheNetherlands
ARTICLE INFO
Articlehistory: Received9June2017
Receivedinrevisedform22August2017 Accepted27August2017
Availableonline1September2017
Keywords:
ImpedanceSpectroscopy
DistributionFunctionofRelaxationTimes FiniteLengthWarburg
Simulation
ABSTRACT
AnanalyticDistributionFunctionofRelaxationTimes(DFRT)isderivedforthefractalFiniteLength Warburg (f-FLW, also called ‘Generalized FLW’) with impedance expression: ZfFLW(v)=Z0tanh
(vt0)n(vt0)n.t0isthecharacteristictimeconstantofthef-FLW.Analysisshowsthatforn!0.5(i.e.
theidealFLW)theDFRTtransformsintoaninfiniteseriesofd-functionsthatappearinthet-domainat positionsgivenbytk=t0/[p2(k1/2)2]withk=1,2,3, ... 1.Themathematicalsurfaceareasofthese
d-functionsareproportionaltotk.ItisfoundthattheFLWimpedancecanbesimulatedbyaninfinite
seriescombinationofparallel(RkC0)-circuits,withRk=C0tk1andtkasdefinedabove.Rk=2tkZ0and
C0=0.5Z01.Z0isthedc-resistancevalueoftheFLW.
A fullanalysis of theseDFRTexpressionsis presented and comparedwithimpedance inversion techniquesbasedonTikhonovregularizationandmulti-(RQ)CNLS-fits(m(RQ)fit).Transformationof simplem(RQ)fitsprovideareasonablyclosepresentationint-spaceofthef-FLW,clearlyshowingthefirst twomajorpeaks.ImpedancereconstructionsfromboththeTikhonovandm(RQ)fitderivedDFRT’sshowa closematchtotheoriginaldata.
©2017TheAuthor.PublishedbyElsevierLtd.ThisisanopenaccessarticleundertheCCBY-NC-ND license(http://creativecommons.org/licenses/by-nc-nd/4.0/).
1.Introduction
OverthelastfiftyyearsElectrochemicalImpedance Spectros-copy(EIS)hasbeendevelopedintoanimportantresearchtoolfor studying electrochemically active systems [1]. The principles, practisesand applicationsofEIS havenowbeenlaiddownin a number ofimportant textbooks [2–5]. Impedancespectroscopy hasalsobecomeamajorresearchmethodforcharacterisingSolid Oxide Fuel Cells (SOFC) [6] and Solid Oxide Electrolyser Cells (SOEC)[7]atafundamentallevelforwell-definedhalfcells,aswell asonasystemslevelforcompletecells.Duetothenatureofthe complex,porouselectrodes,wheretheadsorption,chargetransfer andtransportprocessesaredistributedoverasignificantpartof themicrostructure,theimpedancegraphsarenoteasytomodel withso-called equivalentcircuitsusingcomplexnonlinearleast squares(CNLS)fitting routines[8–12]. Recentlythere hasbeen renewed interest into the transformation of the frequency dispersionintoadistributionfunctionofrelaxationtimes(DFRT)
[13–29].EspeciallythegroupofIvers-Tifféehaspioneeredtheuse of DFRT analysis in the study of SOFC’s [15,17,18] and Li-ion batteries[19,21].TheDFRTprovidesamodel-freerepresentationof essential relaxation times that are directly connected to the
physicaltransportand(charge)transferprocesses.ForSOFC,SOEC andotherfuelcells,thestudyofthepositionandthepeakheightas functionoftemperatureand/orpartialpressureofthegasphase components provides insight in the transport processes in the electrodes.InbatteryresearchthechangeoftheDFRTwiththe ‘StateofCharge’providesessentialinformation,whileforSolarCell research the relation between illumination level and the correspondingDFRTcouldbeabletorevealfundamental process-es.Furthermore,itcanbeaninterestingapproachtouseDFRTto studyaginginavarietyofelectrochemicalsystems.
Thetransformationtothe
t
-domainisdefinedby: Zðv
iÞ ¼R1þRp Zþ1 1 Gðt
Þ 1þjv
it
dlnt
ð1ÞZ(
v
)isthemeasuredimpedance,R1isthehighfrequencycut-off resistance, Rp is thepolarization resistanceand G(
t
) is thesoughtdistributionfunctionofrelaxationtimes(DFRT).G(
t
)isa normalizedrealfunctionwith:RG(t
)dlnt
=1.Thet
-domainisthe inverseofthefrequencydomainwitht
=(2p
f)1.Unfortunately, Eq.(1)isknownasanill-posedinverseproblemforwhichmany solutionsarepossible.Imposingthe(logical)restrictionthatG(t
) must always be positive reduces the number of possibilities. SeveralmethodsexisttoarriveataviableDFRT:FourierTransform (FT) [15,17,28,29], Tikhonov Regularization (TR) [16,18–27], E-mailaddress:b.a.boukamp@utwente.nl(B.A. Boukamp).http://dx.doi.org/10.1016/j.electacta.2017.08.154
0013-4686/©2017TheAuthor.PublishedbyElsevierLtd.ThisisanopenaccessarticleundertheCCBY-NC-NDlicense(http://creativecommons.org/licenses/by-nc-nd/4.0/).
ElectrochimicaActa252(2017)154–163
ContentslistsavailableatScienceDirect
Electrochimica
Acta
Maximum Entropy (ME) [13,14] and multiple-(RQ) CNLS-fit, abbreviated‘m(RQ)fit’[28,29].In theFourier Transformmethod awindowfunctionmustbeapplied,whichwilldampenunwanted oscillations [28]. The width and shape of this window has significant influence on the shape of the DFRT. But also the TikhonovRegularizationand theMaximumEntropyrequire the adjustment of a ‘smoothing’ parameter in order to obtain an acceptableDFRT.Recentlyacriticalcomparison betweentheFT andm(RQ)fithasbeenpublishedbythisauthor[28].Inasecond paper [29] a comparison was made between the FT and TR methodsandthem(RQ)fit.FortheTR-transformationsapublicly availableMatLabapplicationwasused[30].
AquitedifferentapproachistakenbythegroupofTsur[31–33]. Theyuseevolutionaryprogramming(dubbed‘ISGP’,ref[33])for constructingadistributionfunctionbasedonaselectedsetfroma largelibraryofplausiblefunctions.Intheevolutionary program-mingthenumberandtypesoffunctions,andtheirparameters,are automatically adjusted until the impedance conversion of the DFRTmatches,withinpredefinedcriteria,theoriginal measure-mentdata.Theclearadvantageisthatnosmoothingparameteris neededinordertoarriveatanoptimalresult.Thedrawbackcould bethelackofaphysicalinterpretationofcertainfunctionsthat havenoclearcounterpartintheimpedancerepresentation.There areafewimpedancefunctionsthatdo haveanexactDFRT:the (RQ)orZARC[2],whichisaparallelcombinationofaresistance andaconstantphaseelement(symbol:Q),withZQð
v
Þ¼Z0ðjv
Þn;theHavriliak-Negami(H-N)response[29,34,35]andtheGerischer
[29,36]orchemicalimpedance[37],whichisaspecialcaseofthe H-Nrelation.Itisimportanttonotethattheparallelcombination of a resistance with a capacitance, (RC), is represented in the
t
-domainbyad
-function.TheareaunderaDFRTcurve,definedasRRG(
t
)dln(t
),isequal tothedc-resistanceoftheimpedancecounterpart(i.e.Rforthe ‘(RQ)’orZARC),whichfollowsfromthenormalizationofG(t
)and Eq.(1).Fora(RC)circuit,however,theresistanceinformationis inaccessibleinthet
-domainastheDFRTbecomesad
-function, althoughthemathematicalsurfaceareaisequaltoRp1.Hence,apurecapacitanceinparallelwitha resistancein theimpedance complicatestheconstructionofaDFRT.Asimpleapproachisto approximatethe
d
-functionbyasharpGaussfunction[28,29]: RGðRCÞðt
Þ R Wp effiffiffiffip
lnðt0 =tÞ W 2 ð2Þ wheret
0=RC.W was set at 0.15 as a good compromise forvisibility(width)andpeakheight(nottoolargewithrespectto other peaks) and small error with respect to the impedance reconstructionfrom theDFRT [28]. A smallervaluefor Wwill decreasethewidthandincreasethepeakheight,whilethearea underthecurveremainsequaltoR.
BesidestheGerischerimpedance,theFinite LengthWarburg (FLW,[38])isalsoacommondiffusionbasedimpedancefunction. Itpresentsthefrequencydispersionforone-dimensionaldiffusion througha layer witha fixed activity for the mobileion atone interface.Stoynov[39]has, basedonaboundedconstantphase element(BCP),derivedafractalformforthefinitelengthWarburg, indicatedherebyf-FLW(seethenextsection).Itseemsthatfor thesetwoboundeddiffusiondispersionsnoexactDFRThasbeen publishedsofar.InthispublicationageneralDFRTforthef-FLWis derivedusingaspecialtransformationmethod[40].Forthisanew frequencydependentvariableisintroduced:
z¼lnð
v
t
0Þ¼lnt
0t
ð3Þ with
t
0thecharacteristictimeconstantforthedispersionrelationthatistobetransformed.TheDFRT,G(
t
),isthenobtainedfromtheimaginary part of the impedance expression by the following transformprocess: Z0Gð
t
Þ¼ 1p
Zimag zþjp
2 þZimag zjp
2 h i ð4Þ Z0 represents the dc-resistance of the impedance function.Using thisapproachitis quitesimpletoderivethewell-known DFRT for a (RQ) circuit [41] from the imaginary part Zimagð
v
Þ¼Z0v
nsinðnp
=2Þ:RGðRQÞð
t
Þ¼2Rp
sinðn
p
Þcoshðnlnð
t
0=t
ÞÞþcosðnp
Þ ð5ÞThecharacteristictimeconstantisgivenby:
t
0=(R/Z0)1/n,Z0istheCPEconstantdefinedabove.Eq.(5) representsasymmetric peak centredaround
t
0 ona log(t
)scale.For n=1, i.e.theCPEbecomesa purecapacitance,a
d
-functionisobtained.Forlower valuesofnthespreadofthepeakincreasesrapidly[28].In the following section the impedance expression for one-dimensional diffusion through a mixed conducting layer is presented.FromthisthefractalFLW,alsoknownas‘generalized FLW’,isderived.Thecompactimpedanceexpressionisdividedinto aseparaterealandimaginarypart,neededforthetransformtoa DFRTusingEq.(4).ItisnotpossibletodirectlyderivetheDFRTfor theidealFLWusingthetransformmethodofEq.(4),asisindicated in Section 4. Therefore first the general DFRT of a f-FLW is developedandthelimitingcasesfor
t
!0andt
!1areexplored. The DFRT for the ideal FLW is then obtained using a special approachforn!0.5.FinallytheDFRT’sobtainedfromTikhonov Regularization and multi-(RQ) CNLS-fits (i.e. ‘m(RQ)fit’) are comparedtotheexactDFRTrepresentations.2.FLWdispersionrelation
The FiniteLength Warburg is derived for a linear diffusion problem with a special boundary condition, see Fig. 1. The electrodeisa mixedionicandelectronicconductor(MIEC).It is assumed thattheelectronicconductivityis sufficientlylarge,so that it can be ignored in the derivation. In the presented 3-electrodesetuptheactivityofthemobileionismeasuredatthe electrode/electrolyte interface with respect to the (standard) activity of the reference electrode. The current through the
Fig.1.ModelfortheFiniteLengthWarburg,basedon1-dimensionalbounded diffusion.Asanexamplethediffusionofamobilecation,M+
,inamixedionicand electronicconductor(MIEC)ispresented.Butthismodelappliesequallywellto oxygeniondiffusion.Theelectrolyte/MIECinterfaceisatx=0.Thebacksideofthe MIECiscoveredbyafixedactivity(concentration)ofthemobileion,aM,eq.The
currentpassesbetweentheworkingandcounterelectrode.Theactivity(potential) ofthemobileionismeasuredbetweenthereferenceandworkingelectrode.
working electrode is provided by the counter electrode. The electrodeareaisA,thethicknessisland ~Disthechemicaldiffusion coefficient.Atthebackplaneatx=l,theconcentration(oractivity) ofthemobileionisfixedtotheequilibriumconcentration,Ceq.(or
fixedactivity,aeq.)Thispresentsanideallyfasttransferofmobile
ionsatthebackplane.Fortheelectrode/electrolyteinterfacethe particlefluxisdescribedbyFick’sfirstlaw:
J¼~DdCdðx;xtÞjx¼0 ð6Þ
Thetimedependentconcentrationintheelectrodeisgivenby Fick’ssecondlaw:
dCðx;tÞ dt ¼ ~D
d2Cðx;tÞ
dx2 ð7Þ
Theboundaryconditionforx=lisexpressedby:
Cðx;tÞjx¼l¼Ceq: ð8Þ
Applyingasmallpotentialperturbation,
D
E,willcauseasmall changeintheconcentration,withC(x,t)=Ceq.+D
c(x,t).Thechangeinactivity,a(t),attheelectrode/electrolyteinterfaceisgivenbythe Nernstrelation,which becomeslinear fora smallperturbation, withzthechargeofthemobileion:
D
E¼RT zFln aðtÞ aeq: ¼ RT zFln aeq:þD
aðtÞ aeq: RT zFaeq:D
aðtÞ ð9ÞTherelationbetweenactivityandconcentrationisgivenbythe ThermodynamicFactor,
G
:D
aD
c da dc¼ a C dlna dlnc¼ a CG
ð10ÞThethermodynamicfactor,
G
,canbeobtainedfromaso-called titrationexperimentwheretheNernstpotential(oroxygenpartial pressure)ismeasuredasfunctionofaddedconcentrationofthe mobile ion. The solution of the set of diffusion and boundary equationsiseasilyobtainedbyLaplacetransform,seeAppendixA. TheLaplacetransformleadsdirectlytotheimpedanceexpression fortheFLW(Eq.(A7)):ZFLWð
v
Þ ¼ Z0 ffiffiffiffiffiffiffiffiffiffiffi jvt
0 p tanh ffiffiffiffiffiffiffiffiffiffiffijvt
0 p with: Z0¼ RTl z2F2 ACeq~D d ln a d ln c and:t
0¼ l2 ~D ð11Þwherezisthechargeofthemobileion.F,RandThavetheirusual meaning. An example of the dispersion in the impedance representation,forZ0=1
V
andt
0=1s,isgiveninFig.2.Forhighfrequencies a semi-infinite diffusion (or Warburg: Z(
v
)=Z0
(jvt
0)0.5) is obtained, i.e. the perturbation does not reachthebackplaneatx=l.Atlowfrequenciesthedispersionbecomes veryclosetoa semi-circle,i.e. aR(RC)circuit.Here the‘Circuit DescriptionCode’isused(CDC,[42]).Itisimportanttonotethatin thisderivationitisassumedthattheelectronicconductivityofthe electrode is significantlylarger thanthe ionicconductivity and hence can be ignored. When the electronic conductivity is relativelylow then thespecial derivation by Jamnikand Maier
[43,44]mustbeused,whichleadstoamorecomplexdispersion relation.
3.FractalFiniteLengthWarburg
Thefractal-FLWisaphenomenologicalexpression,inanalogy withtheparallelcombinationofaresistanceandaconstantphase element(CPE)or‘ZARC’element[39].Eq.(11)isthenrephrasedby: ZfFLWð
v
Þ¼ j Z0vt
0ð Þntanhðj
vt
0Þn; n0:5 ð12ÞInAppendixBtheseparationintoarealandimaginarypartis presented,whichresultsin(Eq.(B4)):
ZfFLWð
v
Þ¼ Z0vt
0ð Þn
Csinh 2
a
þSsin 2b
jðSsinh 2a
Csin 2b
Þcosh2
a
þcos2b
ð13ÞWith C=cos(n
p
/2), S=sin(np
/2),a
=(vt
0)nC andb
=(vt
0)nS.Fig.2 alsoshows thedispersions fora f-FLWwithn=0.45 and n=0.4.
4.DerivationoftheDFRT
ThetransformationprocedurepresentedbyEqs.(3)and(4)in theintroduction,willbeusedforderivingtheDFRTforthef-FLW. Insertingthevariablezj1
2
p
,withz=ln(t
0/t
),intheimaginarypartofEq.(13)resultsin: Z0GfFLWðzÞ¼ Z0
p
ZfDLW;im zþjp
2 þZfDLW;im zjp
2 h i ¼Z0 GþfFLWðt
ÞþGfFLWðt
Þ h i ð14Þ FromthecompleteanalysisofEq.(14)itwasfoundthatonlythe GþfFLWðt
Þtermneedstobecalculated.ThetransformtoGfFLWðt
Þyieldsexactlythesame realpartwithan oppositesignfor the imaginarypart.ThisisnotsurprizingastheDFRTshouldbeareal function. Hence only the real part of GþfFLWð
t
Þ needs to be consideredwithGfFLWðt
Þ¼2< GþfFLWðt
Þ
Replacing
v
by(z+jp
/2)inthefirstpartofEq.(13)leadstothe followingtransformation: Z0vt
0 ð Þn¼>Z0 ezþj p 2 n ¼Z0enz cos np
2 jsin np
2 h i ¼Z0 Q½CjS ð15ÞwithQ=(
t
0/t
)n.Here‘=>’signifiesthetransformationstep.ThetransformofthedenominatorofEq.(13)yields,seeAppendixCfor thederivation,Eq.(C3):
cosh2
a
þcos2b
¼>cosh2QC2þcosh2QS2cosBþjsinh2QC2sinh2QS2sinB ð16Þ
Fig.2.ImpedancedispersionsofatrueFLW(n=0.5)andtwofractalFLW’swith n=0.45andn=0.4.Frequencyrange:1mHz100kHz.
withB=2QCS=Qsin(n
p
).Thetransformoftheimaginarypartof the numerator of Eq. (13) is derived in the same manner, seeAppendixD,Eq.(D3):
S sinh2
a
C sin2b
¼>Ssinh2QC2cosBCcosh2QS2sinBþ þjnScosh2QC2sinBCsinh2QS2cosBoð17Þ Inthefollowingsectionsthetermsnumeratoranddenominator willbeabbreviatedinequationsbyNum.andDenom.respectively. MultiplyingthisresultwiththeZ0
Q1(CjS)fromEq.(15)gives:Num:¼Z0 Q CS sinh2QC2 sinh2QS2 cosB þ S2cosh2QC2 C2cosh2QS2 sinB n o þ þj S2sinh2QC2 þC2sinh2QS2 cosB þCS cosh2QC2 þcosh2QS2 sinB n o 2 4 3 5 ð18Þ It is interesting to note that for n=0.5, i.e. C=S=p1/
2, the
denominatorchangesto:2QcoshðQÞcosðQÞ,whiletherealpartof thenumerator becomes zero!Eqs. (16) and (18) can easily be combinedinaspreadsheettocalculatethedistributionfunction foralimitedrangeof
t
-values.TheDFRT’sforn=0.45,0.48and 0.49arepresentedinFig.3.ItshowsforeachDFRTonemajorpeak, situatedatalowervaluethanthecharacteristictimeconstant,t
0,followedbysmallerpeaksfordecreasing
t
.Theappearanceofthe DFRT’s are similar to those presented by Leonide et al. [17], although no mathematical expression is presented in that publication. For too small values oft
, the arguments of the hyperbolicfunctionsbecometoolargetobeevaluatedinaspread sheet.Henceadifferentapproachisneededforsmallt
.4.1.Analysisforsmall
t
Forsmall
t
-values theargumentQbecomes large,hencethe sinh() and the cosh() functions can be approximated by the positiveexponentials,e.g.:sinh 2QC2cosh 2QC212e2QC2
ð19Þ ThusEqs.(16),(18)canbereducedbyreplacingthehyperbolic functionsbytheirrespectiveexponentialfunctions.Furthermore, boththenumeratoranddenominatorcanbedividedbytheright handsideofEq.(19).NowarelativelysimplefunctionfortheDFRT isfound(inwhichthefactor2hasbeentakencareof):
Z0Gð
t
Þ¼ Z0p
t
t
0 n sinðnp
Þ1Y22cosðnp
ÞYsinð2Qsinðnp
ÞÞ 1þ2Ycosð2Qsinðnp
ÞÞþY2 ð20Þ withY¼exp½2Qcosðnp
Þ.Writtenoutinfull,Eq.(20)becomes:Z0Gð
t
Þ¼ Z0p
t
t
0 n sinðnpÞ 1e4ðtt0Þ n cosðnpÞ 2cosðnpÞe2ðtt0Þ n cosðnpÞsin 2t0 t nsinðn pÞ 1þ2e2ðtt0Þ n cosðnpÞcos 2 t0 t nsinðn pÞ þe4ðtt0Þ n cosðnpÞ ð21Þ In Fig. 4 the DFRTof the simplifiedfunction of Eq. (21) is comparedtotheDFRTobtainedfromasimulationusingthefull expressions defined by thecombination of Eqs. (16), (18).It is surprizingthat,withintheaccuracyofExcel,theresultsareexactly identicalovertheentiret
-rangeforwhichEqs.(16),(18)canbe evaluated.HenceEq.(21)canbetakenasasimplifiedexpression thatexactlyrepresentstheDFRTofaf-FLW.Forverysmallvaluesoft
(e.g.t
<t
0103forn=0.45)Eq.(21)canbefurtherreducedtoasimplepowerrelation,presentedasthedashedlinesinFig.5: Z0Gð
t
Þ¼ Z0p
t
t
0 n sinðnp
Þ ð22ÞWhereastheDFRTshowsforn=0.45afewdecreasingpeaksfor
t
<t
0,thenumberofpeaksincreaserapidlyfornapproaching0.5.ThisisclearlydemonstratedbytheDFRTforn=0.48inFig.5.The limiting
t
-valuebelowwhichEq.(22)canbeappliedshiftsrapidly forn!0.5tot
=0,orln(t
)=1.AlthoughitseemsthatEq.(22)givesafiniteresultforn!0.5,theupperlimitin
t
forwhichthis equationcanbeusedgoestozero.4.2.Analysisforlarge
t
Forlarge
t
(i.e.t
>>t
0)theargumentsoftheexponentialandgoniometricfunctionsofQapproachzero,hencethesefunctions mustbeexpandedintheirseries.Itisessentialtoincludehigher ordertermscontainingQ3.WithwritingEq.(21)as:
Z0Gð
t
Þ¼Z0
p
Num:ANum:B
Denom: ð23Þ
Fig.3.CalculatedDFRT’sforn=0.45,n=0.48andn=0.49.Themajorpeaksfall belowthecharacteristictimeconstantt0.
Fig.4.ComparisonbetweentheexactDFRT,eqs(23,25)andthesimplifiedsolution, eq.(28)forn=0.45.
theseparatecontributionscanbeevaluated: Num:A¼2CS 1e4QcosðnpÞ
¼4Qcosðn
p
Þsinðnp
Þ8Q2cos2ðnp
Þsinðnp
Þþ323 Q 3 cos3ðn
p
Þsinðnp
Þ ð24Þ and:Num:B ¼2cosðn
p
Þe2QcosðnpÞsinð2Qsinðnp
ÞÞ¼4Qcosðn
p
Þsinðnp
Þ8Q2cos2ðnp
Þsinðnp
Þþþ8Q3
cos3ðn
p
Þsinðnp
Þ83Q3cosðnp
Þsin3ðn
p
Þð25Þ ThedenominatorofEq.(21)canbeexpandedto:
Denom:¼Qh1þ2e2QcosðnpÞcosð2Qsinðn
p
ÞÞþe4QcosðnpÞi4Q ð26Þ
CombinationofEqs.(24)–(26)leadstothelimitingexpression for
t
>>t
0: Z0Gðt
Þ¼ Z0p
8 3Q 3cosðn
p
Þsinðnp
Þcos2ðnp
Þþsin2ðnp
Þ h i 4Q ¼Z0 3p
t
0t
2n sinð2np
Þ ð27ÞForn=0.5Eq.(27)becomeszero.
5.DerivationoftheDFRTfortheidealFLW Analysisforlimitn!0.5
Insertingn=0.5directlyintoEq.(18)doesnotyieldaresultas thehyperbolicfunctionswillreducetocosh(Q)andsinh(Q),which willcanceloutinthenumeratorofEq.(18).Inordertoanalysethe DFRTforn=0.5,oneshouldapproachthisvaluebyreplacingnby: n¼0:5
d
p
ð28ÞAsaresultfor
d
!0thefollowingrelationssimplifyto: sinðnp
Þ¼sinp
2d
¼cosðd
Þ1 cosðnp
Þ¼cosp
2d
¼sinðd
Þd
ð29ÞUsingthefirsttermsofthegoniometricandexponentialseries expansions,Eq.(21)canberewrittenas:
Z0Gð
t
Þ¼Z0
p
2
d
þ2d
2d
Q1sinð2QÞ1þð12Q
d
Þcosð2QÞ2Qd
ð30ÞForn=0.5,i.e.
d
=0,thedenominatorofEq.(30)reducesto:Denom:¼1þcosð2QÞ ð31Þ
ForQ=
p
(k-0.5),withk=1,2, ...,thedenominatorbecomes zero,whichresultsinad
-functionfortheDFRT.HenceG(t
)will becomeasumofd
-functions.Thet
kpositionsaregivenby:t
k¼t
0
p
2ðk0:5Þ2; k¼1; 2; ::: ð32ÞThefirst
d
-functionoccursat0.4053t
0.Forsmallvaluesofd
(i.e.
d
<103)itispossibletocalculatethemaximumvaluesforG(
t
k).Fig.6showstheresultsforthefirst10maximainG(t
)fordecreasingvalues of
d
. Fromthis a clear relationbetween the maxima,t
andd
isobtained:ð33Þ Thesurfaceareaunderthepeakrepresentsthecorresponding dc-resistance value of the related impedance expression. The problemhereisthatthereisnoclearwaytoseparatetheareas undertheconsecutivepeaks(seeFig.5forthecasen=0.48)andto relatethesetocorrespondingresistancevalues.When
d
!0,the peakscollapsetod
-functionswithnoobservablesurfacearea.In the impedance representation, however, ad
-function is repre-sentedbyacapacitanceandresistanceinparallel,e.g.:(RC).Hence itseemsplausibletosimulatetheDFRTforaFLWwithaninfinite seriesofd
-functionsthatarerepresentedinthefrequencydomain byaninfiniteseriesof(RC)circuitswithtimeconstantsdefinedby Eq.(32).Inasimulationeffortinaspreadsheetitwasobservedthat the following relation leads to the Finite Length Warburg dispersion: ZFLWðv
Þ¼Z0 X1 k¼1 2t
k 1jvt
k 1þv
2t
2 k ð34ÞFig.7showsthesimulationofaFLWwithZ0=1
V
andt
0=1s,togetherwiththeapproximationusingEq.(34)forthefirstten
t
kvalues.Uptoafrequencyoff20
t
10 Hztheerrorislessthan1%fora10-(RC)simulation.Fora100-(RC)simulationthe1%errorlies abovef2103
t
10 Hz.6.Discussion
AremarkableresultistheidentitybetweenthecomplexDFRT, formedbythequotientofEq.(18)andEq.(16),andthesimplified Eq.(21),whichis obtainedthroughan approximationfor small
t
-values.Noefforthasbeenundertakentoprovidearigorousproof Fig.5. ComparisonbetweenthetwoDFRT’sforn=0.45andn=0.48,derivedfromeq.(21).Thesmalltapproximation,eq.(22),ispresentedbythedashedlines.
Fig.6.RelationbetweenG(tk)maxandd.Thefirsttenmaximaareshown.
for this identity, as this would be outside the scope of this contribution.
ThederivedDFRTequations providea usefuladditiontothe libraryoffunctionsusedintheevolutionaryalgorithm method. Thisreversedtransformmethod,developedbythegroupofTsur
[31–33],hasbeendescribedbrieflyintheintroduction.Butthese equationsarealsousefulforthecomparisonofregularinversions ofimpedancedatatothe
t
-domain.Theorderofmismatchshows how reliable the DFRT transform methods are. First we will considerthef-FLWwithn=0.45.Thesimulation,withZ0=1V
andt
0=1s,isperformedoverthefrequencyrange0.1mHz1MHz.FortheDFRTanalysiswithTikhonovregularizationthepublically availableMatLabapplication‘DRTtools’[30]isused.Thisprogram hasmanyadjustableparameters,but herethedefault settingis used except for theregularization parameter, RP.In a previous study[29]itwasobservedthatanincreaseinRPdiminishedthe (unwanted)oscillationsintheDFRT,butalsowidenedandlowered thepeakssignificantly.
The m(RQ)fit was performed with the CNLS-fit program ‘EqCWin’[46].Inthisprocedurefirstapartofthelowfrequency end isfitted toa R(RAQA) circuit.Next the(RAQA)dispersionis
subtractedfromtheoveralldispersion.AgainaR(RBQB)circuitis
fitted to an appropriate low frequency part of the altered dispersion. Next the combination of a R(RBQB)(RAQA) circuit is
re-fittedtoalowfrequencysectionoftheoriginaldata,resultingin anew,improvedsetofcircuitparameters.The(RQ)(RQ)-partofthe fitted R(RQ)(RQ) circuit is again subtracted from the original dispersionand anewR(RCQC)circuitis fittedtoanappropriate
sectionoftheremainderofthefrequencydispersion.Thisprocess ofiterativeaddinga(RQ)tothesubtractedmulti-(RQ)iscontinued untilanacceptablematchbetweendataand modelisobtained. Each(RQ)circuithasacounterpartinthe
t
-domainaccordingto Eq. (5), hence adding the separate RiGi(t
) functions yields a(close)approximationof theDFRT.In someinstances theCNLS procedureresultedintheshiftofa(RQ)intoa(RC)combination (i.e.: n=1). In those cases the corresponding Ri
Gi(t
) wascalculatedusingtheGaussapproximationofEq.(2).Thenumber of(RQ)and(RC)combinationsinthem(RQ)fitareindicatedasx (RQ)-y(RC),withxandythenumberofrespective(RQ)and(RC) sub-circuits.Thehighfrequencycut-offresistance,R1,whichis
partoftheCNLS-fit,hasbeenomittedfromthedescriptionofthe circuitsusedforthetransformation.
Fig.8showstheexactDFRT(Eq.(21))andtwoTR-DFRT’swith regularization parameters, RP=10-6 and RP=104. Also a DFRT
derivedfromam(RQ)fitwitha6(RQ)circuit(CNLS-
x
2=3.4109)
is presented. The m(RQ)fit follows the two major peaks quite closely inshape andposition.BothTRresults showlowerpeak height and a somewhat increased FWHM (full width at half maximum)forthemainpeak.WithRP=106thesecondpeakis poorlyreproducedwhileRP=104givesabetterresult,butwith lowerpeakheights.
A good checkof thevalidity of theDFRTis thecomparison betweenthereversedtransform,i.e.applicationofEq.(1)toobtain theimpedancefromG(
t
),andtheoriginalimpedancedata[28,29]. Thereconstructionprocedurehasbeendescribedinref.[28].Both theTR-DFRT,withRP=106,andthem(RQ)fit-DFRTwitha6(RQ) circuit show a very good match with the simulated f-FLW impedance data. This can clearly be seen from the so-called differencesgraphsofFig.9A.Heretherelativerealandimaginary Fig.7.DispersionoftheFLWandthesimulationwithaseriesof10(RC)-circuits.Intheinsertthefirst7(RC)’sareshown.Thebluelineistheenvelopeoftheseparate10-(RC) contributions.Fig.8.ComparisonofDFRT’sforaf-FLWwithn=0.45.RedlineistheexactDFRT. GreenandpurplelinesarefortheTikhonovregularizationswithresp.RP=106and
104.Bluelineisderivedfromthem(RQ)fitwitha6(RQ)circuit.Dashedlinesare scaledwithafactor5forbettervisibility.
differences,
D
real andD
imag, are plotted against log f. Thedifferencesaredefinedas[1]:
D
real;i¼Zreal;iZreconreal;i:
jZij ;
D
imag;i¼Zimag;iZrecon:imag;i
jZij ð35Þ
Zrecon:i represents the reconstructed impedance. The
recon-structedm(RQ)fitdatashowasignificantdeviationin
D
realabove1kHz.Thisisduetothe‘limited’rangeofthecalculatedDFRTinthe
t
-domain (from 107 to103s). The areaunder the DFRTcurverepresentsthetotal dc-resistance(polarisation),whichfallsjust shortof1
V
,i.e.Z0.Addingacorrectionof0.176mV
totherealpartofthereconstructeddatashowsasignificantimprovementinthe residualsplot,seeFig.9B.Butoneshouldalsorealizethatabove 2kHz(i.e.2
103/t
0)theimpedancemagnitudeislessthan1%ofZ0.
Thefitted6(RQ)impedanceshowsanexcellentfittothesimulated datawitha
x
2=3.4109,wellbelowthegeneralnoiselevel.The
highfrequencydeviation(Fig.9A)ofthereconstructedimpedance showsthatthe
t
-rangeoftheDFRTshouldbeextendedfurtherin ordertocoverthetotalpolarizationresistance(orZ0).ThecaseoftheexactFLW(n=0.5)ismorecomplicatedasthe frequencydispersioninvolvescapacitivebehaviour,seeEq.(34). TheTikhonovregularizationshowsratherbroadpeaks,seeFig.10. Thepositionsofthefirstten
d
-functions,i.e.therootsofEq.(32), arepresentedbytheverticalgreylines.Theheightsrepresentthe relativedifferencesin‘strength’,whichareproportionaltot
k.Them(RQ)fityieldeda4(RQ)-2(RC)circuit(CNLS-
x
2=1.8108).Inthe
transformtothe
t
-domainthetwo(RC)circuitswere approximat-edwithGaussfunctions,Eq.(2),withwidthparameterW=0.15. Thesetwomajorpeakscoincidequitewellwiththepositionsofthe firsttwod
-functions,ascanbeseeninFig.10.Therelativeheights alsocorrespondquitewelltotheratiot
1/t
2.TheTRandthem(RQ)fit transforms show quite some differences, however, the reconstructedimpedancesagree remarkablywellwiththeFLW simulation. Fig. 11 shows the residuals graph for the TR with RP=106andthem(RQ)fitwitha4(RQ)-2(RC)circuit.Thisonce again shows that it is difficult to define a unique DFRT from invertedimpedancedata.
Inref.[28]ithasbeenarguedthattheDFRTisonlycapableof showingmajorrelaxationpeaks.Itisnotaprocedurethatallowsto
detectminorcontributionstothefrequencydispersion.Themain poweroftheDFRTisshowingtheshiftin
t
-positionand/orchange inheightofthemajorpeakswithtemperatureorpartialpressure ofoneofthegascomponents[15].Diffusionisanimportantprocessinfuelcellelectrodesandcan bepartoftheratecontrollingprocess.Thefinitelengthdiffusionor FLWimpedancehassomesimilarityinshapewiththeGerischer impedance, see inset in Fig. 12. The latter is a semi-infinite diffusionwithacoupledsidereaction,whichresultsinafinite dc-resistance. The exact DFRT’s are, however, quite different. The Gerischer DFRT is characterized by an asymptotic function for
t
!t
0andisnon-existing(zero)fort
>t
0[29],whiletheDFRTforaFLWisaninfiniteseriesof
d
-functions,asshowninFig.12.The TikhonovregularizationsratherpoorlyreproducetheexactDFRT’s, although the reconstructed impedances closely match the simulatedimpedances.Them(RQ)fitsfor theFLWshowa good approximationofthepositionand heightratiofor thefirsttwo peaks.ButfortheGerischerapoorapproximationofitsanalytical DFRTisobtainedwithtwopeaksthatlieclosertogetherthanfor the FLW DFRT [28,29]. This couldbe used as a discrimination between a FLW and a Gerischer, but interference of other relaxationprocessescouldobscurethisobservation.For the Gerischer impedance the asymptote in the DFRT coincideswiththecharacteristictimeconstant,
t
0.FortheFLWitisfoundfromEq.(32)thatthepositionofthefirst
d
-functionatt
1isafactor2.47smallerthanthecharacteristictimeconstant,
t
0.Thetimeconstantassociatedwiththeminimumintheimaginarypart Fig.9.A)Relativedifferencegraphfortheimpedancereconstructionsfromthe
TikhonovregularizationofFig.8(RP=10-6)andthem(RQ)fitwitha6(RQ)circuit.
Opensymbolsrefertotherighthandaxis.B)Relativedifferencesfortheadjustedm (RQ)fitreconstruction,seetextforexplanation.
Fig.10.DFRT’sfortheexactFLW.RedandgreenlinesrepresenttheTikhonov regularizationswithresp.RP=106and104.Bluelineisforam(RQ)fitwitha4
(RQ)-2(RC)circuit.Vertical grey linesrepresentthepositions ofthe first ten d-functions of the exact DFRT. The height differencesrepresent the relative magnitudesofthed-functions.
Fig.11. Difference graphfor thereconstructed impedancesfor the Tikhonov regularization(RP=106)andthem(RQ)fitderivedDFRT’s,ascomparedtothe simulatedFLWdispersion.Them(RQ)fitwasbasedona4(RQ)-2(RC)circuit.
oftheFLW,
t
min,isafactor2.53smallerthant
0.Thist
minisfoundfromthederivativeofZFLW,im.withrespectto
v
.TherootfordZFLW, im/dv
=0canbefoundthroughaNewton-Raphsoniteration[47].Thedifferencebetween
t
minandt
1isduetothehigherorder(RC)-circuits(seeFig.7)thatareincludedin
t
min.Thepositionoft
minforthe Gerischer impedance can be obtained directly from the derivative:
t
0/t
min=p3[47].ItappearsthattheTikhonovregularization,TR,alwaysshowsa veryrapiddecayoutsidetheinversefrequencyrange.Thislimited
t
-rangeeasesthereconstructionoftheimpedancefromtheDFRT. Inthem(RQ)fitmethodn-valuesfortheCPEcanbeobtainedthat aresignificantlylessthan0.7.Fig.16inref.[28]clearlyshowsthat thesignificantt
-rangeoftheDFRTofsucha(RQ)rapidlyincreases withdecreasingn.Forn=0.5(Warburgdiffusion)theFWHMofthe curveextendsalreadyover2.3decades,whilethe10%peakheight valuesoftheDFRTcurvespanmorethan 5decades.Hencethe reconstructionrequires theintegration over a significant largert
-range, or the application of a resistance correction for the incompletecoverageoftheareaundertheDFRTcurve,ashasbeen demonstratedabove.7.Conclusions
For thefractal FiniteLength Warburg (bounded one-dimen-sionaldiffusion)arelativelycomplexanalyticDFRTexpressionis derivedusingthetransformprocedureintroducedbyFuossand Kirkwood[40]. Aremarkablepoint is thepositionof themajor peakat
t
1,whichlieswellbelowthecharacteristictimeconstantt
0.FromEq.(32)aratioof2.47isobtained,slightlybelowtheratioof2.53obtainedfromthefrequencyoftheminimuminZFLW,im.,
t
min=(2p
fmin)1,and the characteristic time constantt
0. TheDFRTofthef-FLWischaracterizedbyamajorpeakfollowedbya seriesofminorpeaks,thenumberandsharpnessstronglydepends onthevalueofn.Forn=0.5theidealFLWisobtained,butadirect derivationoftheDFRTisnotpossible.Thoroughanalysisof the casewherenapproaches0.5showsthattheDFRTofanidealFLW existsofaninfiniteseriesof
d
-functions.AsaresulttheidealFLW canbedescribedasaninfinitesumof(RC)’swithR=2Z0t
0/[p
2(k1/
2)2], k=1, 2, ... 1, and C0=(2Z0)1. Although the Tikhonov
regularization presents a DFRT withrather broad peaks in the
t
-domain,thereconstructedimpedanceshowsaclosematchwith theoriginaldata.Usingthem(RQ)fitapproachagood approxima-tionoftheDFRTcanbeobtained.Asstatedbeforeinref.[29],scientistsshouldbeawareofthe inherentlimitsoftheDistributionFunctionof RelaxationTimes analysis method, although with prudent use it can help in understandingtheessentialelectrochemicalprocessesincomplex systems like fuel- and electrolyser cells, complex electrodes, batteriesandsolarcells.
Acknowledgement
TheauthorwantstoexpresshisgratitudetoDr.Wanforsharing onthe Internet theuserfriendly MatLabapplication ‘DRTtools’
[30].
AppendixA.LaplacetransformofFiniteLengthDiffusionModel The Laplace transforms of Eqs. (6)–(9), with p as Laplace variable,leadto:
J¼~Dd
D
cðx;pÞ dx jx¼0 pD
cðx;pÞ¼ ~Dd 2D
cðx;pÞ dx2D
cðx;pÞjx¼l¼0 ðA1ÞThegeneralsolutionfortheLaplacetransformofFick-2isgiven by:
D
cðx;pÞ¼j
sinh x ffiffiffiffiffiffiffiffiffi p=~D q þz
cosh x ffiffiffiffiffiffiffiffiffi p=~D q ðA2Þ Fromtheboundaryconditionatx=l(Eq.(A1))itfollowsthat:z
¼j
tanh l ffiffiffiffi p ~D r ðA3Þ ThisleadstoanewexpressionforFick-1inLaplacespace: J¼~Dj
ffiffiffiffi p ~D r cosh x ffiffiffiffi p ~D r þz
ffiffiffiffi p ~D r sinh x ffiffiffiffi p ~D r jx¼0 ¼~Dj
ffiffiffiffi p ~D r ðA4Þ Fig.12.ComparisonbetweentheexactDFRT’sfortheFLW(heavygraylines)andtheGerischer(heavyblackline).TheapproximateDFRT’sforthem(RQ)fitmethod(FLW: blue,Gerischer:red)andtheTikhonovregularization(dashedlines)arealsodisplayed.InsertshowstheimpedancedispersionoftheFLWandtheGerischerwithZ0=1VandThepotentialattheelectrode/electrolyteinterfacecanthenbe expressed by (with z the charge of the mobile ion and
G
the thermodynamicfactor):D
EðpÞ¼RTzFCaeq:dlnaeq:dlnC¼
RT
G
zFCeq:
D
cðx;pÞjx¼0 ðA5ÞTheimpedanceinLaplacespacecanbedefinedbydividingthe voltagebythecurrent,I(p)=zFAJ(p),whereAisthesurfacearea:
ZðpÞ¼z
D
FAEðpÞJ ðpÞ¼ RTG
z2F2 ACeqj
sinh x ffiffiffiffip ~D rj
cosh x ffiffiffiffip ~D r tanh l ffiffiffiffi p ~D rj
ffiffiffiffiffiffiffip ~D q jx¼0 ¼ RTG
z2F2 A Ceq ffiffiffiffiffiffiffi p~D q tanh l ffiffiffiffi p ~D r ðA6Þ TheadvantageoftheLaplacespaceisthatthevariablepcanbe writtenasp=s+jv
,wheresrepresentsthetransienttoasteady statesituationandv
represents the frequencydependent part (withv
=2p
f). Hence, the FLW transfer function follows from Eq.(A6): ZFLWðv
Þ FLW ¼ RTG
z2F2AC eq ffiffiffiffiffiffiffiffiffi jv
~D q tanh l ffiffiffiffiffiffi jv
~D s ! ¼ Z0 ffiffiffiffiffiffiffiffiffiffiffi jvt
0 p tanh ffiffiffiffiffiffiffiffiffiffiffijvt
0 p ðA7Þ AppendixB.Derivationofthefullimpedanceexpression foraf-FLWTheterm(j
vt
0)ninEq.(12)canbewrittenas:ðj
vt
0Þn¼ðvt
0Þncosn
p
2 þjð
vt
0Þnsinn
p
2 ¼
a
þjb
ðB1ÞSubstitution ofEq.(B1)inEq.(12)andexpandingthetanh() functionyields: ZfFLWð
v
Þ¼ j Z0vt
0 ð Þn:tanhðjvt
0Þn ¼ Z0a
þjb
:tanhða
þjb
Þ ¼Z0ða
jb
Þvt
0 ð Þ2n : eaþjbeajb eaþjbþeajb ðB2ÞApplicationoftheEulerformula,ejx=cosx+jsinx,leadstothe
followingexpression: ZfFLWð
v
Þ¼Z0ða
jb
Þvt
0ð Þ2n :
eacos
b
eacosb
þjðeasinb
þeasinb
Þeacos
b
þeacosb
þjðeasinb
easinb
ÞðB3Þ Multiplying by the complex conjugate of the denominator resultsinaseparaterealandimaginarypart:
ZfFLWð
v
Þ¼Z0ða
jb
Þvt
0 ð Þ2n : sinh2a
þjsin2b
cosh2a
þcos2b
¼ ¼ Z0vt
0 ð Þn: cos np
2 sinh2a
þsin np
2 sin2b
cosh2a
þcos2b
2 6 4 jsin np
2 sinh2a
cos np
2 sin2b
cosh2a
þcos2b
ðB4ÞFor n=0.5 this equation reduces to the well-known FLW-function: ZFLWð
v
Þ¼ Z0 ffiffiffiffiffiffiffiffiffiffiffiffi 2vt
0 p sinh ffiffiffiffiffiffiffiffiffiffiffiffi 2vt
0 p þsin ffiffiffiffiffiffiffiffiffiffiffiffi2vt
0 p cosh ffiffiffiffiffiffiffiffiffiffiffiffi2vt
0 p þcos ffiffiffiffiffiffiffiffiffiffiffiffi2vt
0 p jsinh ffiffiffiffiffiffiffiffiffiffiffiffi 2vt
0 p sin ffiffiffiffiffiffiffiffiffiffiffiffi2vt
0 p cosh ffiffiffiffiffiffiffiffiffiffiffiffi2vt
0 p þcos ffiffiffiffiffiffiffiffiffiffiffiffi2vt
0 p " # ðB5Þ AppendixC.Transformationofthedenominator,Eq.(16)ThetransformofthedenominatorofEq.(B4)willbedonein twoparts.Firstthecosh()functionistransformed,followedbythe transformofthecosine:
cosh2
a
¼>12 e2t
0t
n CþjS ½ C þe2t
0t
n CþjS ½ C 2 4 3 5 ¼ ¼12 e2QC2 cos2QCSþjsin2QCS ð Þ h þe2QC2 ðcos2QCSjsin2QCSÞ¼¼cosh2QC2cosBþjsinh2QC2sinB ðC1Þ Transformofthecosineyields:
cos2
b
¼>12 e2jt
0t
n CþjS ð ÞS þe2jt
0t
n CþjS ð ÞS 2 4 3 5 ¼12 ej2QCS2QS2 þej2QCSþ2QS2 h i ¼ ¼12 e2QS2ðcosBþjsinBÞþe2QS2 ðcosBjsinBÞ h i ¼ ¼cosh2QS2cosBjsinh2QS2sinB ðC2Þ CombiningEqs.(C1),(C2)yieldsthefollowingexpressionfor thedenominator:Denom:¼cosh2QC2þcosh2QS2cosB
þjsinh2QC2sinh2QS2sinB ðC3Þ
AppendixD.Transformationofthenumerator,Eq.(18)
Developmentofthenumeratoroftheimaginarypart,excluding thetermbeforethebracketsinEq.(B4),isalsodoneintwoparts:
Ssinh2
a
¼>S 2 e 2t
0t
n ðCþjSÞC e2t
0t
n ðCþjSÞC 2 4 3 5 ¼S2 e2QC2þj2QCS e2QC2j2QCS h i ¼2S e2QC2ðcosBþjsinBÞe2QC2 cosBjsinB ð Þ h i ¼ ¼Shsinh2QC2cosBþjcosh2QC2sinBi ðD1Þ And: Csin2b
¼>2jChejð2QCSþj2QS2Þejð2QCSþj2QS2Þi ¼2jC ej2QCS2QS2Þ ej2QCSþ2QS2Þ h i ¼ ¼2jC e2QS2ðcosBþjsinBÞe2QS2 ðcosBjsinBÞ h i ¼ ¼Chcosh2QS2sinBþjsinh2QS2cosBi ðD2ÞCombinationofEqs.(D1),(D2)yields:
Ssin2
a
þCsin2b
¼>hSsinh2QC2cosBCcosh2QS2sinBiþ þjhScosh2QC2sinBCsinh2QS2cosBiðD3Þ References
[1]B.A.Boukamp,ElectrochemicalImpedanceSpectroscopyinSolidStateIonics; RecentAdvances,SolidStateIonics169(2004)65–73.
[2]E. Barsoukov, J.R. Macdonald (Eds.), Impedance Spectroscopy: Theory, Experiment,andApplications,2ndEdition,JohnWiley&Sons,Hoboken, NJ,2005.
[3]M.E.Orazem,B. Tribollet, ElectrochemicalImpedance Spectroscopy, John Wiley&Sons,Hoboken,NJ,2008.
[4]V.F.Lvovich,ImpedanceSpectroscopyApplicationstoElectrochemicaland DielectricPhenomena,JohnWiley&Sons,2012.
[5]A.Lasia,ElectrochemicalImpedanceSpectroscopyitsApplications,Springer, NewYork/Heidelberg,2014.
[6]Q.-A.Huang,R.Hui,B.Wang,J.Zhang,AreviewofACimpedancemodelingand validationinSOFCdiagnosis,Electrochim.Acta52(2007)8144–8164. [7]A.Nechache,M.Cassir,A.Ringuedé,Solidoxideelectrolysiscellanalysisby
meansofelectrochemicalimpedancespectroscopy:Areview,J.PowerSources 258(2014)164–181.
[8]J.R.Macdonald,L.D.PotterJr.,Aflexibleprocedureforanalyzingimpedance spectroscopyresults:Descriptionandillustrations,SolidStateIonics24(1987) 61–79.
[9]B.A. Boukamp, A non-linear least squares fit procedure for analysis of immittancedataofelectrochemicalsystems,SolidStateIonics20(1986)31– 44.
[10]B.A.Boukamp,Apackage forimpedance/admittanceanalysis,Solid State Ionics18–19(1986)136–140.
[11]K.E.D.Wapenaar,J.Schoonman,ConductivityenhancementinBal-xLaxF2+x solidelectrolytes,SolidStateIonics5(1981)637–640.
[12] J.R. Dygas,G. Fafilek, H. Durakpasa, M.W. Breiter, Automated setup for impedancemeasurementsofelectrochemicalcellswith2electrodes,J.Appl. Electrochem.23(1993)553–558.
[13]T.Hörlin,Maximumentropyinimpedance spectroscopyofnon-inductive systems,SolidStateIonics67(1993)85–96.
[14]T.Hörlin,Deconvolutionandmaximumentropyinimpedancespectroscopyof noninductivesystems,SolidStateIonics107(1998)241–253.
[15]H.Schichlein,A.C.Müller,M.Voigts,A.Krügel,E.Ivers-Tiffée,Deconvolutionof electrochemicalimpedancespectrafortheidentificationofelectrodereaction mechanismsinsolidoxidefuelcells,J.Appl.Electrochem.32(2002)875–882. [16]J.Macutkevic,J.Banys,A.Matulis,Determinationofthedistributionofthe relaxationtimesfromdielectricspectra,NonlinearAnal.Model.Control9 (2004)75–88.
[17]A.Leonide,V.Sonn,A.Weber,E.Ivers-Tiffée,‘EvaluationandModelingofthe CellResistanceinAnode-SupportedSolidOxideFuelCells’,J.Electrochem.Soc. 155(2008)B36–B41.
[18]V.Sonn,A.Leonide,E.Ivers-Tiffée,‘CombinedDeconvolutionandCNLSFitting ApproachAppliedontheImpedanceResponseofTechnicalNi/8YSZCermet Electrodes’,J.Electrochem.Soc.155(2008)B675–B679.
[19]J.P.Schmidt,T.Chrobak,M.Ender,J.Illig,D.Klotz,E.Ivers-Tiffée,‘Studieson LiFePO4ascathodematerialusingimpedancespectroscopy’,J.PowerSources 196(2011)5342–5348.
[20]H.Sumi,T.Yamaguchi,K.Hamamoto,T.Suzuki,Y.Fujishiro,T.Matsui,K. Eguchi,‘ACimpedancecharacteristicsforanode-supportedmicrotubularsolid oxidefuelcells’,ElectrochimicaActa67(2012)159–165.
[21] J.Illig,J.P.Schmidt,M.Weiss,A.Weber,E.Ivers-Tiffée,‘Understandingthe impedancespectrumof18650LiFePO4-cells’,J.PowerSources239(2013)670– 679.
[22]T.Ramos,M.Søgaard,M.B.Mogensen,‘ElectrochemicalCharacterizationofNi/ ScYSZElectrodesasSOFCAnodes’,J.Electrochem.Soc.161(2014)F434–F444. [23]S.Kazlauskas,A.Kežionis,T.Šalkus,A.F.Orliukas,ElectricalpropertiesofYSZ
andCaSZsinglecrystals,SolidStateIonics231(2013)37–42.
[24]M. Saccoccio, T.H. Wan, C. Chen, F. Ciucci, Optimal Regularization in DistributionofRelaxationTimesappliedtoElectrochemicalImpedance Spectroscopy:RidgeandLassoRegressionMethodsATheoreticaland ExperimentalStudy,Electrochim.Acta147(2014)470–482.
[25]T.H.Wan,M.Saccoccio,C.Chen,F.Ciucci,InfluenceoftheDiscretization MethodsontheDistributionofRelaxationTimesDeconvolution:
ImplementingRadialBasisFunctionswithDRTtools,Electrochim.Acta184 (2015)483–499.
[26]Y. Zhang, Y. Chen, M. Yan, F. Chen, Reconstruction of relaxation time distributionfromlinearelectrochemicalimpedancespectroscopy,J.Power Sources283(2015)464–477.
[27]Y.Zhang,Y.Chen,M.Li,M.Yan,M.Ni,Ch.Xia,Ahigh-precisionapproachto reconstructdistributionofrelaxationtimesfromelectrochemicalimpedance spectroscopy,J.PowerSources308(2016)1–6.
[28]B.A.Boukamp,Fouriertransformdistributionfunctionofrelaxationtimes; applicationandlimitations,Electrochim.Acta154(2015)35–46.
[29]B.A.Boukamp,A.Rolle,Analysisandapplicationofdistributionofrelaxation timesinsolidstateionics,SolidStateIonics302(2017)12–18.
[30]DRTtoolsandmanual:https://sites.google.com/site/drttools/.
[31]S.Hershkovitz,S.Baltianski,Y.Tsur,Harnessingevolutionaryprogrammingfor impedancespectroscopyanalysis:Acasestudyofmixedionic-electronic conductors,SolidStateIonics188(2011)104–109.
[32]Z.Drach,S.Hershkovitz,D.Ferrero,P.Leone,A.Lanzini,M.Santarelli,Y.Tsur, Impedancespectroscopyanalysisinspiredbyevolutionaryprogrammingasa diagnostictoolforSOECandSOFC,SolidStateIonics288(2016)307–310. [33]A.K.Baral,Y.Tsur,ImpedancespectroscopyofGddopedceriaanalyzedby
geneticprogramming(ISGP)method,SolidStateIonics304(2017)145–149. [34]A.Bello,E.Laredo,M.Grimau,Distributionofrelaxationtimesfromdielectric spectroscopyusingMonteCarlosimulatedannealing:Applicationtoa-PVDF, Phys.Rev.B60(1999)12764–12774.
[35]A.Bello,E.Laredo,M.Grimau,Relaxationtimedistributionfromtimeand frequencydomaindielectricspectroscopyinpoly(aryletheretherketone),J. Chem.Phys.113(2000)863–868.
[36]B.A.Boukamp,H.J.Bouwmeester,InterpretationoftheGerischerimpedancein solidstateionics,SolidStateIonics157(2003)29–33.
[37]Y.Lu,C.Kreller,S.B.Adler,Measurementandmodelingoftheimpedance characteristicsofporousLa1-xSrxCoO3-delectrodes,J.Electrochem.Soc.156 (2009)B513–B525.
[38]D.R.Franceschetti,J.R.Macdonald,R.P.Buck,Interpretationof Finite-Length-Warburg-typeimpedancesinsupportedandunsupportedelectrochemical cellswithkineticallyreversibleelectrodes,J.Electrochem.Soc.138(1991) 1368–1371.
[39]Z. Stoynov, Impedance modelling and data processing: structural and parametricalestimation,Electrochim.Acta35(1990)1493–1499.
[40]R.M.Fuoss,J.G.Kirkwood,ElectricalPropertiesofSolids.VIII.DipoleMoments inPolyvinylChloride-DiphenylSystems,J.Am.Chem.Soc.63(1941)385–394. [41]F.Dion,A.Lasia,Theuseofregularizationmethodsinthedeconvolutionof underlyingdistributionsinelectrochemicalprocesses,J.Electroanal.Chem. 475(1999)28–37.
[42]CDC_Explained.pdf, available from: www.utwente.nl/tnw/ims/publication/ downloads/.
[43]J. Jamnik, J. Maier, Treatment of the Impedance of Mixed Conductors EquivalentCircuitModelandExplicitApproximateSolutions,J.
Electrochem.Soc.146(1999)4183–4188.
[44]J. Jamnik, J. Maier, Generalised equivalent circuitsfor mass and charge transport:chemicalcapacitanceanditsimplications,Phys.Chem.Chem.Phys. 3(2001)1668–1678.
[46]A.Wisse,B.A.Boukamp,EqCWin95:www.wisseq.nl/portfolio.aspx(inDutch). [47]B.A. Boukamp, private communication: www.utwente.nl/tnw/ims/