Computational geo-electromagnetics: methods, models, and forecasts: volume 5 1st edition viacheslav

Page 1


https://ebookmass.com/product/computational-geoelectromagnetics-methods-models-and-forecasts-volume-5-1stedition-viacheslav-v-spichak/

Instant digital products (PDF, ePub, MOBI) ready for you

Download now and discover formats that fit your needs...

Deterministic and Stochastic Modeling in Computational Electromagnetics Dragan Poljak

https://ebookmass.com/product/deterministic-and-stochastic-modelingin-computational-electromagnetics-dragan-poljak/

ebookmass.com

Computational Frameworks. Systems, Models and Applications

1st Edition Edition Mamadou Kaba Traore (Eds.)

https://ebookmass.com/product/computational-frameworks-systems-modelsand-applications-1st-edition-edition-mamadou-kaba-traore-eds/

ebookmass.com

Computational Models of Reading: A Handbook 1st Edition Erik D. Reichle

https://ebookmass.com/product/computational-models-of-reading-ahandbook-1st-edition-erik-d-reichle/

ebookmass.com

Proceedings of the Third Annual National MultiDisciplinary Conference, V-CMT 2016 Dr. Rohini Kelkar

https://ebookmass.com/product/proceedings-of-the-third-annualnational-multi-disciplinary-conference-v-cmt-2016-dr-rohini-kelkar/

ebookmass.com

Electrotechnology Practice Jeffery Hampson

https://ebookmass.com/product/electrotechnology-practice-jefferyhampson/

ebookmass.com

The Secret Listener Yuan-Tsung Chen

https://ebookmass.com/product/the-secret-listener-yuan-tsung-chen/

ebookmass.com

Vampire the Requiem, Second Edition - Secrets of the Covenants Rose Bailey

https://ebookmass.com/product/vampire-the-requiem-second-editionsecrets-of-the-covenants-rose-bailey/

ebookmass.com

The Beauty of Darkness (Crônicas de Amor e Ódio 3) Pearson

Mary E.

https://ebookmass.com/product/the-beauty-of-darkness-cronicas-de-amore-odio-3-pearson-mary-e/

ebookmass.com

Roman Receptions of Sappho Thea S Thorsen

https://ebookmass.com/product/roman-receptions-of-sappho-thea-sthorsen/

ebookmass.com

Cowboy Wolf Outlaw Kait Ballenger

https://ebookmass.com/product/cowboy-wolf-outlaw-kait-ballenger-5/ ebookmass.com

COMPUTATIONALGEOELECTROMAGNETICS Methods,Models,and Forecasts

VIACHESLAVV.SPICHAK

Elsevier

Radarweg29,POBox211,1000AEAmsterdam,Netherlands

TheBoulevard,LangfordLane,Kidlington,OxfordOX51GB,UnitedKingdom

50HampshireStreet,5thFloor,Cambridge,MA02139,UnitedStates

Copyright 2020ElsevierInc.Allrightsreserved.

Nopartofthispublicationmaybereproducedortransmittedinanyformorbyanymeans,electronicormechanical, includingphotocopying,recording,oranyinformationstorageandretrievalsystem,withoutpermissioninwriting fromthePublisher.Detailsonhowtoseekpermission,furtherinformationaboutthePublisher’spermissions policiesandourarrangementswithorganizationssuchastheCopyrightClearanceCenterandtheCopyright LicensingAgency,canbefoundatourwebsite: www.elsevier.com/permissions .

ThisbookandtheindividualcontributionscontainedinitareprotectedundercopyrightbythePublisher(otherthan asmaybenotedherein).

Notices

Knowledgeandbestpracticeinthis fieldareconstantlychanging.Asnewresearchandexperiencebroadenour understanding,changesinresearchmethods,professionalpractices,ormedicaltreatmentmaybecomenecessary.

Practitionersandresearchersmustalwaysrelyontheirownexperienceandknowledgeinevaluatingandusingany information,methods,compounds,orexperimentsdescribedherein.Inusingsuchinformationormethodsthey shouldbemindfuloftheirownsafetyandthesafetyofothers,includingpartiesforwhomtheyhaveaprofessional responsibility.

Tothefullestextentofthelaw,neitherthePublisher,northeauthors,contributors,oreditors,assumeanyliability foranyinjuryand/ordamagetopersonsorpropertyasamatterofproductsliability,negligenceorotherwise,or fromanyuseoroperationofanymethods,products,instructions,orideascontainedinthematerialherein.

LibraryofCongressCataloging-in-PublicationData

AcatalogrecordforthisbookisavailablefromtheLibraryofCongress

BritishLibraryCataloguing-in-PublicationData

AcataloguerecordforthisbookisavailablefromtheBritishLibrary

ISBN:978-0-12-819631-1

ISSN:2468-547X

ForinformationonallElsevierpublicationsvisitourwebsiteat https://www.elsevier.com/books-and-journals

Publisher:CandiceJanco

AcquisitionsEditor:AmyShapiro

EditorialProjectManager:AndreaDulberger

ProductionProjectManager:OmerMukthar

CoverDesigner:ChristianJ.Bilbow

TypesetbyTNQTechnologies

Preface

Themainstreamapproachinthepresent-dayinterpretation ofelectromagnetic(EM)dataistosolvetheoreticallynonunique inverseproblemusingsomeregularizationtechnique.Many examplescouldbefoundinthemonographspublishedinthe past20years(e.g., Zhdanov,2002,2009;SimpsonandBahr,2005; BerdichevskyandDmitriev,2009;ChaveandJones,2012;Spichak,2015).Thesituationisespeciallycomplicatedinthe3-D and4-Dcases,whenthenumberoftheunknownsincreases dramatically(e.g., Spichak,1999;SpiesandOristaglio,1999; ZhdanovandWannamaker,2002; Spichak,2019).Thehugevolumeandhighcomplexityofthedatatobeprocessednecessitatesthedevelopmentandapplicationofefficientquantitative methodsandcomputationaltools,inparticular,parallelization ofappropriatecodes,exploitinghighperformancecomputer resources(whicharenotalwaysathand).

Meanwhile,ourknowledgeoftheEarth’sinterioraswellas theefficiencyofsolutionsofnumerousrelatedproblemsstrongly dependsonaccuracyofEMdatainterpretation.Theapproaches usedtothisendshouldtakeintoaccountthepurposeofthe study,volume,andqualityoftheavailabledata,priorgeological andgeophysicalinformation,scaling,signaltonoiseratio,etc.If onedoesnothavearegulararrayofhighqualityandevenly distributedEMsoundingdata,reliableanduserfriendlysoftware fortheir3-Dinversionandmulti-processorworkstationfor parallelcalculations,readingthisvolumemayhelptoextract maximuminformationfromtheavailabledata.

Incontrarytopreviouslypublishedmonographsongeoelectromagnetics,thisvolumeoffersthetechniquesandappropriatealgorithmsforbuildinggeoelectricalmodelsunderthe conditionsofsparseorirregularlydistributedEMdata,andlack ofpriorgeologicalorgeophysicalinformation.Secondly,itprovidesmethodologicalguidelinesforinterpretationofelectromagneticsoundingdatadependingongoalsofthestudy.Finally, onemayfindcomputationalalgorithmsforusingelectrical resistivityasaproxyparameterforforecastingpetrophysical propertiesofrocksbeyondboreholes.

Accordingly,thebookhasthreeparts.InPartI(Methodology ofEMdatainterpretation)twoalternativeapproachestoEMdata inversionareconsidered:theBayesianstatisticalinversionand neuralnetwork basedalgorithms.Theformeroneprovides flexibletoolsfortakingintoaccountpriorinformationandexpert

estimatesduringEMdatainversionandquantifyingitsresultsin termsofposteriorparameters’uncertainties.Thelatteroneis especiallyusefulinthecaseofconstructing3-Dmodels(in particular,intermsofmacroparameters)fromsparse,irregularly distributedorasingle-profileelectromagneticdata.Finally,a reviewofthemethodsusedforjointanalysisandinversionofEM andothergeophysicaldataispresented.

InPartII(Modelsofgeologicalmedium)theapproaches consideredinthePartIareappliedtostudymethodologicalissues ofEMmodelingvolcanoes(byexamplesofVesuvio,Kilauea, Elbrus,Komagatake,Hengill),geothermalandhydrocarbonreservoirs.ConceptualmodelsoftheIcelandictypecrust,alensin theuppercrustandcopper-porphyryoreformationaresuggested basedonjointanalysisofEMandothergeophysicaldata.

InPartIII(Forecastingpetrophysicalpropertiesofrocks)the techniquesforestimatingtemperature,seismicvelocities,and porosityfromtheelectricalresistivityasproxyparameterare considered.ThispartissupplementedbyAppendix,which includesusefulempiricalrelationsbetweenelectricalresistivity, seismicvelocitiesandporosity.

Thestudiesaddressedinthisvolumeweresupportedby grantsfromRussianFoundationforBasicResearch,Russian AcademyofSciences,OYOCorporation(Japan),BRGM(France), EinsteinConsortium(Italy),INTASandVIFrameworkProgram ofEuropeanCommunity.Iamverythankfulforthefunding supportprovidedbytheseorganizations.Iamdeeplygratefulto Prof.M.Zhdanov(UtahUniversity)whoinitiatedmyinterestin geo-electromagneticsin1979andwasmyPhDthesismentor,I amindeptedtoProf.T.Mogi(SapporoUniversity)forproviding friendlysupportofmystudiesincomputationalgeo-electromagneticsduringalmost30years.,Itismygreatpleasureto acknowledgeDr.H.Shima(OYOCorporation,Tokyo),Dr.M. Mareschale(l’E ´ colePolytechniquedeMontre ´ al),Dr.R.Lewis (HobartUniversity),Prof.D.Patella(UniversityFedericoIIdi Napoli),Prof.A.Siniscalchi(BariUniversity),Prof.M.Menvielle (UniversityParisSud),Prof.T.Harinarayana(InstituteofScience andTechnology,Hyderabad),Dr.A.Manzella(InstituteofGeoscienceandEarthResources,Pisa),Dr.H.Eysteinsson(ISOR, Reykjavik),Dr.G.Gugunava(M.NodiainstituteofGeophysics, Tbilisi)Dr.P.Calcagno(BRGM,Orleans),Dr.A.Genter(ESGeothermie,Electricite ´ deStrasbourg)forfruitfulscientificcollaborationintheframeworkofjointprojects.Whiledeveloping computercodesforEMdataanalysisandinterpretationI

receivedmuchassistancefromformeremployeesofmyLaboratoryDrsE.FomenkoandI.Popova.

IacknowledgeMarinaNazarenkoforpartialEnglishtranslationofthisvolumeandAlexandraGoidinaforassistancewith computergraphics.

ViacheslavV.Spichak Moscow,July,2019

References

Berdichevsky,M.N.,Dmitriev,V.I.,2008.ModelsandMethodsof Magnetotellurics.Springer-Verlag,Berlin.

Chave,A.,Jones,A.(Eds.),2012.Themagnetotelluricmethod:theoryand practice.CambridgeUniversityPress,NewYork.

Simpson,F.,Bahr,K.,2005.Practicalmagnetotellurics.CambridgeUniversity Press,NewYork.

Spichak,V.V.,1999.Magnetotelluricfieldsinthree-dimensionalgeoelectrical models.ScientificWorld,Moscow.

Spichak,V.V.(Ed.),2015.ElectromagneticSoundingoftheEarth’sInterior: Theory,Modeling,Practice.ElsevierB.V.,Amsterdam.

Spichak,V.V.,2019.ElectromagnetictomographyoftheEarth’sinterior.Scientific World,Moscow.

Spies,B.,Oristaglio,M.(Eds.),3DElectromagnetics.SEGPubl.,GD7,Tulsa,USA.

Zhdanov,M.S.,2002.Geophysicalinversetheoryandregularizationproblems. Elsevier,Amsterdam.

Zhdanov,M.S.,2009.Geophysicalelectromagnetictheoryandmethods.Elsevier, Amsterdam.

Zhdanov,M.S.,Wannamaker,P.(Eds.),2002.Three-Dimensional Electromagnetics.Elsevier,Amsterdam.

3-DEMforwardmodeling techniques

1.1Introduction

Tosolvetheforward3-Delectromagnetic(EM)problem(inafrequencydomain)impliestodeterminetheelectromagnetic field(E; H)satisfyingtheMaxwellequations

atagivenfrequency u,fromthedistributionofgeneralized complexelectricalconductivity e sðr; uÞ¼ s i uε (m isamagnetic permeability, ε isapermittivity, jext ðr; uÞ istheexternalcurrent density)specifiedinacertainspatialregionthatincludesan inhomogeneity.Inthischapterweassumeforsimplicitythat j ext ¼ 0; m ¼ m0 ¼ 4p,10 7 H/m,where m0 isthefree-spacemagnetic permeability,andneglectthedisplacementcurrents,i.e., e sh s.In otherwords,wewillconsidermodelingoftheelectromagnetic fieldsinducedin3-Dearthbynaturalsource planewave(socalled “magnetotelluric” (MT) fields)thoughtheissuesdiscussed belowinmostcasesreferequallytocontrolledsourceexcitation.

Variousapproachestothesolutionofthisproblemexist,allof themfallingintotwolargegroupscomprisingintegralequations anddifferentialequationstechniques.Currently,numerical computer-aidedmodelingofelectromagnetic fieldsbecomesa powerfulandrelativelyeasy-of-accesstoolforanalyzing complicatedsituations.Whereasformerlythescopeofgeophysicalconsiderationwascon finedtosimplifiedmodelmedia(1-D 2-D),theadventindesigningpowerfulcomputerfacilities madeitpossibletoposetheproblemofnumericalmodelingof electromagnetic fieldsexcitedby3-Dsourcesinatwo-or three-dimensionalmediumorby2-Dsourcesinathreedimensionalmedium(seereviewpapersby Avdeev(2005, 2015) ; Boerner(2010); Newman(2014) andreferencestherein).

Inthischapter,wewillconsiderbrieflythenumerical(Sections 1.2 1.4)andanalog(Section 1.5)approachestotheforward problemsolution.Numericalmethods,inturn,aresubdivided

intotwolargegroups:integralequationmethods(Section 1.2)and differentialequationmethods(Section 1.3).Acomparativeanalysisofthesetwogroupsofmethodsandhybridapproaches(Section 1.4)basedontheircombinationispresent.

Section 1.6 describesmaincharacteristicfeaturesofabalance techniqueforEM field’scalculationinmediawitharbitrary3-D distributionoftheelectricalconductivity.Atthesametime,ifthe electricalconductivitymodelshowsacertainsymmetrytype,the above-mentionedpurposecanbeachievedusingsmallercomputerresources.Inparticular,for3-Dmodelswithaverticalaxial symmetry,thevectorproblemreducestoascalarone.Section 1.7 addressesanalgorithmfornumericalcalculationofquasistationaryelectromagnetic fieldsin3-Daxiallysymmetricmedia basedona finiteelementmodificationofthebalancemethod.

1.2Methodsofintegralequations

Theintegralequationmethodforthenumericalcomputationof electromagnetic fieldswaspioneeredby Dmitriev(1969).The basicideasofthisapproachweredevelopedby Raiche(1974), Hohmann(1975), Tabarovsky(1975) ,and Weidelt(1975).Subsequently,theintegralequationmethodwassuccessfullyelaborated by DmitrievandFarzan(1980), TingandHohmann(1981), Hohmann(1983) , Wannamakeretal.(1984a,b), Hvozdara(1985), Hvozdaraetal.(1987), Wannamaker(1991), Xiong(1992) , Pankratovetal.(1995), Singer(1995), SingerandFainberg(1995, 1997), XiongandTripp(1995) , Avdeevetal.(1997) , Farquharsonand Oldenburg(1999) , Leeetal.(1999), Portniaguineetal.(1999), Xiongetal.(1999a ,b), HursanandZhdanov(2002), Zhdanov etal.(2000, 2006), Singer(2008), Endoetal.(2009), Avdeev (2015), Kruglyakovetal.(2016).

Paralleltothemethodofvolumeintegralequations(VIE)that involvesintegrationovertheentirevolumeoftheanomalous domain,amethodofsurfaceintegralequations(SIE)implying integrationonlyovertheboundaryofthedomainhasbeendeveloped(OshiroandMitzner,1967; DmitriedandZakharov,1970; Tabarovsky,1971; Smagin,1980; LiuandLamontagne,1999).

Letusconsiderthesetwoapproaches.

1.2.1Themethodofvolumeintegralequations

Volumeintegralequation(VIE)techniqueisbasedonthenumericalsolutionofthesecond-typeFredholmequation,whichcanbe derivedfromMaxwellequationsusingGreenfunctions:

where Eb isabackground fieldcalculatedwiththeassumption that s ¼ sb (backgroundelectricalconductivity), c G e ðr =r0 Þ isthe electrical-typeGreentensor, r and r0 arecoordinatesoftheobservationandsourcepoint,respectively,and V istheregioninquestion.(Hereinafter,forthesakeofde finitenessonlytheequation fortheelectrical field E willbeconsidered.Oncethenumerical solutiontothisequationisobtained,themagnetic field H can bereadilycalculatedfrom,e.g.,thesecondMaxwellequation (seeinthisconnectionSection 1.6.4)).

Thepivotalideaofthisapproachisasfollows.Theanomalous regionisdividedbyaspatialgridintocells.Withineachcell,the fieldisassumedtobeconstant.Therefore,inthesecondtermof (1.2) ,itcanbetakenoutsidetheintegralsign.Toobtainasystem oflinearalgebraicequationsforthe field,onehastoonlycalculate therespectivetensorcoef ficientswithineachunitcell.

Letusdiscussthemainadvantagesanddrawbacksofthis method.Theadvantagescompriseitsgreaterphysicaltransparencycompared,e.g.,tothemethodsofdifferentialequations. Inaddition,asisapparentfrom (1.2),thesecondtermonthe right-handsideisnonzeroonlyif sssb ;i.e.,integrationmustbe carriedoutonlyovertheanomalousregionalone.Finally,thenumericalapproximationof (1.2) doesnotinvolvetheunstableprocedureofnumericaldifferentiation,whichistypicalofmostthe differentialequationmethods.

Ontheotherhand,signi ficantdifficultiesariseinthenumericalapproximationofthesecondtermintheright-handsideof Eq. (1.2).Inparticular,determinationoftheGreentensor G e ^ ðr =r0 Þ componentsisanontrivialcomputationalproblem, solutionofwhichinvolvestheHankeltransforms(Hohmann, 1975; Weidelt,1975; FarquharsonandOldenburg,1999;etc.)or thelinear filtrationtechnique(DasandVerma,1981; Vermaand Das,1982).Computationofthetensorcoefficientscanbeas timeconsumingasitselfsolvingthesystemoflinearalgebraic equations.Therefore,itisa “bottleneck” ofthetechnique,and thesuccessofthewholeapproachthusdependsonhoweffectivelythisproblemissolved.

Anotherdifficultyariseswhilesolvingthesystemoflinear algebraicequations.Densematrixofthesystemmakeslimited computerresourcesacriticalfactor.Inturn,thelimitationon thedimensionalityofalgebraicsystemultimatelyentailslimitationstothemathematicalmodel:wecanonlycomputethe fields

forbodiesthatarenottoolarge. Xiong(1992) overcamethisdrawbackbypartitioningthescatteringmatrixintomanyblocksubmatricesandsolvingthewholesystembyablockiterative method.Thisreducesthecomputermemoryrequirementsand timeofcomputation.(Itisworthmentioninginthisconnection thatmuchmoredrasticreductionofthecomputerresources couldbeachievedbyusingtheso-called “sparsematrixtechnique” (PoggioandMiller,1973)thatimpliesignoringtheinteractionamongsufficientlyremoteareasofthemodelingdomain;the lattercorrespondstozeroingthoseelementsofthematrixthatare smallcomparedtothediagonalones.)Anothersimplification (whichalsoreducestheclassofobjectsbeingmodeled)istosolve theproblemwithinthelong-waveapproximation.Thisenables one, firstly,tosimplifycomputationoftheGreentensor(Hvozdara,1981)and,secondly,touseinthenumericalsolutionof (1.2) theBornapproximation(Torres-Verdin,1985; Torres-Verdin andBostick,1992),localizednonlinearapproximations(Habashy etal.,1993; Torres-VerdinandHabashy,1994),quasilinear approximationandseries(ZhdanovandFang,1996, 1997)and quasianalyticalseries(Zhdanovetal.,2000). Portniaguineetal. (1999) suggestedawayofreducingthetimeofcomputationsby theuseof “compressionmatrix” thatconvertstheoriginaldense matrixintothesparseonebyconstructinganinterpolationpyramidinmultipledimensions.Finally, Endoetal.(2009) suggesteda multidomainapproach,while Zhdanovetal.(2006) proposedan algorithmbasedonusinginhomogeneousbackgroundconductivity,whichincreasecompetitivenessofthisapproachincomparisonwithdifferentialequationtechniques(Zhdanovetal., 2013).

InevaluatingtheadvantagesanddrawbacksoftheVIE methodonawhole,itshouldbenotedthatthismethodseems givingthemostaccurateresults inacomparativelyshorttime foranomaliesthataresmallcomparedtothewavelengthwithin theanomalousregionandhaveasimpleshape.

1.2.2Themethodofsurfaceintegralequations

Surfaceintegralequation(SIE)techniquemakesuseofelectromagnetic fieldrepresentationasanintegraloverthesurfaceof adomainbymeansofStratton Chuformulas(Dmitrievand Zakharov,1970),methodofauxiliary(fictitious)sources(Tabarovsky,1971),methodofpotentials(Smagin,1980),orthemethod employingtheHelmholtzscalarequations(LiuandLamontagne, 1999).Afterpassingtothelimitwiththeobservationpointtending totheboundaryofthedomainfrominsideandfromoutsidein

turn,onecan,usingthecontinuityofthetangentialcomponents ofelectromagnetic fieldontheboundary,obtainthenecessary equationsonlyovertheboundaryofdomain V withrespecttounknown fielddensities.NumericalsolutionoftheseequationsinvolvestheKrylov Bogolyubovmethod,variationalapproaches, Bubnov Galerkin typemethods,etc. SmaginandTsetsokho (1982) haveobtainedthesystemoflinearalgebraicequationsusingthecollocationmethodandthedensitiessoughtforwere approximatedbysmooth finitefunctions,whichforma finite partitionofunityoverthesurfaceofthedomain.

Thistechniquehasanobviousadvantageovertheintegral equationmethod.Numericalsolutiontotheequationswritten overthesurfaceofthedomainratherthanoveritsvolumeinvolvesaconsiderablereductioninthedimensionalityofthesystemoflinearalgebraicequations.Thisisespeciallyimportant whenmodelinglargeinhomogeneitiesbecausetheratioofthedimensionsofrespectivematricesdecreaseswithanincreaseinthe characteristiclineardimension L as 1/L

Despitethisadvantage,however,theuseofSIEtechniqueis limitedtomodelinghomogeneousbodieswithrotationalsymmetry,embeddedinahomogeneousspace(halfspace)(Oshiro andMitzner,1967; LiuandLamontagne,1999).Thisisdueto theimperfectionofthemathematicalapparatusformodeling inhomogeneitieswithanarbitrarydistributionofanomalouselectricalconductivity sa ¼ s(x,y,z) sb(z)embeddedinahorizontally layeredsectionandtothedifficultiesarisingatthestepofnumericalapproximationofsingularintegralequationsandofsolving theresultingsystemoflinearalgebraicequationswithanillconditionedmatrix.

Therefore,integralequationmethodsaremosthelpfulwhen modelinginhomogeneitiesofcomparativelysimpleshapeand smallsize(comparedtothewavelengthwithinthem)embedded inahorizontallylayeredmediumwithasmallnumberoflayers.

1.3Methodsofdifferentialequations

Advancesingeoelectricscallforthcomputing3-Delectromagnetic fieldsincomplicatedsituationswhentheinhomogeneityexhibits anarbitraryshapeanddimensions(inparticular,itmaynotbe local)andtheanomalouselectricalconductivityinthemodel variesarbitrarily,as,e.g.,inregionalmodels.Inthiscase,as pointedoutabove,integralequationmethodsareoflimiteduse.

Theonlyviableapproachtosuchproblemsinvolveseithera directsolutionofdifferentialequationswithpartialderivatives,

ortheuseofso-calleddifferentialequationmethods,ofwhich twomainonesbeingthe finitedifference(FD)techniqueand the finiteelement(FE)technique.Letuslistthemainfeatures oftheirapplicationtotheproblemofnumericalmodelingof 3-Delectromagnetic fields.

1.3.1Thefinitedifferencetechnique

TheFDtechniquebasedonthe finitedifferenceapproximationof derivativesisusedforthenumericalsolutionofadifferential equationwithrespecttotheelectricalormagnetic fieldina certainspatialregionthatcontainsaninhomogeneity:

whichfollowdirectlyfrom (1.1).

Theadvantagesofthistechniqueareitsgreaterversatility comparedtointegralequationmethods,thesimplicityofitsnumericalimplementation,bandstructureofthematrixofresulting systemoflinearalgebraicequations.Thelattercircumstanceis vitalbecauseitsignificantlyreducesthetimeandamountof operationalcomputermemoryrequiredforsolvingtheproblem and,consequently,enablesmodelingthetargetsoflargesizeor withcomplicatedconductivitydistributioninit.

Themainfactorsinfluencingtheeffectivenessofthismethod areasfollows: -choiceoftheappropriategoverningequation; -selectionofthediscretizationscheme; -approximationofthesecondderivativesin (1.3) or (1.4); -specifyingboundaryconditionsthatarevalidatnonindefinite distancefromtheanomaly; -conservationofcurrentsandmagnetic field flux; -accuracywhensolvingtheproblemsatfrequenciesapproachingthestaticlimit(similarproblemismetbytheFE technique);

-accuracywhensolvingtheproblemswithsharpconductivity contrasts.

Sincethe first3-Dmodelingresultswereobtainedwiththis methodbyF.Jones(JonesandPascoe,1972; LinesandJones, 1973; Jones,1974; JonesandLokken,1975; HibbsandJones,

1978),anumberofimprovementsweremadethatsignificantly increasedtheeffectivenessofFDtechnique(ZhdanovandSpichak,1980; Zhdanovetal.,1982; Lametal.,1982; Spichak, 1983a ,b; ZhdanovandSpichak,1989, 1992; MackieandMadden, 1993; Mackieetal.,1993, 1994; DruskinandKnizhnerman,1994; AlumbaughandNewman,1996; Smith,1996a,b; LaBreque,1999; Spichak,1999; Weidelt,1999; Weaveretal.,1999; Xiongetal., 1999b ; Sasaki,2001; WangandFang,2001; FomenkoandMogi, 2002; NewmanandAlumbaugh,2002; WeissandNewman,2002; Streich,2009; Mittet,2010; YavichandZhdanov,2016; Wang andTan,2017; Caoetal.,2018).

ZhdanovandSpichak(1980) suggestedabalancetechniquefor approximationoftheFDequationinelectrical field(seeSection 1.6 later),whichenabledtheaccuracyoftheresultstoincrease duetoreductionfromtheunstablecalculationofthesecondorderderivativestothe first-orderones. Spichak(1983a) proposed touseacurrentdivergence-freeconditioninthebalancetechnique,whichraisedtheaccuracyoftheresultsduetodisappearanceofthesecondterminEq. (1.3) inregionswithzero conductivitygradientandalsoatthestaticlimit. Zhdanovetal. (1982) and Spichak(1985, 1999, 2006) introducedtheasymptotic boundaryconditionsthatgreatlydiminishedthedimensionsof themodelingdomainandreallyincreasedtheaccuracyoftheforwardmodeling. Smith(1996a) proposedtouseastaggeredgrid method(pioneeredby Yee(1966))thatguaranteesautomaticconservationofcurrentsandmagnetic fluxonthegrid(though,atthe costofsomeinconveniencewhencomputingsome fieldcomponentsattheearthsurfaceorthe fieldtransformationsinvolving thecomponentsspecifiedondifferentgrids). Davydychevaand Druskin(1999) , Weidelt(1999),and WangandTan(2017) extendedthisapproachtotheanisotropicmedia.

DruskinandKnizhnerman(1994), Druskinetal.(1999) , Jin etal.(1999) , Zhouetal.(2013) useaspectralLanczosdecompositionmethod(SLDM)withKrylovsubspacesgeneratedfromthe inverseoftheMaxwelloperator.SLDMenablesaccelerationof theforwardmodelingowingtopossibilityofgettingthesolution forthewholefrequencyrangepracticallyatthecostofsolution forasinglefrequency.

Smith(1996b) developedastaticcorrectionprocedurethat explicitlyenforcestheelectricalcurrentandmagnetic field divergence-freeconditionsthat,inturn,increasestheaccuracyof theEM fieldcalculationwhenthefrequencytendstozero.Inorder toacceleratethesolutionoftheforwardproblematlowinduction numbers(LINs), NewmanandAlumbaugh(2002) proposedanLIN preconditioner.Itisbasedonsplittingtheelectrical fieldintocurl-

freeanddivergence-freeprojectionsthatremovesthenullspaceof thediscretecurl-curloperatorinthesolutionprocess.

Anumberofefficientsolversandpreconditionersareused presentlyinordertoachievebothagoodaccuracyandhigh convergencerateoftheiterationprocess,especiallywhenhigh conductivitycontrastsincreasetheconditionnumberofthematrixofthesystemoflinearequations(SLE). Fomenko(1999), Varentsov(1999), Siripunvarapornetal.(2002), GrayverandStreich (2012), VarilsuhaandCandansayar(2018) carriedoutacomparativeanalysisofdifferentpreconditionersandsolvers.

Finally, Zhdanovetal.(1982) and Spichak(1999) constructed internalcriterionsforcontrollingtheaccuracyoftheforward modelingresults(seeSection 1.6.5.1 later).

1.3.2Thefiniteelementtechnique

The finiteelementtechniqueisespeciallyusefulforEMmodeling intheregionswithcomplicatedgeometryorreliefearthsurface. Duetothis finiteelementmethodbecameverypopularincomputationalelectromagnetics(Reddyetal.,1977; Pridmoreetal.,1981; Boyseetal.,1992; Livelybrooks,1993; Mogi,1996; Haber,1999; Sugengetal.,1999; Zunoubietal.,1999; ZysermanandSantos, 2000; Badeaetal.,2001; Jin,2002; MitsuhataandUchida,2004; FarquharsonandMiensopust,2011; MukherjeeandEverett, 2011; Schwarzbachetal.,2011; Silvaetal.,2012; Umetal.,2012, 2013; VieiradaSilvaetal.,2012; Kordyetal.,2013; Persovaetal., 2013; Puzyrevetal.,2013; Renetal.,2013a,b; Rivera-Rios etal.,2013; AnsariandFarquharson,2014; Caietal.,2014; JahandariandFarquharson,2014, 2015; Caietal.,2017a,b; Jahandari etal.,2017; Liuetal.,2018a,b; Xiaoetal.,2019).

Intheframeworkofthisapproach,thesolutionofEq. (1.1) in the finitedomain V isequivalentto findingthestationarypointof theenergyfunctional:

(Notethatwearedealingwiththedeterminationofastationary point,nottheminimumofthefunctional F becausethedifferential operatorontheright-handsideof (1.5) isnotpositivelydefinedand hencedoesnotsatisfytheconditionsoftheMinimumtheorem).

Haber(1999), Badeaetal.(2001), AnsariandFarquharson (2014), JahandariandFarquharson(2015) usetothisenda potential currentformulationoftheinitialproblembasedon theHelmholtzdecomposition.Theresultingmatrixhasaproperty

ofthediagonaldominancethat,inturn,allowsitsefficientsolutionevenforveryhighconductivitycontrasts.

Analternativewaytoovercomethesameproblemisbasedon usingofasinglevectorshapefunctionateachedgeofthegrid insteadofthreescalarfunctionsde finedincornernodes.Due tothis,onehastoonlysolvefortangentialcomponentsofthe electrical fieldalongtheedgesofcells(“edge-based finiteelements”).Thus,avoidingasolutionfornormalcomponentsof the fieldallowsmodelingforhighcontrasts(Sugentetal.,1999; MukherjeeandEverett,2011; Kordyetal.,2013; Rivera-Rios etal.,2013; Caietal.,2014; Xiaoetal.,2018).

JahandariandFarquharson(2014, 2015) haveintroduced mixed “finitevolume” schemesbasedonboththeEM fieldand thepotentialformulationsofMaxwell’sequations(Nedelec, 1986).Comparisonbetweenthreestaggeredgrid finitevolume andtwoedge based finiteelementmodelingschemes(Jahandari etal.,2017)showsthatwhiletheyarecomparableintermsofaccuracyandcomputationresources,the finiteelementsschemes areslightlymoreaccuratebutalsomoreexpensivethanthe finite volumeones.

Itisalsoimportanttooptimizethehexahedralirregular meshes. Persovaetal.(2013) eliminate “ unnecessary” nodes, while Cherevatovaetal.(2013) representsafullgridasavertical stackofsubgridseachofwhichisastandardstaggeredgrid.Itis refinedonlyinthehorizontaldirection,uniformlyacrossthe verticallayers,allowingonlyafactoroftwobetweenvertically adjacentsubgrids. Schwarzbachetal.(2011), Renetal.(2013a), and Liuetal.(2018b) usetothisendadaptiverefinementprocedure,whichiterativelyrefinesthe finiteelementgridguidedby globalposteriorerrorestimators.Basingonnumericalexperiments Renetal.(2013a) concludedthattheerrorestimatorusing facejumpsofnormalcomponentsofcurrentdensityembedded inthegoal-orientedadaptiverefinementprocedureshowsthe mostefficientperformance.

Inordertoreducethetotalcomputationtime, Zysermanand Santos,(2000) proposeda “mixedhybriddomaindecomposed iterativenonconforming” methodthatisbasedontheiterative decompositionofthemodeldomainandsolvingthelocallinear systemsofequations.Thisapproachallowsreductionofmemory andtimerequiredforsolutionoftheappropriatesystemoflinear equations.Noteinthisconnectionthatmoresignificant reductionofthecomputationtime(notonlyintheFEbasedalgorithms)couldbeachievedmymeansofusingparallelization schemes(AlumbaughandNewman,1996; Newmanand

Alumbaugh,1999; Wilsonetal.,1999; Kordyetal.,2013; Puzyrev etal.,2013; Newman,2014; Caietal.,2017a).

Despitethefactthat,theoretically, fi niteelementsarewell suitedtomodelingthegeologicalsectionswithcomplexdistributionsofelectricalconductivity,the fl exibilityofthemethod isattainedthroughconsiderablecomputingeffortsassociated withtheuseofcomplicated fi niteelements.Tocharacterizedifferentialequationmethodsingen eral,itshouldbeemphasized thattheyaregenerallymoreversatilecomparedtointegral equationmethods.However, toincreasetheircomputing ef fi ciency,anumberofproblemsdiscussedabovehavetobe properlysolved.

1.4Hybridschemes

Onewaytorefinethetechniquesofnumericalmodelingof3-D electromagnetic fieldsistoapplyhybrid(“mixed” or “nonclassical ”)approachesthatblendtheadvantagesofthedifferential andintegralequationmethods.Thisimpliesthedifferential equationmethodtobeusedinsidethemodelingdomain,which permitsconsiderationofamodelwithanarbitrarydistributionof theelectricalconductivityandintegralrelationsbetween field componentstobeemployedattheboundary,whichensures thepossibilityoflimitingthemodelingdomaintowithinan areathatonlyslightlyexceedstheinhomogeneityindimensions.

The firststeponthiswaywasmadeby Weidelt(1975) who usedtheGreenfunctionsapparatustoremovefromaconsiderationtheregionsofthenormalsectionaboveandbelowthe inhomogeneities:

where S isthesurfaceofthediscontinuityseparatingthelayers, Ea istheanomalous(secondaryorscattered) field(Ea hE Eb ),and thepoints r0 areselectedintheregiontoberemoved.

Theef fi ciencyofhybridschemeswasfurtherimprovedby reducingthemodelingdomainnotonlyverticallybuthorizontallyaswell.Thus,in(Guptaetal.,1987; Leeetal.,1981; PridmoreandLee,1980 ; Bestetal.,1985 ),anareaoflimiteddimensionsiscoveredoverasmalldistancebya fi niteelement grid.Applyingthe fi niteelementtechniquetotheinnerpartof thedomainandtheintegralequationtechniquetotheouter partgivesrisetotwosystemsofequations:(1) fi niteelement

equationsintheinnernodes,withtherespectivepartofthematrixshowingacharacteristicbandstructureand(2)equations thatrelatetheunknownsontheboundaryandinnernodes (withtherespectivepartofthematrixbeing fi lled)inaccordance withtheformula

where r0 liesontheboundaryofthedomain V.

Renetal.(2014) haveproposeda finiteelementschemewitha hybridboundaryelement(BE),whichenablescompleteremoval ofthevolumediscretizationoftheairspace,which,inturn,providesthecapabilityofsimulatinglarge-scalecomplicatedgeoelectromagneticinductionproblems.Intheframeworkofthis approach,thesurfaceintegralformulaintermsofthereduced electricalvectorpotentialisapproximatedusingthepointcollocationboundaryelementmethod.Numericalexperimentsshow thatatlowfrequencies,wherethequasistaticapproximationis applicable,standardFEmethodsprovetobesuperiortothe hybridBE FEsolutionsintermsofcomputationaltime,because theFEmethodrequiresonlyacoarsediscretizationoftheair domainandoffersanadvantageoussparsityofthesystemmatrix. AtradiomagnetotelluricfrequenciesofafewhundredkHz,the hybridBE FEschemeturnstobesuperiortotheFEmethod becauseitavoidsexplicitstorageofthesystemmatricesaswell asdensevolumediscretizationoftheairdomainrequiredbyFE methodsathighfrequencies.

Hybridschemesarecommonlygroupedintotwotypes:direct anditerative.Indirecthybridtechniques,thematrixofthesystemisinverteddirectly,whereasiniterativetechniques,the valuesoftheunknownsatthedomainboundariesareat first assumedtobeknown,andthenthevaluesintheinnernodes arecomputed,whicharerecalculatedby (1.7) intoboundary values,etc.Asappearsfromtheestimationperformedby Lee etal.(1981),theiterativehybridschemesaregenerallymore sparingintermsoftheoperationalmemoryresourcesandCPU timeused.Similarlytoothernumericalmodelingtechniquestheir furtherreductionrequiresusingparallelcomputing.

Therefore,despitecertainadvantagesofhybridsystems,their practicalapplicationentailsanumberofdifficulties.Inparticular,therequirementsoncomputerresourcesremainrather high,convergenceoftheiterativeprocedureisnotguaranteed (ifthemostresource-savingiterativeschemeisconsidered)and nointernalcriteriafortheaccuracyoftheresultsobtainedare

used(Thelastobservation,whichalsoappliestotherestofthe numericalapproachesjustmentioned,isperhapsmostcritical forhybridschemes).

1.5Analog(physical)modelingapproaches

Analog,orphysical,modelingimpliesaphysicalnatureforthe modeloftheprimary fieldsource,themedium,andtheinhomogeneity.Twobasicapproachesaredistinguishedhere:thecontinuousmediamethodandtheelectricalcircuitmethod(Tetelbaum andTetelbaum,1979).Intheformerapproach,themodelisdeterminedbythe fieldofanelectricalcurrentinacontinuousmedium (Dosso,1966),whilethelatterinvolvestheuseofelectricalcircuits withconcentratedparameters(Brewitt-TaylorandJohns,1980).

Thecontinuousmediamethodisbasedontheuseoftheelectrodynamicsimilitudecriterion(Stratton,1941):

mus L2 ¼ inv ; (1.8)

where L isthecharacteristiclineardimensionofthemodel.Among theadvantagesofthisapproach,oneshouldrecallitssimplicity andlowcost,theuniquenessoftheequipment,andthepossibility ofmodelingthemediawithsharpelectricalconductivitycontrasts (Berdichevskyetal.,1987; Farquharsonetal.,2006).

Someoftheseadvantages,however,provetobedrawbacksif theissueisaddressedinabroadercontext,namely,fromthe standpointofelectromagnetic fieldsmodelinginrealsituations. Theextremelysmallnumberofphysicalmodelinginstallations allovertheworldvirtuallyobviatestheirapplicationasatoolfor analyzingtheobserved fields,letalonetherepeateduseofthe modelingresults,whichisonlypossiblethroughcouplingphysical installationstocomputingfacilities.Inaddition,oneshouldnotea numberoftechnologicdifficulties,suchasthechoiceofmaterials withasufficientlywiderangeofelectricalconductivityvariations butnotsubjecttofrequencydispersion,thepainstakingtechnologicimplementationofmodelsformultilayermedia,thepoorly developedmodelingtechniqueforusewithhardmaterialsand low-meltingmetals,andthedifficultiesarisinginobtainingexperimentalcurvesoverabroadfrequencyrange.

Theotherimportantapproachinphysicalmodelingistheelectricalcircuitmethod,whichinactualpracticeisrealizedintwomodifications.The firstisbasedondiscretizationofthemodeling domainfollowedbytherepresentationoftheelementaryvolumes obtainedbymeansofelementsofanelectricalcircuit(replacement

schemes)andmeasurementofvoltagesandcurrentsinthecircuit, whichmodeltheelectricalandmagnetic fields,respectively.The otherisbasedonobtainingreplacementcircuitsdirectlyfrom finite differenceequationsthatdescribethe fieldbeingmodeled(BrewittTaylorandJohns,1980).

Aswaspointedoutin(TetelbaumandTetelbaum,1979,p. 211), “theprimeadvantageofelectricalcircuitsisthattheyenable onetomodelthree-dimensional fields,whicharedescribedby equationswitharight-handside”.Anotherimportantadvantage ofthisapproachisthatitcanbeusedinhybridanalog/numerical installations.Itisalsoworthwhiletonotethatwiththisapproach, theuseofthediacopticsconceptof Kron(1972) mayprove noteworthy.

Despitetheoutwardmeritsoftheaboveapproach,however,it isnotfreeofthecharacteristic flawsofboththeotherphysical modelingtechniques(measurementerrorsandtechnicaldifficultiesinconstructingthemodels)andmathematicalmodeling approaches(e.g.,errorsensuingfromthediscretizationofthe modelingdomainand fieldequations).

Wehavediscussedthemaintechniquesandapproaches currentlyinuseformodeling3-Delectromagnetic fields. Collatingtheiradvantagesanddrawbacksshowsthedifferential equationmethodstobethemostversatileandbestsuitedfor modelingthebroadestspectrumofpracticalsituations.However, theirapplicationinvolvesanumberoftheoretical,methodological,andcomputingproblemsthataffectthemodelingefficiency. Inthefollowingsections,thesecrucialissueswillbeaddressedin thecontextofthebalancetechnique.

1.6BalancetechniqueforEMfield computation

Inpublications(ZhdanovandSpichak,1980, 1989; Spichak, 1983a , 1985, 1999),basicprincipleswereformulatedofconstructinganalgorithmforthisproblembearingonamoderate-speed computerwithalimitedcorememory(forinstance,minicomputers).Inwhatfollows,wewilldescribethisalgorithmand demonstrateitsapplicationtosolutionoftheforwardproblem in3-Daxialsymmetricsituations.

1.6.1Governingequations

Letsomedomain U intheEarth’scrustoruppermantlebe isotropic,nonmagnetic(mhm0 isthepermeabilityoffreespace),

andtohaveathree-dimensionaldistributionofelectricalconductivity sðx ; y ; z Þ thatcanberepresentedintheform:

sðP Þ¼ 0 @

sðz Þ; P ˛U1

sðx ; z Þ; P ˛U2

sðx ; y ; z Þ; P ˛U3 ;

where U ¼ U1 WU2 WU3 , P ¼ P ðx ; y ; z Þ; with U3 u 0.Theelectromagnetic fieldinthedomain U isinducedbyaplanewaveverticallyincidentontheEarth’ssurface.Thetimedependenceofthe fieldisgivenbyfactor expð i ut Þ.

Giventheconductivitydistribution sðx ; y ; z Þ,itisnecessaryto determineelectricalandmagnetic fieldseverywhereinthe domain P ¼ UWU0 ,where U0 istheloweratmosphereadjacent to U.

Overperiodsofinteresttogeophysics,the fieldinthedomain S isquasistationaryandsatisfiestheMaxwellequations: V H ¼ sE; (1.9) V E ¼ i um0 H (1.10)

Eq. (1.9) (1.10) yieldtheelectrical fieldequations: DE VðV , EÞþ k 2

where k ¼ði um0 Þ1=2 ; Rek > 0

TakingthedivergencefromthetwosidesofEq. (1.8),we derive sðV , EÞþðE; VsÞ¼ 0(1.12)

Withdueaccounttakenof (1.12),Eq. (1.11) takestheform:

E þ VðE; Vln sÞþ k 2 E ¼ 0 (1.13)

Todeterminetheelectromagnetic fieldinthedomain S we havetosolveaboundaryvalueproblemforthe field E satisfying Eq. (1.13) within S andthencalculate H.

1.6.2Boundaryconditions

Letusnowconsiderthechoiceofboundaryconditionsatthe boundariesofthedomain S: Thevaluesofelectrical fieldorof itsnormalderivativearenotknownbeforehand.So,acommon approachisbasedontheassumptionthatanomalous electromagnetic fieldvanishesattheboundarieslocatedat finite distancefromtheanomaly(Dirichlettypeboundaryconditions).

D

However,thiscausestwoproblems: first,itisdifficulttodeterminethisdistancecorrectlyand,second,inordertoensureits correctnessitisnecessarytoplacetheboundariesofthemodeling domainfarenoughfromtheanomaly(especiallyinthenonconductingparts),which,inturn,leadstounnecessaryincreasing ofthecomputationtime.

Oneofpossiblesolutionsmakeuseofintegralboundaryconditions(Weidelt,1972),butitcouldbeverydifficulttousethemin practicebecauseoftheirentailingconsiderablecomputing difficulties.Inthisconnection,itwouldbeinterestingtoconsider anothertypeofboundaryconditionsthatarebasedontheimplicitaccountofthecharacteroftheanomalous fielddecaying farfromthemediumheterogeneities. Spichak(2006) suggested ageneralapproach,whichmaybeusedtoconstructthedifferentialboundaryconditionsforthevector fields,satisfyingthe Helmholtzequation.Itisbasedontheexpansionofthe fieldsin aseriesofmultipolesanduseof “canceling ” operators.

Inparticular,thealgorithminquestionmakesuseofthe asymptoticboundaryconditionsofthe firstorder:

ikr þ r v vr ðE Eb Þ¼ 0; (1.14)

where Eb isthebackgroundelectrical fieldcorrespondingtothe casewhen shsðz Þ forall P ˛ P; r isthedistancebetweenthe pointslyingontheboundaryofthedomain P : andtheorigin ofthecoordinates.

Basingontheresultsofthenumericalmodeling Spichak (2006) hasconcludedthatitispossibletomakethefollowingrecommendationsusefulforapplicationoftheboundaryconditions whensolvingtheexternalboundaryvalueproblemsof geoelectromagnetics:

-theearthresponsesarenotequallyaffectedbytoosmalldistancebetweenthetargetandthemodelingdomainboundaries:theapparentresistivityismoresensitiveinthecaseof theresistivetargetwhilethephasesaremoresensitiveinthe caseoftheconductiveone; -the “safe” distanceforusingtheDirichlet-typeboundaryconditions E !a ¼ 0shouldnotbelessthan0.5 d,where d isaskin depthinthebackgroundstructure.

1.6.3Discretizationscheme

Toobtainasystemoflinearalgebraicequationsanditssubsequent computer-aidedsolution,atransitionfromcontinuoustodiscrete

valuesisrequired.Variousmechanismsforsuchatransitionare mainlybasedon(1)theapplicationofTaylorseries,(2)thevariationalformulation,and(3)integrationoftheprimaryequation. Eachoftheseapproacheshasitsadvantagesandrespective fields ofapplicability.Thus,the firsttechniqueapplicabletodifferential equationsingeneralcaseisusedmostoftenindefiningtheorder ofapproximationindifferenceschemes.Thevariationalformulationisinvitinginthat,initscontext,naturalboundaryconditions areadirectconsequenceoftherelevantfunctionalbeingstationary (Pridmoreetal.,1981).Lastly,thethirdapproach,oftenreferredto asthe “balancetechnique,” isapplicableingeneralcaseandyields particularlysimplediscreteschemeswithinternalboundariesand nonuniformgrids(Spichak,1983a).LetusrefertothislasttechniquetoobtaindiscreteanalogsofEq. (1.13).

Providedthattheelectricalconductivityintheregion variesstepwise,derivativesofthefunctionbeingsoughtfor mayundergobreaks.Itthusisworthwhiletocomputethe valuesofthefunctionitselfatt henodesofacertaingridwhile specifyingvaluesoftheelectricalconductivityfunctionatthe nodesofanintermediategrid( Brewitt-TaylorandWeaver, 1976 )withnoworrytosatisfyinternalboundaryconditions. Similarideaisrealizedinthestaggeredgridsapproach mentionedearlier.

Toderivediscreteequationsforspacegridnodes,letus proceedfromthecontinuousvectorfunction E tothediscrete one Ul ;m;n ,de finedatthenodesofthemaingrid.Integrating Eq. (1.13) withrespecttothevolumeofanelementarycellin thevicinityofthenode(l ; m; n)(Fig.1.1),weobtainanequation forelectricalcurrentbalance:

Substitutingthederivativesof U and s inEq. (1.15) by finite differencesandapproximatingtheintegralsbythetrapezium rule,wearriveatalinearalgebraicequationrelatingthevalues ofthevectorfunctionUonlyinsevenadjacentnodes:

where b Dði Þ l ;m;n (i ¼ 1,2,.6)arethematrixcoefficientshavingasize (3 3)anddeterminedbythegridgeometry,distributionofthe electricalconductivity s andEM fieldfrequency. NotethatnumericalapproximationofEq. (1.13) hasanumber ofadvantagescomparedtotheapproximationofthecorrespondingsecond-orderdifferentialEq. (1.11): -thebalanceofcurrentsisholdwithineachcellofthespatial grid; -theuseofEq. (1.13) makesitpossibletoavoidtheapproximationofmixedsecondderivativesin (1.11) andexplicitlyenforcescurrentdivergence-freeconditionsateachgridnode (comparewithstaggeredgridapproximation(Smith,1996a )); -theapproximationofthethirdterminEq. (1.13) isaccurate enoughinthevicinityofelectricalconductivitycontrastsas well;

Figure1.1 Elementarycellofaspatialgrid.

StructureofmatrixD:L,M,N numberofnodesinX,Y,andZ axes,respectively;Q ¼ L M N, b D

Þ (3 3) submatrix.

-intheregionswhere s ¼ 0or Vs ¼ 0thesecondtermofthe equationautomaticallyvanishesandinapproximationof (1.15) thetotalerrordecreases.

ThematrixofacorrespondingSLEhasablock-bandedshape andisverysparse(Fig.1.2).

Suchsystemsaresolvedmoreefficientlybyiterativemethods, whichenablethemosteconomicaluseoftheCPUcorememory. Inparticular,in(Spichak,1999)SLE (1.16) wassolvedusingconjugategradienttechniqueBiCGstab(Steijpenetal.,1994)afterits preconditioningbymeansofthediagonal(Jacobi)scaling.

1.6.4Calculationofthemagneticfield

Themagnetic field H canbedetermineddirectlyfromEq. (1.4) or, asnotedabove,itcanbesimplyrecalculatedfromthedetermined electrical fieldbyitsdifferentiationintermsofthe finitedifference (or finiteelement)approximationofEq. (1.10).Indoingso,we havetoovercomethedifficultyarisinginregionswithlargeconductivitygradients,whichmaycauseblundersinthecalculation oftherelevantelectrical fieldderivativesand,eventually,incorrectvaluesofthemagnetic field.Particularly,inthecalculation ofthehorizontalcomponents Hx and Hy attheEarth’ssurface accordingtoformulas (1.10),theapproximationofderivatives dEx =dz and dEy =dz isnotstable.

Apossiblewaytotacklethisproblemliesinconstructinga splineoverasetofvaluesofthegridfunctionandthenusing theanalyticallyobtainedderivativestocomputethemagnetic

Figure1.2

fieldcomponentsfromEq. (1.10).Thistechniquegivessatisfactoryresultsintwo-dimensionalcasewhenthesolutioninvolves onlyonecomponentoftheelectrical field,butitprovestobeunstableinthree-dimensionalcase(Pridmoreetal.,1981).

Anotherapproachliesincomputingtheanomalousmagnetic fieldbymeansofnumericalintegrationofexcesscurrentscirculatingintheanomalousregion Va :

where Ds ¼ s sb .

Despitethefactthatthisprocedureisstable,inordertodeterminethemagnetic fieldtherelevantGreenstensormustbe known.Iftheelectrical fieldwascalculatedusingoneoftheintegralequationtechniques(seeSection 1.2),thencomputingthe magnetic fieldwouldnotrequireanyadditionalcalculationsof theGreenstensor.However,ifthecomputationisperformedby meansofadifferentialequationtechnique(seeSection 1.3), determiningthemagnetic fieldby Formula(1.16) wouldrequire computingtheGreentensor,whichinvolvesconsiderablecomputationexpenses.

Fromthispointofview,thewayproposedby Spichak(1999) appearstobemorepreferable.Inthisapproach,magnetic field componentsaredeterminedintwosteps.First,inthecentersof thegridcells,theverticalcomponentofthemagnetic field Hz is found,andthen,usingHilberttransforms(see,forinstance, (Zhdanov,1988)),tangentialcomponents Hx and Hy are computedatthegridnodes:

where r ¼½ðx x0 Þ2 þðy y0 Þ2 1=2 ;thevalueof Hz isprecalculatedaccordingto Formula(1.10); H b x and H b y arethebackground magnetic fieldcomponentsattheEarth’ssurface.

Theabovealgorithmfortheforwardproblemsolutionwas realizedasasoftwarepackageFDM3D(Spichak,1983b )and widelyusedformodelingEM fieldsin3-Dmedia(Spichak,1999, 2015).

1.6.5Controllingtheaccuracyoftheresults

1.6.5.1Criteriaforaccuracy

Inmostcases,theaccuracyofmodelingEM fieldscanonlybe evaluatedindirectly,becausetheexistingcontroltechniques provide,asarule,necessarybutnotsuf ficientconditionsfor theaccuracyoftheresults.

“External” and “internal” techniquesareemployedforaccuracycontrol.Externaltechniquesinclude (a) comparisonwiththeresultsofotherauthors; (b) comparisonwiththeresultsobtainedbyothermethods, includinganalyticalones.

Thusfar,thesetwocriteriahavebeenusedmainlyfor checkingthecorrectnessofsolutions.Theresultsofnumerous comparisons,however,clearlyshowtheaboveaccuracycriteria tobeinsufficient.Controltechniquesthatenabletheaccuracy estimationbyaninternalmeansofoneoranothernumerical approachshouldbeapplied.

Aspointedoutby PoggioandMiller(1973), “theideaof numericalevaluationoftheaccuracyofatechniqueattimesappearstobeinternallyinconsistent,becauseitseemsthatnumericalresultscannotbeusedtoconfirmtheirowncorrectness.In reality,thereareseveralapproachesthatenableamanifestincorrectnessofnumericalresultstobedetected.” Suchcriteriainclude testingthefollowing:

-thereciprocitytheorem; -theenergyconservationlaw;

-theaccuracytowhichtheequationsandboundaryconditions aresatisfied; -convergenceofthesolutionwithdecreasingsizeofgridcells, numberofunknowns,etc.

Finally,onemayconsiderasamixedtype(inemployingdifferentialequationtechniques)thecriterionproposedby Zhdanov etal.(1982).Itisbasedonanestimationoftheaccuracytowhich theintegralidentity

issatisfiedbysubstitutinginitthesolutionobtainedbythe finite differenceor finiteelementtechniques.Unfortunately,numerical implementationofthiscriterionrequiresatime-consuming computationofGreentensorforalayeredmedium,whichreducesitspracticalvalue.

Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.