7/21/2015
ErrorAnalysis
PRACTICALMATHEMATICS
ErrorAnalysis
Contents
SystematicErrors
RandomErrors
Resolution
PropagationofErrors
SignificantDigitsinNumericalValues
SuggestedReading
SystematicErrors
[Link]
occurbecause:
thereissomethingwrongwiththeinstrumentoritsdatahandlingsystem,or
becausetheinstrumentiswronglyusedbytheexperimenter.
Twotypesofsystematicerrorcanoccurwithinstrumentshavingalinearresponse:
1. Offsetorzerosettingerrorinwhichtheinstrumentdoesnotreadzerowhenthequantitytobe
measurediszero.
2. Multiplierorscalefactorerrorinwhichtheinstrumentconsistentlyreadschangesinthequantity
tobemeasuredgreaterorlessthantheactualchanges.
[Link]
calibrationoftheinstrumentisnotknowncorrectly.
[Link](fullline).
Brokenlineshowsresponseofanidealinstrumentwithouterror.
Examplesofsystematicerrorscausedbythewronguseofinstrumentsare:
errorsinmeasurementsoftemperatureduetopoorthermalcontactbetweenthethermometerand
thesubstancewhosetemperatureistobefound,
errorsinmeasurementsofsolarradiationbecausetreesorbuildingsshadetheradiometer.
Theaccuracyofameasurementishowclosethemeasurementistothetruevalueofthequantitybeing
[Link]
1/4
7/21/2015
ErrorAnalysis
[Link],whicharedifficultto
detectevenforexperiencedresearchworkers.
RandomErrors
Randomerrorsinexperimentalmeasurementsarecausedbyunknownandunpredictablechangesinthe
[Link].
Examplesofcausesofrandomerrorsare:
electronicnoiseinthecircuitofanelectricalinstrument,
irregularchangesintheheatlossratefromasolarcollectorduetochangesinthewind.
RandomerrorsoftenhaveaGaussiannormaldistribution(seeFig.2).Insuchcasesstatisticalmethods
[Link]
bestestimateofthatquantity,andthestandarddeviationsofthemeasurementsshowstheaccuracyof
[Link]/sqrt(n),wherenisthenumberofmeasurements.
[Link].m=meanofmeasurements.s=standarddeviationof
measurements.68%ofthemeasurementslieintheintervalms<x<m+s95%liewithinm2s<x
<m+2sand99.7%liewithinm3s<x<m+3s.
Theprecisionofameasurementishowcloseanumberofmeasurementsofthesamequantityagreewith
[Link]
measurements.
Resolution
Theresolutionofaninstrumentisthesmallestpossibledifferencebetweenseparatevaluesoftheoutput
(inotherwordshowfinethescaleoftheinstrumentis).
Ahighresolutionisnecessaryforhighaccuracy,butitdoesnotgivehighaccuracywhentherearelarge
[Link]
measurementstobemade,butthereisnoneedforaresolutiongreaterthanthesystematicandrandom
errorsintheexperiments.
PropagationofErrors
Whenmeasurementsareusedtocalculatearesult,theaccuracyoftheresultislimitedbythepropagation
[Link]:
1. errorsintheoriginalmeasurements,
[Link]
2/4
7/21/2015
ErrorAnalysis
2. roundofferrorsinthecalculation.
Inmanycalculationswemayusetwoorthreesignificantdigitsmorethanthosejustifiedbytheaccuracy
[Link],whenthenumberofarithmeticaloperationsisverylarge(whichis
possiblewhenacomputerisused),greatcareisneededtomakesurethattheerrorsdonotaccumulate
andgiveinaccurateresults.
[Link]
[Link],theerrorinasum(oradifference)isthesum(or
difference)oftheerrorsintheseparateterms:
d(x+y)=dx+dy,d(xy)=dxdy.
Sincewedonotknowwhethertheerrorsarepositiveornegative,theseequationsarereplacedby
equationsfortheabsoluteerrorsasfollows:
|d(x+y)|=|dx|+|dy|.
Thefractional,orrelative,errorinaproduct(orquotent)isthesum(ordifference)ofthefractional,or
relative,errorsinthefactorsoftheproduct(orquotient):
d(x.y)/(x.y)=dx/x+dy/y,d(x/y)/(x/y)=dx/xdy/y.
Asbefore,sincewedonotknowwhethertheerrorsarepositiveornegative,weuseequationsforthe
absolutevaluesoftherelativeerrors,asfollows:
|d(x.y)/(x.y)|=|dx/x|+|dy/y|,|d(x/y)/(x/y)|=|dx/x|+|dy/y|.
Ifaquantityuiscalculatedfromseveralmeasurementsx1,...,xnbymeansofafunctionfasfollows:
u=f(x1,...,xn),
then
|du|=|u1dx1|+...+|undxn|,
whereu1,...,unarethevaluesofthepartialderivativesoffwithrespecttox1,...,xn.
SignificantDigitsinNumericalValues
[Link]
that15.6hasthreesignificantdigits,[Link]
additionofmoredigits(forexample,15.620478)wouldbemeaningless.
Inmultiplicationsanddivisions,thenumberofsignificantdigitsinthefinalresultshouldbethesameas
thenumberofsignificantdigitsintheleastaccuratefactor.
EXAMPLE
(36.4792.6)/14.85=6.3868956=[Link]
[Link]
3/4
7/21/2015
ErrorAnalysis
intermediatestepsofthecalculation,thefinalresulthasonlytwosignificantdigits,becausethe
originalfactor2.6hasonlytwosignificantdigits.
Inadditionsandsubtractions,thenumberofdigitsafterthedecimalpointinthefinalresultshouldbethe
sameasthesmallestnumberofdigitsafterthedecimalpointinthetermsofthesumsordifferences.
EXAMPLE
17.524+2.43.56=16.364=[Link]
decimalpointbecausetheoriginalterm2.4hasonlyonedigitafterthedecimalpoint.
EXERCISES
1. FindthevolumeVofacircularcylinderwithradiusr=1.26cmandheighth=7.3cm.(V=pi.r2.h)
2. Findthesum:0.056102+11.8101
SuggestedReading
1. Topping,J.(1972).ErrorsofObservationandTheirTreatment(4thEdition),London:Chapman
andHall.[KMUTTLibrarycallnumber:QA275TOP1972]
2. Acton,FormanS.(1959).AnalysisofStraightLineData,NewYork:Dover.[KMUTTLibrary
callnumber:QA276ACT1959]
[Link]
4/4